diff options
Diffstat (limited to 'src/gallium/drivers/nv40')
23 files changed, 1292 insertions, 862 deletions
diff --git a/src/gallium/drivers/nv40/Makefile b/src/gallium/drivers/nv40/Makefile index 2a9de4a2dc..3369a21574 100644 --- a/src/gallium/drivers/nv40/Makefile +++ b/src/gallium/drivers/nv40/Makefile @@ -11,8 +11,17 @@ DRIVER_SOURCES = \ nv40_fragtex.c \ nv40_miptree.c \ nv40_query.c \ + nv40_screen.c \ nv40_state.c \ + nv40_state_blend.c \ + nv40_state_clip.c \ nv40_state_emit.c \ + nv40_state_fb.c \ + nv40_state_rasterizer.c \ + nv40_state_scissor.c \ + nv40_state_stipple.c \ + nv40_state_viewport.c \ + nv40_state_zsa.c \ nv40_surface.c \ nv40_vbo.c \ nv40_vertprog.c diff --git a/src/gallium/drivers/nv40/nv40_context.c b/src/gallium/drivers/nv40/nv40_context.c index 8b5cc693de..203c843a01 100644 --- a/src/gallium/drivers/nv40/nv40_context.c +++ b/src/gallium/drivers/nv40/nv40_context.c @@ -4,84 +4,7 @@ #include "pipe/p_util.h" #include "nv40_context.h" - -#define NV4X_GRCLASS4097_CHIPSETS 0x00000baf -#define NV4X_GRCLASS4497_CHIPSETS 0x00005450 -#define NV6X_GRCLASS4497_CHIPSETS 0x00000088 - -static const char * -nv40_get_name(struct pipe_context *pipe) -{ - struct nv40_context *nv40 = nv40_context(pipe); - static char buffer[128]; - - snprintf(buffer, sizeof(buffer), "NV%02X", nv40->chipset); - return buffer; -} - -static const char * -nv40_get_vendor(struct pipe_context *pipe) -{ - return "nouveau"; -} - -static int -nv40_get_param(struct pipe_context *pipe, int param) -{ - switch (param) { - case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: - return 16; - case PIPE_CAP_NPOT_TEXTURES: - return 1; - case PIPE_CAP_TWO_SIDED_STENCIL: - return 1; - case PIPE_CAP_GLSL: - return 0; - case PIPE_CAP_S3TC: - return 0; - case PIPE_CAP_ANISOTROPIC_FILTER: - return 1; - case PIPE_CAP_POINT_SPRITE: - return 1; - case PIPE_CAP_MAX_RENDER_TARGETS: - return 4; - case PIPE_CAP_OCCLUSION_QUERY: - return 1; - case PIPE_CAP_TEXTURE_SHADOW_MAP: - return 1; - case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: - return 13; - case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: - return 10; - case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: - return 13; - default: - NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param); - return 0; - } -} - -static float -nv40_get_paramf(struct pipe_context *pipe, int param) -{ - switch (param) { - case PIPE_CAP_MAX_LINE_WIDTH: - case PIPE_CAP_MAX_LINE_WIDTH_AA: - return 10.0; - case PIPE_CAP_MAX_POINT_WIDTH: - case PIPE_CAP_MAX_POINT_WIDTH_AA: - return 64.0; - case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: - return 16.0; - case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: - return 16.0; - case PIPE_CAP_BITMAP_TEXCOORD_BIAS: - return 0.0; - default: - NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param); - return 0.0; - } -} +#include "nv40_screen.h" static void nv40_flush(struct pipe_context *pipe, unsigned flags) @@ -97,7 +20,7 @@ nv40_flush(struct pipe_context *pipe, unsigned flags) } if (flags & PIPE_FLUSH_WAIT) { - nvws->notifier_reset(nv40->hw->sync, 0); + nvws->notifier_reset(nv40->screen->sync, 0); BEGIN_RING(curie, 0x104, 1); OUT_RING (0); BEGIN_RING(curie, 0x100, 1); @@ -107,149 +30,7 @@ nv40_flush(struct pipe_context *pipe, unsigned flags) FIRE_RING(); if (flags & PIPE_FLUSH_WAIT) - nvws->notifier_wait(nv40->hw->sync, 0, 0, 2000); -} - -static void -nv40_channel_takedown(struct nv40_channel_context *cnv40) -{ - struct nouveau_winsys *nvws = cnv40->nvws; - - nvws->res_free(&cnv40->vp_exec_heap); - nvws->res_free(&cnv40->vp_data_heap); - nvws->res_free(&cnv40->query_heap); - nvws->notifier_free(&cnv40->query); - nvws->notifier_free(&cnv40->sync); - nvws->grobj_free(&cnv40->curie); - free(cnv40); -} - -static struct nv40_channel_context * -nv40_channel_init(struct pipe_winsys *ws, struct nouveau_winsys *nvws, - unsigned chipset) -{ - struct nv40_channel_context *cnv40 = NULL; - struct nouveau_stateobj *so; - unsigned curie_class = 0; - int ret; - - switch (chipset & 0xf0) { - case 0x40: - if (NV4X_GRCLASS4097_CHIPSETS & (1 << (chipset & 0x0f))) - curie_class = NV40TCL; - else - if (NV4X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) - curie_class = NV44TCL; - break; - case 0x60: - if (NV6X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) - curie_class = NV44TCL; - break; - default: - break; - } - - if (!curie_class) { - NOUVEAU_ERR("Unknown nv4x chipset: nv%02x\n", chipset); - return NULL; - } - - cnv40 = CALLOC(1, sizeof(struct nv40_channel_context)); - if (!cnv40) - return NULL; - cnv40->chipset = chipset; - cnv40->nvws = nvws; - - /* Notifier for sync purposes */ - ret = nvws->notifier_alloc(nvws, 1, &cnv40->sync); - if (ret) { - NOUVEAU_ERR("Error creating notifier object: %d\n", ret); - nv40_channel_takedown(cnv40); - return NULL; - } - - /* Query objects */ - ret = nvws->notifier_alloc(nvws, 32, &cnv40->query); - if (ret) { - NOUVEAU_ERR("Error initialising query objects: %d\n", ret); - nv40_channel_takedown(cnv40); - return NULL; - } - - ret = nvws->res_init(&cnv40->query_heap, 0, 32); - if (ret) { - NOUVEAU_ERR("Error initialising query object heap: %d\n", ret); - nv40_channel_takedown(cnv40); - return NULL; - } - - /* Vtxprog resources */ - if (nvws->res_init(&cnv40->vp_exec_heap, 0, 512) || - nvws->res_init(&cnv40->vp_data_heap, 0, 256)) { - nv40_channel_takedown(cnv40); - return NULL; - } - - /* 3D object */ - ret = nvws->grobj_alloc(nvws, curie_class, &cnv40->curie); - if (ret) { - NOUVEAU_ERR("Error creating 3D object: %d\n", ret); - return FALSE; - } - - /* Static curie initialisation */ - so = so_new(128, 0); - so_method(so, cnv40->curie, NV40TCL_DMA_NOTIFY, 1); - so_data (so, cnv40->sync->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_TEXTURE0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_COLOR1, 1); - so_data (so, nvws->channel->vram->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_COLOR0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_VTXBUF0, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->gart->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_FENCE, 2); - so_data (so, 0); - so_data (so, cnv40->query->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_UNK01AC, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); - so_method(so, cnv40->curie, NV40TCL_DMA_COLOR2, 2); - so_data (so, nvws->channel->vram->handle); - so_data (so, nvws->channel->vram->handle); - - so_method(so, cnv40->curie, 0x1ea4, 3); - so_data (so, 0x00000010); - so_data (so, 0x01000100); - so_data (so, 0xff800006); - - /* vtxprog output routing */ - so_method(so, cnv40->curie, 0x1fc4, 1); - so_data (so, 0x06144321); - so_method(so, cnv40->curie, 0x1fc8, 2); - so_data (so, 0xedcba987); - so_data (so, 0x00000021); - so_method(so, cnv40->curie, 0x1fd0, 1); - so_data (so, 0x00171615); - so_method(so, cnv40->curie, 0x1fd4, 1); - so_data (so, 0x001b1a19); - - so_method(so, cnv40->curie, 0x1ef8, 1); - so_data (so, 0x0020ffff); - so_method(so, cnv40->curie, 0x1d64, 1); - so_data (so, 0x00d30000); - so_method(so, cnv40->curie, 0x1e94, 1); - so_data (so, 0x00000001); - - so_emit(nvws, so); - so_ref(NULL, &so); - nvws->push_flush(nvws->channel, 0); - - return cnv40; + nvws->notifier_wait(nv40->screen->sync, 0, 0, 2000); } static void @@ -259,40 +40,30 @@ nv40_destroy(struct pipe_context *pipe) if (nv40->draw) draw_destroy(nv40->draw); - - if (nv40->hw) { - if (--nv40->hw->refcount == 0) - nv40_channel_takedown(nv40->hw); - } - free(nv40); } struct pipe_context * -nv40_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws, - unsigned chipset) +nv40_create(struct pipe_screen *pscreen, unsigned pctx_id) { + struct nv40_screen *screen = nv40_screen(pscreen); + struct pipe_winsys *ws = pscreen->winsys; struct nv40_context *nv40; + unsigned chipset = screen->chipset; + struct nouveau_winsys *nvws = screen->nvws; nv40 = CALLOC(1, sizeof(struct nv40_context)); if (!nv40) return NULL; - - nv40->hw = nv40_channel_init(ws, nvws, chipset); - if (!nv40->hw) { - nv40_destroy(&nv40->pipe); - return NULL; - } + nv40->screen = screen; + nv40->pctx_id = pctx_id; nv40->chipset = chipset; nv40->nvws = nvws; nv40->pipe.winsys = ws; + nv40->pipe.screen = pscreen; nv40->pipe.destroy = nv40_destroy; - nv40->pipe.get_name = nv40_get_name; - nv40->pipe.get_vendor = nv40_get_vendor; - nv40->pipe.get_param = nv40_get_param; - nv40->pipe.get_paramf = nv40_get_paramf; nv40->pipe.draw_arrays = nv40_draw_arrays; nv40->pipe.draw_elements = nv40_draw_elements; nv40->pipe.clear = nv40_clear; diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h index f511759e3b..e118776306 100644 --- a/src/gallium/drivers/nv40/nv40_context.h +++ b/src/gallium/drivers/nv40/nv40_context.h @@ -11,7 +11,7 @@ #include "nouveau/nouveau_gldefs.h" #define NOUVEAU_PUSH_CONTEXT(ctx) \ - struct nv40_channel_context *ctx = nv40->hw + struct nv40_screen *ctx = nv40->screen #include "nouveau/nouveau_push.h" #include "nouveau/nouveau_stateobj.h" @@ -22,6 +22,45 @@ #define NOUVEAU_MSG(fmt, args...) \ fprintf(stderr, "nouveau: "fmt, ##args); +enum nv40_state_index { + NV40_STATE_FB = 0, + NV40_STATE_VIEWPORT = 1, + NV40_STATE_BLEND = 2, + NV40_STATE_RAST = 3, + NV40_STATE_ZSA = 4, + NV40_STATE_BCOL = 5, + NV40_STATE_CLIP = 6, + NV40_STATE_SCISSOR = 7, + NV40_STATE_STIPPLE = 8, + NV40_STATE_FRAGPROG = 9, + NV40_STATE_VERTPROG = 10, + NV40_STATE_FRAGTEX0 = 11, + NV40_STATE_FRAGTEX1 = 12, + NV40_STATE_FRAGTEX2 = 13, + NV40_STATE_FRAGTEX3 = 14, + NV40_STATE_FRAGTEX4 = 15, + NV40_STATE_FRAGTEX5 = 16, + NV40_STATE_FRAGTEX6 = 17, + NV40_STATE_FRAGTEX7 = 18, + NV40_STATE_FRAGTEX8 = 19, + NV40_STATE_FRAGTEX9 = 20, + NV40_STATE_FRAGTEX10 = 21, + NV40_STATE_FRAGTEX11 = 22, + NV40_STATE_FRAGTEX12 = 23, + NV40_STATE_FRAGTEX13 = 24, + NV40_STATE_FRAGTEX14 = 25, + NV40_STATE_FRAGTEX15 = 26, + NV40_STATE_VERTTEX0 = 27, + NV40_STATE_VERTTEX1 = 28, + NV40_STATE_VERTTEX2 = 29, + NV40_STATE_VERTTEX3 = 30, + NV40_STATE_VTXBUF = 31, + NV40_STATE_VTXFMT = 32, + NV40_STATE_MAX = 33 +}; + +#include "nv40_screen.h" + #define NV40_NEW_BLEND (1 << 0) #define NV40_NEW_RAST (1 << 1) #define NV40_NEW_ZSA (1 << 2) @@ -34,68 +73,70 @@ #define NV40_NEW_VERTPROG (1 << 9) #define NV40_NEW_FRAGPROG (1 << 10) #define NV40_NEW_ARRAYS (1 << 11) +#define NV40_NEW_UCP (1 << 12) -struct nv40_channel_context { - struct nouveau_winsys *nvws; - unsigned refcount; +#define NV40_FALLBACK_TNL (1 << 0) +#define NV40_FALLBACK_RAST (1 << 1) + +struct nv40_rasterizer_state { + struct pipe_rasterizer_state pipe; + struct nouveau_stateobj *so; +}; - unsigned chipset; +struct nv40_zsa_state { + struct pipe_depth_stencil_alpha_state pipe; + struct nouveau_stateobj *so; +}; + +struct nv40_blend_state { + struct pipe_blend_state pipe; + struct nouveau_stateobj *so; +}; - /* HW graphics objects */ - struct nouveau_grobj *curie; - struct nouveau_notifier *sync; - /* Query object resources */ - struct nouveau_notifier *query; - struct nouveau_resource *query_heap; +struct nv40_state { + unsigned scissor_enabled; + unsigned stipple_enabled; + unsigned fp_samplers; - /* Vtxprog resources */ - struct nouveau_resource *vp_exec_heap; - struct nouveau_resource *vp_data_heap; + uint64_t dirty; + struct nouveau_stateobj *hw[NV40_STATE_MAX]; }; struct nv40_context { struct pipe_context pipe; + struct nouveau_winsys *nvws; + struct nv40_screen *screen; + unsigned pctx_id; - struct nv40_channel_context *hw; struct draw_context *draw; int chipset; - uint32_t dirty; - + /* HW state derived from pipe states */ + struct nv40_state state; + unsigned fallback; + + /* Context state */ + unsigned dirty; + struct pipe_scissor_state scissor; + unsigned stipple[32]; + struct pipe_clip_state clip; + struct nv40_vertex_program *vertprog; + struct nv40_fragment_program *fragprog; + struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; + struct nv40_rasterizer_state *rasterizer; + struct nv40_zsa_state *zsa; + struct nv40_blend_state *blend; + struct pipe_blend_color blend_colour; + struct pipe_viewport_state viewport; + struct pipe_framebuffer_state framebuffer; + struct pipe_buffer *idxbuf; + unsigned idxbuf_format; struct nv40_sampler_state *tex_sampler[PIPE_MAX_SAMPLERS]; struct nv40_miptree *tex_miptree[PIPE_MAX_SAMPLERS]; unsigned dirty_samplers; - unsigned fp_samplers; - unsigned vp_samplers; - - struct nouveau_stateobj *so_framebuffer; - struct nouveau_stateobj *so_fragtex[16]; - struct nouveau_stateobj *so_vtxbuf; - struct nouveau_stateobj *so_blend; - struct nouveau_stateobj *so_rast; - struct nouveau_stateobj *so_zsa; - struct nouveau_stateobj *so_bcol; - struct nouveau_stateobj *so_scissor; - struct nouveau_stateobj *so_viewport; - struct nouveau_stateobj *so_stipple; - - struct { - struct nv40_vertex_program *active; - - struct nv40_vertex_program *current; - struct pipe_buffer *constant_buf; - } vertprog; - - struct { - struct nv40_fragment_program *active; - - struct nv40_fragment_program *current; - struct pipe_buffer *constant_buf; - } fragprog; - struct pipe_vertex_buffer vtxbuf[PIPE_ATTRIB_MAX]; struct pipe_vertex_element vtxelt[PIPE_ATTRIB_MAX]; }; @@ -106,27 +147,29 @@ nv40_context(struct pipe_context *pipe) return (struct nv40_context *)pipe; } +struct nv40_state_entry { + boolean (*validate)(struct nv40_context *nv40); + struct { + unsigned pipe; + unsigned hw; + } dirty; +}; + extern void nv40_init_state_functions(struct nv40_context *nv40); extern void nv40_init_surface_functions(struct nv40_context *nv40); extern void nv40_init_miptree_functions(struct nv40_context *nv40); extern void nv40_init_query_functions(struct nv40_context *nv40); +extern void nv40_screen_init_miptree_functions(struct pipe_screen *pscreen); + /* nv40_draw.c */ extern struct draw_stage *nv40_draw_render_stage(struct nv40_context *nv40); /* nv40_vertprog.c */ -extern void nv40_vertprog_translate(struct nv40_context *, - struct nv40_vertex_program *); -extern void nv40_vertprog_bind(struct nv40_context *, - struct nv40_vertex_program *); extern void nv40_vertprog_destroy(struct nv40_context *, struct nv40_vertex_program *); /* nv40_fragprog.c */ -extern void nv40_fragprog_translate(struct nv40_context *, - struct nv40_fragment_program *); -extern void nv40_fragprog_bind(struct nv40_context *, - struct nv40_fragment_program *); extern void nv40_fragprog_destroy(struct nv40_context *, struct nv40_fragment_program *); @@ -136,6 +179,19 @@ extern void nv40_fragtex_bind(struct nv40_context *); /* nv40_state.c and friends */ extern void nv40_emit_hw_state(struct nv40_context *nv40); extern void nv40_state_tex_update(struct nv40_context *nv40); +extern struct nv40_state_entry nv40_state_clip; +extern struct nv40_state_entry nv40_state_rasterizer; +extern struct nv40_state_entry nv40_state_scissor; +extern struct nv40_state_entry nv40_state_stipple; +extern struct nv40_state_entry nv40_state_fragprog; +extern struct nv40_state_entry nv40_state_vertprog; +extern struct nv40_state_entry nv40_state_blend; +extern struct nv40_state_entry nv40_state_blend_colour; +extern struct nv40_state_entry nv40_state_zsa; +extern struct nv40_state_entry nv40_state_viewport; +extern struct nv40_state_entry nv40_state_framebuffer; +extern struct nv40_state_entry nv40_state_fragtex; +extern struct nv40_state_entry nv40_state_vbo; /* nv40_vbo.c */ extern boolean nv40_draw_arrays(struct pipe_context *, unsigned mode, diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index 07a418c1e9..3c4ea7e99e 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -668,7 +668,7 @@ nv40_fragprog_parse_decl_output(struct nv40_fpc *fpc, return TRUE; } -void +static void nv40_fragprog_translate(struct nv40_context *nv40, struct nv40_fragment_program *fp) { @@ -750,28 +750,80 @@ nv40_fragprog_translate(struct nv40_context *nv40, fp->insn[fpc->inst_offset + 3] = 0x00000000; fp->translated = TRUE; - fp->on_hw = FALSE; out_err: tgsi_parse_free(&parse); free(fpc); } -void -nv40_fragprog_bind(struct nv40_context *nv40, struct nv40_fragment_program *fp) +static void +nv40_fragprog_upload(struct nv40_context *nv40, + struct nv40_fragment_program *fp) +{ + struct pipe_winsys *ws = nv40->pipe.winsys; + const uint32_t le = 1; + uint32_t *map; + int i; + + map = ws->buffer_map(ws, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); + +#if 0 + for (i = 0; i < fp->insn_len; i++) { + NOUVEAU_ERR("%d 0x%08x\n", i, fp->insn[i]); + } +#endif + + if ((*(const uint8_t *)&le)) { + for (i = 0; i < fp->insn_len; i++) { + map[i] = fp->insn[i]; + } + } else { + /* Weird swapping for big-endian chips */ + for (i = 0; i < fp->insn_len; i++) { + map[i] = ((fp->insn[i] & 0xffff) << 16) | + ((fp->insn[i] >> 16) & 0xffff); + } + } + + ws->buffer_unmap(ws, fp->buffer); +} + +static boolean +nv40_fragprog_validate(struct nv40_context *nv40) { + struct nv40_fragment_program *fp = nv40->fragprog; + struct pipe_buffer *constbuf = + nv40->constbuf[PIPE_SHADER_FRAGMENT]; struct pipe_winsys *ws = nv40->pipe.winsys; struct nouveau_stateobj *so; int i; + if (fp->translated) + goto update_constants; + + nv40_fragprog_translate(nv40, fp); if (!fp->translated) { - nv40_fragprog_translate(nv40, fp); - if (!fp->translated) - assert(0); + nv40->fallback |= NV40_FALLBACK_RAST; + return FALSE; } + fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); + nv40_fragprog_upload(nv40, fp); + + so = so_new(4, 1); + so_method(so, nv40->screen->curie, NV40TCL_FP_ADDRESS, 1); + so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, + NV40TCL_FP_ADDRESS_DMA0, NV40TCL_FP_ADDRESS_DMA1); + so_method(so, nv40->screen->curie, NV40TCL_FP_CONTROL, 1); + so_data (so, fp->fp_control); + so_ref(so, &fp->so); + +update_constants: if (fp->nr_consts) { - float *map = ws->buffer_map(ws, nv40->fragprog.constant_buf, - PIPE_BUFFER_USAGE_CPU_READ); + boolean new_consts = FALSE; + float *map; + + map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); for (i = 0; i < fp->nr_consts; i++) { struct nv40_fragment_program_data *fpd = &fp->consts[i]; uint32_t *p = &fp->insn[fpd->offset]; @@ -780,56 +832,20 @@ nv40_fragprog_bind(struct nv40_context *nv40, struct nv40_fragment_program *fp) if (!memcmp(p, cb, 4 * sizeof(float))) continue; memcpy(p, cb, 4 * sizeof(float)); - fp->on_hw = 0; + new_consts = TRUE; } - ws->buffer_unmap(ws, nv40->fragprog.constant_buf); - } + ws->buffer_unmap(ws, constbuf); - if (!fp->on_hw) { - const uint32_t le = 1; - uint32_t *map; - - if (!fp->buffer) - fp->buffer = ws->buffer_create(ws, 0x100, 0, - fp->insn_len * 4); - map = ws->buffer_map(ws, fp->buffer, - PIPE_BUFFER_USAGE_CPU_WRITE); - -#if 0 - for (i = 0; i < fp->insn_len; i++) { - NOUVEAU_ERR("%d 0x%08x\n", i, fp->insn[i]); - } -#endif - - if ((*(const uint8_t *)&le)) { - for (i = 0; i < fp->insn_len; i++) { - map[i] = fp->insn[i]; - } - } else { - /* Weird swapping for big-endian chips */ - for (i = 0; i < fp->insn_len; i++) { - map[i] = ((fp->insn[i] & 0xffff) << 16) | - ((fp->insn[i] >> 16) & 0xffff); - } - } - - ws->buffer_unmap(ws, fp->buffer); - fp->on_hw = TRUE; + if (new_consts) + nv40_fragprog_upload(nv40, fp); } - so = so_new(4, 1); - so_method(so, nv40->hw->curie, NV40TCL_FP_ADDRESS, 1); - so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - NV40TCL_FP_ADDRESS_DMA0, NV40TCL_FP_ADDRESS_DMA1); - so_method(so, nv40->hw->curie, NV40TCL_FP_CONTROL, 1); - so_data (so, fp->fp_control); - - so_emit(nv40->nvws, so); - so_ref(so, &fp->so); - so_ref(NULL, &so); + if (fp->so != nv40->state.hw[NV40_STATE_FRAGPROG]) { + so_ref(fp->so, &nv40->state.hw[NV40_STATE_FRAGPROG]); + return TRUE; + } - nv40->fragprog.active = fp; + return FALSE; } void @@ -840,3 +856,11 @@ nv40_fragprog_destroy(struct nv40_context *nv40, free(fp->insn); } +struct nv40_state_entry nv40_state_fragprog = { + .validate = nv40_fragprog_validate, + .dirty = { + .pipe = NV40_NEW_FRAGPROG, + .hw = NV40_STATE_FRAGPROG + } +}; + diff --git a/src/gallium/drivers/nv40/nv40_fragtex.c b/src/gallium/drivers/nv40/nv40_fragtex.c index 5af5fbe746..436f954cec 100644 --- a/src/gallium/drivers/nv40/nv40_fragtex.c +++ b/src/gallium/drivers/nv40/nv40_fragtex.c @@ -41,6 +41,7 @@ static struct nv40_texture_format * nv40_fragtex_format(uint pipe_format) { struct nv40_texture_format *tf = nv40_texture_formats; + char fs[128]; while (tf->defined) { if (tf->pipe == pipe_format) @@ -48,11 +49,13 @@ nv40_fragtex_format(uint pipe_format) tf++; } + pf_sprint_name(fs, pipe_format); + NOUVEAU_ERR("unknown texture format %s\n", fs); return NULL; } -static void +static struct nouveau_stateobj * nv40_fragtex_build(struct nv40_context *nv40, int unit) { struct nv40_sampler_state *ps = nv40->tex_sampler[unit]; @@ -90,7 +93,7 @@ nv40_fragtex_build(struct nv40_context *nv40, int unit) break; default: NOUVEAU_ERR("Unknown target %d\n", pt->target); - return; + return NULL; } if (swizzled) { @@ -103,7 +106,7 @@ nv40_fragtex_build(struct nv40_context *nv40, int unit) txs = tf->swizzle; so = so_new(16, 2); - so_method(so, nv40->hw->curie, NV40TCL_TEX_OFFSET(unit), 8); + so_method(so, nv40->screen->curie, NV40TCL_TEX_OFFSET(unit), 8); so_reloc (so, nv40mt->buffer, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0); so_reloc (so, nv40mt->buffer, txf, tex_flags | NOUVEAU_BO_OR, NV40TCL_TEX_FORMAT_DMA0, NV40TCL_TEX_FORMAT_DMA1); @@ -114,28 +117,30 @@ nv40_fragtex_build(struct nv40_context *nv40, int unit) so_data (so, (pt->width[0] << NV40TCL_TEX_SIZE0_W_SHIFT) | pt->height[0]); so_data (so, ps->bcol); - so_method(so, nv40->hw->curie, NV40TCL_TEX_SIZE1(unit), 1); + so_method(so, nv40->screen->curie, NV40TCL_TEX_SIZE1(unit), 1); so_data (so, (pt->depth[0] << NV40TCL_TEX_SIZE1_DEPTH_SHIFT) | txp); - so_emit(nv40->nvws, so); - so_ref (so, &nv40->so_fragtex[unit]); - so_ref (NULL, &so); + return so; } -void -nv40_fragtex_bind(struct nv40_context *nv40) +static boolean +nv40_fragtex_validate(struct nv40_context *nv40) { - struct nv40_fragment_program *fp = nv40->fragprog.active; + struct nv40_fragment_program *fp = nv40->fragprog; + struct nv40_state *state = &nv40->state; + struct nouveau_stateobj *so; unsigned samplers, unit; - samplers = nv40->fp_samplers & ~fp->samplers; + samplers = state->fp_samplers & ~fp->samplers; while (samplers) { unit = ffs(samplers) - 1; samplers &= ~(1 << unit); - so_ref(NULL, &nv40->so_fragtex[unit]); - BEGIN_RING(curie, NV40TCL_TEX_ENABLE(unit), 1); - OUT_RING (0); + so = so_new(2, 0); + so_method(so, nv40->screen->curie, NV40TCL_TEX_ENABLE(unit), 1); + so_data (so, 0); + so_ref(so, &nv40->state.hw[NV40_STATE_FRAGTEX0 + unit]); + state->dirty |= (1ULL << (NV40_STATE_FRAGTEX0 + unit)); } samplers = nv40->dirty_samplers & fp->samplers; @@ -143,9 +148,20 @@ nv40_fragtex_bind(struct nv40_context *nv40) unit = ffs(samplers) - 1; samplers &= ~(1 << unit); - nv40_fragtex_build(nv40, unit); + so = nv40_fragtex_build(nv40, unit); + so_ref(so, &nv40->state.hw[NV40_STATE_FRAGTEX0 + unit]); + state->dirty |= (1ULL << (NV40_STATE_FRAGTEX0 + unit)); } - nv40->fp_samplers = fp->samplers; + nv40->state.fp_samplers = fp->samplers; + return FALSE; } +struct nv40_state_entry nv40_state_fragtex = { + .validate = nv40_fragtex_validate, + .dirty = { + .pipe = NV40_NEW_SAMPLER | NV40_NEW_FRAGPROG, + .hw = 0 + } +}; + diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c index 92e6b3a43d..94ba05b710 100644 --- a/src/gallium/drivers/nv40/nv40_miptree.c +++ b/src/gallium/drivers/nv40/nv40_miptree.c @@ -54,15 +54,18 @@ nv40_miptree_layout(struct nv40_miptree *nv40mt) } static struct pipe_texture * -nv40_miptree_create(struct pipe_context *pipe, const struct pipe_texture *pt) +nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt) { - struct pipe_winsys *ws = pipe->winsys; + struct pipe_winsys *ws = pscreen->winsys; struct nv40_miptree *mt; mt = MALLOC(sizeof(struct nv40_miptree)); if (!mt) return NULL; mt->base = *pt; + mt->base.refcount = 1; + mt->base.screen = pscreen; + nv40_miptree_layout(mt); mt->buffer = ws->buffer_create(ws, 256, PIPE_BUFFER_USAGE_PIXEL, @@ -76,9 +79,9 @@ nv40_miptree_create(struct pipe_context *pipe, const struct pipe_texture *pt) } static void -nv40_miptree_release(struct pipe_context *pipe, struct pipe_texture **pt) +nv40_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **pt) { - struct pipe_winsys *ws = pipe->winsys; + struct pipe_winsys *ws = pscreen->winsys; struct pipe_texture *mt = *pt; *pt = NULL; @@ -95,10 +98,52 @@ nv40_miptree_release(struct pipe_context *pipe, struct pipe_texture **pt) } } +static void +nv40_miptree_update(struct pipe_context *pipe, struct pipe_texture *mt) +{ +} + +static struct pipe_surface * +nv40_miptree_surface(struct pipe_screen *pscreen, struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice) +{ + struct pipe_winsys *ws = pscreen->winsys; + struct nv40_miptree *nv40mt = (struct nv40_miptree *)pt; + struct pipe_surface *ps; + + ps = ws->surface_alloc(ws); + if (!ps) + return NULL; + pipe_buffer_reference(ws, &ps->buffer, nv40mt->buffer); + ps->format = pt->format; + ps->cpp = pt->cpp; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->pitch = nv40mt->level[level].pitch / ps->cpp; + + if (pt->target == PIPE_TEXTURE_CUBE) { + ps->offset = nv40mt->level[level].image_offset[face]; + } else + if (pt->target == PIPE_TEXTURE_3D) { + ps->offset = nv40mt->level[level].image_offset[zslice]; + } else { + ps->offset = nv40mt->level[level].image_offset[0]; + } + + return ps; +} + void nv40_init_miptree_functions(struct nv40_context *nv40) { - nv40->pipe.texture_create = nv40_miptree_create; - nv40->pipe.texture_release = nv40_miptree_release; + nv40->pipe.texture_update = nv40_miptree_update; +} + +void +nv40_screen_init_miptree_functions(struct pipe_screen *pscreen) +{ + pscreen->texture_create = nv40_miptree_create; + pscreen->texture_release = nv40_miptree_release; + pscreen->get_tex_surface = nv40_miptree_surface; } diff --git a/src/gallium/drivers/nv40/nv40_query.c b/src/gallium/drivers/nv40/nv40_query.c index 8bca2788b9..0317845624 100644 --- a/src/gallium/drivers/nv40/nv40_query.c +++ b/src/gallium/drivers/nv40/nv40_query.c @@ -45,9 +45,9 @@ nv40_query_begin(struct pipe_context *pipe, struct pipe_query *pq) assert(q->type == PIPE_QUERY_OCCLUSION_COUNTER); - if (nv40->nvws->res_alloc(nv40->hw->query_heap, 1, NULL, &q->object)) + if (nv40->nvws->res_alloc(nv40->screen->query_heap, 1, NULL, &q->object)) assert(0); - nv40->nvws->notifier_reset(nv40->hw->query, q->object->start); + nv40->nvws->notifier_reset(nv40->screen->query, q->object->start); BEGIN_RING(curie, NV40TCL_QUERY_RESET, 1); OUT_RING (1); @@ -82,17 +82,17 @@ nv40_query_result(struct pipe_context *pipe, struct pipe_query *pq, if (!q->ready) { unsigned status; - status = nvws->notifier_status(nv40->hw->query, + status = nvws->notifier_status(nv40->screen->query, q->object->start); if (status != NV_NOTIFY_STATE_STATUS_COMPLETED) { if (wait == FALSE) return FALSE; - nvws->notifier_wait(nv40->hw->query, q->object->start, + nvws->notifier_wait(nv40->screen->query, q->object->start, NV_NOTIFY_STATE_STATUS_COMPLETED, 0); } - q->result = nvws->notifier_retval(nv40->hw->query, + q->result = nvws->notifier_retval(nv40->screen->query, q->object->start); q->ready = TRUE; nvws->res_free(&q->object); diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c new file mode 100644 index 0000000000..268ca83ce0 --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_screen.c @@ -0,0 +1,279 @@ +#include "pipe/p_screen.h" +#include "pipe/p_util.h" + +#include "nv40_context.h" +#include "nv40_screen.h" + +#define NV4X_GRCLASS4097_CHIPSETS 0x00000baf +#define NV4X_GRCLASS4497_CHIPSETS 0x00005450 +#define NV6X_GRCLASS4497_CHIPSETS 0x00000088 + +static const char * +nv40_screen_get_name(struct pipe_screen *pscreen) +{ + struct nv40_screen *screen = nv40_screen(pscreen); + static char buffer[128]; + + snprintf(buffer, sizeof(buffer), "NV%02X", screen->chipset); + return buffer; +} + +static const char * +nv40_screen_get_vendor(struct pipe_screen *pscreen) +{ + return "nouveau"; +} + +static int +nv40_screen_get_param(struct pipe_screen *pscreen, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return 16; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 0; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 1; + case PIPE_CAP_POINT_SPRITE: + return 1; + case PIPE_CAP_MAX_RENDER_TARGETS: + return 4; + case PIPE_CAP_OCCLUSION_QUERY: + return 1; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 13; + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 10; + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 13; + default: + NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param); + return 0; + } +} + +static float +nv40_screen_get_paramf(struct pipe_screen *pscreen, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 10.0; + case PIPE_CAP_MAX_POINT_WIDTH: + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 64.0; + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 16.0; + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; + case PIPE_CAP_BITMAP_TEXCOORD_BIAS: + return 0.0; + default: + NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param); + return 0.0; + } +} + +static boolean +nv40_screen_surface_format_supported(struct pipe_screen *pscreen, + enum pipe_format format, uint type) +{ + switch (type) { + case PIPE_SURFACE: + switch (format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_Z24S8_UNORM: + case PIPE_FORMAT_Z16_UNORM: + return TRUE; + default: + break; + } + break; + case PIPE_TEXTURE: + switch (format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_A1R5G5B5_UNORM: + case PIPE_FORMAT_A4R4G4B4_UNORM: + case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_U_L8: + case PIPE_FORMAT_U_A8: + case PIPE_FORMAT_U_I8: + case PIPE_FORMAT_U_A8_L8: + case PIPE_FORMAT_Z16_UNORM: + case PIPE_FORMAT_Z24S8_UNORM: + return TRUE; + default: + break; + } + break; + default: + assert(0); + }; + + return FALSE; +} + +static void +nv40_screen_destroy(struct pipe_screen *pscreen) +{ + struct nv40_screen *screen = nv40_screen(pscreen); + struct nouveau_winsys *nvws = screen->nvws; + + nvws->res_free(&screen->vp_exec_heap); + nvws->res_free(&screen->vp_data_heap); + nvws->res_free(&screen->query_heap); + nvws->notifier_free(&screen->query); + nvws->notifier_free(&screen->sync); + nvws->grobj_free(&screen->curie); + + FREE(pscreen); +} + +struct pipe_screen * +nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws, + unsigned chipset) +{ + struct nv40_screen *screen = CALLOC_STRUCT(nv40_screen); + struct nouveau_stateobj *so; + unsigned curie_class; + int ret; + + if (!screen) + return NULL; + screen->chipset = chipset; + screen->nvws = nvws; + + /* 3D object */ + switch (chipset & 0xf0) { + case 0x40: + if (NV4X_GRCLASS4097_CHIPSETS & (1 << (chipset & 0x0f))) + curie_class = NV40TCL; + else + if (NV4X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) + curie_class = NV44TCL; + break; + case 0x60: + if (NV6X_GRCLASS4497_CHIPSETS & (1 << (chipset & 0x0f))) + curie_class = NV44TCL; + break; + default: + break; + } + + if (!curie_class) { + NOUVEAU_ERR("Unknown