From aa5db684382bd8662a83ca09ed000e4a5a1013f9 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 16 Jul 2009 14:14:32 +0100 Subject: softpipe: remove backwards dependency from tilecache to softpipe The tile cache is a utility, it shouldn't know anything about the entity which is making use of it (ie softpipe). Remove softpipe parameter to all the tilecache function calls, and also remove the need to keep a softpipe pointer in the sampler structs. --- src/gallium/drivers/softpipe/sp_context.c | 8 ++---- src/gallium/drivers/softpipe/sp_flush.c | 6 ++--- src/gallium/drivers/softpipe/sp_quad_blend.c | 6 ++--- src/gallium/drivers/softpipe/sp_quad_colormask.c | 3 +-- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 2 +- src/gallium/drivers/softpipe/sp_quad_output.c | 3 +-- src/gallium/drivers/softpipe/sp_quad_stencil.c | 2 +- src/gallium/drivers/softpipe/sp_state_derived.c | 22 ++++++++++++++++ src/gallium/drivers/softpipe/sp_state_sampler.c | 2 +- src/gallium/drivers/softpipe/sp_state_surface.c | 4 +-- src/gallium/drivers/softpipe/sp_tex_sample.c | 32 ++++++++--------------- src/gallium/drivers/softpipe/sp_tex_sample.h | 5 ++-- src/gallium/drivers/softpipe/sp_tile_cache.c | 20 ++++++-------- src/gallium/drivers/softpipe/sp_tile_cache.h | 12 +++------ 14 files changed, 62 insertions(+), 65 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 86df320ea8..f085889d3a 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -73,8 +73,8 @@ softpipe_unmap_transfers(struct softpipe_context *sp) uint i; for (i = 0; i < sp->framebuffer.nr_cbufs; i++) - sp_flush_tile_cache(sp, sp->cbuf_cache[i]); - sp_flush_tile_cache(sp, sp->zsbuf_cache); + sp_flush_tile_cache(sp->cbuf_cache[i]); + sp_flush_tile_cache(sp->zsbuf_cache); for (i = 0; i < sp->framebuffer.nr_cbufs; i++) { sp_tile_cache_unmap_transfers(sp->cbuf_cache[i]); @@ -254,8 +254,6 @@ softpipe_create( struct pipe_screen *screen ) /* vertex shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples_vertex; - softpipe->tgsi.vert_samplers[i].unit = i; - softpipe->tgsi.vert_samplers[i].sp = softpipe; softpipe->tgsi.vert_samplers[i].cache = softpipe->tex_cache[i]; softpipe->tgsi.vert_samplers_list[i] = &softpipe->tgsi.vert_samplers[i]; } @@ -263,8 +261,6 @@ softpipe_create( struct pipe_screen *screen ) /* fragment shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples_fragment; - softpipe->tgsi.frag_samplers[i].unit = i; - softpipe->tgsi.frag_samplers[i].sp = softpipe; softpipe->tgsi.frag_samplers[i].cache = softpipe->tex_cache[i]; softpipe->tgsi.frag_samplers_list[i] = &softpipe->tgsi.frag_samplers[i]; } diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c index 4a14d49686..732277a2c5 100644 --- a/src/gallium/drivers/softpipe/sp_flush.c +++ b/src/gallium/drivers/softpipe/sp_flush.c @@ -52,17 +52,17 @@ softpipe_flush( struct pipe_context *pipe, if (flags & PIPE_FLUSH_TEXTURE_CACHE) { for (i = 0; i < softpipe->num_textures; i++) { - sp_flush_tile_cache(softpipe, softpipe->tex_cache[i]); + sp_flush_tile_cache(softpipe->tex_cache[i]); } } if (flags & PIPE_FLUSH_RENDER_CACHE) { for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) if (softpipe->cbuf_cache[i]) - sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]); + sp_flush_tile_cache(softpipe->cbuf_cache[i]); if (softpipe->zsbuf_cache) - sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache); + sp_flush_tile_cache(softpipe->zsbuf_cache); /* Need this call for hardware buffers before swapbuffers. * diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index b1e18805c7..04b5daf3a4 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -130,8 +130,7 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad) uint *dst4 = (uint *) dst; uint *res4 = (uint *) res; struct softpipe_cached_tile * - tile = sp_get_cached_tile(softpipe, - softpipe->cbuf_cache[cbuf], + tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], quad->input.x0, quad->input.y0); float (*quadColor)[4] = quad->output.color[cbuf]; uint i, j; @@ -260,8 +259,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { float source[4][QUAD_SIZE], dest[4][QUAD_SIZE]; struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe, - softpipe->cbuf_cache[cbuf], + = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], quad->input.x0, quad->input.y0); float (*quadColor)[4] = quad->output.color[cbuf]; uint i, j; diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c index dc90e5d5e9..89efbe3b02 100644 --- a/src/gallium/drivers/softpipe/sp_quad_colormask.c +++ b/src/gallium/drivers/softpipe/sp_quad_colormask.c @@ -54,8 +54,7 @@ colormask_quad(struct quad_stage *qs, struct quad_header *quad) for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { float dest[4][QUAD_SIZE]; struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe, - softpipe->cbuf_cache[cbuf], + = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], quad->input.x0, quad->input.y0); float (*quadColor)[4] = quad->output.color[cbuf]; uint i, j; diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index d463930bae..768b9275b3 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -60,7 +60,7 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) unsigned zmask = 0; unsigned j; struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); + = sp_get_cached_tile(softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); assert(ps); /* shouldn't get here if there's no zbuffer */ diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c index 92d5f9f3c1..dd8f5377e9 100644 --- a/src/gallium/drivers/softpipe/sp_quad_output.c +++ b/src/gallium/drivers/softpipe/sp_quad_output.c @@ -50,8 +50,7 @@ output_quad(struct quad_stage *qs, struct quad_header *quad) /* loop over colorbuffer outputs */ for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe, - softpipe->cbuf_cache[cbuf], + = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], quad->input.x0, quad->input.y0); float (*quadColor)[4] = quad->output.color[cbuf]; int i, j; diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c index 5e9d447737..34a8d9e9f6 100644 --- a/src/gallium/drivers/softpipe/sp_quad_stencil.c +++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c @@ -206,7 +206,7 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad) ubyte ref, wrtMask, valMask; ubyte stencilVals[QUAD_SIZE]; struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); + = sp_get_cached_tile(softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); uint j; uint face = quad->input.facing; diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 75551000c9..75be99768c 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -184,11 +184,33 @@ compute_cliprect(struct softpipe_context *sp) } +static void +update_tgsi_samplers( struct softpipe_context *softpipe ) +{ + unsigned i; + + /* vertex shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + softpipe->tgsi.vert_samplers[i].sampler = softpipe->sampler[i]; + softpipe->tgsi.vert_samplers[i].texture = softpipe->texture[i]; + } + + /* fragment shader samplers */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i]; + softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i]; + } +} + /* Hopefully this will remain quite simple, otherwise need to pull in * something like the state tracker mechanism. */ void softpipe_update_derived( struct softpipe_context *softpipe ) { + if (softpipe->dirty & (SP_NEW_SAMPLER | + SP_NEW_TEXTURE)) + update_tgsi_samplers( softpipe ); + if (softpipe->dirty & (SP_NEW_RASTERIZER | SP_NEW_FS | SP_NEW_VS)) diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c index cb517b02e4..aa2f3f2ccd 100644 --- a/src/gallium/drivers/softpipe/sp_state_sampler.c +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -97,7 +97,7 @@ softpipe_set_sampler_textures(struct pipe_context *pipe, struct pipe_texture *tex = i < num ? texture[i] : NULL; pipe_texture_reference(&softpipe->texture[i], tex); - sp_tile_cache_set_texture(pipe, softpipe->tex_cache[i], tex); + sp_tile_cache_set_texture(softpipe->tex_cache[i], tex); } softpipe->num_textures = num; diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c index 7c06d864a7..1621a27614 100644 --- a/src/gallium/drivers/softpipe/sp_state_surface.c +++ b/src/gallium/drivers/softpipe/sp_state_surface.c @@ -53,7 +53,7 @@ softpipe_set_framebuffer_state(struct pipe_context *pipe, /* check if changing cbuf */ if (sp->framebuffer.cbufs[i] != fb->cbufs[i]) { /* flush old */ - sp_flush_tile_cache(sp, sp->cbuf_cache[i]); + sp_flush_tile_cache(sp->cbuf_cache[i]); /* assign new */ sp->framebuffer.cbufs[i] = fb->cbufs[i]; @@ -68,7 +68,7 @@ softpipe_set_framebuffer_state(struct pipe_context *pipe, /* zbuf changing? */ if (sp->framebuffer.zsbuf != fb->zsbuf) { /* flush old */ - sp_flush_tile_cache(sp, sp->zsbuf_cache); + sp_flush_tile_cache(sp->zsbuf_cache); /* assign new */ sp->framebuffer.zsbuf = fb->zsbuf; diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index a1d3bade27..3daa88eedd 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -668,10 +668,8 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - struct softpipe_context *sp = samp->sp; - const uint unit = samp->unit; - const struct pipe_texture *texture = sp->texture[unit]; - const struct pipe_sampler_state *sampler = sp->sampler[unit]; + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; if (x < 0 || x >= (int) texture->width[level] || y < 0 || y >= (int) texture->height[level] || @@ -685,7 +683,7 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, const int tx = x % TILE_SIZE; const int ty = y % TILE_SIZE; const struct softpipe_cached_tile *tile - = sp_get_cached_tile_tex(sp, samp->cache, + = sp_get_cached_tile_tex(samp->cache, x, y, z, face, level); rgba[0][j] = tile->data.color[ty][tx][0]; rgba[1][j] = tile->data.color[ty][tx][1]; @@ -840,10 +838,8 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, const unsigned faces[4]) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_context *sp = samp->sp; - const uint unit = samp->unit; - const struct pipe_texture *texture = sp->texture[unit]; - const struct pipe_sampler_state *sampler = sp->sampler[unit]; + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; unsigned level0, level1, j, imgFilter; int width, height; float levelBlend; @@ -992,10 +988,8 @@ sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler, float rgba[NUM_CHANNELS][QUAD_SIZE]) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_context *sp = samp->sp; - const uint unit = samp->unit; - const struct pipe_texture *texture = sp->texture[unit]; - const struct pipe_sampler_state *sampler = sp->sampler[unit]; + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; /* get/map pipe_surfaces corresponding to 3D tex slices */ unsigned level0, level1, j, imgFilter; int width, height, depth; @@ -1139,10 +1133,8 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, float rgba[NUM_CHANNELS][QUAD_SIZE]) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_context *sp = samp->sp; - const uint unit = samp->unit; - const struct pipe_texture *texture = sp->texture[unit]; - const struct pipe_sampler_state *sampler = sp->sampler[unit]; + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; const uint face = 0; unsigned level0, level1, j, imgFilter; int width, height; @@ -1216,10 +1208,8 @@ sp_get_samples(struct tgsi_sampler *tgsi_sampler, float rgba[NUM_CHANNELS][QUAD_SIZE]) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_context *sp = samp->sp; - const uint unit = samp->unit; - const struct pipe_texture *texture = sp->texture[unit]; - const struct pipe_sampler_state *sampler = sp->sampler[unit]; + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; if (!texture) return; diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index 40d8eb2c2a..3c5beb560f 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -39,8 +39,9 @@ struct sp_shader_sampler { struct tgsi_sampler base; /**< base class */ - uint unit; - struct softpipe_context *sp; + const struct pipe_texture *texture; + const struct pipe_sampler_state *sampler; + struct softpipe_tile_cache *cache; }; diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 1f9b8f1f4f..306284808c 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -236,8 +236,7 @@ sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc) * Specify the texture to cache. */ void -sp_tile_cache_set_texture(struct pipe_context *pipe, - struct softpipe_tile_cache *tc, +sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, struct pipe_texture *texture) { uint i; @@ -344,8 +343,7 @@ clear_tile(struct softpipe_cached_tile *tile, * Actually clear the tiles which were flagged as being in a clear state. */ static void -sp_tile_cache_flush_clear(struct pipe_context *pipe, - struct softpipe_tile_cache *tc) +sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc) { struct pipe_transfer *pt = tc->transfer; const uint w = tc->transfer->width; @@ -382,8 +380,7 @@ sp_tile_cache_flush_clear(struct pipe_context *pipe, * any tiles "flagged" as cleared will be "really" cleared. */ void -sp_flush_tile_cache(struct softpipe_context *softpipe, - struct softpipe_tile_cache *tc) +sp_flush_tile_cache(struct softpipe_tile_cache *tc) { struct pipe_transfer *pt = tc->transfer; int inuse = 0, pos; @@ -409,7 +406,7 @@ sp_flush_tile_cache(struct softpipe_context *softpipe, } #if TILE_CLEAR_OPTIMIZATION - sp_tile_cache_flush_clear(&softpipe->pipe, tc); + sp_tile_cache_flush_clear(tc); #endif } else if (tc->texture) { @@ -431,8 +428,7 @@ sp_flush_tile_cache(struct softpipe_context *softpipe, * \param x, y position of tile, in pixels */ struct softpipe_cached_tile * -sp_get_cached_tile(struct softpipe_context *softpipe, - struct softpipe_tile_cache *tc, int x, int y) +sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) { struct pipe_transfer *pt = tc->transfer; @@ -513,11 +509,11 @@ tex_cache_pos(int x, int y, int z, int face, int level) * Tiles are read-only and indexed with more params. */ const struct softpipe_cached_tile * -sp_get_cached_tile_tex(struct softpipe_context *sp, - struct softpipe_tile_cache *tc, int x, int y, int z, +sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, + int x, int y, int z, int face, int level) { - struct pipe_screen *screen = sp->pipe.screen; + struct pipe_screen *screen = tc->screen; /* tile pos in framebuffer: */ const int tile_x = x & ~(TILE_SIZE - 1); const int tile_y = y & ~(TILE_SIZE - 1); diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index 8f247d0e58..639cde6705 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -80,25 +80,21 @@ extern void sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc); extern void -sp_tile_cache_set_texture(struct pipe_context *pipe, - struct softpipe_tile_cache *tc, +sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, struct pipe_texture *texture); extern void -sp_flush_tile_cache(struct softpipe_context *softpipe, - struct softpipe_tile_cache *tc); +sp_flush_tile_cache(struct softpipe_tile_cache *tc); extern void sp_tile_cache_clear(struct softpipe_tile_cache *tc, const float *rgba, uint clearValue); extern struct softpipe_cached_tile * -sp_get_cached_tile(struct softpipe_context *softpipe, - struct softpipe_tile_cache *tc, int x, int y); +sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y); extern const struct softpipe_cached_tile * -sp_get_cached_tile_tex(struct softpipe_context *softpipe, - struct softpipe_tile_cache *tc, int x, int y, int z, +sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, int x, int y, int z, int face, int level); -- cgit v1.2.3 From 0ac879dca797360570543d5bd0fd64f8fb8e566e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 16 Jul 2009 17:51:02 +0100 Subject: util: _debug_printf should print even when DEBUG is not defined The leading underscore is meaningful... This function is used by _warning and _error functions as well as the more common debug_printf(). debug_printf (without underscore) gets turned off when DEBUG is disabled, but warning/error messages still use this function to get their message out. --- src/gallium/auxiliary/util/u_debug.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c index a5ca0b72bd..96d400c839 100644 --- a/src/gallium/auxiliary/util/u_debug.c +++ b/src/gallium/auxiliary/util/u_debug.c @@ -143,11 +143,9 @@ void _debug_vprintf(const char *format, va_list ap) #elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT) /* TODO */ #else /* !PIPE_SUBSYSTEM_WINDOWS */ -#ifdef DEBUG fflush(stdout); vfprintf(stderr, format, ap); #endif -#endif } -- cgit v1.2.3 From 07bb026900a6c01226217ceee1d4d1426c040d6e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 16 Jul 2009 17:57:00 +0100 Subject: gallium/xlib: use XSHM for swapbuffers Makes some difference, but suprisingly little. Barely worth the effort. --- src/gallium/winsys/xlib/xlib_softpipe.c | 98 ++++++++++++++++++--------------- 1 file changed, 53 insertions(+), 45 deletions(-) diff --git a/src/gallium/winsys/xlib/xlib_softpipe.c b/src/gallium/winsys/xlib/xlib_softpipe.c index 44b8464518..1de93c4e8c 100644 --- a/src/gallium/winsys/xlib/xlib_softpipe.c +++ b/src/gallium/winsys/xlib/xlib_softpipe.c @@ -75,9 +75,6 @@ struct xmesa_pipe_winsys { struct pipe_winsys base; /* struct xmesa_visual *xm_visual; */ -#ifdef USE_XSHM - int shm; -#endif }; @@ -93,11 +90,6 @@ xm_buffer( struct pipe_buffer *buf ) /** * X Shared Memory Image extension code */ -#ifdef USE_XSHM -#define XSHM_ENABLED(b) ((b)->shm) -#else -#define XSHM_ENABLED(b) 0 -#endif #ifdef USE_XSHM @@ -116,23 +108,23 @@ mesaHandleXError(Display *dpy, XErrorEvent *event) } -static GLboolean alloc_shm(struct xm_buffer *buf, unsigned size) +static char *alloc_shm(struct xm_buffer *buf, unsigned size) { XShmSegmentInfo *const shminfo = & buf->shminfo; shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777); if (shminfo->shmid < 0) { - return GL_FALSE; + return NULL; } shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0); if (shminfo->shmaddr == (char *) -1) { shmctl(shminfo->shmid, IPC_RMID, 0); - return GL_FALSE; + return NULL; } shminfo->readOnly = False; - return GL_TRUE; + return shminfo->shmaddr; } @@ -258,25 +250,30 @@ xlib_softpipe_display_surface(struct xmesa_buffer *b, return; #ifdef USE_XSHM - if (XSHM_ENABLED(xm_buf) && (xm_buf->tempImage == NULL)) { - assert(surf->texture->block.width == 1); - assert(surf->texture->block.height == 1); - alloc_shm_ximage(xm_buf, b, spt->stride[surf->level] / - surf->texture->block.size, surf->height); - } -#endif + if (xm_buf->shm) + { + if (xm_buf->tempImage == NULL) + { + assert(surf->texture->block.width == 1); + assert(surf->texture->block.height == 1); + alloc_shm_ximage(xm_buf, b, spt->stride[surf->level] / + surf->texture->block.size, surf->height); + } - ximage = (XSHM_ENABLED(xm_buf)) ? xm_buf->tempImage : b->tempImage; - ximage->data = xm_buf->data; + ximage = xm_buf->tempImage; + ximage->data = xm_buf->data; - /* display image in Window */ -#ifdef USE_XSHM - if (XSHM_ENABLED(xm_buf)) { + /* _debug_printf("XSHM\n"); */ XShmPutImage(b->xm_visual->display, b->drawable, b->gc, ximage, 0, 0, 0, 0, surf->width, surf->height, False); - } else + } + else #endif { + /* display image in Window */ + ximage = b->tempImage; + ximage->data = xm_buf->data; + /* check that the XImage has been previously initialized */ assert(ximage->format); assert(ximage->bitmap_unit); @@ -286,6 +283,7 @@ xlib_softpipe_display_surface(struct xmesa_buffer *b, ximage->height = surf->height; ximage->bytes_per_line = spt->stride[surf->level]; + /* _debug_printf("XPUT\n"); */ XPutImage(b->xm_visual->display, b->drawable, b->gc, ximage, 0, 0, 0, 0, surf->width, surf->height); } @@ -321,21 +319,6 @@ xm_buffer_create(struct pipe_winsys *pws, unsigned size) { struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer); -#ifdef USE_XSHM - struct xmesa_pipe_winsys *xpws = (struct xmesa_pipe_winsys *) pws; - - buffer->shminfo.shmid = -1; - buffer->shminfo.shmaddr = (char *) -1; - - if (xpws->shm && (usage & PIPE_BUFFER_USAGE_PIXEL) != 0) { - buffer->shm = xpws->shm; - - if (alloc_shm(buffer, size)) { - buffer->data = buffer->shminfo.shmaddr; - buffer->shm = 1; - } - } -#endif pipe_reference_init(&buffer->base.reference, 1); buffer->base.alignment = alignment; @@ -362,9 +345,6 @@ xm_user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes) buffer->base.size = bytes; buffer->userBuffer = TRUE; buffer->data = ptr; -#ifdef USE_XSHM - buffer->shm = 0; -#endif return &buffer->base; } @@ -379,16 +359,44 @@ xm_surface_buffer_create(struct pipe_winsys *winsys, { const unsigned alignment = 64; struct pipe_format_block block; - unsigned nblocksx, nblocksy; + unsigned nblocksx, nblocksy, size; pf_get_block(format, &block); nblocksx = pf_get_nblocksx(&block, width); nblocksy = pf_get_nblocksy(&block, height); *stride = align(nblocksx * block.size, alignment); + size = *stride * nblocksy; + +#ifdef USE_XSHM + if (!debug_get_bool_option("XLIB_NO_SHM", FALSE)) + { + struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer); + + pipe_reference_init(&buffer->base.reference, 1); + buffer->base.alignment = alignment; + buffer->base.usage = usage; + buffer->base.size = size; + buffer->userBuffer = FALSE; + buffer->shminfo.shmid = -1; + buffer->shminfo.shmaddr = (char *) -1; + buffer->shm = TRUE; + + buffer->data = alloc_shm(buffer, size); + if (!buffer->data) + goto out; + + return &buffer->base; + + out: + if (buffer) + FREE(buffer); + } +#endif + return winsys->buffer_create(winsys, alignment, usage, - *stride * nblocksy); + size); } -- cgit v1.2.3 From b5d583efeff5f195bff48c95125a225c273189e2 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 17 Jul 2009 10:44:22 +0100 Subject: softpipe: make some small steps to flush texture cache less frequently No performance gain yet, but the code is a bit cleaner. --- src/gallium/drivers/softpipe/sp_context.h | 3 +- src/gallium/drivers/softpipe/sp_state_derived.c | 14 +++++ src/gallium/drivers/softpipe/sp_texture.c | 3 +- src/gallium/drivers/softpipe/sp_texture.h | 2 +- src/gallium/drivers/softpipe/sp_tile_cache.c | 80 +++++++++++++++---------- src/gallium/drivers/softpipe/sp_tile_cache.h | 3 + 6 files changed, 70 insertions(+), 35 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index 7888c2f644..f66f0b7849 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -149,7 +149,8 @@ struct softpipe_context { struct softpipe_tile_cache *cbuf_cache[PIPE_MAX_COLOR_BUFS]; struct softpipe_tile_cache *zsbuf_cache; - + + unsigned tex_timestamp; struct softpipe_tile_cache *tex_cache[PIPE_MAX_SAMPLERS]; unsigned use_sse : 1; diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 75be99768c..629a1f8e29 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -32,6 +32,7 @@ #include "draw/draw_vertex.h" #include "draw/draw_private.h" #include "sp_context.h" +#include "sp_screen.h" #include "sp_state.h" @@ -200,6 +201,10 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i]; softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i]; } + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + sp_tile_cache_validate_texture( softpipe->tex_cache[i] ); + } } /* Hopefully this will remain quite simple, otherwise need to pull in @@ -207,6 +212,15 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) */ void softpipe_update_derived( struct softpipe_context *softpipe ) { + struct softpipe_screen *sp_screen = softpipe_screen(softpipe->pipe.screen); + + /* Check for updated textures. + */ + if (softpipe->tex_timestamp != sp_screen->timestamp) { + softpipe->tex_timestamp = sp_screen->timestamp; + softpipe->dirty |= SP_NEW_TEXTURE; + } + if (softpipe->dirty & (SP_NEW_SAMPLER | SP_NEW_TEXTURE)) update_tgsi_samplers( softpipe ); diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c index 7a533dad9f..0c84375bf1 100644 --- a/src/gallium/drivers/softpipe/sp_texture.c +++ b/src/gallium/drivers/softpipe/sp_texture.c @@ -227,7 +227,8 @@ softpipe_get_tex_surface(struct pipe_screen *screen, if (ps->usage & (PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_GPU_WRITE)) { /* Mark the surface as dirty. The tile cache will look for this. */ - spt->modified = TRUE; + spt->timestamp++; + softpipe_screen(screen)->timestamp++; } ps->face = face; diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h index 893aa7d11d..42df722a2d 100644 --- a/src/gallium/drivers/softpipe/sp_texture.h +++ b/src/gallium/drivers/softpipe/sp_texture.h @@ -48,7 +48,7 @@ struct softpipe_texture */ struct pipe_buffer *buffer; - boolean modified; + unsigned timestamp; }; struct softpipe_transfer diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 306284808c..79b1e036be 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -54,7 +54,10 @@ struct softpipe_tile_cache struct pipe_surface *surface; /**< the surface we're caching */ struct pipe_transfer *transfer; void *transfer_map; + struct pipe_texture *texture; /**< if caching a texture */ + unsigned timestamp; + struct softpipe_cached_tile entries[NUM_ENTRIES]; uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32]; float clear_color[4]; /**< for color bufs */ @@ -231,6 +234,23 @@ sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc) } } +void +sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc) +{ + if (tc->texture) { + struct softpipe_texture *spt = softpipe_texture(tc->texture); + if (spt->timestamp != tc->timestamp) { + /* texture was modified, invalidate all cached tiles */ + uint i; + _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].x = -3; + } + + tc->timestamp = spt->timestamp; + } + } +} /** * Specify the texture to cache. @@ -243,27 +263,29 @@ sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, assert(!tc->transfer); - pipe_texture_reference(&tc->texture, texture); + if (tc->texture != texture) { + pipe_texture_reference(&tc->texture, texture); - if (tc->tex_trans) { - struct pipe_screen *screen = tc->tex_trans->texture->screen; + if (tc->tex_trans) { + struct pipe_screen *screen = tc->tex_trans->texture->screen; + + if (tc->tex_trans_map) { + screen->transfer_unmap(screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } - if (tc->tex_trans_map) { - screen->transfer_unmap(screen, tc->tex_trans); - tc->tex_trans_map = NULL; + screen->tex_transfer_destroy(tc->tex_trans); + tc->tex_trans = NULL; } - screen->tex_transfer_destroy(tc->tex_trans); - tc->tex_trans = NULL; - } + /* mark as entries as invalid/empty */ + /* XXX we should try to avoid this when the teximage hasn't changed */ + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].x = -1; + } - /* mark as entries as invalid/empty */ - /* XXX we should try to avoid this when the teximage hasn't changed */ - for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].x = -1; + tc->tex_face = -1; /* any invalid value here */ } - - tc->tex_face = -1; /* any invalid value here */ } @@ -443,7 +465,7 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) if (tile_x != tile->x || tile_y != tile->y) { - if (tile->x != -1) { + if (tile->x >= 0) { /* put dirty tile back in framebuffer */ if (tc->depth_stencil) { pipe_put_tile_raw(pt, @@ -522,30 +544,24 @@ sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, face, level); struct softpipe_cached_tile *tile = tc->entries + pos; - if (tc->texture) { - struct softpipe_texture *spt = softpipe_texture(tc->texture); - if (spt->modified) { - /* texture was modified, invalidate all cached tiles */ - uint p; - for (p = 0; p < NUM_ENTRIES; p++) { - tile = tc->entries + p; - tile->x = -1; - } - spt->modified = FALSE; - } - } - if (tile_x != tile->x || tile_y != tile->y || z != tile->z || face != tile->face || level != tile->level) { - /* cache miss */ + /* cache miss. Most misses are because we've invaldiated the + * texture cache previously -- most commonly on binding a new + * texture. Currently we effectively flush the cache on texture + * bind. + */ #if 0 - printf("miss at %u x=%d y=%d z=%d face=%d level=%d\n", pos, - x/TILE_SIZE, y/TILE_SIZE, z, face, level); + _debug_printf("miss at %u: x=%d y=%d z=%d face=%d level=%d\n" + " tile %u: x=%d y=%d z=%d face=%d level=%d\n", + pos, x/TILE_SIZE, y/TILE_SIZE, z, face, level, + pos, tile->x, tile->y, tile->z, tile->face, tile->level); #endif + /* check if we need to get a new transfer */ if (!tc->tex_trans || tc->tex_face != face || diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index 639cde6705..0d165b4ad7 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -83,6 +83,9 @@ extern void sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, struct pipe_texture *texture); +void +sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc); + extern void sp_flush_tile_cache(struct softpipe_tile_cache *tc); -- cgit v1.2.3 From 73e7356385a703c214b35fbb29aaf3108764f033 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 17 Jul 2009 10:47:32 +0100 Subject: softpipe: simplify flush_spans No loss of performance, but simpler code. --- src/gallium/drivers/softpipe/sp_setup.c | 72 +++++++++------------------------ 1 file changed, 19 insertions(+), 53 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index de3ae3c369..f29fcbd5a2 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -428,65 +428,31 @@ static void flush_spans( struct setup_context *setup ) int minleft, maxright; int x; - switch (setup->span.y_flags) { - case 0x3: - /* both odd and even lines written (both quad rows) */ - minleft = block(MIN2(xleft0, xleft1)); - maxright = block(MAX2(xright0, xright1)); - for (x = minleft; x <= maxright; x += 2) { - /* determine which of the four pixels is inside the span bounds */ - uint mask = 0x0; - if (x >= xleft0 && x < xright0) - mask |= MASK_TOP_LEFT; - if (x >= xleft1 && x < xright1) - mask |= MASK_BOTTOM_LEFT; - if (x+1 >= xleft0 && x+1 < xright0) - mask |= MASK_TOP_RIGHT; - if (x+1 >= xleft1 && x+1 < xright1) - mask |= MASK_BOTTOM_RIGHT; - if (mask) - EMIT_QUAD( setup, x, setup->span.y, mask ); - } - break; - - case 0x1: - /* only even line written (quad top row) */ - minleft = block(xleft0); - maxright = block(xright0); - for (x = minleft; x <= maxright; x += 2) { - uint mask = 0x0; - if (x >= xleft0 && x < xright0) - mask |= MASK_TOP_LEFT; - if (x+1 >= xleft0 && x+1 < xright0) - mask |= MASK_TOP_RIGHT; - if (mask) - EMIT_QUAD( setup, x, setup->span.y, mask ); - } - break; - - case 0x2: - /* only odd line written (quad bottom row) */ - minleft = block(xleft1); - maxright = block(xright1); - for (x = minleft; x <= maxright; x += 2) { - uint mask = 0x0; - if (x >= xleft1 && x < xright1) - mask |= MASK_BOTTOM_LEFT; - if (x+1 >= xleft1 && x+1 < xright1) - mask |= MASK_BOTTOM_RIGHT; - if (mask) - EMIT_QUAD( setup, x, setup->span.y, mask ); - } - break; - - default: - return; + minleft = block(MIN2(xleft0, xleft1)); + maxright = block(MAX2(xright0, xright1)); + + for (x = minleft; x <= maxright; x += 2) { + /* determine which of the four pixels is inside the span bounds */ + uint mask = 0x0; + if (x >= xleft0 && x < xright0) + mask |= MASK_TOP_LEFT; + if (x >= xleft1 && x < xright1) + mask |= MASK_BOTTOM_LEFT; + if (x+1 >= xleft0 && x+1 < xright0) + mask |= MASK_TOP_RIGHT; + if (x+1 >= xleft1 && x+1 < xright1) + mask |= MASK_BOTTOM_RIGHT; + if (mask) + EMIT_QUAD( setup, x, setup->span.y, mask ); } + setup->span.y = 0; setup->span.y_flags = 0; setup->span.right[0] = 0; setup->span.right[1] = 0; + setup->span.left[0] = 1; /* greater than right[0] */ + setup->span.left[1] = 1; /* greater than right[1] */ } -- cgit v1.2.3 From 0ed99f45529178c77e47838f226231ea1bc9b918 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 17 Jul 2009 12:03:51 +0100 Subject: softpipe: use bitwise logic to setup quad masks in sp_setup --- src/gallium/drivers/softpipe/sp_setup.c | 65 ++++++++++++++++++++------------- 1 file changed, 40 insertions(+), 25 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index f29fcbd5a2..8ac22bd302 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -388,19 +388,19 @@ emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job ) emit_quad( setup, &quad, thread ); } -#define EMIT_QUAD(setup,x,y,mask) do {\ +#define EMIT_QUAD(setup,x,y,qmask) do {\ setup->quad.input.x0 = x;\ setup->quad.input.y0 = y;\ - setup->quad.inout.mask = mask;\ + setup->quad.inout.mask = qmask;\ add_quad_job( &setup->que, &setup->quad, emit_quad_job );\ } while (0) #else -#define EMIT_QUAD(setup,x,y,mask) do {\ +#define EMIT_QUAD(setup,x,y,qmask) do {\ setup->quad.input.x0 = x;\ setup->quad.input.y0 = y;\ - setup->quad.inout.mask = mask;\ + setup->quad.inout.mask = qmask;\ emit_quad( setup, &setup->quad, 0 );\ } while (0) @@ -421,29 +421,42 @@ static INLINE int block( int x ) */ static void flush_spans( struct setup_context *setup ) { + const int step = 30; const int xleft0 = setup->span.left[0]; const int xleft1 = setup->span.left[1]; const int xright0 = setup->span.right[0]; const int xright1 = setup->span.right[1]; - int minleft, maxright; + + int minleft = block(MIN2(xleft0, xleft1)); + int maxright = MAX2(xright0, xright1); int x; - minleft = block(MIN2(xleft0, xleft1)); - maxright = block(MAX2(xright0, xright1)); - - for (x = minleft; x <= maxright; x += 2) { - /* determine which of the four pixels is inside the span bounds */ - uint mask = 0x0; - if (x >= xleft0 && x < xright0) - mask |= MASK_TOP_LEFT; - if (x >= xleft1 && x < xright1) - mask |= MASK_BOTTOM_LEFT; - if (x+1 >= xleft0 && x+1 < xright0) - mask |= MASK_TOP_RIGHT; - if (x+1 >= xleft1 && x+1 < xright1) - mask |= MASK_BOTTOM_RIGHT; - if (mask) - EMIT_QUAD( setup, x, setup->span.y, mask ); + for (x = minleft; x < maxright; x += step) { + unsigned skip_left0 = CLAMP(xleft0 - x, 0, step); + unsigned skip_left1 = CLAMP(xleft1 - x, 0, step); + unsigned skip_right0 = CLAMP(x + step - xright0, 0, step); + unsigned skip_right1 = CLAMP(x + step - xright1, 0, step); + unsigned lx = x; + + unsigned skipmask_left0 = (1U << skip_left0) - 1U; + unsigned skipmask_left1 = (1U << skip_left1) - 1U; + + /* These calculations fail when step == 32 and skip_right == 0. + */ + unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0); + unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1); + + unsigned mask0 = ~skipmask_left0 & ~skipmask_right0; + unsigned mask1 = ~skipmask_left1 & ~skipmask_right1; + + while (mask0 | mask1) { + unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2); + if (quadmask) + EMIT_QUAD( setup, lx, setup->span.y, quadmask ); + mask0 >>= 2; + mask1 >>= 2; + lx += 2; + } } @@ -451,8 +464,8 @@ static void flush_spans( struct setup_context *setup ) setup->span.y_flags = 0; setup->span.right[0] = 0; setup->span.right[1] = 0; - setup->span.left[0] = 1; /* greater than right[0] */ - setup->span.left[1] = 1; /* greater than right[1] */ + setup->span.left[0] = 1000000; /* greater than right[0] */ + setup->span.left[1] = 1000000; /* greater than right[1] */ } @@ -810,11 +823,10 @@ static void subtriangle( struct setup_context *setup, /* clip top/bottom */ start_y = sy; - finish_y = sy + lines; - if (start_y < miny) start_y = miny; + finish_y = sy + lines; if (finish_y > maxy) finish_y = maxy; @@ -1495,6 +1507,9 @@ struct setup_context *setup_create_context( struct softpipe_context *softpipe ) setup->quad.coef = setup->coef; setup->quad.posCoef = &setup->posCoef; + setup->span.left[0] = 1000000; /* greater than right[0] */ + setup->span.left[1] = 1000000; /* greater than right[1] */ + #if SP_NUM_QUAD_THREADS > 1 setup->que.first = 0; setup->que.last = 0; -- cgit v1.2.3 From 13e2d35764e0c8de3356ee663885568fc00424f0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 17 Jul 2009 12:12:04 +0100 Subject: softpipe: remove unused vars in sp_setup.c --- src/gallium/drivers/softpipe/sp_setup.c | 5 ----- 1 file changed, 5 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index 8ac22bd302..d05c9ad57c 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -178,8 +178,6 @@ struct setup_context { int left[2]; /**< [0] = row0, [1] = row1 */ int right[2]; int y; - unsigned y_flags; - unsigned mask; /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */ } span; #if DEBUG_FRAGS @@ -461,7 +459,6 @@ static void flush_spans( struct setup_context *setup ) setup->span.y = 0; - setup->span.y_flags = 0; setup->span.right[0] = 0; setup->span.right[1] = 0; setup->span.left[0] = 1000000; /* greater than right[0] */ @@ -863,7 +860,6 @@ static void subtriangle( struct setup_context *setup, setup->span.left[_y&1] = left; setup->span.right[_y&1] = right; - setup->span.y_flags |= 1<<(_y&1); } } @@ -939,7 +935,6 @@ void setup_tri( struct setup_context *setup, setup->quad.input.prim = QUAD_PRIM_TRI; setup->span.y = 0; - setup->span.y_flags = 0; setup->span.right[0] = 0; setup->span.right[1] = 0; /* setup->span.z_mode = tri_z_mode( setup->ctx ); */ -- cgit v1.2.3 From f911c3b9897b90132c8621a72bfeb824eb3b01e5 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 22 Jul 2009 15:08:42 +0100 Subject: softpipe: shortcircuit repeated lookups of the same tile The sp_tile_cache is often called repeatedly to look up the same tile. Add a cache (to the cache) of the single tile most recently retreived and make a quick inline check to see if this matches the subsequent request. Add a tile_address bitfield struct to make this check easier. --- src/gallium/auxiliary/util/u_math.h | 12 +++ src/gallium/drivers/softpipe/sp_tex_sample.c | 12 ++- src/gallium/drivers/softpipe/sp_tile_cache.c | 153 ++++++++++++--------------- src/gallium/drivers/softpipe/sp_tile_cache.h | 88 ++++++++++++++- 4 files changed, 168 insertions(+), 97 deletions(-) diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h index e5003af01d..d30fa3c2d5 100644 --- a/src/gallium/auxiliary/util/u_math.h +++ b/src/gallium/auxiliary/util/u_math.h @@ -366,6 +366,18 @@ unsigned ffs( unsigned u ) #endif +/* Could also binary search for the highest bit. + */ +static INLINE unsigned +util_unsigned_logbase2(unsigned n) +{ + unsigned log2 = 0; + while (n >>= 1) + ++log2; + return log2; +} + + /** * Return float bits. */ diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 3daa88eedd..46c56b0c83 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -680,11 +680,13 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, rgba[3][j] = sampler->border_color[3]; } else { - const int tx = x % TILE_SIZE; - const int ty = y % TILE_SIZE; - const struct softpipe_cached_tile *tile - = sp_get_cached_tile_tex(samp->cache, - x, y, z, face, level); + const unsigned tx = x % TILE_SIZE; + const unsigned ty = y % TILE_SIZE; + const struct softpipe_cached_tile *tile; + + tile = sp_get_cached_tile_tex(samp->cache, + tile_address(x, y, z, face, level)); + rgba[0][j] = tile->data.color[ty][tx][0]; rgba[1][j] = tile->data.color[ty][tx][1]; rgba[2][j] = tile->data.color[ty][tx][2]; diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 79b1e036be..06d9c6a80a 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -40,36 +40,6 @@ #include "sp_texture.h" #include "sp_tile_cache.h" -#define NUM_ENTRIES 50 - - -/** XXX move these */ -#define MAX_WIDTH 2048 -#define MAX_HEIGHT 2048 - - -struct softpipe_tile_cache -{ - struct pipe_screen *screen; - struct pipe_surface *surface; /**< the surface we're caching */ - struct pipe_transfer *transfer; - void *transfer_map; - - struct pipe_texture *texture; /**< if caching a texture */ - unsigned timestamp; - - struct softpipe_cached_tile entries[NUM_ENTRIES]; - uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32]; - float clear_color[4]; /**< for color bufs */ - uint clear_val; /**< for z+stencil, or packed color clear value */ - boolean depth_stencil; /**< Is the surface a depth/stencil format? */ - - struct pipe_transfer *tex_trans; - void *tex_trans_map; - int tex_face, tex_level, tex_z; - - struct softpipe_cached_tile tile; /**< scratch tile for clears */ -}; /** @@ -124,9 +94,9 @@ sp_create_tile_cache( struct pipe_screen *screen ) if (tc) { tc->screen = screen; for (pos = 0; pos < NUM_ENTRIES; pos++) { - tc->entries[pos].x = - tc->entries[pos].y = -1; + tc->entries[pos].addr.bits.invalid = 1; } + tc->last_tile = &tc->entries[0]; /* any tile */ } return tc; } @@ -244,7 +214,7 @@ sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc) uint i; _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].x = -3; + tc->entries[i].addr.bits.invalid = 1; } tc->timestamp = spt->timestamp; @@ -281,7 +251,7 @@ sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, /* mark as entries as invalid/empty */ /* XXX we should try to avoid this when the teximage hasn't changed */ for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].x = -1; + tc->entries[i].addr.bits.invalid = 1; } tc->tex_face = -1; /* any invalid value here */ @@ -411,18 +381,22 @@ sp_flush_tile_cache(struct softpipe_tile_cache *tc) /* caching a drawing transfer */ for (pos = 0; pos < NUM_ENTRIES; pos++) { struct softpipe_cached_tile *tile = tc->entries + pos; - if (tile->x >= 0) { + if (!tile->addr.bits.invalid) { if (tc->depth_stencil) { pipe_put_tile_raw(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, tile->data.depth32, 0/*STRIDE*/); } else { pipe_put_tile_rgba(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, (float *) tile->data.color); } - tile->x = tile->y = -1; /* mark as empty */ + tile->addr.bits.invalid = 1; /* mark as empty */ inuse++; } } @@ -434,7 +408,7 @@ sp_flush_tile_cache(struct softpipe_tile_cache *tc) else if (tc->texture) { /* caching a texture, mark all entries as empty */ for (pos = 0; pos < NUM_ENTRIES; pos++) { - tc->entries[pos].x = -1; + tc->entries[pos].addr.bits.invalid = 1; } tc->tex_face = -1; } @@ -453,34 +427,34 @@ struct softpipe_cached_tile * sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) { struct pipe_transfer *pt = tc->transfer; - + /* tile pos in framebuffer: */ - const int tile_x = x & ~(TILE_SIZE - 1); - const int tile_y = y & ~(TILE_SIZE - 1); - + union tile_address addr = tile_address( x, y, 0, 0, 0 ); /* cache pos/entry: */ const int pos = CACHE_POS(x, y); struct softpipe_cached_tile *tile = tc->entries + pos; - if (tile_x != tile->x || - tile_y != tile->y) { + if (addr.value != tile->addr.value) { - if (tile->x >= 0) { + if (tile->addr.bits.invalid == 0) { /* put dirty tile back in framebuffer */ if (tc->depth_stencil) { pipe_put_tile_raw(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, tile->data.depth32, 0/*STRIDE*/); } else { pipe_put_tile_rgba(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, (float *) tile->data.color); } } - tile->x = tile_x; - tile->y = tile_y; + tile->addr = addr; if (is_clear_flag_set(tc->clear_flags, x, y)) { /* don't get tile from framebuffer, just clear it */ @@ -496,12 +470,16 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) /* get new tile data from transfer */ if (tc->depth_stencil) { pipe_get_tile_raw(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, tile->data.depth32, 0/*STRIDE*/); } else { pipe_get_tile_rgba(pt, - tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->addr.bits.x * TILE_SIZE, + tile->addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, (float *) tile->data.color); } } @@ -519,36 +497,31 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) * XXX There's probably lots of ways in which we can improve this. */ static INLINE uint -tex_cache_pos(int x, int y, int z, int face, int level) +tex_cache_pos( union tile_address addr ) { - uint entry = x + y * 9 + z * 3 + face + level * 7; + uint entry = (addr.bits.x + + addr.bits.y * 9 + + addr.bits.z * 3 + + addr.bits.face + + addr.bits.level * 7); + return entry % NUM_ENTRIES; } - /** * Similar to sp_get_cached_tile() but for textures. * Tiles are read-only and indexed with more params. */ const struct softpipe_cached_tile * -sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, - int x, int y, int z, - int face, int level) +sp_find_cached_tile_tex(struct softpipe_tile_cache *tc, + union tile_address addr ) { struct pipe_screen *screen = tc->screen; - /* tile pos in framebuffer: */ - const int tile_x = x & ~(TILE_SIZE - 1); - const int tile_y = y & ~(TILE_SIZE - 1); - /* cache pos/entry: */ - const uint pos = tex_cache_pos(x / TILE_SIZE, y / TILE_SIZE, z, - face, level); - struct softpipe_cached_tile *tile = tc->entries + pos; + struct softpipe_cached_tile *tile; + + tile = tc->entries + tex_cache_pos( addr ); - if (tile_x != tile->x || - tile_y != tile->y || - z != tile->z || - face != tile->face || - level != tile->level) { + if (addr.value != tile->addr.value) { /* cache miss. Most misses are because we've invaldiated the * texture cache previously -- most commonly on binding a new @@ -559,14 +532,14 @@ sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, _debug_printf("miss at %u: x=%d y=%d z=%d face=%d level=%d\n" " tile %u: x=%d y=%d z=%d face=%d level=%d\n", pos, x/TILE_SIZE, y/TILE_SIZE, z, face, level, - pos, tile->x, tile->y, tile->z, tile->face, tile->level); + pos, tile->addr.bits.x, tile->addr.bits.y, tile->z, tile->face, tile->level); #endif /* check if we need to get a new transfer */ if (!tc->tex_trans || - tc->tex_face != face || - tc->tex_level != level || - tc->tex_z != z) { + tc->tex_face != addr.bits.face || + tc->tex_level != addr.bits.level || + tc->tex_z != addr.bits.z) { /* get new transfer (view into texture) */ if (tc->tex_trans) { @@ -579,28 +552,32 @@ sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, tc->tex_trans = NULL; } - tc->tex_trans = screen->get_tex_transfer(screen, tc->texture, face, level, z, - PIPE_TRANSFER_READ, 0, 0, - tc->texture->width[level], - tc->texture->height[level]); + tc->tex_trans = + screen->get_tex_transfer(screen, tc->texture, + addr.bits.face, + addr.bits.level, + addr.bits.z, + PIPE_TRANSFER_READ, 0, 0, + tc->texture->width[addr.bits.level], + tc->texture->height[addr.bits.level]); + tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans); - tc->tex_face = face; - tc->tex_level = level; - tc->tex_z = z; + tc->tex_face = addr.bits.face; + tc->tex_level = addr.bits.level; + tc->tex_z = addr.bits.z; } /* get tile from the transfer (view into texture) */ pipe_get_tile_rgba(tc->tex_trans, - tile_x, tile_y, TILE_SIZE, TILE_SIZE, + addr.bits.x * TILE_SIZE, + addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, (float *) tile->data.color); - tile->x = tile_x; - tile->y = tile_y; - tile->z = z; - tile->face = face; - tile->level = level; + tile->addr = addr; } + tc->last_tile = tile; return tile; } @@ -633,6 +610,6 @@ sp_tile_cache_clear(struct softpipe_tile_cache *tc, const float *rgba, for (pos = 0; pos < NUM_ENTRIES; pos++) { struct softpipe_cached_tile *tile = tc->entries + pos; - tile->x = tile->y = -1; + tile->addr.bits.invalid = 1; } } diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index 0d165b4ad7..3017fcbebc 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -44,11 +44,25 @@ struct softpipe_tile_cache; #define TILE_SIZE 64 +/* If we need to support > 4096, just expand this to be a 64 bit + * union, or consider tiling in Z as well. + */ +union tile_address { + struct { + unsigned x:6; /* 4096 / TILE_SIZE */ + unsigned y:6; /* 4096 / TILE_SIZE */ + unsigned z:12; /* 4096 -- z not tiled */ + unsigned face:3; + unsigned level:4; + unsigned invalid:1; + } bits; + unsigned value; +}; + struct softpipe_cached_tile { - int x, y; /**< pos of tile in window coords */ - int z, face, level; /**< Extra texture indexes */ + union tile_address addr; union { float color[TILE_SIZE][TILE_SIZE][4]; uint color32[TILE_SIZE][TILE_SIZE]; @@ -59,6 +73,39 @@ struct softpipe_cached_tile } data; }; +#define NUM_ENTRIES 50 + + +/** XXX move these */ +#define MAX_WIDTH 2048 +#define MAX_HEIGHT 2048 + + +struct softpipe_tile_cache +{ + struct pipe_screen *screen; + struct pipe_surface *surface; /**< the surface we're caching */ + struct pipe_transfer *transfer; + void *transfer_map; + + struct pipe_texture *texture; /**< if caching a texture */ + unsigned timestamp; + + struct softpipe_cached_tile entries[NUM_ENTRIES]; + uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32]; + float clear_color[4]; /**< for color bufs */ + uint clear_val; /**< for z+stencil, or packed color clear value */ + boolean depth_stencil; /**< Is the surface a depth/stencil format? */ + + struct pipe_transfer *tex_trans; + void *tex_trans_map; + int tex_face, tex_level, tex_z; + + struct softpipe_cached_tile tile; /**< scratch tile for clears */ + + struct softpipe_cached_tile *last_tile; /**< most recently retrieved tile */ +}; + extern struct softpipe_tile_cache * sp_create_tile_cache( struct pipe_screen *screen ); @@ -97,8 +144,41 @@ extern struct softpipe_cached_tile * sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y); extern const struct softpipe_cached_tile * -sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, int x, int y, int z, - int face, int level); +sp_find_cached_tile_tex(struct softpipe_tile_cache *tc, + union tile_address addr ); + +static INLINE const union tile_address +tile_address( unsigned x, + unsigned y, + unsigned z, + unsigned face, + unsigned level ) +{ + union tile_address addr; + + addr.value = 0; + addr.bits.x = x / TILE_SIZE; + addr.bits.y = y / TILE_SIZE; + addr.bits.z = z; + addr.bits.face = face; + addr.bits.level = level; + + return addr; +} + +/* Quickly retrieve tile if it matches last lookup. + */ +static INLINE const struct softpipe_cached_tile * +sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, + union tile_address addr ) +{ + if (tc->last_tile->addr.value == addr.value) + return tc->last_tile; + + return sp_find_cached_tile_tex( tc, addr ); +} + + #endif /* SP_TILE_CACHE_H */ -- cgit v1.2.3 From 19097907ef042b97bbbda39b34bf3212f4cf154a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 22 Jul 2009 15:36:25 +0100 Subject: softpipe: also shortcircuit non-texture tile lookups --- src/gallium/drivers/softpipe/sp_tile_cache.c | 33 ++++++++++++++-------------- src/gallium/drivers/softpipe/sp_tile_cache.h | 16 +++++++++++++- 2 files changed, 31 insertions(+), 18 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 06d9c6a80a..77d02fa3e7 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -49,7 +49,7 @@ * a LRU replacement policy. */ #define CACHE_POS(x, y) \ - (((x) / TILE_SIZE + ((y) / TILE_SIZE) * 5) % NUM_ENTRIES) + (((x) + (y) * 5) % NUM_ENTRIES) @@ -57,12 +57,10 @@ * Is the tile at (x,y) in cleared state? */ static INLINE uint -is_clear_flag_set(const uint *bitvec, int x, int y) +is_clear_flag_set(const uint *bitvec, union tile_address addr) { int pos, bit; - x /= TILE_SIZE; - y /= TILE_SIZE; - pos = y * (MAX_WIDTH / TILE_SIZE) + x; + pos = addr.bits.y * (MAX_WIDTH / TILE_SIZE) + addr.bits.x; assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32); bit = bitvec[pos / 32] & (1 << (pos & 31)); return bit; @@ -73,12 +71,10 @@ is_clear_flag_set(const uint *bitvec, int x, int y) * Mark the tile at (x,y) as not cleared. */ static INLINE void -clear_clear_flag(uint *bitvec, int x, int y) +clear_clear_flag(uint *bitvec, union tile_address addr) { int pos; - x /= TILE_SIZE; - y /= TILE_SIZE; - pos = y * (MAX_WIDTH / TILE_SIZE) + x; + pos = addr.bits.y * (MAX_WIDTH / TILE_SIZE) + addr.bits.x; assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32); bitvec[pos / 32] &= ~(1 << (pos & 31)); } @@ -349,13 +345,15 @@ sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc) /* push the tile to all positions marked as clear */ for (y = 0; y < h; y += TILE_SIZE) { for (x = 0; x < w; x += TILE_SIZE) { - if (is_clear_flag_set(tc->clear_flags, x, y)) { + union tile_address addr = tile_address(x, y, 0, 0, 0); + + if (is_clear_flag_set(tc->clear_flags, addr)) { pipe_put_tile_raw(pt, x, y, TILE_SIZE, TILE_SIZE, tc->tile.data.color32, 0/*STRIDE*/); /* do this? */ - clear_clear_flag(tc->clear_flags, x, y); + clear_clear_flag(tc->clear_flags, addr); numCleared++; } @@ -424,14 +422,14 @@ sp_flush_tile_cache(struct softpipe_tile_cache *tc) * \param x, y position of tile, in pixels */ struct softpipe_cached_tile * -sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) +sp_find_cached_tile(struct softpipe_tile_cache *tc, + union tile_address addr ) { struct pipe_transfer *pt = tc->transfer; - /* tile pos in framebuffer: */ - union tile_address addr = tile_address( x, y, 0, 0, 0 ); /* cache pos/entry: */ - const int pos = CACHE_POS(x, y); + const int pos = CACHE_POS(addr.bits.x, + addr.bits.y); struct softpipe_cached_tile *tile = tc->entries + pos; if (addr.value != tile->addr.value) { @@ -456,7 +454,7 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) tile->addr = addr; - if (is_clear_flag_set(tc->clear_flags, x, y)) { + if (is_clear_flag_set(tc->clear_flags, addr)) { /* don't get tile from framebuffer, just clear it */ if (tc->depth_stencil) { clear_tile(tile, pt->format, tc->clear_val); @@ -464,7 +462,7 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) else { clear_tile_rgba(tile, pt->format, tc->clear_color); } - clear_clear_flag(tc->clear_flags, x, y); + clear_clear_flag(tc->clear_flags, addr); } else { /* get new tile data from transfer */ @@ -485,6 +483,7 @@ sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y) } } + tc->last_tile = tile; return tile; } diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index 3017fcbebc..ac2aae5875 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -141,7 +141,8 @@ sp_tile_cache_clear(struct softpipe_tile_cache *tc, const float *rgba, uint clearValue); extern struct softpipe_cached_tile * -sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y); +sp_find_cached_tile(struct softpipe_tile_cache *tc, + union tile_address addr ); extern const struct softpipe_cached_tile * sp_find_cached_tile_tex(struct softpipe_tile_cache *tc, @@ -179,6 +180,19 @@ sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, } +static INLINE struct softpipe_cached_tile * +sp_get_cached_tile(struct softpipe_tile_cache *tc, + int x, int y ) +{ + union tile_address addr = tile_address( x, y, 0, 0, 0 ); + + if (tc->last_tile->addr.value == addr.value) + return tc->last_tile; + + return sp_find_cached_tile( tc, addr ); +} + + #endif /* SP_TILE_CACHE_H */ -- cgit v1.2.3 From 93a026d4baf90266f4c9cc48d039b4d65ce1ab6d Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 23 Jul 2009 11:14:39 +0100 Subject: softpipe: avoid flushing depth buffer cache on swapbuffers There's no need to push out depth buffer contents on swapbuffers. Note that this change doesn't throw away depth buffer changes, it simply holds them in the cache over calls to swapbuffers. The hope is that swapbuffers will be followed by a clear() which means in that case we won't have to write the changes out. --- src/gallium/drivers/softpipe/sp_context.c | 5 +---- src/gallium/drivers/softpipe/sp_flush.c | 20 +++++++++++++++----- 2 files changed, 16 insertions(+), 9 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index f085889d3a..1b2c6aded0 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -72,13 +72,10 @@ softpipe_unmap_transfers(struct softpipe_context *sp) { uint i; - for (i = 0; i < sp->framebuffer.nr_cbufs; i++) - sp_flush_tile_cache(sp->cbuf_cache[i]); - sp_flush_tile_cache(sp->zsbuf_cache); - for (i = 0; i < sp->framebuffer.nr_cbufs; i++) { sp_tile_cache_unmap_transfers(sp->cbuf_cache[i]); } + sp_tile_cache_unmap_transfers(sp->zsbuf_cache); } diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c index 732277a2c5..679ad0cd3d 100644 --- a/src/gallium/drivers/softpipe/sp_flush.c +++ b/src/gallium/drivers/softpipe/sp_flush.c @@ -56,14 +56,16 @@ softpipe_flush( struct pipe_context *pipe, } } - if (flags & PIPE_FLUSH_RENDER_CACHE) { + if (flags & PIPE_FLUSH_SWAPBUFFERS) { + /* If this is a swapbuffers, just flush color buffers. + * + * The zbuffer changes are not discarded, but held in the cache + * in the hope that a later clear will wipe them out. + */ for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) if (softpipe->cbuf_cache[i]) sp_flush_tile_cache(softpipe->cbuf_cache[i]); - if (softpipe->zsbuf_cache) - sp_flush_tile_cache(softpipe->zsbuf_cache); - /* Need this call for hardware buffers before swapbuffers. * * there should probably be another/different flush-type function @@ -71,7 +73,15 @@ softpipe_flush( struct pipe_context *pipe, * to unmap surfaces when flushing. */ softpipe_unmap_transfers(softpipe); - + } + else if (flags & PIPE_FLUSH_RENDER_CACHE) { + for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) + if (softpipe->cbuf_cache[i]) + sp_flush_tile_cache(softpipe->cbuf_cache[i]); + + if (softpipe->zsbuf_cache) + sp_flush_tile_cache(softpipe->zsbuf_cache); + softpipe->dirty_render_cache = FALSE; } -- cgit v1.2.3 From 6153a1c28f118be1a74ffee0e19c16fb83b5cab7 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 16:12:48 +0100 Subject: softpipe: rip out old mulithread support --- src/gallium/drivers/softpipe/sp_context.c | 48 +++--- src/gallium/drivers/softpipe/sp_context.h | 13 +- src/gallium/drivers/softpipe/sp_quad_pipe.c | 48 ++---- src/gallium/drivers/softpipe/sp_setup.c | 253 ++-------------------------- 4 files changed, 57 insertions(+), 305 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 1b2c6aded0..4418ef0ff4 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -88,19 +88,17 @@ static void softpipe_destroy( struct pipe_context *pipe ) if (softpipe->draw) draw_destroy( softpipe->draw ); - for (i = 0; i < SP_NUM_QUAD_THREADS; i++) { - softpipe->quad[i].polygon_stipple->destroy( softpipe->quad[i].polygon_stipple ); - softpipe->quad[i].earlyz->destroy( softpipe->quad[i].earlyz ); - softpipe->quad[i].shade->destroy( softpipe->quad[i].shade ); - softpipe->quad[i].alpha_test->destroy( softpipe->quad[i].alpha_test ); - softpipe->quad[i].depth_test->destroy( softpipe->quad[i].depth_test ); - softpipe->quad[i].stencil_test->destroy( softpipe->quad[i].stencil_test ); - softpipe->quad[i].occlusion->destroy( softpipe->quad[i].occlusion ); - softpipe->quad[i].coverage->destroy( softpipe->quad[i].coverage ); - softpipe->quad[i].blend->destroy( softpipe->quad[i].blend ); - softpipe->quad[i].colormask->destroy( softpipe->quad[i].colormask ); - softpipe->quad[i].output->destroy( softpipe->quad[i].output ); - } + softpipe->quad.polygon_stipple->destroy( softpipe->quad.polygon_stipple ); + softpipe->quad.earlyz->destroy( softpipe->quad.earlyz ); + softpipe->quad.shade->destroy( softpipe->quad.shade ); + softpipe->quad.alpha_test->destroy( softpipe->quad.alpha_test ); + softpipe->quad.depth_test->destroy( softpipe->quad.depth_test ); + softpipe->quad.stencil_test->destroy( softpipe->quad.stencil_test ); + softpipe->quad.occlusion->destroy( softpipe->quad.occlusion ); + softpipe->quad.coverage->destroy( softpipe->quad.coverage ); + softpipe->quad.blend->destroy( softpipe->quad.blend ); + softpipe->quad.colormask->destroy( softpipe->quad.colormask ); + softpipe->quad.output->destroy( softpipe->quad.output ); for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) sp_destroy_tile_cache(softpipe->cbuf_cache[i]); @@ -234,19 +232,17 @@ softpipe_create( struct pipe_screen *screen ) /* setup quad rendering stages */ - for (i = 0; i < SP_NUM_QUAD_THREADS; i++) { - softpipe->quad[i].polygon_stipple = sp_quad_polygon_stipple_stage(softpipe); - softpipe->quad[i].earlyz = sp_quad_earlyz_stage(softpipe); - softpipe->quad[i].shade = sp_quad_shade_stage(softpipe); - softpipe->quad[i].alpha_test = sp_quad_alpha_test_stage(softpipe); - softpipe->quad[i].depth_test = sp_quad_depth_test_stage(softpipe); - softpipe->quad[i].stencil_test = sp_quad_stencil_test_stage(softpipe); - softpipe->quad[i].occlusion = sp_quad_occlusion_stage(softpipe); - softpipe->quad[i].coverage = sp_quad_coverage_stage(softpipe); - softpipe->quad[i].blend = sp_quad_blend_stage(softpipe); - softpipe->quad[i].colormask = sp_quad_colormask_stage(softpipe); - softpipe->quad[i].output = sp_quad_output_stage(softpipe); - } + softpipe->quad.polygon_stipple = sp_quad_polygon_stipple_stage(softpipe); + softpipe->quad.earlyz = sp_quad_earlyz_stage(softpipe); + softpipe->quad.shade = sp_quad_shade_stage(softpipe); + softpipe->quad.alpha_test = sp_quad_alpha_test_stage(softpipe); + softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe); + softpipe->quad.stencil_test = sp_quad_stencil_test_stage(softpipe); + softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe); + softpipe->quad.coverage = sp_quad_coverage_stage(softpipe); + softpipe->quad.blend = sp_quad_blend_stage(softpipe); + softpipe->quad.colormask = sp_quad_colormask_stage(softpipe); + softpipe->quad.output = sp_quad_output_stage(softpipe); /* vertex shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index f66f0b7849..414d903a37 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -39,17 +39,6 @@ #include "sp_tex_sample.h" -/** - * This is a temporary variable for testing draw-stage polygon stipple. - * If zero, do stipple in sp_quad_stipple.c - */ -#define USE_DRAW_STAGE_PSTIPPLE 1 - -/* Number of threads working on individual quads. - * Setting to 1 disables this feature. - */ -#define SP_NUM_QUAD_THREADS 1 - struct softpipe_vbuf_render; struct draw_context; struct draw_stage; @@ -129,7 +118,7 @@ struct softpipe_context { struct quad_stage *output; struct quad_stage *first; /**< points to one of the above stages */ - } quad[SP_NUM_QUAD_THREADS]; + } quad; /** TGSI exec things */ struct { diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index b5f69b7426..3a3359d303 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -31,35 +31,29 @@ #include "pipe/p_shader_tokens.h" static void -sp_push_quad_first( - struct softpipe_context *sp, - struct quad_stage *quad, - uint i ) +sp_push_quad_first( struct softpipe_context *sp, + struct quad_stage *quad ) { - quad->next = sp->quad[i].first; - sp->quad[i].first = quad; + quad->next = sp->quad.first; + sp->quad.first = quad; } static void -sp_build_depth_stencil( - struct softpipe_context *sp, - uint i ) +sp_build_depth_stencil( struct softpipe_context *sp ) { if (sp->depth_stencil->stencil[0].enabled || sp->depth_stencil->stencil[1].enabled) { - sp_push_quad_first( sp, sp->quad[i].stencil_test, i ); + sp_push_quad_first( sp, sp->quad.stencil_test ); } else if (sp->depth_stencil->depth.enabled && sp->framebuffer.zsbuf) { - sp_push_quad_first( sp, sp->quad[i].depth_test, i ); + sp_push_quad_first( sp, sp->quad.depth_test ); } } void sp_build_quad_pipeline(struct softpipe_context *sp) { - uint i; - boolean early_depth_test = sp->depth_stencil->depth.enabled && sp->framebuffer.zsbuf && @@ -68,51 +62,43 @@ sp_build_quad_pipeline(struct softpipe_context *sp) !sp->fs->info.writes_z; /* build up the pipeline in reverse order... */ - for (i = 0; i < SP_NUM_QUAD_THREADS; i++) { - sp->quad[i].first = sp->quad[i].output; + sp->quad.first = sp->quad.output; if (sp->blend->colormask != 0xf) { - sp_push_quad_first( sp, sp->quad[i].colormask, i ); + sp_push_quad_first( sp, sp->quad.colormask ); } if (sp->blend->blend_enable || sp->blend->logicop_enable) { - sp_push_quad_first( sp, sp->quad[i].blend, i ); + sp_push_quad_first( sp, sp->quad.blend ); } if (sp->active_query_count) { - sp_push_quad_first( sp, sp->quad[i].occlusion, i ); + sp_push_quad_first( sp, sp->quad.occlusion ); } if (sp->rasterizer->poly_smooth || sp->rasterizer->line_smooth || sp->rasterizer->point_smooth) { - sp_push_quad_first( sp, sp->quad[i].coverage, i ); + sp_push_quad_first( sp, sp->quad.coverage ); } if (!early_depth_test) { - sp_build_depth_stencil( sp, i ); + sp_build_depth_stencil( sp ); } if (sp->depth_stencil->alpha.enabled) { - sp_push_quad_first( sp, sp->quad[i].alpha_test, i ); + sp_push_quad_first( sp, sp->quad.alpha_test ); } /* XXX always enable shader? */ if (1) { - sp_push_quad_first( sp, sp->quad[i].shade, i ); + sp_push_quad_first( sp, sp->quad.shade ); } if (early_depth_test) { - sp_build_depth_stencil( sp, i ); - sp_push_quad_first( sp, sp->quad[i].earlyz, i ); - } - -#if !USE_DRAW_STAGE_PSTIPPLE - if (sp->rasterizer->poly_stipple_enable) { - sp_push_quad_first( sp, sp->quad[i].polygon_stipple, i ); + sp_build_depth_stencil( sp ); + sp_push_quad_first( sp, sp->quad.earlyz ); } -#endif - } } diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index d05c9ad57c..eaf84ed9de 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -61,87 +61,7 @@ struct edge { int lines; /**< number of lines on this edge */ }; -#if SP_NUM_QUAD_THREADS > 1 -/* Set to 1 if you want other threads to be instantly - * notified of pending jobs. - */ -#define INSTANT_NOTEMPTY_NOTIFY 0 - -struct thread_info -{ - struct setup_context *setup; - uint id; - pipe_thread handle; -}; - -struct quad_job; - -typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job ); - -struct quad_job -{ - struct quad_header_input input; - struct quad_header_inout inout; - quad_job_routine routine; -}; - -#define NUM_QUAD_JOBS 64 - -struct quad_job_que -{ - struct quad_job jobs[NUM_QUAD_JOBS]; - uint first; - uint last; - pipe_mutex que_mutex; - pipe_condvar que_notfull_condvar; - pipe_condvar que_notempty_condvar; - uint jobs_added; - uint jobs_done; - pipe_condvar que_done_condvar; -}; - -static void -add_quad_job( struct quad_job_que *que, struct quad_header *quad, quad_job_routine routine ) -{ -#if INSTANT_NOTEMPTY_NOTIFY - boolean empty; -#endif - - /* Wait for empty slot, see if the que is empty. - */ - pipe_mutex_lock( que->que_mutex ); - while ((que->last + 1) % NUM_QUAD_JOBS == que->first) { -#if !INSTANT_NOTEMPTY_NOTIFY - pipe_condvar_broadcast( que->que_notempty_condvar ); -#endif - pipe_condvar_wait( que->que_notfull_condvar, que->que_mutex ); - } -#if INSTANT_NOTEMPTY_NOTIFY - empty = que->last == que->first; -#endif - que->jobs_added++; - pipe_mutex_unlock( que->que_mutex ); - - /* Submit new job. - */ - que->jobs[que->last].input = quad->input; - que->jobs[que->last].inout = quad->inout; - que->jobs[que->last].routine = routine; - que->last = (que->last + 1) % NUM_QUAD_JOBS; - -#if INSTANT_NOTEMPTY_NOTIFY - /* If the que was empty, notify consumers there's a job to be done. - */ - if (empty) { - pipe_mutex_lock( que->que_mutex ); - pipe_condvar_broadcast( que->que_notempty_condvar ); - pipe_mutex_unlock( que->que_mutex ); - } -#endif -} - -#endif /** * Triangle setup info (derived from draw_stage). @@ -169,11 +89,6 @@ struct setup_context { struct tgsi_interp_coef posCoef; /* For Z, W */ struct quad_header quad; -#if SP_NUM_QUAD_THREADS > 1 - struct quad_job_que que; - struct thread_info threads[SP_NUM_QUAD_THREADS]; -#endif - struct { int left[2]; /**< [0] = row0, [1] = row1 */ int right[2]; @@ -188,67 +103,6 @@ struct setup_context { unsigned winding; /* which winding to cull */ }; -#if SP_NUM_QUAD_THREADS > 1 - -static PIPE_THREAD_ROUTINE( quad_thread, param ) -{ - struct thread_info *info = (struct thread_info *) param; - struct quad_job_que *que = &info->setup->que; - - for (;;) { - struct quad_job job; - boolean full; - - /* Wait for an available job. - */ - pipe_mutex_lock( que->que_mutex ); - while (que->last == que->first) - pipe_condvar_wait( que->que_notempty_condvar, que->que_mutex ); - - /* See if the que is full. - */ - full = (que->last + 1) % NUM_QUAD_JOBS == que->first; - - /* Take a job and remove it from que. - */ - job = que->jobs[que->first]; - que->first = (que->first + 1) % NUM_QUAD_JOBS; - - /* Notify the producer if the que is not full. - */ - if (full) - pipe_condvar_signal( que->que_notfull_condvar ); - pipe_mutex_unlock( que->que_mutex ); - - job.routine( info->setup, info->id, &job ); - - /* Notify the producer if that's the last finished job. - */ - pipe_mutex_lock( que->que_mutex ); - que->jobs_done++; - if (que->jobs_added == que->jobs_done) - pipe_condvar_signal( que->que_done_condvar ); - pipe_mutex_unlock( que->que_mutex ); - } - - return NULL; -} - -#define WAIT_FOR_COMPLETION(setup) \ - do {\ - pipe_mutex_lock( setup->que.que_mutex );\ - if (!INSTANT_NOTEMPTY_NOTIFY)\ - pipe_condvar_broadcast( setup->que.que_notempty_condvar );\ - while (setup->que.jobs_added != setup->que.jobs_done)\ - pipe_condvar_wait( setup->que.que_done_condvar, setup->que.que_mutex );\ - pipe_mutex_unlock( setup->que.que_mutex );\ - } while (0) - -#else - -#define WAIT_FOR_COMPLETION(setup) ((void) 0) - -#endif @@ -311,39 +165,17 @@ quad_clip( struct setup_context *setup, struct quad_header *quad ) * Emit a quad (pass to next stage) with clipping. */ static INLINE void -clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) +clip_emit_quad( struct setup_context *setup, struct quad_header *quad ) { quad_clip( setup, quad ); + if (quad->inout.mask) { struct softpipe_context *sp = setup->softpipe; - sp->quad[thread].first->run( sp->quad[thread].first, quad ); + sp->quad.first->run( sp->quad.first, quad ); } } -#if SP_NUM_QUAD_THREADS > 1 - -static void -clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job ) -{ - struct quad_header quad; - - quad.input = job->input; - quad.inout = job->inout; - quad.coef = setup->quad.coef; - quad.posCoef = setup->quad.posCoef; - quad.nr_attrs = setup->quad.nr_attrs; - clip_emit_quad( setup, &quad, thread ); -} - -#define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, &setup->quad, clip_emit_quad_job ) - -#else - -#define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 ) - -#endif - /** * Emit a quad (pass to next stage). No clipping is done. */ @@ -361,7 +193,7 @@ emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) if (mask & 4) setup->numFragsEmitted++; if (mask & 8) setup->numFragsEmitted++; #endif - sp->quad[thread].first->run( sp->quad[thread].first, quad ); + sp->quad.first->run( sp->quad.first, quad ); #if DEBUG_FRAGS mask = quad->inout.mask; if (mask & 1) setup->numFragsWritten++; @@ -371,38 +203,15 @@ emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) #endif } -#if SP_NUM_QUAD_THREADS > 1 - -static void -emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job ) -{ - struct quad_header quad; - - quad.input = job->input; - quad.inout = job->inout; - quad.coef = setup->quad.coef; - quad.posCoef = setup->quad.posCoef; - quad.nr_attrs = setup->quad.nr_attrs; - emit_quad( setup, &quad, thread ); -} - -#define EMIT_QUAD(setup,x,y,qmask) do {\ - setup->quad.input.x0 = x;\ - setup->quad.input.y0 = y;\ - setup->quad.inout.mask = qmask;\ - add_quad_job( &setup->que, &setup->quad, emit_quad_job );\ - } while (0) -#else +#define EMIT_QUAD(setup,x,y,qmask) \ +do { \ + setup->quad.input.x0 = x; \ + setup->quad.input.y0 = y; \ + setup->quad.inout.mask = qmask; \ + emit_quad( setup, &setup->quad, 0 ); \ +} while (0) -#define EMIT_QUAD(setup,x,y,qmask) do {\ - setup->quad.input.x0 = x;\ - setup->quad.input.y0 = y;\ - setup->quad.inout.mask = qmask;\ - emit_quad( setup, &setup->quad, 0 );\ - } while (0) - -#endif /** * Given an X or Y coordinate, return the block/quad coordinate that it @@ -956,8 +765,6 @@ void setup_tri( struct setup_context *setup, flush_spans( setup ); - WAIT_FOR_COMPLETION(setup); - #if DEBUG_FRAGS printf("Tri: %u frags emitted, %u written\n", setup->numFragsEmitted, @@ -1101,7 +908,7 @@ plot(struct setup_context *setup, int x, int y) /* flush prev quad, start new quad */ if (setup->quad.input.x0 != -1) - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); setup->quad.input.x0 = quadX; setup->quad.input.y0 = quadY; @@ -1223,10 +1030,8 @@ setup_line(struct setup_context *setup, /* draw final quad */ if (setup->quad.inout.mask) { - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } - - WAIT_FOR_COMPLETION(setup); } @@ -1334,7 +1139,7 @@ setup_point( struct setup_context *setup, setup->quad.input.x0 = (int) x - ix; setup->quad.input.y0 = (int) y - iy; setup->quad.inout.mask = (1 << ix) << (2 * iy); - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } else { if (round) { @@ -1395,7 +1200,7 @@ setup_point( struct setup_context *setup, if (setup->quad.inout.mask) { setup->quad.input.x0 = ix; setup->quad.input.y0 = iy; - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } } } @@ -1442,19 +1247,16 @@ setup_point( struct setup_context *setup, setup->quad.inout.mask = mask; setup->quad.input.x0 = ix; setup->quad.input.y0 = iy; - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } } } } - - WAIT_FOR_COMPLETION(setup); } void setup_prepare( struct setup_context *setup ) { struct softpipe_context *sp = setup->softpipe; - unsigned i; if (sp->dirty) { softpipe_update_derived(sp); @@ -1463,9 +1265,7 @@ void setup_prepare( struct setup_context *setup ) /* Note: nr_attrs is only used for debugging (vertex printing) */ setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw); - for (i = 0; i < SP_NUM_QUAD_THREADS; i++) { - sp->quad[i].first->begin( sp->quad[i].first ); - } + sp->quad.first->begin( sp->quad.first ); if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES && sp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL && @@ -1493,9 +1293,6 @@ void setup_destroy_context( struct setup_context *setup ) struct setup_context *setup_create_context( struct softpipe_context *softpipe ) { struct setup_context *setup = CALLOC_STRUCT(setup_context); -#if SP_NUM_QUAD_THREADS > 1 - uint i; -#endif setup->softpipe = softpipe; @@ -1505,22 +1302,6 @@ struct setup_context *setup_create_context( struct softpipe_context *softpipe ) setup->span.left[0] = 1000000; /* greater than right[0] */ setup->span.left[1] = 1000000; /* greater than right[1] */ -#if SP_NUM_QUAD_THREADS > 1 - setup->que.first = 0; - setup->que.last = 0; - pipe_mutex_init( setup->que.que_mutex ); - pipe_condvar_init( setup->que.que_notfull_condvar ); - pipe_condvar_init( setup->que.que_notempty_condvar ); - setup->que.jobs_added = 0; - setup->que.jobs_done = 0; - pipe_condvar_init( setup->que.que_done_condvar ); - for (i = 0; i < SP_NUM_QUAD_THREADS; i++) { - setup->threads[i].setup = setup; - setup->threads[i].id = i; - setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] ); - } -#endif - return setup; } -- cgit v1.2.3 From ab9fb5167023a26566b53e98f206dd73a18000f3 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 16:49:35 +0100 Subject: softpipe: expand quad pipeline to process >1 quad at a time This is part one -- we still only pass a single quad down, but the code can now cope with more. The quads must all be from the same tile. --- src/gallium/drivers/softpipe/sp_quad_alpha_test.c | 106 ++-- src/gallium/drivers/softpipe/sp_quad_blend.c | 730 +++++++++++----------- src/gallium/drivers/softpipe/sp_quad_colormask.c | 15 +- src/gallium/drivers/softpipe/sp_quad_coverage.c | 48 +- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 23 +- src/gallium/drivers/softpipe/sp_quad_earlyz.c | 28 +- src/gallium/drivers/softpipe/sp_quad_fs.c | 40 +- src/gallium/drivers/softpipe/sp_quad_occlusion.c | 10 +- src/gallium/drivers/softpipe/sp_quad_output.c | 49 +- src/gallium/drivers/softpipe/sp_quad_pipe.c | 88 +-- src/gallium/drivers/softpipe/sp_quad_pipe.h | 4 +- src/gallium/drivers/softpipe/sp_quad_stencil.c | 185 +++--- src/gallium/drivers/softpipe/sp_quad_stipple.c | 48 +- src/gallium/drivers/softpipe/sp_setup.c | 4 +- 14 files changed, 745 insertions(+), 633 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c index 0845bae0e6..3a282208b6 100644 --- a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c @@ -9,76 +9,80 @@ #include "pipe/p_defines.h" #include "util/u_memory.h" +#define ALPHATEST( FUNC, COMP ) \ + static void \ + alpha_test_quads_##FUNC( struct quad_stage *qs, \ + struct quad_header *quads[], \ + unsigned nr ) \ + { \ + const float ref = qs->softpipe->depth_stencil->alpha.ref_value; \ + const uint cbuf = 0; /* only output[0].alpha is tested */ \ + unsigned pass_nr = 0; \ + unsigned i; \ + \ + for (i = 0; i < nr; i++) { \ + const float *aaaa = quads[i]->output.color[cbuf][3]; \ + unsigned passMask = 0; \ + \ + if (aaaa[0] COMP ref) passMask |= (1 << 0); \ + if (aaaa[1] COMP ref) passMask |= (1 << 1); \ + if (aaaa[2] COMP ref) passMask |= (1 << 2); \ + if (aaaa[3] COMP ref) passMask |= (1 << 3); \ + \ + quads[i]->inout.mask &= passMask; \ + \ + if (quads[i]->inout.mask) \ + quads[pass_nr++] = quads[i]; \ + } \ + \ + if (pass_nr) \ + qs->next->run(qs->next, quads, pass_nr); \ + } + + +ALPHATEST( LESS, < ) +ALPHATEST( EQUAL, == ) +ALPHATEST( LEQUAL, <= ) +ALPHATEST( GREATER, > ) +ALPHATEST( NOTEQUAL, != ) +ALPHATEST( GEQUAL, >= ) + +/* XXX: Incorporate into shader using KILP. + */ static void -alpha_test_quad(struct quad_stage *qs, struct quad_header *quad) +alpha_test_quad(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { - struct softpipe_context *softpipe = qs->softpipe; - const float ref = softpipe->depth_stencil->alpha.ref_value; - unsigned passMask = 0x0, j; - const uint cbuf = 0; /* only output[0].alpha is tested */ - const float *aaaa = quad->output.color[cbuf][3]; - - switch (softpipe->depth_stencil->alpha.func) { - case PIPE_FUNC_NEVER: - break; + switch (qs->softpipe->depth_stencil->alpha.func) { case PIPE_FUNC_LESS: - /* - * If mask were an array [4] we could do this SIMD-style: - * passMask = (quad->outputs.color[0][3] <= vec4(ref)); - */ - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] < ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_LESS( qs, quads, nr ); break; case PIPE_FUNC_EQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] == ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_EQUAL( qs, quads, nr ); break; case PIPE_FUNC_LEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] <= ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_LEQUAL( qs, quads, nr ); break; case PIPE_FUNC_GREATER: - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] > ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_GREATER( qs, quads, nr ); break; case PIPE_FUNC_NOTEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] != ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_NOTEQUAL( qs, quads, nr ); break; case PIPE_FUNC_GEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (aaaa[j] >= ref) { - passMask |= (1 << j); - } - } + alpha_test_quads_GEQUAL( qs, quads, nr ); break; case PIPE_FUNC_ALWAYS: - passMask = MASK_ALL; + assert(0); /* should be caught earlier */ + qs->next->run(qs->next, quads, nr); break; + case PIPE_FUNC_NEVER: default: - assert(0); + assert(0); /* should be caught earlier */ + return; } - - quad->inout.mask &= passMask; - - if (quad->inout.mask) - qs->next->run(qs->next, quad); } diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index 04b5daf3a4..fdf1bb4552 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -117,10 +117,16 @@ do { \ static void -logicop_quad(struct quad_stage *qs, struct quad_header *quad) +logicop_quad(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; uint cbuf; + struct softpipe_cached_tile * + tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], + quads[0]->input.x0, + quads[0]->input.y0); /* loop over colorbuffer outputs */ for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { @@ -129,165 +135,161 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad) uint *src4 = (uint *) src; uint *dst4 = (uint *) dst; uint *res4 = (uint *) res; - struct softpipe_cached_tile * - tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quad->input.x0, quad->input.y0); - float (*quadColor)[4] = quad->output.color[cbuf]; uint i, j; - /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); - for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; + for (i = 0; i < nr; i++) { + struct quad_header *quad = quads[i]; + float (*quadColor)[4] = quad->output.color[cbuf]; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } } - } - /* convert to ubyte */ - for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */ - dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */ - dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */ - dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */ - dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */ - - src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */ - src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */ - src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */ - src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */ - } + /* convert to ubyte */ + for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */ + dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */ + dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */ + dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */ + dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */ + + src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */ + src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */ + src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */ + src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */ + } - switch (softpipe->blend->logicop_func) { - case PIPE_LOGICOP_CLEAR: - for (j = 0; j < 4; j++) - res4[j] = 0; - break; - case PIPE_LOGICOP_NOR: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] | dst4[j]); - break; - case PIPE_LOGICOP_AND_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j] & dst4[j]; - break; - case PIPE_LOGICOP_COPY_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j]; - break; - case PIPE_LOGICOP_AND_REVERSE: - for (j = 0; j < 4; j++) - res4[j] = src4[j] & ~dst4[j]; - break; - case PIPE_LOGICOP_INVERT: - for (j = 0; j < 4; j++) - res4[j] = ~dst4[j]; - break; - case PIPE_LOGICOP_XOR: - for (j = 0; j < 4; j++) - res4[j] = dst4[j] ^ src4[j]; - break; - case PIPE_LOGICOP_NAND: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] & dst4[j]); - break; - case PIPE_LOGICOP_AND: - for (j = 0; j < 4; j++) - res4[j] = src4[j] & dst4[j]; - break; - case PIPE_LOGICOP_EQUIV: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] ^ dst4[j]); - break; - case PIPE_LOGICOP_NOOP: - for (j = 0; j < 4; j++) - res4[j] = dst4[j]; - break; - case PIPE_LOGICOP_OR_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j] | dst4[j]; - break; - case PIPE_LOGICOP_COPY: - for (j = 0; j < 4; j++) - res4[j] = src4[j]; - break; - case PIPE_LOGICOP_OR_REVERSE: - for (j = 0; j < 4; j++) - res4[j] = src4[j] | ~dst4[j]; - break; - case PIPE_LOGICOP_OR: - for (j = 0; j < 4; j++) - res4[j] = src4[j] | dst4[j]; - break; - case PIPE_LOGICOP_SET: - for (j = 0; j < 4; j++) - res4[j] = ~0; - break; - default: - assert(0); - } + switch (softpipe->blend->logicop_func) { + case PIPE_LOGICOP_CLEAR: + for (j = 0; j < 4; j++) + res4[j] = 0; + break; + case PIPE_LOGICOP_NOR: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] | dst4[j]); + break; + case PIPE_LOGICOP_AND_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_COPY_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j]; + break; + case PIPE_LOGICOP_AND_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & ~dst4[j]; + break; + case PIPE_LOGICOP_INVERT: + for (j = 0; j < 4; j++) + res4[j] = ~dst4[j]; + break; + case PIPE_LOGICOP_XOR: + for (j = 0; j < 4; j++) + res4[j] = dst4[j] ^ src4[j]; + break; + case PIPE_LOGICOP_NAND: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] & dst4[j]); + break; + case PIPE_LOGICOP_AND: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_EQUIV: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] ^ dst4[j]); + break; + case PIPE_LOGICOP_NOOP: + for (j = 0; j < 4; j++) + res4[j] = dst4[j]; + break; + case PIPE_LOGICOP_OR_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_COPY: + for (j = 0; j < 4; j++) + res4[j] = src4[j]; + break; + case PIPE_LOGICOP_OR_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | ~dst4[j]; + break; + case PIPE_LOGICOP_OR: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_SET: + for (j = 0; j < 4; j++) + res4[j] = ~0; + break; + default: + assert(0); + } - for (j = 0; j < 4; j++) { - quadColor[j][0] = ubyte_to_float(res[j][0]); - quadColor[j][1] = ubyte_to_float(res[j][1]); - quadColor[j][2] = ubyte_to_float(res[j][2]); - quadColor[j][3] = ubyte_to_float(res[j][3]); + for (j = 0; j < 4; j++) { + quadColor[j][0] = ubyte_to_float(res[j][0]); + quadColor[j][1] = ubyte_to_float(res[j][1]); + quadColor[j][2] = ubyte_to_float(res[j][2]); + quadColor[j][3] = ubyte_to_float(res[j][3]); + } } } - - /* pass quad to next stage */ - qs->next->run(qs->next, quad); } - - static void -blend_quad(struct quad_stage *qs, struct quad_header *quad) +blend_quads(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { static const float zero[4] = { 0, 0, 0, 0 }; static const float one[4] = { 1, 1, 1, 1 }; - struct softpipe_context *softpipe = qs->softpipe; uint cbuf; - if (softpipe->blend->logicop_enable) { - logicop_quad(qs, quad); - return; - } - /* loop over colorbuffer outputs */ for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { float source[4][QUAD_SIZE], dest[4][QUAD_SIZE]; struct softpipe_cached_tile *tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quad->input.x0, quad->input.y0); - float (*quadColor)[4] = quad->output.color[cbuf]; - uint i, j; - - /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); - for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; + quads[0]->input.x0, + quads[0]->input.y0); + uint q, i, j; + + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[cbuf]; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } } - } - /* - * Compute src/first term RGB - */ - switch (softpipe->blend->rgb_src_factor) { - case PIPE_BLENDFACTOR_ONE: - VEC4_COPY(source[0], quadColor[0]); /* R */ - VEC4_COPY(source[1], quadColor[1]); /* G */ - VEC4_COPY(source[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */ - VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */ - VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: + /* + * Compute src/first term RGB + */ + switch (softpipe->blend->rgb_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[0], quadColor[0]); /* R */ + VEC4_COPY(source[1], quadColor[1]); /* G */ + VEC4_COPY(source[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: { const float *alpha = quadColor[3]; VEC4_MUL(source[0], quadColor[0], alpha); /* R */ @@ -295,12 +297,12 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], alpha); /* B */ } break; - case PIPE_BLENDFACTOR_DST_COLOR: - VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */ - VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */ - VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */ - break; - case PIPE_BLENDFACTOR_DST_ALPHA: + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: { const float *alpha = dest[3]; VEC4_MUL(source[0], quadColor[0], alpha); /* R */ @@ -308,7 +310,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], alpha); /* B */ } break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: { const float *alpha = quadColor[3]; float diff[4], temp[4]; @@ -319,7 +321,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], temp); /* B */ } break; - case PIPE_BLENDFACTOR_CONST_COLOR: + case PIPE_BLENDFACTOR_CONST_COLOR: { float comp[4]; VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ @@ -330,7 +332,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], comp); /* B */ } break; - case PIPE_BLENDFACTOR_CONST_ALPHA: + case PIPE_BLENDFACTOR_CONST_ALPHA: { float alpha[4]; VEC4_SCALAR(alpha, softpipe->blend_color.color[3]); @@ -339,18 +341,18 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], alpha); /* B */ } break; - case PIPE_BLENDFACTOR_SRC1_COLOR: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_SRC1_ALPHA: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(source[0], zero); /* R */ - VEC4_COPY(source[1], zero); /* G */ - VEC4_COPY(source[2], zero); /* B */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: + case PIPE_BLENDFACTOR_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[0], zero); /* R */ + VEC4_COPY(source[1], zero); /* G */ + VEC4_COPY(source[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: { float inv_comp[4]; VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ @@ -361,7 +363,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ } break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: { float inv_alpha[4]; VEC4_SUB(inv_alpha, one, quadColor[3]); @@ -370,7 +372,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ } break; - case PIPE_BLENDFACTOR_INV_DST_ALPHA: + case PIPE_BLENDFACTOR_INV_DST_ALPHA: { float inv_alpha[4]; VEC4_SUB(inv_alpha, one, dest[3]); @@ -379,7 +381,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ } break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: + case PIPE_BLENDFACTOR_INV_DST_COLOR: { float inv_comp[4]; VEC4_SUB(inv_comp, one, dest[0]); /* R */ @@ -390,7 +392,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ } break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: + case PIPE_BLENDFACTOR_INV_CONST_COLOR: { float inv_comp[4]; /* R */ @@ -404,7 +406,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_comp); } break; - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: { float inv_alpha[4]; VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]); @@ -413,73 +415,73 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ } break; - case PIPE_BLENDFACTOR_INV_SRC1_COLOR: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: - assert(0); /* to do */ - break; - default: - assert(0); - } + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + assert(0); /* to do */ + break; + default: + assert(0); + } - /* - * Compute src/first term A - */ - switch (softpipe->blend->alpha_src_factor) { - case PIPE_BLENDFACTOR_ONE: - VEC4_COPY(source[3], quadColor[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_SRC_ALPHA: + /* + * Compute src/first term A + */ + switch (softpipe->blend->alpha_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[3], quadColor[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: { const float *alpha = quadColor[3]; VEC4_MUL(source[3], quadColor[3], alpha); /* A */ } break; - case PIPE_BLENDFACTOR_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - /* multiply alpha by 1.0 */ - VEC4_COPY(source[3], quadColor[3]); /* A */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_CONST_ALPHA: + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + /* multiply alpha by 1.0 */ + VEC4_COPY(source[3], quadColor[3]); /* A */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: { float comp[4]; VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ VEC4_MUL(source[3], quadColor[3], comp); /* A */ } break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(source[3], zero); /* A */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: { float inv_alpha[4]; VEC4_SUB(inv_alpha, one, quadColor[3]); VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ } break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_DST_ALPHA: + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: { float inv_alpha[4]; VEC4_SUB(inv_alpha, one, dest[3]); VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ } break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: { float inv_comp[4]; /* A */ @@ -487,42 +489,42 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(source[3], quadColor[3], inv_comp); } break; - default: - assert(0); - } + default: + assert(0); + } - /* - * Compute dest/second term RGB - */ - switch (softpipe->blend->rgb_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - /* dest = dest * 1 NO-OP, leave dest as-is */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */ - VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */ - VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: - VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */ - VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */ - VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */ - break; - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */ - VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */ - VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */ - break; - case PIPE_BLENDFACTOR_DST_COLOR: - VEC4_MUL(dest[0], dest[0], dest[0]); /* R */ - VEC4_MUL(dest[1], dest[1], dest[1]); /* G */ - VEC4_MUL(dest[2], dest[2], dest[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - assert(0); /* illegal */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: + /* + * Compute dest/second term RGB + */ + switch (softpipe->blend->rgb_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */ + VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */ + VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(dest[0], dest[0], dest[0]); /* R */ + VEC4_MUL(dest[1], dest[1], dest[1]); /* G */ + VEC4_MUL(dest[2], dest[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: { float comp[4]; VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ @@ -533,7 +535,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], comp); /* B */ } break; - case PIPE_BLENDFACTOR_CONST_ALPHA: + case PIPE_BLENDFACTOR_CONST_ALPHA: { float comp[4]; VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ @@ -542,17 +544,17 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], comp); /* B */ } break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(dest[0], zero); /* R */ - VEC4_COPY(dest[1], zero); /* G */ - VEC4_COPY(dest[2], zero); /* B */ - break; - case PIPE_BLENDFACTOR_SRC1_COLOR: - case PIPE_BLENDFACTOR_SRC1_ALPHA: - /* XXX what are these? */ - assert(0); - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[0], zero); /* R */ + VEC4_COPY(dest[1], zero); /* G */ + VEC4_COPY(dest[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + case PIPE_BLENDFACTOR_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: { float inv_comp[4]; VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ @@ -563,7 +565,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ } break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: { float one_minus_alpha[QUAD_SIZE]; VEC4_SUB(one_minus_alpha, one, quadColor[3]); @@ -572,7 +574,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ } break; - case PIPE_BLENDFACTOR_INV_DST_ALPHA: + case PIPE_BLENDFACTOR_INV_DST_ALPHA: { float inv_comp[4]; VEC4_SUB(inv_comp, one, dest[3]); /* A */ @@ -581,7 +583,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ } break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: + case PIPE_BLENDFACTOR_INV_DST_COLOR: { float inv_comp[4]; VEC4_SUB(inv_comp, one, dest[0]); /* R */ @@ -592,7 +594,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], inv_comp); /* B */ } break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: + case PIPE_BLENDFACTOR_INV_CONST_COLOR: { float inv_comp[4]; /* R */ @@ -606,7 +608,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], inv_comp); } break; - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: { float inv_comp[4]; VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); @@ -615,138 +617,154 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad) VEC4_MUL(dest[2], dest[2], inv_comp); } break; - case PIPE_BLENDFACTOR_INV_SRC1_COLOR: - case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: - /* XXX what are these? */ - assert(0); - break; - default: - assert(0); - } + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + default: + assert(0); + } - /* - * Compute dest/second term A - */ - switch (softpipe->blend->alpha_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - /* dest = dest * 1 NO-OP, leave dest as-is */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_SRC_ALPHA: - VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */ - break; - case PIPE_BLENDFACTOR_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(dest[3], dest[3], dest[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - assert(0); /* illegal */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_CONST_ALPHA: + /* + * Compute dest/second term A + */ + switch (softpipe->blend->alpha_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[3], dest[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: { float comp[4]; VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ VEC4_MUL(dest[3], dest[3], comp); /* A */ } break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(dest[3], zero); /* A */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: { float one_minus_alpha[QUAD_SIZE]; VEC4_SUB(one_minus_alpha, one, quadColor[3]); VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */ } break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_DST_ALPHA: + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: { float inv_comp[4]; VEC4_SUB(inv_comp, one, dest[3]); /* A */ VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */ } break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: { float inv_comp[4]; VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); VEC4_MUL(dest[3], dest[3], inv_comp); } break; - default: - assert(0); - } + default: + assert(0); + } - /* - * Combine RGB terms - */ - switch (softpipe->blend->rgb_func) { - case PIPE_BLEND_ADD: - VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ - VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ - VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_SUBTRACT: - VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */ - VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */ - VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_REVERSE_SUBTRACT: - VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */ - VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */ - VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */ - break; - case PIPE_BLEND_MIN: - VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */ - VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */ - VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_MAX: - VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */ - VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */ - VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */ - break; - default: - assert(0); - } + /* + * Combine RGB terms + */ + switch (softpipe->blend->rgb_func) { + case PIPE_BLEND_ADD: + VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */ + VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */ + VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */ + break; + default: + assert(0); + } - /* - * Combine A terms - */ - switch (softpipe->blend->alpha_func) { - case PIPE_BLEND_ADD: - VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_SUBTRACT: - VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_REVERSE_SUBTRACT: - VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */ - break; - case PIPE_BLEND_MIN: - VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_MAX: - VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */ - break; - default: - assert(0); + /* + * Combine A terms + */ + switch (softpipe->blend->alpha_func) { + case PIPE_BLEND_ADD: + VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */ + break; + default: + assert(0); + } } - } /* cbuf loop */ +} + + +static void +blend_quad(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + struct softpipe_context *softpipe = qs->softpipe; + + if (softpipe->blend->logicop_enable) { + logicop_quad(qs, quads, nr); + } + else if (softpipe->blend->blend_enable) { + blend_quads(qs, quads, nr ); + } /* pass blended quad to next stage */ - qs->next->run(qs->next, quad); + qs->next->run(qs->next, quads, nr); } diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c index 89efbe3b02..ac74287473 100644 --- a/src/gallium/drivers/softpipe/sp_quad_colormask.c +++ b/src/gallium/drivers/softpipe/sp_quad_colormask.c @@ -84,12 +84,23 @@ colormask_quad(struct quad_stage *qs, struct quad_header *quad) if (!(softpipe->blend->colormask & PIPE_MASK_A)) COPY_4V(quadColor[3], dest[3]); } +} + +static void +colormask_quads(struct quad_stage *qs, struct quad_header *quads[], + unsigned nr) +{ + unsigned i; + + for (i = 0; i < nr; i++) + colormask_quad(qs, quads[i]); /* pass quad to next stage */ - qs->next->run(qs->next, quad); + qs->next->run(qs->next, quads, nr); } + static void colormask_begin(struct quad_stage *qs) { qs->next->begin(qs->next); @@ -108,7 +119,7 @@ struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = colormask_begin; - stage->run = colormask_quad; + stage->run = colormask_quads; stage->destroy = colormask_destroy; return stage; diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c index 4aeee85870..eda5ce8e63 100644 --- a/src/gallium/drivers/softpipe/sp_quad_coverage.c +++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c @@ -42,33 +42,47 @@ /** * Multiply quad's alpha values by the fragment coverage. */ -static void +static INLINE void coverage_quad(struct quad_stage *qs, struct quad_header *quad) { struct softpipe_context *softpipe = qs->softpipe; - const uint prim = quad->input.prim; + uint cbuf; + + /* loop over colorbuffer outputs */ + for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { + float (*quadColor)[4] = quad->output.color[cbuf]; + unsigned j; + for (j = 0; j < QUAD_SIZE; j++) { + assert(quad->input.coverage[j] >= 0.0); + assert(quad->input.coverage[j] <= 1.0); + quadColor[3][j] *= quad->input.coverage[j]; + } + } +} + + +/* XXX: Incorporate into shader after alpha_test. + */ +static void +coverage_run(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + struct softpipe_context *softpipe = qs->softpipe; + const uint prim = quads[0]->input.prim; + unsigned i; if ((softpipe->rasterizer->poly_smooth && prim == QUAD_PRIM_TRI) || (softpipe->rasterizer->line_smooth && prim == QUAD_PRIM_LINE) || (softpipe->rasterizer->point_smooth && prim == QUAD_PRIM_POINT)) { - uint cbuf; - - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { - float (*quadColor)[4] = quad->output.color[cbuf]; - unsigned j; - for (j = 0; j < QUAD_SIZE; j++) { - assert(quad->input.coverage[j] >= 0.0); - assert(quad->input.coverage[j] <= 1.0); - quadColor[3][j] *= quad->input.coverage[j]; - } - } + + for (i = 0; i < nr; i++) + coverage_quad( qs, quads[i] ); } - qs->next->run(qs->next, quad); + qs->next->run(qs->next, quads, nr); } - static void coverage_begin(struct quad_stage *qs) { qs->next->begin(qs->next); @@ -87,7 +101,7 @@ struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = coverage_begin; - stage->run = coverage_quad; + stage->run = coverage_run; stage->destroy = coverage_destroy; return stage; diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index 768b9275b3..8f223a7eae 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -49,7 +49,7 @@ * Try to effectively do that with codegen... */ -void +boolean sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) { struct softpipe_context *softpipe = qs->softpipe; @@ -193,6 +193,8 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) } quad->inout.mask &= zmask; + if (quad->inout.mask == 0) + return FALSE; if (softpipe->depth_stencil->depth.writemask) { @@ -252,16 +254,25 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) assert(0); } } + + return TRUE; } static void -depth_test_quad(struct quad_stage *qs, struct quad_header *quad) +depth_test_quads(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { - sp_depth_test_quad(qs, quad); + unsigned i, pass = 0; - if (quad->inout.mask) - qs->next->run(qs->next, quad); + for (i = 0; i < nr; i++) { + if (sp_depth_test_quad(qs, quads[i])) + quads[pass++] = quads[i]; + } + + if (pass) + qs->next->run(qs->next, quads, pass); } @@ -283,7 +294,7 @@ struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = depth_test_begin; - stage->run = depth_test_quad; + stage->run = depth_test_quads; stage->destroy = depth_test_destroy; return stage; diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c index 496fd39ed1..1048d44984 100644 --- a/src/gallium/drivers/softpipe/sp_quad_earlyz.c +++ b/src/gallium/drivers/softpipe/sp_quad_earlyz.c @@ -43,20 +43,26 @@ static void earlyz_quad( struct quad_stage *qs, - struct quad_header *quad ) + struct quad_header *quads[], + unsigned nr ) { - const float fx = (float) quad->input.x0; - const float fy = (float) quad->input.y0; - const float dzdx = quad->posCoef->dadx[2]; - const float dzdy = quad->posCoef->dady[2]; - const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy; + const float a0z = quads[0]->posCoef->a0[2]; + const float dzdx = quads[0]->posCoef->dadx[2]; + const float dzdy = quads[0]->posCoef->dady[2]; + unsigned i; - quad->output.depth[0] = z0; - quad->output.depth[1] = z0 + dzdx; - quad->output.depth[2] = z0 + dzdy; - quad->output.depth[3] = z0 + dzdx + dzdy; + for (i = 0; i < nr; i++) { + const float fx = (float) quads[i]->input.x0; + const float fy = (float) quads[i]->input.y0; + const float z0 = a0z + dzdx * fx + dzdy * fy; - qs->next->run( qs->next, quad ); + quads[i]->output.depth[0] = z0; + quads[i]->output.depth[1] = z0 + dzdx; + quads[i]->output.depth[2] = z0 + dzdy; + quads[i]->output.depth[3] = z0 + dzdx + dzdy; + } + + qs->next->run( qs->next, quads, nr ); } static void diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c index 28f8d1a60e..ea5ed3bbd0 100644 --- a/src/gallium/drivers/softpipe/sp_quad_fs.c +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -68,21 +68,18 @@ quad_shade_stage(struct quad_stage *qs) /** * Execute fragment shader for the four fragments in the quad. */ -static void +static boolean shade_quad(struct quad_stage *qs, struct quad_header *quad) { struct quad_shade_stage *qss = quad_shade_stage( qs ); struct softpipe_context *softpipe = qs->softpipe; struct tgsi_exec_machine *machine = qss->machine; boolean z_written; - - /* Consts do not require 16 byte alignment. */ - machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT]; - - machine->InterpCoefs = quad->coef; /* run shader */ quad->inout.mask &= softpipe->fs->run( softpipe->fs, machine, quad ); + if (quad->inout.mask == 0) + return FALSE; /* store outputs */ z_written = FALSE; @@ -129,11 +126,34 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) quad->output.depth[3] = z0 + dzdx + dzdy; } - /* shader may cull fragments */ - if (quad->inout.mask) { - qs->next->run( qs->next, quad ); + return TRUE; +} + +static void +shade_quads(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + struct quad_shade_stage *qss = quad_shade_stage( qs ); + struct softpipe_context *softpipe = qs->softpipe; + struct tgsi_exec_machine *machine = qss->machine; + + unsigned i, pass = 0; + + machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT]; + machine->InterpCoefs = quads[0]->coef; + + for (i = 0; i < nr; i++) { + if (shade_quad(qs, quads[i])) + quads[pass++] = quads[i]; } + + if (pass) + qs->next->run(qs->next, quads, pass); } + + + /** @@ -174,7 +194,7 @@ sp_quad_shade_stage( struct softpipe_context *softpipe ) qss->stage.softpipe = softpipe; qss->stage.begin = shade_begin; - qss->stage.run = shade_quad; + qss->stage.run = shade_quads; qss->stage.destroy = shade_destroy; qss->machine = tgsi_exec_machine_create(); diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/drivers/softpipe/sp_quad_occlusion.c index dfa7ff3b1d..4adeb16546 100644 --- a/src/gallium/drivers/softpipe/sp_quad_occlusion.c +++ b/src/gallium/drivers/softpipe/sp_quad_occlusion.c @@ -50,13 +50,15 @@ static unsigned count_bits( unsigned val ) } static void -occlusion_count_quad(struct quad_stage *qs, struct quad_header *quad) +occlusion_count_quads(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; + unsigned i; - softpipe->occlusion_count += count_bits(quad->inout.mask); + for (i = 0; i < nr; i++) + softpipe->occlusion_count += count_bits(quads[i]->inout.mask); - qs->next->run(qs->next, quad); + qs->next->run(qs->next, quads, nr); } @@ -78,7 +80,7 @@ struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = occlusion_begin; - stage->run = occlusion_count_quad; + stage->run = occlusion_count_quads; stage->destroy = occlusion_destroy; return stage; diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c index dd8f5377e9..79a222ff58 100644 --- a/src/gallium/drivers/softpipe/sp_quad_output.c +++ b/src/gallium/drivers/softpipe/sp_quad_output.c @@ -38,11 +38,8 @@ * taking mask into account. */ static void -output_quad(struct quad_stage *qs, struct quad_header *quad) +output_quad(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { - /* in-tile pos: */ - const int itx = quad->input.x0 % TILE_SIZE; - const int ity = quad->input.y0 % TILE_SIZE; struct softpipe_context *softpipe = qs->softpipe; uint cbuf; @@ -51,25 +48,35 @@ output_quad(struct quad_stage *qs, struct quad_header *quad) for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { struct softpipe_cached_tile *tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quad->input.x0, quad->input.y0); - float (*quadColor)[4] = quad->output.color[cbuf]; - int i, j; + quads[0]->input.x0, + quads[0]->input.y0); + int i, j, q; /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - if (quad->inout.mask & (1 << j)) { - int x = itx + (j & 1); - int y = ity + (j >> 1); - for (i = 0; i < 4; i++) { /* loop over color chans */ - tile->data.color[y][x][i] = quadColor[i][j]; - } - if (0) { - debug_printf("sp write pixel %d,%d: %g, %g, %g\n", - quad->input.x0 + x, - quad->input.y0 + y, - quadColor[0][j], - quadColor[1][j], - quadColor[2][j]); + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[cbuf]; + + /* in-tile pos: */ + const int itx = quad->input.x0 % TILE_SIZE; + const int ity = quad->input.y0 % TILE_SIZE; + + + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + if (0) { + debug_printf("sp write pixel %d,%d: %g, %g, %g\n", + quad->input.x0 + x, + quad->input.y0 + y, + quadColor[0][j], + quadColor[1][j], + quadColor[2][j]); + } } } } diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index 3a3359d303..594fade455 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -55,50 +55,52 @@ void sp_build_quad_pipeline(struct softpipe_context *sp) { boolean early_depth_test = - sp->depth_stencil->depth.enabled && - sp->framebuffer.zsbuf && - !sp->depth_stencil->alpha.enabled && - !sp->fs->info.uses_kill && - !sp->fs->info.writes_z; + sp->depth_stencil->depth.enabled && + sp->framebuffer.zsbuf && + !sp->depth_stencil->alpha.enabled && + !sp->fs->info.uses_kill && + !sp->fs->info.writes_z; /* build up the pipeline in reverse order... */ - sp->quad.first = sp->quad.output; - - if (sp->blend->colormask != 0xf) { - sp_push_quad_first( sp, sp->quad.colormask ); - } - - if (sp->blend->blend_enable || - sp->blend->logicop_enable) { - sp_push_quad_first( sp, sp->quad.blend ); - } - - if (sp->active_query_count) { - sp_push_quad_first( sp, sp->quad.occlusion ); - } - - if (sp->rasterizer->poly_smooth || - sp->rasterizer->line_smooth || - sp->rasterizer->point_smooth) { - sp_push_quad_first( sp, sp->quad.coverage ); - } - - if (!early_depth_test) { - sp_build_depth_stencil( sp ); - } - - if (sp->depth_stencil->alpha.enabled) { - sp_push_quad_first( sp, sp->quad.alpha_test ); - } - - /* XXX always enable shader? */ - if (1) { - sp_push_quad_first( sp, sp->quad.shade ); - } - - if (early_depth_test) { - sp_build_depth_stencil( sp ); - sp_push_quad_first( sp, sp->quad.earlyz ); - } + + /* Color combine + */ + sp->quad.first = sp->quad.output; + + if (sp->blend->colormask != 0xf) { + sp_push_quad_first( sp, sp->quad.colormask ); + } + + if (sp->blend->blend_enable || + sp->blend->logicop_enable) { + sp_push_quad_first( sp, sp->quad.blend ); + } + + if (sp->rasterizer->poly_smooth || + sp->rasterizer->line_smooth || + sp->rasterizer->point_smooth) { + sp_push_quad_first( sp, sp->quad.coverage ); + } + + /* Shade/Depth/Stencil/Alpha + */ + if (sp->active_query_count) { + sp_push_quad_first( sp, sp->quad.occlusion ); + } + + if (!early_depth_test) { + sp_build_depth_stencil( sp ); + } + + if (sp->depth_stencil->alpha.enabled) { + sp_push_quad_first( sp, sp->quad.alpha_test ); + } + + sp_push_quad_first( sp, sp->quad.shade ); + + if (early_depth_test) { + sp_build_depth_stencil( sp ); + sp_push_quad_first( sp, sp->quad.earlyz ); + } } diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.h b/src/gallium/drivers/softpipe/sp_quad_pipe.h index 0e40586ffc..add31ba705 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.h +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.h @@ -49,7 +49,7 @@ struct quad_stage { void (*begin)(struct quad_stage *qs); /** the stage action */ - void (*run)(struct quad_stage *qs, struct quad_header *quad); + void (*run)(struct quad_stage *qs, struct quad_header *quad[], unsigned nr); void (*destroy)(struct quad_stage *qs); }; @@ -69,6 +69,6 @@ struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe ); void sp_build_quad_pipeline(struct softpipe_context *sp); -void sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad); +boolean sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad); #endif /* SP_QUAD_PIPE_H */ diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c index 34a8d9e9f6..706dd2f756 100644 --- a/src/gallium/drivers/softpipe/sp_quad_stencil.c +++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c @@ -198,7 +198,8 @@ apply_stencil_op(ubyte stencilVals[QUAD_SIZE], * depth testing. */ static void -stencil_test_quad(struct quad_stage *qs, struct quad_header *quad) +stencil_test_quad(struct quad_stage *qs, struct quad_header *quads[], + unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; struct pipe_surface *ps = softpipe->framebuffer.zsbuf; @@ -206,9 +207,12 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad) ubyte ref, wrtMask, valMask; ubyte stencilVals[QUAD_SIZE]; struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); - uint j; - uint face = quad->input.facing; + = sp_get_cached_tile(softpipe->zsbuf_cache, + quads[0]->input.x0, + quads[0]->input.y0); + uint face = quads[0]->input.facing; + uint pass = 0; + uint j, q; if (!softpipe->depth_stencil->stencil[1].enabled) { /* single-sided stencil test, use front (face=0) state */ @@ -227,103 +231,110 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad) assert(ps); /* shouldn't get here if there's no stencil buffer */ - /* get stencil values from cached tile */ - switch (ps->format) { - case PIPE_FORMAT_S8Z24_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.depth32[y][x] >> 24; - } - break; - case PIPE_FORMAT_Z24S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.depth32[y][x] & 0xff; - } - break; - case PIPE_FORMAT_S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.stencil8[y][x]; + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + + /* get stencil values from cached tile */ + switch (ps->format) { + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.depth32[y][x] >> 24; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.depth32[y][x] & 0xff; + } + break; + case PIPE_FORMAT_S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.stencil8[y][x]; + } + break; + default: + assert(0); } - break; - default: - assert(0); - } - /* do the stencil test first */ - { - unsigned passMask, failMask; - passMask = do_stencil_test(stencilVals, func, ref, valMask); - failMask = quad->inout.mask & ~passMask; - quad->inout.mask &= passMask; + /* do the stencil test first */ + { + unsigned passMask, failMask; + passMask = do_stencil_test(stencilVals, func, ref, valMask); + failMask = quad->inout.mask & ~passMask; + quad->inout.mask &= passMask; - if (failOp != PIPE_STENCIL_OP_KEEP) { - apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask); + if (failOp != PIPE_STENCIL_OP_KEEP) { + apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask); + } } - } - if (quad->inout.mask) { - /* now the pixels that passed the stencil test are depth tested */ - if (softpipe->depth_stencil->depth.enabled) { - const unsigned origMask = quad->inout.mask; + if (quad->inout.mask) { + /* now the pixels that passed the stencil test are depth tested */ + if (softpipe->depth_stencil->depth.enabled) { + const unsigned origMask = quad->inout.mask; - sp_depth_test_quad(qs, quad); /* quad->mask is updated */ + sp_depth_test_quad(qs, quad); /* quad->mask is updated */ - /* update stencil buffer values according to z pass/fail result */ - if (zFailOp != PIPE_STENCIL_OP_KEEP) { - const unsigned failMask = origMask & ~quad->inout.mask; - apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask); - } + /* update stencil buffer values according to z pass/fail result */ + if (zFailOp != PIPE_STENCIL_OP_KEEP) { + const unsigned failMask = origMask & ~quad->inout.mask; + apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask); + } - if (zPassOp != PIPE_STENCIL_OP_KEEP) { - const unsigned passMask = origMask & quad->inout.mask; - apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask); + if (zPassOp != PIPE_STENCIL_OP_KEEP) { + const unsigned passMask = origMask & quad->inout.mask; + apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask); + } + } + else { + /* no depth test, apply Zpass operator to stencil buffer values */ + apply_stencil_op(stencilVals, quad->inout.mask, zPassOp, ref, wrtMask); } - } - else { - /* no depth test, apply Zpass operator to stencil buffer values */ - apply_stencil_op(stencilVals, quad->inout.mask, zPassOp, ref, wrtMask); - } - - } - /* put new stencil values into cached tile */ - switch (ps->format) { - case PIPE_FORMAT_S8Z24_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint s8z24 = tile->data.depth32[y][x]; - s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff); - tile->data.depth32[y][x] = s8z24; } - break; - case PIPE_FORMAT_Z24S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint z24s8 = tile->data.depth32[y][x]; - z24s8 = (z24s8 & 0xffffff00) | stencilVals[j]; - tile->data.depth32[y][x] = z24s8; - } - break; - case PIPE_FORMAT_S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - tile->data.stencil8[y][x] = stencilVals[j]; + + /* put new stencil values into cached tile */ + switch (ps->format) { + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + uint s8z24 = tile->data.depth32[y][x]; + s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff); + tile->data.depth32[y][x] = s8z24; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + uint z24s8 = tile->data.depth32[y][x]; + z24s8 = (z24s8 & 0xffffff00) | stencilVals[j]; + tile->data.depth32[y][x] = z24s8; + } + break; + case PIPE_FORMAT_S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.stencil8[y][x] = stencilVals[j]; + } + break; + default: + assert(0); } - break; - default: - assert(0); + + if (quad->inout.mask) + quads[pass++] = q; } - if (quad->inout.mask) - qs->next->run(qs->next, quad); + if (pass) + qs->next->run(qs->next, quads, pass); } diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c index 07162db7b6..05665d8e26 100644 --- a/src/gallium/drivers/softpipe/sp_quad_stipple.c +++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c @@ -14,40 +14,46 @@ * Apply polygon stipple to quads produced by triangle rasterization */ static void -stipple_quad(struct quad_stage *qs, struct quad_header *quad) +stipple_quad(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { static const uint bit31 = 1 << 31; static const uint bit30 = 1 << 30; + unsigned pass = nr; - if (quad->input.prim == QUAD_PRIM_TRI) { + if (quads[0]->input.prim == QUAD_PRIM_TRI) { struct softpipe_context *softpipe = qs->softpipe; - /* need to invert Y to index into OpenGL's stipple pattern */ - const int col0 = quad->input.x0 % 32; - const int y0 = quad->input.y0; - const int y1 = y0 + 1; - const uint stipple0 = softpipe->poly_stipple.stipple[y0 % 32]; - const uint stipple1 = softpipe->poly_stipple.stipple[y1 % 32]; + unsigned q; - /* turn off quad mask bits that fail the stipple test */ - if ((stipple0 & (bit31 >> col0)) == 0) - quad->inout.mask &= ~MASK_TOP_LEFT; + pass = 0; - if ((stipple0 & (bit30 >> col0)) == 0) - quad->inout.mask &= ~MASK_TOP_RIGHT; + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; - if ((stipple1 & (bit31 >> col0)) == 0) - quad->inout.mask &= ~MASK_BOTTOM_LEFT; + const int col0 = quad->input.x0 % 32; + const int y0 = quad->input.y0; + const int y1 = y0 + 1; + const uint stipple0 = softpipe->poly_stipple.stipple[y0 % 32]; + const uint stipple1 = softpipe->poly_stipple.stipple[y1 % 32]; - if ((stipple1 & (bit30 >> col0)) == 0) - quad->inout.mask &= ~MASK_BOTTOM_RIGHT; + /* turn off quad mask bits that fail the stipple test */ + if ((stipple0 & (bit31 >> col0)) == 0) + quad->inout.mask &= ~MASK_TOP_LEFT; - if (!quad->inout.mask) { - /* all fragments failed stipple test, end of quad pipeline */ - return; + if ((stipple0 & (bit30 >> col0)) == 0) + quad->inout.mask &= ~MASK_TOP_RIGHT; + + if ((stipple1 & (bit31 >> col0)) == 0) + quad->inout.mask &= ~MASK_BOTTOM_LEFT; + + if ((stipple1 & (bit30 >> col0)) == 0) + quad->inout.mask &= ~MASK_BOTTOM_RIGHT; + + if (quad->inout.mask) + quads[pass++] = quad; } } - qs->next->run(qs->next, quad); + qs->next->run(qs->next, quads, pass); } diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index eaf84ed9de..23dcae89c6 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -172,7 +172,7 @@ clip_emit_quad( struct setup_context *setup, struct quad_header *quad ) if (quad->inout.mask) { struct softpipe_context *sp = setup->softpipe; - sp->quad.first->run( sp->quad.first, quad ); + sp->quad.first->run( sp->quad.first, &quad, 1 ); } } @@ -193,7 +193,7 @@ emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) if (mask & 4) setup->numFragsEmitted++; if (mask & 8) setup->numFragsEmitted++; #endif - sp->quad.first->run( sp->quad.first, quad ); + sp->quad.first->run( sp->quad.first, &quad, 1 ); #if DEBUG_FRAGS mask = quad->inout.mask; if (mask & 1) setup->numFragsWritten++; -- cgit v1.2.3 From a1dbd7aa159e266592a1e52504680992327ca9e0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 18:17:05 +0100 Subject: softpipe: actually pass >1 quad from triangle routine First attempt --- src/gallium/drivers/softpipe/sp_context.h | 11 +- src/gallium/drivers/softpipe/sp_prim_vbuf.c | 2 + src/gallium/drivers/softpipe/sp_quad.h | 6 +- src/gallium/drivers/softpipe/sp_quad_blend.c | 9 +- src/gallium/drivers/softpipe/sp_quad_coverage.c | 10 +- src/gallium/drivers/softpipe/sp_quad_pipe.c | 6 +- src/gallium/drivers/softpipe/sp_quad_stipple.c | 44 +++--- src/gallium/drivers/softpipe/sp_setup.c | 186 ++++++++++++------------ 8 files changed, 135 insertions(+), 139 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index 414d903a37..153a648b0e 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -96,7 +96,16 @@ struct softpipe_context { /** Which vertex shader output slot contains point size */ int psize_slot; - unsigned reduced_api_prim; /**< PIPE_PRIM_POINTS, _LINES or _TRIANGLES */ + /* The reduced version of the primitive supplied by the state + * tracker. + */ + unsigned reduced_api_prim; + + /* The reduced primitive after unfilled triangles, wide-line + * decomposition, etc, are taken into account. This is the + * primitive actually rasterized. + */ + unsigned reduced_prim; /** Derived from scissor and surface bounds: */ struct pipe_scissor_state cliprect; diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c index 42021789ea..1dd63d99ff 100644 --- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c @@ -44,6 +44,7 @@ #include "draw/draw_context.h" #include "draw/draw_vbuf.h" #include "util/u_memory.h" +#include "util/u_prim.h" #define SP_MAX_VBUF_INDEXES 1024 @@ -167,6 +168,7 @@ sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim) setup_prepare( setup_ctx ); + cvbr->softpipe->reduced_prim = u_reduced_prim(prim); cvbr->prim = prim; return TRUE; diff --git a/src/gallium/drivers/softpipe/sp_quad.h b/src/gallium/drivers/softpipe/sp_quad.h index bd6c6cb912..a3236bd116 100644 --- a/src/gallium/drivers/softpipe/sp_quad.h +++ b/src/gallium/drivers/softpipe/sp_quad.h @@ -97,10 +97,10 @@ struct quad_header { struct quad_header_inout inout; struct quad_header_output output; - const struct tgsi_interp_coef *coef; + /* Redundant/duplicated: + */ const struct tgsi_interp_coef *posCoef; - - unsigned nr_attrs; + const struct tgsi_interp_coef *coef; }; #endif /* SP_QUAD_H */ diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index fdf1bb4552..8ef8666c0e 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -123,10 +123,6 @@ logicop_quad(struct quad_stage *qs, { struct softpipe_context *softpipe = qs->softpipe; uint cbuf; - struct softpipe_cached_tile * - tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quads[0]->input.x0, - quads[0]->input.y0); /* loop over colorbuffer outputs */ for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { @@ -137,6 +133,11 @@ logicop_quad(struct quad_stage *qs, uint *res4 = (uint *) res; uint i, j; + struct softpipe_cached_tile * + tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], + quads[0]->input.x0, + quads[0]->input.y0); + for (i = 0; i < nr; i++) { struct quad_header *quad = quads[i]; float (*quadColor)[4] = quad->output.color[cbuf]; diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c index eda5ce8e63..f06a385b3c 100644 --- a/src/gallium/drivers/softpipe/sp_quad_coverage.c +++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c @@ -69,16 +69,10 @@ coverage_run(struct quad_stage *qs, unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; - const uint prim = quads[0]->input.prim; unsigned i; - if ((softpipe->rasterizer->poly_smooth && prim == QUAD_PRIM_TRI) || - (softpipe->rasterizer->line_smooth && prim == QUAD_PRIM_LINE) || - (softpipe->rasterizer->point_smooth && prim == QUAD_PRIM_POINT)) { - - for (i = 0; i < nr; i++) - coverage_quad( qs, quads[i] ); - } + for (i = 0; i < nr; i++) + coverage_quad( qs, quads[i] ); qs->next->run(qs->next, quads, nr); } diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index 594fade455..6fae7d552f 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -76,9 +76,9 @@ sp_build_quad_pipeline(struct softpipe_context *sp) sp_push_quad_first( sp, sp->quad.blend ); } - if (sp->rasterizer->poly_smooth || - sp->rasterizer->line_smooth || - sp->rasterizer->point_smooth) { + if ((sp->rasterizer->poly_smooth && sp->reduced_prim == PIPE_PRIM_TRIANGLES) || + (sp->rasterizer->line_smooth && sp->reduced_prim == PIPE_PRIM_LINES) || + (sp->rasterizer->point_smooth && sp->reduced_prim == PIPE_PRIM_POINTS)) { sp_push_quad_first( sp, sp->quad.coverage ); } diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c index 05665d8e26..a0527a596a 100644 --- a/src/gallium/drivers/softpipe/sp_quad_stipple.c +++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c @@ -20,37 +20,35 @@ stipple_quad(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) static const uint bit30 = 1 << 30; unsigned pass = nr; - if (quads[0]->input.prim == QUAD_PRIM_TRI) { - struct softpipe_context *softpipe = qs->softpipe; - unsigned q; + struct softpipe_context *softpipe = qs->softpipe; + unsigned q; - pass = 0; + pass = 0; - for (q = 0; q < nr; q++) { - struct quad_header *quad = quads[q]; + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; - const int col0 = quad->input.x0 % 32; - const int y0 = quad->input.y0; - const int y1 = y0 + 1; - const uint stipple0 = softpipe->poly_stipple.stipple[y0 % 32]; - const uint stipple1 = softpipe->poly_stipple.stipple[y1 % 32]; + const int col0 = quad->input.x0 % 32; + const int y0 = quad->input.y0; + const int y1 = y0 + 1; + const uint stipple0 = softpipe->poly_stipple.stipple[y0 % 32]; + const uint stipple1 = softpipe->poly_stipple.stipple[y1 % 32]; - /* turn off quad mask bits that fail the stipple test */ - if ((stipple0 & (bit31 >> col0)) == 0) - quad->inout.mask &= ~MASK_TOP_LEFT; + /* turn off quad mask bits that fail the stipple test */ + if ((stipple0 & (bit31 >> col0)) == 0) + quad->inout.mask &= ~MASK_TOP_LEFT; - if ((stipple0 & (bit30 >> col0)) == 0) - quad->inout.mask &= ~MASK_TOP_RIGHT; + if ((stipple0 & (bit30 >> col0)) == 0) + quad->inout.mask &= ~MASK_TOP_RIGHT; - if ((stipple1 & (bit31 >> col0)) == 0) - quad->inout.mask &= ~MASK_BOTTOM_LEFT; + if ((stipple1 & (bit31 >> col0)) == 0) + quad->inout.mask &= ~MASK_BOTTOM_LEFT; - if ((stipple1 & (bit30 >> col0)) == 0) - quad->inout.mask &= ~MASK_BOTTOM_RIGHT; + if ((stipple1 & (bit30 >> col0)) == 0) + quad->inout.mask &= ~MASK_BOTTOM_RIGHT; - if (quad->inout.mask) - quads[pass++] = quad; - } + if (quad->inout.mask) + quads[pass++] = quad; } qs->next->run(qs->next, quads, pass); diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index 23dcae89c6..a132911c99 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -62,6 +62,8 @@ struct edge { }; +#define MAX_QUADS 16 + /** * Triangle setup info (derived from draw_stage). @@ -84,10 +86,14 @@ struct setup_context { struct edge emaj; float oneoverarea; + int facing; + + struct quad_header quad[MAX_QUADS]; + struct quad_header *quad_ptrs[MAX_QUADS]; + unsigned count; struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; struct tgsi_interp_coef posCoef; /* For Z, W */ - struct quad_header quad; struct { int left[2]; /**< [0] = row0, [1] = row1 */ @@ -176,41 +182,6 @@ clip_emit_quad( struct setup_context *setup, struct quad_header *quad ) } } -/** - * Emit a quad (pass to next stage). No clipping is done. - */ -static INLINE void -emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) -{ - struct softpipe_context *sp = setup->softpipe; -#if DEBUG_FRAGS - uint mask = quad->inout.mask; -#endif - -#if DEBUG_FRAGS - if (mask & 1) setup->numFragsEmitted++; - if (mask & 2) setup->numFragsEmitted++; - if (mask & 4) setup->numFragsEmitted++; - if (mask & 8) setup->numFragsEmitted++; -#endif - sp->quad.first->run( sp->quad.first, &quad, 1 ); -#if DEBUG_FRAGS - mask = quad->inout.mask; - if (mask & 1) setup->numFragsWritten++; - if (mask & 2) setup->numFragsWritten++; - if (mask & 4) setup->numFragsWritten++; - if (mask & 8) setup->numFragsWritten++; -#endif -} - - -#define EMIT_QUAD(setup,x,y,qmask) \ -do { \ - setup->quad.input.x0 = x; \ - setup->quad.input.y0 = y; \ - setup->quad.inout.mask = qmask; \ - emit_quad( setup, &setup->quad, 0 ); \ -} while (0) /** @@ -219,7 +190,12 @@ do { \ */ static INLINE int block( int x ) { - return x & ~1; + return x & ~(2-1); +} + +static INLINE int block_x( int x ) +{ + return x & ~(16-1); } @@ -228,13 +204,15 @@ static INLINE int block( int x ) */ static void flush_spans( struct setup_context *setup ) { - const int step = 30; + const int step = 16; const int xleft0 = setup->span.left[0]; const int xleft1 = setup->span.left[1]; const int xright0 = setup->span.right[0]; const int xright1 = setup->span.right[1]; + struct quad_stage *pipe = setup->softpipe->quad.first; + - int minleft = block(MIN2(xleft0, xleft1)); + int minleft = block_x(MIN2(xleft0, xleft1)); int maxright = MAX2(xright0, xright1); int x; @@ -244,7 +222,8 @@ static void flush_spans( struct setup_context *setup ) unsigned skip_right0 = CLAMP(x + step - xright0, 0, step); unsigned skip_right1 = CLAMP(x + step - xright1, 0, step); unsigned lx = x; - + unsigned q = 0; + unsigned skipmask_left0 = (1U << skip_left0) - 1U; unsigned skipmask_left1 = (1U << skip_left1) - 1U; @@ -256,13 +235,22 @@ static void flush_spans( struct setup_context *setup ) unsigned mask0 = ~skipmask_left0 & ~skipmask_right0; unsigned mask1 = ~skipmask_left1 & ~skipmask_right1; - while (mask0 | mask1) { - unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2); - if (quadmask) - EMIT_QUAD( setup, lx, setup->span.y, quadmask ); - mask0 >>= 2; - mask1 >>= 2; - lx += 2; + if (mask0 | mask1) { + do { + unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2); + if (quadmask) { + setup->quad[q].input.x0 = lx; + setup->quad[q].input.y0 = setup->span.y; + setup->quad[q].inout.mask = quadmask; + setup->quad_ptrs[q] = &setup->quad[q]; + q++; + } + mask0 >>= 2; + mask1 >>= 2; + lx += 2; + } while (mask0 | mask1); + + pipe->run( pipe, setup->quad_ptrs, q ); } } @@ -281,7 +269,7 @@ static void print_vertex(const struct setup_context *setup, { int i; debug_printf(" Vertex: (%p)\n", v); - for (i = 0; i < setup->quad.nr_attrs; i++) { + for (i = 0; i < setup->quad[0].nr_attrs; i++) { debug_printf(" %d: %f %f %f %f\n", i, v[i][0], v[i][1], v[i][2], v[i][3]); if (util_is_inf_or_nan(v[i][0])) { @@ -386,7 +374,9 @@ static boolean setup_sort_vertices( struct setup_context *setup, * - the GLSL gl_FrontFacing fragment attribute (bool) * - two-sided stencil test */ - setup->quad.input.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW); + setup->facing = + ((det > 0.0) ^ + (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW)); return TRUE; } @@ -573,7 +563,7 @@ static void setup_tri_coefficients( struct setup_context *setup ) } if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { - setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing; + setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef[fragSlot].dadx[0] = 0.0; setup->coef[fragSlot].dady[0] = 0.0; } @@ -741,7 +731,7 @@ void setup_tri( struct setup_context *setup, setup_tri_coefficients( setup ); setup_tri_edges( setup ); - setup->quad.input.prim = QUAD_PRIM_TRI; + assert(setup->softpipe->reduced_prim == PIPE_PRIM_TRIANGLES); setup->span.y = 0; setup->span.right[0] = 0; @@ -881,7 +871,7 @@ setup_line_coefficients(struct setup_context *setup, } if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { - setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing; + setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef[fragSlot].dadx[0] = 0.0; setup->coef[fragSlot].dady[0] = 0.0; } @@ -902,20 +892,20 @@ plot(struct setup_context *setup, int x, int y) const int quadY = y - iy; const int mask = (1 << ix) << (2 * iy); - if (quadX != setup->quad.input.x0 || - quadY != setup->quad.input.y0) + if (quadX != setup->quad[0].input.x0 || + quadY != setup->quad[0].input.y0) { /* flush prev quad, start new quad */ - if (setup->quad.input.x0 != -1) - clip_emit_quad( setup, &setup->quad ); + if (setup->quad[0].input.x0 != -1) + clip_emit_quad( setup, &setup->quad[0] ); - setup->quad.input.x0 = quadX; - setup->quad.input.y0 = quadY; - setup->quad.inout.mask = 0x0; + setup->quad[0].input.x0 = quadX; + setup->quad[0].input.y0 = quadY; + setup->quad[0].inout.mask = 0x0; } - setup->quad.inout.mask |= mask; + setup->quad[0].inout.mask |= mask; } @@ -975,17 +965,18 @@ setup_line(struct setup_context *setup, assert(dx >= 0); assert(dy >= 0); + assert(setup->softpipe->reduced_prim == PIPE_PRIM_LINES); + + setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1; + setup->quad[0].inout.mask = 0x0; - setup->quad.input.x0 = setup->quad.input.y0 = -1; - setup->quad.inout.mask = 0x0; - setup->quad.input.prim = QUAD_PRIM_LINE; /* XXX temporary: set coverage to 1.0 so the line appears * if AA mode happens to be enabled. */ - setup->quad.input.coverage[0] = - setup->quad.input.coverage[1] = - setup->quad.input.coverage[2] = - setup->quad.input.coverage[3] = 1.0; + setup->quad[0].input.coverage[0] = + setup->quad[0].input.coverage[1] = + setup->quad[0].input.coverage[2] = + setup->quad[0].input.coverage[3] = 1.0; if (dx > dy) { /*** X-major line ***/ @@ -1029,8 +1020,8 @@ setup_line(struct setup_context *setup, } /* draw final quad */ - if (setup->quad.inout.mask) { - clip_emit_quad( setup, &setup->quad ); + if (setup->quad[0].inout.mask) { + clip_emit_quad( setup, &setup->quad[0] ); } } @@ -1078,6 +1069,8 @@ setup_point( struct setup_context *setup, if (softpipe->no_rast) return; + assert(setup->softpipe->reduced_prim == PIPE_PRIM_POINTS); + /* For points, all interpolants are constant-valued. * However, for point sprites, we'll need to setup texcoords appropriately. * XXX: which coefficients are the texcoords??? @@ -1124,22 +1117,21 @@ setup_point( struct setup_context *setup, } if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { - setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing; + setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef[fragSlot].dadx[0] = 0.0; setup->coef[fragSlot].dady[0] = 0.0; } } - setup->quad.input.prim = QUAD_PRIM_POINT; if (halfSize <= 0.5 && !round) { /* special case for 1-pixel points */ const int ix = ((int) x) & 1; const int iy = ((int) y) & 1; - setup->quad.input.x0 = (int) x - ix; - setup->quad.input.y0 = (int) y - iy; - setup->quad.inout.mask = (1 << ix) << (2 * iy); - clip_emit_quad( setup, &setup->quad ); + setup->quad[0].input.x0 = (int) x - ix; + setup->quad[0].input.y0 = (int) y - iy; + setup->quad[0].inout.mask = (1 << ix) << (2 * iy); + clip_emit_quad( setup, &setup->quad[0] ); } else { if (round) { @@ -1159,15 +1151,15 @@ setup_point( struct setup_context *setup, for (ix = ixmin; ix <= ixmax; ix += 2) { float dx, dy, dist2, cover; - setup->quad.inout.mask = 0x0; + setup->quad[0].inout.mask = 0x0; dx = (ix + 0.5f) - x; dy = (iy + 0.5f) - y; dist2 = dx * dx + dy * dy; if (dist2 <= rmax2) { cover = 1.0F - (dist2 - rmin2) * cscale; - setup->quad.input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f); - setup->quad.inout.mask |= MASK_TOP_LEFT; + setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_TOP_LEFT; } dx = (ix + 1.5f) - x; @@ -1175,8 +1167,8 @@ setup_point( struct setup_context *setup, dist2 = dx * dx + dy * dy; if (dist2 <= rmax2) { cover = 1.0F - (dist2 - rmin2) * cscale; - setup->quad.input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f); - setup->quad.inout.mask |= MASK_TOP_RIGHT; + setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_TOP_RIGHT; } dx = (ix + 0.5f) - x; @@ -1184,8 +1176,8 @@ setup_point( struct setup_context *setup, dist2 = dx * dx + dy * dy; if (dist2 <= rmax2) { cover = 1.0F - (dist2 - rmin2) * cscale; - setup->quad.input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f); - setup->quad.inout.mask |= MASK_BOTTOM_LEFT; + setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT; } dx = (ix + 1.5f) - x; @@ -1193,14 +1185,14 @@ setup_point( struct setup_context *setup, dist2 = dx * dx + dy * dy; if (dist2 <= rmax2) { cover = 1.0F - (dist2 - rmin2) * cscale; - setup->quad.input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f); - setup->quad.inout.mask |= MASK_BOTTOM_RIGHT; + setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f); + setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT; } - if (setup->quad.inout.mask) { - setup->quad.input.x0 = ix; - setup->quad.input.y0 = iy; - clip_emit_quad( setup, &setup->quad ); + if (setup->quad[0].inout.mask) { + setup->quad[0].input.x0 = ix; + setup->quad[0].input.y0 = iy; + clip_emit_quad( setup, &setup->quad[0] ); } } } @@ -1244,10 +1236,10 @@ setup_point( struct setup_context *setup, mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); } - setup->quad.inout.mask = mask; - setup->quad.input.x0 = ix; - setup->quad.input.y0 = iy; - clip_emit_quad( setup, &setup->quad ); + setup->quad[0].inout.mask = mask; + setup->quad[0].input.x0 = ix; + setup->quad[0].input.y0 = iy; + clip_emit_quad( setup, &setup->quad[0] ); } } } @@ -1262,9 +1254,6 @@ void setup_prepare( struct setup_context *setup ) softpipe_update_derived(sp); } - /* Note: nr_attrs is only used for debugging (vertex printing) */ - setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw); - sp->quad.first->begin( sp->quad.first ); if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES && @@ -1293,11 +1282,14 @@ void setup_destroy_context( struct setup_context *setup ) struct setup_context *setup_create_context( struct softpipe_context *softpipe ) { struct setup_context *setup = CALLOC_STRUCT(setup_context); + unsigned i; setup->softpipe = softpipe; - setup->quad.coef = setup->coef; - setup->quad.posCoef = &setup->posCoef; + for (i = 0; i < MAX_QUADS; i++) { + setup->quad[i].coef = setup->coef; + setup->quad[i].posCoef = &setup->posCoef; + } setup->span.left[0] = 1000000; /* greater than right[0] */ setup->span.left[1] = 1000000; /* greater than right[1] */ -- cgit v1.2.3 From 333ec94380af502b1c492f61dcc1897bcf43a96c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 18:46:17 +0100 Subject: softpipe: example fastpaths in blending --- src/gallium/drivers/softpipe/sp_quad_blend.c | 132 ++++++++++++++++++++++++--- 1 file changed, 121 insertions(+), 11 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index 8ef8666c0e..1ef7529cff 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -117,9 +117,9 @@ do { \ static void -logicop_quad(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +logicop_quads(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; uint cbuf; @@ -241,13 +241,102 @@ logicop_quad(struct quad_stage *qs, } } } + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quads, nr); +} + +static void +blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + static const float one[4] = { 1, 1, 1, 1 }; + float one_minus_alpha[QUAD_SIZE]; + float dest[4][QUAD_SIZE]; + float source[4][QUAD_SIZE]; + uint i, j, q; + + struct softpipe_cached_tile *tile + = sp_get_cached_tile(qs->softpipe->cbuf_cache[0], + quads[0]->input.x0, + quads[0]->input.y0); + + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[0]; + const float *alpha = quadColor[3]; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } + + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + VEC4_MUL(source[3], quadColor[3], alpha); /* A */ + + VEC4_SUB(one_minus_alpha, one, alpha); + VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ + VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ + VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ + VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* B */ + + VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ + VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ + } + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quads, nr); } +static void +blend_single_add_one_one(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + float dest[4][QUAD_SIZE]; + uint i, j, q; + + struct softpipe_cached_tile *tile + = sp_get_cached_tile(qs->softpipe->cbuf_cache[0], + quads[0]->input.x0, + quads[0]->input.y0); + + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[0]; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } + + VEC4_ADD_SAT(quadColor[0], quadColor[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */ + VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */ + } + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quads, nr); +} static void -blend_quads(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +blend_quads_fallback(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { static const float zero[4] = { 0, 0, 0, 0 }; static const float one[4] = { 1, 1, 1, 1 }; @@ -747,6 +836,9 @@ blend_quads(struct quad_stage *qs, } } } /* cbuf loop */ + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quads, nr); } @@ -756,21 +848,39 @@ blend_quad(struct quad_stage *qs, unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; + const struct pipe_blend_state *blend = softpipe->blend; if (softpipe->blend->logicop_enable) { - logicop_quad(qs, quads, nr); + qs->run = logicop_quads; } - else if (softpipe->blend->blend_enable) { - blend_quads(qs, quads, nr ); + else { + qs->run = blend_quads_fallback; + + if (blend->rgb_src_factor == blend->alpha_src_factor && + blend->rgb_dst_factor == blend->alpha_dst_factor && + blend->rgb_func == blend->alpha_func && + softpipe->framebuffer.nr_cbufs == 1) + { + if (blend->alpha_func == PIPE_BLEND_ADD) { + if (blend->rgb_src_factor == PIPE_BLENDFACTOR_ONE && + blend->rgb_dst_factor == PIPE_BLENDFACTOR_ONE) { + qs->run = blend_single_add_one_one; + } + else if (blend->rgb_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA && + blend->rgb_dst_factor == PIPE_BLENDFACTOR_INV_SRC_ALPHA) + qs->run = blend_single_add_src_alpha_inv_src_alpha; + + } + } } - /* pass blended quad to next stage */ - qs->next->run(qs->next, quads, nr); + qs->run(qs, quads, nr); } static void blend_begin(struct quad_stage *qs) { + qs->run = blend_quad; qs->next->begin(qs->next); } -- cgit v1.2.3 From 42f1757189ba965e6d917d1124d0d6cf78b19a70 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 20:18:52 +0100 Subject: softpipe: fix typo --- src/gallium/drivers/softpipe/sp_quad_stencil.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c index 706dd2f756..d9ee80e59a 100644 --- a/src/gallium/drivers/softpipe/sp_quad_stencil.c +++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c @@ -330,7 +330,7 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quads[], } if (quad->inout.mask) - quads[pass++] = q; + quads[pass++] = quad; } if (pass) -- cgit v1.2.3 From a2f7ab1d155da52c689f7c6390c233e4eae44643 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 20:19:18 +0100 Subject: softpipe: move all color-combine code into sp_quad_blend.c Consolidate the read-modify-write color combining code from the blend, colormask and output stages. --- src/gallium/drivers/softpipe/Makefile | 14 +- src/gallium/drivers/softpipe/SConscript | 2 - src/gallium/drivers/softpipe/sp_context.c | 4 - src/gallium/drivers/softpipe/sp_context.h | 2 - src/gallium/drivers/softpipe/sp_quad_blend.c | 1352 ++++++++++++---------- src/gallium/drivers/softpipe/sp_quad_bufloop.c | 74 -- src/gallium/drivers/softpipe/sp_quad_colormask.c | 126 -- src/gallium/drivers/softpipe/sp_quad_coverage.c | 1 - src/gallium/drivers/softpipe/sp_quad_output.c | 109 -- src/gallium/drivers/softpipe/sp_quad_pipe.c | 15 +- 10 files changed, 727 insertions(+), 972 deletions(-) delete mode 100644 src/gallium/drivers/softpipe/sp_quad_bufloop.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_colormask.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_output.c diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile index 516e3992fd..bdc1a5819f 100644 --- a/src/gallium/drivers/softpipe/Makefile +++ b/src/gallium/drivers/softpipe/Makefile @@ -15,17 +15,15 @@ C_SOURCES = \ sp_prim_setup.c \ sp_prim_vbuf.c \ sp_quad_pipe.c \ - sp_quad_alpha_test.c \ - sp_quad_blend.c \ - sp_quad_colormask.c \ - sp_quad_coverage.c \ - sp_quad_depth_test.c \ + sp_quad_stipple.c \ sp_quad_earlyz.c \ + sp_quad_depth_test.c \ + sp_quad_stencil.c \ sp_quad_fs.c \ + sp_quad_alpha_test.c \ sp_quad_occlusion.c \ - sp_quad_output.c \ - sp_quad_stencil.c \ - sp_quad_stipple.c \ + sp_quad_coverage.c \ + sp_quad_blend.c \ sp_screen.c \ sp_setup.c \ sp_state_blend.c \ diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript index f8720638a7..dcc25732ba 100644 --- a/src/gallium/drivers/softpipe/SConscript +++ b/src/gallium/drivers/softpipe/SConscript @@ -18,13 +18,11 @@ softpipe = env.ConvenienceLibrary( 'sp_quad_alpha_test.c', 'sp_quad_blend.c', 'sp_quad_pipe.c', - 'sp_quad_colormask.c', 'sp_quad_coverage.c', 'sp_quad_depth_test.c', 'sp_quad_earlyz.c', 'sp_quad_fs.c', 'sp_quad_occlusion.c', - 'sp_quad_output.c', 'sp_quad_stencil.c', 'sp_quad_stipple.c', 'sp_query.c', diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 4418ef0ff4..28a0dd62ac 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -97,8 +97,6 @@ static void softpipe_destroy( struct pipe_context *pipe ) softpipe->quad.occlusion->destroy( softpipe->quad.occlusion ); softpipe->quad.coverage->destroy( softpipe->quad.coverage ); softpipe->quad.blend->destroy( softpipe->quad.blend ); - softpipe->quad.colormask->destroy( softpipe->quad.colormask ); - softpipe->quad.output->destroy( softpipe->quad.output ); for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) sp_destroy_tile_cache(softpipe->cbuf_cache[i]); @@ -241,8 +239,6 @@ softpipe_create( struct pipe_screen *screen ) softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe); softpipe->quad.coverage = sp_quad_coverage_stage(softpipe); softpipe->quad.blend = sp_quad_blend_stage(softpipe); - softpipe->quad.colormask = sp_quad_colormask_stage(softpipe); - softpipe->quad.output = sp_quad_output_stage(softpipe); /* vertex shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index 153a648b0e..b76ff610a3 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -123,8 +123,6 @@ struct softpipe_context { struct quad_stage *occlusion; struct quad_stage *coverage; struct quad_stage *blend; - struct quad_stage *colormask; - struct quad_stage *output; struct quad_stage *first; /**< points to one of the above stages */ } quad; diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index 1ef7529cff..e1f0e77255 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -117,135 +117,677 @@ do { \ static void -logicop_quads(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +logicop_quad(struct quad_stage *qs, + float (*quadColor)[4], + float (*dest)[4]) { struct softpipe_context *softpipe = qs->softpipe; - uint cbuf; + ubyte src[4][4], dst[4][4], res[4][4]; + uint *src4 = (uint *) src; + uint *dst4 = (uint *) dst; + uint *res4 = (uint *) res; + uint j; + + + /* convert to ubyte */ + for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */ + dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */ + dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */ + dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */ + dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */ + + src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */ + src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */ + src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */ + src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */ + } + + switch (softpipe->blend->logicop_func) { + case PIPE_LOGICOP_CLEAR: + for (j = 0; j < 4; j++) + res4[j] = 0; + break; + case PIPE_LOGICOP_NOR: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] | dst4[j]); + break; + case PIPE_LOGICOP_AND_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_COPY_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j]; + break; + case PIPE_LOGICOP_AND_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & ~dst4[j]; + break; + case PIPE_LOGICOP_INVERT: + for (j = 0; j < 4; j++) + res4[j] = ~dst4[j]; + break; + case PIPE_LOGICOP_XOR: + for (j = 0; j < 4; j++) + res4[j] = dst4[j] ^ src4[j]; + break; + case PIPE_LOGICOP_NAND: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] & dst4[j]); + break; + case PIPE_LOGICOP_AND: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_EQUIV: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] ^ dst4[j]); + break; + case PIPE_LOGICOP_NOOP: + for (j = 0; j < 4; j++) + res4[j] = dst4[j]; + break; + case PIPE_LOGICOP_OR_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_COPY: + for (j = 0; j < 4; j++) + res4[j] = src4[j]; + break; + case PIPE_LOGICOP_OR_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | ~dst4[j]; + break; + case PIPE_LOGICOP_OR: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_SET: + for (j = 0; j < 4; j++) + res4[j] = ~0; + break; + default: + assert(0); + } + + for (j = 0; j < 4; j++) { + quadColor[j][0] = ubyte_to_float(res[j][0]); + quadColor[j][1] = ubyte_to_float(res[j][1]); + quadColor[j][2] = ubyte_to_float(res[j][2]); + quadColor[j][3] = ubyte_to_float(res[j][3]); + } +} + + + +static void +blend_quad(struct quad_stage *qs, + float (*quadColor)[4], + float (*dest)[4]) +{ + static const float zero[4] = { 0, 0, 0, 0 }; + static const float one[4] = { 1, 1, 1, 1 }; + struct softpipe_context *softpipe = qs->softpipe; + float source[4][QUAD_SIZE]; + + /* + * Compute src/first term RGB + */ + switch (softpipe->blend->rgb_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[0], quadColor[0]); /* R */ + VEC4_COPY(source[1], quadColor[1]); /* G */ + VEC4_COPY(source[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const float *alpha = quadColor[3]; + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + { + const float *alpha = dest[3]; + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + { + const float *alpha = quadColor[3]; + float diff[4], temp[4]; + VEC4_SUB(diff, one, dest[3]); + VEC4_MIN(temp, alpha, diff); + VEC4_MUL(source[0], quadColor[0], temp); /* R */ + VEC4_MUL(source[1], quadColor[1], temp); /* G */ + VEC4_MUL(source[2], quadColor[2], temp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], comp); /* R */ + VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], comp); /* G */ + VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float alpha[4]; + VEC4_SCALAR(alpha, softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[0], zero); /* R */ + VEC4_COPY(source[1], zero); /* G */ + VEC4_COPY(source[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, quadColor[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, dest[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, dest[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, dest[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + { + float inv_comp[4]; + /* R */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); + VEC4_MUL(source[0], quadColor[0], inv_comp); + /* G */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); + VEC4_MUL(source[1], quadColor[1], inv_comp); + /* B */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); + VEC4_MUL(source[2], quadColor[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_alpha[4]; + VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + assert(0); /* to do */ + break; + default: + assert(0); + } + + /* + * Compute src/first term A + */ + switch (softpipe->blend->alpha_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[3], quadColor[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const float *alpha = quadColor[3]; + VEC4_MUL(source[3], quadColor[3], alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + /* multiply alpha by 1.0 */ + VEC4_COPY(source[3], quadColor[3]); /* A */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(source[3], quadColor[3], comp); /* A */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, quadColor[3]); + VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, dest[3]); + VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + /* A */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(source[3], quadColor[3], inv_comp); + } + break; + default: + assert(0); + } + - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { + /* + * Compute dest/second term RGB + */ + switch (softpipe->blend->rgb_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */ + VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */ + VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(dest[0], dest[0], dest[0]); /* R */ + VEC4_MUL(dest[1], dest[1], dest[1]); /* G */ + VEC4_MUL(dest[2], dest[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ + VEC4_MUL(dest[0], dest[0], comp); /* R */ + VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ + VEC4_MUL(dest[1], dest[1], comp); /* G */ + VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ + VEC4_MUL(dest[2], dest[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(dest[0], dest[0], comp); /* R */ + VEC4_MUL(dest[1], dest[1], comp); /* G */ + VEC4_MUL(dest[2], dest[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[0], zero); /* R */ + VEC4_COPY(dest[1], zero); /* G */ + VEC4_COPY(dest[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + case PIPE_BLENDFACTOR_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ + VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ + VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ + VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ + VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ + VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quadColor[3]); + VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ + VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ + VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[3]); /* A */ + VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ + VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ + VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[0]); /* R */ + VEC4_MUL(dest[0], dest[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, dest[1]); /* G */ + VEC4_MUL(dest[1], dest[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, dest[2]); /* B */ + VEC4_MUL(dest[2], dest[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + { + float inv_comp[4]; + /* R */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); + VEC4_MUL(dest[0], dest[0], inv_comp); + /* G */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); + VEC4_MUL(dest[1], dest[1], inv_comp); + /* B */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); + VEC4_MUL(dest[2], dest[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(dest[0], dest[0], inv_comp); + VEC4_MUL(dest[1], dest[1], inv_comp); + VEC4_MUL(dest[2], dest[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + default: + assert(0); + } + + /* + * Compute dest/second term A + */ + switch (softpipe->blend->alpha_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[3], dest[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(dest[3], dest[3], comp); /* A */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quadColor[3]); + VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[3]); /* A */ + VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(dest[3], dest[3], inv_comp); + } + break; + default: + assert(0); + } + + /* + * Combine RGB terms + */ + switch (softpipe->blend->rgb_func) { + case PIPE_BLEND_ADD: + VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */ + VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */ + VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */ + break; + default: + assert(0); + } + + /* + * Combine A terms + */ + switch (softpipe->blend->alpha_func) { + case PIPE_BLEND_ADD: + VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */ + break; + default: + assert(0); + } +} + +static void +colormask_quad(struct quad_stage *qs, + float (*quadColor)[4], + float (*dest)[4]) +{ + struct softpipe_context *softpipe = qs->softpipe; + + /* R */ + if (!(softpipe->blend->colormask & PIPE_MASK_R)) + COPY_4V(quadColor[0], dest[0]); + + /* G */ + if (!(softpipe->blend->colormask & PIPE_MASK_G)) + COPY_4V(quadColor[1], dest[1]); + + /* B */ + if (!(softpipe->blend->colormask & PIPE_MASK_B)) + COPY_4V(quadColor[2], dest[2]); + + /* A */ + if (!(softpipe->blend->colormask & PIPE_MASK_A)) + COPY_4V(quadColor[3], dest[3]); +} + + +static void +blend_fallback(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + struct softpipe_context *softpipe = qs->softpipe; + const struct pipe_blend_state *blend = softpipe->blend; + unsigned cbuf; + + for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) + { float dest[4][QUAD_SIZE]; - ubyte src[4][4], dst[4][4], res[4][4]; - uint *src4 = (uint *) src; - uint *dst4 = (uint *) dst; - uint *res4 = (uint *) res; - uint i, j; - - struct softpipe_cached_tile * - tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quads[0]->input.x0, - quads[0]->input.y0); - - for (i = 0; i < nr; i++) { - struct quad_header *quad = quads[i]; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], + quads[0]->input.x0, + quads[0]->input.y0); + uint q, i, j; + + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; float (*quadColor)[4] = quad->output.color[cbuf]; + const int itx = (quad->input.x0 & (TILE_SIZE-1)); + const int ity = (quad->input.y0 & (TILE_SIZE-1)); - /* get/swizzle dest colors */ + /* get/swizzle dest colors + */ for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + int x = itx + (j & 1); + int y = ity + (j >> 1); for (i = 0; i < 4; i++) { dest[i][j] = tile->data.color[y][x][i]; } } - /* convert to ubyte */ - for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */ - dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */ - dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */ - dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */ - dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */ - - src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */ - src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */ - src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */ - src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */ - } - switch (softpipe->blend->logicop_func) { - case PIPE_LOGICOP_CLEAR: - for (j = 0; j < 4; j++) - res4[j] = 0; - break; - case PIPE_LOGICOP_NOR: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] | dst4[j]); - break; - case PIPE_LOGICOP_AND_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j] & dst4[j]; - break; - case PIPE_LOGICOP_COPY_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j]; - break; - case PIPE_LOGICOP_AND_REVERSE: - for (j = 0; j < 4; j++) - res4[j] = src4[j] & ~dst4[j]; - break; - case PIPE_LOGICOP_INVERT: - for (j = 0; j < 4; j++) - res4[j] = ~dst4[j]; - break; - case PIPE_LOGICOP_XOR: - for (j = 0; j < 4; j++) - res4[j] = dst4[j] ^ src4[j]; - break; - case PIPE_LOGICOP_NAND: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] & dst4[j]); - break; - case PIPE_LOGICOP_AND: - for (j = 0; j < 4; j++) - res4[j] = src4[j] & dst4[j]; - break; - case PIPE_LOGICOP_EQUIV: - for (j = 0; j < 4; j++) - res4[j] = ~(src4[j] ^ dst4[j]); - break; - case PIPE_LOGICOP_NOOP: - for (j = 0; j < 4; j++) - res4[j] = dst4[j]; - break; - case PIPE_LOGICOP_OR_INVERTED: - for (j = 0; j < 4; j++) - res4[j] = ~src4[j] | dst4[j]; - break; - case PIPE_LOGICOP_COPY: - for (j = 0; j < 4; j++) - res4[j] = src4[j]; - break; - case PIPE_LOGICOP_OR_REVERSE: - for (j = 0; j < 4; j++) - res4[j] = src4[j] | ~dst4[j]; - break; - case PIPE_LOGICOP_OR: - for (j = 0; j < 4; j++) - res4[j] = src4[j] | dst4[j]; - break; - case PIPE_LOGICOP_SET: - for (j = 0; j < 4; j++) - res4[j] = ~0; - break; - default: - assert(0); + if (blend->logicop_enable) { + logicop_quad( qs, quadColor, dest ); + } + else if (blend->blend_enable) { + blend_quad( qs, quadColor, dest ); } - for (j = 0; j < 4; j++) { - quadColor[j][0] = ubyte_to_float(res[j][0]); - quadColor[j][1] = ubyte_to_float(res[j][1]); - quadColor[j][2] = ubyte_to_float(res[j][2]); - quadColor[j][3] = ubyte_to_float(res[j][3]); + if (blend->colormask != 0xf) + colormask_quad( qs, quadColor, dest ); + + /* Output color values + */ + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + } } } } - - /* pass blended quad to next stage */ - qs->next->run(qs->next, quads, nr); } + static void blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, struct quad_header *quads[], @@ -266,11 +808,13 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, struct quad_header *quad = quads[q]; float (*quadColor)[4] = quad->output.color[0]; const float *alpha = quadColor[3]; + const int itx = (quad->input.x0 & (TILE_SIZE-1)); + const int ity = (quad->input.y0 & (TILE_SIZE-1)); /* get/swizzle dest colors */ for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + int x = itx + (j & 1); + int y = ity + (j >> 1); for (i = 0; i < 4; i++) { dest[i][j] = tile->data.color[y][x][i]; } @@ -291,10 +835,17 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ - } - /* pass blended quad to next stage */ - qs->next->run(qs->next, quads, nr); + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + } + } + } } static void @@ -313,11 +864,13 @@ blend_single_add_one_one(struct quad_stage *qs, for (q = 0; q < nr; q++) { struct quad_header *quad = quads[q]; float (*quadColor)[4] = quad->output.color[0]; + const int itx = (quad->input.x0 & (TILE_SIZE-1)); + const int ity = (quad->input.y0 & (TILE_SIZE-1)); /* get/swizzle dest colors */ for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); + int x = itx + (j & 1); + int y = ity + (j >> 1); for (i = 0; i < 4; i++) { dest[i][j] = tile->data.color[y][x][i]; } @@ -327,539 +880,71 @@ blend_single_add_one_one(struct quad_stage *qs, VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */ VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */ VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */ - } - /* pass blended quad to next stage */ - qs->next->run(qs->next, quads, nr); + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + } + } + } } + static void -blend_quads_fallback(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +single_output_color(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { - static const float zero[4] = { 0, 0, 0, 0 }; - static const float one[4] = { 1, 1, 1, 1 }; - struct softpipe_context *softpipe = qs->softpipe; - uint cbuf; - - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { - float source[4][QUAD_SIZE], dest[4][QUAD_SIZE]; - struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quads[0]->input.x0, - quads[0]->input.y0); - uint q, i, j; + uint i, j, q; - for (q = 0; q < nr; q++) { - struct quad_header *quad = quads[q]; - float (*quadColor)[4] = quad->output.color[cbuf]; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(qs->softpipe->cbuf_cache[0], + quads[0]->input.x0, + quads[0]->input.y0); - /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); - for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; + for (q = 0; q < nr; q++) { + struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[0]; + const int itx = (quad->input.x0 & (TILE_SIZE-1)); + const int ity = (quad->input.y0 & (TILE_SIZE-1)); + + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; } } - - /* - * Compute src/first term RGB - */ - switch (softpipe->blend->rgb_src_factor) { - case PIPE_BLENDFACTOR_ONE: - VEC4_COPY(source[0], quadColor[0]); /* R */ - VEC4_COPY(source[1], quadColor[1]); /* G */ - VEC4_COPY(source[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */ - VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */ - VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: - { - const float *alpha = quadColor[3]; - VEC4_MUL(source[0], quadColor[0], alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_DST_COLOR: - VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */ - VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */ - VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */ - break; - case PIPE_BLENDFACTOR_DST_ALPHA: - { - const float *alpha = dest[3]; - VEC4_MUL(source[0], quadColor[0], alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - { - const float *alpha = quadColor[3]; - float diff[4], temp[4]; - VEC4_SUB(diff, one, dest[3]); - VEC4_MIN(temp, alpha, diff); - VEC4_MUL(source[0], quadColor[0], temp); /* R */ - VEC4_MUL(source[1], quadColor[1], temp); /* G */ - VEC4_MUL(source[2], quadColor[2], temp); /* B */ - } - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - { - float comp[4]; - VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ - VEC4_MUL(source[0], quadColor[0], comp); /* R */ - VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ - VEC4_MUL(source[1], quadColor[1], comp); /* G */ - VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ - VEC4_MUL(source[2], quadColor[2], comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_CONST_ALPHA: - { - float alpha[4]; - VEC4_SCALAR(alpha, softpipe->blend_color.color[3]); - VEC4_MUL(source[0], quadColor[0], alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_SRC1_COLOR: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_SRC1_ALPHA: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(source[0], zero); /* R */ - VEC4_COPY(source[1], zero); /* G */ - VEC4_COPY(source[2], zero); /* B */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ - VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ - VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ - VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ - VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ - VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - { - float inv_alpha[4]; - VEC4_SUB(inv_alpha, one, quadColor[3]); - VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_ALPHA: - { - float inv_alpha[4]; - VEC4_SUB(inv_alpha, one, dest[3]); - VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, dest[0]); /* R */ - VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ - VEC4_SUB(inv_comp, one, dest[1]); /* G */ - VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ - VEC4_SUB(inv_comp, one, dest[2]); /* B */ - VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - { - float inv_comp[4]; - /* R */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); - VEC4_MUL(source[0], quadColor[0], inv_comp); - /* G */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); - VEC4_MUL(source[1], quadColor[1], inv_comp); - /* B */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); - VEC4_MUL(source[2], quadColor[2], inv_comp); - } - break; - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: - { - float inv_alpha[4]; - VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]); - VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_SRC1_COLOR: - assert(0); /* to do */ - break; - case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: - assert(0); /* to do */ - break; - default: - assert(0); - } - - /* - * Compute src/first term A - */ - switch (softpipe->blend->alpha_src_factor) { - case PIPE_BLENDFACTOR_ONE: - VEC4_COPY(source[3], quadColor[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_SRC_ALPHA: - { - const float *alpha = quadColor[3]; - VEC4_MUL(source[3], quadColor[3], alpha); /* A */ - } - break; - case PIPE_BLENDFACTOR_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - /* multiply alpha by 1.0 */ - VEC4_COPY(source[3], quadColor[3]); /* A */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_CONST_ALPHA: - { - float comp[4]; - VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ - VEC4_MUL(source[3], quadColor[3], comp); /* A */ - } - break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(source[3], zero); /* A */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - { - float inv_alpha[4]; - VEC4_SUB(inv_alpha, one, quadColor[3]); - VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_DST_ALPHA: - { - float inv_alpha[4]; - VEC4_SUB(inv_alpha, one, dest[3]); - VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ - } - break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: - { - float inv_comp[4]; - /* A */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); - VEC4_MUL(source[3], quadColor[3], inv_comp); - } - break; - default: - assert(0); - } - - - /* - * Compute dest/second term RGB - */ - switch (softpipe->blend->rgb_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - /* dest = dest * 1 NO-OP, leave dest as-is */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */ - VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */ - VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: - VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */ - VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */ - VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */ - break; - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */ - VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */ - VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */ - break; - case PIPE_BLENDFACTOR_DST_COLOR: - VEC4_MUL(dest[0], dest[0], dest[0]); /* R */ - VEC4_MUL(dest[1], dest[1], dest[1]); /* G */ - VEC4_MUL(dest[2], dest[2], dest[2]); /* B */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - assert(0); /* illegal */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - { - float comp[4]; - VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ - VEC4_MUL(dest[0], dest[0], comp); /* R */ - VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ - VEC4_MUL(dest[1], dest[1], comp); /* G */ - VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ - VEC4_MUL(dest[2], dest[2], comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_CONST_ALPHA: - { - float comp[4]; - VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ - VEC4_MUL(dest[0], dest[0], comp); /* R */ - VEC4_MUL(dest[1], dest[1], comp); /* G */ - VEC4_MUL(dest[2], dest[2], comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(dest[0], zero); /* R */ - VEC4_COPY(dest[1], zero); /* G */ - VEC4_COPY(dest[2], zero); /* B */ - break; - case PIPE_BLENDFACTOR_SRC1_COLOR: - case PIPE_BLENDFACTOR_SRC1_ALPHA: - /* XXX what are these? */ - assert(0); - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ - VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ - VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ - VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ - VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ - VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - { - float one_minus_alpha[QUAD_SIZE]; - VEC4_SUB(one_minus_alpha, one, quadColor[3]); - VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ - VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ - VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_ALPHA: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, dest[3]); /* A */ - VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ - VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ - VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, dest[0]); /* R */ - VEC4_MUL(dest[0], dest[0], inv_comp); /* R */ - VEC4_SUB(inv_comp, one, dest[1]); /* G */ - VEC4_MUL(dest[1], dest[1], inv_comp); /* G */ - VEC4_SUB(inv_comp, one, dest[2]); /* B */ - VEC4_MUL(dest[2], dest[2], inv_comp); /* B */ - } - break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - { - float inv_comp[4]; - /* R */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); - VEC4_MUL(dest[0], dest[0], inv_comp); - /* G */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); - VEC4_MUL(dest[1], dest[1], inv_comp); - /* B */ - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); - VEC4_MUL(dest[2], dest[2], inv_comp); - } - break; - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: - { - float inv_comp[4]; - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); - VEC4_MUL(dest[0], dest[0], inv_comp); - VEC4_MUL(dest[1], dest[1], inv_comp); - VEC4_MUL(dest[2], dest[2], inv_comp); - } - break; - case PIPE_BLENDFACTOR_INV_SRC1_COLOR: - case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: - /* XXX what are these? */ - assert(0); - break; - default: - assert(0); - } - - /* - * Compute dest/second term A - */ - switch (softpipe->blend->alpha_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - /* dest = dest * 1 NO-OP, leave dest as-is */ - break; - case PIPE_BLENDFACTOR_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_SRC_ALPHA: - VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */ - break; - case PIPE_BLENDFACTOR_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_DST_ALPHA: - VEC4_MUL(dest[3], dest[3], dest[3]); /* A */ - break; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: - assert(0); /* illegal */ - break; - case PIPE_BLENDFACTOR_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_CONST_ALPHA: - { - float comp[4]; - VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ - VEC4_MUL(dest[3], dest[3], comp); /* A */ - } - break; - case PIPE_BLENDFACTOR_ZERO: - VEC4_COPY(dest[3], zero); /* A */ - break; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - { - float one_minus_alpha[QUAD_SIZE]; - VEC4_SUB(one_minus_alpha, one, quadColor[3]); - VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */ - } - break; - case PIPE_BLENDFACTOR_INV_DST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_DST_ALPHA: - { - float inv_comp[4]; - VEC4_SUB(inv_comp, one, dest[3]); /* A */ - VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */ - } - break; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: - /* fall-through */ - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: - { - float inv_comp[4]; - VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); - VEC4_MUL(dest[3], dest[3], inv_comp); - } - break; - default: - assert(0); - } - - /* - * Combine RGB terms - */ - switch (softpipe->blend->rgb_func) { - case PIPE_BLEND_ADD: - VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ - VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ - VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_SUBTRACT: - VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */ - VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */ - VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_REVERSE_SUBTRACT: - VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */ - VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */ - VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */ - break; - case PIPE_BLEND_MIN: - VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */ - VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */ - VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */ - break; - case PIPE_BLEND_MAX: - VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */ - VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */ - VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */ - break; - default: - assert(0); - } - - /* - * Combine A terms - */ - switch (softpipe->blend->alpha_func) { - case PIPE_BLEND_ADD: - VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_SUBTRACT: - VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_REVERSE_SUBTRACT: - VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */ - break; - case PIPE_BLEND_MIN: - VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */ - break; - case PIPE_BLEND_MAX: - VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */ - break; - default: - assert(0); - } } - } /* cbuf loop */ - - /* pass blended quad to next stage */ - qs->next->run(qs->next, quads, nr); + } } static void -blend_quad(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +choose_blend_quad(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { struct softpipe_context *softpipe = qs->softpipe; const struct pipe_blend_state *blend = softpipe->blend; - if (softpipe->blend->logicop_enable) { - qs->run = logicop_quads; - } - else { - qs->run = blend_quads_fallback; + qs->run = blend_fallback; - if (blend->rgb_src_factor == blend->alpha_src_factor && - blend->rgb_dst_factor == blend->alpha_dst_factor && - blend->rgb_func == blend->alpha_func && - softpipe->framebuffer.nr_cbufs == 1) + if (!softpipe->blend->logicop_enable && + softpipe->blend->colormask == 0xf) + { + if (!blend->blend_enable) { + qs->run = single_output_color; + } + else if (blend->rgb_src_factor == blend->alpha_src_factor && + blend->rgb_dst_factor == blend->alpha_dst_factor && + blend->rgb_func == blend->alpha_func && + softpipe->framebuffer.nr_cbufs == 1) { if (blend->alpha_func == PIPE_BLEND_ADD) { if (blend->rgb_src_factor == PIPE_BLENDFACTOR_ONE && @@ -871,7 +956,7 @@ blend_quad(struct quad_stage *qs, qs->run = blend_single_add_src_alpha_inv_src_alpha; } - } + } } qs->run(qs, quads, nr); @@ -880,8 +965,7 @@ blend_quad(struct quad_stage *qs, static void blend_begin(struct quad_stage *qs) { - qs->run = blend_quad; - qs->next->begin(qs->next); + qs->run = choose_blend_quad; } @@ -897,7 +981,7 @@ struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = blend_begin; - stage->run = blend_quad; + stage->run = choose_blend_quad; stage->destroy = blend_destroy; return stage; diff --git a/src/gallium/drivers/softpipe/sp_quad_bufloop.c b/src/gallium/drivers/softpipe/sp_quad_bufloop.c deleted file mode 100644 index 953d8516b9..