diff options
| author | Michal Krol <michal@vmware.com> | 2010-03-10 15:49:30 +0100 |
|---|---|---|
| committer | Michal Krol <michal@vmware.com> | 2010-03-10 15:49:30 +0100 |
| commit | 3ce4375912c8ea488460e593e07c5bb15b92dca9 (patch) | |
| tree | 1011fa439bd829fd46a44fd99478135848800e73 /src/gallium/drivers/r300 | |
| parent | f59f28093ea827bd234d8e1a36bdd56a9fce5f09 (diff) | |
| parent | 9b348d0ed125a22be3f318ac60cef6f201edfdab (diff) | |
Merge branch 'master' into gallium-sampler-view
Conflicts:
src/gallium/auxiliary/Makefile
src/gallium/auxiliary/SConscript
src/gallium/auxiliary/tgsi/tgsi_exec.c
src/gallium/auxiliary/util/u_blitter.c
src/gallium/drivers/i915/i915_context.h
src/gallium/drivers/i965/brw_context.h
src/gallium/drivers/llvmpipe/lp_context.h
src/gallium/drivers/nv50/nv50_context.h
src/gallium/drivers/nv50/nv50_state_validate.c
src/gallium/drivers/nv50/nv50_tex.c
src/gallium/drivers/r300/r300_blit.c
src/gallium/drivers/r300/r300_context.h
src/gallium/drivers/r300/r300_emit.c
src/gallium/drivers/r300/r300_state.c
src/gallium/drivers/softpipe/sp_context.h
src/gallium/drivers/svga/svga_context.h
src/gallium/drivers/svga/svga_pipe_sampler.c
Diffstat (limited to 'src/gallium/drivers/r300')
25 files changed, 1935 insertions, 1417 deletions
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile index afddcb161f..61b54af4dd 100644 --- a/src/gallium/drivers/r300/Makefile +++ b/src/gallium/drivers/r300/Makefile @@ -19,7 +19,8 @@ C_SOURCES = \ r300_state_invariant.c \ r300_vs.c \ r300_texture.c \ - r300_tgsi_to_rc.c + r300_tgsi_to_rc.c \ + r300_transfer.c LIBRARY_INCLUDES = \ -I$(TOP)/src/mesa/drivers/dri/r300/compiler \ @@ -32,7 +33,5 @@ EXTRA_OBJECTS = \ include ../../Makefile.template -.PHONY : $(COMPILER_ARCHIVE) - $(COMPILER_ARCHIVE): $(MAKE) -C $(TOP)/src/mesa/drivers/dri/r300/compiler diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript index 183aa17f9b..27b2e30993 100644 --- a/src/gallium/drivers/r300/SConscript +++ b/src/gallium/drivers/r300/SConscript @@ -30,6 +30,7 @@ r300 = env.ConvenienceLibrary( 'r300_vs.c', 'r300_texture.c', 'r300_tgsi_to_rc.c', + 'r300_transfer.c', ] + r300compiler) + r300compiler Export('r300') diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c index c48684fe51..a60b12844d 100644 --- a/src/gallium/drivers/r300/r300_blit.c +++ b/src/gallium/drivers/r300/r300_blit.c @@ -33,7 +33,10 @@ static void r300_blitter_save_states(struct r300_context* r300) util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref)); util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state); util_blitter_save_fragment_shader(r300->blitter, r300->fs); - util_blitter_save_vertex_shader(r300->blitter, r300->vs); + util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state); + util_blitter_save_viewport(r300->blitter, &r300->viewport); + util_blitter_save_clip(r300->blitter, &r300->clip); + util_blitter_save_vertex_elements(r300->blitter, r300->velems); } /* Clear currently bound buffers. */ @@ -98,6 +101,8 @@ static void r300_hw_copy(struct pipe_context* pipe, unsigned width, unsigned height) { struct r300_context* r300 = r300_context(pipe); + struct r300_textures_state* state = + (struct r300_textures_state*)r300->textures_state.state; /* Yeah we have to save all those states to ensure this blitter operation * is really transparent. The states will be restored by the blitter once @@ -106,7 +111,7 @@ static void r300_hw_copy(struct pipe_context* pipe, util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); util_blitter_save_fragment_sampler_states( - r300->blitter, r300->sampler_count, (void**)r300->sampler_states); + r300->blitter, state->sampler_count, (void**)state->sampler_states); util_blitter_save_fragment_sampler_views( r300->blitter, r300->fragment_sampler_view_count, @@ -139,10 +144,10 @@ void r300_surface_copy(struct pipe_context* pipe, new_format = PIPE_FORMAT_I8_UNORM; break; case 2: - new_format = PIPE_FORMAT_A4R4G4B4_UNORM; + new_format = PIPE_FORMAT_B4G4R4A4_UNORM; break; case 4: - new_format = PIPE_FORMAT_A8R8G8B8_UNORM; + new_format = PIPE_FORMAT_B8G8R8A8_UNORM; break; default: debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n" diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index f631b4ed27..923e1e541f 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -59,7 +59,8 @@ static void r300_destroy_context(struct pipe_context* context) FREE(r300->fb_state.state); FREE(r300->rs_block_state.state); FREE(r300->scissor_state.state); - FREE(r300->vertex_format_state.state); + FREE(r300->textures_state.state); + FREE(r300->vap_output_state.state); FREE(r300->viewport_state.state); FREE(r300->ztop_state.state); FREE(r300); @@ -70,11 +71,8 @@ r300_is_texture_referenced(struct pipe_context *pipe, struct pipe_texture *texture, unsigned face, unsigned level) { - struct pipe_buffer* buf = 0; - - r300_get_texture_buffer(pipe->screen, texture, &buf, NULL); - - return pipe->is_buffer_referenced(pipe, buf); + return pipe->is_buffer_referenced(pipe, + ((struct r300_texture *)texture)->buffer); } static unsigned int @@ -84,7 +82,14 @@ r300_is_buffer_referenced(struct pipe_context *pipe, /* This only checks to see whether actual hardware buffers are * referenced. Since we use managed BOs and transfers, it's actually not * possible for pipe_buffers to ever reference the actual hardware, so - * buffers are never referenced. */ + * buffers are never referenced. + */ + + /* XXX: that doesn't make sense given that + * r300_is_texture_referenced is implemented on top of this + * function and hardware can certainly refer to textures + * directly... + */ return 0; } @@ -96,39 +101,54 @@ static void r300_flush_cb(void *data) } #define R300_INIT_ATOM(atomname, atomsize) \ - r300->atomname##_state.name = #atomname; \ - r300->atomname##_state.state = NULL; \ - r300->atomname##_state.size = atomsize; \ - r300->atomname##_state.emit = r300_emit_##atomname##_state; \ - r300->atomname##_state.dirty = FALSE; \ - insert_at_tail(&r300->atom_list, &r300->atomname##_state); + r300->atomname.name = #atomname; \ + r300->atomname.state = NULL; \ + r300->atomname.size = atomsize; \ + r300->atomname.emit = r300_emit_##atomname; \ + r300->atomname.dirty = FALSE; \ + insert_at_tail(&r300->atom_list, &r300->atomname); static void r300_setup_atoms(struct r300_context* r300) { + boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; + boolean has_tcl = r300_screen(r300->context.screen)->caps->has_tcl; + /* Create the actual atom list. * * Each atom is examined and emitted in the order it appears here, which * can affect performance and conformance if not handled with care. * - * Some atoms never change size, others change every emit. This is just - * an upper bound on each atom, to keep the emission machinery from - * underallocating space. */ + * Some atoms never change size, others change every emit - those have + * the size of 0 here. */ make_empty_list(&r300->atom_list); - R300_INIT_ATOM(invariant, 71); - R300_INIT_ATOM(ztop, 2); - R300_INIT_ATOM(blend, 8); - R300_INIT_ATOM(blend_color, 3); - R300_INIT_ATOM(clip, 29); - R300_INIT_ATOM(dsa, 8); - R300_INIT_ATOM(fb, 56); - R300_INIT_ATOM(rs, 25); - R300_INIT_ATOM(scissor, 3); - R300_INIT_ATOM(viewport, 9); - R300_INIT_ATOM(rs_block, 21); - R300_INIT_ATOM(vertex_format, 26); + R300_INIT_ATOM(invariant_state, 71); + R300_INIT_ATOM(ztop_state, 2); + R300_INIT_ATOM(blend_state, 8); + R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2); + R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2); + R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6); + R300_INIT_ATOM(fb_state, 0); + R300_INIT_ATOM(rs_state, 0); + R300_INIT_ATOM(scissor_state, 3); + R300_INIT_ATOM(viewport_state, 9); + R300_INIT_ATOM(rs_block_state, 0); + R300_INIT_ATOM(vertex_stream_state, 0); + R300_INIT_ATOM(vap_output_state, 6); + R300_INIT_ATOM(pvs_flush, 2); + R300_INIT_ATOM(vs_state, 0); + R300_INIT_ATOM(texture_cache_inval, 2); + R300_INIT_ATOM(textures_state, 0); /* Some non-CSO atoms need explicit space to store the state locally. */ + r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state); + r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state); r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state); + r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block); + r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state); + r300->textures_state.state = CALLOC_STRUCT(r300_textures_state); + r300->vap_output_state.state = CALLOC_STRUCT(r300_vap_output_state); + r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state); + r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state); } struct pipe_context* r300_create_context(struct pipe_screen* screen, @@ -141,6 +161,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, if (!r300) return NULL; + r300screen->ctx = (struct pipe_context*)r300; + r300->winsys = radeon_winsys; r300->context.winsys = (struct pipe_winsys*)radeon_winsys; @@ -178,14 +200,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300_setup_atoms(r300); - r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state); - r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state); - r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block); - r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state); - r300->vertex_format_state.state = CALLOC_STRUCT(r300_vertex_info); - r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state); - r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state); - /* Open up the OQ BO. */ r300->oqbo = screen->buffer_create(screen, 4096, PIPE_BUFFER_USAGE_VERTEX, 4096); @@ -202,7 +216,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->invariant_state.dirty = TRUE; r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300); - r300->dirty_state = R300_NEW_KITCHEN_SINK; r300->dirty_hw++; r300->blitter = util_blitter_create(&r300->context); diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index 2edf65797a..3b70bcda82 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -45,7 +45,7 @@ struct r300_atom { /* Opaque state. */ void* state; /* Emit the state to the context. */ - void (*emit)(struct r300_context*, void*); + void (*emit)(struct r300_context*, unsigned, void*); /* Upper bound on number of dwords to emit. */ unsigned size; /* Whether this atom should be emitted. */ @@ -86,7 +86,6 @@ struct r300_rs_state { uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ uint32_t antialiasing_config; /* R300_GB_AA_CONFIG: 0x4020 */ uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ - uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ float depth_scale; /* R300_SU_POLY_OFFSET_FRONT_SCALE: 0x42a4 */ /* R300_SU_POLY_OFFSET_BACK_SCALE: 0x42ac */ @@ -119,12 +118,58 @@ struct r300_sampler_state { unsigned min_lod, max_lod; }; -struct r300_texture_state { +struct r300_texture_format_state { uint32_t format0; /* R300_TX_FORMAT0: 0x4480 */ uint32_t format1; /* R300_TX_FORMAT1: 0x44c0 */ uint32_t format2; /* R300_TX_FORMAT2: 0x4500 */ }; +#define R300_MAX_TEXTURE_LEVELS 13 + +struct r300_texture_fb_state { + /* Colorbuffer. */ + uint32_t colorpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_COLORPITCH[0-3]*/ + uint32_t us_out_fmt; /* R300_US_OUT_FMT[0-3] */ + + /* Zbuffer. */ + uint32_t depthpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_DEPTHPITCH */ + uint32_t zb_format; /* R300_ZB_FORMAT */ +}; + +struct r300_textures_state { + /* Textures. */ + struct r300_texture *textures[8]; + int texture_count; + /* Sampler states. */ + struct r300_sampler_state *sampler_states[8]; + int sampler_count; + + /* These is the merge of the texture and sampler states. */ + unsigned count; + uint32_t tx_enable; /* R300_TX_ENABLE: 0x4101 */ + struct r300_texture_sampler_state { + uint32_t format[3]; /* R300_TX_FORMAT[0-2] */ + uint32_t filter[2]; /* R300_TX_FILTER[0-1] */ + uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ + uint32_t tile_config; /* R300_TX_OFFSET (subset thereof) */ + } regs[8]; +}; + +struct r300_vertex_stream_state { + /* R300_VAP_PROG_STREAK_CNTL_[0-7] */ + uint32_t vap_prog_stream_cntl[8]; + /* R300_VAP_PROG_STREAK_CNTL_EXT_[0-7] */ + uint32_t vap_prog_stream_cntl_ext[8]; + + unsigned count; +}; + +struct