nv4x chipset: nv%02x\n", chipset); + return NULL; + } + + ret = nvws->grobj_alloc(nvws, curie_class, &screen->curie); + if (ret) { + NOUVEAU_ERR("Error creating 3D object: %d\n", ret); + return FALSE; + } + + /* Notifier for sync purposes */ + ret = nvws->notifier_alloc(nvws, 1, &screen->sync); + if (ret) { + NOUVEAU_ERR("Error creating notifier object: %d\n", ret); + nv40_screen_destroy(&screen->pipe); + return NULL; + } + + /* Query objects */ + ret = nvws->notifier_alloc(nvws, 32, &screen->query); + if (ret) { + NOUVEAU_ERR("Error initialising query objects: %d\n", ret); + nv40_screen_destroy(&screen->pipe); + return NULL; + } + + ret = nvws->res_init(&screen->query_heap, 0, 32); + if (ret) { + NOUVEAU_ERR("Error initialising query object heap: %d\n", ret); + nv40_screen_destroy(&screen->pipe); + return NULL; + } + + /* Vtxprog resources */ + if (nvws->res_init(&screen->vp_exec_heap, 0, 512) || + nvws->res_init(&screen->vp_data_heap, 0, 256)) { + nv40_screen_destroy(&screen->pipe); + return NULL; + } + + /* Static curie initialisation */ + so = so_new(128, 0); + so_method(so, screen->curie, NV40TCL_DMA_NOTIFY, 1); + so_data (so, screen->sync->handle); + so_method(so, screen->curie, NV40TCL_DMA_TEXTURE0, 2); + so_data (so, nvws->channel->vram->handle); + so_data (so, nvws->channel->gart->handle); + so_method(so, screen->curie, NV40TCL_DMA_COLOR1, 1); + so_data (so, nvws->channel->vram->handle); + so_method(so, screen->curie, NV40TCL_DMA_COLOR0, 2); + so_data (so, nvws->channel->vram->handle); + so_data (so, nvws->channel->vram->handle); + so_method(so, screen->curie, NV40TCL_DMA_VTXBUF0, 2); + so_data (so, nvws->channel->vram->handle); + so_data (so, nvws->channel->gart->handle); + so_method(so, screen->curie, NV40TCL_DMA_FENCE, 2); + so_data (so, 0); + so_data (so, screen->query->handle); + so_method(so, screen->curie, NV40TCL_DMA_UNK01AC, 2); + so_data (so, nvws->channel->vram->handle); + so_data (so, nvws->channel->vram->handle); + so_method(so, screen->curie, NV40TCL_DMA_COLOR2, 2); + so_data (so, nvws->channel->vram->handle); + so_data (so, nvws->channel->vram->handle); + + so_method(so, screen->curie, 0x1ea4, 3); + so_data (so, 0x00000010); + so_data (so, 0x01000100); + so_data (so, 0xff800006); + + /* vtxprog output routing */ + so_method(so, screen->curie, 0x1fc4, 1); + so_data (so, 0x06144321); + so_method(so, screen->curie, 0x1fc8, 2); + so_data (so, 0xedcba987); + so_data (so, 0x00000021); + so_method(so, screen->curie, 0x1fd0, 1); + so_data (so, 0x00171615); + so_method(so, screen->curie, 0x1fd4, 1); + so_data (so, 0x001b1a19); + + so_method(so, screen->curie, 0x1ef8, 1); + so_data (so, 0x0020ffff); + so_method(so, screen->curie, 0x1d64, 1); + so_data (so, 0x00d30000); + so_method(so, screen->curie, 0x1e94, 1); + so_data (so, 0x00000001); + + so_emit(nvws, so); + so_ref(NULL, &so); + nvws->push_flush(nvws->channel, 0); + + screen->pipe.winsys = ws; + screen->pipe.destroy = nv40_screen_destroy; + + screen->pipe.get_name = nv40_screen_get_name; + screen->pipe.get_vendor = nv40_screen_get_vendor; + screen->pipe.get_param = nv40_screen_get_param; + screen->pipe.get_paramf = nv40_screen_get_paramf; + + screen->pipe.is_format_supported = nv40_screen_surface_format_supported; + + nv40_screen_init_miptree_functions(&screen->pipe); + + return &screen->pipe; +} + diff --git a/src/gallium/drivers/nv40/nv40_screen.h b/src/gallium/drivers/nv40/nv40_screen.h new file mode 100644 index 0000000000..3ea78aadfd --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_screen.h @@ -0,0 +1,36 @@ +#ifndef __NV40_SCREEN_H__ +#define __NV40_SCREEN_H__ + +#include "pipe/p_screen.h" + +struct nv40_screen { + struct pipe_screen pipe; + + struct nouveau_winsys *nvws; + unsigned chipset; + + unsigned cur_pctx; + + /* HW graphics objects */ + struct nouveau_grobj *curie; + struct nouveau_notifier *sync; + + /* Query object resources */ + struct nouveau_notifier *query; + struct nouveau_resource *query_heap; + + /* Vtxprog resources */ + struct nouveau_resource *vp_exec_heap; + struct nouveau_resource *vp_data_heap; + + /* Current 3D state of channel */ + struct nouveau_stateobj *state[NV40_STATE_MAX]; +}; + +static INLINE struct nv40_screen * +nv40_screen(struct pipe_screen *screen) +{ + return (struct nv40_screen *)screen; +} + +#endif diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c index 713f31dbb1..caa2f9df0c 100644 --- a/src/gallium/drivers/nv40/nv40_state.c +++ b/src/gallium/drivers/nv40/nv40_state.c @@ -10,7 +10,8 @@ nv40_blend_state_create(struct pipe_context *pipe, const struct pipe_blend_state *cso) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_grobj *curie = nv40->hw->curie; + struct nouveau_grobj *curie = nv40->screen->curie; + struct nv40_blend_state *bso = CALLOC(1, sizeof(*bso)); struct nouveau_stateobj *so = so_new(16, 0); if (cso->blend_enable) { @@ -46,7 +47,9 @@ nv40_blend_state_create(struct pipe_context *pipe, so_method(so, curie, NV40TCL_DITHER_ENABLE, 1); so_data (so, cso->dither ? 1 : 0); - return (void *)so; + so_ref(so, &bso->so); + bso->pipe = *cso; + return (void *)bso; } static void @@ -54,16 +57,17 @@ nv40_blend_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv40_context *nv40 = nv40_context(pipe); - so_ref(hwcso, &nv40->so_blend); + nv40->blend = hwcso; nv40->dirty |= NV40_NEW_BLEND; } static void nv40_blend_state_delete(struct pipe_context *pipe, void *hwcso) { - struct nouveau_stateobj *so = hwcso; + struct nv40_blend_state *bso = hwcso; - so_ref(NULL, &so); + so_ref(NULL, &bso->so); + FREE(bso); } @@ -255,12 +259,13 @@ nv40_sampler_state_bind(struct pipe_context *pipe, unsigned unit, nv40->tex_sampler[unit] = ps; nv40->dirty_samplers |= (1 << unit); + nv40->dirty |= NV40_NEW_SAMPLER; } static void nv40_sampler_state_delete(struct pipe_context *pipe, void *hwcso) { - free(hwcso); + FREE(hwcso); } static void @@ -271,6 +276,7 @@ nv40_set_sampler_texture(struct pipe_context *pipe, unsigned unit, nv40->tex_miptree[unit] = (struct nv40_miptree *)miptree; nv40->dirty_samplers |= (1 << unit); + nv40->dirty |= NV40_NEW_SAMPLER; } static void * @@ -278,33 +284,32 @@ nv40_rasterizer_state_create(struct pipe_context *pipe, const struct pipe_rasterizer_state *cso) { struct nv40_context *nv40 = nv40_context(pipe); + struct nv40_rasterizer_state *rsso = CALLOC(1, sizeof(*rsso)); struct nouveau_stateobj *so = so_new(32, 0); + struct nouveau_grobj *curie = nv40->screen->curie; /*XXX: ignored: * light_twoside - * offset_cw/ccw -nohw - * scissor * point_smooth -nohw * multisample - * offset_units / offset_scale */ - so_method(so, nv40->hw->curie, NV40TCL_SHADE_MODEL, 1); + so_method(so, curie, NV40TCL_SHADE_MODEL, 1); so_data (so, cso->flatshade ? NV40TCL_SHADE_MODEL_FLAT : NV40TCL_SHADE_MODEL_SMOOTH); - so_method(so, nv40->hw->curie, NV40TCL_LINE_WIDTH, 2); + so_method(so, curie, NV40TCL_LINE_WIDTH, 2); so_data (so, (unsigned char)(cso->line_width * 8.0) & 0xff); so_data (so, cso->line_smooth ? 1 : 0); - so_method(so, nv40->hw->curie, NV40TCL_LINE_STIPPLE_ENABLE, 2); + so_method(so, curie, NV40TCL_LINE_STIPPLE_ENABLE, 2); so_data (so, cso->line_stipple_enable ? 1 : 0); so_data (so, (cso->line_stipple_pattern << 16) | cso->line_stipple_factor); - so_method(so, nv40->hw->curie, NV40TCL_POINT_SIZE, 1); + so_method(so, curie, NV40TCL_POINT_SIZE, 1); so_data (so, fui(cso->point_size)); - so_method(so, nv40->hw->curie, NV40TCL_POLYGON_MODE_FRONT, 6); + so_method(so, curie, NV40TCL_POLYGON_MODE_FRONT, 6); if (cso->front_winding == PIPE_WINDING_CCW) { so_data(so, nvgl_polygon_mode(cso->fill_ccw)); so_data(so, nvgl_polygon_mode(cso->fill_cw)); @@ -345,10 +350,32 @@ nv40_rasterizer_state_create(struct pipe_context *pipe, so_data(so, cso->poly_smooth ? 1 : 0); so_data(so, cso->cull_mode != PIPE_WINDING_NONE ? 1 : 0); - so_method(so, nv40->hw->curie, NV40TCL_POLYGON_STIPPLE_ENABLE, 1); + so_method(so, curie, NV40TCL_POLYGON_STIPPLE_ENABLE, 1); so_data (so, cso->poly_stipple_enable ? 1 : 0); - so_method(so, nv40->hw->curie, NV40TCL_POINT_SPRITE, 1); + so_method(so, curie, NV40TCL_POLYGON_OFFSET_POINT_ENABLE, 3); + if ((cso->offset_cw && cso->fill_cw == PIPE_POLYGON_MODE_POINT) || + (cso->offset_ccw && cso->fill_ccw == PIPE_POLYGON_MODE_POINT)) + so_data(so, 1); + else + so_data(so, 0); + if ((cso->offset_cw && cso->fill_cw == PIPE_POLYGON_MODE_LINE) || + (cso->offset_ccw && cso->fill_ccw == PIPE_POLYGON_MODE_LINE)) + so_data(so, 1); + else + so_data(so, 0); + if ((cso->offset_cw && cso->fill_cw == PIPE_POLYGON_MODE_FILL) || + (cso->offset_ccw && cso->fill_ccw == PIPE_POLYGON_MODE_FILL)) + so_data(so, 1); + else + so_data(so, 0); + if (cso->offset_cw || cso->offset_ccw) { + so_method(so, curie, NV40TCL_POLYGON_OFFSET_FACTOR, 2); + so_data (so, fui(cso->offset_scale)); + so_data (so, fui(cso->offset_units * 2)); + } + + so_method(so, curie, NV40TCL_POINT_SPRITE, 1); if (cso->point_sprite) { unsigned psctl = (1 << 0), i; @@ -362,7 +389,9 @@ nv40_rasterizer_state_create(struct pipe_context *pipe, so_data(so, 0); } - return (void *)so; + so_ref(so, &rsso->so); + rsso->pipe = *cso; + return (void *)rsso; } static void @@ -370,16 +399,17 @@ nv40_rasterizer_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv40_context *nv40 = nv40_context(pipe); - so_ref(hwcso, &nv40->so_rast); + nv40->rasterizer = hwcso; nv40->dirty |= NV40_NEW_RAST; } static void nv40_rasterizer_state_delete(struct pipe_context *pipe, void *hwcso) { - struct nouveau_stateobj *so = hwcso; + struct nv40_rasterizer_state *rsso = hwcso; - so_ref(NULL, &so); + so_ref(NULL, &rsso->so); + FREE(rsso); } static void * @@ -387,20 +417,21 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe, const struct pipe_depth_stencil_alpha_state *cso) { struct nv40_context *nv40 = nv40_context(pipe); + struct nv40_zsa_state *zsaso = CALLOC(1, sizeof(*zsaso)); struct nouveau_stateobj *so = so_new(32, 0); - so_method(so, nv40->hw->curie, NV40TCL_DEPTH_FUNC, 3); + so_method(so, nv40->screen->curie, NV40TCL_DEPTH_FUNC, 3); so_data (so, nvgl_comparison_op(cso->depth.func)); so_data (so, cso->depth.writemask ? 1 : 0); so_data (so, cso->depth.enabled ? 1 : 0); - so_method(so, nv40->hw->curie, NV40TCL_ALPHA_TEST_ENABLE, 3); + so_method(so, nv40->screen->curie, NV40TCL_ALPHA_TEST_ENABLE, 3); so_data (so, cso->alpha.enabled ? 