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_bufloop.c +++ /dev/null @@ -1,74 +0,0 @@ - -#include "util/u_memory.h" -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_quad_pipe.h" - - -/** - * Loop over colorbuffers, passing quad to next stage each time. - */ -static void -cbuf_loop_quad(struct quad_stage *qs, struct quad_header *quad) -{ - struct softpipe_context *softpipe = qs->softpipe; - float tmp[PIPE_MAX_COLOR_BUFS][4][QUAD_SIZE]; - unsigned i; - - assert(sizeof(quad->outputs.color) == sizeof(tmp)); - assert(softpipe->framebuffer.nr_cbufs <= PIPE_MAX_COLOR_BUFS); - - /* make copy of original colors since they can get modified - * by blending and masking. - * XXX we won't have to do this if the fragment program actually emits - * N separate colors and we're drawing to N color buffers (MRT). - * But if we emitted one color and glDrawBuffer(GL_FRONT_AND_BACK) is - * in effect, we need to save/restore colors like this. - */ - memcpy(tmp, quad->outputs.color, sizeof(tmp)); - - for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) { - /* set current cbuffer */ -#if 0 /* obsolete & going away */ - softpipe->current_cbuf = i; -#endif - - /* pass blended quad to next stage */ - qs->next->run(qs->next, quad); - - /* restore quad's colors for next buffer */ - memcpy(quad->outputs.color, tmp, sizeof(tmp)); - } -} - - -static void cbuf_loop_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void cbuf_loop_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -/** - * Create the colorbuffer loop stage. - * This is used to implement multiple render targets and GL_FRONT_AND_BACK - * rendering. - */ -struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = cbuf_loop_begin; - stage->run = cbuf_loop_quad; - stage->destroy = cbuf_loop_destroy; - - return stage; -} - diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c deleted file mode 100644 index ac74287473..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_colormask.c +++ /dev/null @@ -1,126 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -/** - * \brief quad colormask stage - * \author Brian Paul - */ - -#include "pipe/p_defines.h" -#include "util/u_math.h" -#include "util/u_memory.h" -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_quad_pipe.h" -#include "sp_tile_cache.h" - - - -/** - * XXX colormask could be rolled into blending... - */ -static void -colormask_quad(struct quad_stage *qs, struct quad_header *quad) -{ - struct softpipe_context *softpipe = qs->softpipe; - uint cbuf; - - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { - float dest[4][QUAD_SIZE]; - struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quad->input.x0, quad->input.y0); - float (*quadColor)[4] = quad->output.color[cbuf]; - uint i, j; - - /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1); - int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1); - for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; - } - } - - /* R */ - if (!(softpipe->blend->colormask & PIPE_MASK_R)) - COPY_4V(quadColor[0], dest[0]); - - /* G */ - if (!(softpipe->blend->colormask & PIPE_MASK_G)) - COPY_4V(quadColor[1], dest[1]); - - /* B */ - if (!(softpipe->blend->colormask & PIPE_MASK_B)) - COPY_4V(quadColor[2], dest[2]); - - /* A */ - if (!(softpipe->blend->colormask & PIPE_MASK_A)) - COPY_4V(quadColor[3], dest[3]); - } -} - -static void -colormask_quads(struct quad_stage *qs, struct quad_header *quads[], - unsigned nr) -{ - unsigned i; - - for (i = 0; i < nr; i++) - colormask_quad(qs, quads[i]); - - /* pass quad to next stage */ - qs->next->run(qs->next, quads, nr); -} - - - -static void colormask_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void colormask_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = colormask_begin; - stage->run = colormask_quads; - stage->destroy = colormask_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c index f06a385b3c..989e997f81 100644 --- a/src/gallium/drivers/softpipe/sp_quad_coverage.c +++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c @@ -68,7 +68,6 @@ coverage_run(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { - struct softpipe_context *softpipe = qs->softpipe; unsigned i; for (i = 0; i < nr; i++) diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c deleted file mode 100644 index 79a222ff58..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_output.c +++ /dev/null @@ -1,109 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -#include "util/u_memory.h" -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_quad_pipe.h" -#include "sp_tile_cache.h" - - -/** - * Last step of quad processing: write quad colors to the framebuffer, - * taking mask into account. - */ -static void -output_quad(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) -{ - - struct softpipe_context *softpipe = qs->softpipe; - uint cbuf; - - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { - struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->cbuf_cache[cbuf], - quads[0]->input.x0, - quads[0]->input.y0); - int i, j, q; - - /* get/swizzle dest colors */ - for (q = 0; q < nr; q++) { - struct quad_header *quad = quads[q]; - float (*quadColor)[4] = quad->output.color[cbuf]; - - /* in-tile pos: */ - const int itx = quad->input.x0 % TILE_SIZE; - const int ity = quad->input.y0 % TILE_SIZE; - - - for (j = 0; j < QUAD_SIZE; j++) { - if (quad->inout.mask & (1 << j)) { - int x = itx + (j & 1); - int y = ity + (j >> 1); - for (i = 0; i < 4; i++) { /* loop over color chans */ - tile->data.color[y][x][i] = quadColor[i][j]; - } - if (0) { - debug_printf("sp write pixel %d,%d: %g, %g, %g\n", - quad->input.x0 + x, - quad->input.y0 + y, - quadColor[0][j], - quadColor[1][j], - quadColor[2][j]); - } - } - } - } - } -} - - -static void output_begin(struct quad_stage *qs) -{ - assert(qs->next == NULL); -} - - -static void output_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = output_begin; - stage->run = output_quad; - stage->destroy = output_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index 6fae7d552f..d138d417ac 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -65,25 +65,16 @@ sp_build_quad_pipeline(struct softpipe_context *sp) /* Color combine */ - sp->quad.first = sp->quad.output; - - if (sp->blend->colormask != 0xf) { - sp_push_quad_first( sp, sp->quad.colormask ); - } - - if (sp->blend->blend_enable || - sp->blend->logicop_enable) { - sp_push_quad_first( sp, sp->quad.blend ); - } + sp->quad.first = sp->quad.blend; + /* Shade/Depth/Stencil/Alpha + */ if ((sp->rasterizer->poly_smooth && sp->reduced_prim == PIPE_PRIM_TRIANGLES) || (sp->rasterizer->line_smooth && sp->reduced_prim == PIPE_PRIM_LINES) || (sp->rasterizer->point_smooth && sp->reduced_prim == PIPE_PRIM_POINTS)) { sp_push_quad_first( sp, sp->quad.coverage ); } - /* Shade/Depth/Stencil/Alpha - */ if (sp->active_query_count) { sp_push_quad_first( sp, sp->quad.occlusion ); } -- cgit v1.2.3 From ade8984f5023b05412f2467add4a59d14af53185 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 25 Jul 2009 10:01:06 +0100 Subject: softpipe: cleanup framebuffer state routine slightly --- src/gallium/drivers/softpipe/sp_state_surface.c | 50 +++++-------------------- 1 file changed, 10 insertions(+), 40 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c index 1621a27614..c8f55c3cec 100644 --- a/src/gallium/drivers/softpipe/sp_state_surface.c +++ b/src/gallium/drivers/softpipe/sp_state_surface.c @@ -75,51 +75,21 @@ softpipe_set_framebuffer_state(struct pipe_context *pipe, /* update cache */ sp_tile_cache_set_surface(sp->zsbuf_cache, fb->zsbuf); - } - -#if 0 - /* XXX combined depth/stencil here */ - - /* sbuf changing? */ - if (sp->framebuffer.sbuf != fb->sbuf) { - /* flush old */ - sp_flush_tile_cache(sp, sp->sbuf_cache_sep); - - /* assign new */ - sp->framebuffer.sbuf = fb->sbuf; - - /* update cache */ - if (fb->sbuf != fb->zbuf) { - /* separate stencil buf */ - sp->sbuf_cache = sp->sbuf_cache_sep; - sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf); - } - else { - /* combined depth/stencil */ - sp->sbuf_cache = sp->zbuf_cache; - sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf); - } - } -#endif - /* Tell draw module how deep the Z/depth buffer is */ - { - int depth_bits; - double mrd; + /* Tell draw module how deep the Z/depth buffer is */ if (sp->framebuffer.zsbuf) { + int depth_bits; + double mrd; depth_bits = pf_get_component_bits(sp->framebuffer.zsbuf->format, PIPE_FORMAT_COMP_Z); + if (depth_bits > 16) { + mrd = 0.0000001; + } + else { + mrd = 0.00002; + } + draw_set_mrd(sp->draw, mrd); } - else { - depth_bits = 0; - } - if (depth_bits > 16) { - mrd = 0.0000001; - } - else { - mrd = 0.00002; - } - draw_set_mrd(sp->draw, mrd); } sp->framebuffer.width = fb->width; -- cgit v1.2.3 From 85613cc4f14de968ddd503610c5b8fcc77234c81 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 25 Jul 2009 11:01:48 +0100 Subject: softpipe: fix error in scissor state dependencies --- src/gallium/drivers/softpipe/sp_state_derived.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 629a1f8e29..0226501267 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -165,11 +165,19 @@ softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe) static void compute_cliprect(struct softpipe_context *sp) { + /* SP_NEW_FRAMEBUFFER + */ uint surfWidth = sp->framebuffer.width; uint surfHeight = sp->framebuffer.height; + /* SP_NEW_RASTERIZER + */ if (sp->rasterizer->scissor) { - /* clip to scissor rect */ + + /* SP_NEW_SCISSOR + * + * clip to scissor rect: + */ sp->cliprect.minx = MAX2(sp->scissor.minx, 0); sp->cliprect.miny = MAX2(sp->scissor.miny, 0); sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth); @@ -231,7 +239,7 @@ void softpipe_update_derived( struct softpipe_context *softpipe ) invalidate_vertex_layout( softpipe ); if (softpipe->dirty & (SP_NEW_SCISSOR | - SP_NEW_DEPTH_STENCIL_ALPHA | + SP_NEW_RASTERIZER | SP_NEW_FRAMEBUFFER)) compute_cliprect(softpipe); -- cgit v1.2.3 From bac8e34c9e4077d370923773d67fe565ce154849 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 27 Jul 2009 08:17:45 +0100 Subject: softpipe: move all depth/stencil/alpha pixel processing into one stage --- src/gallium/drivers/softpipe/Makefile | 5 - src/gallium/drivers/softpipe/sp_context.c | 12 - src/gallium/drivers/softpipe/sp_context.h | 6 - src/gallium/drivers/softpipe/sp_quad_alpha_test.c | 112 ---- src/gallium/drivers/softpipe/sp_quad_coverage.c | 101 ---- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 659 ++++++++++++++++++---- src/gallium/drivers/softpipe/sp_quad_earlyz.c | 94 --- src/gallium/drivers/softpipe/sp_quad_fs.c | 46 +- src/gallium/drivers/softpipe/sp_quad_occlusion.c | 87 --- src/gallium/drivers/softpipe/sp_quad_pipe.c | 46 +- src/gallium/drivers/softpipe/sp_quad_pipe.h | 2 - src/gallium/drivers/softpipe/sp_quad_stencil.c | 363 ------------ src/gallium/drivers/softpipe/sp_state_derived.c | 4 +- 13 files changed, 580 insertions(+), 957 deletions(-) delete mode 100644 src/gallium/drivers/softpipe/sp_quad_alpha_test.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_coverage.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_earlyz.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_occlusion.c delete mode 100644 src/gallium/drivers/softpipe/sp_quad_stencil.c diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile index bdc1a5819f..48522abe98 100644 --- a/src/gallium/drivers/softpipe/Makefile +++ b/src/gallium/drivers/softpipe/Makefile @@ -16,13 +16,8 @@ C_SOURCES = \ sp_prim_vbuf.c \ sp_quad_pipe.c \ sp_quad_stipple.c \ - sp_quad_earlyz.c \ sp_quad_depth_test.c \ - sp_quad_stencil.c \ sp_quad_fs.c \ - sp_quad_alpha_test.c \ - sp_quad_occlusion.c \ - sp_quad_coverage.c \ sp_quad_blend.c \ sp_screen.c \ sp_setup.c \ diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 28a0dd62ac..e35c6b3aec 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -88,14 +88,8 @@ static void softpipe_destroy( struct pipe_context *pipe ) if (softpipe->draw) draw_destroy( softpipe->draw ); - softpipe->quad.polygon_stipple->destroy( softpipe->quad.polygon_stipple ); - softpipe->quad.earlyz->destroy( softpipe->quad.earlyz ); softpipe->quad.shade->destroy( softpipe->quad.shade ); - softpipe->quad.alpha_test->destroy( softpipe->quad.alpha_test ); softpipe->quad.depth_test->destroy( softpipe->quad.depth_test ); - softpipe->quad.stencil_test->destroy( softpipe->quad.stencil_test ); - softpipe->quad.occlusion->destroy( softpipe->quad.occlusion ); - softpipe->quad.coverage->destroy( softpipe->quad.coverage ); softpipe->quad.blend->destroy( softpipe->quad.blend ); for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) @@ -230,14 +224,8 @@ softpipe_create( struct pipe_screen *screen ) /* setup quad rendering stages */ - softpipe->quad.polygon_stipple = sp_quad_polygon_stipple_stage(softpipe); - softpipe->quad.earlyz = sp_quad_earlyz_stage(softpipe); softpipe->quad.shade = sp_quad_shade_stage(softpipe); - softpipe->quad.alpha_test = sp_quad_alpha_test_stage(softpipe); softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe); - softpipe->quad.stencil_test = sp_quad_stencil_test_stage(softpipe); - softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe); - softpipe->quad.coverage = sp_quad_coverage_stage(softpipe); softpipe->quad.blend = sp_quad_blend_stage(softpipe); /* vertex shader samplers */ diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index b76ff610a3..fa3306c020 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -114,14 +114,8 @@ struct softpipe_context { /** Software quad rendering pipeline */ struct { - struct quad_stage *polygon_stipple; - struct quad_stage *earlyz; struct quad_stage *shade; - struct quad_stage *alpha_test; - struct quad_stage *stencil_test; struct quad_stage *depth_test; - struct quad_stage *occlusion; - struct quad_stage *coverage; struct quad_stage *blend; struct quad_stage *first; /**< points to one of the above stages */ diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c deleted file mode 100644 index 3a282208b6..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c +++ /dev/null @@ -1,112 +0,0 @@ - -/** - * quad alpha test - */ - -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_quad_pipe.h" -#include "pipe/p_defines.h" -#include "util/u_memory.h" - -#define ALPHATEST( FUNC, COMP ) \ - static void \ - alpha_test_quads_##FUNC( struct quad_stage *qs, \ - struct quad_header *quads[], \ - unsigned nr ) \ - { \ - const float ref = qs->softpipe->depth_stencil->alpha.ref_value; \ - const uint cbuf = 0; /* only output[0].alpha is tested */ \ - unsigned pass_nr = 0; \ - unsigned i; \ - \ - for (i = 0; i < nr; i++) { \ - const float *aaaa = quads[i]->output.color[cbuf][3]; \ - unsigned passMask = 0; \ - \ - if (aaaa[0] COMP ref) passMask |= (1 << 0); \ - if (aaaa[1] COMP ref) passMask |= (1 << 1); \ - if (aaaa[2] COMP ref) passMask |= (1 << 2); \ - if (aaaa[3] COMP ref) passMask |= (1 << 3); \ - \ - quads[i]->inout.mask &= passMask; \ - \ - if (quads[i]->inout.mask) \ - quads[pass_nr++] = quads[i]; \ - } \ - \ - if (pass_nr) \ - qs->next->run(qs->next, quads, pass_nr); \ - } - - -ALPHATEST( LESS, < ) -ALPHATEST( EQUAL, == ) -ALPHATEST( LEQUAL, <= ) -ALPHATEST( GREATER, > ) -ALPHATEST( NOTEQUAL, != ) -ALPHATEST( GEQUAL, >= ) - - -/* XXX: Incorporate into shader using KILP. - */ -static void -alpha_test_quad(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) -{ - switch (qs->softpipe->depth_stencil->alpha.func) { - case PIPE_FUNC_LESS: - alpha_test_quads_LESS( qs, quads, nr ); - break; - case PIPE_FUNC_EQUAL: - alpha_test_quads_EQUAL( qs, quads, nr ); - break; - case PIPE_FUNC_LEQUAL: - alpha_test_quads_LEQUAL( qs, quads, nr ); - break; - case PIPE_FUNC_GREATER: - alpha_test_quads_GREATER( qs, quads, nr ); - break; - case PIPE_FUNC_NOTEQUAL: - alpha_test_quads_NOTEQUAL( qs, quads, nr ); - break; - case PIPE_FUNC_GEQUAL: - alpha_test_quads_GEQUAL( qs, quads, nr ); - break; - case PIPE_FUNC_ALWAYS: - assert(0); /* should be caught earlier */ - qs->next->run(qs->next, quads, nr); - break; - case PIPE_FUNC_NEVER: - default: - assert(0); /* should be caught earlier */ - return; - } -} - - -static void alpha_test_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void alpha_test_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage * -sp_quad_alpha_test_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = alpha_test_begin; - stage->run = alpha_test_quad; - stage->destroy = alpha_test_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c deleted file mode 100644 index 989e997f81..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_coverage.c +++ /dev/null @@ -1,101 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - - -/** - * \brief Apply AA coverage to quad alpha valus - * \author Brian Paul - */ - - -#include "pipe/p_defines.h" -#include "util/u_memory.h" -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_quad_pipe.h" - - -/** - * Multiply quad's alpha values by the fragment coverage. - */ -static INLINE void -coverage_quad(struct quad_stage *qs, struct quad_header *quad) -{ - struct softpipe_context *softpipe = qs->softpipe; - uint cbuf; - - /* loop over colorbuffer outputs */ - for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { - float (*quadColor)[4] = quad->output.color[cbuf]; - unsigned j; - for (j = 0; j < QUAD_SIZE; j++) { - assert(quad->input.coverage[j] >= 0.0); - assert(quad->input.coverage[j] <= 1.0); - quadColor[3][j] *= quad->input.coverage[j]; - } - } -} - - -/* XXX: Incorporate into shader after alpha_test. - */ -static void -coverage_run(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) -{ - unsigned i; - - for (i = 0; i < nr; i++) - coverage_quad( qs, quads[i] ); - - qs->next->run(qs->next, quads, nr); -} - -static void coverage_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void coverage_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = coverage_begin; - stage->run = coverage_run; - stage->destroy = coverage_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index 8f223a7eae..bf65799a81 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -31,61 +31,109 @@ #include "pipe/p_defines.h" #include "util/u_memory.h" +#include "tgsi/tgsi_scan.h" #include "sp_context.h" #include "sp_quad.h" #include "sp_surface.h" #include "sp_quad_pipe.h" #include "sp_tile_cache.h" +#include "sp_state.h" /* for sp_fragment_shader */ -/** - * Do depth testing for a quad. - * Not static since it's used by the stencil code. - */ +struct depth_data { + struct pipe_surface *ps; + enum pipe_format format; + unsigned bzzzz[QUAD_SIZE]; /**< Z values fetched from depth buffer */ + unsigned qzzzz[QUAD_SIZE]; /**< Z values from the quad */ + ubyte stencilVals[QUAD_SIZE]; + struct softpipe_cached_tile *tile; +}; -/* - * To increase efficiency, we should probably have multiple versions - * of this function that are specifically for Z16, Z32 and FP Z buffers. - * Try to effectively do that with codegen... - */ -boolean -sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) + +static void +get_depth_stencil_values( struct depth_data *data, + const struct quad_header *quad ) { - struct softpipe_context *softpipe = qs->softpipe; - struct pipe_surface *ps = softpipe->framebuffer.zsbuf; - const enum pipe_format format = ps->format; - unsigned bzzzz[QUAD_SIZE]; /**< Z values fetched from depth buffer */ - unsigned qzzzz[QUAD_SIZE]; /**< Z values from the quad */ - unsigned zmask = 0; unsigned j; - struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->zsbuf_cache, quad->input.x0, quad->input.y0); + const struct softpipe_cached_tile *tile = data->tile; + + switch (data->format) { + case PIPE_FORMAT_Z16_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + data->bzzzz[j] = tile->data.depth16[y][x]; + } + break; + case PIPE_FORMAT_Z32_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + data->bzzzz[j] = tile->data.depth32[y][x]; + } + break; + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + data->bzzzz[j] = tile->data.depth32[y][x] & 0xffffff; + data->stencilVals[j] = tile->data.depth32[y][x] >> 24; + } + break; + case PIPE_FORMAT_Z24X8_UNORM: + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + data->bzzzz[j] = tile->data.depth32[y][x] >> 8; + data->stencilVals[j] = tile->data.depth32[y][x] & 0xff; + } + break; + default: + assert(0); + } +} - assert(ps); /* shouldn't get here if there's no zbuffer */ +/* If the shader has not been run, interpolate the depth values + * ourselves. + */ +static void +interpolate_quad_depth( struct quad_header *quad ) +{ + const float fx = (float) quad->input.x0; + const float fy = (float) quad->input.y0; + const float dzdx = quad->posCoef->dadx[2]; + const float dzdy = quad->posCoef->dady[2]; + const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy; + + quad->output.depth[0] = z0; + quad->output.depth[1] = z0 + dzdx; + quad->output.depth[2] = z0 + dzdy; + quad->output.depth[3] = z0 + dzdx + dzdy; +} - /* - * Convert quad's float depth values to int depth values (qzzzz). + +static void +convert_quad_depth( struct depth_data *data, + const struct quad_header *quad ) +{ + unsigned j; + + /* Convert quad's float depth values to int depth values (qzzzz). * If the Z buffer stores integer values, we _have_ to do the depth * compares with integers (not floats). Otherwise, the float->int->float * conversion of Z values (which isn't an identity function) will cause * Z-fighting errors. - * - * Also, get the zbuffer values (bzzzz) from the cached tile. */ - switch (format) { + switch (data->format) { case PIPE_FORMAT_Z16_UNORM: { float scale = 65535.0; for (j = 0; j < QUAD_SIZE; j++) { - qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); - } - - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - bzzzz[j] = tile->data.depth16[y][x]; + data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); } } break; @@ -94,47 +142,247 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) double scale = (double) (uint) ~0UL; for (j = 0; j < QUAD_SIZE; j++) { - qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); - } - - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - bzzzz[j] = tile->data.depth32[y][x]; + data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); } } break; case PIPE_FORMAT_X8Z24_UNORM: - /* fall-through */ case PIPE_FORMAT_S8Z24_UNORM: { float scale = (float) ((1 << 24) - 1); for (j = 0; j < QUAD_SIZE; j++) { - qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); - } - - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - bzzzz[j] = tile->data.depth32[y][x] & 0xffffff; + data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); } } break; case PIPE_FORMAT_Z24X8_UNORM: - /* fall-through */ case PIPE_FORMAT_Z24S8_UNORM: { float scale = (float) ((1 << 24) - 1); for (j = 0; j < QUAD_SIZE; j++) { - qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); + data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale); } + } + break; + default: + assert(0); + } +} - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - bzzzz[j] = tile->data.depth32[y][x] >> 8; + + +static void +write_depth_stencil_values( struct depth_data *data, + struct quad_header *quad ) +{ + struct softpipe_cached_tile *tile = data->tile; + unsigned j; + + /* put updated Z values back into cached tile */ + switch (data->format) { + case PIPE_FORMAT_Z16_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.depth16[y][x] = (ushort) data->bzzzz[j]; + } + break; + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_Z32_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.depth32[y][x] = data->bzzzz[j]; + } + break; + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.depth32[y][x] = (data->stencilVals[j] << 24) | data->bzzzz[j]; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.depth32[y][x] = (data->bzzzz[j] << 8) | data->stencilVals[j]; + } + break; + case PIPE_FORMAT_Z24X8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->input.x0 % TILE_SIZE + (j & 1); + int y = quad->input.y0 % TILE_SIZE + (j >> 1); + tile->data.depth32[y][x] = data->bzzzz[j] << 8; + } + break; + default: + assert(0); + } +} + + + + +/** Only 8-bit stencil supported */ +#define STENCIL_MAX 0xff + + +/** + * Do the basic stencil test (compare stencil buffer values against the + * reference value. + * + * \param data->stencilVals the stencil values from the stencil buffer + * \param func the stencil func (PIPE_FUNC_x) + * \param ref the stencil reference value + * \param valMask the stencil value mask indicating which bits of the stencil + * values and ref value are to be used. + * \return mask indicating which pixels passed the stencil test + */ +static unsigned +do_stencil_test(struct depth_data *data, + unsigned func, + unsigned ref, unsigned valMask) +{ + unsigned passMask = 0x0; + unsigned j; + + ref &= valMask; + + switch (func) { + case PIPE_FUNC_NEVER: + /* passMask = 0x0 */ + break; + case PIPE_FUNC_LESS: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref < (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_EQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref == (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_LEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref <= (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GREATER: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref > (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_NOTEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref != (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref >= (data->stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_ALWAYS: + passMask = MASK_ALL; + break; + default: + assert(0); + } + + return passMask; +} + + +/** + * Apply the stencil operator to stencil values. + * + * \param data->stencilVals the stencil buffer values (read and written) + * \param mask indicates which pixels to update + * \param op the stencil operator (PIPE_STENCIL_OP_x) + * \param ref the stencil reference value + * \param wrtMask writemask controlling which bits are changed in the + * stencil values + */ +static void +apply_stencil_op(struct depth_data *data, + unsigned mask, unsigned op, ubyte ref, ubyte wrtMask) +{ + unsigned j; + ubyte newstencil[QUAD_SIZE]; + + for (j = 0; j < QUAD_SIZE; j++) { + newstencil[j] = data->stencilVals[j]; + } + + switch (op) { + case PIPE_STENCIL_OP_KEEP: + /* no-op */ + break; + case PIPE_STENCIL_OP_ZERO: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = 0; + } + } + break; + case PIPE_STENCIL_OP_REPLACE: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = ref; + } + } + break; + case PIPE_STENCIL_OP_INCR: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + if (data->stencilVals[j] < STENCIL_MAX) { + newstencil[j] = data->stencilVals[j] + 1; + } + } + } + break; + case PIPE_STENCIL_OP_DECR: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + if (data->stencilVals[j] > 0) { + newstencil[j] = data->stencilVals[j] - 1; + } + } + } + break; + case PIPE_STENCIL_OP_INCR_WRAP: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = data->stencilVals[j] + 1; + } + } + break; + case PIPE_STENCIL_OP_DECR_WRAP: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = data->stencilVals[j] - 1; + } + } + break; + case PIPE_STENCIL_OP_INVERT: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = ~data->stencilVals[j]; } } break; @@ -142,6 +390,39 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) assert(0); } + /* + * update the stencil values + */ + if (wrtMask != STENCIL_MAX) { + /* apply bit-wise stencil buffer writemask */ + for (j = 0; j < QUAD_SIZE; j++) { + data->stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & data->stencilVals[j]); + } + } + else { + for (j = 0; j < QUAD_SIZE; j++) { + data->stencilVals[j] = newstencil[j]; + } + } +} + + + +/* + * To increase efficiency, we should probably have multiple versions + * of this function that are specifically for Z16, Z32 and FP Z buffers. + * Try to effectively do that with codegen... + */ + +static boolean +depth_test_quad(struct quad_stage *qs, + struct depth_data *data, + struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + unsigned zmask = 0; + unsigned j; + switch (softpipe->depth_stencil->depth.func) { case PIPE_FUNC_NEVER: /* zmask = 0 */ @@ -151,37 +432,37 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) * Like this: quad->mask &= (quad->outputs.depth < zzzz); */ for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] < bzzzz[j]) + if (data->qzzzz[j] < data->bzzzz[j]) zmask |= 1 << j; } break; case PIPE_FUNC_EQUAL: for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] == bzzzz[j]) + if (data->qzzzz[j] == data->bzzzz[j]) zmask |= 1 << j; } break; case PIPE_FUNC_LEQUAL: for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] <= bzzzz[j]) + if (data->qzzzz[j] <= data->bzzzz[j]) zmask |= (1 << j); } break; case PIPE_FUNC_GREATER: for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] > bzzzz[j]) + if (data->qzzzz[j] > data->bzzzz[j]) zmask |= (1 << j); } break; case PIPE_FUNC_NOTEQUAL: for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] != bzzzz[j]) + if (data->qzzzz[j] != data->bzzzz[j]) zmask |= (1 << j); } break; case PIPE_FUNC_GEQUAL: for (j = 0; j < QUAD_SIZE; j++) { - if (qzzzz[j] >= bzzzz[j]) + if (data->qzzzz[j] >= data->bzzzz[j]) zmask |= (1 << j); } break; @@ -196,83 +477,231 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) if (quad->inout.mask == 0) return FALSE; + /* Update our internal copy only if writemask set. Even if + * depth.writemask is FALSE, may still need to write out buffer + * data due to stencil changes. + */ if (softpipe->depth_stencil->depth.writemask) { - - /* This is also efficient with sse / spe instructions: - */ for (j = 0; j < QUAD_SIZE; j++) { - if (quad->inout.mask & (1 << j)) { - bzzzz[j] = qzzzz[j]; - } + if (quad->inout.mask & (1 << j)) { + data->bzzzz[j] = data->qzzzz[j]; + } } + } - /* put updated Z values back into cached tile */ - switch (format) { - case PIPE_FORMAT_Z16_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - tile->data.depth16[y][x] = (ushort) bzzzz[j]; - } - break; - case PIPE_FORMAT_X8Z24_UNORM: - /* fall-through */ - /* (yes, this falls through to a different case than above) */ - case PIPE_FORMAT_Z32_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - tile->data.depth32[y][x] = bzzzz[j]; - } - break; - case PIPE_FORMAT_S8Z24_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint s8z24 = tile->data.depth32[y][x]; - s8z24 = (s8z24 & 0xff000000) | bzzzz[j]; - tile->data.depth32[y][x] = s8z24; - } - break; - case PIPE_FORMAT_Z24S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint z24s8 = tile->data.depth32[y][x]; - z24s8 = (z24s8 & 0xff) | (bzzzz[j] << 8); - tile->data.depth32[y][x] = z24s8; + return TRUE; +} + + + +/** + * Do stencil (and depth) testing. Stenciling depends on the outcome of + * depth testing. + */ +static boolean +depth_stencil_test_quad(struct quad_stage *qs, + struct depth_data *data, + struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + unsigned func, zFailOp, zPassOp, failOp; + ubyte ref, wrtMask, valMask; + uint face = quad->input.facing; + + if (!softpipe->depth_stencil->stencil[1].enabled) { + /* single-sided stencil test, use front (face=0) state */ + face = 0; + } + + /* choose front or back face function, operator, etc */ + /* XXX we could do these initializations once per primitive */ + func = softpipe->depth_stencil->stencil[face].func; + failOp = softpipe->depth_stencil->stencil[face].fail_op; + zFailOp = softpipe->depth_stencil->stencil[face].zfail_op; + zPassOp = softpipe->depth_stencil->stencil[face].zpass_op; + ref = softpipe->depth_stencil->stencil[face].ref_value; + wrtMask = softpipe->depth_stencil->stencil[face].writemask; + valMask = softpipe->depth_stencil->stencil[face].valuemask; + + + /* do the stencil test first */ + { + unsigned passMask, failMask; + passMask = do_stencil_test(data, func, ref, valMask); + failMask = quad->inout.mask & ~passMask; + quad->inout.mask &= passMask; + + if (failOp != PIPE_STENCIL_OP_KEEP) { + apply_stencil_op(data, failMask, failOp, ref, wrtMask); + } + } + + if (quad->inout.mask) { + /* now the pixels that passed the stencil test are depth tested */ + if (softpipe->depth_stencil->depth.enabled) { + const unsigned origMask = quad->inout.mask; + + depth_test_quad(qs, data, quad); /* quad->mask is updated */ + + /* update stencil buffer values according to z pass/fail result */ + if (zFailOp != PIPE_STENCIL_OP_KEEP) { + const unsigned failMask = origMask & ~quad->inout.mask; + apply_stencil_op(data, failMask, zFailOp, ref, wrtMask); } - break; - case PIPE_FORMAT_Z24X8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - tile->data.depth32[y][x] = bzzzz[j] << 8; + + if (zPassOp != PIPE_STENCIL_OP_KEEP) { + const unsigned passMask = origMask & quad->inout.mask; + apply_stencil_op(data, passMask, zPassOp, ref, wrtMask); } - break; - default: - assert(0); + } + else { + /* no depth test, apply Zpass operator to stencil buffer values */ + apply_stencil_op(data, quad->inout.mask, zPassOp, ref, wrtMask); } } - return TRUE; + return quad->inout.mask != 0; } +#define ALPHATEST( FUNC, COMP ) \ + static int \ + alpha_test_quads_##FUNC( struct quad_stage *qs, \ + struct quad_header *quads[], \ + unsigned nr ) \ + { \ + const float ref = qs->softpipe->depth_stencil->alpha.ref_value; \ + const uint cbuf = 0; /* only output[0].alpha is tested */ \ + unsigned pass_nr = 0; \ + unsigned i; \ + \ + for (i = 0; i < nr; i++) { \ + const float *aaaa = quads[i]->output.color[cbuf][3]; \ + unsigned passMask = 0; \ + \ + if (aaaa[0] COMP ref) passMask |= (1 << 0); \ + if (aaaa[1] COMP ref) passMask |= (1 << 1); \ + if (aaaa[2] COMP ref) passMask |= (1 << 2); \ + if (aaaa[3] COMP ref) passMask |= (1 << 3); \ + \ + quads[i]->inout.mask &= passMask; \ + \ + if (quads[i]->inout.mask) \ + quads[pass_nr++] = quads[i]; \ + } \ + \ + return pass_nr; \ + } + + +ALPHATEST( LESS, < ) +ALPHATEST( EQUAL, == ) +ALPHATEST( LEQUAL, <= ) +ALPHATEST( GREATER, > ) +ALPHATEST( NOTEQUAL, != ) +ALPHATEST( GEQUAL, >= ) + + +/* XXX: Incorporate into shader using KILP. + */ +static int +alpha_test_quads(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + switch (qs->softpipe->depth_stencil->alpha.func) { + case PIPE_FUNC_LESS: + return alpha_test_quads_LESS( qs, quads, nr ); + case PIPE_FUNC_EQUAL: + return alpha_test_quads_EQUAL( qs, quads, nr ); + break; + case PIPE_FUNC_LEQUAL: + return alpha_test_quads_LEQUAL( qs, quads, nr ); + case PIPE_FUNC_GREATER: + return alpha_test_quads_GREATER( qs, quads, nr ); + case PIPE_FUNC_NOTEQUAL: + return alpha_test_quads_NOTEQUAL( qs, quads, nr ); + case PIPE_FUNC_GEQUAL: + return alpha_test_quads_GEQUAL( qs, quads, nr ); + case PIPE_FUNC_ALWAYS: + return nr; + case PIPE_FUNC_NEVER: + default: + return 0; + } +} + +static unsigned mask_count[0x8] = +{ + 0, /* 0x0 */ + 1, /* 0x1 */ + 1, /* 0x2 */ + 2, /* 0x3 */ + 1, /* 0x4 */ + 2, /* 0x5 */ + 2, /* 0x6 */ + 3, /* 0x7 */ +}; + + + static void depth_test_quads(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { unsigned i, pass = 0; + const struct sp_fragment_shader *fs = qs->softpipe->fs; + boolean interp_depth = !fs->info.writes_z; + struct depth_data data; - for (i = 0; i < nr; i++) { - if (sp_depth_test_quad(qs, quads[i])) + + if (qs->softpipe->depth_stencil->alpha.enabled) { + nr = alpha_test_quads(qs, quads, nr); + } + + if (qs->softpipe->framebuffer.zsbuf && + (qs->softpipe->depth_stencil->depth.enabled || + qs->softpipe->depth_stencil->stencil[0].enabled)) { + + data.ps = qs->softpipe->framebuffer.zsbuf; + data.format = data.ps->format; + data.tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, + quads[0]->input.x0, + quads[0]->input.y0); + + for (i = 0; i < nr; i++) { + get_depth_stencil_values(&data, quads[i]); + + if (qs->softpipe->depth_stencil->depth.enabled) { + if (interp_depth) + interpolate_quad_depth(quads[i]); + + convert_quad_depth(&data, quads[i]); + } + + if (qs->softpipe->depth_stencil->stencil[0].enabled) { + if (!depth_stencil_test_quad(qs, &data, quads[i])) + continue; + } + else { + if (!depth_test_quad(qs, &data, quads[i])) + continue; + } + + if (qs->softpipe->depth_stencil->stencil[0].enabled || + qs->softpipe->depth_stencil->depth.writemask) + write_depth_stencil_values(&data, quads[i]); + + qs->softpipe->occlusion_count += mask_count[quads[i]->inout.mask]; quads[pass++] = quads[i]; + } + + nr = pass; } - - if (pass) - qs->next->run(qs->next, quads, pass); + + if (nr) + qs->next->run(qs->next, quads, nr); } diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c deleted file mode 100644 index 1048d44984..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_earlyz.c +++ /dev/null @@ -1,94 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -/** - * \brief Quad early-z testing - */ - -#include "pipe/p_defines.h" -#include "util/u_memory.h" -#include "sp_quad.h" -#include "sp_quad_pipe.h" - - -/** - * All this stage does is compute the quad's Z values (which is normally - * done by the shading stage). - * The next stage will do the actual depth test. - */ -static void -earlyz_quad( - struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr ) -{ - const float a0z = quads[0]->posCoef->a0[2]; - const float dzdx = quads[0]->posCoef->dadx[2]; - const float dzdy = quads[0]->posCoef->dady[2]; - unsigned i; - - for (i = 0; i < nr; i++) { - const float fx = (float) quads[i]->input.x0; - const float fy = (float) quads[i]->input.y0; - const float z0 = a0z + dzdx * fx + dzdy * fy; - - quads[i]->output.depth[0] = z0; - quads[i]->output.depth[1] = z0 + dzdx; - quads[i]->output.depth[2] = z0 + dzdy; - quads[i]->output.depth[3] = z0 + dzdx + dzdy; - } - - qs->next->run( qs->next, quads, nr ); -} - -static void -earlyz_begin( - struct quad_stage *qs ) -{ - qs->next->begin( qs->next ); -} - -static void -earlyz_destroy( - struct quad_stage *qs ) -{ - FREE( qs ); -} - -struct quad_stage * -sp_quad_earlyz_stage( - struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT( quad_stage ); - - stage->softpipe = softpipe; - stage->begin = earlyz_begin; - stage->run = earlyz_quad; - stage->destroy = earlyz_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c index ea5ed3bbd0..56a8f55d77 100644 --- a/src/gallium/drivers/softpipe/sp_quad_fs.c +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -111,24 +111,31 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) } } - if (!z_written) { - /* compute Z values now, as in the quad earlyz stage */ - /* XXX we should really only do this if the earlyz stage is not used */ - const float fx = (float) quad->input.x0; - const float fy = (float) quad->input.y0; - const float dzdx = quad->posCoef->dadx[2]; - const float dzdy = quad->posCoef->dady[2]; - const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy; - - quad->output.depth[0] = z0; - quad->output.depth[1] = z0 + dzdx; - quad->output.depth[2] = z0 + dzdy; - quad->output.depth[3] = z0 + dzdx + dzdy; - } - return TRUE; } + + +static void +coverage_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + uint cbuf; + + /* loop over colorbuffer outputs */ + for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) { + float (*quadColor)[4] = quad->output.color[cbuf]; + unsigned j; + for (j = 0; j < QUAD_SIZE; j++) { + assert(quad->input.coverage[j] >= 0.0); + assert(quad->input.coverage[j] <= 1.0); + quadColor[3][j] *= quad->input.coverage[j]; + } + } +} + + + static void shade_quads(struct quad_stage *qs, struct quad_header *quads[], @@ -144,8 +151,13 @@ shade_quads(struct quad_stage *qs, machine->InterpCoefs = quads[0]->coef; for (i = 0; i < nr; i++) { - if (shade_quad(qs, quads[i])) - quads[pass++] = quads[i]; + if (!shade_quad(qs, quads[i])) + continue; + + if (/*do_coverage*/ 0) + coverage_quad( qs, quads[i] ); + + quads[pass++] = quads[i]; } if (pass) diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/drivers/softpipe/sp_quad_occlusion.c deleted file mode 100644 index 4adeb16546..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_occlusion.c +++ /dev/null @@ -1,87 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - - -/** - * \brief Quad occlusion counter stage - * \author Brian Paul - */ - - -#include "pipe/p_defines.h" -#include "util/u_memory.h" -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_quad_pipe.h" - -static unsigned count_bits( unsigned val ) -{ - unsigned i; - - for (i = 0; val ; val >>= 1) - i += (val & 1); - - return i; -} - -static void -occlusion_count_quads(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) -{ - struct softpipe_context *softpipe = qs->softpipe; - unsigned i; - - for (i = 0; i < nr; i++) - softpipe->occlusion_count += count_bits(quads[i]->inout.mask); - - qs->next->run(qs->next, quads, nr); -} - - -static void occlusion_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void occlusion_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = occlusion_begin; - stage->run = occlusion_count_quads; - stage->destroy = occlusion_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c index d138d417ac..1b5bab4eca 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.c +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c @@ -38,18 +38,6 @@ sp_push_quad_first( struct softpipe_context *sp, sp->quad.first = quad; } -static void -sp_build_depth_stencil( struct softpipe_context *sp ) -{ - if (sp->depth_stencil->stencil[0].enabled || - sp->depth_stencil->stencil[1].enabled) { - sp_push_quad_first( sp, sp->quad.stencil_test ); - } - else if (sp->depth_stencil->depth.enabled && - sp->framebuffer.zsbuf) { - sp_push_quad_first( sp, sp->quad.depth_test ); - } -} void sp_build_quad_pipeline(struct softpipe_context *sp) @@ -61,37 +49,15 @@ sp_build_quad_pipeline(struct softpipe_context *sp) !sp->fs->info.uses_kill && !sp->fs->info.writes_z; - /* build up the pipeline in reverse order... */ - - /* Color combine - */ sp->quad.first = sp->quad.blend; - /* Shade/Depth/Stencil/Alpha - */ - if ((sp->rasterizer->poly_smooth && sp->reduced_prim == PIPE_PRIM_TRIANGLES) || - (sp->rasterizer->line_smooth && sp->reduced_prim == PIPE_PRIM_LINES) || - (sp->rasterizer->point_smooth && sp->reduced_prim == PIPE_PRIM_POINTS)) { - sp_push_quad_first( sp, sp->quad.coverage ); - } - - if (sp->active_query_count) { - sp_push_quad_first( sp, sp->quad.occlusion ); - } - - if (!early_depth_test) { - sp_build_depth_stencil( sp ); - } - - if (sp->depth_stencil->alpha.enabled) { - sp_push_quad_first( sp, sp->quad.alpha_test ); - } - - sp_push_quad_first( sp, sp->quad.shade ); - if (early_depth_test) { - sp_build_depth_stencil( sp ); - sp_push_quad_first( sp, sp->quad.earlyz ); + sp_push_quad_first( sp, sp->quad.shade ); + sp_push_quad_first( sp, sp->quad.depth_test ); + } + else { + sp_push_quad_first( sp, sp->quad.depth_test ); + sp_push_quad_first( sp, sp->quad.shade ); } } diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.h b/src/gallium/drivers/softpipe/sp_quad_pipe.h index add31ba705..c0aa134831 100644 --- a/src/gallium/drivers/softpipe/sp_quad_pipe.h +++ b/src/gallium/drivers/softpipe/sp_quad_pipe.h @@ -69,6 +69,4 @@ struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe ); void sp_build_quad_pipeline(struct softpipe_context *sp); -boolean sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad); - #endif /* SP_QUAD_PIPE_H */ diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c deleted file mode 100644 index d9ee80e59a..0000000000 --- a/src/gallium/drivers/softpipe/sp_quad_stencil.c +++ /dev/null @@ -1,363 +0,0 @@ - -/** - * \brief Quad stencil testing - */ - - -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_tile_cache.h" -#include "sp_quad_pipe.h" -#include "pipe/p_defines.h" -#include "util/u_memory.h" - - -/** Only 8-bit stencil supported */ -#define STENCIL_MAX 0xff - - -/** - * Do the basic stencil test (compare stencil buffer values against the - * reference value. - * - * \param stencilVals the stencil values from the stencil buffer - * \param func the stencil func (PIPE_FUNC_x) - * \param ref the stencil reference value - * \param valMask the stencil value mask indicating which bits of the stencil - * values and ref value are to be used. - * \return mask indicating which pixels passed the stencil test - */ -static unsigned -do_stencil_test(const ubyte stencilVals[QUAD_SIZE], unsigned func, - unsigned ref, unsigned valMask) -{ - unsigned passMask = 0x0; - unsigned j; - - ref &= valMask; - - switch (func) { - case PIPE_FUNC_NEVER: - /* passMask = 0x0 */ - break; - case PIPE_FUNC_LESS: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref < (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_EQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref == (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_LEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref <= (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_GREATER: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref > (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_NOTEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref != (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_GEQUAL: - for (j = 0; j < QUAD_SIZE; j++) { - if (ref >= (stencilVals[j] & valMask)) { - passMask |= (1 << j); - } - } - break; - case PIPE_FUNC_ALWAYS: - passMask = MASK_ALL; - break; - default: - assert(0); - } - - return passMask; -} - - -/** - * Apply the stencil operator to stencil values. - * - * \param stencilVals the stencil buffer values (read and written) - * \param mask indicates which pixels to update - * \param op the stencil operator (PIPE_STENCIL_OP_x) - * \param ref the stencil reference value - * \param wrtMask writemask controlling which bits are changed in the - * stencil values - */ -static void -apply_stencil_op(ubyte stencilVals[QUAD_SIZE], - unsigned mask, unsigned op, ubyte ref, ubyte wrtMask) -{ - unsigned j; - ubyte newstencil[QUAD_SIZE]; - - for (j = 0; j < QUAD_SIZE; j++) { - newstencil[j] = stencilVals[j]; - } - - switch (op) { - case PIPE_STENCIL_OP_KEEP: - /* no-op */ - break; - case PIPE_STENCIL_OP_ZERO: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - newstencil[j] = 0; - } - } - break; - case PIPE_STENCIL_OP_REPLACE: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - newstencil[j] = ref; - } - } - break; - case PIPE_STENCIL_OP_INCR: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - if (stencilVals[j] < STENCIL_MAX) { - newstencil[j] = stencilVals[j] + 1; - } - } - } - break; - case PIPE_STENCIL_OP_DECR: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - if (stencilVals[j] > 0) { - newstencil[j] = stencilVals[j] - 1; - } - } - } - break; - case PIPE_STENCIL_OP_INCR_WRAP: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - newstencil[j] = stencilVals[j] + 1; - } - } - break; - case PIPE_STENCIL_OP_DECR_WRAP: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - newstencil[j] = stencilVals[j] - 1; - } - } - break; - case PIPE_STENCIL_OP_INVERT: - for (j = 0; j < QUAD_SIZE; j++) { - if (mask & (1 << j)) { - newstencil[j] = ~stencilVals[j]; - } - } - break; - default: - assert(0); - } - - /* - * update the stencil values - */ - if (wrtMask != STENCIL_MAX) { - /* apply bit-wise stencil buffer writemask */ - for (j = 0; j < QUAD_SIZE; j++) { - stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & stencilVals[j]); - } - } - else { - for (j = 0; j < QUAD_SIZE; j++) { - stencilVals[j] = newstencil[j]; - } - } -} - - -/** - * Do stencil (and depth) testing. Stenciling depends on the outcome of - * depth testing. - */ -static void -stencil_test_quad(struct quad_stage *qs, struct quad_header *quads[], - unsigned nr) -{ - struct softpipe_context *softpipe = qs->softpipe; - struct pipe_surface *ps = softpipe->framebuffer.zsbuf; - unsigned func, zFailOp, zPassOp, failOp; - ubyte ref, wrtMask, valMask; - ubyte stencilVals[QUAD_SIZE]; - struct softpipe_cached_tile *tile - = sp_get_cached_tile(softpipe->zsbuf_cache, - quads[0]->input.x0, - quads[0]->input.y0); - uint face = quads[0]->input.facing; - uint pass = 0; - uint j, q; - - if (!softpipe->depth_stencil->stencil[1].enabled) { - /* single-sided stencil test, use front (face=0) state */ - face = 0; - } - - /* choose front or back face function, operator, etc */ - /* XXX we could do these initializations once per primitive */ - func = softpipe->depth_stencil->stencil[face].func; - failOp = softpipe->depth_stencil->stencil[face].fail_op; - zFailOp = softpipe->depth_stencil->stencil[face].zfail_op; - zPassOp = softpipe->depth_stencil->stencil[face].zpass_op; - ref = softpipe->depth_stencil->stencil[face].ref_value; - wrtMask = softpipe->depth_stencil->stencil[face].writemask; - valMask = softpipe->depth_stencil->stencil[face].valuemask; - - assert(ps); /* shouldn't get here if there's no stencil buffer */ - - for (q = 0; q < nr; q++) { - struct quad_header *quad = quads[q]; - - /* get stencil values from cached tile */ - switch (ps->format) { - case PIPE_FORMAT_S8Z24_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.depth32[y][x] >> 24; - } - break; - case PIPE_FORMAT_Z24S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.depth32[y][x] & 0xff; - } - break; - case PIPE_FORMAT_S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - stencilVals[j] = tile->data.stencil8[y][x]; - } - break; - default: - assert(0); - } - - /* do the stencil test first */ - { - unsigned passMask, failMask; - passMask = do_stencil_test(stencilVals, func, ref, valMask); - failMask = quad->inout.mask & ~passMask; - quad->inout.mask &= passMask; - - if (failOp != PIPE_STENCIL_OP_KEEP) { - apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask); - } - } - - if (quad->inout.mask) { - /* now the pixels that passed the stencil test are depth tested */ - if (softpipe->depth_stencil->depth.enabled) { - const unsigned origMask = quad->inout.mask; - - sp_depth_test_quad(qs, quad); /* quad->mask is updated */ - - /* update stencil buffer values according to z pass/fail result */ - if (zFailOp != PIPE_STENCIL_OP_KEEP) { - const unsigned failMask = origMask & ~quad->inout.mask; - apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask); - } - - if (zPassOp != PIPE_STENCIL_OP_KEEP) { - const unsigned passMask = origMask & quad->inout.mask; - apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask); - } - } - else { - /* no depth test, apply Zpass operator to stencil buffer values */ - apply_stencil_op(stencilVals, quad->inout.mask, zPassOp, ref, wrtMask); - } - - } - - /* put new stencil values into cached tile */ - switch (ps->format) { - case PIPE_FORMAT_S8Z24_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint s8z24 = tile->data.depth32[y][x]; - s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff); - tile->data.depth32[y][x] = s8z24; - } - break; - case PIPE_FORMAT_Z24S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - uint z24s8 = tile->data.depth32[y][x]; - z24s8 = (z24s8 & 0xffffff00) | stencilVals[j]; - tile->data.depth32[y][x] = z24s8; - } - break; - case PIPE_FORMAT_S8_UNORM: - for (j = 0; j < QUAD_SIZE; j++) { - int x = quad->input.x0 % TILE_SIZE + (j & 1); - int y = quad->input.y0 % TILE_SIZE + (j >> 1); - tile->data.stencil8[y][x] = stencilVals[j]; - } - break; - default: - assert(0); - } - - if (quad->inout.mask) - quads[pass++] = quad; - } - - if (pass) - qs->next->run(qs->next, quads, pass); -} - - -static void stencil_begin(struct quad_stage *qs) -{ - qs->next->begin(qs->next); -} - - -static void stencil_destroy(struct quad_stage *qs) -{ - FREE( qs ); -} - - -struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe ) -{ - struct quad_stage *stage = CALLOC_STRUCT(quad_stage); - - stage->softpipe = softpipe; - stage->begin = stencil_begin; - stage->run = stencil_test_quad; - stage->destroy = stencil_destroy; - - return stage; -} diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 0226501267..8654069bde 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -246,9 +246,7 @@ void softpipe_update_derived( struct softpipe_context *softpipe ) if (softpipe->dirty & (SP_NEW_BLEND | SP_NEW_DEPTH_STENCIL_ALPHA | SP_NEW_FRAMEBUFFER | - SP_NEW_RASTERIZER | - SP_NEW_FS | - SP_NEW_QUERY)) + SP_NEW_FS)) sp_build_quad_pipeline(softpipe); softpipe->dirty = 0; -- cgit v1.2.3 From 1078844d18367b4259cd3b6a3a73e3cd72ea019f Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 27 Jul 2009 11:23:51 +0100 Subject: softpipe: cope with nr_cbufs == 0 Disable blend code when no color buffer --- src/gallium/drivers/softpipe/sp_quad_blend.c | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index e1f0e77255..e243c63fa2 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -924,6 +924,13 @@ single_output_color(struct quad_stage *qs, } } +static void +blend_noop(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ +} + static void choose_blend_quad(struct quad_stage *qs, @@ -934,9 +941,12 @@ choose_blend_quad(struct quad_stage *qs, const struct pipe_blend_state *blend = softpipe->blend; qs->run = blend_fallback; - - if (!softpipe->blend->logicop_enable && - softpipe->blend->colormask == 0xf) + + if (softpipe->framebuffer.nr_cbufs == 0) { + qs->run = blend_noop; + } + else if (!softpipe->blend->logicop_enable && + softpipe->blend->colormask == 0xf) { if (!blend->blend_enable) { qs->run = single_output_color; -- cgit v1.2.3 From c61145820556833dccd728eb6df3397bec7f70da Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 27 Jul 2009 12:11:16 +0100 Subject: softpipe: fastpath for interpolated z16 less depthtesting Because this is interpolated (ie. early) depth, we can build in an assumption about the quads emitted by triangle setup, ie that they are actually linear spans. Interpolate z over those spans in z16 format to save on math & conversion. --- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 143 +++++++++++++++++++++- 1 file changed, 139 insertions(+), 4 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index bf65799a81..506867f4d0 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -646,9 +646,9 @@ static unsigned mask_count[0x8] = static void -depth_test_quads(struct quad_stage *qs, - struct quad_header *quads[], - unsigned nr) +depth_test_quads_fallback(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) { unsigned i, pass = 0; const struct sp_fragment_shader *fs = qs->softpipe->fs; @@ -704,9 +704,144 @@ depth_test_quads(struct quad_stage *qs, qs->next->run(qs->next, quads, nr); } +/* XXX: this function assumes setup function actually emits linear + * spans of quads. It seems a lot more natural to do (early) + * depth-testing on spans rather than quads. + */ +static void +depth_interp_z16_less_write(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + unsigned i, pass = 0; + const unsigned ix = quads[0]->input.x0; + const unsigned iy = quads[0]->input.y0; + const float fx = (float) ix; + const float fy = (float) iy; + const float dzdx = quads[0]->posCoef->dadx[2]; + const float dzdy = quads[0]->posCoef->dady[2]; + const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy; + struct softpipe_cached_tile *tile; + ushort (*depth16)[TILE_SIZE]; + ushort idepth[4], depth_step; + const float scale = 65535.0; + + idepth[0] = (ushort)((z0) * scale); + idepth[1] = (ushort)((z0 + dzdx) * scale); + idepth[2] = (ushort)((z0 + dzdy) * scale); + idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale); + + depth_step = (ushort)(dzdx * 2 * scale); + + tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy); + + depth16 = (ushort (*)[TILE_SIZE]) + &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE]; + + for (i = 0; i < nr; i++) { + unsigned outmask = quads[i]->inout.mask; + unsigned mask = 0; + + if ((outmask & 1) && idepth[0] < depth16[0][0]) { + depth16[0][0] = idepth[0]; + mask |= (1 << 0); + } + + if ((outmask & 2) && idepth[1] < depth16[0][1]) { + depth16[0][1] = idepth[1]; + mask |= (1 << 1); + } + + if ((outmask & 4) && idepth[2] < depth16[1][0]) { + depth16[1][0] = idepth[2]; + mask |= (1 << 2); + } + + if ((outmask & 8) && idepth[3] < depth16[1][1]) { + depth16[1][1] = idepth[3]; + mask |= (1 << 3); + } + + idepth[0] += depth_step; + idepth[1] += depth_step; + idepth[2] += depth_step; + idepth[3] += depth_step; + + depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2]; + + quads[i]->inout.mask = mask; + if (quads[i]->inout.mask) + quads[pass++] = quads[i]; + } + + if (pass) + qs->next->run(qs->next, quads, pass); + +} + + +static void +depth_noop(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + qs->next->run(qs->next, quads, nr); +} + + + +static void +choose_depth_test(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + boolean interp_depth = !qs->softpipe->fs->info.writes_z; + + boolean alpha = qs->softpipe->depth_stencil->alpha.enabled; + + boolean depth = (qs->softpipe->framebuffer.zsbuf && + qs->softpipe->depth_stencil->depth.enabled); + + unsigned depthfunc = qs->softpipe->depth_stencil->depth.func; + + boolean stencil = qs->softpipe->depth_stencil->stencil[0].enabled; + + boolean depthwrite = qs->softpipe->depth_stencil->depth.writemask; + + + qs->run = depth_test_quads_fallback; + + if (!alpha && + !depth && + !stencil) { + qs->run = depth_noop; + } + else if (!alpha && + interp_depth && + depth && + depthfunc == PIPE_FUNC_LESS && + depthwrite && + !stencil) + { + switch (qs->softpipe->framebuffer.zsbuf->format) { + case PIPE_FORMAT_Z16_UNORM: + qs->run = depth_interp_z16_less_write; + break; + default: + break; + } + } + + qs->run( qs, quads, nr ); +} + + + + static void depth_test_begin(struct quad_stage *qs) { + qs->run = choose_depth_test; qs->next->begin(qs->next); } @@ -723,7 +858,7 @@ struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe ) stage->softpipe = softpipe; stage->begin = depth_test_begin; - stage->run = depth_test_quads; + stage->run = choose_depth_test; stage->destroy = depth_test_destroy; return stage; -- cgit v1.2.3 From 6142de393fe34ff0866f8489f1292eb473276f11 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 27 Jul 2009 12:44:58 +0100 Subject: softpipe: example fast paths for simple samplers All these fastpaths are examples of the types of things we'd code-generate in a more sophisticated version of softpipe. --- src/gallium/drivers/softpipe/sp_state_derived.c | 1 + src/gallium/drivers/softpipe/sp_tex_sample.c | 327 +++++++++++++++++++++++- src/gallium/drivers/softpipe/sp_tex_sample.h | 10 +- 3 files changed, 333 insertions(+), 5 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 8654069bde..3ed1de7e17 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -208,6 +208,7 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i]; softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i]; + softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples_fragment; } for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 46c56b0c83..8248576e98 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -662,10 +662,64 @@ choose_mipmap_levels(const struct pipe_texture *texture, * XXX maybe move this into sp_tile_cache.c and merge with the * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... */ +static void +get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, int x, int y, + const float *out[4]) +{ + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + + const struct softpipe_cached_tile *tile + = sp_get_cached_tile_tex(samp->cache, + tile_address(x, y, 0, face, level)); + + y %= TILE_SIZE; + x %= TILE_SIZE; + + out[0] = &tile->data.color[y ][x ][0]; + out[1] = &tile->data.color[y ][x+1][0]; + out[2] = &tile->data.color[y+1][x ][0]; + out[3] = &tile->data.color[y+1][x+1][0]; +} + +static INLINE const float * +get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, int x, int y) +{ + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + + const struct softpipe_cached_tile *tile + = sp_get_cached_tile_tex(samp->cache, + tile_address(x, y, 0, face, level)); + + y %= TILE_SIZE; + x %= TILE_SIZE; + + return &tile->data.color[y][x][0]; +} + + +static void +get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, + unsigned face, unsigned level, + int x0, int y0, + int x1, int y1, + const float *out[4]) +{ + unsigned i; + + for (i = 0; i < 4; i++) { + unsigned tx = (i & 1) ? x1 : x0; + unsigned ty = (i >> 1) ? y1 : y0; + + out[i] = get_texel_2d_ptr( tgsi_sampler, face, level, tx, ty ); + } +} + static void get_texel(const struct tgsi_sampler *tgsi_sampler, - unsigned face, unsigned level, int x, int y, int z, - float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) + unsigned face, unsigned level, int x, int y, int z, + float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; @@ -825,6 +879,193 @@ shadow_compare4(const struct pipe_sampler_state *sampler, } + +static void +sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot - 0.5F; + float v = t[j] * ypot - 0.5F; + + int uflr = util_ifloor(u); + int vflr = util_ifloor(v); + + float xw = u - (float)uflr; + float yw = v - (float)vflr; + + int x0 = uflr & (xpot - 1); + int y0 = vflr & (ypot - 1); + + const float *tx[4]; + + + /* Can we fetch all four at once: + */ + if (x0 % TILE_SIZE != TILE_SIZE-1 && + y0 % TILE_SIZE != TILE_SIZE-1) + { + get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx); + } + else + { + unsigned x1 = (uflr + 1) & (xpot - 1); + unsigned y1 = (vflr + 1) & (ypot - 1); + get_texel_quad_2d_mt(tgsi_sampler, 0, level, + x0, y0, x1, y1, tx); + } + + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw, yw, + tx[0][c], tx[1][c], + tx[2][c], tx[3][c]); + } + } +} + + +static void +sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot - 0.5F; + float v = t[j] * ypot - 0.5F; + + int uflr = util_ifloor(u); + int vflr = util_ifloor(v); + + int x0 = uflr & (xpot - 1); + int y0 = vflr & (ypot - 1); + + const float *out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } + } +} + + +static void +sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + unsigned j; + unsigned level = samp->level; + unsigned xpot = (1<xpot); + unsigned ypot = (1<ypot); + + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot - 0.5F; + float v = t[j] * ypot - 0.5F; + + int x0, y0; + const float *out; + + x0 = util_ifloor(u); + if (x0 < 0) + x0 = 0; + else if (x0 > xpot - 1) + x0 = xpot - 1; + + y0 = util_ifloor(v); + if (y0 < 0) + y0 = 0; + else if (y0 > ypot - 1) + y0 = ypot - 1; + + out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } + } +} + + +static void +sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + int level0, level1; + float lambda; + + lambda = compute_lambda(texture, sampler, s, t, p, lodbias); + level0 = (int)lambda; + level1 = level0 + 1; + + if (lambda < 0.0) { + samp->level = 0; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, lodbias, rgba ); + } + else if (level0 >= texture->last_level) { + samp->level = texture->last_level; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, lodbias, rgba ); + } + else { + float rgba0[4][4]; + float rgba1[4][4]; + int c,j; + + float levelBlend = lambda - level0; /* blending weight between levels */ + + samp->level = level0; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, lodbias, rgba0 ); + + samp->level++; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, lodbias, rgba1 ); + + for (c = 0; c < 4; c++) + for (j = 0; j < 4; j++) + rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); + } +} + /** * Common code for sampling 1D/2D/cube textures. * Could probably extend for 3D... @@ -1254,6 +1495,16 @@ sp_get_samples(struct tgsi_sampler *tgsi_sampler, #endif } +static void +sp_get_samples_fallback(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba); +} /** * Called via tgsi_sampler::get_samples() when running a fragment shader. @@ -1267,7 +1518,77 @@ sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { - sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba); + struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + + tgsi_sampler->get_samples = sp_get_samples_fallback; + + /* Try to hook in a faster sampler. Ultimately we'll have to + * code-generate these. Luckily most of this looks like it is + * orthogonal state within the sampler. + */ + if (texture->target == PIPE_TEXTURE_2D && + sampler->min_img_filter == sampler->mag_img_filter && + sampler->wrap_s == sampler->wrap_t && + sampler->compare_mode == FALSE && + sampler->normalized_coords) + { + samp->xpot = util_unsigned_logbase2( samp->texture->width[0] ); + samp->ypot = util_unsigned_logbase2( samp->texture->height[0] ); + + if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { + samp->level = CLAMP((int) sampler->min_lod, + 0, (int) texture->last_level); + + if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_NEAREST: + tgsi_sampler->get_samples = sp_get_samples_2d_nearest_repeat_POT; + break; + case PIPE_TEX_FILTER_LINEAR: + tgsi_sampler->get_samples = sp_get_samples_2d_linear_repeat_POT; + break; + default: + break; + } + } + else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_NEAREST: + tgsi_sampler->get_samples = sp_get_samples_2d_nearest_clamp_POT; + break; + default: + break; + } + } + } + else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { + if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { + switch (sampler->min_img_filter) { + case PIPE_TEX_FILTER_LINEAR: + /* This one not working yet: + */ + if (0) + tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT; + break; + default: + break; + } + } + } + } + else if (0) { + _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n", + texture->target, PIPE_TEXTURE_2D, + sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE, + sampler->min_img_filter, sampler->mag_img_filter, + sampler->wrap_s, sampler->wrap_t, + sampler->compare_mode, FALSE, + sampler->normalized_coords, TRUE); + } + + tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba ); } diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index 3c5beb560f..0650c7830b 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -39,6 +39,12 @@ struct sp_shader_sampler { struct tgsi_sampler base; /**< base class */ + /* For sp_get_samples_2d_linear_POT: + */ + unsigned xpot; + unsigned ypot; + unsigned level; + const struct pipe_texture *texture; const struct pipe_sampler_state *sampler; @@ -47,10 +53,10 @@ struct sp_shader_sampler -static INLINE const struct sp_shader_sampler * +static INLINE struct sp_shader_sampler * sp_shader_sampler(const struct tgsi_sampler *sampler) { - return (const struct sp_shader_sampler *) sampler; + return (struct sp_shader_sampler *) sampler; } -- cgit v1.2.3 From 5fdac2dcea09c654725666b3cab5f59dfc9e31a5 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Mon, 27 Jul 2009 15:51:15 +0100 Subject: softpipe: fix off-by-one in nearest texcoord routines Stray '- 0.5' copied from linear versions. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 8248576e98..4651d781a9 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -955,8 +955,8 @@ sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, for (j = 0; j < QUAD_SIZE; j++) { int c; - float u = s[j] * xpot - 0.5F; - float v = t[j] * ypot - 0.5F; + float u = s[j] * xpot; + float v = t[j] * ypot; int uflr = util_ifloor(u); int vflr = util_ifloor(v); @@ -990,8 +990,8 @@ sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, for (j = 0; j < QUAD_SIZE; j++) { int c; - float u = s[j] * xpot - 0.5F; - float v = t[j] * ypot - 0.5F; + float u = s[j] * xpot; + float v = t[j] * ypot; int x0, y0; const float *out; -- cgit v1.2.3 From 572c2fb5bb6cec71ef42e93416251a6a6c183de0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 30 Jul 2009 11:34:36 +0100 Subject: softpipe: remove unused variable in shade_quad --- src/gallium/drivers/softpipe/sp_quad_fs.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c index 56a8f55d77..e1bc0712de 100644 --- a/src/gallium/drivers/softpipe/sp_quad_fs.c +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -74,7 +74,6 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) struct quad_shade_stage *qss = quad_shade_stage( qs ); struct softpipe_context *softpipe = qs->softpipe; struct tgsi_exec_machine *machine = qss->machine; - boolean z_written; /* run shader */ quad->inout.mask &= softpipe->fs->run( softpipe->fs, machine, quad ); @@ -82,7 +81,6 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) return FALSE; /* store outputs */ - z_written = FALSE; { const ubyte *sem_name = softpipe->fs->info.output_semantic_name; const ubyte *sem_index = softpipe->fs->info.output_semantic_index; @@ -104,7 +102,6 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) for (j = 0; j < 4; j++) { quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j]; } - z_written = TRUE; } break; } -- cgit v1.2.3 From 73a6178a73a4cc34195348a537d3f94aab6a43e1 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 30 Jul 2009 11:35:08 +0100 Subject: softpipe: add depth-lequal z16 path --- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 107 ++++++++++++++++++++-- 1 file changed, 100 insertions(+), 7 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index 506867f4d0..9cffea2c9e 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -780,6 +780,81 @@ depth_interp_z16_less_write(struct quad_stage *qs, } +static void +depth_interp_z16_lequal_write(struct quad_stage *qs, + struct quad_header *quads[], + unsigned nr) +{ + unsigned i, pass = 0; + const unsigned ix = quads[0]->input.x0; + const unsigned iy = quads[0]->input.y0; + const float fx = (float) ix; + const float fy = (float) iy; + const float dzdx = quads[0]->posCoef->dadx[2]; + const float dzdy = quads[0]->posCoef->dady[2]; + const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy; + struct softpipe_cached_tile *tile; + ushort (*depth16)[TILE_SIZE]; + ushort idepth[4], depth_step; + const float scale = 65535.0; + + idepth[0] = (ushort)((z0) * scale); + idepth[1] = (ushort)((z0 + dzdx) * scale); + idepth[2] = (ushort)((z0 + dzdy) * scale); + idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale); + + depth_step = (ushort)(dzdx * 2 * scale); + + tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy); + + depth16 = (ushort (*)[TILE_SIZE]) + &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE]; + + for (i = 0; i < nr; i++) { + unsigned outmask = quads[i]->inout.mask; + unsigned mask = 0; + + if ((outmask & 1) && idepth[0] <= depth16[0][0]) { + depth16[0][0] = idepth[0]; + mask |= (1 << 0); + } + + if ((outmask & 2) && idepth[1] <= depth16[0][1]) { + depth16[0][1] = idepth[1]; + mask |= (1 << 1); + } + + if ((outmask & 4) && idepth[2] <= depth16[1][0]) { + depth16[1][0] = idepth[2]; + mask |= (1 << 2); + } + + if ((outmask & 8) && idepth[3] <= depth16[1][1]) { + depth16[1][1] = idepth[3]; + mask |= (1 << 3); + } + + idepth[0] += depth_step; + idepth[1] += depth_step; + idepth[2] += depth_step; + idepth[3] += depth_step; + + depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2]; + + quads[i]->inout.mask = mask; + if (quads[i]->inout.mask) + quads[pass++] = quads[i]; + } + + if (pass) + qs->next->run(qs->next, quads, pass); + +} + + + + + static void depth_noop(struct quad_stage *qs, struct quad_header *quads[], @@ -809,8 +884,6 @@ choose_depth_test(struct quad_stage *qs, boolean depthwrite = qs->softpipe->depth_stencil->depth.writemask; - qs->run = depth_test_quads_fallback; - if (!alpha && !depth && !stencil) { @@ -819,18 +892,38 @@ choose_depth_test(struct quad_stage *qs, else if (!alpha && interp_depth && depth && - depthfunc == PIPE_FUNC_LESS && depthwrite && !stencil) { - switch (qs->softpipe->framebuffer.zsbuf->format) { - case PIPE_FORMAT_Z16_UNORM: - qs->run = depth_interp_z16_less_write; + switch (depthfunc) { + case PIPE_FUNC_LESS: + switch (qs->softpipe->framebuffer.zsbuf->format) { + case PIPE_FORMAT_Z16_UNORM: + qs->run = depth_interp_z16_less_write; + break; + default: + qs->run = depth_test_quads_fallback; + break; + } break; - default: + case PIPE_FUNC_LEQUAL: + switch (qs->softpipe->framebuffer.zsbuf->format) { + case PIPE_FORMAT_Z16_UNORM: + qs->run = depth_interp_z16_lequal_write; + break; + default: + qs->run = depth_test_quads_fallback; + break; + } break; + default: + qs->run = depth_test_quads_fallback; } } + else { + qs->run = depth_test_quads_fallback; + } + qs->run( qs, quads, nr ); } -- cgit v1.2.3 From 1295cf423e21dad04a947960782ffa8db2739709 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 30 Jul 2009 11:35:50 +0100 Subject: softpipe: rearrange blend fastpaths --- src/gallium/drivers/softpipe/sp_quad_blend.c | 82 +++++++++------------------- 1 file changed, 27 insertions(+), 55 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index e243c63fa2..b8ed086734 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -793,10 +793,7 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { - static const float one[4] = { 1, 1, 1, 1 }; - float one_minus_alpha[QUAD_SIZE]; - float dest[4][QUAD_SIZE]; - float source[4][QUAD_SIZE]; + float source[4]; uint i, j, q; struct softpipe_cached_tile *tile @@ -806,45 +803,26 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, for (q = 0; q < nr; q++) { struct quad_header *quad = quads[q]; - float (*quadColor)[4] = quad->output.color[0]; - const float *alpha = quadColor[3]; const int itx = (quad->input.x0 & (TILE_SIZE-1)); const int ity = (quad->input.y0 & (TILE_SIZE-1)); + float (*swzColor)[4] = quad->output.color[0]; - /* get/swizzle dest colors */ - for (j = 0; j < QUAD_SIZE; j++) { - int x = itx + (j & 1); - int y = ity + (j >> 1); - for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; - } - } - - VEC4_MUL(source[0], quadColor[0], alpha); /* R */ - VEC4_MUL(source[1], quadColor[1], alpha); /* G */ - VEC4_MUL(source[2], quadColor[2], alpha); /* B */ - VEC4_MUL(source[3], quadColor[3], alpha); /* A */ + for (j = 0; j < 4; j++) { + if (quad->inout.mask & (1<data.color[ity + (j>>1)][itx + (j&1)]; + const float alpha = swzColor[3][j]; + const float one_minus_alpha = 1.0 - alpha; - VEC4_SUB(one_minus_alpha, one, alpha); - VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ - VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ - VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ - VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* B */ - - VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ - VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ - VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ - VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ + for (i = 0; i < 4; i++) { + dest[i] *= one_minus_alpha; + dest[i] += swzColor[i][j] * alpha; - for (j = 0; j < QUAD_SIZE; j++) { - if (quad->inout.mask & (1 << j)) { - int x = itx + (j & 1); - int y = ity + (j >> 1); - for (i = 0; i < 4; i++) { /* loop over color chans */ - tile->data.color[y][x][i] = quadColor[i][j]; + /* XXX: redundant, will be clamped later for argb8 surfaces: + */ + dest[i] = CLAMP(dest[i], 0.0, 1.0); } } - } + } } } @@ -863,33 +841,27 @@ blend_single_add_one_one(struct quad_stage *qs, for (q = 0; q < nr; q++) { struct quad_header *quad = quads[q]; - float (*quadColor)[4] = quad->output.color[0]; const int itx = (quad->input.x0 & (TILE_SIZE-1)); const int ity = (quad->input.y0 & (TILE_SIZE-1)); - + float (*dest)[64][4] = (float (*)[64][4])&tile->data.color[ity][itx]; + float (*swzColor)[4] = quad->output.color[0]; + float quadColor[4][4]; + /* get/swizzle dest colors */ for (j = 0; j < QUAD_SIZE; j++) { - int x = itx + (j & 1); - int y = ity + (j >> 1); for (i = 0; i < 4; i++) { - dest[i][j] = tile->data.color[y][x][i]; + quadColor[i][j] = swzColor[j][i]; } } - VEC4_ADD_SAT(quadColor[0], quadColor[0], dest[0]); /* R */ - VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */ - VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */ - VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */ - - for (j = 0; j < QUAD_SIZE; j++) { - if (quad->inout.mask & (1 << j)) { - int x = itx + (j & 1); - int y = ity + (j >> 1); - for (i = 0; i < 4; i++) { /* loop over color chans */ - tile->data.color[y][x][i] = quadColor[i][j]; - } - } - } + if (quad->inout.mask & 1) + VEC4_ADD_SAT(dest[0][0], quadColor[0], dest[0][0]); + if (quad->inout.mask & 2) + VEC4_ADD_SAT(dest[0][1], quadColor[1], dest[0][1]); + if (quad->inout.mask & 4) + VEC4_ADD_SAT(dest[1][0], quadColor[2], dest[1][0]); + if (quad->inout.mask & 8) + VEC4_ADD_SAT(dest[1][1], quadColor[3], dest[1][1]); } } -- cgit v1.2.3 From 95f7ed4638d4e379783abdd5b250e203b6b1b435 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 30 Jul 2009 11:59:32 +0100 Subject: softpipe: setup quad outputs from with fs->run --- src/gallium/drivers/softpipe/sp_fs_exec.c | 34 ++++++++++++++++++++++++++++- src/gallium/drivers/softpipe/sp_fs_sse.c | 35 +++++++++++++++++++++++++++++- src/gallium/drivers/softpipe/sp_quad_fs.c | 36 ++----------------------------- 3 files changed, 69 insertions(+), 36 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c index 9ee86fe787..91e04687c5 100644 --- a/src/gallium/drivers/softpipe/sp_fs_exec.c +++ b/src/gallium/drivers/softpipe/sp_fs_exec.c @@ -126,7 +126,39 @@ exec_run( const struct sp_fragment_shader *base, (float)quad->input.x0, (float)quad->input.y0, &machine->QuadPos); - return tgsi_exec_machine_run( machine ); + quad->inout.mask &= tgsi_exec_machine_run( machine ); + if (quad->inout.mask == 0) + return FALSE; + + /* store outputs */ + { + const ubyte *sem_name = base->info.output_semantic_name; + const ubyte *sem_index = base->info.output_semantic_index; + const uint n = base->info.num_outputs; + uint i; + for (i = 0; i < n; i++) { + switch (sem_name[i]) { + case TGSI_SEMANTIC_COLOR: + { + uint cbuf = sem_index[i]; + memcpy(quad->output.color[cbuf], + &machine->Outputs[i].xyzw[0].f[0], + sizeof(quad->output.color[0]) ); + } + break; + case TGSI_SEMANTIC_POSITION: + { + uint j; + for (j = 0; j < 4; j++) { + quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j]; + } + } + break; + } + } + } + + return TRUE; } diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c index f4fa0905d7..364bb94a5f 100644 --- a/src/gallium/drivers/softpipe/sp_fs_sse.c +++ b/src/gallium/drivers/softpipe/sp_fs_sse.c @@ -104,7 +104,40 @@ fs_sse_run( const struct sp_fragment_shader *base, // , &machine->QuadPos ); - return ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]); + quad->inout.mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]); + if (quad->inout.mask == 0) + return FALSE; + + + /* store outputs */ + { + const ubyte *sem_name = shader->base.info.output_semantic_name; + const ubyte *sem_index = shader->base.info.output_semantic_index; + const uint n = shader->base.info.num_outputs; + uint i; + for (i = 0; i < n; i++) { + switch (sem_name[i]) { + case TGSI_SEMANTIC_COLOR: + { + uint cbuf = sem_index[i]; + memcpy(quad->output.color[cbuf], + &machine->Outputs[i].xyzw[0].f[0], + sizeof(quad->output.color[0]) ); + } + break; + case TGSI_SEMANTIC_POSITION: + { + uint j; + for (j = 0; j < 4; j++) { + quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j]; + } + } + break; + } + } + } + + return TRUE; } diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c index e1bc0712de..1e7533d0f9 100644 --- a/src/gallium/drivers/softpipe/sp_quad_fs.c +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -68,7 +68,7 @@ quad_shade_stage(struct quad_stage *qs) /** * Execute fragment shader for the four fragments in the quad. */ -static boolean +static INLINE boolean shade_quad(struct quad_stage *qs, struct quad_header *quad) { struct quad_shade_stage *qss = quad_shade_stage( qs ); @@ -76,39 +76,7 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad) struct tgsi_exec_machine *machine = qss->machine; /* run shader */ - quad->inout.mask &= softpipe->fs->run( softpipe->fs, machine, quad ); - if (quad->inout.mask == 0) - return FALSE; - - /* store outputs */ - { - const ubyte *sem_name = softpipe->fs->info.output_semantic_name; - const ubyte *sem_index = softpipe->fs->info.output_semantic_index; - const uint n = qss->stage.softpipe->fs->info.num_outputs; - uint i; - for (i = 0; i < n; i++) { - switch (sem_name[i]) { - case TGSI_SEMANTIC_COLOR: - { - uint cbuf = sem_index[i]; - memcpy(quad->output.color[cbuf], - &machine->Outputs[i].xyzw[0].f[0], - sizeof(quad->output.color[0]) ); - } - break; - case TGSI_SEMANTIC_POSITION: - { - uint j; - for (j = 0; j < 4; j++) { - quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j]; - } - } - break; - } - } - } - - return TRUE; + return softpipe->fs->run( softpipe->fs, machine, quad ); } -- cgit v1.2.3 From b5c389721aec09c260789e6371910937f15ef1a0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 11 Aug 2009 18:03:01 +0100 Subject: softpipe: remove gallivm fragment shaders However we do llvm integration, it will be different & more comprehensive than this. --- src/gallium/drivers/softpipe/Makefile | 1 - src/gallium/drivers/softpipe/sp_fs_llvm.c | 205 ----------------------------- src/gallium/drivers/softpipe/sp_state_fs.c | 7 +- 3 files changed, 2 insertions(+), 211 deletions(-) delete mode 100644 src/gallium/drivers/softpipe/sp_fs_llvm.c diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile index 48522abe98..a6ed7ea6a2 100644 --- a/src/gallium/drivers/softpipe/Makefile +++ b/src/gallium/drivers/softpipe/Makefile @@ -6,7 +6,6 @@ LIBNAME = softpipe C_SOURCES = \ sp_fs_exec.c \ sp_fs_sse.c \ - sp_fs_llvm.c \ sp_clear.c \ sp_flush.c \ sp_query.c \ diff --git a/src/gallium/drivers/softpipe/sp_fs_llvm.c b/src/gallium/drivers/softpipe/sp_fs_llvm.c deleted file mode 100644 index 95c0d982d1..0000000000 --- a/src/gallium/drivers/softpipe/sp_fs_llvm.c +++ /dev/null @@ -1,205 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -/** - * Execute fragment shader using LLVM code generation. - * Authors: - * Zack Rusin - */ - -#include "sp_context.h" -#include "sp_state.h" -#include "sp_fs.h" - -#include "pipe/p_state.h" -#include "pipe/p_defines.h" -#include "util/u_memory.h" -#include "tgsi/tgsi_sse2.h" - -#if 0 - -/** - * Subclass of sp_fragment_shader - */ -struct sp_llvm_fragment_shader -{ - struct sp_fragment_shader base; - struct gallivm_prog *llvm_prog; -}; - - -static void -shade_quad_llvm(struct quad_stage *qs, - struct quad_header *quad) -{ - struct quad_shade_stage *qss = quad_shade_stage(qs); - struct softpipe_context *softpipe = qs->softpipe; - float dests[4][16][4] ALIGN16_ATTRIB; - float inputs[4][16][4] ALIGN16_ATTRIB; - const float fx = (float) quad->x0; - const float fy = (float) quad->y0; - struct gallivm_prog *llvm = qss->llvm_prog; - - inputs[0][0][0] = fx; - inputs[1][0][0] = fx + 1.0f; - inputs[2][0][0] = fx; - inputs[3][0][0] = fx + 1.0f; - - inputs[0][0][1] = fy; - inputs[1][0][1] = fy; - inputs[2][0][1] = fy + 1.0f; - inputs[3][0][1] = fy + 1.0f; - - - gallivm_prog_inputs_interpolate(llvm, inputs, quad->coef); - -#if DLLVM - debug_printf("MASK = %d\n", quad->mask); - for (int i = 0; i < 4; ++i) { - for (int j = 0; j < 2; ++j) { - debug_printf("IN(%d,%d) [%f %f %f %f]\n", i, j, - inputs[i][j][0], inputs[i][j][1], inputs[i][j][2], inputs[i][j][3]); - } - } -#endif - - quad->mask &= - gallivm_fragment_shader_exec(llvm, fx, fy, dests, inputs, - softpipe->mapped_constants[PIPE_SHADER_FRAGMENT], - qss->samplers); -#if DLLVM - debug_printf("OUT LLVM = 1[%f %f %f %f], 2[%f %f %f %f]\n", - dests[0][0][0], dests[0][0][1], dests[0][0][2], dests[0][0][3], - dests[0][1][0], dests[0][1][1], dests[0][1][2], dests[0][1][3]); -#endif - - /* store result color */ - if (qss->colorOutSlot >= 0) { - unsigned i; - /* XXX need to handle multiple color outputs someday */ - allvmrt(qss->stage.softpipe->fs->info.output_semantic_name[qss->colorOutSlot] - == TGSI_SEMANTIC_COLOR); - for (i = 0; i < QUAD_SIZE; ++i) { - quad->outputs.color[0][0][i] = dests[i][qss->colorOutSlot][0]; - quad->outputs.color[0][1][i] = dests[i][qss->colorOutSlot][1]; - quad->outputs.color[0][2][i] = dests[i][qss->colorOutSlot][2]; - quad->outputs.color[0][3][i] = dests[i][qss->colorOutSlot][3]; - } - } -#if DLLVM - for (int i = 0; i < QUAD_SIZE; ++i) { - debug_printf("QLLVM%d(%d) [%f, %f, %f, %f]\n", i, qss->colorOutSlot, - quad->outputs.color[0][0][i], - quad->outputs.color[0][1][i], - quad->outputs.color[0][2][i], - quad->outputs.color[0][3][i]); - } -#endif - - /* store result Z */ - if (qss->depthOutSlot >= 0) { - /* output[slot] is new Z */ - uint i; - for (i = 0; i < 4; i++) { - quad->outputs.depth[i] = dests[i][0][2]; - } - } - else { - /* copy input Z (which was interpolated by the executor) to output Z */ - uint i; - for (i = 0; i < 4; i++) { - quad->outputs.depth[i] = inputs[i][0][2]; - } - } -#if DLLVM - debug_printf("D [%f, %f, %f, %f] mask = %d\n", - quad->outputs.depth[0], - quad->outputs.depth[1], - quad->outputs.depth[2], - quad->outputs.depth[3], quad->mask); -#endif - - /* shader may cull fragments */ - if( quad->mask ) { - qs->next->run( qs->next, quad ); - } -} - - -unsigned -run_llvm_fs( const struct sp_fragment_shader *base, - struct foo *machine ) -{ -} - - -void -delete_llvm_fs( struct sp_fragment_shader *base ) -{ - FREE(base); -} - - -struct sp_fragment_shader * -softpipe_create_fs_llvm(struct softpipe_context *softpipe, - const struct pipe_shader_state *templ) -{ - struct sp_llvm_fragment_shader *shader = NULL; - - /* LLVM fragment shaders currently disabled: - */ - state = CALLOC_STRUCT(sp_llvm_shader_state); - if (!state) - return NULL; - - state->llvm_prog = 0; - - if (!gallivm_global_cpu_engine()) { - gallivm_cpu_engine_create(state->llvm_prog); - } - else - gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog); - - if (shader) { - shader->base.run = run_llvm_fs; - shader->base.delete = delete_llvm_fs; - } - - return shader; -} - - -#else - -struct sp_fragment_shader * -softpipe_create_fs_llvm(struct softpipe_context *softpipe, - const struct pipe_shader_state *templ) -{ - return NULL; -} - -#endif diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c index 4330c20393..108ac8b9bb 100644 --- a/src/gallium/drivers/softpipe/sp_state_fs.c +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -51,12 +51,9 @@ softpipe_create_fs_state(struct pipe_context *pipe, tgsi_dump(templ->tokens, 0); /* codegen */ - state = softpipe_create_fs_llvm( softpipe, templ ); + state = softpipe_create_fs_sse( softpipe, templ ); if (!state) { - state = softpipe_create_fs_sse( softpipe, templ ); - if (!state) { - state = softpipe_create_fs_exec( softpipe, templ ); - } + state = softpipe_create_fs_exec( softpipe, templ ); } assert(state); -- cgit v1.2.3 From da319095f2ca8869657ebda0db54eb9b2f7393ce Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 11 Aug 2009 18:06:16 +0100 Subject: softpipe: reduce textual differences between exec and sse shader paths Unshare one function (setup_pos_vector) as we want to push this code into the generated shader in the SSE case. --- src/gallium/drivers/softpipe/sp_fs_exec.c | 51 ++++++++++++++++--------------- src/gallium/drivers/softpipe/sp_fs_sse.c | 50 +++++++++++++++++++++++++----- 2 files changed, 69 insertions(+), 32 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c index 91e04687c5..c469ac6340 100644 --- a/src/gallium/drivers/softpipe/sp_fs_exec.c +++ b/src/gallium/drivers/softpipe/sp_fs_exec.c @@ -59,15 +59,34 @@ sp_exec_fragment_shader(const struct sp_fragment_shader *base) } +static void +exec_prepare( const struct sp_fragment_shader *base, + struct tgsi_exec_machine *machine, + struct tgsi_sampler **samplers ) +{ + /* + * Bind tokens/shader to the interpreter's machine state. + * Avoid redundant binding. + */ + if (machine->Tokens != base->shader.tokens) { + tgsi_exec_machine_bind_shader( machine, + base->shader.tokens, + PIPE_MAX_SAMPLERS, + samplers ); + } +} + + + /** * Compute quad X,Y,Z,W for the four fragments in a quad. * * This should really be part of the compiled shader. */ -void -sp_setup_pos_vector(const struct tgsi_interp_coef *coef, - float x, float y, - struct tgsi_exec_vector *quadpos) +static void +setup_pos_vector(const struct tgsi_interp_coef *coef, + float x, float y, + struct tgsi_exec_vector *quadpos) { uint chan; /* do X */ @@ -95,24 +114,6 @@ sp_setup_pos_vector(const struct tgsi_interp_coef *coef, } -static void -exec_prepare( const struct sp_fragment_shader *base, - struct tgsi_exec_machine *machine, - struct tgsi_sampler **samplers ) -{ - /* - * Bind tokens/shader to the interpreter's machine state. - * Avoid redundant binding. - */ - if (machine->Tokens != base->shader.tokens) { - tgsi_exec_machine_bind_shader( machine, - base->shader.tokens, - PIPE_MAX_SAMPLERS, - samplers ); - } -} - - /* TODO: hide the machine struct in here somewhere, remove from this * interface: */ @@ -122,9 +123,9 @@ exec_run( const struct sp_fragment_shader *base, struct quad_header *quad ) { /* Compute X, Y, Z, W vals for this quad */ - sp_setup_pos_vector(quad->posCoef, - (float)quad->input.x0, (float)quad->input.y0, - &machine->QuadPos); + setup_pos_vector(quad->posCoef, + (float)quad->input.x0, (float)quad->input.y0, + &machine->QuadPos); quad->inout.mask &= tgsi_exec_machine_run( machine ); if (quad->inout.mask == 0) diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c index 364bb94a5f..9d3e4670ee 100644 --- a/src/gallium/drivers/softpipe/sp_fs_sse.c +++ b/src/gallium/drivers/softpipe/sp_fs_sse.c @@ -76,6 +76,43 @@ fs_sse_prepare( const struct sp_fragment_shader *base, } + +/** + * Compute quad X,Y,Z,W for the four fragments in a quad. + * + * This should really be part of the compiled shader. + */ +static void +setup_pos_vector(const struct tgsi_interp_coef *coef, + float x, float y, + struct tgsi_exec_vector *quadpos) +{ + uint chan; + /* do X */ + quadpos->xyzw[0].f[0] = x; + quadpos->xyzw[0].f[1] = x + 1; + quadpos->xyzw[0].f[2] = x; + quadpos->xyzw[0].f[3] = x + 1; + + /* do Y */ + quadpos->xyzw[1].f[0] = y; + quadpos->xyzw[1].f[1] = y; + quadpos->xyzw[1].f[2] = y + 1; + quadpos->xyzw[1].f[3] = y + 1; + + /* do Z and W for all fragments in the quad */ + for (chan = 2; chan < 4; chan++) { + const float dadx = coef->dadx[chan]; + const float dady = coef->dady[chan]; + const float a0 = coef->a0[chan] + dadx * x + dady * y; + quadpos->xyzw[chan].f[0] = a0; + quadpos->xyzw[chan].f[1] = a0 + dadx; + quadpos->xyzw[chan].f[2] = a0 + dady; + quadpos->xyzw[chan].f[3] = a0 + dadx + dady; + } +} + + /* TODO: codegenerate the whole run function, skip this wrapper. * TODO: break dependency on tgsi_exec_machine struct * TODO: push Position calculation into the generated shader @@ -89,9 +126,9 @@ fs_sse_run( const struct sp_fragment_shader *base, struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base); /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */ - sp_setup_pos_vector(quad->posCoef, - (float)quad->input.x0, (float)quad->input.y0, - machine->Temps); + setup_pos_vector(quad->posCoef, + (float)quad->input.x0, (float)quad->input.y0, + machine->Temps); /* init kill mask */ tgsi_set_kill_mask(machine, 0x0); @@ -108,12 +145,11 @@ fs_sse_run( const struct sp_fragment_shader *base, if (quad->inout.mask == 0) return FALSE; - /* store outputs */ { - const ubyte *sem_name = shader->base.info.output_semantic_name; - const ubyte *sem_index = shader->base.info.output_semantic_index; - const uint n = shader->base.info.num_outputs; + const ubyte *sem_name = base->info.output_semantic_name; + const ubyte *sem_index = base->info.output_semantic_index; + const uint n = base->info.num_outputs; uint i; for (i = 0; i < n; i++) { switch (sem_name[i]) { -- cgit v1.2.3 From 99ec78d9462d2a553982d0ea15d538b36b1c123b Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 11 Aug 2009 18:23:28 +0100 Subject: Revert "softpipe: rearrange blend fastpaths" This reverts commit 1295cf423e21dad04a947960782ffa8db2739709. The original formulation was easier to understand & work with. Will revisit this later. --- src/gallium/drivers/softpipe/sp_quad_blend.c | 82 +++++++++++++++++++--------- 1 file changed, 55 insertions(+), 27 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c index b8ed086734..e243c63fa2 100644 --- a/src/gallium/drivers/softpipe/sp_quad_blend.c +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -793,7 +793,10 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, struct quad_header *quads[], unsigned nr) { - float source[4]; + static const float one[4] = { 1, 1, 1, 1 }; + float one_minus_alpha[QUAD_SIZE]; + float dest[4][QUAD_SIZE]; + float source[4][QUAD_SIZE]; uint i, j, q; struct softpipe_cached_tile *tile @@ -803,26 +806,45 @@ blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, for (q = 0; q < nr; q++) { struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[0]; + const float *alpha = quadColor[3]; const int itx = (quad->input.x0 & (TILE_SIZE-1)); const int ity = (quad->input.y0 & (TILE_SIZE-1)); - float (*swzColor)[4] = quad->output.color[0]; - for (j = 0; j < 4; j++) { - if (quad->inout.mask & (1<data.color[ity + (j>>1)][itx + (j&1)]; - const float alpha = swzColor[3][j]; - const float one_minus_alpha = 1.0 - alpha; + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } - for (i = 0; i < 4; i++) { - dest[i] *= one_minus_alpha; - dest[i] += swzColor[i][j] * alpha; + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + VEC4_MUL(source[3], quadColor[3], alpha); /* A */ + + VEC4_SUB(one_minus_alpha, one, alpha); + VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ + VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ + VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ + VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* B */ + + VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */ + VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */ - /* XXX: redundant, will be clamped later for argb8 surfaces: - */ - dest[i] = CLAMP(dest[i], 0.0, 1.0); + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; } } - } + } } } @@ -841,27 +863,33 @@ blend_single_add_one_one(struct quad_stage *qs, for (q = 0; q < nr; q++) { struct quad_header *quad = quads[q]; + float (*quadColor)[4] = quad->output.color[0]; const int itx = (quad->input.x0 & (TILE_SIZE-1)); const int ity = (quad->input.y0 & (TILE_SIZE-1)); - float (*dest)[64][4] = (float (*)[64][4])&tile->data.color[ity][itx]; - float (*swzColor)[4] = quad->output.color[0]; - float quadColor[4][4]; - + /* get/swizzle dest colors */ for (j = 0; j < QUAD_SIZE; j++) { + int x = itx + (j & 1); + int y = ity + (j >> 1); for (i = 0; i < 4; i++) { - quadColor[i][j] = swzColor[j][i]; + dest[i][j] = tile->data.color[y][x][i]; } } - if (quad->inout.mask & 1) - VEC4_ADD_SAT(dest[0][0], quadColor[0], dest[0][0]); - if (quad->inout.mask & 2) - VEC4_ADD_SAT(dest[0][1], quadColor[1], dest[0][1]); - if (quad->inout.mask & 4) - VEC4_ADD_SAT(dest[1][0], quadColor[2], dest[1][0]); - if (quad->inout.mask & 8) - VEC4_ADD_SAT(dest[1][1], quadColor[3], dest[1][1]); + VEC4_ADD_SAT(quadColor[0], quadColor[0], dest[0]); /* R */ + VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */ + VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */ + VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */ + + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->inout.mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + } + } } } -- cgit v1.2.3 From d12bae9368e0c44a9943d9b37ab848ea307d70c7 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 18 Aug 2009 16:21:12 +0100 Subject: softpipe: move flatshade-first check out of loop --- src/gallium/drivers/softpipe/sp_prim_vbuf.c | 80 ++++++++++++++++++----------- 1 file changed, 50 insertions(+), 30 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c index 1dd63d99ff..76524a8d41 100644 --- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c @@ -239,14 +239,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) break; case PIPE_PRIM_TRIANGLES: - for (i = 2; i < nr; i += 3) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 3) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-1], stride), get_vert(vertex_buffer, indices[i-0], stride), get_vert(vertex_buffer, indices[i-2], stride) ); } - else { + } + else { + for (i = 2; i < nr; i += 3) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-2], stride), get_vert(vertex_buffer, indices[i-1], stride), @@ -256,14 +258,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) break; case PIPE_PRIM_TRIANGLE_STRIP: - for (i = 2; i < nr; i += 1) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i+(i&1)-1], stride), get_vert(vertex_buffer, indices[i-(i&1)], stride), get_vert(vertex_buffer, indices[i-2], stride) ); } - else { + } + else { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i+(i&1)-2], stride), get_vert(vertex_buffer, indices[i-(i&1)-1], stride), @@ -273,14 +277,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) break; case PIPE_PRIM_TRIANGLE_FAN: - for (i = 2; i < nr; i += 1) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-0], stride), get_vert(vertex_buffer, indices[0], stride), get_vert(vertex_buffer, indices[i-1], stride) ); } - else { + } + else { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[0], stride), get_vert(vertex_buffer, indices[i-1], stride), @@ -290,8 +296,8 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) break; case PIPE_PRIM_QUADS: - for (i = 3; i < nr; i += 4) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 4) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-2], stride), get_vert(vertex_buffer, indices[i-1], stride), @@ -301,7 +307,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) get_vert(vertex_buffer, indices[i-0], stride), get_vert(vertex_buffer, indices[i-3], stride) ); } - else { + } + else { + for (i = 3; i < nr; i += 4) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-3], stride), get_vert(vertex_buffer, indices[i-2], stride), @@ -316,8 +324,8 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) break; case PIPE_PRIM_QUAD_STRIP: - for (i = 3; i < nr; i += 2) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 2) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-0], stride), get_vert(vertex_buffer, indices[i-1], stride), @@ -327,7 +335,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) get_vert(vertex_buffer, indices[i-0], stride), get_vert(vertex_buffer, indices[i-3], stride) ); } - else { + } + else { + for (i = 3; i < nr; i += 2) { setup_tri( setup_ctx, get_vert(vertex_buffer, indices[i-3], stride), get_vert(vertex_buffer, indices[i-2], stride), @@ -423,14 +433,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) break; case PIPE_PRIM_TRIANGLES: - for (i = 2; i < nr; i += 3) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 3) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-1, stride), get_vert(vertex_buffer, i-0, stride), get_vert(vertex_buffer, i-2, stride) ); } - else { + } + else { + for (i = 2; i < nr; i += 3) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-2, stride), get_vert(vertex_buffer, i-1, stride), @@ -440,14 +452,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) break; case PIPE_PRIM_TRIANGLE_STRIP: - for (i = 2; i < nr; i++) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i++) { setup_tri( setup_ctx, get_vert(vertex_buffer, i+(i&1)-1, stride), get_vert(vertex_buffer, i-(i&1), stride), get_vert(vertex_buffer, i-2, stride) ); } - else { + } + else { + for (i = 2; i < nr; i++) { setup_tri( setup_ctx, get_vert(vertex_buffer, i+(i&1)-2, stride), get_vert(vertex_buffer, i-(i&1)-1, stride), @@ -457,14 +471,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) break; case PIPE_PRIM_TRIANGLE_FAN: - for (i = 2; i < nr; i += 1) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-0, stride), get_vert(vertex_buffer, 0, stride), get_vert(vertex_buffer, i-1, stride) ); } - else { + } + else { + for (i = 2; i < nr; i += 1) { setup_tri( setup_ctx, get_vert(vertex_buffer, 0, stride), get_vert(vertex_buffer, i-1, stride), @@ -474,8 +490,8 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) break; case PIPE_PRIM_QUADS: - for (i = 3; i < nr; i += 4) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 4) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-2, stride), get_vert(vertex_buffer, i-1, stride), @@ -485,7 +501,9 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) get_vert(vertex_buffer, i-0, stride), get_vert(vertex_buffer, i-3, stride) ); } - else { + } + else { + for (i = 3; i < nr; i += 4) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-3, stride), get_vert(vertex_buffer, i-2, stride), @@ -499,8 +517,8 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) break; case PIPE_PRIM_QUAD_STRIP: - for (i = 3; i < nr; i += 2) { - if (softpipe->rasterizer->flatshade_first) { + if (softpipe->rasterizer->flatshade_first) { + for (i = 3; i < nr; i += 2) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-0, stride), get_vert(vertex_buffer, i-1, stride), @@ -510,7 +528,9 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) get_vert(vertex_buffer, i-0, stride), get_vert(vertex_buffer, i-3, stride) ); } - else { + } + else { + for (i = 3; i < nr; i += 2) { setup_tri( setup_ctx, get_vert(vertex_buffer, i-3, stride), get_vert(vertex_buffer, i-2, stride), -- cgit v1.2.3 From 80c78472ad43f4288c9ef5076074ba9d31a39885 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 29 Jul 2009 07:40:50 +0100 Subject: softpipe: split texture and surface tile caches These do similar jobs but with largely disjoint code. Will want to evolve them separately going forward. --- src/gallium/drivers/softpipe/Makefile | 1 + src/gallium/drivers/softpipe/SConscript | 1 + src/gallium/drivers/softpipe/sp_context.c | 4 +- src/gallium/drivers/softpipe/sp_context.h | 3 +- src/gallium/drivers/softpipe/sp_flush.c | 3 +- src/gallium/drivers/softpipe/sp_state_derived.c | 2 +- src/gallium/drivers/softpipe/sp_state_sampler.c | 4 +- src/gallium/drivers/softpipe/sp_tex_sample.c | 16 +- src/gallium/drivers/softpipe/sp_tex_sample.h | 2 +- src/gallium/drivers/softpipe/sp_tex_tile_cache.c | 274 +++++++++++++++++++++++ src/gallium/drivers/softpipe/sp_tex_tile_cache.h | 161 +++++++++++++ src/gallium/drivers/softpipe/sp_tile_cache.c | 171 +------------- src/gallium/drivers/softpipe/sp_tile_cache.h | 42 +--- 13 files changed, 460 insertions(+), 224 deletions(-) create mode 100644 src/gallium/drivers/softpipe/sp_tex_tile_cache.c create mode 100644 src/gallium/drivers/softpipe/sp_tex_tile_cache.h diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile index a6ed7ea6a2..3da9be6957 100644 --- a/src/gallium/drivers/softpipe/Makefile +++ b/src/gallium/drivers/softpipe/Makefile @@ -30,6 +30,7 @@ C_SOURCES = \ sp_state_vertex.c \ sp_texture.c \ sp_tex_sample.c \ + sp_tex_tile_cache.c \ sp_tile_cache.c \ sp_surface.c diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript index dcc25732ba..30c099813e 100644 --- a/src/gallium/drivers/softpipe/SConscript +++ b/src/gallium/drivers/softpipe/SConscript @@ -37,6 +37,7 @@ softpipe = env.ConvenienceLibrary( 'sp_state_vertex.c', 'sp_surface.c', 'sp_tex_sample.c', + 'sp_tex_tile_cache.c', 'sp_texture.c', 'sp_tile_cache.c', ]) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index e35c6b3aec..396d4c6557 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -97,7 +97,7 @@ static void softpipe_destroy( struct pipe_context *pipe ) sp_destroy_tile_cache(softpipe->zsbuf_cache); for (i = 0; i < PIPE_MAX_SAMPLERS; i++) - sp_destroy_tile_cache(softpipe->tex_cache[i]); + sp_destroy_tex_tile_cache(softpipe->tex_cache[i]); for (i = 0; i < Elements(softpipe->constants); i++) { if (softpipe->constants[i].buffer) { @@ -220,7 +220,7 @@ softpipe_create( struct pipe_screen *screen ) softpipe->zsbuf_cache = sp_create_tile_cache( screen ); for (i = 0; i < PIPE_MAX_SAMPLERS; i++) - softpipe->tex_cache[i] = sp_create_tile_cache( screen ); + softpipe->tex_cache[i] = sp_create_tex_tile_cache( screen ); /* setup quad rendering stages */ diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index fa3306c020..683c3aef9b 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -43,6 +43,7 @@ struct softpipe_vbuf_render; struct draw_context; struct draw_stage; struct softpipe_tile_cache; +struct softpipe_tex_tile_cache; struct sp_fragment_shader; struct sp_vertex_shader; @@ -141,7 +142,7 @@ struct softpipe_context { struct softpipe_tile_cache *zsbuf_cache; unsigned tex_timestamp; - struct softpipe_tile_cache *tex_cache[PIPE_MAX_SAMPLERS]; + struct softpipe_tex_tile_cache *tex_cache[PIPE_MAX_SAMPLERS]; unsigned use_sse : 1; unsigned dump_fs : 1; diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c index 679ad0cd3d..e38b767cf2 100644 --- a/src/gallium/drivers/softpipe/sp_flush.c +++ b/src/gallium/drivers/softpipe/sp_flush.c @@ -37,6 +37,7 @@ #include "sp_surface.h" #include "sp_state.h" #include "sp_tile_cache.h" +#include "sp_tex_tile_cache.h" #include "sp_winsys.h" @@ -52,7 +53,7 @@ softpipe_flush( struct pipe_context *pipe, if (flags & PIPE_FLUSH_TEXTURE_CACHE) { for (i = 0; i < softpipe->num_textures; i++) { - sp_flush_tile_cache(softpipe->tex_cache[i]); + sp_flush_tex_tile_cache(softpipe->tex_cache[i]); } } diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 3ed1de7e17..88a6e4fbdf 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -212,7 +212,7 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) } for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - sp_tile_cache_validate_texture( softpipe->tex_cache[i] ); + sp_tex_tile_cache_validate_texture( softpipe->tex_cache[i] ); } } diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c index aa2f3f2ccd..a725925264 100644 --- a/src/gallium/drivers/softpipe/sp_state_sampler.c +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -37,7 +37,7 @@ #include "sp_context.h" #include "sp_state.h" #include "sp_texture.h" -#include "sp_tile_cache.h" +#include "sp_tex_tile_cache.h" #include "draw/draw_context.h" @@ -97,7 +97,7 @@ softpipe_set_sampler_textures(struct pipe_context *pipe, struct pipe_texture *tex = i < num ? texture[i] : NULL; pipe_texture_reference(&softpipe->texture[i], tex); - sp_tile_cache_set_texture(softpipe->tex_cache[i], tex); + sp_tex_tile_cache_set_texture(softpipe->tex_cache[i], tex); } softpipe->num_textures = num; diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 4651d781a9..8bed573e8c 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -38,7 +38,7 @@ #include "sp_surface.h" #include "sp_texture.h" #include "sp_tex_sample.h" -#include "sp_tile_cache.h" +#include "sp_tex_tile_cache.h" #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "util/u_math.h" @@ -659,7 +659,7 @@ choose_mipmap_levels(const struct pipe_texture *texture, * \param rgba the quad to put the texel/color into * \param j which element of the rgba quad to write to * - * XXX maybe move this into sp_tile_cache.c and merge with the + * XXX maybe move this into sp_tex_tile_cache.c and merge with the * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... */ static void @@ -669,9 +669,9 @@ get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_cached_tile *tile + const struct softpipe_tex_cached_tile *tile = sp_get_cached_tile_tex(samp->cache, - tile_address(x, y, 0, face, level)); + tex_tile_address(x, y, 0, face, level)); y %= TILE_SIZE; x %= TILE_SIZE; @@ -688,9 +688,9 @@ get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, { const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct softpipe_cached_tile *tile + const struct softpipe_tex_cached_tile *tile = sp_get_cached_tile_tex(samp->cache, - tile_address(x, y, 0, face, level)); + tex_tile_address(x, y, 0, face, level)); y %= TILE_SIZE; x %= TILE_SIZE; @@ -736,10 +736,10 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, else { const unsigned tx = x % TILE_SIZE; const unsigned ty = y % TILE_SIZE; - const struct softpipe_cached_tile *tile; + const struct softpipe_tex_cached_tile *tile; tile = sp_get_cached_tile_tex(samp->cache, - tile_address(x, y, z, face, level)); + tex_tile_address(x, y, z, face, level)); rgba[0][j] = tile->data.color[ty][tx][0]; rgba[1][j] = tile->data.color[ty][tx][1]; diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index 0650c7830b..1756d3a4ee 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -48,7 +48,7 @@ struct sp_shader_sampler const struct pipe_texture *texture; const struct pipe_sampler_state *sampler; - struct softpipe_tile_cache *cache; + struct softpipe_tex_tile_cache *cache; }; diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c new file mode 100644 index 0000000000..c2dd68c7a2 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c @@ -0,0 +1,274 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture tile caching. + * + * Author: + * Brian Paul + */ + +#include "pipe/p_inlines.h" +#include "util/u_memory.h" +#include "util/u_tile.h" +#include "sp_context.h" +#include "sp_surface.h" +#include "sp_texture.h" +#include "sp_tex_tile_cache.h" + + + +struct softpipe_tex_tile_cache * +sp_create_tex_tile_cache( struct pipe_screen *screen ) +{ + struct softpipe_tex_tile_cache *tc; + uint pos; + + tc = CALLOC_STRUCT( softpipe_tex_tile_cache ); + if (tc) { + tc->screen = screen; + for (pos = 0; pos < NUM_ENTRIES; pos++) { + tc->entries[pos].addr.bits.invalid = 1; + } + tc->last_tile = &tc->entries[0]; /* any tile */ + } + return tc; +} + + +void +sp_destroy_tex_tile_cache(struct softpipe_tex_tile_cache *tc) +{ + struct pipe_screen *screen; + uint pos; + + for (pos = 0; pos < NUM_ENTRIES; pos++) { + /*assert(tc->entries[pos].x < 0);*/ + } + if (tc->transfer) { + screen = tc->transfer->texture->screen; + screen->tex_transfer_destroy(tc->transfer); + } + if (tc->tex_trans) { + screen = tc->tex_trans->texture->screen; + screen->tex_transfer_destroy(tc->tex_trans); + } + + FREE( tc ); +} + + + + +void +sp_tex_tile_cache_map_transfers(struct softpipe_tex_tile_cache *tc) +{ + if (tc->tex_trans && !tc->tex_trans_map) + tc->tex_trans_map = tc->screen->transfer_map(tc->screen, tc->tex_trans); +} + + +void +sp_tex_tile_cache_unmap_transfers(struct softpipe_tex_tile_cache *tc) +{ + if (tc->tex_trans_map) { + tc->screen->transfer_unmap(tc->screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } +} + +void +sp_tex_tile_cache_validate_texture(struct softpipe_tex_tile_cache *tc) +{ + if (tc->texture) { + struct softpipe_texture *spt = softpipe_texture(tc->texture); + if (spt->timestamp != tc->timestamp) { + /* texture was modified, invalidate all cached tiles */ + uint i; + _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].addr.bits.invalid = 1; + } + + tc->timestamp = spt->timestamp; + } + } +} + +/** + * Specify the texture to cache. + */ +void +sp_tex_tile_cache_set_texture(struct softpipe_tex_tile_cache *tc, + struct pipe_texture *texture) +{ + uint i; + + assert(!tc->transfer); + + if (tc->texture != texture) { + pipe_texture_reference(&tc->texture, texture); + + if (tc->tex_trans) { + struct pipe_screen *screen = tc->tex_trans->texture->screen; + + if (tc->tex_trans_map) { + screen->transfer_unmap(screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } + + screen->tex_transfer_destroy(tc->tex_trans); + tc->tex_trans = NULL; + } + + /* mark as entries as invalid/empty */ + /* XXX we should try to avoid this when the teximage hasn't changed */ + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].addr.bits.invalid = 1; + } + + tc->tex_face = -1; /* any invalid value here */ + } +} + + + + +/** + * Flush the tile cache: write all dirty tiles back to the transfer. + * any tiles "flagged" as cleared will be "really" cleared. + */ +void +sp_flush_tex_tile_cache(struct softpipe_tex_tile_cache *tc) +{ + int pos; + + if (tc->texture) { + /* caching a texture, mark all entries as empty */ + for (pos = 0; pos < NUM_ENTRIES; pos++) { + tc->entries[pos].addr.bits.invalid = 1; + } + tc->tex_face = -1; + } + +} + + +/** + * Given the texture face, level, zslice, x and y values, compute + * the cache entry position/index where we'd hope to find the + * cached texture tile. + * This is basically a direct-map cache. + * XXX There's probably lots of ways in which we can improve this. + */ +static INLINE uint +tex_cache_pos( union tex_tile_address addr ) +{ + uint entry = (addr.bits.x + + addr.bits.y * 9 + + addr.bits.z * 3 + + addr.bits.face + + addr.bits.level * 7); + + return entry % NUM_ENTRIES; +} + +/** + * Similar to sp_get_cached_tile() but for textures. + * Tiles are read-only and indexed with more params. + */ +const struct softpipe_tex_cached_tile * +sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc, + union tex_tile_address addr ) +{ + struct pipe_screen *screen = tc->screen; + struct softpipe_tex_cached_tile *tile; + + tile = tc->entries + tex_cache_pos( addr ); + + if (addr.value != tile->addr.value) { + + /* cache miss. Most misses are because we've invaldiated the + * texture cache previously -- most commonly on binding a new + * texture. Currently we effectively flush the cache on texture + * bind. + */ +#if 0 + _debug_printf("miss at %u: x=%d y=%d z=%d face=%d level=%d\n" + " tile %u: x=%d y=%d z=%d face=%d level=%d\n", + pos, x/TILE_SIZE, y/TILE_SIZE, z, face, level, + pos, tile->addr.bits.x, tile->addr.bits.y, tile->z, tile->face, tile->level); +#endif + + /* check if we need to get a new transfer */ + if (!tc->tex_trans || + tc->tex_face != addr.bits.face || + tc->tex_level != addr.bits.level || + tc->tex_z != addr.bits.z) { + /* get new transfer (view into texture) */ + + if (tc->tex_trans) { + if (tc->tex_trans_map) { + tc->screen->transfer_unmap(tc->screen, tc->tex_trans); + tc->tex_trans_map = NULL; + } + + screen->tex_transfer_destroy(tc->tex_trans); + tc->tex_trans = NULL; + } + + tc->tex_trans = + screen->get_tex_transfer(screen, tc->texture, + addr.bits.face, + addr.bits.level, + addr.bits.z, + PIPE_TRANSFER_READ, 0, 0, + tc->texture->width[addr.bits.level], + tc->texture->height[addr.bits.level]); + + tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans); + + tc->tex_face = addr.bits.face; + tc->tex_level = addr.bits.level; + tc->tex_z = addr.bits.z; + } + + /* get tile from the transfer (view into texture) */ + pipe_get_tile_rgba(tc->tex_trans, + addr.bits.x * TILE_SIZE, + addr.bits.y * TILE_SIZE, + TILE_SIZE, TILE_SIZE, + (float *) tile->data.color); + tile->addr = addr; + } + + tc->last_tile = tile; + return tile; +} + + + diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.h b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h new file mode 100644 index 0000000000..c6003f3550 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h @@ -0,0 +1,161 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SP_TEX_TILE_CACHE_H +#define SP_TEX_TILE_CACHE_H + + +#include "pipe/p_compiler.h" + + +struct softpipe_context; +struct softpipe_tex_tile_cache; + + +/** + * Cache tile size (width and height). This needs to be a power of two. + */ +#define TILE_SIZE 64 + + +/* If we need to support > 4096, just expand this to be a 64 bit + * union, or consider tiling in Z as well. + */ +union tex_tile_address { + struct { + unsigned x:6; /* 4096 / TILE_SIZE */ + unsigned y:6; /* 4096 / TILE_SIZE */ + unsigned z:12; /* 4096 -- z not tiled */ + unsigned face:3; + unsigned level:4; + unsigned invalid:1; + } bits; + unsigned value; +}; + + +struct softpipe_tex_cached_tile +{ + union tex_tile_address addr; + union { + float color[TILE_SIZE][TILE_SIZE][4]; + } data; +}; + +#define NUM_ENTRIES 50 + + +/** XXX move these */ +#define MAX_WIDTH 2048 +#define MAX_HEIGHT 2048 + + +struct softpipe_tex_tile_cache +{ + struct pipe_screen *screen; + struct pipe_transfer *transfer; + void *transfer_map; + + struct pipe_texture *texture; /**< if caching a texture */ + unsigned timestamp; + + struct softpipe_tex_cached_tile entries[NUM_ENTRIES]; + + struct pipe_transfer *tex_trans; + void *tex_trans_map; + int tex_face, tex_level, tex_z; + + struct softpipe_tex_cached_tile *last_tile; /**< most recently retrieved tile */ +}; + + +extern struct softpipe_tex_tile_cache * +sp_create_tex_tile_cache( struct pipe_screen *screen ); + +extern void +sp_destroy_tex_tile_cache(struct softpipe_tex_tile_cache *tc); + + +extern void +sp_tex_tile_cache_map_transfers(struct softpipe_tex_tile_cache *tc); + +extern void +sp_tex_tile_cache_unmap_transfers(struct softpipe_tex_tile_cache *tc); + +extern void +sp_tex_tile_cache_set_texture(struct softpipe_tex_tile_cache *tc, + struct pipe_texture *texture); + +void +sp_tex_tile_cache_validate_texture(struct softpipe_tex_tile_cache *tc); + +extern void +sp_flush_tex_tile_cache(struct softpipe_tex_tile_cache *tc); + + + +extern const struct softpipe_tex_cached_tile * +sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc, + union tex_tile_address addr ); + +static INLINE const union tex_tile_address +tex_tile_address( unsigned x, + unsigned y, + unsigned z, + unsigned face, + unsigned level ) +{ + union tex_tile_address addr; + + addr.value = 0; + addr.bits.x = x / TILE_SIZE; + addr.bits.y = y / TILE_SIZE; + addr.bits.z = z; + addr.bits.face = face; + addr.bits.level = level; + + return addr; +} + +/* Quickly retrieve tile if it matches last lookup. + */ +static INLINE const struct softpipe_tex_cached_tile * +sp_get_cached_tile_tex(struct softpipe_tex_tile_cache *tc, + union tex_tile_address addr ) +{ + if (tc->last_tile->addr.value == addr.value) + return tc->last_tile; + + return sp_find_cached_tile_tex( tc, addr ); +} + + + + + +#endif /* SP_TEX_TILE_CACHE_H */ + diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 77d02fa3e7..2d82badcec 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -26,7 +26,7 @@ **************************************************************************/ /** - * Texture tile caching. + * Render target tile caching. * * Author: * Brian Paul @@ -37,7 +37,6 @@ #include "util/u_tile.h" #include "sp_context.h" #include "sp_surface.h" -#include "sp_texture.h" #include "sp_tile_cache.h" @@ -111,10 +110,6 @@ sp_destroy_tile_cache(struct softpipe_tile_cache *tc) screen = tc->transfer->texture->screen; screen->tex_transfer_destroy(tc->transfer); } - if (tc->tex_trans) { - screen = tc->tex_trans->texture->screen; - screen->tex_transfer_destroy(tc->tex_trans); - } FREE( tc ); } @@ -127,8 +122,6 @@ void sp_tile_cache_set_surface(struct softpipe_tile_cache *tc, struct pipe_surface *ps) { - assert(!tc->texture); - if (tc->transfer) { struct pipe_screen *screen = tc->transfer->texture->screen; @@ -180,9 +173,6 @@ sp_tile_cache_map_transfers(struct softpipe_tile_cache *tc) { if (tc->transfer && !tc->transfer_map) tc->transfer_map = tc->screen->transfer_map(tc->screen, tc->transfer); - - if (tc->tex_trans && !tc->tex_trans_map) - tc->tex_trans_map = tc->screen->transfer_map(tc->screen, tc->tex_trans); } @@ -193,68 +183,8 @@ sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc) tc->screen->transfer_unmap(tc->screen, tc->transfer); tc->transfer_map = NULL; } - - if (tc->tex_trans_map) { - tc->screen->transfer_unmap(tc->screen, tc->tex_trans); - tc->tex_trans_map = NULL; - } -} - -void -sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc) -{ - if (tc->texture) { - struct softpipe_texture *spt = softpipe_texture(tc->texture); - if (spt->timestamp != tc->timestamp) { - /* texture was modified, invalidate all cached tiles */ - uint i; - _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); - for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].addr.bits.invalid = 1; - } - - tc->timestamp = spt->timestamp; - } - } -} - -/** - * Specify the texture to cache. - */ -void -sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, - struct pipe_texture *texture) -{ - uint i; - - assert(!tc->transfer); - - if (tc->texture != texture) { - pipe_texture_reference(&tc->texture, texture); - - if (tc->tex_trans) { - struct pipe_screen *screen = tc->tex_trans->texture->screen; - - if (tc->tex_trans_map) { - screen->transfer_unmap(screen, tc->tex_trans); - tc->tex_trans_map = NULL; - } - - screen->tex_transfer_destroy(tc->tex_trans); - tc->tex_trans = NULL; - } - - /* mark as entries as invalid/empty */ - /* XXX we should try to avoid this when the teximage hasn't changed */ - for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].addr.bits.invalid = 1; - } - - tc->tex_face = -1; /* any invalid value here */ - } } - /** * Set pixels in a tile to the given clear color/value, float. */ @@ -345,7 +275,7 @@ sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc) /* push the tile to all positions marked as clear */ for (y = 0; y < h; y += TILE_SIZE) { for (x = 0; x < w; x += TILE_SIZE) { - union tile_address addr = tile_address(x, y, 0, 0, 0); + union tile_address addr = tile_address(x, y); if (is_clear_flag_set(tc->clear_flags, addr)) { pipe_put_tile_raw(pt, @@ -403,13 +333,6 @@ sp_flush_tile_cache(struct softpipe_tile_cache *tc) sp_tile_cache_flush_clear(tc); #endif } - else if (tc->texture) { - /* caching a texture, mark all entries as empty */ - for (pos = 0; pos < NUM_ENTRIES; pos++) { - tc->entries[pos].addr.bits.invalid = 1; - } - tc->tex_face = -1; - } #if 0 debug_printf("flushed tiles in use: %d\n", inuse); @@ -488,97 +411,7 @@ sp_find_cached_tile(struct softpipe_tile_cache *tc, } -/** - * Given the texture face, level, zslice, x and y values, compute - * the cache entry position/index where we'd hope to find the - * cached texture tile. - * This is basically a direct-map cache. - * XXX There's probably lots of ways in which we can improve this. - */ -static INLINE uint -tex_cache_pos( union tile_address addr ) -{ - uint entry = (addr.bits.x + - addr.bits.y * 9 + - addr.bits.z * 3 + - addr.bits.face + - addr.bits.level * 7); - - return entry % NUM_ENTRIES; -} - -/** - * Similar to sp_get_cached_tile() but for textures. - * Tiles are read-only and indexed with more params. - */ -const struct softpipe_cached_tile * -sp_find_cached_tile_tex(struct softpipe_tile_cache *tc, - union tile_address addr ) -{ - struct pipe_screen *screen = tc->screen; - struct softpipe_cached_tile *tile; - - tile = tc->entries + tex_cache_pos( addr ); - - if (addr.value != tile->addr.value) { - - /* cache miss. Most misses are because we've invaldiated the - * texture cache previously -- most commonly on binding a new - * texture. Currently we effectively flush the cache on texture - * bind. - */ -#if 0 - _debug_printf("miss at %u: x=%d y=%d z=%d face=%d level=%d\n" - " tile %u: x=%d y=%d z=%d face=%d level=%d\n", - pos, x/TILE_SIZE, y/TILE_SIZE, z, face, level, - pos, tile->addr.bits.x, tile->addr.bits.y, tile->z, tile->face, tile->level); -#endif - - /* check if we need to get a new transfer */ - if (!tc->tex_trans || - tc->tex_face != addr.bits.face || - tc->tex_level != addr.bits.level || - tc->tex_z != addr.bits.z) { - /* get new transfer (view into texture) */ - - if (tc->tex_trans) { - if (tc->tex_trans_map) { - tc->screen->transfer_unmap(tc->screen, tc->tex_trans); - tc->tex_trans_map = NULL; - } - - screen->tex_transfer_destroy(tc->tex_trans); - tc->tex_trans = NULL; - } - - tc->tex_trans = - screen->get_tex_transfer(screen, tc->texture, - addr.bits.face, - addr.bits.level, - addr.bits.z, - PIPE_TRANSFER_READ, 0, 0, - tc->texture->width[addr.bits.level], - tc->texture->height[addr.bits.level]); - - tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans); - tc->tex_face = addr.bits.face; - tc->tex_level = addr.bits.level; - tc->tex_z = addr.bits.z; - } - - /* get tile from the transfer (view into texture) */ - pipe_get_tile_rgba(tc->tex_trans, - addr.bits.x * TILE_SIZE, - addr.bits.y * TILE_SIZE, - TILE_SIZE, TILE_SIZE, - (float *) tile->data.color); - tile->addr = addr; - } - - tc->last_tile = tile; - return tile; -} /** diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index ac2aae5875..3b0be274d5 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -51,10 +51,8 @@ union tile_address { struct { unsigned x:6; /* 4096 / TILE_SIZE */ unsigned y:6; /* 4096 / TILE_SIZE */ - unsigned z:12; /* 4096 -- z not tiled */ - unsigned face:3; - unsigned level:4; unsigned invalid:1; + unsigned pad:19; } bits; unsigned value; }; @@ -88,19 +86,12 @@ struct softpipe_tile_cache struct pipe_transfer *transfer; void *transfer_map; - struct pipe_texture *texture; /**< if caching a texture */ - unsigned timestamp; - struct softpipe_cached_tile entries[NUM_ENTRIES]; uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32]; float clear_color[4]; /**< for color bufs */ uint clear_val; /**< for z+stencil, or packed color clear value */ boolean depth_stencil; /**< Is the surface a depth/stencil format? */ - struct pipe_transfer *tex_trans; - void *tex_trans_map; - int tex_face, tex_level, tex_z; - struct softpipe_cached_tile tile; /**< scratch tile for clears */ struct softpipe_cached_tile *last_tile; /**< most recently retrieved tile */ @@ -126,13 +117,6 @@ sp_tile_cache_map_transfers(struct softpipe_tile_cache *tc); extern void sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc); -extern void -sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, - struct pipe_texture *texture); - -void -sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc); - extern void sp_flush_tile_cache(struct softpipe_tile_cache *tc); @@ -144,47 +128,27 @@ extern struct softpipe_cached_tile * sp_find_cached_tile(struct softpipe_tile_cache *tc, union tile_address addr ); -extern const struct softpipe_cached_tile * -sp_find_cached_tile_tex(struct softpipe_tile_cache *tc, - union tile_address addr ); static INLINE const union tile_address tile_address( unsigned x, - unsigned y, - unsigned z, - unsigned face, - unsigned level ) + unsigned y ) { union tile_address addr; addr.value = 0; addr.bits.x = x / TILE_SIZE; addr.bits.y = y / TILE_SIZE; - addr.bits.z = z; - addr.bits.face = face; - addr.bits.level = level; return addr; } /* Quickly retrieve tile if it matches last lookup. */ -static INLINE const struct softpipe_cached_tile * -sp_get_cached_tile_tex(struct softpipe_tile_cache *tc, - union tile_address addr ) -{ - if (tc->last_tile->addr.value == addr.value) - return tc->last_tile; - - return sp_find_cached_tile_tex( tc, addr ); -} - - static INLINE struct softpipe_cached_tile * sp_get_cached_tile(struct softpipe_tile_cache *tc, int x, int y ) { - union tile_address addr = tile_address( x, y, 0, 0, 0 ); + union tile_address addr = tile_address( x, y ); if (tc->last_tile->addr.value == addr.value) return tc->last_tile; -- cgit v1.2.3 From c84abe36a93312cfa061ce1bd005e43eb9f6a6df Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Wed, 29 Jul 2009 23:06:22 +0100 Subject: softpipe: fix typo in clear_tile --- src/gallium/drivers/softpipe/sp_tile_cache.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 2d82badcec..c520aef44f 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -225,7 +225,7 @@ clear_tile(struct softpipe_cached_tile *tile, switch (pf_get_size(format)) { case 1: - memset(tile->data.any, 0, TILE_SIZE * TILE_SIZE); + memset(tile->data.any, clear_value, TILE_SIZE * TILE_SIZE); break; case 2: if (clear_value == 0) { -- cgit v1.2.3 From 4f409da3456070946eda2d8ff5153b3b4306bb46 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 20 Aug 2009 11:25:20 +0100 Subject: softpipe: optimized path for simple mipmap sampling linear-mip-linear-repeat-POT sampling faspath, provides a very nice speedup to apps that do this common type of texturing. Test case: demos/terrain, turn fog off, turn texturing on. Without patch: 12 fps With patch: 20 fps. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 95 +++++++++++++++++++++------- 1 file changed, 71 insertions(+), 24 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 4651d781a9..b32ac9dabb 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -921,8 +921,8 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, } else { - unsigned x1 = (uflr + 1) & (xpot - 1); - unsigned y1 = (vflr + 1) & (ypot - 1); + unsigned x1 = (x0 + 1) & (xpot - 1); + unsigned y1 = (y0 + 1) & (ypot - 1); get_texel_quad_2d_mt(tgsi_sampler, 0, level, x0, y0, x1, y1, tx); } @@ -1028,42 +1028,92 @@ sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; const struct pipe_sampler_state *sampler = samp->sampler; - int level0, level1; + int level0; float lambda; lambda = compute_lambda(texture, sampler, s, t, p, lodbias); level0 = (int)lambda; - level1 = level0 + 1; if (lambda < 0.0) { samp->level = 0; sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, lodbias, rgba ); + s, t, p, 0, rgba ); } else if (level0 >= texture->last_level) { samp->level = texture->last_level; sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, lodbias, rgba ); + s, t, p, 0, rgba ); } else { - float rgba0[4][4]; - float rgba1[4][4]; - int c,j; - float levelBlend = lambda - level0; /* blending weight between levels */ + unsigned xpot = 1 << (samp->xpot - level0); + unsigned ypot = 1 << (samp->ypot - level0); + unsigned xpot1 = 1 << (samp->xpot - (level0+1)); + unsigned ypot1 = 1 << (samp->ypot - (level0+1)); + unsigned j; - samp->level = level0; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, lodbias, rgba0 ); + for (j = 0; j < QUAD_SIZE; j++) { + int c; + + float u = s[j] * xpot - 0.5F; + float v = t[j] * ypot - 0.5F; - samp->level++; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, lodbias, rgba1 ); + int uflr = util_ifloor(u); + int vflr = util_ifloor(v); + + float xw = u - (float)uflr; + float yw = v - (float)vflr; + + int x0 = uflr & (xpot - 1); + int y0 = vflr & (ypot - 1); + + const float *tx0[4]; + const float *tx1[4]; + + if (x0 % TILE_SIZE != TILE_SIZE-1 && + y0 % TILE_SIZE != TILE_SIZE-1) + { + get_texel_quad_2d(tgsi_sampler, 0, level0, x0, y0, tx0); + } + else + { + unsigned x1 = (x0 + 1) & (xpot - 1); + unsigned y1 = (y0 + 1) & (ypot - 1); + get_texel_quad_2d_mt(tgsi_sampler, 0, level0, + x0, y0, x1, y1, tx0); + } - for (c = 0; c < 4; c++) - for (j = 0; j < 4; j++) - rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); - } + x0 /= 2; + y0 /= 2; + /* also need to adjust xw, yw?? */ + + if (x0 % TILE_SIZE != TILE_SIZE-1 && + y0 % TILE_SIZE != TILE_SIZE-1) + { + get_texel_quad_2d(tgsi_sampler, 0, level0+1, x0, y0, tx1); + } + else + { + unsigned x1 = (x0 + 1) & (xpot1 - 1); + unsigned y1 = (y0 + 1) & (ypot1 - 1); + get_texel_quad_2d_mt(tgsi_sampler, 0, level0+1, + x0, y0, x1, y1, tx1); + } + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + float rgba0 = lerp_2d(xw, yw, + tx0[0][c], tx0[1][c], + tx0[2][c], tx0[3][c]); + + float rgba1 = lerp_2d(xw, yw, + tx1[0][c], tx1[1][c], + tx1[2][c], tx1[3][c]); + + rgba[c][j] = lerp(levelBlend, rgba0, rgba1); + } + } + } } /** @@ -1567,10 +1617,7 @@ sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { switch (sampler->min_img_filter) { case PIPE_TEX_FILTER_LINEAR: - /* This one not working yet: - */ - if (0) - tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT; + tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT; break; default: break; -- cgit v1.2.3 From 79a7ddb57a04cde5a4a0c27eb4a9b6889d12622a Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 20 Aug 2009 15:46:51 +0100 Subject: softpipe: fix glitch in texel lookups on fastpaths Fixes two issues - firstly for mipmap levels with one or more dimensions smaller than tilesize, the code was sampling off the edge of the texture (but still within the tile). Secondly, in the linear_mipmap_linear case, both the default code and new fastpath were incorrect. This change fixes the fastpath and adds a comment to the default path, which still needs to be fixed. Basically the issue is that the coordinates in the smaller texture level are/were being computed by just dividing thecoordinates from the larger texture level by two, as in: x0[j] /= 2; y0[j] /= 2; x1[j] /= 2; y1[j] /= 2; The issues with this are signficant. Initially x1 is most often equal to x0+1, but after this, it will likely be equal to x0, so we will not actually be performing the linear blend within the smaller mipmap. The fastpath code avoided this (recalculated x1), but was still using the weighting factors from the larger mipmap level (xw, yw), which were incorrect. Change the fastpath code to do two full, independent linear samples of the two mipmap levels before blending. The default code needs to do the same thing. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 87 ++++++++-------------------- 1 file changed, 23 insertions(+), 64 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index b32ac9dabb..24f3311772 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -894,6 +894,9 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, unsigned xpot = 1 << (samp->xpot - level); unsigned ypot = 1 << (samp->ypot - level); + unsigned xmax = MIN2(TILE_SIZE, xpot) - 1; + unsigned ymax = MIN2(TILE_SIZE, ypot) - 1; + for (j = 0; j < QUAD_SIZE; j++) { int c; @@ -914,8 +917,7 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, /* Can we fetch all four at once: */ - if (x0 % TILE_SIZE != TILE_SIZE-1 && - y0 % TILE_SIZE != TILE_SIZE-1) + if (x0 < xmax && y0 < ymax) { get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx); } @@ -1045,72 +1047,22 @@ sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler s, t, p, 0, rgba ); } else { - float levelBlend = lambda - level0; /* blending weight between levels */ - unsigned xpot = 1 << (samp->xpot - level0); - unsigned ypot = 1 << (samp->ypot - level0); - unsigned xpot1 = 1 << (samp->xpot - (level0+1)); - unsigned ypot1 = 1 << (samp->ypot - (level0+1)); - unsigned j; - - for (j = 0; j < QUAD_SIZE; j++) { - int c; - - float u = s[j] * xpot - 0.5F; - float v = t[j] * ypot - 0.5F; - - int uflr = util_ifloor(u); - int vflr = util_ifloor(v); - - float xw = u - (float)uflr; - float yw = v - (float)vflr; - - int x0 = uflr & (xpot - 1); - int y0 = vflr & (ypot - 1); - - const float *tx0[4]; - const float *tx1[4]; - - if (x0 % TILE_SIZE != TILE_SIZE-1 && - y0 % TILE_SIZE != TILE_SIZE-1) - { - get_texel_quad_2d(tgsi_sampler, 0, level0, x0, y0, tx0); - } - else - { - unsigned x1 = (x0 + 1) & (xpot - 1); - unsigned y1 = (y0 + 1) & (ypot - 1); - get_texel_quad_2d_mt(tgsi_sampler, 0, level0, - x0, y0, x1, y1, tx0); - } + float levelBlend = lambda - level0; + float rgba0[4][4]; + float rgba1[4][4]; + int c,j; - x0 /= 2; - y0 /= 2; - /* also need to adjust xw, yw?? */ + samp->level = level0; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba0 ); - if (x0 % TILE_SIZE != TILE_SIZE-1 && - y0 % TILE_SIZE != TILE_SIZE-1) - { - get_texel_quad_2d(tgsi_sampler, 0, level0+1, x0, y0, tx1); - } - else - { - unsigned x1 = (x0 + 1) & (xpot1 - 1); - unsigned y1 = (y0 + 1) & (ypot1 - 1); - get_texel_quad_2d_mt(tgsi_sampler, 0, level0+1, - x0, y0, x1, y1, tx1); - } + samp->level = level0+1; + sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, + s, t, p, 0, rgba1 ); - /* interpolate R, G, B, A */ + for (j = 0; j < QUAD_SIZE; j++) { for (c = 0; c < 4; c++) { - float rgba0 = lerp_2d(xw, yw, - tx0[0][c], tx0[1][c], - tx0[2][c], tx0[3][c]); - - float rgba1 = lerp_2d(xw, yw, - tx1[0][c], tx1[1][c], - tx1[2][c], tx1[3][c]); - - rgba[c][j] = lerp(levelBlend, rgba0, rgba1); + rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); } } } @@ -1209,6 +1161,13 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, if (level0 != level1) { /* get texels from second mipmap level and blend */ float rgba2[4][4]; + + /* XXX: This is incorrect -- will often end up with (x0 + * == x1 && y0 == y1), meaning that we fetch the same + * texel four times and linearly interpolate between + * identical values. The correct approach would be to + * call linear_texcoord again for the second level. + */ x0[j] /= 2; y0[j] /= 2; x1[j] /= 2; -- cgit v1.2.3 From 1fd40e506c2207664f0c3f435e4614472ea4c540 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 20 Aug 2009 18:12:44 +0100 Subject: softpipe: slightly optimized tiling calculation --- src/gallium/drivers/softpipe/sp_tex_sample.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 24f3311772..90371d6353 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -893,9 +893,8 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); unsigned ypot = 1 << (samp->ypot - level); - - unsigned xmax = MIN2(TILE_SIZE, xpot) - 1; - unsigned ymax = MIN2(TILE_SIZE, ypot) - 1; + unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */ + unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */ for (j = 0; j < QUAD_SIZE; j++) { int c; -- cgit v1.2.3 From 0d9979d9ec5b931856d29c4ec44edb1f4931d1ac Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 20 Aug 2009 18:13:25 +0100 Subject: softpipe: fix xpot calculation typo in sp_get_samples_2d_nearest_clamp_POT --- src/gallium/drivers/softpipe/sp_tex_sample.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 90371d6353..2987548fb3 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -985,8 +985,8 @@ sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); unsigned j; unsigned level = samp->level; - unsigned xpot = (1<xpot); - unsigned ypot = (1<ypot); + unsigned xpot = 1 << (samp->xpot - level); + unsigned ypot = 1 << (samp->ypot - level); for (j = 0; j < QUAD_SIZE; j++) { int c; -- cgit v1.2.3 From 00c835918259f8d41c3f74eca679a972713b11b2 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 20 Aug 2009 18:36:57 +0100 Subject: softpipe: allow the existing sampler routines to be hooked up directly Let eg. sp_get_samples_rect be hooked directly in as the tgsi sampler routine. Add a field to determine whether this is a vertex or fragment sampling call, and massage parameters to match the tgsi call. --- src/gallium/drivers/softpipe/sp_context.c | 6 +- src/gallium/drivers/softpipe/sp_state_derived.c | 3 +- src/gallium/drivers/softpipe/sp_tex_sample.c | 193 ++++++++++-------------- src/gallium/drivers/softpipe/sp_tex_sample.h | 21 +-- 4 files changed, 94 insertions(+), 129 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index e35c6b3aec..a0196955c8 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -230,14 +230,16 @@ softpipe_create( struct pipe_screen *screen ) /* vertex shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples_vertex; + softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples; + softpipe->tgsi.vert_samplers[i].processor = TGSI_PROCESSOR_VERTEX; softpipe->tgsi.vert_samplers[i].cache = softpipe->tex_cache[i]; softpipe->tgsi.vert_samplers_list[i] = &softpipe->tgsi.vert_samplers[i]; } /* fragment shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples_fragment; + softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; + softpipe->tgsi.frag_samplers[i].processor = TGSI_PROCESSOR_FRAGMENT; softpipe->tgsi.frag_samplers[i].cache = softpipe->tex_cache[i]; softpipe->tgsi.frag_samplers_list[i] = &softpipe->tgsi.frag_samplers[i]; } diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 3ed1de7e17..1f6e2ccb83 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -202,13 +202,14 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { softpipe->tgsi.vert_samplers[i].sampler = softpipe->sampler[i]; softpipe->tgsi.vert_samplers[i].texture = softpipe->texture[i]; + softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; } /* fragment shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i]; softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i]; - softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples_fragment; + softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; } for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 2987548fb3..6c75158d59 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -41,6 +41,7 @@ #include "sp_tile_cache.h" #include "pipe/p_context.h" #include "pipe/p_defines.h" +#include "pipe/p_shader_tokens.h" #include "util/u_math.h" #include "util/u_memory.h" @@ -521,15 +522,20 @@ choose_cube_face(float rx, float ry, float rz, float *newS, float *newT) * This is only done for fragment shaders, not vertex shaders. */ static float -compute_lambda(const struct pipe_texture *tex, - const struct pipe_sampler_state *sampler, +compute_lambda(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias) { + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; float rho, lambda; + if (samp->processor == TGSI_PROCESSOR_VERTEX) + return lodbias; + assert(sampler->normalized_coords); assert(s); @@ -538,7 +544,7 @@ compute_lambda(const struct pipe_texture *tex, float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]; dsdx = fabsf(dsdx); dsdy = fabsf(dsdy); - rho = MAX2(dsdx, dsdy) * tex->width[0]; + rho = MAX2(dsdx, dsdy) * texture->width[0]; } if (t) { float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]; @@ -546,7 +552,7 @@ compute_lambda(const struct pipe_texture *tex, float max; dtdx = fabsf(dtdx); dtdy = fabsf(dtdy); - max = MAX2(dtdx, dtdy) * tex->height[0]; + max = MAX2(dtdx, dtdy) * texture->height[0]; rho = MAX2(rho, max); } if (p) { @@ -555,7 +561,7 @@ compute_lambda(const struct pipe_texture *tex, float max; dpdx = fabsf(dpdx); dpdy = fabsf(dpdy); - max = MAX2(dpdx, dpdy) * tex->depth[0]; + max = MAX2(dpdx, dpdy) * texture->depth[0]; rho = MAX2(rho, max); } @@ -579,16 +585,18 @@ compute_lambda(const struct pipe_texture *tex, * \param imgFilter Returns either the min or mag filter, depending on lambda */ static void -choose_mipmap_levels(const struct pipe_texture *texture, - const struct pipe_sampler_state *sampler, +choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, unsigned *level0, unsigned *level1, float *levelBlend, unsigned *imgFilter) { + const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { /* no mipmap selection needed */ *level0 = *level1 = CLAMP((int) sampler->min_lod, @@ -598,7 +606,7 @@ choose_mipmap_levels(const struct pipe_texture *texture, /* non-mipmapped texture, but still need to determine if doing * minification or magnification. */ - float lambda = compute_lambda(texture, sampler, s, t, p, lodbias); + float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); if (lambda <= 0.0) { *imgFilter = sampler->mag_img_filter; } @@ -611,14 +619,7 @@ choose_mipmap_levels(const struct pipe_texture *texture, } } else { - float lambda; - - if (computeLambda) - /* fragment shader */ - lambda = compute_lambda(texture, sampler, s, t, p, lodbias); - else - /* vertex shader */ - lambda = lodbias; /* not really a bias, but absolute LOD */ + float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); if (lambda <= 0.0) { /* XXX threshold depends on the filter */ /* magnifying */ @@ -1028,11 +1029,10 @@ sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler { struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; int level0; float lambda; - lambda = compute_lambda(texture, sampler, s, t, p, lodbias); + lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); level0 = (int)lambda; if (lambda < 0.0) { @@ -1072,11 +1072,10 @@ sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler * Could probably extend for 3D... */ static void -sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, +sp_get_samples_2d_common(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE], const unsigned faces[4]) @@ -1088,7 +1087,8 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, int width, height; float levelBlend; - choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias, + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, &level0, &level1, &levelBlend, &imgFilter); assert(sampler->normalized_coords); @@ -1199,42 +1199,39 @@ sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler, static INLINE void -sp_get_samples_1d(const struct tgsi_sampler *sampler, +sp_get_samples_1d(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { static const unsigned faces[4] = {0, 0, 0, 0}; static const float tzero[4] = {0, 0, 0, 0}; sp_get_samples_2d_common(sampler, s, tzero, NULL, - computeLambda, lodbias, rgba, faces); + lodbias, rgba, faces); } static INLINE void -sp_get_samples_2d(const struct tgsi_sampler *sampler, +sp_get_samples_2d(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { static const unsigned faces[4] = {0, 0, 0, 0}; sp_get_samples_2d_common(sampler, s, t, p, - computeLambda, lodbias, rgba, faces); + lodbias, rgba, faces); } static INLINE void -sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler, +sp_get_samples_3d(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { @@ -1247,7 +1244,8 @@ sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler, float levelBlend; const uint face = 0; - choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias, + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, &level0, &level1, &levelBlend, &imgFilter); assert(sampler->normalized_coords); @@ -1356,11 +1354,10 @@ sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler, static void -sp_get_samples_cube(const struct tgsi_sampler *sampler, +sp_get_samples_cube(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { @@ -1370,16 +1367,15 @@ sp_get_samples_cube(const struct tgsi_sampler *sampler, faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j); } sp_get_samples_2d_common(sampler, ssss, tttt, NULL, - computeLambda, lodbias, rgba, faces); + lodbias, rgba, faces); } static void -sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, +sp_get_samples_rect(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { @@ -1391,7 +1387,8 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, int width, height; float levelBlend; - choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias, + choose_mipmap_levels(tgsi_sampler, s, t, p, + lodbias, &level0, &level1, &levelBlend, &imgFilter); /* texture RECTS cannot be mipmapped */ @@ -1447,90 +1444,77 @@ sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler, /** - * Common code for vertex/fragment program texture sampling. + * Error condition handler */ static INLINE void +sp_get_samples_null(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + int i,j; + + for (i = 0; i < 4; i++) + for (j = 0; j < 4; j++) + rgba[i][j] = 1.0; +} + +/** + * Called via tgsi_sampler::get_samples() when using a sampler for the + * first time. Determine the actual sampler function, link it in and + * call it. + */ +void sp_get_samples(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], - boolean computeLambda, float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; const struct pipe_sampler_state *sampler = samp->sampler; - if (!texture) - return; + /* Default to the 'undefined' case: + */ + tgsi_sampler->get_samples = sp_get_samples_null; + + if (!texture) { + assert(0); /* is this legal?? */ + goto out; + } + + if (!sampler->normalized_coords) { + assert (texture->target == PIPE_TEXTURE_2D); + tgsi_sampler->get_samples = sp_get_samples_rect; + goto out; + } switch (texture->target) { case PIPE_TEXTURE_1D: - assert(sampler->normalized_coords); - sp_get_samples_1d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba); + tgsi_sampler->get_samples = sp_get_samples_1d; break; case PIPE_TEXTURE_2D: - if (sampler->normalized_coords) - sp_get_samples_2d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba); - else - sp_get_samples_rect(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba); + tgsi_sampler->get_samples = sp_get_samples_2d; break; case PIPE_TEXTURE_3D: - assert(sampler->normalized_coords); - sp_get_samples_3d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba); + tgsi_sampler->get_samples = sp_get_samples_3d; break; case PIPE_TEXTURE_CUBE: - assert(sampler->normalized_coords); - sp_get_samples_cube(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba); + tgsi_sampler->get_samples = sp_get_samples_cube; break; default: assert(0); + break; } -#if 0 /* DEBUG */ - { - int i; - printf("Sampled at %f, %f, %f:\n", s[0], t[0], p[0]); - for (i = 0; i < 4; i++) { - printf("Frag %d: %f %f %f %f\n", i, - rgba[0][i], - rgba[1][i], - rgba[2][i], - rgba[3][i]); - } - } -#endif -} - -static void -sp_get_samples_fallback(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) -{ - sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba); -} - -/** - * Called via tgsi_sampler::get_samples() when running a fragment shader. - * Get four filtered RGBA values from the sampler's texture. - */ -void -sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) -{ - struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; - - tgsi_sampler->get_samples = sp_get_samples_fallback; + /* Do this elsewhere: + */ + samp->xpot = util_unsigned_logbase2( samp->texture->width[0] ); + samp->ypot = util_unsigned_logbase2( samp->texture->height[0] ); /* Try to hook in a faster sampler. Ultimately we'll have to * code-generate these. Luckily most of this looks like it is @@ -1542,9 +1526,6 @@ sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, sampler->compare_mode == FALSE && sampler->normalized_coords) { - samp->xpot = util_unsigned_logbase2( samp->texture->width[0] ); - samp->ypot = util_unsigned_logbase2( samp->texture->height[0] ); - if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level); @@ -1593,21 +1574,7 @@ sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, sampler->normalized_coords, TRUE); } +out: tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba ); } - -/** - * Called via tgsi_sampler::get_samples() when running a vertex shader. - * Get four filtered RGBA values from the sampler's texture. - */ -void -sp_get_samples_vertex(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) -{ - sp_get_samples(tgsi_sampler, s, t, p, FALSE, lodbias, rgba); -} diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index 0650c7830b..c73ae44131 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -39,6 +39,8 @@ struct sp_shader_sampler { struct tgsi_sampler base; /**< base class */ + unsigned processor; + /* For sp_get_samples_2d_linear_POT: */ unsigned xpot; @@ -60,21 +62,14 @@ sp_shader_sampler(const struct tgsi_sampler *sampler) } -extern void -sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]); extern void -sp_get_samples_vertex(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]); +sp_get_samples(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]); #endif /* SP_TEX_SAMPLE_H */ -- cgit v1.2.3 From b1cc196e6d18494348c2974aad5d85d1b8281ce0 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 21 Aug 2009 18:07:35 +0100 Subject: util: add util_is_power_of_two function --- src/gallium/auxiliary/util/u_math.h | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h index d30fa3c2d5..163522d3ef 100644 --- a/src/gallium/auxiliary/util/u_math.h +++ b/src/gallium/auxiliary/util/u_math.h @@ -273,6 +273,14 @@ util_fast_pow(float x, float y) return util_fast_exp2(util_fast_log2(x) * y); } +/* Note that this counts zero as a power of two. + */ +static INLINE boolean +util_is_power_of_two( unsigned v ) +{ + return (v & (v-1)) == 0; +} + /** -- cgit v1.2.3 From 4fc7d0345a18042a79686940fb7cc4e698cc9192 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 21 Aug 2009 17:13:11 +0100 Subject: softpipe: rework texture sampling code Split into component pieces, stitch together at runtime using function pointers. Make it possible to utilize the existing fastpaths as image-level filters for generic mip-filtering routines. Remove special case for rectangle filtering, as it can now be handled by the 2d path. As most of the mesa demo texturing was already covered by fast paths, its harder to find examples of speedups, but tunnel gets a boost as mip-nearest filtering is now able to access the img_2d_linear_wrap_POT functions for sampling within a mipmap level. --- src/gallium/drivers/softpipe/sp_context.c | 15 - src/gallium/drivers/softpipe/sp_context.h | 23 +- src/gallium/drivers/softpipe/sp_state.h | 1 + src/gallium/drivers/softpipe/sp_state_blend.c | 4 +- src/gallium/drivers/softpipe/sp_state_derived.c | 18 +- src/gallium/drivers/softpipe/sp_state_fs.c | 3 + src/gallium/drivers/softpipe/sp_state_sampler.c | 104 +- src/gallium/drivers/softpipe/sp_tex_sample.c | 2198 ++++++++++++----------- src/gallium/drivers/softpipe/sp_tex_sample.h | 94 +- src/gallium/drivers/softpipe/sp_texture.c | 8 +- src/gallium/drivers/softpipe/sp_texture.h | 4 + 11 files changed, 1361 insertions(+), 1111 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index a0196955c8..ef8faab3bd 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -228,21 +228,6 @@ softpipe_create( struct pipe_screen *screen ) softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe); softpipe->quad.blend = sp_quad_blend_stage(softpipe); - /* vertex shader samplers */ - for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples; - softpipe->tgsi.vert_samplers[i].processor = TGSI_PROCESSOR_VERTEX; - softpipe->tgsi.vert_samplers[i].cache = softpipe->tex_cache[i]; - softpipe->tgsi.vert_samplers_list[i] = &softpipe->tgsi.vert_samplers[i]; - } - - /* fragment shader samplers */ - for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; - softpipe->tgsi.frag_samplers[i].processor = TGSI_PROCESSOR_FRAGMENT; - softpipe->tgsi.frag_samplers[i].cache = softpipe->tex_cache[i]; - softpipe->tgsi.frag_samplers_list[i] = &softpipe->tgsi.frag_samplers[i]; - } /* * Create drawing context and plug our rendering stage into it. diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index fa3306c020..068a892f25 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -36,7 +36,6 @@ #include "draw/draw_vertex.h" #include "sp_quad_pipe.h" -#include "sp_tex_sample.h" struct softpipe_vbuf_render; @@ -51,12 +50,12 @@ struct softpipe_context { struct pipe_context pipe; /**< base class */ /** Constant state objects */ - const struct pipe_blend_state *blend; - const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS]; - const struct pipe_depth_stencil_alpha_state *depth_stencil; - const struct pipe_rasterizer_state *rasterizer; - const struct sp_fragment_shader *fs; - const struct sp_vertex_shader *vs; + struct pipe_blend_state *blend; + struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + struct pipe_depth_stencil_alpha_state *depth_stencil; + struct pipe_rasterizer_state *rasterizer; + struct sp_fragment_shader *fs; + struct sp_vertex_shader *vs; /** Other rendering state */ struct pipe_blend_color blend_color; @@ -123,10 +122,8 @@ struct softpipe_context { /** TGSI exec things */ struct { - struct sp_shader_sampler vert_samplers[PIPE_MAX_SAMPLERS]; - struct sp_shader_sampler *vert_samplers_list[PIPE_MAX_SAMPLERS]; - struct sp_shader_sampler frag_samplers[PIPE_MAX_SAMPLERS]; - struct sp_shader_sampler *frag_samplers_list[PIPE_MAX_SAMPLERS]; + struct sp_sampler_varient *vert_samplers_list[PIPE_MAX_SAMPLERS]; + struct sp_sampler_varient *frag_samplers_list[PIPE_MAX_SAMPLERS]; } tgsi; /** The primitive drawing context */ @@ -155,5 +152,9 @@ softpipe_context( struct pipe_context *pipe ) return (struct softpipe_context *)pipe; } +void +softpipe_reset_sampler_varients(struct softpipe_context *softpipe); + + #endif /* SP_CONTEXT_H */ diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h index 9776e978e3..77ee3c1136 100644 --- a/src/gallium/drivers/softpipe/sp_state.h +++ b/src/gallium/drivers/softpipe/sp_state.h @@ -87,6 +87,7 @@ struct sp_fragment_shader { struct sp_vertex_shader { struct pipe_shader_state shader; struct draw_vertex_shader *draw_data; + int max_sampler; /* -1 if no samplers */ }; diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c index 384fe559af..efed082f82 100644 --- a/src/gallium/drivers/softpipe/sp_state_blend.c +++ b/src/gallium/drivers/softpipe/sp_state_blend.c @@ -45,7 +45,7 @@ void softpipe_bind_blend_state( struct pipe_context *pipe, { struct softpipe_context *softpipe = softpipe_context(pipe); - softpipe->blend = (const struct pipe_blend_state *)blend; + softpipe->blend = (struct pipe_blend_state *)blend; softpipe->dirty |= SP_NEW_BLEND; } @@ -86,7 +86,7 @@ softpipe_bind_depth_stencil_state(struct pipe_context *pipe, { struct softpipe_context *softpipe = softpipe_context(pipe); - softpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil; + softpipe->depth_stencil = (struct pipe_depth_stencil_alpha_state *)depth_stencil; softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA; } diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 1f6e2ccb83..5310928332 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -198,19 +198,7 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) { unsigned i; - /* vertex shader samplers */ - for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.vert_samplers[i].sampler = softpipe->sampler[i]; - softpipe->tgsi.vert_samplers[i].texture = softpipe->texture[i]; - softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; - } - - /* fragment shader samplers */ - for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i]; - softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i]; - softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples; - } + softpipe_reset_sampler_varients( softpipe ); for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { sp_tile_cache_validate_texture( softpipe->tex_cache[i] ); @@ -232,7 +220,9 @@ void softpipe_update_derived( struct softpipe_context *softpipe ) } if (softpipe->dirty & (SP_NEW_SAMPLER | - SP_NEW_TEXTURE)) + SP_NEW_TEXTURE | + SP_NEW_FS | + SP_NEW_VS)) update_tgsi_samplers( softpipe ); if (softpipe->dirty & (SP_NEW_RASTERIZER | diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c index 108ac8b9bb..3a45321923 100644 --- a/src/gallium/drivers/softpipe/sp_state_fs.c +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -34,6 +34,7 @@ #include "pipe/internal/p_winsys_screen.h" #include "pipe/p_shader_tokens.h" #include "draw/draw_context.h" +#include "draw/draw_vs.h" #include "tgsi/tgsi_dump.h" #include "tgsi/tgsi_scan.h" #include "tgsi/tgsi_parse.h" @@ -108,6 +109,8 @@ softpipe_create_vs_state(struct pipe_context *pipe, if (state->draw_data == NULL) goto fail; + state->max_sampler = state->draw_data->info.file_max[TGSI_FILE_SAMPLER]; + return state; fail: diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c index aa2f3f2ccd..714e638048 100644 --- a/src/gallium/drivers/softpipe/sp_state_sampler.c +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -38,15 +38,32 @@ #include "sp_state.h" #include "sp_texture.h" #include "sp_tile_cache.h" +#include "sp_tex_sample.h" #include "draw/draw_context.h" +struct sp_sampler { + struct pipe_sampler_state base; + struct sp_sampler_varient *varients; + struct sp_sampler_varient *current; +}; + +static struct sp_sampler *sp_sampler( struct pipe_sampler_state *sampler ) +{ + return (struct sp_sampler *)sampler; +} + void * softpipe_create_sampler_state(struct pipe_context *pipe, const struct pipe_sampler_state *sampler) { - return mem_dup(sampler, sizeof(*sampler)); + struct sp_sampler *sp_sampler = CALLOC_STRUCT(sp_sampler); + + sp_sampler->base = *sampler; + sp_sampler->varients = NULL; + + return (void *)sp_sampler; } @@ -106,10 +123,95 @@ softpipe_set_sampler_textures(struct pipe_context *pipe, } + +static struct sp_sampler_varient * +get_sampler_varient( struct sp_sampler *sampler, + struct pipe_texture *texture, + unsigned processor ) +{ + struct softpipe_texture *sp_texture = softpipe_texture(texture); + struct sp_sampler_varient *v = NULL; + union sp_sampler_key key; + + key.bits.target = sp_texture->base.target; + key.bits.is_pot = sp_texture->pot; + key.bits.processor = processor; + key.bits.pad = 0; + + if (sampler->current && + key.value == sampler->current->key.value) { + v = sampler->current; + } + + if (v == NULL) { + for (v = sampler->varients; v; v = v->next) + if (v->key.value == key.value) + break; + + if (v == NULL) { + v = sp_create_sampler_varient( &sampler->base, key ); + v->next = sampler->varients; + sampler->varients = v; + } + } + + sampler->current = v; + return v; +} + + + + +void +softpipe_reset_sampler_varients(struct softpipe_context *softpipe) +{ + int i; + + /* It's a bit hard to build these samplers ahead of time -- don't + * really know which samplers are going to be used for vertex and + * fragment programs. + */ + for (i = 0; i <= softpipe->vs->max_sampler; i++) { + if (softpipe->sampler[i]) { + softpipe->tgsi.vert_samplers_list[i] = + get_sampler_varient( sp_sampler(softpipe->sampler[i]), + softpipe->texture[i], + TGSI_PROCESSOR_VERTEX ); + + sp_sampler_varient_bind_texture( softpipe->tgsi.vert_samplers_list[i], + softpipe->tex_cache[i], + softpipe->texture[i] ); + } + } + + for (i = 0; i <= softpipe->fs->info.file_max[TGSI_FILE_SAMPLER]; i++) { + if (softpipe->sampler[i]) { + softpipe->tgsi.frag_samplers_list[i] = + get_sampler_varient( sp_sampler(softpipe->sampler[i]), + softpipe->texture[i], + TGSI_PROCESSOR_FRAGMENT ); + + sp_sampler_varient_bind_texture( softpipe->tgsi.frag_samplers_list[i], + softpipe->tex_cache[i], + softpipe->texture[i] ); + } + } +} + + + void softpipe_delete_sampler_state(struct pipe_context *pipe, void *sampler) { + struct sp_sampler *sp_sampler = (struct sp_sampler *)sampler; + struct sp_sampler_varient *v, *tmp; + + for (v = sp_sampler->varients; v; v = tmp) { + tmp = v->next; + sp_sampler_varient_destroy(v); + } + FREE( sampler ); } diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 6c75158d59..7bc689a298 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -31,6 +31,7 @@ * * Authors: * Brian Paul + * Keith Whitwell */ #include "sp_context.h" @@ -116,133 +117,157 @@ lerp_3d(float a, float b, float c, * \param icoord returns the integer texcoords * \return integer texture index */ -static INLINE void -nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, +static void +wrap_nearest_repeat(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + + /* s limited to [0,1) */ + /* i limited to [0,size-1] */ + for (ch = 0; ch < 4; ch++) { + int i = util_ifloor(s[ch] * size); + icoord[ch] = REMAINDER(i, size); + } +} + + +static void +wrap_nearest_clamp(const float s[4], unsigned size, int icoord[4]) { uint ch; - switch (wrapMode) { - case PIPE_TEX_WRAP_REPEAT: - /* s limited to [0,1) */ - /* i limited to [0,size-1] */ - for (ch = 0; ch < 4; ch++) { - int i = util_ifloor(s[ch] * size); - icoord[ch] = REMAINDER(i, size); - } - return; - case PIPE_TEX_WRAP_CLAMP: + /* s limited to [0,1] */ + /* i limited to [0,size-1] */ + for (ch = 0; ch < 4; ch++) { + if (s[ch] <= 0.0F) + icoord[ch] = 0; + else if (s[ch] >= 1.0F) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(s[ch] * size); + } +} + + +static void +wrap_nearest_clamp_to_edge(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + if (s[ch] < min) + icoord[ch] = 0; + else if (s[ch] > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(s[ch] * size); + } +} + + +static void +wrap_nearest_clamp_to_border(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + /* s limited to [min,max] */ + /* i limited to [-1, size] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + if (s[ch] <= min) + icoord[ch] = -1; + else if (s[ch] >= max) + icoord[ch] = size; + else + icoord[ch] = util_ifloor(s[ch] * size); + } +} + +static void +wrap_nearest_mirror_repeat(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const int flr = util_ifloor(s[ch]); + float u; + if (flr & 1) + u = 1.0F - (s[ch] - (float) flr); + else + u = s[ch] - (float) flr; + if (u < min) + icoord[ch] = 0; + else if (u > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } +} + +static void +wrap_nearest_mirror_clamp(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { /* s limited to [0,1] */ /* i limited to [0,size-1] */ - for (ch = 0; ch < 4; ch++) { - if (s[ch] <= 0.0F) - icoord[ch] = 0; - else if (s[ch] >= 1.0F) - icoord[ch] = size - 1; - else - icoord[ch] = util_ifloor(s[ch] * size); - } - return; - case PIPE_TEX_WRAP_CLAMP_TO_EDGE: - { - /* s limited to [min,max] */ - /* i limited to [0, size-1] */ - const float min = 1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - if (s[ch] < min) - icoord[ch] = 0; - else if (s[ch] > max) - icoord[ch] = size - 1; - else - icoord[ch] = util_ifloor(s[ch] * size); - } - } - return; - case PIPE_TEX_WRAP_CLAMP_TO_BORDER: - { - /* s limited to [min,max] */ - /* i limited to [-1, size] */ - const float min = -1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - if (s[ch] <= min) - icoord[ch] = -1; - else if (s[ch] >= max) - icoord[ch] = size; - else - icoord[ch] = util_ifloor(s[ch] * size); - } - } - return; - case PIPE_TEX_WRAP_MIRROR_REPEAT: - { - const float min = 1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - const int flr = util_ifloor(s[ch]); - float u; - if (flr & 1) - u = 1.0F - (s[ch] - (float) flr); - else - u = s[ch] - (float) flr; - if (u < min) - icoord[ch] = 0; - else if (u > max) - icoord[ch] = size - 1; - else - icoord[ch] = util_ifloor(u * size); - } - } - return; - case PIPE_TEX_WRAP_MIRROR_CLAMP: - for (ch = 0; ch < 4; ch++) { - /* s limited to [0,1] */ - /* i limited to [0,size-1] */ - const float u = fabsf(s[ch]); - if (u <= 0.0F) - icoord[ch] = 0; - else if (u >= 1.0F) - icoord[ch] = size - 1; - else - icoord[ch] = util_ifloor(u * size); - } - return; - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: - { - /* s limited to [min,max] */ - /* i limited to [0, size-1] */ - const float min = 1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - const float u = fabsf(s[ch]); - if (u < min) - icoord[ch] = 0; - else if (u > max) - icoord[ch] = size - 1; - else - icoord[ch] = util_ifloor(u * size); - } - } - return; - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: - { - /* s limited to [min,max] */ - /* i limited to [0, size-1] */ - const float min = -1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - const float u = fabsf(s[ch]); - if (u < min) - icoord[ch] = -1; - else if (u > max) - icoord[ch] = size; - else - icoord[ch] = util_ifloor(u * size); - } - } - return; - default: - assert(0); + const float u = fabsf(s[ch]); + if (u <= 0.0F) + icoord[ch] = 0; + else if (u >= 1.0F) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } +} + +static void +wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const float u = fabsf(s[ch]); + if (u < min) + icoord[ch] = 0; + else if (u > max) + icoord[ch] = size - 1; + else + icoord[ch] = util_ifloor(u * size); + } +} + + +static void +wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + for (ch = 0; ch < 4; ch++) { + const float u = fabsf(s[ch]); + if (u < min) + icoord[ch] = -1; + else if (u > max) + icoord[ch] = size; + else + icoord[ch] = util_ifloor(u * size); } } @@ -257,125 +282,151 @@ nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, * \param w returns blend factor/weight between texture indexes * \param icoord returns the computed integer texture coords */ -static INLINE void -linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, +static void +wrap_linear_repeat(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + + for (ch = 0; ch < 4; ch++) { + float u = s[ch] * size - 0.5F; + icoord0[ch] = REMAINDER(util_ifloor(u), size); + icoord1[ch] = REMAINDER(icoord0[ch] + 1, size); + w[ch] = FRAC(u); + } +} + +static void +wrap_linear_clamp(const float s[4], unsigned size, int icoord0[4], int icoord1[4], float w[4]) { uint ch; + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.0F, 1.0F); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } +} - switch (wrapMode) { - case PIPE_TEX_WRAP_REPEAT: - for (ch = 0; ch < 4; ch++) { - float u = s[ch] * size - 0.5F; - icoord0[ch] = REMAINDER(util_ifloor(u), size); - icoord1[ch] = REMAINDER(icoord0[ch] + 1, size); - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_CLAMP: - for (ch = 0; ch < 4; ch++) { - float u = CLAMP(s[ch], 0.0F, 1.0F); - u = u * size - 0.5f; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_CLAMP_TO_EDGE: - for (ch = 0; ch < 4; ch++) { - float u = CLAMP(s[ch], 0.0F, 1.0F); - u = u * size - 0.5f; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - if (icoord0[ch] < 0) - icoord0[ch] = 0; - if (icoord1[ch] >= (int) size) - icoord1[ch] = size - 1; - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_CLAMP_TO_BORDER: - { - const float min = -1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - float u = CLAMP(s[ch], min, max); - u = u * size - 0.5f; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - w[ch] = FRAC(u); - } - } - break;; - case PIPE_TEX_WRAP_MIRROR_REPEAT: - for (ch = 0; ch < 4; ch++) { - const int flr = util_ifloor(s[ch]); - float u; - if (flr & 1) - u = 1.0F - (s[ch] - (float) flr); - else - u = s[ch] - (float) flr; - u = u * size - 0.5F; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - if (icoord0[ch] < 0) - icoord0[ch] = 0; - if (icoord1[ch] >= (int) size) - icoord1[ch] = size - 1; - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_MIRROR_CLAMP: - for (ch = 0; ch < 4; ch++) { - float u = fabsf(s[ch]); - if (u >= 1.0F) - u = (float) size; - else - u *= size; - u -= 0.5F; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: - for (ch = 0; ch < 4; ch++) { - float u = fabsf(s[ch]); - if (u >= 1.0F) - u = (float) size; - else - u *= size; - u -= 0.5F; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - if (icoord0[ch] < 0) - icoord0[ch] = 0; - if (icoord1[ch] >= (int) size) - icoord1[ch] = size - 1; - w[ch] = FRAC(u); - } - break;; - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: - { - const float min = -1.0F / (2.0F * size); - const float max = 1.0F - min; - for (ch = 0; ch < 4; ch++) { - float u = fabsf(s[ch]); - if (u <= min) - u = min * size; - else if (u >= max) - u = max * size; - else - u *= size; - u -= 0.5F; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - w[ch] = FRAC(u); - } - } - break;; - default: - assert(0); +static void +wrap_linear_clamp_to_edge(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.0F, 1.0F); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } +} + +static void +wrap_linear_clamp_to_border(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], min, max); + u = u * size - 0.5f; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } +} + + +static void +wrap_linear_mirror_repeat(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { + const int flr = util_ifloor(s[ch]); + float u; + if (flr & 1) + u = 1.0F - (s[ch] - (float) flr); + else + u = s[ch] - (float) flr; + u = u * size - 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } +} + +static void +wrap_linear_mirror_clamp(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); + } +} + +static void +wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord0[ch] < 0) + icoord0[ch] = 0; + if (icoord1[ch] >= (int) size) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); + } +} + +static void +wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) +{ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = fabsf(s[ch]); + if (u <= min) + u = min * size; + else if (u >= max) + u = max * size; + else + u *= size; + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); } } @@ -384,27 +435,26 @@ linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size, * For RECT textures / unnormalized texcoords * Only a subset of wrap modes supported. */ -static INLINE void -nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size, +static