r300_vap_output_state { + uint32_t vap_vtx_state_cntl; /* R300_VAP_VTX_STATE_CNTL: 0x2180 */ + uint32_t vap_vsm_vtx_assm; /* R300_VAP_VSM_VTX_ASSM: 0x2184 */ + uint32_t vap_out_vtx_fmt[2]; /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */ +}; + struct r300_viewport_state { float xscale; /* R300_VAP_VPORT_XSCALE: 0x2098 */ float xoffset; /* R300_VAP_VPORT_XOFFSET: 0x209c */ @@ -141,11 +186,6 @@ struct r300_ztop_state { #define R300_NEW_FRAGMENT_SHADER 0x00000020 #define R300_NEW_FRAGMENT_SHADER_CONSTANTS 0x00000040 -#define R300_NEW_SAMPLER 0x00000200 -#define R300_ANY_NEW_SAMPLERS 0x0001fe00 -#define R300_NEW_TEXTURE 0x00040000 -#define R300_ANY_NEW_TEXTURES 0x03fc0000 -#define R300_NEW_VERTEX_SHADER 0x08000000 #define R300_NEW_VERTEX_SHADER_CONSTANTS 0x10000000 #define R300_NEW_QUERY 0x40000000 #define R300_NEW_KITCHEN_SINK 0x7fffffff @@ -198,16 +238,16 @@ struct r300_texture { struct pipe_texture tex; /* Offsets into the buffer. */ - unsigned offset[PIPE_MAX_TEXTURE_LEVELS]; + unsigned offset[R300_MAX_TEXTURE_LEVELS]; /* A pitch for each mip-level */ - unsigned pitch[PIPE_MAX_TEXTURE_LEVELS]; + unsigned pitch[R300_MAX_TEXTURE_LEVELS]; /* Size of one zslice or face based on the texture target */ - unsigned layer_size[PIPE_MAX_TEXTURE_LEVELS]; + unsigned layer_size[R300_MAX_TEXTURE_LEVELS]; /* Whether the mipmap level is macrotiled. */ - enum r300_buffer_tiling mip_macrotile[PIPE_MAX_TEXTURE_LEVELS]; + enum r300_buffer_tiling mip_macrotile[R300_MAX_TEXTURE_LEVELS]; /** * If non-zero, override the natural texture layout with @@ -231,7 +271,8 @@ struct r300_texture { struct pipe_buffer* buffer; /* Registers carrying texture format data. */ - struct r300_texture_state state; + struct r300_texture_format_state state; + struct r300_texture_fb_state fb_state; /* Buffer tiling */ enum r300_buffer_tiling microtile, macrotile; @@ -247,6 +288,13 @@ struct r300_vertex_info { uint32_t vap_prog_stream_cntl_ext[8]; }; +struct r300_vertex_element_state { + unsigned count; + struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS]; + + struct r300_vertex_stream_state vertex_stream; +}; + extern struct pipe_viewport_state r300_viewport_identity; struct r300_context { @@ -271,9 +319,6 @@ struct r300_context { struct r300_query *query_current; struct r300_query query_list; - /* Vertex formatting information. */ - struct r300_atom vertex_format_state; - /* Various CSO state objects. */ /* Beginning of atom list. */ struct r300_atom atom_list; @@ -295,20 +340,25 @@ struct r300_context { struct r300_atom rs_state; /* RS block state. */ struct r300_atom rs_block_state; - /* Sampler states. */ - struct r300_sampler_state* sampler_states[8]; - int sampler_count; /* Scissor state. */ struct r300_atom scissor_state; /* Sampler view states. */ struct pipe_sampler_view* fragment_sampler_views[8]; int fragment_sampler_view_count; + /* Vertex stream formatting state. */ + struct r300_atom vertex_stream_state; + /* VAP (vertex shader) output mapping state. */ + struct r300_atom vap_output_state; /* Vertex shader. */ - struct r300_vertex_shader* vs; + struct r300_atom vs_state; /* Viewport state. */ struct r300_atom viewport_state; /* ZTOP state. */ struct r300_atom ztop_state; + /* PVS flush. */ + struct r300_atom pvs_flush; + /* Texture cache invalidate. */ + struct r300_atom texture_cache_inval; /* Invariant state. This must be emitted to get the engine started. */ struct r300_atom invariant_state; @@ -316,22 +366,29 @@ struct r300_context { /* Vertex buffers for Gallium. */ struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; int vertex_buffer_count; + int vertex_buffer_max_index; /* Vertex elements for Gallium. */ - struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; - int vertex_element_count; + struct r300_vertex_element_state *velems; + + /* Vertex info for Draw. */ + struct vertex_info vertex_info; struct pipe_stencil_ref stencil_ref; + struct pipe_clip_state clip; + + struct pipe_viewport_state viewport; + /* Bitmask of dirty state objects. */ uint32_t dirty_state; /* Flag indicating whether or not the HW is dirty. */ uint32_t dirty_hw; - /* Whether the TCL engine should be in bypass mode. */ - boolean tcl_bypass; /* Whether polygon offset is enabled. */ boolean polygon_offset_enabled; /* Z buffer bit depth. */ uint32_t zbuffer_bpp; + /* Whether scissor is enabled. */ + boolean scissor_enabled; }; /* Convenience cast wrapper. */ diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 4310ff1411..fc8a8a2773 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -32,17 +32,17 @@ #include "r300_emit.h" #include "r300_fs.h" #include "r300_screen.h" -#include "r300_state_inlines.h" #include "r300_vs.h" -void r300_emit_blend_state(struct r300_context* r300, void* state) +void r300_emit_blend_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_blend_state* blend = (struct r300_blend_state*)state; struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)r300->fb_state.state; CS_LOCALS(r300); - BEGIN_CS(8); + BEGIN_CS(size); OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop); OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); if (fb->nr_cbufs) { @@ -59,26 +59,28 @@ void r300_emit_blend_state(struct r300_context* r300, void* state) END_CS; } -void r300_emit_blend_color_state(struct r300_context* r300, void* state) +void r300_emit_blend_color_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state; struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); if (r300screen->caps->is_r500) { - BEGIN_CS(3); + BEGIN_CS(size); OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); OUT_CS(bc->blend_color_red_alpha); OUT_CS(bc->blend_color_green_blue); END_CS; } else { - BEGIN_CS(2); + BEGIN_CS(size); OUT_CS_REG(R300_RB3D_BLEND_COLOR, bc->blend_color); END_CS; } } -void r300_emit_clip_state(struct r300_context* r300, void* state) +void r300_emit_clip_state(struct r300_context* r300, + unsigned size, void* state) { struct pipe_clip_state* clip = (struct pipe_clip_state*)state; int i; @@ -86,7 +88,7 @@ void r300_emit_clip_state(struct r300_context* r300, void* state) CS_LOCALS(r300); if (r300screen->caps->has_tcl) { - BEGIN_CS(5 + (6 * 4)); + BEGIN_CS(size); OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, (r300screen->caps->is_r500 ? R500_PVS_UCP_START : R300_PVS_UCP_START)); @@ -101,14 +103,14 @@ void r300_emit_clip_state(struct r300_context* r300, void* state) R300_PS_UCP_MODE_CLIP_AS_TRIFAN); END_CS; } else { - BEGIN_CS(2); + BEGIN_CS(size); OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); END_CS; } } -void r300_emit_dsa_state(struct r300_context* r300, void* state) +void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state) { struct r300_dsa_state* dsa = (struct r300_dsa_state*)state; struct r300_screen* r300screen = r300_screen(r300->context.screen); @@ -117,7 +119,7 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state) struct pipe_stencil_ref stencil_ref = r300->stencil_ref; CS_LOCALS(r300); - BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6); + BEGIN_CS(size); OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); OUT_CS_REG_SEQ(R300_ZB_CNTL, 3); @@ -144,6 +146,8 @@ static const float * get_shader_constant( { struct r300_viewport_state* viewport = (struct r300_viewport_state*)r300->viewport_state.state; + struct r300_textures_state* texstate = + (struct r300_textures_state*)r300->textures_state.state; static float vec[4] = { 0.0, 0.0, 0.0, 1.0 }; struct pipe_texture *tex; @@ -171,23 +175,15 @@ static const float * get_shader_constant( break; case RC_STATE_R300_VIEWPORT_SCALE: - if (r300->tcl_bypass) { - vec[0] = 1; - vec[1] = 1; - vec[2] = 1; - } else { - vec[0] = viewport->xscale; - vec[1] = viewport->yscale; - vec[2] = viewport->zscale; - } + vec[0] = viewport->xscale; + vec[1] = viewport->yscale; + vec[2] = viewport->zscale; break; case RC_STATE_R300_VIEWPORT_OFFSET: - if (!r300->tcl_bypass) { - vec[0] = viewport->xoffset; - vec[1] = viewport->yoffset; - vec[2] = viewport->zoffset; - } + vec[0] = viewport->xoffset; + vec[1] = viewport->yoffset; + vec[2] = viewport->zoffset; break; default: @@ -379,7 +375,7 @@ void r500_emit_fs_constant_buffer(struct r300_context* r300, END_CS; } -void r300_emit_fb_state(struct r300_context* r300, void* state) +void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state) { struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state; struct r300_screen* r300screen = r300_screen(r300->context.screen); @@ -388,7 +384,7 @@ void r300_emit_fb_state(struct r300_context* r300, void* state) int i; CS_LOCALS(r300); - BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 6); + BEGIN_CS(size); /* Flush and free renderbuffer caches. */ OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, @@ -422,14 +418,13 @@ void r300_emit_fb_state(struct r300_context* r300, void* state) OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1); - OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] | - r300_translate_colorformat(tex->tex.format) | - R300_COLOR_TILE(tex->mip_macrotile[surf->level]) | - R300_COLOR_MICROTILE(tex->microtile), + OUT_CS_RELOC(tex->buffer, tex->fb_state.colorpitch[surf->level], 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), - r300_translate_out_fmt(surf->format)); + OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), tex->fb_state.us_out_fmt); + } + for (; i < 4; i++) { + OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED); } /* Set up a zbuffer. */ @@ -441,15 +436,15 @@ void r300_emit_fb_state(struct r300_context* r300, void* state) OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1); OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format)); + OUT_CS_REG(R300_ZB_FORMAT, tex->fb_state.zb_format); OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); - OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] | - R300_DEPTHMACROTILE(tex->mip_macrotile[surf->level]) | - R300_DEPTHMICROTILE(tex->microtile), + OUT_CS_RELOC(tex->buffer, tex->fb_state.depthpitch[surf->level], 0, RADEON_GEM_DOMAIN_VRAM, 0); } + OUT_CS_REG(R300_GA_POINT_MINMAX, + (MAX2(fb->width, fb->height) * 6) << R300_GA_POINT_MINMAX_MAX_SHIFT); END_CS; } @@ -579,21 +574,19 @@ void r300_emit_query_end(struct r300_context* r300) r300_emit_query_finish(r300, query); } -void r300_emit_rs_state(struct r300_context* r300, void* state) +void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state) { struct r300_rs_state* rs = (struct r300_rs_state*)state; float scale, offset; CS_LOCALS(r300); - BEGIN_CS(18 + (rs->polygon_offset_enable ? 