1 : 0); so_data (so, nvgl_comparison_op(cso->alpha.func)); so_data (so, float_to_ubyte(cso->alpha.ref)); if (cso->stencil[0].enabled) { - so_method(so, nv40->hw->curie, NV40TCL_STENCIL_FRONT_ENABLE, 8); + so_method(so, nv40->screen->curie, NV40TCL_STENCIL_FRONT_ENABLE, 8); so_data (so, cso->stencil[0].enabled ? 1 : 0); so_data (so, cso->stencil[0].write_mask); so_data (so, nvgl_comparison_op(cso->stencil[0].func)); @@ -410,12 +441,12 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe, so_data (so, nvgl_stencil_op(cso->stencil[0].zfail_op)); so_data (so, nvgl_stencil_op(cso->stencil[0].zpass_op)); } else { - so_method(so, nv40->hw->curie, NV40TCL_STENCIL_FRONT_ENABLE, 1); + so_method(so, nv40->screen->curie, NV40TCL_STENCIL_FRONT_ENABLE, 1); so_data (so, 0); } if (cso->stencil[1].enabled) { - so_method(so, nv40->hw->curie, NV40TCL_STENCIL_BACK_ENABLE, 8); + so_method(so, nv40->screen->curie, NV40TCL_STENCIL_BACK_ENABLE, 8); so_data (so, cso->stencil[1].enabled ? 1 : 0); so_data (so, cso->stencil[1].write_mask); so_data (so, nvgl_comparison_op(cso->stencil[1].func)); @@ -425,11 +456,13 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe, so_data (so, nvgl_stencil_op(cso->stencil[1].zfail_op)); so_data (so, nvgl_stencil_op(cso->stencil[1].zpass_op)); } else { - so_method(so, nv40->hw->curie, NV40TCL_STENCIL_BACK_ENABLE, 1); + so_method(so, nv40->screen->curie, NV40TCL_STENCIL_BACK_ENABLE, 1); so_data (so, 0); } - return (void *)so; + so_ref(so, &zsaso->so); + zsaso->pipe = *cso; + return (void *)zsaso; } static void @@ -437,16 +470,17 @@ nv40_depth_stencil_alpha_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv40_context *nv40 = nv40_context(pipe); - so_ref(hwcso, &nv40->so_zsa); + nv40->zsa = hwcso; nv40->dirty |= NV40_NEW_ZSA; } static void nv40_depth_stencil_alpha_state_delete(struct pipe_context *pipe, void *hwcso) { - struct nouveau_stateobj *so = hwcso; + struct nv40_zsa_state *zsaso = hwcso; - so_ref(NULL, &so); + so_ref(NULL, &zsaso->so); + FREE(zsaso); } static void * @@ -465,9 +499,8 @@ static void nv40_vp_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv40_context *nv40 = nv40_context(pipe); - struct nv40_vertex_program *vp = hwcso; - nv40->vertprog.current = vp; + nv40->vertprog = hwcso; nv40->dirty |= NV40_NEW_VERTPROG; } @@ -478,7 +511,7 @@ nv40_vp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv40_vertex_program *vp = hwcso; nv40_vertprog_destroy(nv40, vp); - free(vp); + FREE(vp); } static void * @@ -497,9 +530,8 @@ static void nv40_fp_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv40_context *nv40 = nv40_context(pipe); - struct nv40_fragment_program *fp = hwcso; - nv40->fragprog.current = fp; + nv40->fragprog = hwcso; nv40->dirty |= NV40_NEW_FRAGPROG; } @@ -510,7 +542,7 @@ nv40_fp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv40_fragment_program *fp = hwcso; nv40_fragprog_destroy(nv40, fp); - free(fp); + FREE(fp); } static void @@ -518,16 +550,8 @@ nv40_set_blend_color(struct pipe_context *pipe, const struct pipe_blend_color *bcol) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_stateobj *so = so_new(2, 0); - - so_method(so, nv40->hw->curie, NV40TCL_BLEND_COLOR, 1); - so_data (so, ((float_to_ubyte(bcol->color[3]) << 24) | - (float_to_ubyte(bcol->color[0]) << 16) | - (float_to_ubyte(bcol->color[1]) << 8) | - (float_to_ubyte(bcol->color[2]) << 0))); - so_ref(so, &nv40->so_bcol); - so_ref(NULL, &so); + nv40->blend_colour = *bcol; nv40->dirty |= NV40_NEW_BCOL; } @@ -535,6 +559,10 @@ static void nv40_set_clip_state(struct pipe_context *pipe, const struct pipe_clip_state *clip) { + struct nv40_context *nv40 = nv40_context(pipe); + + nv40->clip = *clip; + nv40->dirty |= NV40_NEW_UCP; } static void @@ -544,11 +572,11 @@ nv40_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, struct nv40_context *nv40 = nv40_context(pipe); if (shader == PIPE_SHADER_VERTEX) { - nv40->vertprog.constant_buf = buf->buffer; + nv40->constbuf[PIPE_SHADER_VERTEX] = buf->buffer; nv40->dirty |= NV40_NEW_VERTPROG; } else if (shader == PIPE_SHADER_FRAGMENT) { - nv40->fragprog.constant_buf = buf->buffer; + nv40->constbuf[PIPE_SHADER_FRAGMENT] = buf->buffer; nv40->dirty |= NV40_NEW_FRAGPROG; } } @@ -558,146 +586,8 @@ nv40_set_framebuffer_state(struct pipe_context *pipe, const struct pipe_framebuffer_state *fb) { struct nv40_context *nv40 = nv40_context(pipe); - struct pipe_surface *rt[4], *zeta; - uint32_t rt_enable, rt_format, w, h; - int i, colour_format = 0, zeta_format = 0; - struct nouveau_stateobj *so = so_new(64, 10); - unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM; - - rt_enable = 0; - for (i = 0; i < 4; i++) { - if (!fb->cbufs[i]) - continue; - - if (colour_format) { - assert(w == fb->cbufs[i]->width); - assert(h == fb->cbufs[i]->height); - assert(colour_format == fb->cbufs[i]->format); - } else { - w = fb->cbufs[i]->width; - h = fb->cbufs[i]->height; - colour_format = fb->cbufs[i]->format; - rt_enable |= (NV40TCL_RT_ENABLE_COLOR0 << i); - rt[i] = fb->cbufs[i]; - } - } - - if (rt_enable & (NV40TCL_RT_ENABLE_COLOR1 | NV40TCL_RT_ENABLE_COLOR2 | - NV40TCL_RT_ENABLE_COLOR3)) - rt_enable |= NV40TCL_RT_ENABLE_MRT; - - if (fb->zsbuf) { - if (colour_format) { - assert(w == fb->zsbuf->width); - assert(h == fb->zsbuf->height); - } else { - w = fb->zsbuf->width; - h = fb->zsbuf->height; - } - zeta_format = fb->zsbuf->format; - zeta = fb->zsbuf; - } - - rt_format = NV40TCL_RT_FORMAT_TYPE_LINEAR; - - switch (colour_format) { - case PIPE_FORMAT_A8R8G8B8_UNORM: - case 0: - rt_format |= NV40TCL_RT_FORMAT_COLOR_A8R8G8B8; - break; - case PIPE_FORMAT_R5G6B5_UNORM: - rt_format |= NV40TCL_RT_FORMAT_COLOR_R5G6B5; - break; - default: - assert(0); - } - - switch (zeta_format) { - case PIPE_FORMAT_Z16_UNORM: - rt_format |= NV40TCL_RT_FORMAT_ZETA_Z16; - break; - case PIPE_FORMAT_Z24S8_UNORM: - case 0: - rt_format |= NV40TCL_RT_FORMAT_ZETA_Z24S8; - break; - default: - assert(0); - } - - if (rt_enable & NV40TCL_RT_ENABLE_COLOR0) { - so_method(so, nv40->hw->curie, NV40TCL_DMA_COLOR0, 1); - so_reloc (so, rt[0]->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->hw->curie, NV40TCL_COLOR0_PITCH, 2); - so_data (so, rt[0]->pitch * rt[0]->cpp); - so_reloc (so, rt[0]->buffer, rt[0]->offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - } - - if (rt_enable & NV40TCL_RT_ENABLE_COLOR1) { - so_method(so, nv40->hw->curie, NV40TCL_DMA_COLOR1, 1); - so_reloc (so, rt[1]->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->hw->curie, NV40TCL_COLOR1_OFFSET, 2); - so_reloc (so, rt[1]->buffer, rt[1]->offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_data (so, rt[1]->pitch * rt[1]->cpp); - } - - if (rt_enable & NV40TCL_RT_ENABLE_COLOR2) { - so_method(so, nv40->hw->curie, NV40TCL_DMA_COLOR2, 1); - so_reloc (so, rt[2]->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->hw->curie, NV40TCL_COLOR2_OFFSET, 1); - so_reloc (so, rt[2]->buffer, rt[2]->offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->hw->curie, NV40TCL_COLOR2_PITCH, 1); - so_data (so, rt[2]->pitch * rt[2]->cpp); - } - - if (rt_enable & NV40TCL_RT_ENABLE_COLOR3) { - so_method(so, nv40->hw->curie, NV40TCL_DMA_COLOR3, 1); - so_reloc (so, rt[3]->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->hw->curie, NV40TCL_COLOR3_OFFSET, 1); - so_reloc (so, rt[3]->buffer, rt[3]->offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->hw->curie, NV40TCL_COLOR3_PITCH, 1); - so_data (so, rt[3]->pitch * rt[3]->cpp); - } - - if (zeta_format) { - so_method(so, nv40->hw->curie, NV40TCL_DMA_ZETA, 1); - so_reloc (so, zeta->buffer, 0, rt_flags | NOUVEAU_BO_OR, - nv40->nvws->channel->vram->handle, - nv40->nvws->channel->gart->handle); - so_method(so, nv40->hw->curie, NV40TCL_ZETA_OFFSET, 1); - so_reloc (so, zeta->buffer, zeta->offset, rt_flags | - NOUVEAU_BO_LOW, 0, 0); - so_method(so, nv40->hw->curie, NV40TCL_ZETA_PITCH, 1); - so_data (so, zeta->pitch * zeta->cpp); - } - - so_method(so, nv40->hw->curie, NV40TCL_RT_ENABLE, 1); - so_data (so, rt_enable); - so_method(so, nv40->hw->curie, NV40TCL_RT_HORIZ, 3); - so_data (so, (w << 16) | 0); - so_data (so, (h << 16) | 0); - so_data (so, rt_format); - so_method(so, nv40->hw->curie, NV40TCL_VIEWPORT_HORIZ, 2); - so_data (so, (w << 16) | 0); - so_data (so, (h << 16) | 0); - so_method(so, nv40->hw->curie, NV40TCL_VIEWPORT_CLIP_HORIZ(0), 2); - so_data (so, ((w - 1) << 16) | 0); - so_data (so, ((h - 1) << 16) | 0); - - so_ref(so, &nv40->so_framebuffer); - so_ref(NULL, &so); + nv40->framebuffer = *fb; nv40->dirty |= NV40_NEW_FB; } @@ -706,15 +596,8 @@ nv40_set_polygon_stipple(struct pipe_context *pipe, const struct pipe_poly_stipple *stipple) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_stateobj *so = so_new(33, 0); - unsigned i; - so_method(so, nv40->hw->curie, NV40TCL_POLYGON_STIPPLE_PATTERN(0), 32); - for (i = 0; i < 32; i++) - so_data(so, stipple->stipple[i]); - - so_ref(so, &nv40->so_stipple); - so_ref(NULL, &so); + memcpy(nv40->stipple, stipple->stipple, 4 * 32); nv40->dirty |= NV40_NEW_STIPPLE; } @@ -723,14 +606,8 @@ nv40_set_scissor_state(struct pipe_context *pipe, const struct pipe_scissor_state *s) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_stateobj *so = so_new(3, 0); - - so_method(so, nv40->hw->curie, NV40TCL_SCISSOR_HORIZ, 2); - so_data (so, ((s->maxx - s->minx) << 16) | s->minx); - so_data (so, ((s->maxy - s->miny) << 16) | s->miny); - so_ref(so, &nv40->so_scissor); - so_ref(NULL, &so); + nv40->scissor = *s; nv40->dirty |= NV40_NEW_SCISSOR; } @@ -739,20 +616,8 @@ nv40_set_viewport_state(struct pipe_context *pipe, const struct pipe_viewport_state *vpt) { struct nv40_context *nv40 = nv40_context(pipe); - struct nouveau_stateobj *so = so_new(9, 0); - - so_method(so, nv40->hw->curie, NV40TCL_VIEWPORT_TRANSLATE_X, 8); - so_data (so, fui(vpt->translate[0])); - so_data (so, fui(vpt->translate[1])); - so_data (so, fui(vpt->translate[2])); - so_data (so, fui(vpt->translate[3])); - so_data (so, fui(vpt->scale[0])); - so_data (so, fui(vpt->scale[1])); - so_data (so, fui(vpt->scale[2])); - so_data (so, fui(vpt->scale[3])); - - so_ref(so, &nv40->so_viewport); - so_ref(NULL, &so); + + nv40->viewport = *vpt; nv40->dirty |= NV40_NEW_VIEWPORT; } @@ -763,7 +628,6 @@ nv40_set_vertex_buffer(struct pipe_context *pipe, unsigned index, struct nv40_context *nv40 = nv40_context(pipe); nv40->vtxbuf[index] = *vb; - nv40->dirty |= NV40_NEW_ARRAYS; } @@ -774,7 +638,6 @@ nv40_set_vertex_element(struct pipe_context *pipe, unsigned index, struct nv40_context *nv40 = nv40_context(pipe); nv40->vtxelt[index] = *ve; - nv40->dirty |= NV40_NEW_ARRAYS; } diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h index e82ab9de98..e5217fe91c 100644 --- a/src/gallium/drivers/nv40/nv40_state.h +++ b/src/gallium/drivers/nv40/nv40_state.h @@ -39,6 +39,7 @@ struct nv40_vertex_program { uint32_t ir; uint32_t or; + struct nouveau_stateobj *so; }; struct nv40_fragment_program_data { @@ -50,7 +51,6 @@ struct nv40_fragment_program { const struct pipe_shader_state *pipe; boolean translated; - boolean on_hw; unsigned samplers; uint32_t *insn; diff --git a/src/gallium/drivers/nv40/nv40_state_blend.c b/src/gallium/drivers/nv40/nv40_state_blend.c new file mode 100644 index 0000000000..95e6d7394f --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_blend.c @@ -0,0 +1,40 @@ +#include "nv40_context.h" + +static boolean +nv40_state_blend_validate(struct nv40_context *nv40) +{ + so_ref(nv40->blend->so, &nv40->state.hw[NV40_STATE_BLEND]); + return TRUE; +} + +struct nv40_state_entry nv40_state_blend = { + .validate = nv40_state_blend_validate, + .dirty = { + .pipe = NV40_NEW_BLEND, + .hw = NV40_STATE_BLEND + } +}; + +static boolean +nv40_state_blend_colour_validate(struct nv40_context *nv40) +{ + struct nouveau_stateobj *so = so_new(2, 0); + struct pipe_blend_color *bcol = &nv40->blend_colour; + + so_method(so, nv40->screen->curie, NV40TCL_BLEND_COLOR, 1); + so_data (so, ((float_to_ubyte(bcol->color[3]) << 24) | + (float_to_ubyte(bcol->color[0]) << 16) | + (float_to_ubyte(bcol->color[1]) << 8) | + (float_to_ubyte(bcol->color[2]) << 0))); + + so_ref(so, &nv40->state.hw[NV40_STATE_BCOL]); + return TRUE; +} + +struct nv40_state_entry nv40_state_blend_colour = { + .validate = nv40_state_blend_colour_validate, + .dirty = { + .pipe = NV40_NEW_BCOL, + .hw = NV40_STATE_BCOL + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_clip.c b/src/gallium/drivers/nv40/nv40_state_clip.c new file mode 100644 index 0000000000..93e690161f --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_clip.c @@ -0,0 +1,18 @@ +#include "nv40_context.h" + +static boolean +nv40_state_clip_validate(struct nv40_context *nv40) +{ + if (nv40->clip.nr) + nv40->fallback |= NV40_FALLBACK_TNL; + + return FALSE; +} + +struct nv40_state_entry nv40_state_clip = { + .validate = nv40_state_clip_validate, + .dirty = { + .pipe = NV40_NEW_UCP, + .hw = 0 + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_emit.c b/src/gallium/drivers/nv40/nv40_state_emit.c index a10c995548..9f268640e0 100644 --- a/src/gallium/drivers/nv40/nv40_state_emit.c +++ b/src/gallium/drivers/nv40/nv40_state_emit.c @@ -1,77 +1,108 @@ #include "nv40_context.h" #include "nv40_state.h" -/* Emit relocs for every referenced buffer. - * - * This is to ensure the bufmgr has an accurate idea of how - * the buffer is used. These relocs appear in the push buffer as - * NOPs, and will only be turned into state changes if a buffer - * actually moves. - */ +static struct nv40_state_entry *render_states[] = { + &nv40_state_framebuffer, + &nv40_state_rasterizer, + &nv40_state_clip, + &nv40_state_scissor, + &nv40_state_stipple, + &nv40_state_fragprog, + &nv40_state_fragtex, + &nv40_state_vertprog, + &nv40_state_blend, + &nv40_state_blend_colour, + &nv40_state_zsa, + &nv40_state_viewport, + &nv40_state_vbo, + NULL +}; + static void -nv40_state_emit_dummy_relocs(struct nv40_context *nv40) +nv40_state_validate(struct nv40_context *nv40) { - unsigned i; - - so_emit_reloc_markers(nv40->nvws, nv40->so_framebuffer); - for (i = 0; i < 16; i++) { - if (!(nv40->fp_samplers & (1 << i))) - continue; - so_emit_reloc_markers(nv40->nvws, nv40->so_fragtex[i]); - } - so_emit_reloc_markers(nv40->nvws, nv40->fragprog.active->so); -} + struct nv40_state_entry **states = render_states; + unsigned last_fallback; -void -nv40_emit_hw_state(struct nv40_context *nv40) -{ - if (nv40->dirty & NV40_NEW_FB) - so_emit(nv40->nvws, nv40->so_framebuffer); + last_fallback = nv40->fallback; + nv40->fallback = 0; - if (nv40->dirty & NV40_NEW_BLEND) - so_emit(nv40->nvws, nv40->so_blend); + while (*states) { + struct nv40_state_entry *e = *states; - if (nv40->dirty & NV40_NEW_RAST) - so_emit(nv40->nvws, nv40->so_rast); + if (nv40->dirty & e->dirty.pipe) { + if (e->validate(nv40)) + nv40->state.dirty |= (1ULL << e->dirty.hw); + } - if (nv40->dirty & NV40_NEW_ZSA) - so_emit(nv40->nvws, nv40->so_zsa); + states++; + } + nv40->dirty = 0; - if (nv40->dirty & NV40_NEW_BCOL) - so_emit(nv40->nvws, nv40->so_bcol); + if (nv40->fallback & NV40_FALLBACK_TNL && + !(last_fallback & NV40_FALLBACK_TNL)) { + NOUVEAU_ERR("XXX: hwtnl->swtnl\n"); + } else + if (last_fallback & NV40_FALLBACK_TNL && + !(nv40->fallback & NV40_FALLBACK_TNL)) { + NOUVEAU_ERR("XXX: swtnl->hwtnl\n"); + } - if (nv40->dirty & NV40_NEW_SCISSOR) - so_emit(nv40->nvws, nv40->so_scissor); + if (nv40->fallback & NV40_FALLBACK_RAST && + !(last_fallback & NV40_FALLBACK_RAST)) { + NOUVEAU_ERR("XXX: hwrast->swrast\n"); + } else + if (last_fallback & NV40_FALLBACK_RAST && + !(nv40->fallback & NV40_FALLBACK_RAST)) { + NOUVEAU_ERR("XXX: swrast->hwrast\n"); + } +} - if (nv40->dirty & NV40_NEW_VIEWPORT) - so_emit(nv40->nvws, nv40->so_viewport); +static void +nv40_state_emit(struct nv40_context *nv40) +{ + struct nv40_state *state = &nv40->state; + struct nv40_screen *screen = nv40->screen; + unsigned i, samplers; - if (nv40->dirty & NV40_NEW_STIPPLE) - so_emit(nv40->nvws, nv40->so_stipple); + if (nv40->pctx_id != screen->cur_pctx) { + for (i = 0; i < NV40_STATE_MAX; i++) { + if (state->hw[i] && screen->state[i] != state->hw[i]) + state->dirty |= (1ULL << i); + } - if (nv40->dirty & NV40_NEW_FRAGPROG) { - nv40_fragprog_bind(nv40, nv40->fragprog.current); - /*XXX: clear NV40_NEW_FRAGPROG if no new program uploaded */ + screen->cur_pctx = nv40->pctx_id; } - if (nv40->dirty_samplers || (nv40->dirty & NV40_NEW_FRAGPROG)) { - nv40_fragtex_bind(nv40); + while (state->dirty) { + unsigned idx = ffsll(state->dirty) - 1; - BEGIN_RING(curie, NV40TCL_TEX_CACHE_CTL, 1); - OUT_RING (2); - BEGIN_RING(curie, NV40TCL_TEX_CACHE_CTL, 1); - OUT_RING (1); - nv40->dirty &= ~NV40_NEW_FRAGPROG; + so_ref (state->hw[idx], &nv40->screen->state[idx]); + so_emit(nv40->nvws, nv40->screen->state[idx]); + state->dirty &= ~(1ULL << idx); } - if (nv40->dirty & NV40_NEW_VERTPROG) { - nv40_vertprog_bind(nv40, nv40->vertprog.current); - nv40->dirty &= ~NV40_NEW_VERTPROG; + so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FB]); + for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) { + if (!(samplers & (1 << i))) + continue; + so_emit_reloc_markers(nv40->nvws, + state->hw[NV40_STATE_FRAGTEX0+i]); + samplers &= ~(1ULL << i); } + so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FRAGPROG]); + so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_VTXBUF]); +} - nv40->dirty_samplers = 0; - nv40->dirty = 0; +void +nv40_emit_hw_state(struct nv40_context *nv40) +{ + nv40_state_validate(nv40); + nv40_state_emit(nv40); - nv40_state_emit_dummy_relocs(nv40); + BEGIN_RING(curie, NV40TCL_TEX_CACHE_CTL, 1); + OUT_RING (2); + BEGIN_RING(curie, NV40TCL_TEX_CACHE_CTL, 1); + OUT_RING (1); } diff --git a/src/gallium/drivers/nv40/nv40_state_fb.c b/src/gallium/drivers/nv40/nv40_state_fb.c new file mode 100644 index 0000000000..71795ab182 --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_fb.c @@ -0,0 +1,155 @@ +#include "nv40_context.h" + +static boolean +nv40_state_framebuffer_validate(struct nv40_context *nv40) +{ + struct pipe_framebuffer_state *fb = &nv40->framebuffer; + struct pipe_surface *rt[4], *zeta; + uint32_t rt_enable, rt_format, w, h; + int i, colour_format = 0, zeta_format = 0; + struct nouveau_stateobj *so = so_new(64, 10); + unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM; + + rt_enable = 0; + for (i = 0; i < 4; i++) { + if (!fb->cbufs[i]) + continue; + + if (colour_format) { + assert(w == fb->cbufs[i]->width); + assert(h == fb->cbufs[i]->height); + assert(colour_format == fb->cbufs[i]->format); + } else { + w = fb->cbufs[i]->width; + h = fb->cbufs[i]->height; + colour_format = fb->cbufs[i]->format; + rt_enable |= (NV40TCL_RT_ENABLE_COLOR0 << i); + rt[i] = fb->cbufs[i]; + } + } + + if (rt_enable & (NV40TCL_RT_ENABLE_COLOR1 | NV40TCL_RT_ENABLE_COLOR2 | + NV40TCL_RT_ENABLE_COLOR3)) + rt_enable |= NV40TCL_RT_ENABLE_MRT; + + if (fb->zsbuf) { + if (colour_format) { + assert(w == fb->zsbuf->width); + assert(h == fb->zsbuf->height); + } else { + w = fb->zsbuf->width; + h = fb->zsbuf->height; + } + + zeta_format = fb->zsbuf->format; + zeta = fb->zsbuf; + } + + rt_format = NV40TCL_RT_FORMAT_TYPE_LINEAR; + + switch (colour_format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + case 0: + rt_format |= NV40TCL_RT_FORMAT_COLOR_A8R8G8B8; + break; + case PIPE_FORMAT_R5G6B5_UNORM: + rt_format |= NV40TCL_RT_FORMAT_COLOR_R5G6B5; + break; + default: + assert(0); + } + + switch (zeta_format) { + case PIPE_FORMAT_Z16_UNORM: + rt_format |= NV40TCL_RT_FORMAT_ZETA_Z16; + break; + case PIPE_FORMAT_Z24S8_UNORM: + case 0: + rt_format |= NV40TCL_RT_FORMAT_ZETA_Z24S8; + break; + default: + assert(0); + } + + if (rt_enable & NV40TCL_RT_ENABLE_COLOR0) { + so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR0, 1); + so_reloc (so, rt[0]->buffer, 0, rt_flags | NOUVEAU_BO_OR, + nv40->nvws->channel->vram->handle, + nv40->nvws->channel->gart->handle); + so_method(so, nv40->screen->curie, NV40TCL_COLOR0_PITCH, 2); + so_data (so, rt[0]->pitch * rt[0]->cpp); + so_reloc (so, rt[0]->buffer, rt[0]->offset, rt_flags | + NOUVEAU_BO_LOW, 0, 0); + } + + if (rt_enable & NV40TCL_RT_ENABLE_COLOR1) { + so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR1, 1); + so_reloc (so, rt[1]->buffer, 0, rt_flags | NOUVEAU_BO_OR, + nv40->nvws->channel->vram->handle, + nv40->nvws->channel->gart->handle); + so_method(so, nv40->screen->curie, NV40TCL_COLOR1_OFFSET, 2); + so_reloc (so, rt[1]->buffer, rt[1]->offset, rt_flags | + NOUVEAU_BO_LOW, 0, 0); + so_data (so, rt[1]->pitch * rt[1]->cpp); + } + + if (rt_enable & NV40TCL_RT_ENABLE_COLOR2) { + so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR2, 1); + so_reloc (so, rt[2]->buffer, 0, rt_flags | NOUVEAU_BO_OR, + nv40->nvws->channel->vram->handle, + nv40->nvws->channel->gart->handle); + so_method(so, nv40->screen->curie, NV40TCL_COLOR2_OFFSET, 