void +wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4]) { uint ch; - switch (wrapMode) { - case PIPE_TEX_WRAP_CLAMP: - for (ch = 0; ch < 4; ch++) { - int i = util_ifloor(s[ch]); - icoord[ch]= CLAMP(i, 0, (int) size-1); - } - return; - case PIPE_TEX_WRAP_CLAMP_TO_EDGE: - /* fall-through */ - case PIPE_TEX_WRAP_CLAMP_TO_BORDER: - for (ch = 0; ch < 4; ch++) { - icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) ); - } - return; - default: - assert(0); + for (ch = 0; ch < 4; ch++) { + int i = util_ifloor(s[ch]); + icoord[ch]= CLAMP(i, 0, (int) size-1); + } +} + +/* Handles clamp_to_edge and clamp_to_border: + */ +static void +wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size, + int icoord[4]) +{ + uint ch; + for (ch = 0; ch < 4; ch++) { + icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) ); } } @@ -413,157 +463,82 @@ nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size, * For RECT textures / unnormalized texcoords. * Only a subset of wrap modes supported. */ -static INLINE void -linear_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size, +static void +wrap_linear_unorm_clamp(const float s[4], unsigned size, int icoord0[4], int icoord1[4], float w[4]) { uint ch; - switch (wrapMode) { - case PIPE_TEX_WRAP_CLAMP: - for (ch = 0; ch < 4; ch++) { - /* Not exactly what the spec says, but it matches NVIDIA output */ - float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f); - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - w[ch] = FRAC(u); - } - return; - case PIPE_TEX_WRAP_CLAMP_TO_EDGE: - /* fall-through */ - case PIPE_TEX_WRAP_CLAMP_TO_BORDER: - for (ch = 0; ch < 4; ch++) { - float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F); - u -= 0.5F; - icoord0[ch] = util_ifloor(u); - icoord1[ch] = icoord0[ch] + 1; - if (icoord1[ch] > (int) size - 1) - icoord1[ch] = size - 1; - w[ch] = FRAC(u); - } - break; - default: - assert(0); + for (ch = 0; ch < 4; ch++) { + /* Not exactly what the spec says, but it matches NVIDIA output */ + float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f); + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + w[ch] = FRAC(u); } } - -static unsigned -choose_cube_face(float rx, float ry, float rz, float *newS, float *newT) +static void +wrap_linear_unorm_clamp_to_border( const float s[4], unsigned size, + int icoord0[4], int icoord1[4], float w[4]) { - /* - major axis - direction target sc tc ma - ---------- ------------------------------- --- --- --- - +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx - -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx - +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry - -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry - +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz - -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz - */ - const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz); - unsigned face; - float sc, tc, ma; - - if (arx > ary && arx > arz) { - if (rx >= 0.0F) { - face = PIPE_TEX_FACE_POS_X; - sc = -rz; - tc = -ry; - ma = arx; - } - else { - face = PIPE_TEX_FACE_NEG_X; - sc = rz; - tc = -ry; - ma = arx; - } - } - else if (ary > arx && ary > arz) { - if (ry >= 0.0F) { - face = PIPE_TEX_FACE_POS_Y; - sc = rx; - tc = rz; - ma = ary; - } - else { - face = PIPE_TEX_FACE_NEG_Y; - sc = rx; - tc = -rz; - ma = ary; - } - } - else { - if (rz > 0.0F) { - face = PIPE_TEX_FACE_POS_Z; - sc = rx; - tc = -ry; - ma = arz; - } - else { - face = PIPE_TEX_FACE_NEG_Z; - sc = -rx; - tc = -ry; - ma = arz; - } + uint ch; + for (ch = 0; ch < 4; ch++) { + float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F); + u -= 0.5F; + icoord0[ch] = util_ifloor(u); + icoord1[ch] = icoord0[ch] + 1; + if (icoord1[ch] > (int) size - 1) + icoord1[ch] = size - 1; + w[ch] = FRAC(u); } +} + - *newS = ( sc / ma + 1.0F ) * 0.5F; - *newT = ( tc / ma + 1.0F ) * 0.5F; - return face; -} /** * Examine the quad's texture coordinates to compute the partial * derivatives w.r.t X and Y, then compute lambda (level of detail). - * - * This is only done for fragment shaders, not vertex shaders. */ static float -compute_lambda(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias) +compute_lambda_1d(const struct sp_sampler_varient *samp, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; const struct pipe_sampler_state *sampler = samp->sampler; - float rho, lambda; + float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]); + float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]); + float rho = MAX2(dsdx, dsdy) * texture->width[0]; + float lambda; - if (samp->processor == TGSI_PROCESSOR_VERTEX) - return lodbias; + lambda = util_fast_log2(rho); + lambda += lodbias + sampler->lod_bias; + lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod); - assert(sampler->normalized_coords); + return lambda; +} - assert(s); - { - float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]; - float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]; - dsdx = fabsf(dsdx); - dsdy = fabsf(dsdy); - rho = MAX2(dsdx, dsdy) * texture->width[0]; - } - if (t) { - float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]; - float dtdy = t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]; - float max; - dtdx = fabsf(dtdx); - dtdy = fabsf(dtdy); - max = MAX2(dtdx, dtdy) * texture->height[0]; - rho = MAX2(rho, max); - } - if (p) { - float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]; - float dpdy = p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]; - float max; - dpdx = fabsf(dpdx); - dpdy = fabsf(dpdy); - max = MAX2(dpdx, dpdy) * texture->depth[0]; - rho = MAX2(rho, max); - } +static float +compute_lambda_2d(const struct sp_sampler_varient *samp, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) +{ + const struct pipe_texture *texture = samp->texture; + const struct pipe_sampler_state *sampler = samp->sampler; + float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]); + float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]); + float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]); + float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]); + float maxx = MAX2(dsdx, dsdy) * texture->width[0]; + float maxy = MAX2(dtdx, dtdy) * texture->height[0]; + float rho = MAX2(maxx, maxy); + float lambda; lambda = util_fast_log2(rho); lambda += lodbias + sampler->lod_bias; @@ -573,88 +548,56 @@ compute_lambda(struct tgsi_sampler *tgsi_sampler, } -/** - * Do several things here: - * 1. Compute lambda from the texcoords, if needed - * 2. Determine if we're minifying or magnifying - * 3. If minifying, choose mipmap levels - * 4. Return image filter to use within mipmap images - * \param level0 Returns first mipmap level to sample from - * \param level1 Returns second mipmap level to sample from - * \param levelBlend Returns blend factor between levels, in [0,1] - * \param imgFilter Returns either the min or mag filter, depending on lambda - */ -static void -choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - unsigned *level0, unsigned *level1, float *levelBlend, - unsigned *imgFilter) +static float +compute_lambda_3d(const struct sp_sampler_varient *samp, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); const struct pipe_texture *texture = samp->texture; const struct pipe_sampler_state *sampler = samp->sampler; + float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]); + float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]); + float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]); + float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]); + float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]); + float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]); + float maxx = MAX2(dsdx, dsdy) * texture->width[0]; + float maxy = MAX2(dtdx, dtdy) * texture->height[0]; + float maxz = MAX2(dpdx, dpdy) * texture->depth[0]; + float rho, lambda; - if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { - /* no mipmap selection needed */ - *level0 = *level1 = CLAMP((int) sampler->min_lod, - 0, (int) texture->last_level); - - if (sampler->min_img_filter != sampler->mag_img_filter) { - /* non-mipmapped texture, but still need to determine if doing - * minification or magnification. - */ - float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); - if (lambda <= 0.0) { - *imgFilter = sampler->mag_img_filter; - } - else { - *imgFilter = sampler->min_img_filter; - } - } - else { - *imgFilter = sampler->mag_img_filter; - } - } - else { - float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); + rho = MAX2(maxx, maxy); + rho = MAX2(rho, maxz); - if (lambda <= 0.0) { /* XXX threshold depends on the filter */ - /* magnifying */ - *imgFilter = sampler->mag_img_filter; - *level0 = *level1 = 0; - } - else { - /* minifying */ - *imgFilter = sampler->min_img_filter; - - /* choose mipmap level(s) and compute the blend factor between them */ - if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) { - /* Nearest mipmap level */ - const int lvl = (int) (lambda + 0.5); - *level0 = - *level1 = CLAMP(lvl, 0, (int) texture->last_level); - } - else { - /* Linear interpolation between mipmap levels */ - const int lvl = (int) lambda; - *level0 = CLAMP(lvl, 0, (int) texture->last_level); - *level1 = CLAMP(lvl + 1, 0, (int) texture->last_level); - *levelBlend = FRAC(lambda); /* blending weight between levels */ - } - } - } + lambda = util_fast_log2(rho); + lambda += lodbias + sampler->lod_bias; + lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod); + + return lambda; } -/** - * Get a texel from a texture, using the texture tile cache. - * - * \param face the cube face in 0..5 - * \param level the mipmap level - * \param x the x coord of texel within 2D image + +static float +compute_lambda_vert(const struct sp_sampler_varient *samp, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) +{ + return lodbias; +} + + + +/** + * Get a texel from a texture, using the texture tile cache. + * + * \param face the cube face in 0..5 + * \param level the mipmap level + * \param x the x coord of texel within 2D image * \param y the y coord of texel within 2D image * \param z which slice of a 3D texture * \param rgba the quad to put the texel/color into @@ -663,12 +606,12 @@ choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler, * XXX maybe move this into sp_tile_cache.c and merge with the * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... */ -static void +static INLINE void get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, unsigned face, unsigned level, int x, int y, const float *out[4]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct softpipe_cached_tile *tile = sp_get_cached_tile_tex(samp->cache, @@ -687,7 +630,7 @@ static INLINE const float * get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, unsigned face, unsigned level, int x, int y) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct softpipe_cached_tile *tile = sp_get_cached_tile_tex(samp->cache, @@ -700,7 +643,7 @@ get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, } -static void +static INLINE void get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, unsigned face, unsigned level, int x0, int y0, @@ -717,12 +660,12 @@ get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, } } -static void +static INLINE void get_texel(const struct tgsi_sampler *tgsi_sampler, unsigned face, unsigned level, int x, int y, int z, float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; const struct pipe_sampler_state *sampler = samp->sampler; @@ -756,140 +699,18 @@ get_texel(const struct tgsi_sampler *tgsi_sampler, } -/** - * Compare texcoord 'p' (aka R) against texture value 'rgba[0]' - * When we sampled the depth texture, the depth value was put into all - * RGBA channels. We look at the red channel here. - * \param rgba quad of (depth) texel values - * \param p texture 'P' components for four pixels in quad - * \param j which pixel in the quad to test [0..3] - */ -static INLINE void -shadow_compare(const struct pipe_sampler_state *sampler, - float rgba[NUM_CHANNELS][QUAD_SIZE], - const float p[QUAD_SIZE], - uint j) -{ - int k; - switch (sampler->compare_func) { - case PIPE_FUNC_LESS: - k = p[j] < rgba[0][j]; - break; - case PIPE_FUNC_LEQUAL: - k = p[j] <= rgba[0][j]; - break; - case PIPE_FUNC_GREATER: - k = p[j] > rgba[0][j]; - break; - case PIPE_FUNC_GEQUAL: - k = p[j] >= rgba[0][j]; - break; - case PIPE_FUNC_EQUAL: - k = p[j] == rgba[0][j]; - break; - case PIPE_FUNC_NOTEQUAL: - k = p[j] != rgba[0][j]; - break; - case PIPE_FUNC_ALWAYS: - k = 1; - break; - case PIPE_FUNC_NEVER: - k = 0; - break; - default: - k = 0; - assert(0); - break; - } - /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ - rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k; - rgba[3][j] = 1.0F; -} -/** - * As above, but do four z/texture comparisons. - */ static INLINE void -shadow_compare4(const struct pipe_sampler_state *sampler, - float rgba[NUM_CHANNELS][QUAD_SIZE], - const float p[QUAD_SIZE]) -{ - int j, k0, k1, k2, k3; - float val; - - /* compare four texcoords vs. four texture samples */ - switch (sampler->compare_func) { - case PIPE_FUNC_LESS: - k0 = p[0] < rgba[0][0]; - k1 = p[1] < rgba[0][1]; - k2 = p[2] < rgba[0][2]; - k3 = p[3] < rgba[0][3]; - break; - case PIPE_FUNC_LEQUAL: - k0 = p[0] <= rgba[0][0]; - k1 = p[1] <= rgba[0][1]; - k2 = p[2] <= rgba[0][2]; - k3 = p[3] <= rgba[0][3]; - break; - case PIPE_FUNC_GREATER: - k0 = p[0] > rgba[0][0]; - k1 = p[1] > rgba[0][1]; - k2 = p[2] > rgba[0][2]; - k3 = p[3] > rgba[0][3]; - break; - case PIPE_FUNC_GEQUAL: - k0 = p[0] >= rgba[0][0]; - k1 = p[1] >= rgba[0][1]; - k2 = p[2] >= rgba[0][2]; - k3 = p[3] >= rgba[0][3]; - break; - case PIPE_FUNC_EQUAL: - k0 = p[0] == rgba[0][0]; - k1 = p[1] == rgba[0][1]; - k2 = p[2] == rgba[0][2]; - k3 = p[3] == rgba[0][3]; - break; - case PIPE_FUNC_NOTEQUAL: - k0 = p[0] != rgba[0][0]; - k1 = p[1] != rgba[0][1]; - k2 = p[2] != rgba[0][2]; - k3 = p[3] != rgba[0][3]; - break; - case PIPE_FUNC_ALWAYS: - k0 = k1 = k2 = k3 = 1; - break; - case PIPE_FUNC_NEVER: - k0 = k1 = k2 = k3 = 0; - break; - default: - k0 = k1 = k2 = k3 = 0; - assert(0); - break; - } - - /* convert four pass/fail values to an intensity in [0,1] */ - val = 0.25F * (k0 + k1 + k2 + k3); - - /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ - for (j = 0; j < 4; j++) { - rgba[0][j] = rgba[1][j] = rgba[2][j] = val; - rgba[3][j] = 1.0F; - } -} - - - -static void -sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); @@ -940,15 +761,15 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, } -static void -sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +static INLINE void +img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); @@ -975,15 +796,15 @@ sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, } -static void -sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +static INLINE void +img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); @@ -1018,238 +839,78 @@ sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, } } - static void -sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - int level0; - float lambda; - - lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias); - level0 = (int)lambda; + unsigned level0, j; + int width; + int x[4]; - if (lambda < 0.0) { - samp->level = 0; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, 0, rgba ); - } - else if (level0 >= texture->last_level) { - samp->level = texture->last_level; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, 0, rgba ); - } - else { - float levelBlend = lambda - level0; - float rgba0[4][4]; - float rgba1[4][4]; - int c,j; + level0 = samp->level; + width = texture->width[level0]; - samp->level = level0; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, 0, rgba0 ); + assert(width > 0); - samp->level = level0+1; - sp_get_samples_2d_linear_repeat_POT( tgsi_sampler, - s, t, p, 0, rgba1 ); + samp->nearest_texcoord_s(s, width, x); - for (j = 0; j < QUAD_SIZE; j++) { - for (c = 0; c < 4; c++) { - rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); - } - } + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, 0, level0, x[j], 0, 0, rgba, j); } } -/** - * Common code for sampling 1D/2D/cube textures. - * Could probably extend for 3D... - */ + static void -sp_get_samples_2d_common(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE], - const unsigned faces[4]) +img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; - unsigned level0, level1, j, imgFilter; + const unsigned *faces = samp->faces; /* zero when not cube-mapping */ + unsigned level0, j; int width, height; - float levelBlend; - - choose_mipmap_levels(tgsi_sampler, s, t, p, - lodbias, - &level0, &level1, &levelBlend, &imgFilter); - - assert(sampler->normalized_coords); + int x[4], y[4]; + level0 = samp->level; width = texture->width[level0]; height = texture->height[level0]; assert(width > 0); - switch (imgFilter) { - case PIPE_TEX_FILTER_NEAREST: - { - int x[4], y[4]; - nearest_texcoord_4(sampler->wrap_s, s, width, x); - nearest_texcoord_4(sampler->wrap_t, t, height, y); - - for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(sampler, rgba, p, j); - } - - if (level0 != level1) { - /* get texels from second mipmap level and blend */ - float rgba2[4][4]; - unsigned c; - x[j] /= 2; - y[j] /= 2; - get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0, - rgba2, j); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ - shadow_compare(sampler, rgba2, p, j); - } - - for (c = 0; c < NUM_CHANNELS; c++) { - rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); - } - } - } - } - break; - case PIPE_TEX_FILTER_LINEAR: - case PIPE_TEX_FILTER_ANISO: - { - int x0[4], y0[4], x1[4], y1[4]; - float xw[4], yw[4]; /* weights */ - - linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw); - linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw); - - for (j = 0; j < QUAD_SIZE; j++) { - float tx[4][4]; /* texels */ - int c; - get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0); - get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1); - get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2); - get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare4(sampler, tx, p); - } - - /* interpolate R, G, B, A */ - for (c = 0; c < 4; c++) { - rgba[c][j] = lerp_2d(xw[j], yw[j], - tx[c][0], tx[c][1], - tx[c][2], tx[c][3]); - } + samp->nearest_texcoord_s(s, width, x); + samp->nearest_texcoord_t(t, height, y); - if (level0 != level1) { - /* get texels from second mipmap level and blend */ - float rgba2[4][4]; - - /* XXX: This is incorrect -- will often end up with (x0 - * == x1 && y0 == y1), meaning that we fetch the same - * texel four times and linearly interpolate between - * identical values. The correct approach would be to - * call linear_texcoord again for the second level. - */ - x0[j] /= 2; - y0[j] /= 2; - x1[j] /= 2; - y1[j] /= 2; - get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0); - get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1); - get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2); - get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ - shadow_compare4(sampler, tx, p); - } - - /* interpolate R, G, B, A */ - for (c = 0; c < 4; c++) { - rgba2[c][j] = lerp_2d(xw[j], yw[j], - tx[c][0], tx[c][1], tx[c][2], tx[c][3]); - } - - for (c = 0; c < NUM_CHANNELS; c++) { - rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); - } - } - } - } - break; - default: - assert(0); + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j); } } -static INLINE void -sp_get_samples_1d(struct tgsi_sampler *sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) -{ - static const unsigned faces[4] = {0, 0, 0, 0}; - static const float tzero[4] = {0, 0, 0, 0}; - sp_get_samples_2d_common(sampler, s, tzero, NULL, - lodbias, rgba, faces); -} - - -static INLINE void -sp_get_samples_2d(struct tgsi_sampler *sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) -{ - static const unsigned faces[4] = {0, 0, 0, 0}; - sp_get_samples_2d_common(sampler, s, t, p, - lodbias, rgba, faces); -} - - -static INLINE void -sp_get_samples_3d(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +static void +img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; - /* get/map pipe_surfaces corresponding to 3D tex slices */ - unsigned level0, level1, j, imgFilter; + unsigned level0, j; int width, height, depth; - float levelBlend; - const uint face = 0; - - choose_mipmap_levels(tgsi_sampler, s, t, p, - lodbias, - &level0, &level1, &levelBlend, &imgFilter); - - assert(sampler->normalized_coords); + int x[4], y[4], z[4]; + level0 = samp->level; width = texture->width[level0]; height = texture->height[level0]; depth = texture->depth[level0]; @@ -1258,323 +919,746 @@ sp_get_samples_3d(struct tgsi_sampler *tgsi_sampler, assert(height > 0); assert(depth > 0); - switch (imgFilter) { - case PIPE_TEX_FILTER_NEAREST: - { - int x[4], y[4], z[4]; - nearest_texcoord_4(sampler->wrap_s, s, width, x); - nearest_texcoord_4(sampler->wrap_t, t, height, y); - nearest_texcoord_4(sampler->wrap_r, p, depth, z); - for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j); - if (level0 != level1) { - /* get texels from second mipmap level and blend */ - float rgba2[4][4]; - unsigned c; - x[j] /= 2; - y[j] /= 2; - z[j] /= 2; - get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j); - for (c = 0; c < NUM_CHANNELS; c++) { - rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]); - } - } - } - } - break; - case PIPE_TEX_FILTER_LINEAR: - case PIPE_TEX_FILTER_ANISO: - { - int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4]; - float xw[4], yw[4], zw[4]; /* interpolation weights */ - linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw); - linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw); - linear_texcoord_4(sampler->wrap_r, p, depth, z0, z1, zw); - - for (j = 0; j < QUAD_SIZE; j++) { - int c; - float tx0[4][4], tx1[4][4]; - get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0); - get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1); - get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2); - get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3); - get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0); - get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1); - get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2); - get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3); - - /* interpolate R, G, B, A */ - for (c = 0; c < 4; c++) { - rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j], - tx0[c][0], tx0[c][1], - tx0[c][2], tx0[c][3], - tx1[c][0], tx1[c][1], - tx1[c][2], tx1[c][3]); - } + samp->nearest_texcoord_s(s, width, x); + samp->nearest_texcoord_t(t, height, y); + samp->nearest_texcoord_p(p, depth, z); - if (level0 != level1) { - /* get texels from second mipmap level and blend */ - float rgba2[4][4]; - x0[j] /= 2; - y0[j] /= 2; - z0[j] /= 2; - x1[j] /= 2; - y1[j] /= 2; - z1[j] /= 2; - get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0); - get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1); - get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2); - get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3); - get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0); - get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1); - get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2); - get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3); - - /* interpolate R, G, B, A */ - for (c = 0; c < 4; c++) { - rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j], - tx0[c][0], tx0[c][1], - tx0[c][2], tx0[c][3], - tx1[c][0], tx1[c][1], - tx1[c][2], tx1[c][3]); - } - - /* blend mipmap levels */ - for (c = 0; c < NUM_CHANNELS; c++) { - rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]); - } - } - } - } - break; - default: - assert(0); + for (j = 0; j < QUAD_SIZE; j++) { + get_texel(tgsi_sampler, 0, level0, x[j], y[j], z[j], rgba, j); } } static void -sp_get_samples_cube(struct tgsi_sampler *sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - unsigned faces[QUAD_SIZE], j; - float ssss[4], tttt[4]; + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + unsigned level0, j; + int width; + int x0[4], x1[4]; + float xw[4]; /* weights */ + + + level0 = samp->level; + width = texture->width[level0]; + + assert(width > 0); + + samp->linear_texcoord_s(s, width, x0, x1, xw); + + for (j = 0; j < QUAD_SIZE; j++) { - faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j); + float tx[4][4]; /* texels */ + int c; + get_texel(tgsi_sampler, 0, level0, x0[j], 0, 0, tx, 0); + get_texel(tgsi_sampler, 0, level0, x1[j], 0, 0, tx, 1); + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp(xw[j], tx[c][0], tx[c][1]); + } } - sp_get_samples_2d_common(sampler, ssss, tttt, NULL, - lodbias, rgba, faces); } - static void -sp_get_samples_rect(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; - const uint face = 0; - unsigned level0, level1, j, imgFilter; + const unsigned *faces = samp->faces; /* zero when not cube-mapping */ + unsigned level0, j; int width, height; - float levelBlend; + int x0[4], y0[4], x1[4], y1[4]; + float xw[4], yw[4]; /* weights */ - choose_mipmap_levels(tgsi_sampler, s, t, p, - lodbias, - &level0, &level1, &levelBlend, &imgFilter); - - /* texture RECTS cannot be mipmapped */ - assert(level0 == level1); + level0 = samp->level; width = texture->width[level0]; height = texture->height[level0]; assert(width > 0); - switch (imgFilter) { - case PIPE_TEX_FILTER_NEAREST: - { - int x[4], y[4]; - nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x); - nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y); - for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare(sampler, rgba, p, j); - } - } + samp->linear_texcoord_s(s, width, x0, x1, xw); + samp->linear_texcoord_s(t, height, y0, y1, yw); + + for (j = 0; j < QUAD_SIZE; j++) { + float tx[4][4]; /* texels */ + int c; + get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0); + get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1); + get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2); + get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3); + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw[j], yw[j], + tx[c][0], tx[c][1], + tx[c][2], tx[c][3]); } - break; - case PIPE_TEX_FILTER_LINEAR: - case PIPE_TEX_FILTER_ANISO: - { - int x0[4], y0[4], x1[4], y1[4]; - float xw[4], yw[4]; /* weights */ - linear_texcoord_unnorm_4(sampler->wrap_s, s, width, x0, x1, xw); - linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw); - for (j = 0; j < QUAD_SIZE; j++) { - float tx[4][4]; /* texels */ - int c; - get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0); - get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1); - get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2); - get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3); - if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - shadow_compare4(sampler, tx, p); - } - for (c = 0; c < 4; c++) { - rgba[c][j] = lerp_2d(xw[j], yw[j], - tx[c][0], tx[c][1], tx[c][2], tx[c][3]); - } + } +} + + +static void +img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + unsigned level0, j; + int width, height, depth; + int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4]; + float xw[4], yw[4], zw[4]; /* interpolation weights */ + + level0 = samp->level; + width = texture->width[level0]; + height = texture->height[level0]; + depth = texture->depth[level0]; + + assert(width > 0); + assert(height > 0); + assert(depth > 0); + + samp->linear_texcoord_s(s, width, x0, x1, xw); + samp->linear_texcoord_s(t, height, y0, y1, yw); + samp->linear_texcoord_s(p, depth, z0, z1, zw); + + for (j = 0; j < QUAD_SIZE; j++) { + float tx0[4][4], tx1[4][4]; + int c; + + get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z0[j], tx0, 0); + get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z0[j], tx0, 1); + get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z0[j], tx0, 2); + get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z0[j], tx0, 3); + get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z1[j], tx1, 0); + get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z1[j], tx1, 1); + get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z1[j], tx1, 2); + get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z1[j], tx1, 3); + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j], + tx0[c][0], tx0[c][1], + tx0[c][2], tx0[c][3], + tx1[c][0], tx1[c][1], + tx1[c][2], tx1[c][3]); + } + } +} + + + + + + + +static void +mip_filter_linear(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + int level0; + float lambda; + + lambda = samp->compute_lambda(samp, s, t, p, lodbias); + level0 = (int)lambda; + + if (lambda < 0.0) { + samp->level = 0; + samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } + else if (level0 >= texture->last_level) { + samp->level = texture->last_level; + samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } + else { + float levelBlend = lambda - level0; + float rgba0[4][4]; + float rgba1[4][4]; + int c,j; + + samp->level = level0; + samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba0 ); + + samp->level = level0+1; + samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba1 ); + + for (j = 0; j < QUAD_SIZE; j++) { + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); } } - break; - default: - assert(0); } } -/** - * Error condition handler + +static void +mip_filter_nearest(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + float lambda; + + lambda = samp->compute_lambda(samp, s, t, p, lodbias); + + if (lambda < 0.0) { + samp->level = 0; + samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } + else { + samp->level = (int)(lambda + 0.5) ; + samp->level = MIN2(samp->level, (int)texture->last_level); + samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } +} + + +static void +mip_filter_none(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + float lambda = samp->compute_lambda(samp, s, t, p, lodbias); + + if (lambda < 0.0) { + samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } + else { + samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba ); + } +} + + + +/* Specialized version of mip_filter_linear with hard-wired calls to + * 2d lambda calculation and 2d_linear_repeat_POT img filters. */ -static INLINE void -sp_get_samples_null(struct tgsi_sampler *tgsi_sampler, - const float s[QUAD_SIZE], - const float t[QUAD_SIZE], - const float p[QUAD_SIZE], - float lodbias, - float rgba[NUM_CHANNELS][QUAD_SIZE]) +static void +mip_filter_linear_2d_linear_repeat_POT( + struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) { - int i,j; + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + int level0; + float lambda; - for (i = 0; i < 4; i++) - for (j = 0; j < 4; j++) - rgba[i][j] = 1.0; + lambda = compute_lambda_2d(samp, s, t, p, lodbias); + level0 = (int)lambda; + + /* Catches both negative and large values of level0: + */ + if ((unsigned)level0 >= texture->last_level) { + if (level0 < 0) + samp->level = 0; + else + samp->level = texture->last_level; + + img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba ); + } + else { + float levelBlend = lambda - level0; + float rgba0[4][4]; + float rgba1[4][4]; + int c,j; + + samp->level = level0; + img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba0 ); + + samp->level = level0+1; + img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba1 ); + + for (j = 0; j < QUAD_SIZE; j++) { + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]); + } + } + } } -/** - * Called via tgsi_sampler::get_samples() when using a sampler for the - * first time. Determine the actual sampler function, link it in and - * call it. + + +/* Compare stage in the little sampling pipeline. */ -void -sp_get_samples(struct tgsi_sampler *tgsi_sampler, +static void +sample_compare(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { - struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler); - const struct pipe_texture *texture = samp->texture; + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_sampler_state *sampler = samp->sampler; + int j, k0, k1, k2, k3; + float val; - /* Default to the 'undefined' case: - */ - tgsi_sampler->get_samples = sp_get_samples_null; + samp->mip_filter( tgsi_sampler, s, t, p, lodbias, rgba ); - if (!texture) { - assert(0); /* is this legal?? */ - goto out; - } - if (!sampler->normalized_coords) { - assert (texture->target == PIPE_TEXTURE_2D); - tgsi_sampler->get_samples = sp_get_samples_rect; - goto out; - } + /** + * Compare texcoord 'p' (aka R) against texture value 'rgba[0]' + * When we sampled the depth texture, the depth value was put into all + * RGBA channels. We look at the red channel here. + */ - switch (texture->target) { - case PIPE_TEXTURE_1D: - tgsi_sampler->get_samples = sp_get_samples_1d; + /* compare four texcoords vs. four texture samples */ + switch (sampler->compare_func) { + case PIPE_FUNC_LESS: + k0 = p[0] < rgba[0][0]; + k1 = p[1] < rgba[0][1]; + k2 = p[2] < rgba[0][2]; + k3 = p[3] < rgba[0][3]; break; - case PIPE_TEXTURE_2D: - tgsi_sampler->get_samples = sp_get_samples_2d; + case PIPE_FUNC_LEQUAL: + k0 = p[0] <= rgba[0][0]; + k1 = p[1] <= rgba[0][1]; + k2 = p[2] <= rgba[0][2]; + k3 = p[3] <= rgba[0][3]; break; - case PIPE_TEXTURE_3D: - tgsi_sampler->get_samples = sp_get_samples_3d; + case PIPE_FUNC_GREATER: + k0 = p[0] > rgba[0][0]; + k1 = p[1] > rgba[0][1]; + k2 = p[2] > rgba[0][2]; + k3 = p[3] > rgba[0][3]; break; - case PIPE_TEXTURE_CUBE: - tgsi_sampler->get_samples = sp_get_samples_cube; + case PIPE_FUNC_GEQUAL: + k0 = p[0] >= rgba[0][0]; + k1 = p[1] >= rgba[0][1]; + k2 = p[2] >= rgba[0][2]; + k3 = p[3] >= rgba[0][3]; + break; + case PIPE_FUNC_EQUAL: + k0 = p[0] == rgba[0][0]; + k1 = p[1] == rgba[0][1]; + k2 = p[2] == rgba[0][2]; + k3 = p[3] == rgba[0][3]; + break; + case PIPE_FUNC_NOTEQUAL: + k0 = p[0] != rgba[0][0]; + k1 = p[1] != rgba[0][1]; + k2 = p[2] != rgba[0][2]; + k3 = p[3] != rgba[0][3]; + break; + case PIPE_FUNC_ALWAYS: + k0 = k1 = k2 = k3 = 1; + break; + case PIPE_FUNC_NEVER: + k0 = k1 = k2 = k3 = 0; break; default: + k0 = k1 = k2 = k3 = 0; assert(0); break; } - /* Do this elsewhere: - */ - samp->xpot = util_unsigned_logbase2( samp->texture->width[0] ); - samp->ypot = util_unsigned_logbase2( samp->texture->height[0] ); + /* convert four pass/fail values to an intensity in [0,1] */ + val = 0.25F * (k0 + k1 + k2 + k3); + + /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ + for (j = 0; j < 4; j++) { + rgba[0][j] = rgba[1][j] = rgba[2][j] = val; + rgba[3][j] = 1.0F; + } +} + +/* Calculate cube faces. + */ +static void +sample_cube(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + unsigned j; + float ssss[4], tttt[4]; + + /* + major axis + direction target sc tc ma + ---------- ------------------------------- --- --- --- + +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx + -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx + +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry + -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry + +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz + -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz + */ + for (j = 0; j < QUAD_SIZE; j++) { + float rx = s[j]; + float ry = t[j]; + float rz = p[j]; + const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz); + unsigned face; + float sc, tc, ma; + + if (arx > ary && arx > arz) { + if (rx >= 0.0F) { + face = PIPE_TEX_FACE_POS_X; + sc = -rz; + tc = -ry; + ma = arx; + } + else { + face = PIPE_TEX_FACE_NEG_X; + sc = rz; + tc = -ry; + ma = arx; + } + } + else if (ary > arx && ary > arz) { + if (ry >= 0.0F) { + face = PIPE_TEX_FACE_POS_Y; + sc = rx; + tc = rz; + ma = ary; + } + else { + face = PIPE_TEX_FACE_NEG_Y; + sc = rx; + tc = -rz; + ma = ary; + } + } + else { + if (rz > 0.0F) { + face = PIPE_TEX_FACE_POS_Z; + sc = rx; + tc = -ry; + ma = arz; + } + else { + face = PIPE_TEX_FACE_NEG_Z; + sc = -rx; + tc = -ry; + ma = arz; + } + } + + ssss[j] = ( sc / ma + 1.0F ) * 0.5F; + tttt[j] = ( tc / ma + 1.0F ) * 0.5F; + samp->faces[j] = face; + } - /* Try to hook in a faster sampler. Ultimately we'll have to - * code-generate these. Luckily most of this looks like it is - * orthogonal state within the sampler. + /* In our little pipeline, the compare stage is next. If compare + * is not active, this will point somewhere deeper into the + * pipeline, eg. to mip_filter or even img_filter. */ - if (texture->target == PIPE_TEXTURE_2D && - sampler->min_img_filter == sampler->mag_img_filter && - sampler->wrap_s == sampler->wrap_t && - sampler->compare_mode == FALSE && - sampler->normalized_coords) - { - if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { - samp->level = CLAMP((int) sampler->min_lod, - 0, (int) texture->last_level); - - if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { - switch (sampler->min_img_filter) { + samp->compare(tgsi_sampler, ssss, tttt, NULL, lodbias, rgba); +} + + + + +static wrap_nearest_func get_nearest_unorm_wrap( unsigned mode ) +{ + switch (mode) { + case PIPE_TEX_WRAP_CLAMP: + return wrap_nearest_unorm_clamp; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return wrap_nearest_unorm_clamp_to_border; + default: + assert(0); + return wrap_nearest_unorm_clamp; + } +} + + +static wrap_nearest_func get_nearest_wrap( unsigned mode ) +{ + switch (mode) { + case PIPE_TEX_WRAP_REPEAT: + return wrap_nearest_repeat; + case PIPE_TEX_WRAP_CLAMP: + return wrap_nearest_clamp; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + return wrap_nearest_clamp_to_edge; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return wrap_nearest_clamp_to_border; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + return wrap_nearest_mirror_repeat; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + return wrap_nearest_mirror_clamp; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + return wrap_nearest_mirror_clamp_to_edge; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + return wrap_nearest_mirror_clamp_to_border; + default: + assert(0); + return wrap_nearest_repeat; + } +} + +static wrap_linear_func get_linear_unorm_wrap( unsigned mode ) +{ + switch (mode) { + case PIPE_TEX_WRAP_CLAMP: + return wrap_linear_unorm_clamp; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return wrap_linear_unorm_clamp_to_border; + default: + assert(0); + return wrap_linear_unorm_clamp; + } +} + +static wrap_linear_func get_linear_wrap( unsigned mode ) +{ + switch (mode) { + case PIPE_TEX_WRAP_REPEAT: + return wrap_linear_repeat; + case PIPE_TEX_WRAP_CLAMP: + return wrap_linear_clamp; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + return wrap_linear_clamp_to_edge; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return wrap_linear_clamp_to_border; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + return wrap_linear_mirror_repeat; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + return wrap_linear_mirror_clamp; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + return wrap_linear_mirror_clamp_to_edge; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + return wrap_linear_mirror_clamp_to_border; + default: + assert(0); + return wrap_linear_repeat; + } +} + +static compute_lambda_func get_lambda_func( const union sp_sampler_key key ) +{ + if (key.bits.processor == TGSI_PROCESSOR_VERTEX) + return compute_lambda_vert; + + switch (key.bits.target) { + case PIPE_TEXTURE_1D: + return compute_lambda_1d; + case PIPE_TEXTURE_2D: + return compute_lambda_2d; + case PIPE_TEXTURE_3D: + return compute_lambda_3d; + default: + assert(0); + return compute_lambda_1d; + } +} + +static filter_func get_img_filter( const union sp_sampler_key key, + unsigned filter, + const struct pipe_sampler_state *sampler ) +{ + switch (key.bits.target) { + case PIPE_TEXTURE_1D: + if (filter == PIPE_TEX_FILTER_NEAREST) + return img_filter_1d_nearest; + else + return img_filter_1d_linear; + break; + case PIPE_TEXTURE_2D: + /* Try for fast path: + */ + if (key.bits.is_pot && + sampler->wrap_s == sampler->wrap_t && + sampler->normalized_coords) + { + switch (sampler->wrap_s) { + case PIPE_TEX_WRAP_REPEAT: + switch (filter) { case PIPE_TEX_FILTER_NEAREST: - tgsi_sampler->get_samples = sp_get_samples_2d_nearest_repeat_POT; - break; + return img_filter_2d_nearest_repeat_POT; case PIPE_TEX_FILTER_LINEAR: - tgsi_sampler->get_samples = sp_get_samples_2d_linear_repeat_POT; - break; + return img_filter_2d_linear_repeat_POT; default: break; } - } - else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) { - switch (sampler->min_img_filter) { + break; + case PIPE_TEX_WRAP_CLAMP: + switch (filter) { case PIPE_TEX_FILTER_NEAREST: - tgsi_sampler->get_samples = sp_get_samples_2d_nearest_clamp_POT; - break; + return img_filter_2d_nearest_clamp_POT; default: break; } } } - else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) { - switch (sampler->min_img_filter) { - case PIPE_TEX_FILTER_LINEAR: - tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT; - break; - default: - break; - } - } + /* Fallthrough to default versions: + */ + case PIPE_TEXTURE_CUBE: + if (filter == PIPE_TEX_FILTER_NEAREST) + return img_filter_2d_nearest; + else + return img_filter_2d_linear; + break; + case PIPE_TEXTURE_3D: + if (filter == PIPE_TEX_FILTER_NEAREST) + return img_filter_3d_nearest; + else + return img_filter_3d_linear; + break; + default: + assert(0); + return img_filter_1d_nearest; + } +} + + +void +sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp, + struct softpipe_tile_cache *tex_cache, + const struct pipe_texture *texture ) +{ + const struct pipe_sampler_state *sampler = samp->sampler; + + samp->texture = texture; + samp->cache = tex_cache; + samp->xpot = util_unsigned_logbase2( texture->width[0] ); + samp->ypot = util_unsigned_logbase2( texture->height[0] ); + samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level); +} + +/* Create a sampler varient for a given set of non-orthogonal state. Currently the + */ +struct sp_sampler_varient * +sp_create_sampler_varient( const struct pipe_sampler_state *sampler, + const union sp_sampler_key key ) +{ + struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient); + if (!samp) + return NULL; + + samp->sampler = sampler; + samp->key = key; + + /* Note that (for instance) linear_texcoord_s and + * nearest_texcoord_s may be active at the same time, if the + * sampler min_img_filter differs from its mag_img_filter. + */ + if (sampler->normalized_coords) { + samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s ); + samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t ); + samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r ); + + samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s ); + samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t ); + samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r ); + } + else { + samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s ); + samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t ); + samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r ); + + samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s ); + samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t ); + samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r ); + } + + samp->compute_lambda = get_lambda_func( key ); + + samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler); + samp->mag_img_filter = get_img_filter(key, sampler->min_img_filter, sampler); + + switch (sampler->min_mip_filter) { + case PIPE_TEX_MIPFILTER_NONE: + if (sampler->min_img_filter == sampler->mag_img_filter) + samp->mip_filter = samp->min_img_filter; + else + samp->mip_filter = mip_filter_none; + break; + + case PIPE_TEX_MIPFILTER_NEAREST: + samp->mip_filter = mip_filter_nearest; + break; + + case PIPE_TEX_MIPFILTER_LINEAR: + if (key.bits.is_pot && + sampler->min_img_filter == sampler->mag_img_filter && + sampler->wrap_s == sampler->wrap_t && + sampler->normalized_coords && + sampler->wrap_s == sampler->wrap_t && + sampler->wrap_s == PIPE_TEX_WRAP_REPEAT && + sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR) + { + samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT; + } + else + { + samp->mip_filter = mip_filter_linear; } + break; + } + + if (sampler->compare_mode != FALSE) { + samp->compare = sample_compare; } - else if (0) { - _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n", - texture->target, PIPE_TEXTURE_2D, - sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE, - sampler->min_img_filter, sampler->mag_img_filter, - sampler->wrap_s, sampler->wrap_t, - sampler->compare_mode, FALSE, - sampler->normalized_coords, TRUE); + else { + /* Skip compare operation by promoting the mip_filter function + * pointer: + */ + samp->compare = samp->mip_filter; + } + + if (key.bits.target == PIPE_TEXTURE_CUBE) { + samp->base.get_samples = sample_cube; + } + else { + samp->faces[0] = 0; + samp->faces[1] = 0; + samp->faces[2] = 0; + samp->faces[3] = 0; + + /* Skip cube face determination by promoting the compare + * function pointer: + */ + samp->base.get_samples = samp->compare; } -out: - tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba ); + return samp; } + + + + + diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index c73ae44131..26f80eb88a 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -31,14 +31,61 @@ #include "tgsi/tgsi_exec.h" +struct sp_sampler_varient; + +typedef void (*wrap_nearest_func)(const float s[4], + unsigned size, + int icoord[4]); + +typedef void (*wrap_linear_func)(const float s[4], + unsigned size, + int icoord0[4], + int icoord1[4], + float w[4]); + +typedef float (*compute_lambda_func)(const struct sp_sampler_varient *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias); + +typedef void (*filter_func)(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]); + + +union sp_sampler_key { + struct { + unsigned target:3; + unsigned is_pot:1; + unsigned processor:2; + unsigned pad:26; + } bits; + unsigned value; +}; /** * Subclass of tgsi_sampler */ -struct sp_shader_sampler +struct sp_sampler_varient { struct tgsi_sampler base; /**< base class */ + union sp_sampler_key key; + + /* The owner of this struct: + */ + const struct pipe_sampler_state *sampler; + + + /* Currently bound texture: + */ + const struct pipe_texture *texture; + struct softpipe_tile_cache *cache; + unsigned processor; /* For sp_get_samples_2d_linear_POT: @@ -47,22 +94,51 @@ struct sp_shader_sampler unsigned ypot; unsigned level; - const struct pipe_texture *texture; - const struct pipe_sampler_state *sampler; + unsigned faces[4]; + + wrap_nearest_func nearest_texcoord_s; + wrap_nearest_func nearest_texcoord_t; + wrap_nearest_func nearest_texcoord_p; - struct softpipe_tile_cache *cache; + wrap_linear_func linear_texcoord_s; + wrap_linear_func linear_texcoord_t; + wrap_linear_func linear_texcoord_p; + + filter_func min_img_filter; + filter_func mag_img_filter; + + compute_lambda_func compute_lambda; + + filter_func mip_filter; + filter_func compare; + + /* Linked list: + */ + struct sp_sampler_varient *next; }; +struct sp_sampler; + +/* Create a sampler varient for a given set of non-orthogonal state. Currently the + */ +struct sp_sampler_varient * +sp_create_sampler_varient( const struct pipe_sampler_state *sampler, + const union sp_sampler_key key ); +void sp_sampler_varient_bind_texture( struct sp_sampler_varient *varient, + struct softpipe_tile_cache *tex_cache, + const struct pipe_texture *tex ); -static INLINE struct sp_shader_sampler * -sp_shader_sampler(const struct tgsi_sampler *sampler) -{ - return (struct sp_shader_sampler *) sampler; -} +void sp_sampler_varient_destroy( struct sp_sampler_varient * ); +static INLINE struct sp_sampler_varient * +sp_sampler_varient(const struct tgsi_sampler *sampler) +{ + return (struct sp_sampler_varient *) sampler; +} + extern void sp_get_samples(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c index 0c84375bf1..a3a54dada4 100644 --- a/src/gallium/drivers/softpipe/sp_texture.c +++ b/src/gallium/drivers/softpipe/sp_texture.c @@ -120,16 +120,20 @@ softpipe_displaytarget_layout(struct pipe_screen *screen, static struct pipe_texture * softpipe_texture_create(struct pipe_screen *screen, - const struct pipe_texture *templat) + const struct pipe_texture *template) { struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture); if (!spt) return NULL; - spt->base = *templat; + spt->base = *template; pipe_reference_init(&spt->base.reference, 1); spt->base.screen = screen; + spt->pot = (util_is_power_of_two(template->width[0]) && + util_is_power_of_two(template->height[0]) && + util_is_power_of_two(template->depth[0])); + if (spt->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) { if (!softpipe_displaytarget_layout(screen, spt)) goto fail; diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h index 42df722a2d..4dd0c1239e 100644 --- a/src/gallium/drivers/softpipe/sp_texture.h +++ b/src/gallium/drivers/softpipe/sp_texture.h @@ -48,6 +48,10 @@ struct softpipe_texture */ struct pipe_buffer *buffer; + /* True if texture images are power-of-two in all dimensions: + */ + boolean pot; + unsigned timestamp; }; -- cgit v1.2.3 From 4e5c385d2183e7006c9d7ac0823919156bd4b8e6 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 11:40:33 -0600 Subject: softpipe: fix s/t/p typos --- src/gallium/drivers/softpipe/sp_tex_sample.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 7bc689a298..a626731105 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -990,7 +990,7 @@ img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, assert(width > 0); samp->linear_texcoord_s(s, width, x0, x1, xw); - samp->linear_texcoord_s(t, height, y0, y1, yw); + samp->linear_texcoord_t(t, height, y0, y1, yw); for (j = 0; j < QUAD_SIZE; j++) { float tx[4][4]; /* texels */ @@ -1035,8 +1035,8 @@ img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler, assert(depth > 0); samp->linear_texcoord_s(s, width, x0, x1, xw); - samp->linear_texcoord_s(t, height, y0, y1, yw); - samp->linear_texcoord_s(p, depth, z0, z1, zw); + samp->linear_texcoord_t(t, height, y0, y1, yw); + samp->linear_texcoord_p(p, depth, z0, z1, zw); for (j = 0; j < QUAD_SIZE; j++) { float tx0[4][4], tx1[4][4]; -- cgit v1.2.3 From 41483627f0fd3dc9df2cc55dfd5f3e5987fcfd22 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 11:41:29 -0600 Subject: softpipe: fix min/mag filter typo --- src/gallium/drivers/softpipe/sp_tex_sample.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index a626731105..9502b60479 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1597,7 +1597,7 @@ sp_create_sampler_varient( const struct pipe_sampler_state *sampler, samp->compute_lambda = get_lambda_func( key ); samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler); - samp->mag_img_filter = get_img_filter(key, sampler->min_img_filter, sampler); + samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler); switch (sampler->min_mip_filter) { case PIPE_TEX_MIPFILTER_NONE: -- cgit v1.2.3 From cf102b031e7ef33c8e3ffce2f9dcd064f44e8190 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 11:43:48 -0600 Subject: softpipe: remove redundant comparison, make test easier to understand --- src/gallium/drivers/softpipe/sp_tex_sample.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 9502b60479..51118ae38b 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1614,10 +1614,9 @@ sp_create_sampler_varient( const struct pipe_sampler_state *sampler, case PIPE_TEX_MIPFILTER_LINEAR: if (key.bits.is_pot && sampler->min_img_filter == sampler->mag_img_filter && - sampler->wrap_s == sampler->wrap_t && sampler->normalized_coords && - sampler->wrap_s == sampler->wrap_t && sampler->wrap_s == PIPE_TEX_WRAP_REPEAT && + sampler->wrap_t == PIPE_TEX_WRAP_REPEAT && sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR) { samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT; -- cgit v1.2.3 From ecfa8be150ed276af816467b467e76e026f5b541 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 21 Aug 2009 18:44:27 +0100 Subject: softpipe: add missing sp_sampler_varient_destroy --- src/gallium/drivers/softpipe/sp_tex_sample.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 51118ae38b..a2e2a221e4 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1558,6 +1558,14 @@ sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp, samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level); } + +void +sp_sampler_varient_destroy( struct sp_sampler_varient *samp ) +{ + FREE(samp); +} + + /* Create a sampler varient for a given set of non-orthogonal state. Currently the */ struct sp_sampler_varient * -- cgit v1.2.3 From 87ec83afd58536c31bf02c307f1d5488abc84861 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 11:47:27 -0600 Subject: softpipe: add missing PIPE_TEXTURE_CUBE case in get_lambda_func() Fixes progs/demos/cubemap --- src/gallium/drivers/softpipe/sp_tex_sample.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index a2e2a221e4..f371003708 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1476,6 +1476,7 @@ static compute_lambda_func get_lambda_func( const union sp_sampler_key key ) case PIPE_TEXTURE_1D: return compute_lambda_1d; case PIPE_TEXTURE_2D: + case PIPE_TEXTURE_CUBE: return compute_lambda_2d; case PIPE_TEXTURE_3D: return compute_lambda_3d; -- cgit v1.2.3 From a29447c33d44b3427e0c40a761067c0cc6e71c39 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 12:11:44 -0600 Subject: softpipe: per-unit sampler varients Can't share sampler varients across multiple tex units because the texture pointer is in the sampler varient. That prevents different textures per unit. Fixes progs/demos/multiarb, progs/glsl/samplers, etc. --- src/gallium/drivers/softpipe/sp_state_sampler.c | 24 ++++++++++++++++++++---- src/gallium/drivers/softpipe/sp_tex_sample.c | 3 +++ src/gallium/drivers/softpipe/sp_tex_sample.h | 3 ++- 3 files changed, 25 insertions(+), 5 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c index 714e638048..53210812f4 100644 --- a/src/gallium/drivers/softpipe/sp_state_sampler.c +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -123,9 +123,19 @@ softpipe_set_sampler_textures(struct pipe_context *pipe, } - +/** + * Find/create an sp_sampler_varient object for sampling the given texture, + * sampler and tex unit. + * + * Note that the tex unit is significant. We can't re-use a sampler + * varient for multiple texture units because the sampler varient contains + * the texture object pointer. If the texture object pointer were stored + * somewhere outside the sampler varient, we could re-use samplers for + * multiple texture units. + */ static struct sp_sampler_varient * -get_sampler_varient( struct sp_sampler *sampler, +get_sampler_varient( unsigned unit, + struct sp_sampler *sampler, struct pipe_texture *texture, unsigned processor ) { @@ -133,9 +143,13 @@ get_sampler_varient( struct sp_sampler *sampler, struct sp_sampler_varient *v = NULL; union sp_sampler_key key; + /* if this fails, widen the key.unit field and update this assertion */ + assert(PIPE_MAX_SAMPLERS <= 16); + key.bits.target = sp_texture->base.target; key.bits.is_pot = sp_texture->pot; key.bits.processor = processor; + key.bits.unit = unit; key.bits.pad = 0; if (sampler->current && @@ -174,7 +188,8 @@ softpipe_reset_sampler_varients(struct softpipe_context *softpipe) for (i = 0; i <= softpipe->vs->max_sampler; i++) { if (softpipe->sampler[i]) { softpipe->tgsi.vert_samplers_list[i] = - get_sampler_varient( sp_sampler(softpipe->sampler[i]), + get_sampler_varient( i, + sp_sampler(softpipe->sampler[i]), softpipe->texture[i], TGSI_PROCESSOR_VERTEX ); @@ -187,7 +202,8 @@ softpipe_reset_sampler_varients(struct softpipe_context *softpipe) for (i = 0; i <= softpipe->fs->info.file_max[TGSI_FILE_SAMPLER]; i++) { if (softpipe->sampler[i]) { softpipe->tgsi.frag_samplers_list[i] = - get_sampler_varient( sp_sampler(softpipe->sampler[i]), + get_sampler_varient( i, + sp_sampler(softpipe->sampler[i]), softpipe->texture[i], TGSI_PROCESSOR_FRAGMENT ); diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index f371003708..8f3dc12d0f 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1545,6 +1545,9 @@ static filter_func get_img_filter( const union sp_sampler_key key, } +/** + * Bind the given texture object and texture cache to the sampler varient. + */ void sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp, struct softpipe_tile_cache *tex_cache, diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h index 26f80eb88a..f6cd57ec0a 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.h +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -62,7 +62,8 @@ union sp_sampler_key { unsigned target:3; unsigned is_pot:1; unsigned processor:2; - unsigned pad:26; + unsigned unit:4; + unsigned pad:22; } bits; unsigned value; }; -- cgit v1.2.3 From 46fbc872881081ffcf0b526f8c4a909fd915ad78 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 13:45:16 -0600 Subject: softpipe: remove unneeded const qualifier --- src/gallium/drivers/softpipe/sp_state_fs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c index 3a45321923..a055d6295f 100644 --- a/src/gallium/drivers/softpipe/sp_state_fs.c +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -128,7 +128,7 @@ softpipe_bind_vs_state(struct pipe_context *pipe, void *vs) { struct softpipe_context *softpipe = softpipe_context(pipe); - softpipe->vs = (const struct sp_vertex_shader *)vs; + softpipe->vs = (struct sp_vertex_shader *)vs; draw_bind_vertex_shader(softpipe->draw, (softpipe->vs ? softpipe->vs->draw_data : NULL)); -- cgit v1.2.3 From 4256c5829f8c23f8bd5c7c29491210f0f7813bf9 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 13:47:50 -0600 Subject: softpipe: remove unused #includes, white-space clean-up --- src/gallium/drivers/softpipe/sp_state_fs.