5 : 0)); + BEGIN_CS(size); OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status); OUT_CS_REG(R300_GB_AA_CONFIG, rs->antialiasing_config); OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size); - OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2); - OUT_CS(rs->point_minmax); - OUT_CS(rs->line_control); + OUT_CS_REG(R300_GA_LINE_CNTL, rs->line_control); if (rs->polygon_offset_enable) { scale = rs->depth_scale * 12; @@ -624,7 +617,8 @@ void r300_emit_rs_state(struct r300_context* r300, void* state) END_CS; } -void r300_emit_rs_block_state(struct r300_context* r300, void* state) +void r300_emit_rs_block_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_rs_block* rs = (struct r300_rs_block*)state; unsigned i; @@ -635,7 +629,7 @@ void r300_emit_rs_block_state(struct r300_context* r300, void* state) DBG(r300, DBG_DRAW, "r300: RS emit:\n"); - BEGIN_CS(5 + count*2); + BEGIN_CS(size); if (r300screen->caps->is_r500) { OUT_CS_REG_SEQ(R500_RS_IP_0, count); } else { @@ -666,7 +660,8 @@ void r300_emit_rs_block_state(struct r300_context* r300, void* state) END_CS; } -void r300_emit_scissor_state(struct r300_context* r300, void* state) +void r300_emit_scissor_state(struct r300_context* r300, + unsigned size, void* state) { unsigned minx, miny, maxx, maxy; uint32_t top_left, bottom_right; @@ -680,7 +675,7 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state) maxx = fb->width; maxy = fb->height; - if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) { + if (r300->scissor_enabled) { minx = MAX2(minx, scissor->minx); miny = MAX2(miny, scissor->miny); maxx = MIN2(maxx, scissor->maxx); @@ -720,65 +715,51 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state) (((maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); } - BEGIN_CS(3); + BEGIN_CS(size); OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); OUT_CS(top_left); OUT_CS(bottom_right); END_CS; } -void r300_emit_texture(struct r300_context* r300, - struct r300_sampler_state* sampler, - struct r300_texture* tex, - unsigned offset) +void r300_emit_textures_state(struct r300_context *r300, + unsigned size, void *state) { - uint32_t filter0 = sampler->filter0; - uint32_t format0 = tex->state.format0; - unsigned min_level, max_level; + struct r300_textures_state *allstate = (struct r300_textures_state*)state; + struct r300_texture_sampler_state *texstate; + unsigned i; CS_LOCALS(r300); - /* to emulate 1D textures through 2D ones correctly */ - if (tex->tex.target == PIPE_TEXTURE_1D) { - filter0 &= ~R300_TX_WRAP_T_MASK; - filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); - } + BEGIN_CS(size); + OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable); - if (tex->is_npot) { - /* NPOT textures don't support mip filter, unfortunately. - * This prevents incorrect rendering. */ - filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK; - } else { - /* determine min/max levels */ - /* the MAX_MIP level is the largest (finest) one */ - max_level = MIN2(sampler->max_lod, tex->tex.last_level); - min_level = MIN2(sampler->min_lod, max_level); - format0 |= R300_TX_NUM_LEVELS(max_level); - filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); - } + for (i = 0; i < allstate->count; i++) { + if ((1 << i) & allstate->tx_enable) { + texstate = &allstate->regs[i]; + + OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter[0]); + OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter[1]); + OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4), + texstate->border_color); - BEGIN_CS(16); - OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 | - (offset << 28)); - OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); - OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); - - OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0); - OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1); - OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2); - OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1); - OUT_CS_RELOC(tex->buffer, - R300_TXO_MACRO_TILE(tex->macrotile) | - R300_TXO_MICRO_TILE(tex->microtile), - RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0); + OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format[0]); + OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format[1]); + OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format[2]); + + OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_CS_RELOC(allstate->textures[i]->buffer, texstate->tile_config, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0); + } + } END_CS; } void r300_emit_aos(struct r300_context* r300, unsigned offset) { struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer; - struct pipe_vertex_element *velem = r300->vertex_element; + struct pipe_vertex_element *velem = r300->velems->velem; int i; - unsigned size1, size2, aos_count = r300->vertex_element_count; + unsigned size1, size2, aos_count = r300->velems->count; unsigned packet_size = (aos_count * 3 + 1) / 2; CS_LOCALS(r300); @@ -813,68 +794,110 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset) END_CS; } -void r300_emit_vertex_format_state(struct r300_context* r300, void* state) +void r300_emit_vertex_buffer(struct r300_context* r300) { - struct r300_vertex_info* vertex_info = (struct r300_vertex_info*)state; - unsigned i; CS_LOCALS(r300); - DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n"); + DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, " + "vertex size %d\n", r300->vbo, + r300->vertex_info.size); + /* Set the pointer to our vertex buffer. The emitted values are this: + * PACKET3 [3D_LOAD_VBPNTR] + * COUNT [1] + * FORMAT [size | stride << 8] + * OFFSET [offset into BO] + * VBPNTR [relocated BO] + */ + BEGIN_CS(7); + OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); + OUT_CS(1); + OUT_CS(r300->vertex_info.size | + (r300->vertex_info.size << 8)); + OUT_CS(r300->vbo_offset); + OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); + END_CS; +} - BEGIN_CS(26); - OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_info->vinfo.size); +void r300_emit_vertex_stream_state(struct r300_context* r300, + unsigned size, void* state) +{ + struct r300_vertex_stream_state *streams = + (struct r300_vertex_stream_state*)state; + unsigned i; + CS_LOCALS(r300); - OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2); - OUT_CS(vertex_info->vinfo.hwfmt[0]); - OUT_CS(vertex_info->vinfo.hwfmt[1]); - OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); - OUT_CS(vertex_info->vinfo.hwfmt[2]); - OUT_CS(vertex_info->vinfo.hwfmt[3]); - for (i = 0; i < 4; i++) { - DBG(r300, DBG_DRAW, " : hwfmt%d: 0x%08x\n", i, - vertex_info->vinfo.hwfmt[i]); - } + DBG(r300, DBG_DRAW, "r300: PSC emit:\n"); - OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8); - for (i = 0; i < 8; i++) { - OUT_CS(vertex_info->vap_prog_stream_cntl[i]); + BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count); + for (i = 0; i < streams->count; i++) { + OUT_CS(streams->vap_prog_stream_cntl[i]); DBG(r300, DBG_DRAW, " : prog_stream_cntl%d: 0x%08x\n", i, - vertex_info->vap_prog_stream_cntl[i]); + streams->vap_prog_stream_cntl[i]); } - OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8); - for (i = 0; i < 8; i++) { - OUT_CS(vertex_info->vap_prog_stream_cntl_ext[i]); + OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count); + for (i = 0; i < streams->count; i++) { + OUT_CS(streams->vap_prog_stream_cntl_ext[i]); DBG(r300, DBG_DRAW, " : prog_stream_cntl_ext%d: 0x%08x\n", i, - vertex_info->vap_prog_stream_cntl_ext[i]); + streams->vap_prog_stream_cntl_ext[i]); } END_CS; } +void r300_emit_vap_output_state(struct r300_context* r300, + unsigned size, void* state) +{ + struct r300_vap_output_state *vap_out_state = + (struct r300_vap_output_state*)state; + CS_LOCALS(r300); + + DBG(r300, DBG_DRAW, "r300: VAP emit:\n"); -void r300_emit_vertex_program_code(struct r300_context* r300, - struct r300_vertex_program_code* code) + BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2); + OUT_CS(vap_out_state->vap_vtx_state_cntl); + OUT_CS(vap_out_state->vap_vsm_vtx_assm); + OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); + OUT_CS(vap_out_state->vap_out_vtx_fmt[0]); + OUT_CS(vap_out_state->vap_out_vtx_fmt[1]); + END_CS; +} + +void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state) { - int i; + CS_LOCALS(r300); + + BEGIN_CS(size); + OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); + END_CS; +} + +void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) +{ + struct r300_vertex_shader* vs = (struct r300_vertex_shader*)state; + struct r300_vertex_program_code* code = &vs->code; struct r300_screen* r300screen = r300_screen(r300->context.screen); unsigned instruction_count = code->length / 4; + unsigned i; + + unsigned vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72; + unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1); + unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1); + unsigned temp_count = MAX2(code->num_temporaries, 1); - int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72; - int input_count = MAX2(util_bitcount(code->InputsRead), 1); - int output_count = MAX2(util_bitcount(code->OutputsWritten), 1); - int temp_count = MAX2(code->num_temporaries, 1); - int pvs_num_slots = MIN3(vtx_mem_size / input_count, - vtx_mem_size / output_count, 10); - int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count, + vtx_mem_size / output_count, 10); + unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { - debug_printf("r300: Implementation error: emit_vertex_shader called," + debug_printf("r300: Implementation error: emit_vs_state called," " but has_tcl is FALSE!\n"); return; } - BEGIN_CS(9 + code->length); + BEGIN_CS(size); /* R300_VAP_PVS_CODE_CNTL_0 * R300_VAP_PVS_CONST_CNTL * R300_VAP_PVS_CODE_CNTL_1 @@ -888,8 +911,9 @@ void r300_emit_vertex_program_code(struct r300_context* r300, OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0); OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length); - for (i = 0; i < code->length; i++) + for (i = 0; i < code->length; i++) { OUT_CS(code->body.d[i]); + } OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) | R300_PVS_NUM_CNTLRS(pvs_num_controllers) | @@ -899,12 +923,6 @@ void r300_emit_vertex_program_code(struct r300_context* r300, END_CS; } -void r300_emit_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs) -{ - r300_emit_vertex_program_code(r300, &vs->code); -} - void r300_emit_vs_constant_buffer(struct r300_context* r300, struct rc_constant_list* constants) { @@ -913,7 +931,7 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300, CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { - debug_printf("r300: Implementation error: emit_vertex_shader called," + debug_printf("r300: Implementation error: emit_vs_constant_buffer called," " but has_tcl is FALSE!\n"); return; } @@ -938,83 +956,56 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300, END_CS; } -void r300_emit_viewport_state(struct r300_context* r300, void* state) +void r300_emit_viewport_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_viewport_state* viewport = (struct r300_viewport_state*)state; CS_LOCALS(r300); - if (r300->tcl_bypass) { - BEGIN_CS(2); - OUT_CS_REG(R300_VAP_VTE_CNTL, 0); - END_CS; - } else { - BEGIN_CS(9); - OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); - OUT_CS_32F(viewport->xscale); - OUT_CS_32F(viewport->xoffset); - OUT_CS_32F(viewport->yscale); - OUT_CS_32F(viewport->yoffset); - OUT_CS_32F(viewport->zscale); - OUT_CS_32F(viewport->zoffset); - OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); - END_CS; - } -} - -void r300_emit_texture_count(struct r300_context* r300) -{ - uint32_t tx_enable = 0; - int i; - CS_LOCALS(r300); - - /* Notice that fragment_sampler_view_count and sampler_count are just sizes - * of the respective arrays. We still have to check for the individual - * elements. */ - for (i = 0; i < MIN2(r300->sampler_count, r300->fragment_sampler_view_count); i++) { - if (r300->fragment_sampler_views[i]) { - tx_enable |= 1 << i; - } - } - - BEGIN_CS(2); - OUT_CS_REG(R300_TX_ENABLE, tx_enable); - END_CS; - + BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); + OUT_CS_32F(viewport->xscale); + OUT_CS_32F(viewport->xoffset); + OUT_CS_32F(viewport->yscale); + OUT_CS_32F(viewport->yoffset); + OUT_CS_32F(viewport->zscale); + OUT_CS_32F(viewport->zoffset); + OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + END_CS; } -void r300_emit_ztop_state(struct r300_context* r300, void* state) +void r300_emit_ztop_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_ztop_state* ztop = (struct r300_ztop_state*)state; CS_LOCALS(r300); - BEGIN_CS(2); + BEGIN_CS(size); OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top); END_CS; } -void r300_flush_textures(struct r300_context* r300) +void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state) { CS_LOCALS(r300); - BEGIN_CS(2); + BEGIN_CS(size); OUT_CS_REG(R300_TX_INVALTAGS, 0); END_CS; } -static void r300_flush_pvs(struct r300_context* r300) -{ - CS_LOCALS(r300); - - BEGIN_CS(2); - OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); - END_CS; -} - -void r300_emit_buffer_validate(struct r300_context *r300) +void r300_emit_buffer_validate(struct r300_context *r300, + boolean do_validate_vertex_buffers, + struct pipe_buffer *index_buffer) { struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_textures_state *texstate = + (struct r300_textures_state*)r300->textures_state.state; struct r300_texture* tex; + struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->velems->velem; + struct pipe_buffer *pbuf; unsigned i; boolean invalid = FALSE; @@ -1061,16 +1052,35 @@ validate: goto validate; } } - /* ...and vertex buffer. */ + /* ...vertex buffer for SWTCL path... */ if (r300->vbo) { if (!r300->winsys->add_buffer(r300->winsys, r300->vbo, RADEON_GEM_DOMAIN_GTT, 0)) { r300->context.flush(&r300->context, 0, NULL); goto validate; } - } else { - /* debug_printf("No VBO while emitting dirty state!