1); + so_reloc (so, rt[2]->buffer, rt[2]->offset, rt_flags | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, nv40->screen->curie, NV40TCL_COLOR2_PITCH, 1); + so_data (so, rt[2]->pitch * rt[2]->cpp); + } + + if (rt_enable & NV40TCL_RT_ENABLE_COLOR3) { + so_method(so, nv40->screen->curie, NV40TCL_DMA_COLOR3, 1); + so_reloc (so, rt[3]->buffer, 0, rt_flags | NOUVEAU_BO_OR, + nv40->nvws->channel->vram->handle, + nv40->nvws->channel->gart->handle); + so_method(so, nv40->screen->curie, NV40TCL_COLOR3_OFFSET, 1); + so_reloc (so, rt[3]->buffer, rt[3]->offset, rt_flags | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, nv40->screen->curie, NV40TCL_COLOR3_PITCH, 1); + so_data (so, rt[3]->pitch * rt[3]->cpp); + } + + if (zeta_format) { + so_method(so, nv40->screen->curie, NV40TCL_DMA_ZETA, 1); + so_reloc (so, zeta->buffer, 0, rt_flags | NOUVEAU_BO_OR, + nv40->nvws->channel->vram->handle, + nv40->nvws->channel->gart->handle); + so_method(so, nv40->screen->curie, NV40TCL_ZETA_OFFSET, 1); + so_reloc (so, zeta->buffer, zeta->offset, rt_flags | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, nv40->screen->curie, NV40TCL_ZETA_PITCH, 1); + so_data (so, zeta->pitch * zeta->cpp); + } + + so_method(so, nv40->screen->curie, NV40TCL_RT_ENABLE, 1); + so_data (so, rt_enable); + so_method(so, nv40->screen->curie, NV40TCL_RT_HORIZ, 3); + so_data (so, (w << 16) | 0); + so_data (so, (h << 16) | 0); + so_data (so, rt_format); + so_method(so, nv40->screen->curie, NV40TCL_VIEWPORT_HORIZ, 2); + so_data (so, (w << 16) | 0); + so_data (so, (h << 16) | 0); + so_method(so, nv40->screen->curie, NV40TCL_VIEWPORT_CLIP_HORIZ(0), 2); + so_data (so, ((w - 1) << 16) | 0); + so_data (so, ((h - 1) << 16) | 0); + + so_ref(so, &nv40->state.hw[NV40_STATE_FB]); + return TRUE; +} + +struct nv40_state_entry nv40_state_framebuffer = { + .validate = nv40_state_framebuffer_validate, + .dirty = { + .pipe = NV40_NEW_FB, + .hw = NV40_STATE_FB + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_rasterizer.c b/src/gallium/drivers/nv40/nv40_state_rasterizer.c new file mode 100644 index 0000000000..9ecda5990f --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_rasterizer.c @@ -0,0 +1,17 @@ +#include "nv40_context.h" + +static boolean +nv40_state_rasterizer_validate(struct nv40_context *nv40) +{ + so_ref(nv40->rasterizer->so, + &nv40->state.hw[NV40_STATE_RAST]); + return TRUE; +} + +struct nv40_state_entry nv40_state_rasterizer = { + .validate = nv40_state_rasterizer_validate, + .dirty = { + .pipe = NV40_NEW_RAST, + .hw = NV40_STATE_RAST + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_scissor.c b/src/gallium/drivers/nv40/nv40_state_scissor.c new file mode 100644 index 0000000000..9e9eadc511 --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_scissor.c @@ -0,0 +1,34 @@ +#include "nv40_context.h" + +static boolean +nv40_state_scissor_validate(struct nv40_context *nv40) +{ + struct pipe_rasterizer_state *rast = &nv40->rasterizer->pipe; + struct pipe_scissor_state *s = &nv40->scissor; + struct nouveau_stateobj *so; + + if (nv40->state.hw[NV40_STATE_SCISSOR] && + (rast->scissor == 0 && nv40->state.scissor_enabled == 0)) + return FALSE; + + so = so_new(3, 0); + so_method(so, nv40->screen->curie, NV40TCL_SCISSOR_HORIZ, 2); + if (rast->scissor) { + so_data (so, ((s->maxx - s->minx) << 16) | s->minx); + so_data (so, ((s->maxy - s->miny) << 16) | s->miny); + } else { + so_data (so, 4096 << 16); + so_data (so, 4096 << 16); + } + + so_ref(so, &nv40->state.hw[NV40_STATE_SCISSOR]); + return TRUE; +} + +struct nv40_state_entry nv40_state_scissor = { + .validate = nv40_state_scissor_validate, + .dirty = { + .pipe = NV40_NEW_SCISSOR | NV40_NEW_RAST, + .hw = NV40_STATE_SCISSOR + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_stipple.c b/src/gallium/drivers/nv40/nv40_state_stipple.c new file mode 100644 index 0000000000..b51024ad9b --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_stipple.c @@ -0,0 +1,39 @@ +#include "nv40_context.h" + +static boolean +nv40_state_stipple_validate(struct nv40_context *nv40) +{ + struct pipe_rasterizer_state *rast = &nv40->rasterizer->pipe; + struct nouveau_grobj *curie = nv40->screen->curie; + struct nouveau_stateobj *so; + + if (nv40->state.hw[NV40_STATE_STIPPLE] && + (rast->poly_stipple_enable == 0 && nv40->state.stipple_enabled == 0)) + return FALSE; + + if (rast->poly_stipple_enable) { + unsigned i; + + so = so_new(35, 0); + so_method(so, curie, NV40TCL_POLYGON_STIPPLE_ENABLE, 1); + so_data (so, 1); + so_method(so, curie, NV40TCL_POLYGON_STIPPLE_PATTERN(0), 32); + for (i = 0; i < 32; i++) + so_data(so, nv40->stipple[i]); + } else { + so = so_new(2, 0); + so_method(so, curie, NV40TCL_POLYGON_STIPPLE_ENABLE, 1); + so_data (so, 0); + } + + so_ref(so, &nv40->state.hw[NV40_STATE_STIPPLE]); + return TRUE; +} + +struct nv40_state_entry nv40_state_stipple = { + .validate = nv40_state_stipple_validate, + .dirty = { + .pipe = NV40_NEW_STIPPLE | NV40_NEW_RAST, + .hw = NV40_STATE_STIPPLE, + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_viewport.c b/src/gallium/drivers/nv40/nv40_state_viewport.c new file mode 100644 index 0000000000..3a32533907 --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_viewport.c @@ -0,0 +1,29 @@ +#include "nv40_context.h" + +static boolean +nv40_state_viewport_validate(struct nv40_context *nv40) +{ + struct nouveau_stateobj *so = so_new(9, 0); + struct pipe_viewport_state *vpt = &nv40->viewport; + + so_method(so, nv40->screen->curie, NV40TCL_VIEWPORT_TRANSLATE_X, 8); + so_data (so, fui(vpt->translate[0])); + so_data (so, fui(vpt->translate[1])); + so_data (so, fui(vpt->translate[2])); + so_data (so, fui(vpt->translate[3])); + so_data (so, fui(vpt->scale[0])); + so_data (so, fui(vpt->scale[1])); + so_data (so, fui(vpt->scale[2])); + so_data (so, fui(vpt->scale[3])); + + so_ref(so, &nv40->state.hw[NV40_STATE_VIEWPORT]); + return TRUE; +} + +struct nv40_state_entry nv40_state_viewport = { + .validate = nv40_state_viewport_validate, + .dirty = { + .pipe = NV40_NEW_VIEWPORT, + .hw = NV40_STATE_VIEWPORT + } +}; diff --git a/src/gallium/drivers/nv40/nv40_state_zsa.c b/src/gallium/drivers/nv40/nv40_state_zsa.c new file mode 100644 index 0000000000..fb760677c8 --- /dev/null +++ b/src/gallium/drivers/nv40/nv40_state_zsa.c @@ -0,0 +1,17 @@ +#include "nv40_context.h" + +static boolean +nv40_state_zsa_validate(struct nv40_context *nv40) +{ + so_ref(nv40->zsa->so, + &nv40->state.hw[NV40_STATE_ZSA]); + return TRUE; +} + +struct nv40_state_entry nv40_state_zsa = { + .validate = nv40_state_zsa_validate, + .dirty = { + .pipe = NV40_NEW_ZSA, + .hw = NV40_STATE_ZSA + } +}; diff --git a/src/gallium/drivers/nv40/nv40_surface.c b/src/gallium/drivers/nv40/nv40_surface.c index 9726ab4e4d..e8a6011696 100644 --- a/src/gallium/drivers/nv40/nv40_surface.c +++ b/src/gallium/drivers/nv40/nv40_surface.c @@ -33,76 +33,6 @@ #include "pipe/p_inlines.h" #include "util/p_tile.h" -static boolean -nv40_surface_format_supported(struct pipe_context *pipe, - enum pipe_format format, uint type) -{ - switch (type) { - case PIPE_SURFACE: - switch (format) { - case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_R5G6B5_UNORM: - case PIPE_FORMAT_Z24S8_UNORM: - case PIPE_FORMAT_Z16_UNORM: - return TRUE; - default: - break; - } - break; - case PIPE_TEXTURE: - switch (format) { - case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_A1R5G5B5_UNORM: - case PIPE_FORMAT_A4R4G4B4_UNORM: - case PIPE_FORMAT_R5G6B5_UNORM: - case PIPE_FORMAT_U_L8: - case PIPE_FORMAT_U_A8: - case PIPE_FORMAT_U_I8: - case PIPE_FORMAT_U_A8_L8: - case PIPE_FORMAT_Z16_UNORM: - case PIPE_FORMAT_Z24S8_UNORM: - return TRUE; - default: - break; - } - break; - default: - assert(0); - }; - - return FALSE; -} - -static struct pipe_surface * -nv40_get_tex_surface(struct pipe_context *pipe, struct pipe_texture *pt, - unsigned face, unsigned level, unsigned zslice) -{ - struct pipe_winsys *ws = pipe->winsys; - struct nv40_miptree *nv40mt = (struct nv40_miptree *)pt; - struct pipe_surface *ps; - - ps = ws->surface_alloc(ws); - if (!ps) - return NULL; - pipe_buffer_reference(ws, &ps->buffer, nv40mt->buffer); - ps->format = pt->format; - ps->cpp = pt->cpp; - ps->width = pt->width[level]; - ps->height = pt->height[level]; - ps->pitch = nv40mt->level[level].pitch / ps->cpp; - - if (pt->target == PIPE_TEXTURE_CUBE) { - ps->offset = nv40mt->level[level].image_offset[face]; - } else - if (pt->target == PIPE_TEXTURE_3D) { - ps->offset = nv40mt->level[level].image_offset[zslice]; - } else { - ps->offset = nv40mt->level[level].image_offset[0]; - } - - return ps; -} - static void nv40_surface_copy(struct pipe_context *pipe, unsigned do_flip, struct pipe_surface *dest, unsigned destx, unsigned desty, @@ -130,8 +60,6 @@ nv40_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest, void nv40_init_surface_functions(struct nv40_context *nv40) { - nv40->pipe.is_format_supported = nv40_surface_format_supported; - nv40->pipe.get_tex_surface = nv40_get_tex_surface; nv40->pipe.surface_copy = nv40_surface_copy; nv40->pipe.surface_fill = nv40_surface_fill; } diff --git a/src/gallium/drivers/nv40/nv40_vbo.c b/src/gallium/drivers/nv40/nv40_vbo.c index fa827ef0c5..bedc8c6d4e 100644 --- a/src/gallium/drivers/nv40/nv40_vbo.c +++ b/src/gallium/drivers/nv40/nv40_vbo.c @@ -30,9 +30,50 @@ nv40_vbo_type(uint format) case PIPE_FORMAT_TYPE_UNORM: return NV40TCL_VTXFMT_TYPE_UBYTE; default: - NOUVEAU_ERR("Unknown format 0x%08x\n", format); + { + char fs[128]; + pf_sprint_name(fs, format); + NOUVEAU_ERR("Unknown format %s\n", fs); return NV40TCL_VTXFMT_TYPE_FLOAT; } + } +} + +static boolean +nv40_vbo_set_idxbuf(struct nv40_context *nv40, struct pipe_buffer *ib, + unsigned ib_size) +{ + unsigned type; + + if (!ib) { + nv40->idxbuf = NULL; + nv40->idxbuf_format = 0xdeadbeef; + return FALSE; + } + + /* No support for 8bit indices, no support at all on 0x4497 chips */ + if (nv40->screen->curie->grclass == NV44TCL || ib_size == 1) + return FALSE; + + switch (ib_size) { + case 2: + type = NV40TCL_IDXBUF_FORMAT_TYPE_U16; + break; + case 4: + type = NV40TCL_IDXBUF_FORMAT_TYPE_U32; + break; + default: + return FALSE; + } + + if (ib != nv40->idxbuf || + type != nv40->idxbuf_format) { + nv40->dirty |= NV40_NEW_ARRAYS; + nv40->idxbuf = ib; + nv40->idxbuf_format = type; + } + + return TRUE; } static boolean @@ -97,105 +138,15 @@ nv40_vbo_static_attrib(struct nv40_context *nv40, int attrib, return TRUE; } -static void -nv40_vbo_arrays_update(struct nv40_context *nv40, struct pipe_buffer *ib, - unsigned ib_format) -{ - struct nv40_vertex_program *vp = nv40->vertprog.active; - struct nouveau_stateobj *vtxbuf, *vtxfmt; - unsigned inputs, hw, num_hw; - unsigned vb_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD; - - inputs = vp->ir; - for (hw = 0; hw < 16 && inputs; hw++) { - if (inputs & (1 << hw)) { - num_hw = hw; - inputs &= ~(1 << hw); - } - } - num_hw++; - - vtxbuf = so_new(20, 18); - so_method(vtxbuf, nv40->hw->curie, NV40TCL_VTXBUF_ADDRESS(0), num_hw); - vtxfmt = so_new(17, 0); - so_method(vtxfmt, nv40->hw->curie, NV40TCL_VTXFMT(0), num_hw); - - inputs = vp->ir; - for (hw = 0; hw < num_hw; hw++) { - struct pipe_vertex_element *ve; - struct pipe_vertex_buffer *vb; - - if (!