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c index a055d6295f..256faa94b8 100644 --- a/src/gallium/drivers/softpipe/sp_state_fs.c +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -31,8 +31,6 @@ #include "pipe/p_defines.h" #include "util/u_memory.h" -#include "pipe/internal/p_winsys_screen.h" -#include "pipe/p_shader_tokens.h" #include "draw/draw_context.h" #include "draw/draw_vs.h" #include "tgsi/tgsi_dump.h" @@ -128,7 +126,7 @@ softpipe_bind_vs_state(struct pipe_context *pipe, void *vs) { struct softpipe_context *softpipe = softpipe_context(pipe); - softpipe->vs = (struct sp_vertex_shader *)vs; + softpipe->vs = (struct sp_vertex_shader *) vs; draw_bind_vertex_shader(softpipe->draw, (softpipe->vs ? softpipe->vs->draw_data : NULL)); @@ -142,8 +140,7 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs) { struct softpipe_context *softpipe = softpipe_context(pipe); - struct sp_vertex_shader *state = - (struct sp_vertex_shader *)vs; + struct sp_vertex_shader *state = (struct sp_vertex_shader *) vs; draw_delete_vertex_shader(softpipe->draw, state->draw_data); FREE( state ); -- cgit v1.2.3 From 3adc8c3779895c483ba8a1004939e7dd7d76fa9a Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 14:01:58 -0600 Subject: softpipe: minor code refactoring to remove softpipe/tile cache dependencies The tile cache code now has no hard dependencies on softpipe. --- src/gallium/drivers/softpipe/sp_state_derived.c | 13 ++++++++++++- src/gallium/drivers/softpipe/sp_tile_cache.c | 26 ++++++++++++------------- src/gallium/drivers/softpipe/sp_tile_cache.h | 1 - 3 files changed, 24 insertions(+), 16 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 5310928332..202a2bc94c 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -34,6 +34,8 @@ #include "sp_context.h" #include "sp_screen.h" #include "sp_state.h" +#include "sp_texture.h" +#include "sp_tile_cache.h" /** @@ -201,10 +203,19 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) softpipe_reset_sampler_varients( softpipe ); for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { - sp_tile_cache_validate_texture( softpipe->tex_cache[i] ); + struct softpipe_tile_cache *tc = softpipe->tex_cache[i]; + if (tc->texture) { + struct softpipe_texture *spt = softpipe_texture(tc->texture); + if (spt->timestamp != tc->timestamp) { + sp_tile_cache_validate_texture( tc ); + _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); + tc->timestamp = spt->timestamp; + } + } } } + /* Hopefully this will remain quite simple, otherwise need to pull in * something like the state tracker mechanism. */ diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c index 77d02fa3e7..e075ab6290 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.c +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -35,9 +35,6 @@ #include "pipe/p_inlines.h" #include "util/u_memory.h" #include "util/u_tile.h" -#include "sp_context.h" -#include "sp_surface.h" -#include "sp_texture.h" #include "sp_tile_cache.h" @@ -200,24 +197,25 @@ sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc) } } + +/** + * Invalidate all cached tiles for the cached texture. + * Should be called when the texture is modified. + */ void sp_tile_cache_validate_texture(struct softpipe_tile_cache *tc) { - if (tc->texture) { - struct softpipe_texture *spt = softpipe_texture(tc->texture); - if (spt->timestamp != tc->timestamp) { - /* texture was modified, invalidate all cached tiles */ - uint i; - _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); - for (i = 0; i < NUM_ENTRIES; i++) { - tc->entries[i].addr.bits.invalid = 1; - } + uint i; - tc->timestamp = spt->timestamp; - } + assert(tc); + assert(tc->texture); + + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].addr.bits.invalid = 1; } } + /** * Specify the texture to cache. */ diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h index ac2aae5875..1596cd0ae7 100644 --- a/src/gallium/drivers/softpipe/sp_tile_cache.h +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -34,7 +34,6 @@ #include "pipe/p_compiler.h" -struct softpipe_context; struct softpipe_tile_cache; -- cgit v1.2.3 From d204659c8c725c02212ad4a49275c7447f2d02a6 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 14:04:47 -0600 Subject: softpipe: remove tex sample dependencies on softpipe The texture sampling code doesn't really have any dependencies on the rest of softpipe, just the tile cache. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 8f3dc12d0f..a9efb82491 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -34,17 +34,14 @@ * Keith Whitwell */ -#include "sp_context.h" -#include "sp_quad.h" -#include "sp_surface.h" -#include "sp_texture.h" -#include "sp_tex_sample.h" -#include "sp_tile_cache.h" #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_shader_tokens.h" #include "util/u_math.h" #include "util/u_memory.h" +#include "sp_quad.h" /* only for #define QUAD_* tokens */ +#include "sp_tex_sample.h" +#include "sp_tile_cache.h" -- cgit v1.2.3 From 0f24886f922df3e00094a53b5b37b1588ea84bc0 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Fri, 21 Aug 2009 14:07:37 -0600 Subject: softpipe: remove duplicate #include, move another --- src/gallium/drivers/softpipe/sp_state_sampler.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c index 53210812f4..afc6e1d2eb 100644 --- a/src/gallium/drivers/softpipe/sp_state_sampler.c +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -31,15 +31,14 @@ #include "util/u_memory.h" +#include "draw/draw_context.h" #include "draw/draw_context.h" -#include "sp_context.h" #include "sp_context.h" #include "sp_state.h" #include "sp_texture.h" #include "sp_tile_cache.h" #include "sp_tex_sample.h" -#include "draw/draw_context.h" struct sp_sampler { -- cgit v1.2.3 From 47800c572f199e7857e02e0f999b410c727a275d Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 11:13:20 +0100 Subject: softpipe: add missing header --- src/gallium/drivers/softpipe/sp_context.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 48ec540ebf..3c465c95a5 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -42,6 +42,7 @@ #include "sp_state.h" #include "sp_surface.h" #include "sp_tile_cache.h" +#include "sp_tex_tile_cache.h" #include "sp_texture.h" #include "sp_winsys.h" #include "sp_query.h" -- cgit v1.2.3 From 4fe0fc3eba1f79beda890a5016359d549bab6ad4 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 11:22:41 +0100 Subject: softpipe: remove old prim_setup draw stage Everything now goes through the draw_vbuf handler, the same as regular drivers. --- src/gallium/drivers/softpipe/Makefile | 1 - src/gallium/drivers/softpipe/SConscript | 1 - src/gallium/drivers/softpipe/sp_context.c | 26 ++-- src/gallium/drivers/softpipe/sp_context.h | 5 +- src/gallium/drivers/softpipe/sp_prim_setup.c | 190 ------------------------ src/gallium/drivers/softpipe/sp_prim_setup.h | 85 ----------- src/gallium/drivers/softpipe/sp_prim_vbuf.c | 107 ++++--------- src/gallium/drivers/softpipe/sp_prim_vbuf.h | 4 +- src/gallium/drivers/softpipe/sp_setup.c | 1 - src/gallium/drivers/softpipe/sp_state_derived.c | 25 ++-- 10 files changed, 59 insertions(+), 386 deletions(-) delete mode 100644 src/gallium/drivers/softpipe/sp_prim_setup.c delete mode 100644 src/gallium/drivers/softpipe/sp_prim_setup.h diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile index 3da9be6957..6ab3753762 100644 --- a/src/gallium/drivers/softpipe/Makefile +++ b/src/gallium/drivers/softpipe/Makefile @@ -11,7 +11,6 @@ C_SOURCES = \ sp_query.c \ sp_context.c \ sp_draw_arrays.c \ - sp_prim_setup.c \ sp_prim_vbuf.c \ sp_quad_pipe.c \ sp_quad_stipple.c \ diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript index 30c099813e..153fe44546 100644 --- a/src/gallium/drivers/softpipe/SConscript +++ b/src/gallium/drivers/softpipe/SConscript @@ -12,7 +12,6 @@ softpipe = env.ConvenienceLibrary( 'sp_context.c', 'sp_draw_arrays.c', 'sp_flush.c', - 'sp_prim_setup.c', 'sp_prim_vbuf.c', 'sp_setup.c', 'sp_quad_alpha_test.c', diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 3c465c95a5..6b75ee6002 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -31,13 +31,13 @@ */ #include "draw/draw_context.h" +#include "draw/draw_vbuf.h" #include "pipe/p_defines.h" #include "util/u_math.h" #include "util/u_memory.h" #include "sp_clear.h" #include "sp_context.h" #include "sp_flush.h" -#include "sp_prim_setup.h" #include "sp_prim_vbuf.h" #include "sp_state.h" #include "sp_surface.h" @@ -242,21 +242,21 @@ softpipe_create( struct pipe_screen *screen ) (struct tgsi_sampler **) softpipe->tgsi.vert_samplers_list); - softpipe->setup = sp_draw_render_stage(softpipe); - if (!softpipe->setup) - goto fail; - if (debug_get_bool_option( "SP_NO_RAST", FALSE )) softpipe->no_rast = TRUE; - if (debug_get_bool_option( "SP_NO_VBUF", FALSE )) { - /* Deprecated path -- vbuf is the intended interface to the draw module: - */ - draw_set_rasterize_stage(softpipe->draw, softpipe->setup); - } - else { - sp_init_vbuf(softpipe); - } + softpipe->vbuf_backend = sp_create_vbuf_backend(softpipe); + if (!softpipe->vbuf_backend) + goto fail; + + softpipe->vbuf = draw_vbuf_stage(softpipe->draw, softpipe->vbuf_backend); + if (!softpipe->vbuf) + goto fail; + + draw_set_rasterize_stage(softpipe->draw, softpipe->vbuf); + draw_set_render(softpipe->draw, softpipe->vbuf_backend); + + /* plug in AA line/point stages */ draw_install_aaline_stage(softpipe->draw, &softpipe->pipe); diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index df45d2249f..43a195c8ef 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -129,9 +129,10 @@ struct softpipe_context { /** The primitive drawing context */ struct draw_context *draw; - struct draw_stage *setup; + + /** Draw module backend */ + struct vbuf_render *vbuf_backend; struct draw_stage *vbuf; - struct softpipe_vbuf_render *vbuf_render; boolean dirty_render_cache; diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c deleted file mode 100644 index 038ff04d4f..0000000000 --- a/src/gallium/drivers/softpipe/sp_prim_setup.c +++ /dev/null @@ -1,190 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - -/** - * \brief A draw stage that drives our triangle setup routines from - * within the draw pipeline. One of two ways to drive setup, the - * other being in sp_prim_vbuf.c. - * - * \author Keith Whitwell - * \author Brian Paul - */ - - -#include "sp_context.h" -#include "sp_setup.h" -#include "sp_state.h" -#include "sp_prim_setup.h" -#include "draw/draw_pipe.h" -#include "draw/draw_vertex.h" -#include "util/u_memory.h" - -/** - * Triangle setup info (derived from draw_stage). - * Also used for line drawing (taking some liberties). - */ -struct setup_stage { - struct draw_stage stage; /**< This must be first (base class) */ - - struct setup_context *setup; -}; - - - -/** - * Basically a cast wrapper. - */ -static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) -{ - return (struct setup_stage *)stage; -} - - -typedef const float (*cptrf4)[4]; - -static void -do_tri(struct draw_stage *stage, struct prim_header *prim) -{ - struct setup_stage *setup = setup_stage( stage ); - - setup_tri( setup->setup, - (cptrf4)prim->v[0]->data, - (cptrf4)prim->v[1]->data, - (cptrf4)prim->v[2]->data ); -} - -static void -do_line(struct draw_stage *stage, struct prim_header *prim) -{ - struct setup_stage *setup = setup_stage( stage ); - - setup_line( setup->setup, - (cptrf4)prim->v[0]->data, - (cptrf4)prim->v[1]->data ); -} - -static void -do_point(struct draw_stage *stage, struct prim_header *prim) -{ - struct setup_stage *setup = setup_stage( stage ); - - setup_point( setup->setup, - (cptrf4)prim->v[0]->data ); -} - - - - -static void setup_begin( struct draw_stage *stage ) -{ - struct setup_stage *setup = setup_stage(stage); - - setup_prepare( setup->setup ); - - stage->point = do_point; - stage->line = do_line; - stage->tri = do_tri; -} - - -static void setup_first_point( struct draw_stage *stage, - struct prim_header *header ) -{ - setup_begin(stage); - stage->point( stage, header ); -} - -static void setup_first_line( struct draw_stage *stage, - struct prim_header *header ) -{ - setup_begin(stage); - stage->line( stage, header ); -} - - -static void setup_first_tri( struct draw_stage *stage, - struct prim_header *header ) -{ - setup_begin(stage); - stage->tri( stage, header ); -} - - - -static void setup_flush( struct draw_stage *stage, - unsigned flags ) -{ - stage->point = setup_first_point; - stage->line = setup_first_line; - stage->tri = setup_first_tri; -} - - -static void reset_stipple_counter( struct draw_stage *stage ) -{ -} - - -static void render_destroy( struct draw_stage *stage ) -{ - struct setup_stage *ssetup = setup_stage(stage); - setup_destroy_context(ssetup->setup); - FREE( stage ); -} - - -/** - * Create a new primitive setup/render stage. - */ -struct draw_stage *sp_draw_render_stage( struct softpipe_context *softpipe ) -{ - struct setup_stage *sstage = CALLOC_STRUCT(setup_stage); - - sstage->setup = setup_create_context(softpipe); - sstage->stage.draw = softpipe->draw; - sstage->stage.point = setup_first_point; - sstage->stage.line = setup_first_line; - sstage->stage.tri = setup_first_tri; - sstage->stage.flush = setup_flush; - sstage->stage.reset_stipple_counter = reset_stipple_counter; - sstage->stage.destroy = render_destroy; - - return (struct draw_stage *)sstage; -} - -struct setup_context * -sp_draw_setup_context( struct draw_stage *stage ) -{ - struct setup_stage *ssetup = setup_stage(stage); - return ssetup->setup; -} - -void -sp_draw_flush( struct draw_stage *stage ) -{ - stage->flush( stage, 0 ); -} diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.h b/src/gallium/drivers/softpipe/sp_prim_setup.h deleted file mode 100644 index 49bdd98ed8..0000000000 --- a/src/gallium/drivers/softpipe/sp_prim_setup.h +++ /dev/null @@ -1,85 +0,0 @@ -/************************************************************************** - * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS - * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR - * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, - * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE - * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ - - -#ifndef SP_PRIM_SETUP_H -#define SP_PRIM_SETUP_H - - -/** - * vbuf is a special stage to gather the stream of triangles, lines, points - * together and reconstruct vertex buffers for hardware upload. - * - * First attempt, work in progress. - * - * TODO: - * - separate out vertex buffer building and primitive emit, ie >1 draw per vb. - * - tell vbuf stage how to build hw vertices directly - * - pass vbuf stage a buffer pointer for direct emit to agp/vram. - * - * - * - * Vertices are just an array of floats, with all the attributes - * packed. We currently assume a layout like: - * - * attr[0][0..3] - window position - * attr[1..n][0..3] - remaining attributes. - * - * Attributes are assumed to be 4 floats wide but are packed so that - * all the enabled attributes run contiguously. - */ - - -struct draw_stage; -struct softpipe_context; - - -typedef void (*vbuf_draw_func)( struct pipe_context *pipe, - unsigned prim, - const ushort *elements, - unsigned nr_elements, - const void *vertex_buffer, - unsigned nr_vertices ); - - -extern struct draw_stage * -sp_draw_render_stage( struct softpipe_context *softpipe ); - -extern struct setup_context * -sp_draw_setup_context( struct draw_stage * ); - -extern void -sp_draw_flush( struct draw_stage * ); - - -extern struct draw_stage * -sp_draw_vbuf_stage( struct draw_context *draw_context, - struct pipe_context *pipe, - vbuf_draw_func draw ); - - -#endif /* SP_PRIM_SETUP_H */ diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c index 76524a8d41..e603c20fc4 100644 --- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c @@ -37,10 +37,9 @@ #include "sp_context.h" +#include "sp_setup.h" #include "sp_state.h" #include "sp_prim_vbuf.h" -#include "sp_prim_setup.h" -#include "sp_setup.h" #include "draw/draw_context.h" #include "draw/draw_vbuf.h" #include "util/u_memory.h" @@ -59,6 +58,8 @@ struct softpipe_vbuf_render { struct vbuf_render base; struct softpipe_context *softpipe; + struct setup_context *setup; + uint prim; uint vertex_size; uint nr_vertices; @@ -75,6 +76,11 @@ softpipe_vbuf_render(struct vbuf_render *vbr) } + + + + + static const struct vertex_info * sp_vbuf_get_vertex_info(struct vbuf_render *vbr) { @@ -105,36 +111,6 @@ sp_vbuf_allocate_vertices(struct vbuf_render *vbr, static void sp_vbuf_release_vertices(struct vbuf_render *vbr) { -#if 0 - { - struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); - const struct vertex_info *info = - softpipe_get_vbuf_vertex_info(cvbr->softpipe); - const float *vtx = (const float *) cvbr->vertex_buffer; - uint i, j; - debug_printf("%s (vtx_size = %u, vtx_used = %u)\n", - __FUNCTION__, cvbr->vertex_size, cvbr->nr_vertices); - for (i = 0; i < cvbr->nr_vertices; i++) { - for (j = 0; j < info->num_attribs; j++) { - uint k; - switch (info->attrib[j].emit) { - case EMIT_4F: k = 4; break; - case EMIT_3F: k = 3; break; - case EMIT_2F: k = 2; break; - case EMIT_1F: k = 1; break; - default: assert(0); - } - debug_printf("Vert %u attr %u: ", i, j); - while (k-- > 0) { - debug_printf("%g ", vtx[0]); - vtx++; - } - debug_printf("\n"); - } - } - } -#endif - /* keep the old allocation for next time */ } @@ -160,11 +136,7 @@ static boolean sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim) { struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); - - /* XXX: break this dependency - make setup_context live under - * softpipe, rename the old "setup" draw stage to something else. - */ - struct setup_context *setup_ctx = sp_draw_setup_context(cvbr->softpipe->setup); + struct setup_context *setup_ctx = cvbr->setup; setup_prepare( setup_ctx ); @@ -193,14 +165,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) struct softpipe_context *softpipe = cvbr->softpipe; const unsigned stride = softpipe->vertex_info_vbuf.size * sizeof(float); const void *vertex_buffer = cvbr->vertex_buffer; + struct setup_context *setup_ctx = cvbr->setup; unsigned i; - /* XXX: break this dependency - make setup_context live under - * softpipe, rename the old "setup" draw stage to something else. - */ - struct draw_stage *setup = softpipe->setup; - struct setup_context *setup_ctx = sp_draw_setup_context(setup); - switch (cvbr->prim) { case PIPE_PRIM_POINTS: for (i = 0; i < nr; i++) { @@ -367,11 +334,6 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr) default: assert(0); } - - /* XXX: why are we calling this??? If we had to call something, it - * would be a function in sp_setup.c: - */ - sp_draw_flush( setup ); } @@ -384,17 +346,12 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr) { struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); struct softpipe_context *softpipe = cvbr->softpipe; + struct setup_context *setup_ctx = cvbr->setup; const unsigned stride = softpipe->vertex_info_vbuf.size * sizeof(float); const void *vertex_buffer = (void *) get_vert(cvbr->vertex_buffer, start, stride); unsigned i; - /* XXX: break this dependency - make setup_context live under - * softpipe, rename the old "setup" draw stage to something else. - */ - struct draw_stage *setup = softpipe->setup; - struct setup_context *setup_ctx = sp_draw_setup_context(setup); - switch (cvbr->prim) { case PIPE_PRIM_POINTS: for (i = 0; i < nr; i++) { @@ -568,40 +525,38 @@ static void sp_vbuf_destroy(struct vbuf_render *vbr) { struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); - cvbr->softpipe->vbuf_render = NULL; + setup_destroy_context(cvbr->setup); FREE(cvbr); } /** - * Initialize the post-transform vertex buffer information for the given - * context. + * Create the post-transform vertex handler for the given context. */ -void -sp_init_vbuf(struct softpipe_context *sp) +struct vbuf_render * +sp_create_vbuf_backend(struct softpipe_context *sp) { - assert(sp->draw); + struct softpipe_vbuf_render *cvbr = CALLOC_STRUCT(softpipe_vbuf_render); - sp->vbuf_render = CALLOC_STRUCT(softpipe_vbuf_render); + assert(sp->draw); - sp->vbuf_render->base.max_indices = SP_MAX_VBUF_INDEXES; - sp->vbuf_render->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE; - sp->vbuf_render->base.get_vertex_info = sp_vbuf_get_vertex_info; - sp->vbuf_render->base.allocate_vertices = sp_vbuf_allocate_vertices; - sp->vbuf_render->base.map_vertices = sp_vbuf_map_vertices; - sp->vbuf_render->base.unmap_vertices = sp_vbuf_unmap_vertices; - sp->vbuf_render->base.set_primitive = sp_vbuf_set_primitive; - sp->vbuf_render->base.draw = sp_vbuf_draw; - sp->vbuf_render->base.draw_arrays = sp_vbuf_draw_arrays; - sp->vbuf_render->base.release_vertices = sp_vbuf_release_vertices; - sp->vbuf_render->base.destroy = sp_vbuf_destroy; + cvbr->base.max_indices = SP_MAX_VBUF_INDEXES; + cvbr->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE; - sp->vbuf_render->softpipe = sp; + cvbr->base.get_vertex_info = sp_vbuf_get_vertex_info; + cvbr->base.allocate_vertices = sp_vbuf_allocate_vertices; + cvbr->base.map_vertices = sp_vbuf_map_vertices; + cvbr->base.unmap_vertices = sp_vbuf_unmap_vertices; + cvbr->base.set_primitive = sp_vbuf_set_primitive; + cvbr->base.draw = sp_vbuf_draw; + cvbr->base.draw_arrays = sp_vbuf_draw_arrays; + cvbr->base.release_vertices = sp_vbuf_release_vertices; + cvbr->base.destroy = sp_vbuf_destroy; - sp->vbuf = draw_vbuf_stage(sp->draw, &sp->vbuf_render->base); + cvbr->softpipe = sp; - draw_set_rasterize_stage(sp->draw, sp->vbuf); + cvbr->setup = setup_create_context(cvbr->softpipe); - draw_set_render(sp->draw, &sp->vbuf_render->base); + return &cvbr->base; } diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.h b/src/gallium/drivers/softpipe/sp_prim_vbuf.h index 1de9cc2a89..ad01cc2f28 100644 --- a/src/gallium/drivers/softpipe/sp_prim_vbuf.h +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.h @@ -31,8 +31,8 @@ struct softpipe_context; -extern void -sp_init_vbuf(struct softpipe_context *softpipe); +extern struct vbuf_render * +sp_create_vbuf_backend(struct softpipe_context *softpipe); #endif /* SP_VBUF_H */ diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index a132911c99..bc8366b0e6 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -33,7 +33,6 @@ */ #include "sp_context.h" -#include "sp_prim_setup.h" #include "sp_quad.h" #include "sp_quad_pipe.h" #include "sp_setup.h" diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 04fc125e3d..2a40589e84 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -68,24 +68,19 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe) const struct sp_fragment_shader *spfs = softpipe->fs; const enum interp_mode colorInterp = softpipe->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR; + struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf; + const uint num = draw_num_vs_outputs(softpipe->draw); uint i; - if (softpipe->vbuf) { - /* if using the post-transform vertex buffer, tell draw_vbuf to - * simply emit the whole post-xform vertex as-is: - */ - struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf; - const uint num = draw_num_vs_outputs(softpipe->draw); - uint i; - - /* No longer any need to try and emit draw vertex_header info. - */ - vinfo_vbuf->num_attribs = 0; - for (i = 0; i < num; i++) { - draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i); - } - draw_compute_vertex_size(vinfo_vbuf); + /* Tell draw_vbuf to simply emit the whole post-xform vertex + * as-is. No longer any need to try and emit draw vertex_header + * info. + */ + vinfo_vbuf->num_attribs = 0; + for (i = 0; i < num; i++) { + draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i); } + draw_compute_vertex_size(vinfo_vbuf); /* * Loop over fragment shader inputs, searching for the matching output -- cgit v1.2.3 From 153e474d22d1b440bb6bd7b04dabf244d7455582 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 13:38:10 +0100 Subject: softpipe: lift tex_address construction up to img_filter For fastpaths at least, can avoid recalculating this sometimes. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 60 ++++++++++++++++------------ 1 file changed, 35 insertions(+), 25 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index d233924565..f2d4f7eb8c 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -605,14 +605,14 @@ compute_lambda_vert(const struct sp_sampler_varient *samp, */ static INLINE void get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, - unsigned face, unsigned level, int x, int y, + union tex_tile_address addr, + unsigned x, unsigned y, const float *out[4]) { const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct softpipe_tex_cached_tile *tile - = sp_get_cached_tile_tex(samp->cache, - tex_tile_address(x, y, 0, face, level)); + = sp_get_cached_tile_tex(samp->cache, addr); y %= TILE_SIZE; x %= TILE_SIZE; @@ -625,36 +625,33 @@ get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, static INLINE const float * get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, - unsigned face, unsigned level, int x, int y) + union tex_tile_address addr, int x, int y) { const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct softpipe_tex_cached_tile *tile; - const struct softpipe_tex_cached_tile *tile - = sp_get_cached_tile_tex(samp->cache, - tex_tile_address(x, y, 0, face, level)); - + addr.bits.x = x / TILE_SIZE; + addr.bits.y = y / TILE_SIZE; y %= TILE_SIZE; x %= TILE_SIZE; + tile = sp_get_cached_tile_tex(samp->cache, addr); + return &tile->data.color[y][x][0]; } static INLINE void get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, - unsigned face, unsigned level, + union tex_tile_address addr, int x0, int y0, int x1, int y1, const float *out[4]) { - unsigned i; - - for (i = 0; i < 4; i++) { - unsigned tx = (i & 1) ? x1 : x0; - unsigned ty = (i >> 1) ? y1 : y0; - - out[i] = get_texel_2d_ptr( tgsi_sampler, face, level, tx, ty ); - } + out[0] = get_texel_2d_ptr( tgsi_sampler, addr, x0, y0 ); + out[1] = get_texel_2d_ptr( tgsi_sampler, addr, x1, y0 ); + out[2] = get_texel_2d_ptr( tgsi_sampler, addr, x0, y1 ); + out[3] = get_texel_2d_ptr( tgsi_sampler, addr, x1, y1 ); } static INLINE void @@ -714,7 +711,12 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, unsigned ypot = 1 << (samp->ypot - level); unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */ unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */ - + union tex_tile_address addr; + + addr.value = 0; + addr.bits.level = samp->level; + + for (j = 0; j < QUAD_SIZE; j++) { int c; @@ -730,21 +732,21 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, int x0 = uflr & (xpot - 1); int y0 = vflr & (ypot - 1); - const float *tx[4]; + const float *tx[4]; - /* Can we fetch all four at once: */ if (x0 < xmax && y0 < ymax) { - get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx); + addr.bits.x = x0 / TILE_SIZE; + addr.bits.y = y0 / TILE_SIZE; + get_texel_quad_2d(tgsi_sampler, addr, x0, y0, tx); } else { unsigned x1 = (x0 + 1) & (xpot - 1); unsigned y1 = (y0 + 1) & (ypot - 1); - get_texel_quad_2d_mt(tgsi_sampler, 0, level, - x0, y0, x1, y1, tx); + get_texel_quad_2d_mt(tgsi_sampler, addr, x0, y0, x1, y1, tx); } @@ -771,6 +773,10 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); unsigned ypot = 1 << (samp->ypot - level); + union tex_tile_address addr; + + addr.value = 0; + addr.bits.level = samp->level; for (j = 0; j < QUAD_SIZE; j++) { int c; @@ -784,7 +790,7 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, int x0 = uflr & (xpot - 1); int y0 = vflr & (ypot - 1); - const float *out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + const float *out = get_texel_2d_ptr(tgsi_sampler, addr, x0, y0); for (c = 0; c < 4; c++) { rgba[c][j] = out[c]; @@ -806,6 +812,10 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, unsigned level = samp->level; unsigned xpot = 1 << (samp->xpot - level); unsigned ypot = 1 << (samp->ypot - level); + union tex_tile_address addr; + + addr.value = 0; + addr.bits.level = samp->level; for (j = 0; j < QUAD_SIZE; j++) { int c; @@ -828,7 +838,7 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, else if (y0 > ypot - 1) y0 = ypot - 1; - out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0); + out = get_texel_2d_ptr(tgsi_sampler, addr, x0, y0); for (c = 0; c < 4; c++) { rgba[c][j] = out[c]; -- cgit v1.2.3 From 75312b655567695359063dcf3c28aa48433c575b Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 19:02:17 +0100 Subject: progs/demos: add fps output to cubemap --- progs/demos/cubemap.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/progs/demos/cubemap.c b/progs/demos/cubemap.c index 1f9f290575..0a3e10dac0 100644 --- a/progs/demos/cubemap.c +++ b/progs/demos/cubemap.c @@ -53,6 +53,8 @@ static GLint FrameParity = 0; static GLenum FilterIndex = 0; static GLint ClampIndex = 0; static GLboolean supportFBO = GL_FALSE; +static GLint T0 = 0; +static GLint Frames = 0; static struct { @@ -268,6 +270,20 @@ static void draw( void ) glPopMatrix(); glutSwapBuffers(); + + Frames++; + + { + GLint t = glutGet(GLUT_ELAPSED_TIME); + if (t - T0 >= 5000) { + GLfloat seconds = (t - T0) / 1000.0; + GLfloat fps = Frames / seconds; + printf("%d frames in %6.3f seconds = %6.3f FPS\n", Frames, seconds, fps); + fflush(stdout); + T0 = t; + Frames = 0; + } + } } -- cgit v1.2.3 From 81601d85ef6b82297b046d5aab1b70e75168c2fa Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 19:14:09 +0100 Subject: softpipe: make the various get_texel routines more similar Remove arguments, return const float * by default. Add specialized 3d versions and remove 3d texture support from the others. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 268 ++++++++++++++++++--------- 1 file changed, 176 insertions(+), 92 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index f2d4f7eb8c..8283010740 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -592,30 +592,68 @@ compute_lambda_vert(const struct sp_sampler_varient *samp, /** * Get a texel from a texture, using the texture tile cache. * - * \param face the cube face in 0..5 - * \param level the mipmap level + * \param addr the template tex address containing cube, z, face info. * \param x the x coord of texel within 2D image * \param y the y coord of texel within 2D image - * \param z which slice of a 3D texture * \param rgba the quad to put the texel/color into - * \param j which element of the rgba quad to write to * * XXX maybe move this into sp_tex_tile_cache.c and merge with the * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... */ -static INLINE void -get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, - union tex_tile_address addr, - unsigned x, unsigned y, - const float *out[4]) + + + + +static INLINE const float * +get_texel_2d_no_border(const struct sp_sampler_varient *samp, + union tex_tile_address addr, int x, int y) { - const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct softpipe_tex_cached_tile *tile; + + addr.bits.x = x / TILE_SIZE; + addr.bits.y = y / TILE_SIZE; + y %= TILE_SIZE; + x %= TILE_SIZE; + + tile = sp_get_cached_tile_tex(samp->cache, addr); + + return &tile->data.color[y][x][0]; +} + + +static INLINE const float * +get_texel_2d(const struct sp_sampler_varient *samp, + union tex_tile_address addr, int x, int y) +{ + const struct pipe_texture *texture = samp->texture; + unsigned level = addr.bits.level; + + if (x < 0 || x >= (int) texture->width[level] || + y < 0 || y >= (int) texture->height[level]) { + return samp->sampler->border_color; + } + else { + return get_texel_2d_no_border( samp, addr, x, y ); + } +} - const struct softpipe_tex_cached_tile *tile - = sp_get_cached_tile_tex(samp->cache, addr); +/* Gather a quad of adjacent texels within a tile: + */ +static INLINE void +get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp, + union tex_tile_address addr, + unsigned x, unsigned y, + const float *out[4]) +{ + const struct softpipe_tex_cached_tile *tile; + + addr.bits.x = x / TILE_SIZE; + addr.bits.y = y / TILE_SIZE; y %= TILE_SIZE; x %= TILE_SIZE; + + tile = sp_get_cached_tile_tex(samp->cache, addr); out[0] = &tile->data.color[y ][x ][0]; out[1] = &tile->data.color[y ][x+1][0]; @@ -623,15 +661,50 @@ get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler, out[3] = &tile->data.color[y+1][x+1][0]; } + +/* Gather a quad of potentially non-adjacent texels: + */ +static INLINE void +get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp, + union tex_tile_address addr, + int x0, int y0, + int x1, int y1, + const float *out[4]) +{ + out[0] = get_texel_2d_no_border( samp, addr, x0, y0 ); + out[1] = get_texel_2d_no_border( samp, addr, x1, y0 ); + out[2] = get_texel_2d_no_border( samp, addr, x0, y1 ); + out[3] = get_texel_2d_no_border( samp, addr, x1, y1 ); +} + +/* Can involve a lot of unnecessary checks for border color: + */ +static INLINE void +get_texel_quad_2d(const struct sp_sampler_varient *samp, + union tex_tile_address addr, + int x0, int y0, + int x1, int y1, + const float *out[4]) +{ + out[0] = get_texel_2d( samp, addr, x0, y0 ); + out[1] = get_texel_2d( samp, addr, x1, y0 ); + out[3] = get_texel_2d( samp, addr, x1, y1 ); + out[2] = get_texel_2d( samp, addr, x0, y1 ); +} + + + +/* 3d varients: + */ static INLINE const float * -get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, - union tex_tile_address addr, int x, int y) +get_texel_3d_no_border(const struct sp_sampler_varient *samp, + union tex_tile_address addr, int x, int y, int z) { - const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct softpipe_tex_cached_tile *tile; addr.bits.x = x / TILE_SIZE; addr.bits.y = y / TILE_SIZE; + addr.bits.z = z; y %= TILE_SIZE; x %= TILE_SIZE; @@ -641,61 +714,26 @@ get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler, } -static INLINE void -get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler, - union tex_tile_address addr, - int x0, int y0, - int x1, int y1, - const float *out[4]) -{ - out[0] = get_texel_2d_ptr( tgsi_sampler, addr, x0, y0 ); - out[1] = get_texel_2d_ptr( tgsi_sampler, addr, x1, y0 ); - out[2] = get_texel_2d_ptr( tgsi_sampler, addr, x0, y1 ); - out[3] = get_texel_2d_ptr( tgsi_sampler, addr, x1, y1 ); -} - -static INLINE void -get_texel(const struct tgsi_sampler *tgsi_sampler, - unsigned face, unsigned level, int x, int y, int z, - float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) +static INLINE const float * +get_texel_3d(const struct sp_sampler_varient *samp, + union tex_tile_address addr, int x, int y, int z ) { - const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); const struct pipe_texture *texture = samp->texture; - const struct pipe_sampler_state *sampler = samp->sampler; + unsigned level = addr.bits.level; if (x < 0 || x >= (int) texture->width[level] || y < 0 || y >= (int) texture->height[level] || z < 0 || z >= (int) texture->depth[level]) { - rgba[0][j] = sampler->border_color[0]; - rgba[1][j] = sampler->border_color[1]; - rgba[2][j] = sampler->border_color[2]; - rgba[3][j] = sampler->border_color[3]; + return samp->sampler->border_color; } else { - const unsigned tx = x % TILE_SIZE; - const unsigned ty = y % TILE_SIZE; - const struct softpipe_tex_cached_tile *tile; - - tile = sp_get_cached_tile_tex(samp->cache, - tex_tile_address(x, y, z, face, level)); - - rgba[0][j] = tile->data.color[ty][tx][0]; - rgba[1][j] = tile->data.color[ty][tx][1]; - rgba[2][j] = tile->data.color[ty][tx][2]; - rgba[3][j] = tile->data.color[ty][tx][3]; - if (0) - { - debug_printf("Get texel %f %f %f %f from %s\n", - rgba[0][j], rgba[1][j], rgba[2][j], rgba[3][j], - pf_name(texture->format)); - } + return get_texel_3d_no_border( samp, addr, x, y, z ); } } - - - +/* Some image-filter fastpaths: + */ static INLINE void img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], @@ -738,15 +776,13 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, */ if (x0 < xmax && y0 < ymax) { - addr.bits.x = x0 / TILE_SIZE; - addr.bits.y = y0 / TILE_SIZE; - get_texel_quad_2d(tgsi_sampler, addr, x0, y0, tx); + get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx); } else { unsigned x1 = (x0 + 1) & (xpot - 1); unsigned y1 = (y0 + 1) & (ypot - 1); - get_texel_quad_2d_mt(tgsi_sampler, addr, x0, y0, x1, y1, tx); + get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx); } @@ -790,7 +826,7 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, int x0 = uflr & (xpot - 1); int y0 = vflr & (ypot - 1); - const float *out = get_texel_2d_ptr(tgsi_sampler, addr, x0, y0); + const float *out = get_texel_2d_no_border(samp, addr, x0, y0); for (c = 0; c < 4; c++) { rgba[c][j] = out[c]; @@ -838,7 +874,7 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, else if (y0 > ypot - 1) y0 = ypot - 1; - out = get_texel_2d_ptr(tgsi_sampler, addr, x0, y0); + out = get_texel_2d_no_border(samp, addr, x0, y0); for (c = 0; c < 4; c++) { rgba[c][j] = out[c]; @@ -859,20 +895,37 @@ img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler, unsigned level0, j; int width; int x[4]; + union tex_tile_address addr; level0 = samp->level; width = texture->width[level0]; assert(width > 0); + addr.value = 0; + addr.bits.level = samp->level; + samp->nearest_texcoord_s(s, width, x); for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, 0, level0, x[j], 0, 0, rgba, j); + const float *out = get_texel_2d(samp, addr, x[j], 0); + int c; + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } } } + + +static inline union tex_tile_address face( union tex_tile_address addr, + unsigned face ) +{ + addr.bits.face = face; + return addr; +} + static void img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], @@ -887,18 +940,27 @@ img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, unsigned level0, j; int width, height; int x[4], y[4]; + union tex_tile_address addr; + level0 = samp->level; width = texture->width[level0]; height = texture->height[level0]; assert(width > 0); + + addr.value = 0; + addr.bits.level = samp->level; samp->nearest_texcoord_s(s, width, x); samp->nearest_texcoord_t(t, height, y); for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j); + const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]); + int c; + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } } } @@ -916,6 +978,7 @@ img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler, unsigned level0, j; int width, height, depth; int x[4], y[4], z[4]; + union tex_tile_address addr; level0 = samp->level; width = texture->width[level0]; @@ -930,8 +993,15 @@ img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler, samp->nearest_texcoord_t(t, height, y); samp->nearest_texcoord_p(p, depth, z); + addr.value = 0; + addr.bits.level = samp->level; + for (j = 0; j < QUAD_SIZE; j++) { - get_texel(tgsi_sampler, 0, level0, x[j], y[j], z[j], rgba, j); + const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]); + int c; + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } } } @@ -950,6 +1020,7 @@ img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler, int width; int x0[4], x1[4]; float xw[4]; /* weights */ + union tex_tile_address addr; level0 = samp->level; @@ -957,22 +1028,27 @@ img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler, assert(width > 0); + addr.value = 0; + addr.bits.level = samp->level; + samp->linear_texcoord_s(s, width, x0, x1, xw); for (j = 0; j < QUAD_SIZE; j++) { - float tx[4][4]; /* texels */ + const float *tx0 = get_texel_2d(samp, addr, x0[j], 0); + const float *tx1 = get_texel_2d(samp, addr, x1[j], 0); int c; - get_texel(tgsi_sampler, 0, level0, x0[j], 0, 0, tx, 0); - get_texel(tgsi_sampler, 0, level0, x1[j], 0, 0, tx, 1); /* interpolate R, G, B, A */ for (c = 0; c < 4; c++) { - rgba[c][j] = lerp(xw[j], tx[c][0], tx[c][1]); + rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]); } } } + + + static void img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], @@ -988,6 +1064,7 @@ img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, int width, height; int x0[4], y0[4], x1[4], y1[4]; float xw[4], yw[4]; /* weights */ + union tex_tile_address addr; level0 = samp->level; @@ -996,22 +1073,25 @@ img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, assert(width > 0); + addr.value = 0; + addr.bits.level = samp->level; + samp->linear_texcoord_s(s, width, x0, x1, xw); samp->linear_texcoord_t(t, height, y0, y1, yw); for (j = 0; j < QUAD_SIZE; j++) { - float tx[4][4]; /* texels */ + union tex_tile_address addrj = face(addr, faces[j]); + const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]); + const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]); + const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]); + const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]); int c; - get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0); - get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1); - get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2); - get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3); /* interpolate R, G, B, A */ for (c = 0; c < 4; c++) { rgba[c][j] = lerp_2d(xw[j], yw[j], - tx[c][0], tx[c][1], - tx[c][2], tx[c][3]); + tx0[c], tx1[c], + tx2[c], tx3[c]); } } } @@ -1031,12 +1111,16 @@ img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler, int width, height, depth; int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4]; float xw[4], yw[4], zw[4]; /* interpolation weights */ + union tex_tile_address addr; level0 = samp->level; width = texture->width[level0]; height = texture->height[level0]; depth = texture->depth[level0]; + addr.value = 0; + addr.bits.level = level0; + assert(width > 0); assert(height > 0); assert(depth > 0); @@ -1046,25 +1130,25 @@ img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler, samp->linear_texcoord_p(p, depth, z0, z1, zw); for (j = 0; j < QUAD_SIZE; j++) { - float tx0[4][4], tx1[4][4]; int c; - - get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z0[j], tx0, 0); - get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z0[j], tx0, 1); - get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z0[j], tx0, 2); - get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z0[j], tx0, 3); - get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z1[j], tx1, 0); - get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z1[j], tx1, 1); - get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z1[j], tx1, 2); - get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z1[j], tx1, 3); + const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]); + const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]); + const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]); + const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]); + + const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]); + const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]); + const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]); + const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]); + /* interpolate R, G, B, A */ for (c = 0; c < 4; c++) { rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j], - tx0[c][0], tx0[c][1], - tx0[c][2], tx0[c][3], - tx1[c][0], tx1[c][1], - tx1[c][2], tx1[c][3]); + tx00[c], tx01[c], + tx02[c], tx03[c], + tx10[c], tx11[c], + tx12[c], tx13[c]); } } } -- cgit v1.2.3 From 60adc15ba5633190fc8a68e7c182f06dc7909df4 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 19:17:35 +0100 Subject: softpipe: separate out 2d and cube img filter functions --- src/gallium/drivers/softpipe/sp_tex_sample.c | 92 ++++++++++++++++++++++++++-- 1 file changed, 88 insertions(+), 4 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 8283010740..3bc4599e04 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -917,8 +917,43 @@ img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler, } +static void +img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + unsigned level0, j; + int width, height; + int x[4], y[4]; + union tex_tile_address addr; + level0 = samp->level; + width = texture->width[level0]; + height = texture->height[level0]; + + assert(width > 0); + + addr.value = 0; + addr.bits.level = samp->level; + + samp->nearest_texcoord_s(s, width, x); + samp->nearest_texcoord_t(t, height, y); + + for (j = 0; j < QUAD_SIZE; j++) { + const float *out = get_texel_2d(samp, addr, x[j], y[j]); + int c; + for (c = 0; c < 4; c++) { + rgba[c][j] = out[c]; + } + } +} + static inline union tex_tile_address face( union tex_tile_address addr, unsigned face ) { @@ -927,7 +962,7 @@ static inline union tex_tile_address face( union tex_tile_address addr, } static void -img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, +img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], @@ -1047,10 +1082,54 @@ img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler, } +static void +img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); + const struct pipe_texture *texture = samp->texture; + unsigned level0, j; + int width, height; + int x0[4], y0[4], x1[4], y1[4]; + float xw[4], yw[4]; /* weights */ + union tex_tile_address addr; + + + level0 = samp->level; + width = texture->width[level0]; + height = texture->height[level0]; + + assert(width > 0); + + addr.value = 0; + addr.bits.level = samp->level; + + samp->linear_texcoord_s(s, width, x0, x1, xw); + samp->linear_texcoord_t(t, height, y0, y1, yw); + + for (j = 0; j < QUAD_SIZE; j++) { + const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]); + const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]); + const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]); + const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]); + int c; + + /* interpolate R, G, B, A */ + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(xw[j], yw[j], + tx0[c], tx1[c], + tx2[c], tx3[c]); + } + } +} static void -img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, +img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], @@ -1615,14 +1694,19 @@ static filter_func get_img_filter( const union sp_sampler_key key, } } } - /* Fallthrough to default versions: + /* Otherwise use default versions: */ - case PIPE_TEXTURE_CUBE: if (filter == PIPE_TEX_FILTER_NEAREST) return img_filter_2d_nearest; else return img_filter_2d_linear; break; + case PIPE_TEXTURE_CUBE: + if (filter == PIPE_TEX_FILTER_NEAREST) + return img_filter_cube_nearest; + else + return img_filter_cube_linear; + break; case PIPE_TEXTURE_3D: if (filter == PIPE_TEX_FILTER_NEAREST) return img_filter_3d_nearest; -- cgit v1.2.3 From fd19e8adcd82e88d0fc8d187360b528100fed244 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 19:28:34 +0100 Subject: softpipe: use one fewer divide in sample_cube GCC won't do this for us. Makes a bigger difference to cubemap fps than previous set of compilcated rearrangements. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 3bc4599e04..50460df7cd 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -1543,9 +1543,12 @@ sample_cube(struct tgsi_sampler *tgsi_sampler, } } - ssss[j] = ( sc / ma + 1.0F ) * 0.5F; - tttt[j] = ( tc / ma + 1.0F ) * 0.5F; - samp->faces[j] = face; + { + const float ima = 1.0 / ma; + ssss[j] = ( sc * ima + 1.0F ) * 0.5F; + tttt[j] = ( tc * ima + 1.0F ) * 0.5F; + samp->faces[j] = face; + } } /* In our little pipeline, the compare stage is next. If compare -- cgit v1.2.3 From 67d4a5b15cfd8583c19a5776b0ec1564b60239eb Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 23 Aug 2009 23:50:27 +0100 Subject: mesa/swrast: use one fewer divide in swrast's choose_cube_face also Same change as for softpipe --- src/mesa/swrast/s_texfilter.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/src/mesa/swrast/s_texfilter.c b/src/mesa/swrast/s_texfilter.c index 6b1f934647..216c107e3f 100644 --- a/src/mesa/swrast/s_texfilter.c +++ b/src/mesa/swrast/s_texfilter.c @@ -1905,8 +1905,12 @@ choose_cube_face(const struct gl_texture_object *texObj, } } - newCoord[0] = ( sc / ma + 1.0F ) * 0.5F; - newCoord[1] = ( tc / ma + 1.0F ) * 0.5F; + { + const float ima = 1.0F / ma; + newCoord[0] = ( sc * ima + 1.0F ) * 0.5F; + newCoord[1] = ( tc * ima + 1.0F ) * 0.5F; + } + return (const struct gl_texture_image **) texObj->Image[face]; } -- cgit v1.2.3 From b626176f0613852df908b4b0552b9b67d5830b4e Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 22 Sep 2009 19:26:08 +0100 Subject: softpipe: fix occlusion counting --- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index 9cffea2c9e..ce1bab9341 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -631,7 +631,7 @@ alpha_test_quads(struct quad_stage *qs, } } -static unsigned mask_count[0x8] = +static unsigned mask_count[16] = { 0, /* 0x0 */ 1, /* 0x1 */ @@ -641,6 +641,14 @@ static unsigned mask_count[0x8] = 2, /* 0x5 */ 2, /* 0x6 */ 3, /* 0x7 */ + 1, /* 0x8 */ + 2, /* 0x9 */ + 2, /* 0xa */ + 3, /* 0xb */ + 2, /* 0xc */ + 3, /* 0xd */ + 3, /* 0xe */ + 4, /* 0xf */ }; @@ -693,13 +701,17 @@ depth_test_quads_fallback(struct quad_stage *qs, qs->softpipe->depth_stencil->depth.writemask) write_depth_stencil_values(&data, quads[i]); - qs->softpipe->occlusion_count += mask_count[quads[i]->inout.mask]; quads[pass++] = quads[i]; } nr = pass; } + if (qs->softpipe->active_query_count) { + for (i = 0; i < nr; i++) + qs->softpipe->occlusion_count += mask_count[quads[i]->inout.mask]; + } + if (nr) qs->next->run(qs->next, quads, nr); } @@ -883,6 +895,8 @@ choose_depth_test(struct quad_stage *qs, boolean depthwrite = qs->softpipe->depth_stencil->depth.writemask; + boolean occlusion = qs->softpipe->active_query_count; + if (!alpha && !depth && @@ -893,6 +907,7 @@ choose_depth_test(struct quad_stage *qs, interp_depth && depth && depthwrite && + !occlusion && !stencil) { switch (depthfunc) { -- cgit v1.2.3 From b1139e9ad827d86886772a9c9d83dbb0071c702c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 22 Sep 2009 19:38:34 +0100 Subject: softpipe: fix polygon stipple --- src/gallium/drivers/softpipe/sp_context.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 6b75ee6002..c4b8b33c6a 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -262,10 +262,8 @@ softpipe_create( struct pipe_screen *screen ) draw_install_aaline_stage(softpipe->draw, &softpipe->pipe); draw_install_aapoint_stage(softpipe->draw, &softpipe->pipe); -#if USE_DRAW_STAGE_PSTIPPLE /* Do polygon stipple w/ texture map + frag prog? */ draw_install_pstipple_stage(softpipe->draw, &softpipe->pipe); -#endif sp_init_surface_functions(softpipe); -- cgit v1.2.3 From 207764894b6d565568bc46722e4c239d839a62fc Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 22 Sep 2009 20:47:07 +0100 Subject: softpipe: set quad->facing value --- src/gallium/drivers/softpipe/sp_setup.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index bc8366b0e6..ade125662a 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -240,6 +240,7 @@ static void flush_spans( struct setup_context *setup ) if (quadmask) { setup->quad[q].input.x0 = lx; setup->quad[q].input.y0 = setup->span.y; + setup->quad[q].input.facing = setup->facing; setup->quad[q].inout.mask = quadmask; setup->quad_ptrs[q] = &setup->quad[q]; q++; -- cgit v1.2.3 From fe9ca0f718cbc467e5cee99a2c20a5f257ed2fe1 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 22 Sep 2009 20:47:37 +0100 Subject: softpipe: need to write depth/stencil values even when stencil fails --- src/gallium/drivers/softpipe/sp_quad_depth_test.c | 22 ++++++++++------------ 1 file changed, 10 insertions(+), 12 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c index ce1bab9341..0ca86c4e1c 100644 --- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -498,7 +498,7 @@ depth_test_quad(struct quad_stage *qs, * Do stencil (and depth) testing. Stenciling depends on the outcome of * depth testing. */ -static boolean +static void depth_stencil_test_quad(struct quad_stage *qs, struct depth_data *data, struct quad_header *quad) @@ -545,13 +545,13 @@ depth_stencil_test_quad(struct quad_stage *qs, /* update stencil buffer values according to z pass/fail result */ if (zFailOp != PIPE_STENCIL_OP_KEEP) { - const unsigned failMask = origMask & ~quad->inout.mask; - apply_stencil_op(data, failMask, zFailOp, ref, wrtMask); + const unsigned zFailMask = origMask & ~quad->inout.mask; + apply_stencil_op(data, zFailMask, zFailOp, ref, wrtMask); } if (zPassOp != PIPE_STENCIL_OP_KEEP) { - const unsigned passMask = origMask & quad->inout.mask; - apply_stencil_op(data, passMask, zPassOp, ref, wrtMask); + const unsigned zPassMask = origMask & quad->inout.mask; + apply_stencil_op(data, zPassMask, zPassOp, ref, wrtMask); } } else { @@ -559,8 +559,6 @@ depth_stencil_test_quad(struct quad_stage *qs, apply_stencil_op(data, quad->inout.mask, zPassOp, ref, wrtMask); } } - - return quad->inout.mask != 0; } @@ -689,17 +687,17 @@ depth_test_quads_fallback(struct quad_stage *qs, } if (qs->softpipe->depth_stencil->stencil[0].enabled) { - if (!depth_stencil_test_quad(qs, &data, quads[i])) - continue; + depth_stencil_test_quad(qs, &data, quads[i]); + write_depth_stencil_values(&data, quads[i]); } else { if (!depth_test_quad(qs, &data, quads[i])) continue; + + if (qs->softpipe->depth_stencil->depth.writemask) + write_depth_stencil_values(&data, quads[i]); } - if (qs->softpipe->depth_stencil->stencil[0].enabled || - qs->softpipe->depth_stencil->depth.writemask) - write_depth_stencil_values(&data, quads[i]); quads[pass++] = quads[i]; } -- cgit v1.2.3 From 0670df5cb20c0b6630ab29511d9b2cbe18b47f65 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Tue, 22 Sep 2009 16:42:15 -0600 Subject: softpipe: disable a _debug_printf() --- src/gallium/drivers/softpipe/sp_state_derived.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c index 2a40589e84..856c9ce176 100644 --- a/src/gallium/drivers/softpipe/sp_state_derived.c +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -203,7 +203,9 @@ update_tgsi_samplers( struct softpipe_context *softpipe ) struct softpipe_texture *spt = softpipe_texture(tc->texture); if (spt->timestamp != tc->timestamp) { sp_tex_tile_cache_validate_texture( tc ); + /* _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp); + */ tc->timestamp = spt->timestamp; } } -- cgit v1.2.3 From 5dbedf3d7e99efe35fad308d382670e44cd60e25 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Tue, 22 Sep 2009 16:59:28 -0600 Subject: softpipe: additional assertions --- src/gallium/drivers/softpipe/sp_tex_sample.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index 50460df7cd..be210d5671 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -938,6 +938,7 @@ img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler, height = texture->height[level0]; assert(width > 0); + assert(height > 0); addr.value = 0; addr.bits.level = samp->level; @@ -983,6 +984,7 @@ img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler, height = texture->height[level0]; assert(width > 0); + assert(height > 0); addr.value = 0; addr.bits.level = samp->level; @@ -1104,6 +1106,7 @@ img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler, height = texture->height[level0]; assert(width > 0); + assert(height > 0); addr.value = 0; addr.bits.level = samp->level; @@ -1151,6 +1154,7 @@ img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler, height = texture->height[level0]; assert(width > 0); + assert(height > 0); addr.value = 0; addr.bits.level = samp->level; -- cgit v1.2.3 From 75276ea316610a5737f2115326482024aa09d02a Mon Sep 17 00:00:00 2001 From: root Date: Tue, 22 Sep 2009 20:14:05 -0600 Subject: softpipe: fix bugs in POT texture sampling when texture is not square Before, if level was greater than the logbase2(base size) we were doing a negative bit shift and winding up with garbage values. --- src/gallium/drivers/softpipe/sp_tex_sample.c | 34 +++++++++++++++++++++++----- 1 file changed, 28 insertions(+), 6 deletions(-) diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c index be210d5671..ba9b91a378 100644 --- a/src/gallium/drivers/softpipe/sp_tex_sample.c +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -732,6 +732,20 @@ get_texel_3d(const struct sp_sampler_varient *samp, } +/** + * Given the logbase2 of a mipmap's base level size and a mipmap level, + * return the size (in texels) of that mipmap level. + * For example, if level[0].width = 256 then base_pot will be 8. + * If level = 2, then we'll return 64 (the width at level=2). + * Return 1 if level > base_pot. + */ +static INLINE unsigned +pot_level_size(unsigned base_pot, unsigned level) +{ + return (base_pot >= level) ? (1 << (base_pot - level)) : 1; +} + + /* Some image-filter fastpaths: */ static INLINE void @@ -745,8 +759,8 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler, const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; - unsigned xpot = 1 << (samp->xpot - level); - unsigned ypot = 1 << (samp->ypot - level); + unsigned xpot = pot_level_size(samp->xpot, level); + unsigned ypot = pot_level_size(samp->ypot, level); unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */ unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */ union tex_tile_address addr; @@ -807,8 +821,8 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler, const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; - unsigned xpot = 1 << (samp->xpot - level); - unsigned ypot = 1 << (samp->ypot - level); + unsigned xpot = pot_level_size(samp->xpot, level); + unsigned ypot = pot_level_size(samp->ypot, level); union tex_tile_address addr; addr.value = 0; @@ -846,8 +860,8 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler, const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler); unsigned j; unsigned level = samp->level; - unsigned xpot = 1 << (samp->xpot - level); - unsigned ypot = 1 << (samp->ypot - level); + unsigned xpot = pot_level_size(samp->xpot, level); + unsigned ypot = pot_level_size(samp->ypot, level); union tex_tile_address addr; addr.value = 0; @@ -1311,6 +1325,14 @@ mip_filter_nearest(struct tgsi_sampler *tgsi_sampler, samp->level = MIN2(samp->level, (int)texture->last_level); samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba ); } + +#if 0 + printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n", + rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0], + rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1], + rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2], + rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]); +#endif } -- cgit v1.2.3