\n"); */ } + /* ...vertex buffers for HWTCL path... */ + if (do_validate_vertex_buffers) { + for (i = 0; i < r300->velems->count; i++) { + pbuf = vbuf[velem[i].vertex_buffer_index].buffer; + + if (!r300->winsys->add_buffer(r300->winsys, pbuf, + RADEON_GEM_DOMAIN_GTT, 0)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } + } + /* ...and index buffer for HWTCL path. */ + if (index_buffer) { + if (!r300->winsys->add_buffer(r300->winsys, index_buffer, + RADEON_GEM_DOMAIN_GTT, 0)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } + } + if (!r300->winsys->validate(r300->winsys)) { r300->context.flush(&r300->context, 0, NULL); if (invalid) { @@ -1083,16 +1093,10 @@ validate: } } -/* Emit all dirty state. */ -void r300_emit_dirty_state(struct r300_context* r300) +unsigned r300_get_num_dirty_dwords(struct r300_context *r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); struct r300_atom* atom; - unsigned i, dwords = 1024; - int dirty_tex = 0; - - /* Check the required number of dwords against the space remaining in the - * current CS object. If we need more, then flush. */ + unsigned dwords = 0; foreach(atom, &r300->atom_list) { if (atom->dirty || atom->always_dirty) { @@ -1100,12 +1104,17 @@ void r300_emit_dirty_state(struct r300_context* r300) } } - /* Make sure we have at least 2*1024 spare dwords. */ - /* XXX It would be nice to know the number of dwords we really need to - * XXX emit. */ - while (!r300->winsys->check_cs(r300->winsys, dwords)) { - r300->context.flush(&r300->context, 0, NULL); - } + /* XXX This is the compensation for the non-atomized states. */ + dwords += 1024; + + return dwords; +} + +/* Emit all dirty state. */ +void r300_emit_dirty_state(struct r300_context* r300) +{ + struct r300_screen* r300screen = r300_screen(r300->context.screen); + struct r300_atom* atom; if (r300->dirty_state & R300_NEW_QUERY) { r300_emit_query_start(r300); @@ -1114,7 +1123,7 @@ void r300_emit_dirty_state(struct r300_context* r300) foreach(atom, &r300->atom_list) { if (atom->dirty || atom->always_dirty) { - atom->emit(r300, atom->state); + atom->emit(r300, atom->size, atom->state); atom->dirty = FALSE; } } @@ -1140,43 +1149,9 @@ void r300_emit_dirty_state(struct r300_context* r300) r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS; } - /* Samplers and textures are tracked separately but emitted together. */ - if (r300->dirty_state & - (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) { - r300_emit_texture_count(r300); - - for (i = 0; i < MIN2(r300->sampler_count, r300->fragment_sampler_view_count); i++) { - if (r300->dirty_state & - ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) { - if (r300->fragment_sampler_views[i]) { - r300_emit_texture(r300, - r300->sampler_states[i], - (struct r300_texture *)r300->fragment_sampler_views[i]->texture, - i); - dirty_tex |= r300->dirty_state & (R300_NEW_TEXTURE << i); - } - r300->dirty_state &= - ~((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i)); - } - } - r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES); - } - - if (dirty_tex) { - r300_flush_textures(r300); - } - - if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) { - r300_flush_pvs(r300); - } - - if (r300->dirty_state & R300_NEW_VERTEX_SHADER) { - r300_emit_vertex_shader(r300, r300->vs); - r300->dirty_state &= ~R300_NEW_VERTEX_SHADER; - } - if (r300->dirty_state & R300_NEW_VERTEX_SHADER_CONSTANTS) { - r300_emit_vs_constant_buffer(r300, &r300->vs->code.constants); + struct r300_vertex_shader* vs = r300->vs_state.state; + r300_emit_vs_constant_buffer(r300, &vs->code.constants); r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS; } @@ -1184,8 +1159,10 @@ void r300_emit_dirty_state(struct r300_context* r300) assert(r300->dirty_state == 0); */ - /* Finally, emit the VBO. */ - /* r300_emit_vertex_buffer(r300); */ + /* Emit the VBO for SWTCL. */ + if (!r300screen->caps->has_tcl) { + r300_emit_vertex_buffer(r300); + } r300->dirty_hw++; } diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h index 6b96d9b57c..449e640a88 100644 --- a/src/gallium/drivers/r300/r300_emit.h +++ b/src/gallium/drivers/r300/r300_emit.h @@ -31,13 +31,17 @@ struct r300_vertex_program_code; void r300_emit_aos(struct r300_context* r300, unsigned offset); -void r300_emit_blend_state(struct r300_context* r300, void* state); +void r300_emit_blend_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_blend_color_state(struct r300_context* r300, void* state); +void r300_emit_blend_color_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_clip_state(struct r300_context* r300, void* state); +void r300_emit_clip_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_dsa_state(struct r300_context* r300, void* state); +void r300_emit_dsa_state(struct r300_context* r300, + unsigned size, void* state); void r300_emit_fragment_program_code(struct r300_context* r300, struct rX00_fragment_program_code* generic_code); @@ -51,48 +55,54 @@ void r500_emit_fragment_program_code(struct r300_context* r300, void r500_emit_fs_constant_buffer(struct r300_context* r300, struct rc_constant_list* constants); -void r300_emit_fb_state(struct r300_context* r300, void* state); +void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state); void r300_emit_query_begin(struct r300_context* r300, struct r300_query* query); void r300_emit_query_end(struct r300_context* r300); -void r300_emit_rs_state(struct r300_context* r300, void* state); +void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state); -void r300_emit_rs_block_state(struct r300_context* r300, void* state); +void r300_emit_rs_block_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_scissor_state(struct r300_context* r300, void* state); +void r300_emit_scissor_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_texture(struct r300_context* r300, - struct r300_sampler_state* sampler, - struct r300_texture* tex, - unsigned offset); +void r300_emit_textures_state(struct r300_context *r300, + unsigned size, void *state); void r300_emit_vertex_buffer(struct r300_context* r300); -void r300_emit_vertex_format_state(struct r300_context* r300, void* state); +void r300_emit_vertex_stream_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_vertex_program_code(struct r300_context* r300, - struct r300_vertex_program_code* code); +void r300_emit_vap_output_state(struct r300_context* r300, + unsigned size, void* state); void r300_emit_vs_constant_buffer(struct r300_context* r300, struct rc_constant_list* constants); -void r300_emit_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs); +void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state); -void r300_emit_viewport_state(struct r300_context* r300, void* state); +void r300_emit_viewport_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_texture_count(struct r300_context* r300); +void r300_emit_ztop_state(struct r300_context* r300, + unsigned size, void* state); -void r300_emit_ztop_state(struct r300_context* r300, void* state); +void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state); -void r300_flush_textures(struct r300_context* r300); +void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state); + +unsigned r300_get_num_dirty_dwords(struct r300_context *r300); /* Emit all dirty state. */ void r300_emit_dirty_state(struct r300_context* r300); -void r300_emit_buffer_validate(struct r300_context *r300); +void r300_emit_buffer_validate(struct r300_context *r300, + boolean do_validate_vertex_buffers, + struct pipe_buffer *index_buffer); #endif /* R300_EMIT_H */ diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index e37d309270..70de152713 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -61,6 +61,12 @@ static void r300_flush(struct pipe_context* pipe, atom->dirty = TRUE; } } + + /* Unmark HWTCL state for SWTCL. */ + if (!r300_screen(pipe->screen)->caps->has_tcl) { + r300->vs_state.dirty = FALSE; + r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS; + } } /* reset flushed query */ diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c index ae4c62b2f1..9e71e61c30 100644 --- a/src/gallium/drivers/r300/r300_fs.c +++ b/src/gallium/drivers/r300/r300_fs.c @@ -133,10 +133,13 @@ static void get_compare_state( struct r300_fragment_program_external_state* state, unsigned shadow_samplers) { + struct r300_textures_state *texstate = + (struct r300_textures_state*)r300->textures_state.state; + memset(state, 0, sizeof(*state)); - for (int i = 0; i < r300->sampler_count; i++) { - struct r300_sampler_state* s = r300->sampler_states[i]; + for (int i = 0; i < texstate->sampler_count; i++) { + struct r300_sampler_state* s = texstate->sampler_states[i]; if (s && s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { /* XXX Gallium doesn't provide us with any information regarding @@ -204,6 +207,7 @@ static void r300_translate_fragment_shader( DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n", compiler.Base.ErrorMsg); assert(0); + abort(); } /* And, finally... */ diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index a049da69e2..a249e8b36b 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -113,11 +113,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* index size - when not set the indices are assumed to be 16 bit */ # define R300_VAP_VF_CNTL__INDEX_SIZE_32bit (1<<11) +# define R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS (1<<14) /* number of vertices */ # define R300_VAP_VF_CNTL__NUM_VERTICES__SHIFT 16 #define R500_VAP_INDEX_OFFSET 0x208c +#define R500_VAP_ALT_NUM_VERTICES 0x2088 + #define R300_VAP_OUTPUT_VTX_FMT_0 0x2090 # define R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT (1<<0) # define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT (1<<1) diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index cd4971ae13..9c001ae186 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -42,6 +42,9 @@ /* r300_render: Vertex and index buffer primitive emission. */ #define R300_MAX_VBO_SIZE (1024 * 1024) +/* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */ +//#define ENABLE_ALT_NUM_VERTS + uint32_t r300_translate_primitive(unsigned prim) { switch (prim) { @@ -115,6 +118,18 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, return color_control; } +/* Check if the requested number of dwords is available in the CS and + * if not, flush. Return TRUE if the flush occured. */ +static boolean r300_reserve_cs_space(struct r300_context *r300, + unsigned dwords) +{ + if (!r300->winsys->check_cs(r300->winsys, dwords)) { + r300->context.flush(&r300->context, 0, NULL); + return TRUE; + } + return FALSE; +} + static boolean immd_is_good_idea(struct r300_context *r300, unsigned count) { @@ -128,8 +143,8 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, { struct pipe_vertex_element* velem; struct pipe_vertex_buffer* vbuf; - unsigned vertex_element_count = r300->vertex_element_count; - unsigned i, v, vbi, dw, elem_offset; + unsigned vertex_element_count = r300->velems->count; + unsigned i, v, vbi, dw, elem_offset, dwords; /* Size of the vertex, in dwords. */ unsigned vertex_size = 0; @@ -151,7 +166,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ for (i = 0; i < vertex_element_count; i++) { - velem = &r300->vertex_element[i]; + velem = &r300->velems->velem[i]; offset[i] = velem->src_offset / 4; size[i] = util_format_get_blocksize(velem->src_format) / 4; vertex_size += size[i]; @@ -168,14 +183,19 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, } } + dwords = 9 + count * vertex_size; + + r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords); + r300_emit_buffer_validate(r300, FALSE, NULL); r300_emit_dirty_state(r300); - BEGIN_CS(10 + count * vertex_size); + BEGIN_CS(dwords); OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); - OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); + OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); + OUT_CS(count - 1); + OUT_CS(0); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | r300_translate_primitive(mode)); @@ -183,7 +203,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, /* Emit vertices. */ for (v = 0; v < count; v++) { for (i = 0; i < vertex_element_count; i++) { - velem = &r300->vertex_element[i]; + velem = &r300->velems->velem[i]; vbi = velem->vertex_buffer_index; elem_offset = offset[i] + stride[vbi] * (v + start); @@ -196,7 +216,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300, /* Unmap buffers. */ for (i = 0; i < vertex_element_count; i++) { - vbi = r300->vertex_element[i].vertex_buffer_index; + vbi = r300->velems->velem[i].vertex_buffer_index; if (map[vbi]) { vbuf = &r300->vertex_buffer[vbi]; @@ -210,16 +230,29 @@ static void r300_emit_draw_arrays(struct r300_context *r300, unsigned mode, unsigned count) { +#if defined(ENABLE_ALT_NUM_VERTS) + boolean alt_num_verts = count > 65535; +#else + boolean alt_num_verts = FALSE; +#endif CS_LOCALS(r300); - BEGIN_CS(8); + if (alt_num_verts) { + assert(count < (1 << 24)); + BEGIN_CS(9); + OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); + } else { + BEGIN_CS(7); + } OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); - OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); + OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); + OUT_CS(count - 1); + OUT_CS(0); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | - r300_translate_primitive(mode)); + r300_translate_primitive(mode) | + (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); END_CS; } @@ -234,72 +267,61 @@ static void r300_emit_draw_elements(struct r300_context *r300, { uint32_t count_dwords; uint32_t offset_dwords = indexSize * start / sizeof(uint32_t); +#if defined(ENABLE_ALT_NUM_VERTS) + boolean alt_num_verts = count > 65535; +#else + boolean alt_num_verts = FALSE; +#endif CS_LOCALS(r300); - /* XXX most of these are stupid */ - assert(indexSize == 4 || indexSize == 2); assert((start * indexSize) % 4 == 0); - assert(offset_dwords == 0); + assert(count < (1 << 24)); + + DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n", + count, minIndex, maxIndex); - BEGIN_CS(14); + maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index); + + if (alt_num_verts) { + BEGIN_CS(15); + OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); + } else { + BEGIN_CS(13); + } OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); - OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex); + OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); + OUT_CS(maxIndex); + OUT_CS(minIndex); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); if (indexSize == 4) { - count_dwords = count + start; + count_dwords = count; OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | R300_VAP_VF_CNTL__INDEX_SIZE_32bit | - r300_translate_primitive(mode)); + r300_translate_primitive(mode) | + (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); } else { - count_dwords = (count + start + 1) / 2; + count_dwords = (count + 1) / 2; OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | - r300_translate_primitive(mode)); + r300_translate_primitive(mode) | + (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); } /* INDX_BUFFER is a truly special packet3. * Unlike most other packet3, where the offset is after the count, * the order is reversed, so the relocation ends up carrying the * size of the indexbuf instead of the offset. - * - * XXX Fix offset */ OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | (0 << R300_INDX_BUFFER_SKIP_SHIFT)); - OUT_CS(offset_dwords); + OUT_CS(offset_dwords << 2); OUT_CS_RELOC(indexBuffer, count_dwords, RADEON_GEM_DOMAIN_GTT, 0, 0); END_CS; } -static boolean r300_setup_vertex_buffers(struct r300_context *r300) -{ - struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; - struct pipe_vertex_element *velem = r300->vertex_element; - struct pipe_buffer *pbuf; - -validate: - for (int i = 0; i < r300->vertex_element_count; i++) { - pbuf = vbuf[velem[i].vertex_buffer_index].buffer; - - if (!r300->winsys->add_buffer(r300->winsys, pbuf, - RADEON_GEM_DOMAIN_GTT, 0)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } - } - - if (!r300->winsys->validate(r300->winsys)) { - r300->context.flush(&r300->context, 0, NULL); - return r300->winsys->validate(r300->winsys); - } - - return TRUE; -} - static void r300_shorten_ubyte_elts(struct r300_context* r300, struct pipe_buffer** elts, unsigned count) @@ -343,48 +365,52 @@ void r300_draw_range_elements(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); struct pipe_buffer* orgIndexBuffer = indexBuffer; +#if defined(ENABLE_ALT_NUM_VERTS) + boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 && + count > 65536; +#else + boolean alt_num_verts = FALSE; +#endif + unsigned short_count; if (!u_trim_pipe_prim(mode, &count)) { return; } - if (count > 65535) { - /* XXX: use aux/indices functions to split this into smaller - * primitives. - */ - return; - } - - r300_update_derived_state(r300); - - r300_emit_buffer_validate(r300); - - if (!r300_setup_vertex_buffers(r300)) { - return; - } - if (indexSize == 1) { r300_shorten_ubyte_elts(r300, &indexBuffer, count); indexSize = 2; } - if (!r300->winsys->add_buffer(r300->winsys, indexBuffer, - RADEON_GEM_DOMAIN_GTT, 0)) { - goto cleanup; - } - - if (!r300->winsys->validate(r300->winsys)) { - goto cleanup; - } + r300_update_derived_state(r300); + /* 128 dwords for emit_aos and emit_draw_elements */ + r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128); + r300_emit_buffer_validate(r300, TRUE, indexBuffer); r300_emit_dirty_state(r300); - r300_emit_aos(r300, 0); - r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex, - mode, start, count); + if (alt_num_verts || count <= 65535) { + r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, + maxIndex, mode, start, count); + } else { + do { + short_count = MIN2(count, 65534); + r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, + maxIndex, mode, start, short_count); + + start += short_count; + count -= short_count; + + /* 16 spare dwords are enough for emit_draw_elements. */ + if (count && r300_reserve_cs_space(r300, 16)) { + r300_emit_buffer_validate(r300, TRUE, indexBuffer); + r300_emit_dirty_state(r300); + r300_emit_aos(r300, 0); + } + } while (count); + } -cleanup: if (indexBuffer != orgIndexBuffer) { pipe->screen->buffer_destroy(indexBuffer); } @@ -396,40 +422,60 @@ void r300_draw_elements(struct pipe_context* pipe, unsigned indexSize, unsigned mode, unsigned start, unsigned count) { - pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0, - mode, start, count); + struct r300_context *r300 = r300_context(pipe); + + pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, + r300->vertex_buffer_max_index, + mode, start, count); } void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, unsigned start, unsigned count) { struct r300_context* r300 = r300_context(pipe); +#if defined(ENABLE_ALT_NUM_VERTS) + boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 && + count > 65536; +#else + boolean alt_num_verts = FALSE; +#endif + unsigned short_count; if (!u_trim_pipe_prim(mode, &count)) { return; } - if (count > 65535) { - /* XXX: driver needs to handle this -- use the functions in - * aux/indices to split this into several smaller primitives. - */ - return; - } - r300_update_derived_state(r300); - r300_emit_buffer_validate(r300); - if (immd_is_good_idea(r300, count)) { r300_emit_draw_arrays_immediate(r300, mode, start, count); } else { - if (!r300_setup_vertex_buffers(r300)) { - return; - } - + /* Make sure there are at least 128 spare dwords in the command buffer. + * (most of it being consumed by emit_aos) */ + r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128); + r300_emit_buffer_validate(r300, TRUE, NULL); r300_emit_dirty_state(r300); - r300_emit_aos(r300, start); - r300_emit_draw_arrays(r300, mode, count); + + if (alt_num_verts || count <= 65535) { + r300_emit_aos(r300, start); + r300_emit_draw_arrays(r300, mode, count); + } else { + do { + short_count = MIN2(count, 65535); + r300_emit_aos(r300, start); + r300_emit_draw_arrays(r300, mode, short_count); + + start += short_count; + count -= short_count; + + /* Again, we emit both AOS and draw_arrays so there should be + * at least 128 spare dwords. */ + if (count && r300_reserve_cs_space(r300, 128)) { + r300_emit_buffer_validate(r300, TRUE, NULL); + r300_emit_dirty_state(r300); + } + } while (count); + } } } @@ -559,7 +605,7 @@ r300_render_get_vertex_info(struct vbuf_render* render) r300_update_derived_state(r300); - return (struct vertex_info*)r300->vertex_format_state.state; + return &r300->vertex_info; } static boolean r300_render_allocate_vertices(struct vbuf_render* render, @@ -644,6 +690,8 @@ static void r300_render_draw_arrays(struct vbuf_render* render, CS_LOCALS(r300); + r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2); + r300_emit_buffer_validate(r300, FALSE, NULL); r300_emit_dirty_state(r300); DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count); @@ -662,12 +710,15 @@ static void r300_render_draw(struct vbuf_render* render, struct r300_render* r300render = r300_render(render); struct r300_context* r300 = r300render->r300; int i; + unsigned dwords = 2 + (count+1)/2; CS_LOCALS(r300); + r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords); + r300_emit_buffer_validate(r300, FALSE, NULL); r300_emit_dirty_state(r300); - BEGIN_CS(2 + (count+1)/2); + BEGIN_CS(dwords); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | r300render->hwprim); diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c index 5fa1ea0850..5880eecd5f 100644 --- a/src/gallium/drivers/r300/r300_screen.c +++ b/src/gallium/drivers/r300/r300_screen.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2010 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -20,14 +21,13 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "util/u_inlines.h" #include "util/u_format.h" #include "util/u_memory.h" #include "util/u_simple_screen.h" #include "r300_context.h" -#include "r300_screen.h" #include "r300_texture.h" +#include "r300_transfer.h" #include "radeon_winsys.h" #include "r300_winsys.h" @@ -210,8 +210,9 @@ static boolean r300_is_format_supported(struct pipe_screen* screen, { uint32_t retval = 0; boolean is_r500 = r300_screen(screen)->caps->is_r500; - boolean is_z24 = format == PIPE_FORMAT_Z24X8_UNORM || - format == PIPE_FORMAT_Z24S8_UNORM; + boolean is_z24 = format == PIPE_FORMAT_X8Z24_UNORM || + format == PIPE_FORMAT_S8Z24_UNORM; + boolean is_color2101010 = format == PIPE_FORMAT_R10G10B10A2_UNORM; if (target >= PIPE_MAX_TEXTURE_TYPES) { debug_printf("r300: Implementation error: Received bogus texture " @@ -221,114 +222,34 @@ static boolean r300_is_format_supported(struct pipe_screen* screen, /* Check sampler format support. */ if ((usage & PIPE_TEXTURE_USAGE_SAMPLER) && - (is_r500 || !is_z24) && /* Z24 cannot be sampled from on non-r5xx. */ - r300_translate_texformat(format) != ~0) { + /* Z24 cannot be sampled from on non-r5xx. */ + (is_r500 || !is_z24) && + r300_is_sampler_format_supported(format)) { retval |= PIPE_TEXTURE_USAGE_SAMPLER; } - switch (format) { - /* Supported formats. */ - /* Colorbuffer */ - case PIPE_FORMAT_A8_UNORM: - case PIPE_FORMAT_L8_UNORM: - case PIPE_FORMAT_R5G6B5_UNORM: - case PIPE_FORMAT_A1R5G5B5_UNORM: - case PIPE_FORMAT_A4R4G4B4_UNORM: - case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_X8R8G8B8_UNORM: - case PIPE_FORMAT_R8G8B8A8_UNORM: - case PIPE_FORMAT_R8G8B8X8_UNORM: - case PIPE_FORMAT_I8_UNORM: - retval |= usage & - (PIPE_TEXTURE_USAGE_RENDER_TARGET | - PIPE_TEXTURE_USAGE_DISPLAY_TARGET | - PIPE_TEXTURE_USAGE_PRIMARY); - break; - - /* ZS buffer */ - case PIPE_FORMAT_Z16_UNORM: - case PIPE_FORMAT_Z24X8_UNORM: - case PIPE_FORMAT_Z24S8_UNORM: - retval = usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL; - break; - - /* XXX Add all remaining gallium-supported formats, - * see util/u_format.csv. */ - - default:; + /* Check colorbuffer format support. */ + if ((usage & (PIPE_TEXTURE_USAGE_RENDER_TARGET | + PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_SCANOUT | + PIPE_TEXTURE_USAGE_SHARED)) && + /* 2101010 cannot be rendered to on non-r5xx. */ + (is_r500 || !is_color2101010) && + r300_is_colorbuffer_format_supported(format)) { + retval |= usage & + (PIPE_TEXTURE_USAGE_RENDER_TARGET | + PIPE_TEXTURE_USAGE_DISPLAY_TARGET | + PIPE_TEXTURE_USAGE_SCANOUT | + PIPE_TEXTURE_USAGE_SHARED); } - /* If usage was a mask that contained multiple