(inputs & (1 << hw))) { - so_data(vtxbuf, 0); - so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT); - continue; - } - - ve = &nv40->vtxelt[hw]; - vb = &nv40->vtxbuf[ve->vertex_buffer_index]; - - if (!vb->pitch && nv40_vbo_static_attrib(nv40, hw, ve, vb)) { - so_data(vtxbuf, 0); - so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT); - continue; - } - - so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset, - vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, - 0, NV40TCL_VTXBUF_ADDRESS_DMA1); - so_data (vtxfmt, ((vb->pitch << NV40TCL_VTXFMT_STRIDE_SHIFT) | - (nv40_vbo_ncomp(ve->src_format) << - NV40TCL_VTXFMT_SIZE_SHIFT) | - nv40_vbo_type(ve->src_format))); - } - - if (ib) { - so_method(vtxbuf, nv40->hw->curie, NV40TCL_IDXBUF_ADDRESS, 2); - so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); - so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR, - 0, NV40TCL_IDXBUF_FORMAT_DMA1); - } - - so_emit(nv40->nvws, vtxfmt); - so_emit(nv40->nvws, vtxbuf); - so_ref (vtxbuf, &nv40->so_vtxbuf); - so_ref (NULL, &vtxbuf); - so_ref (NULL, &vtxfmt); -} - -static boolean -nv40_vbo_validate_state(struct nv40_context *nv40, - struct pipe_buffer *ib, unsigned ib_format) -{ - unsigned vdn = nv40->dirty & NV40_NEW_ARRAYS; - - nv40_emit_hw_state(nv40); - if (vdn || ib) { - nv40_vbo_arrays_update(nv40, ib, ib_format); - nv40->dirty &= ~NV40_NEW_ARRAYS; - } - - so_emit_reloc_markers(nv40->nvws, nv40->so_vtxbuf); - - BEGIN_RING(curie, 0x1710, 1); - OUT_RING (0); /* vtx cache flush */ - - return TRUE; -} - boolean nv40_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); unsigned nr; - boolean ret; - ret = nv40_vbo_validate_state(nv40, NULL, 0); - if (!ret) { - NOUVEAU_ERR("state validate failed\n"); - return FALSE; - } + nv40_vbo_set_idxbuf(nv40, NULL, 0); + nv40_emit_hw_state(nv40); BEGIN_RING(curie, NV40TCL_BEGIN_END, 1); OUT_RING (nvgl_primitive(mode)); @@ -302,14 +253,9 @@ nv40_draw_elements_inline(struct pipe_context *pipe, { struct nv40_context *nv40 = nv40_context(pipe); struct pipe_winsys *ws = pipe->winsys; - boolean ret; void *map; - ret = nv40_vbo_validate_state(nv40, NULL, 0); - if (!ret) { - NOUVEAU_ERR("state validate failed\n"); - return FALSE; - } + nv40_emit_hw_state(nv40); map = ws->buffer_map(ws, ib, PIPE_BUFFER_USAGE_CPU_READ); if (!ib) { @@ -345,30 +291,12 @@ nv40_draw_elements_inline(struct pipe_context *pipe, static boolean nv40_draw_elements_vbo(struct pipe_context *pipe, - struct pipe_buffer *ib, unsigned ib_size, unsigned mode, unsigned start, unsigned count) { struct nv40_context *nv40 = nv40_context(pipe); - unsigned nr, type; - boolean ret; - - switch (ib_size) { - case 2: - type = NV40TCL_IDXBUF_FORMAT_TYPE_U16; - break; - case 4: - type = NV40TCL_IDXBUF_FORMAT_TYPE_U32; - break; - default: - NOUVEAU_ERR("invalid idxbuf fmt %d\n", ib_size); - return FALSE; - } + unsigned nr; - ret = nv40_vbo_validate_state(nv40, ib, type); - if (!ret) { - NOUVEAU_ERR("failed state validation\n"); - return FALSE; - } + nv40_emit_hw_state(nv40); BEGIN_RING(curie, NV40TCL_BEGIN_END, 1); OUT_RING (nvgl_primitive(mode)); @@ -406,19 +334,92 @@ nv40_draw_elements(struct pipe_context *pipe, { struct nv40_context *nv40 = nv40_context(pipe); - /* 0x4497 doesn't support real index buffers, and there doesn't appear - * to be support on any chipset for 8-bit indices. - */ - if (nv40->hw->curie->grclass == NV44TCL || indexSize == 1) { + if (nv40_vbo_set_idxbuf(nv40, indexBuffer, indexSize)) { + nv40_draw_elements_vbo(pipe, mode, start, count); + } else { nv40_draw_elements_inline(pipe, indexBuffer, indexSize, mode, start, count); - } else { - nv40_draw_elements_vbo(pipe, indexBuffer, indexSize, - mode, start, count); } pipe->flush(pipe, 0); return TRUE; } +static boolean +nv40_vbo_validate(struct nv40_context *nv40) +{ + struct nv40_vertex_program *vp = nv40->vertprog; + struct nouveau_stateobj *vtxbuf, *vtxfmt; + struct pipe_buffer *ib = nv40->idxbuf; + unsigned ib_format = nv40->idxbuf_format; + unsigned inputs, hw, num_hw; + unsigned vb_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD; + + inputs = vp->ir; + for (hw = 0; hw < 16 && inputs; hw++) { + if (inputs & (1 << hw)) { + num_hw = hw; + inputs &= ~(1 << hw); + } + } + num_hw++; + + vtxbuf = so_new(20, 18); + so_method(vtxbuf, nv40->screen->curie, NV40TCL_VTXBUF_ADDRESS(0), num_hw); + vtxfmt = so_new(17, 0); + so_method(vtxfmt, nv40->screen->curie, NV40TCL_VTXFMT(0), num_hw); + + inputs = vp->ir; + for (hw = 0; hw < num_hw; hw++) { + struct pipe_vertex_element *ve; + struct pipe_vertex_buffer *vb; + + if (!(inputs & (1 << hw))) { + so_data(vtxbuf, 0); + so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT); + continue; + } + + ve = &nv40->vtxelt[hw]; + vb = &nv40->vtxbuf[ve->vertex_buffer_index]; + + if (!vb->pitch && nv40_vbo_static_attrib(nv40, hw, ve, vb)) { + so_data(vtxbuf, 0); + so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT); + continue; + } + + so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset, + vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, + 0, NV40TCL_VTXBUF_ADDRESS_DMA1); + so_data (vtxfmt, ((vb->pitch << NV40TCL_VTXFMT_STRIDE_SHIFT) | + (nv40_vbo_ncomp(ve->src_format) << + NV40TCL_VTXFMT_SIZE_SHIFT) | + nv40_vbo_type(ve->src_format))); + } + + if (ib) { + so_method(vtxbuf, nv40->screen->curie, NV40TCL_IDXBUF_ADDRESS, 2); + so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0); + so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR, + 0, NV40TCL_IDXBUF_FORMAT_DMA1); + } + + so_method(vtxbuf, nv40->screen->curie, 0x1710, 1); + so_data (vtxbuf, 0); + + so_ref(vtxbuf, &nv40->state.hw[NV40_STATE_VTXBUF]); + nv40->state.dirty |= (1ULL << NV40_STATE_VTXBUF); + so_ref(vtxfmt, &nv40->state.hw[NV40_STATE_VTXFMT]); + nv40->state.dirty |= (1ULL << NV40_STATE_VTXFMT); + return FALSE; +} + +struct nv40_state_entry nv40_state_vbo = { + .validate = nv40_vbo_validate, + .dirty = { + .pipe = NV40_NEW_ARRAYS, + .hw = 0, + } +}; diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c index 9f4738b830..5b7a343e55 100644 --- a/src/gallium/drivers/nv40/nv40_vertprog.c +++ b/src/gallium/drivers/nv40/nv40_vertprog.c @@ -558,7 +558,7 @@ nv40_vertprog_prepare(struct nv40_vpc *vpc) return TRUE; } -void +static void nv40_vertprog_translate(struct nv40_context *nv40, struct nv40_vertex_program *vp) { @@ -631,24 +631,32 @@ out_err: free(vpc); } -void -nv40_vertprog_bind(struct nv40_context *nv40, struct nv40_vertex_program *vp) +static boolean +nv40_vertprog_validate(struct nv40_context *nv40) { + struct nv40_vertex_program *vp = nv40->vertprog; + struct pipe_buffer *constbuf = + nv40->constbuf[PIPE_SHADER_VERTEX]; struct nouveau_winsys *nvws = nv40->nvws; struct pipe_winsys *ws = nv40->pipe.winsys; boolean upload_code = FALSE, upload_data = FALSE; int i; /* Translate TGSI shader into hw bytecode */ + if (vp->translated) + goto check_gpu_resources; + + nv40_vertprog_translate(nv40, vp); if (!vp->translated) { - nv40_vertprog_translate(nv40, vp); - if (!vp->translated) - assert(0); + nv40->fallback |= NV40_FALLBACK_TNL; + return FALSE; } +check_gpu_resources: /* Allocate hw vtxprog exec slots */ if (!vp->exec) { - struct nouveau_resource *heap = nv40->hw->vp_exec_heap; + struct nouveau_resource *heap = nv40->screen->vp_exec_heap; + struct nouveau_stateobj *so; uint vplen = vp->nr_insns; if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) { @@ -663,12 +671,20 @@ nv40_vertprog_bind(struct nv40_context *nv40, struct nv40_vertex_program *vp) assert(0); } + so = so_new(5, 0); + so_method(so, nv40->screen->curie, NV40TCL_VP_START_FROM_ID, 1); + so_data (so, vp->exec->start); + so_method(so, nv40->screen->curie, NV40TCL_VP_ATTRIB_EN, 2); + so_data (so, vp->ir); + so_data (so, vp->or); + so_ref(so, &vp->so); + upload_code = TRUE; } /* Allocate hw vtxprog const slots */ if (vp->nr_consts && !vp->data) { - struct nouveau_resource *heap = nv40->hw->vp_data_heap; + struct nouveau_resource *heap = nv40->screen->vp_data_heap; if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) { while (heap->next && heap->size < vp->nr_consts) { @@ -725,8 +741,8 @@ nv40_vertprog_bind(struct nv40_context *nv40, struct nv40_vertex_program *vp) if (vp->nr_consts) { float *map = NULL; - if (nv40->vertprog.constant_buf) { - map = ws->buffer_map(ws, nv40->vertprog.constant_buf, + if (constbuf) { + map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); } @@ -747,9 +763,8 @@ nv40_vertprog_bind(struct nv40_context *nv40, struct nv40_vertex_program *vp) OUT_RINGp ((uint32_t *)vpd->value, 4); } - if (map) { - ws->buffer_unmap(ws, nv40->vertprog.constant_buf); - } + if (constbuf) + ws->buffer_unmap(ws, constbuf); } /* Upload vtxprog */ @@ -770,13 +785,12 @@ nv40_vertprog_bind(struct nv40_context *nv40, struct nv40_vertex_program *vp) } } - BEGIN_RING(curie, NV40TCL_VP_START_FROM_ID, 1); - OUT_RING (vp->exec->start); - BEGIN_RING(curie, NV40TCL_VP_ATTRIB_EN, 2); - OUT_RING (vp->ir); - OUT_RING (vp->or); + if (vp->so != nv40->state.hw[NV40_STATE_VERTPROG]) { + so_ref(vp->so, &nv40->state.hw[NV40_STATE_VERTPROG]); + return TRUE; + } - nv40->vertprog.active = vp; + return FALSE; } void @@ -788,3 +802,11 @@ nv40_vertprog_destroy(struct nv40_context *nv40, struct nv40_vertex_program *vp) free(vp->insns); } +struct nv40_state_entry nv40_state_vertprog = { + .validate = nv40_vertprog_validate, + .dirty = { + .pipe = NV40_NEW_VERTPROG, + .hw = NV40_STATE_VERTPROG, + } +}; + |