bits, and not all of them - * are supported, this will catch that and return FALSE. - * e.g. usage = 2 | 4; retval = 4; (retval >= usage) == FALSE - * - * This also returns FALSE for any unknown formats. - */ - return (retval >= usage); -} - -static struct pipe_transfer* -r300_get_tex_transfer(struct pipe_screen *screen, - struct pipe_texture *texture, - unsigned face, unsigned level, unsigned zslice, - enum pipe_transfer_usage usage, unsigned x, unsigned y, - unsigned w, unsigned h) -{ - struct r300_texture *tex = (struct r300_texture *)texture; - struct r300_transfer *trans; - struct r300_screen *rscreen = r300_screen(screen); - unsigned offset; - - offset = r300_texture_get_offset(tex, level, zslice, face); /* in bytes */ - - trans = CALLOC_STRUCT(r300_transfer); - if (trans) { - pipe_texture_reference(&trans->transfer.texture, texture); - trans->transfer.x = x; - trans->transfer.y = y; - trans->transfer.width = w; - trans->transfer.height = h; - trans->transfer.stride = r300_texture_get_stride(rscreen, tex, level); - trans->transfer.usage = usage; - trans->transfer.zslice = zslice; - trans->transfer.face = face; - - trans->offset = offset; + /* Check depth-stencil format support. */ + if (usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL && + r300_is_zs_format_supported(format)) { + retval |= PIPE_TEXTURE_USAGE_DEPTH_STENCIL; } - return &trans->transfer; -} -static void -r300_tex_transfer_destroy(struct pipe_transfer *trans) -{ - pipe_texture_reference(&trans->texture, NULL); - FREE(trans); -} - -static void* r300_transfer_map(struct pipe_screen* screen, - struct pipe_transfer* transfer) -{ - struct r300_texture* tex = (struct r300_texture*)transfer->texture; - char* map; - enum pipe_format format = tex->tex.format; - - map = pipe_buffer_map(screen, tex->buffer, - pipe_transfer_buffer_flags(transfer)); - - if (!map) { - return NULL; - } - - return map + r300_transfer(transfer)->offset + - transfer->y / util_format_get_blockheight(format) * transfer->stride + - transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format); -} - -static void r300_transfer_unmap(struct pipe_screen* screen, - struct pipe_transfer* transfer) -{ - struct r300_texture* tex = (struct r300_texture*)transfer->texture; - pipe_buffer_unmap(screen, tex->buffer); + return retval == usage; } static void r300_destroy_screen(struct pipe_screen* pscreen) @@ -367,13 +288,11 @@ struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys) r300screen->screen.get_paramf = r300_get_paramf; r300screen->screen.is_format_supported = r300_is_format_supported; r300screen->screen.context_create = r300_create_context; - r300screen->screen.get_tex_transfer = r300_get_tex_transfer; - r300screen->screen.tex_transfer_destroy = r300_tex_transfer_destroy; - r300screen->screen.transfer_map = r300_transfer_map; - r300screen->screen.transfer_unmap = r300_transfer_unmap; r300_init_screen_texture_functions(&r300screen->screen); + r300_init_screen_transfer_functions(&r300screen->screen); u_simple_screen_init(&r300screen->screen); return &r300screen->screen; } + diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h index 502fbfa5a2..484bde6a6b 100644 --- a/src/gallium/drivers/r300/r300_screen.h +++ b/src/gallium/drivers/r300/r300_screen.h @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2010 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -27,6 +28,8 @@ #include "r300_chipset.h" +#define R300_TEXTURE_USAGE_TRANSFER PIPE_TEXTURE_USAGE_CUSTOM + struct radeon_winsys; struct r300_screen { @@ -35,6 +38,10 @@ struct r300_screen { struct radeon_winsys* radeon_winsys; + /* XXX This hack will be removed once texture transfers become part of + * pipe_context. */ + struct pipe_context* ctx; + /* Chipset capabilities */ struct r300_capabilities* caps; @@ -42,25 +49,14 @@ struct r300_screen { unsigned debug; }; -struct r300_transfer { - /* Parent class */ - struct pipe_transfer transfer; - - /* Offset from start of buffer. */ - unsigned offset; -}; /* Convenience cast wrapper. */ static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) { return (struct r300_screen*)screen; } -/* Convenience cast wrapper. */ -static INLINE struct r300_transfer* -r300_transfer(struct pipe_transfer* transfer) -{ - return (struct r300_transfer*)transfer; -} +/* Creates a new r300 screen. */ +struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys); /* Debug functionality. */ diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index e5ec2701c8..09bbf6c60e 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -43,6 +43,12 @@ /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ +#define UPDATE_STATE(cso, atom) \ + if (cso != atom.state) { \ + atom.state = cso; \ + atom.dirty = TRUE; \ + } + static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, unsigned dstRGB, unsigned dstA) { @@ -328,8 +334,7 @@ static void r300_bind_blend_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300->blend_state.state = state; - r300->blend_state.dirty = TRUE; + UPDATE_STATE(state, r300->blend_state); } /* Free blend state. */ @@ -356,7 +361,7 @@ static void r300_set_blend_color(struct pipe_context* pipe, (struct r300_blend_color_state*)r300->blend_color_state.state; union util_color uc; - util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); state->blend_color = uc.ui; /* XXX if FP16 blending is enabled, we should use the FP16 format */ @@ -376,6 +381,8 @@ static void r300_set_clip_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); + r300->clip = *state; + if (r300_screen(pipe->screen)->caps->has_tcl) { memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); r300->clip_state.size = 29; @@ -476,11 +483,8 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); - struct r300_screen* r300screen = r300_screen(pipe->screen); - r300->dsa_state.state = state; - r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; - r300->dsa_state.dirty = TRUE; + UPDATE_STATE(state, r300->dsa_state); } /* Free DSA state. */ @@ -567,6 +571,7 @@ static void { struct r300_context* r300 = r300_context(pipe); struct r300_screen* r300screen = r300_screen(pipe->screen); + struct pipe_framebuffer_state *old_state = r300->fb_state.state; unsigned max_width, max_height; uint32_t zbuffer_bpp = 0; @@ -591,22 +596,30 @@ static void return; } - if (r300->draw) { draw_flush(r300->draw); } - memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); + r300->fb_state.dirty = TRUE; - r300->fb_state.size = (10 * state->nr_cbufs) + (state->zsbuf ? 10 : 0) + 6; + /* If nr_cbufs is changed from zero to non-zero or vice versa... */ + if (!!old_state->nr_cbufs != !!state->nr_cbufs) { + r300->blend_state.dirty = TRUE; + } + /* If zsbuf is set from NULL to non-NULL or vice versa.. */ + if (!!old_state->zsbuf != !!state->zsbuf) { + r300->dsa_state.dirty = TRUE; + } + if (!r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; + } r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); - /* XXX wait what */ - r300->blend_state.dirty = TRUE; - r300->dsa_state.dirty = TRUE; - r300->fb_state.dirty = TRUE; - r300->scissor_state.dirty = TRUE; + memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); + + r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + + (state->zsbuf ? 10 : 0) + 8; /* Polygon offset depends on the zbuffer bit depth. */ if (state->zsbuf && r300->polygon_offset_enabled) { @@ -658,8 +671,10 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) r300->fs = fs; r300_pick_fragment_shader(r300); - if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { - r300->vertex_format_state.dirty = TRUE; + r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + + if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { + r300->vap_output_state.dirty = TRUE; } r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -709,32 +724,14 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->vap_control_status = R300_VC_32BIT_SWAP; #endif - /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. - * Else, enable HW TCL and force Draw's TCL off. */ - if (state->bypass_vs_clip_and_viewport || - !r300screen->caps->has_tcl) { + /* If no TCL engine is present, turn off the HW TCL. */ + if (!r300screen->caps->has_tcl) { rs->vap_control_status |= R300_VAP_TCL_BYPASS; } rs->point_size = pack_float_16_6x(state->point_size) | (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); - /* Point minimum and maximum sizes. This register has to be emitted, - * and it'd be a step backwards to put it in invariant state. */ - if (r300screen->caps->is_r500) { - rs->point_minmax = - ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | - ((int)(4096.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); - } else if (r300screen->caps->is_r400) { - rs->point_minmax = - ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | - ((int)(4021.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); - } else { - rs->point_minmax = - ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | - ((int)(2560.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); - } - rs->line_control = pack_float_16_6x(state->line_width) | R300_GA_LINE_CNTL_END_TYPE_COMP; @@ -817,6 +814,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); struct r300_rs_state* rs = (struct r300_rs_state*)state; + boolean scissor_was_enabled = r300->scissor_enabled; if (r300->draw) { draw_flush(r300->draw); @@ -824,22 +822,18 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) } if (rs) { - r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; + r300->scissor_enabled = rs->rs.scissor; } else { - r300->tcl_bypass = FALSE; r300->polygon_offset_enabled = FALSE; + r300->scissor_enabled = FALSE; } - r300->rs_state.state = rs; - r300->rs_state.dirty = TRUE; - /* XXX Why is this still needed, dammit!? */ - r300->scissor_state.dirty = TRUE; - r300->viewport_state.dirty = TRUE; + UPDATE_STATE(state, r300->rs_state); + r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0); - /* XXX Clean these up when we move to atom emits */ - if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { - r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + if (scissor_was_enabled != r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; } } @@ -871,7 +865,7 @@ static void* state->max_anisotropy > 0); /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ - /* We must pass these to the emit function to clamp them properly. */ + /* We must pass these to the merge function to clamp them properly. */ sampler->min_lod = MAX2((unsigned)state->min_lod, 0); sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); @@ -881,7 +875,7 @@ static void* sampler->filter1 |= r300_anisotropy(state->max_anisotropy); - util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); sampler->border_color = uc.ui; /* R500-specific fixups and optimizations */ @@ -897,23 +891,20 @@ static void r300_bind_sampler_states(struct pipe_context* pipe, void** states) { struct r300_context* r300 = r300_context(pipe); - int i; + struct r300_textures_state* state = + (struct r300_textures_state*)r300->textures_state.state; if (count > 8) { return; } - for (i = 0; i < count; i++) { - if (r300->sampler_states[i] != states[i]) { - r300->sampler_states[i] = (struct r300_sampler_state*)states[i]; - r300->dirty_state |= (R300_NEW_SAMPLER << i); - } - } + memcpy(state->sampler_states, states, sizeof(void*) * count); + state->sampler_count = count; - r300->sampler_count = count; + r300->textures_state.dirty = TRUE; /* Pick a fragment shader based on the texture compare state. */ - if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { + if (r300->fs && count) { if (r300_pick_fragment_shader(r300)) { r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -937,21 +928,22 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, struct pipe_sampler_view** views) { struct r300_context* r300 = r300_context(pipe); + unsigned i; boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; - int i; + boolean dirty_tex = FALSE; /* XXX magic num */ if (count > 8) { return; } - + for (i = 0; i < count; i++) { if (r300->fragment_sampler_views[i] != views[i]) { struct r300_texture *texture; pipe_sampler_view_reference(&r300->fragment_sampler_views[i], views[i]); - r300->dirty_state |= (R300_NEW_TEXTURE << i); + dirty_tex = TRUE; texture = (struct r300_texture *)views[i]->texture; @@ -968,11 +960,16 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, if (r300->fragment_sampler_views[i]) { pipe_sampler_view_reference(&r300->fragment_sampler_views[i], NULL); - r300->dirty_state |= (R300_NEW_TEXTURE << i); } } r300->fragment_sampler_view_count = count; + + r300->textures_state.dirty = TRUE; + + if (dirty_tex) { + r300->texture_cache_inval.dirty = TRUE; + } } static struct pipe_sampler_view * @@ -1009,7 +1006,9 @@ static void r300_set_scissor_state(struct pipe_context* pipe, memcpy(r300->scissor_state.state, state, sizeof(struct pipe_scissor_state)); - r300->scissor_state.dirty = TRUE; + if (r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; + } } static void r300_set_viewport_state(struct pipe_context* pipe, @@ -1019,6 +1018,8 @@ static void r300_set_viewport_state(struct pipe_context* pipe, struct r300_viewport_state* viewport = (struct r300_viewport_state*)r300->viewport_state.state; + r300->viewport = *state; + /* Do the transform in HW. */ viewport->vte_control = R300_VTX_W0_FMT; @@ -1058,27 +1059,32 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, const struct pipe_vertex_buffer* buffers) { struct r300_context* r300 = r300_context(pipe); + unsigned i, max_index = ~0; memcpy(r300->vertex_buffer, buffers, sizeof(struct pipe_vertex_buffer) * count); + + for (i = 0; i < count; i++) { + max_index = MIN2(buffers[i].max_index, max_index); + } + r300->vertex_buffer_count = count; + r300->vertex_buffer_max_index = max_index; if (r300->draw) { draw_flush(r300->draw); draw_set_vertex_buffers(r300->draw, count, buffers); } - - r300->vertex_format_state.dirty = TRUE; } static boolean r300_validate_aos(struct r300_context *r300) { struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; - struct pipe_vertex_element *velem = r300->vertex_element; + struct pipe_vertex_element *velem = r300->velems->velem; int i; /* Check if formats and strides are aligned to the size of DWORD. */ - for (i = 0; i < r300->vertex_element_count; i++) { + for (i = 0; i < r300->velems->count; i++) { if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || util_format_get_blocksize(velem[i].src_format) % 4 != 0) { return FALSE; @@ -1087,20 +1093,209 @@ static boolean r300_validate_aos(struct r300_context *r300) return TRUE; } -static void r300_set_vertex_elements(struct pipe_context* pipe, - unsigned count, - const struct pipe_vertex_element* elements) +static void r300_draw_emit_attrib(struct r300_context* r300, + enum attrib_emit emit, + enum interp_mode interp, + int index) { - struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = r300->vs_state.state; + struct tgsi_shader_info* info = &vs->info; + int output; + + output = draw_find_shader_output(r300->draw, + info->output_semantic_name[index], + info->output_semantic_index[index]); + draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output); +} + +static void r300_draw_emit_all_attribs(struct r300_context* r300) +{ + struct r300_vertex_shader* vs = r300->vs_state.state; + struct r300_shader_semantics* vs_outputs = &vs->outputs; + int i, gen_count; + + /* Position. */ + if (vs_outputs->pos != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->pos); + } else { + assert(0); + } - memcpy(r300->vertex_element, - elements, - sizeof(struct pipe_vertex_element) * count); - r300->vertex_element_count = count; + /* Point size. */ + if (vs_outputs->psize != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS, + vs_outputs->psize); + } + + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, + vs_outputs->color[i]); + } + } + + /* XXX Back-face colors. */ + + /* Texture coordinates. */ + gen_count = 0; + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->generic[i]); + gen_count++; + } + } + + /* Fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->fog); + gen_count++; + } + + /* XXX magic */ + assert(gen_count <= 8); +} + +/* Update the PSC tables. */ +static void r300_vertex_psc(struct r300_vertex_element_state *velems) +{ + struct r300_vertex_stream_state *vstream = &velems->vertex_stream; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i; + + assert(velems->count <= 16); + + /* Vertex shaders have no semantics on their inputs, + * so PSC should just route stuff based on the vertex elements, + * and not on attrib information. */ + for (i = 0; i < velems->count; i++) { + format = velems->velem[i].src_format; + + type = r300_translate_vertex_data_type(format) | + (i << R300_DST_VEC_LOC_SHIFT); + swizzle = r300_translate_vertex_data_swizzle(format); + + if (i & 1) { + vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vstream->vap_prog_stream_cntl[i >> 1] |= type; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; + } + } + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vstream->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); + + vstream->count = (i >> 1) + 1; +} + +/* Update the PSC tables for SW TCL, using Draw. */ +static void r300_swtcl_vertex_psc(struct r300_context *r300, + struct r300_vertex_element_state *velems) +{ + struct r300_vertex_stream_state *vstream = &velems->vertex_stream; + struct r300_vertex_shader* vs = r300->vs_state.state; + struct vertex_info* vinfo = &r300->vertex_info; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i, attrib_count; + int* vs_output_tab = vs->stream_loc_notcl; + + /* For each Draw attribute, route it to the fragment shader according + * to the vs_output_tab. */ + attrib_count = vinfo->num_attribs; + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," + " vs_output_tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + vs_output_tab[i]); + } + + for (i = 0; i < attrib_count; i++) { + /* Make sure we have a proper destination for our attribute. */ + assert(vs_output_tab[i] != -1); + + format = draw_translate_vinfo_format(vinfo->attrib[i].emit); + + /* Obtain the type of data in this attribute. */ + type = r300_translate_vertex_data_type(format) | + vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; + + /* Obtain the swizzle for this attribute. Note that the default + * swizzle in the hardware is not XYZW! */ + swizzle = r300_translate_vertex_data_swizzle(format); + + /* Add the attribute to the PSC table. */ + if (i & 1) { + vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vstream->vap_prog_stream_cntl[i >> 1] |= type; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; + } + } + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vstream->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); + + vstream->count = (i >> 1) + 1; +} + +static void* r300_create_vertex_elements_state(struct pipe_context* pipe, + unsigned count, + const struct pipe_vertex_element* attribs) +{ + struct r300_context *r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); + struct r300_vertex_element_state *velems; + + assert(count <= PIPE_MAX_ATTRIBS); + velems = CALLOC_STRUCT(r300_vertex_element_state); + if (velems != NULL) { + velems->count = count; + memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); + + if (r300screen->caps->has_tcl) { + r300_vertex_psc(velems); + } else { + memset(&r300->vertex_info, 0, sizeof(struct vertex_info)); + r300_draw_emit_all_attribs(r300); + draw_compute_vertex_size(&r300->vertex_info); + r300_swtcl_vertex_psc(r300, velems); + } + } + return velems; +} + +static void r300_bind_vertex_elements_state(struct pipe_context *pipe, + void *state) +{ + struct r300_context *r300 = r300_context(pipe); + struct r300_vertex_element_state *velems = state; + + if (velems == NULL) { + return; + } + + r300->velems = velems; if (r300->draw) { draw_flush(r300->draw); - draw_set_vertex_elements(r300->draw, count, elements); + draw_set_vertex_elements(r300->draw, velems->count, velems->velem); } if (!r300_validate_aos(r300)) { @@ -1108,6 +1303,14 @@ static void r300_set_vertex_elements(struct pipe_context* pipe, assert(0); abort(); } + + UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); + r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; +} + +static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) +{ + FREE(state); } static void* r300_create_vs_state(struct pipe_context* pipe, @@ -1115,64 +1318,71 @@ static void* r300_create_vs_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); - /* Copy state directly into shader. */ - vs->state = *shader; - vs->state.tokens = tgsi_dup_tokens(shader->tokens); - - tgsi_scan_shader(shader->tokens, &vs->info); + struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); + r300_vertex_shader_common_init(vs, shader); - return (void*)vs; + if (r300_screen(pipe->screen)->caps->has_tcl) { + r300_translate_vertex_shader(r300, vs); } else { - return draw_create_vertex_shader(r300->draw, shader); + vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); } + + return vs; } static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; - if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; + if (vs == NULL) { + r300->vs_state.state = NULL; + return; + } + if (vs == r300->vs_state.state) { + return; + } + r300->vs_state.state = vs; - if (vs == NULL) { - r300->vs = NULL; - return; - } else if (!vs->translated) { - r300_translate_vertex_shader(r300, vs); - } + // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block + if (r300->fs) { + r300_vertex_shader_setup_wpos(r300); + } + memcpy(r300->vap_output_state.state, &vs->vap_out, + sizeof(struct r300_vap_output_state)); + r300->vap_output_state.dirty = TRUE; - r300->vs = vs; - if (r300->fs) { - r300_vertex_shader_setup_wpos(r300); - } + /* The majority of the RS block bits is dependent on the vertex shader. */ + r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + + if (r300_screen(pipe->screen)->caps->has_tcl) { + r300->vs_state.dirty = TRUE; + r300->vs_state.size = vs->code.length + 9; - r300->vertex_format_state.dirty = TRUE; + r300->pvs_flush.dirty = TRUE; - r300->dirty_state |= - R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; + r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; } else { draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, - (struct draw_vertex_shader*)shader); + (struct draw_vertex_shader*)vs->draw_vs); } } static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; - rc_constants_destroy(&vs->code.constants); - FREE((void*)vs->state.tokens); - FREE(shader); } else { draw_delete_vertex_shader(r300->draw, - (struct draw_vertex_shader*)shader); + (struct draw_vertex_shader*)vs->draw_vs); } + + FREE((void*)vs->state.tokens); + FREE(shader); } static void r300_set_constant_buffer(struct pipe_context *pipe, @@ -1223,8 +1433,12 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); pipe_buffer_unmap(pipe->screen, buf); - if (shader == PIPE_SHADER_VERTEX) - r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; + if (shader == PIPE_SHADER_VERTEX) { + if (r300screen->caps->has_tcl) { + r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; + r300->pvs_flush.dirty = TRUE; + } + } else if (shader == PIPE_SHADER_FRAGMENT) r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -1273,7 +1487,10 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.set_viewport_state = r300_set_viewport_state; r300->context.set_vertex_buffers = r300_set_vertex_buffers; - r300->context.set_vertex_elements = r300_set_vertex_elements; + + r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; + r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; + r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; r300->context.create_vs_state = r300_create_vs_state; r300->context.bind_vs_state = r300_bind_vs_state; diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 2cbce9210a..6b9f61acd7 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -37,179 +37,6 @@ /* r300_state_derived: Various bits of state which are dependent upon * currently bound CSO data. */ -static void r300_draw_emit_attrib(struct r300_context* r300, - enum attrib_emit emit, - enum interp_mode interp, - int index) -{ - struct tgsi_shader_info* info = &r300->vs->info; - int output; - - output = draw_find_shader_output(r300->draw, - info->output_semantic_name[index], - info->output_semantic_index[index]); - draw_emit_vertex_attr( - (struct vertex_info*)r300->vertex_format_state.state, - emit, interp, output); -} - -static void r300_draw_emit_all_attribs(struct r300_context* r300) -{ - struct r300_shader_semantics* vs_outputs = &r300->vs->outputs; - int i, gen_count; - - /* Position. */ - if (vs_outputs->pos != ATTR_UNUSED) { - r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, - vs_outputs->pos); - } else { - assert(0); - } - - /* Point size. */ - if (vs_outputs->psize != ATTR_UNUSED) { - r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS, - vs_outputs->psize); - } - - /* Colors. */ - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - if (vs_outputs->color[i] != ATTR_UNUSED) { - r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, - vs_outputs->color[i]); - } - } - - /* XXX Back-face colors. */ - - /* Texture coordinates. */ - gen_count = 0; - for (i = 0; i < ATTR_GENERIC_COUNT; i++) { - if (vs_outputs->generic[i] != ATTR_UNUSED) { - r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, - vs_outputs->generic[i]); - gen_count++; - } - } - - /* Fog coordinates. */ - if (vs_outputs->fog != ATTR_UNUSED) { - r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, - vs_outputs->fog); - gen_count++; - } - - /* XXX magic */ - assert(gen_count <= 8); -} - -/* Update the PSC tables. */ -static void r300_vertex_psc(struct r300_context* r300) -{ - struct r300_vertex_info *vformat = - (struct r300_vertex_info*)r300->vertex_format_state.state; - uint16_t type, swizzle; - enum pipe_format format; - unsigned i; - int identity[16] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}; - int* stream_tab; - - /* If TCL is bypassed, map vertex streams to equivalent VS output - * locations. */ - if (r300->tcl_bypass) { - stream_tab = r300->vs->stream_loc_notcl; - } else { - stream_tab = identity; - } - - /* Vertex shaders have no semantics on their inputs, - * so PSC should just route stuff based on the vertex elements, - * and not on attrib information. */ - DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements" - " in psc\n", - r300->vs->info.num_inputs, - r300->vertex_element_count); - - for (i = 0; i < r300->vertex_element_count; i++) { - format = r300->vertex_element[i].src_format; - - type = r300_translate_vertex_data_type(format) | - (stream_tab[i] << R300_DST_VEC_LOC_SHIFT); - swizzle = r300_translate_vertex_data_swizzle(format); - - if (i & 1) { - vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; - } else { - vformat->vap_prog_stream_cntl[i >> 1] |= type; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; - } - } - - assert(i <= 15); - - /* Set the last vector in the PSC. */ - if (i) { - i -= 1; - } - vformat->vap_prog_stream_cntl[i >> 1] |= - (R300_LAST_VEC << (i & 1 ? 16 : 0)); -} - -/* Update the PSC tables for SW TCL, using Draw. */ -static void r300_swtcl_vertex_psc(struct r300_context* r300) -{ - struct r300_vertex_info *vformat = - (struct r300_vertex_info*)r300->vertex_format_state.state; - struct vertex_info* vinfo = &vformat->vinfo; - uint16_t type, swizzle; - enum pipe_format format; - unsigned i, attrib_count; - int* vs_output_tab = r300->vs->stream_loc_notcl; - - /* For each Draw attribute, route it to the fragment shader according - * to the vs_output_tab. */ - attrib_count = vinfo->num_attribs; - DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); - for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," - " vs_output_tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - vs_output_tab[i]); - } - - for (i = 0; i < attrib_count; i++) { - /* Make sure we have a proper destination for our attribute. */ - assert(vs_output_tab[i] != -1); - - format = draw_translate_vinfo_format(vinfo->attrib[i].emit); - - /* Obtain the type of data in this attribute. */ - type = r300_translate_vertex_data_type(format) | - vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; - - /* Obtain the swizzle for this attribute. Note that the default - * swizzle in the hardware is not XYZW! */ - swizzle = r300_translate_vertex_data_swizzle(format); - - /* Add the attribute to the PSC table. */ - if (i & 1) { - vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; - } else { - vformat->vap_prog_stream_cntl[i >> 1] |= type; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; - } - } - - /* Set the last vector in the PSC. */ - if (i) { - i -= 1; - } - vformat->vap_prog_stream_cntl[i >> 1] |= - (R300_LAST_VEC << (i & 1 ? 16 : 0)); -} - static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, boolean swizzle_0001) { @@ -416,33 +243,16 @@ static void r300_update_rs_block(struct r300_context* r300, /* Now, after all that, see if we actually need to update the state. */ if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) { memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block)); - r300->rs_block_state.size = 5 + count; - r300->rs_block_state.dirty = TRUE; + r300->rs_block_state.size = 5 + count*2; } } /* Update the shader-dependant states. */ static void r300_update_derived_shader_state(struct r300_context* r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct r300_vertex_info *vformat = - (struct r300_vertex_info*)r300->vertex_format_state.state; - struct vertex_info* vinfo = &vformat->vinfo; + struct r300_vertex_shader* vs = r300->vs_state.state; - /* Mmm, delicious hax */ - memset(r300->vertex_format_state.state, 0, sizeof(struct r300_vertex_info)); - memcpy(vinfo->hwfmt, r300->vs->hwfmt, sizeof(uint)*4); - - r300_update_rs_block(r300, &r300->vs->outputs, &r300->fs->inputs); - - if (r300screen->caps->has_tcl) { - r300_vertex_psc(r300); - } else { - r300_draw_emit_all_attribs(r300); - draw_compute_vertex_size( - (struct vertex_info*)r300->vertex_format_state.state); - r300_swtcl_vertex_psc(r300); - } + r300_update_rs_block(r300, &vs->outputs, &r300->fs->inputs); } static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa) @@ -516,14 +326,72 @@ static void r300_update_ztop(struct r300_context* r300) r300->ztop_state.dirty = TRUE; } +static void r300_merge_textures_and_samplers(struct r300_context* r300) +{ + struct r300_textures_state *state = + (struct r300_textures_state*)r300->textures_state.state; + struct r300_texture_sampler_state *texstate; + struct r300_sampler_state *sampler; + struct r300_texture *tex; + unsigned min_level, max_level, i, size; + unsigned count = MIN2(state->texture_count, state->sampler_count); + + state->tx_enable = 0; + size = 2; + + for (i = 0; i < count; i++) { + if (state->textures[i] && state->sampler_states[i]) { + state->tx_enable |= 1 << i; + + tex = state->textures[i]; + sampler = state->sampler_states[i]; + + texstate = &state->regs[i]; + memcpy(texstate->format, &tex->state, sizeof(uint32_t)*3); + texstate->filter[0] = sampler->filter0; + texstate->filter[1] = sampler->filter1; + texstate->border_color = sampler->border_color; + texstate->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) | + R300_TXO_MICRO_TILE(tex->microtile); + + /* to emulate 1D textures through 2D ones correctly */ + if (tex->tex.target == PIPE_TEXTURE_1D) { + texstate->filter[0] &= ~R300_TX_WRAP_T_MASK; + texstate->filter[0] |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); + } + + if (tex->is_npot) { + /* NPOT textures don't support mip filter, unfortunately. + * This prevents incorrect rendering. */ + texstate->filter[0] &= ~R300_TX_MIN_FILTER_MIP_MASK; + } else { + /* determine min/max levels */ + /* the MAX_MIP level is the largest (finest) one */ + max_level = MIN2(sampler->max_lod, tex->tex.last_level); + min_level = MIN2(sampler->min_lod, max_level); + texstate->format[0] |= R300_TX_NUM_LEVELS(max_level); + texstate->filter[0] |= R300_TX_MAX_MIP_LEVEL(min_level); + } + + texstate->filter[0] |= i << 28; + + size += 16; + state->count = i+1; + } + } + + r300->textures_state.size = size; +} + void r300_update_derived_state(struct r300_context* r300) { - /* XXX */ - if (r300->dirty_state & - (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER) || - r300->vertex_format_state.dirty || r300->rs_state.dirty) { + if (r300->rs_block_state.dirty) { r300_update_derived_shader_state(r300); } + if (r300->textures_state.dirty) { + r300_merge_textures_and_samplers(r300); + } + r300_update_ztop(r300); } diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h index a608392532..a32924ed0a 100644 --- a/src/gallium/drivers/r300/r300_state_inlines.h +++ b/src/gallium/drivers/r300/r300_state_inlines.h @@ -327,117 +327,6 @@ static INLINE uint32_t r300_anisotropy(unsigned max_aniso) } } -/* Buffer formats. */ - -/* Colorbuffer formats. This is the unswizzled format of the RB3D block's - * output. For the swizzling of the targets, check the shader's format. */ -static INLINE uint32_t r300_translate_colorformat(enum pipe_format format) -{ - switch (format) { - /* 8-bit buffers */ - case PIPE_FORMAT_A8_UNORM: - case PIPE_FORMAT_I8_UNORM: - case PIPE_FORMAT_L8_UNORM: - return R300_COLOR_FORMAT_I8; - /* 16-bit buffers */ - case PIPE_FORMAT_R5G6B5_UNORM: - return R300_COLOR_FORMAT_RGB565; - case PIPE_FORMAT_A1R5G5B5_UNORM: - return R300_COLOR_FORMAT_ARGB1555; - case PIPE_FORMAT_A4R4G4B4_UNORM: - return R300_COLOR_FORMAT_ARGB4444; - /* 32-bit buffers */ - case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_X8R8G8B8_UNORM: - case PIPE_FORMAT_R8G8B8A8_UNORM: - case PIPE_FORMAT_R8G8B8X8_UNORM: - return R300_COLOR_FORMAT_ARGB8888; - /* XXX Not in pipe_format - case PIPE_FORMAT_A32R32G32B32: - return R300_COLOR_FORMAT_ARGB32323232; - case PIPE_FORMAT_A16R16G16B16: - return R300_COLOR_FORMAT_ARGB16161616; - case PIPE_FORMAT_A10R10G10B10_UNORM: - return R500_COLOR_FORMAT_ARGB10101010; - case PIPE_FORMAT_A2R10G10B10_UNORM: - return R500_COLOR_FORMAT_ARGB2101010; - case PIPE_FORMAT_I10_UNORM: - return R500_COLOR_FORMAT_I10; */ - default: - debug_printf("r300: Implementation error: " - "Got unsupported color format %s in %s\n", - util_format_name(format), __FUNCTION__); - assert(0); - break; - } - return 0; -} - -/* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */ -static INLINE uint32_t r300_translate_zsformat(enum pipe_format format) -{ - switch (format) { - /* 16-bit depth, no stencil */ - case PIPE_FORMAT_Z16_UNORM: - return R300_DEPTHFORMAT_16BIT_INT_Z; - /* 24-bit depth, ignored stencil */ - case PIPE_FORMAT_Z24X8_UNORM: - /* 24-bit depth, 8-bit stencil */ - case PIPE_FORMAT_Z24S8_UNORM: - return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; - default: - debug_printf("r300: Implementation error: " - "Got unsupported ZS format %s in %s\n", - util_format_name(format), __FUNCTION__); - assert(0); - break; - } - return 0; -} - -/* Shader output formats. This is essentially the swizzle from the shader - * to the RB3D block. - * - * Note that formats are stored from C3 to C0. */ -static INLINE uint32_t r300_translate_out_fmt(enum pipe_format format) -{ - switch (format) { - case PIPE_FORMAT_R5G6B5_UNORM: - /* C_5_6_5 is missing in US_OUT_FMT, but C4_8 works just fine. */ - case PIPE_FORMAT_A1R5G5B5_UNORM: - /* C_1_5_5_5 is missing in US_OUT_FMT, but C4_8 works just fine. */ - case PIPE_FORMAT_A4R4G4B4_UNORM: - /* C4_4 is missing in US_OUT_FMT, but C4_8 works just fine. */ - case PIPE_FORMAT_A8R8G8B8_UNORM: - case PIPE_FORMAT_X8R8G8B8_UNORM: - return R300_US_OUT_FMT_C4_8 | - R300_C0_SEL_B | R300_C1_SEL_G | - R300_C2_SEL_R | R300_C3_SEL_A; - case PIPE_FORMAT_R8G8B8A8_UNORM: - case PIPE_FORMAT_R8G8B8X8_UNORM: - return R300_US_OUT_FMT_C4_8 | - R300_C0_SEL_A | R300_C1_SEL_B | - R300_C2_SEL_G | R300_C3_SEL_R; - - /* 8-bit outputs */ - case PIPE_FORMAT_A8_UNORM: - return R300_US_OUT_FMT_C4_8 | - R300_C0_SEL_A; - case PIPE_FORMAT_I8_UNORM: - case PIPE_FORMAT_L8_UNORM: - return R300_US_OUT_FMT_C4_8 | - R300_C0_SEL_R; - /* R300_OUT_SIGN(x) */ - default: - debug_printf("r300: Implementation error: " - "Got unsupported output format %s in %s\n", - util_format_name(format), __FUNCTION__); - assert(0); - return R300_US_OUT_FMT_UNUSED; - } - return 0; -} - /* Non-CSO state. (For now.) */ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count) @@ -459,44 +348,16 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count) return 0; } -/* Utility function to count the number of components in RGBAZS formats. - * XXX should go to util or p_format.h */ -static INLINE unsigned pf_component_count(enum pipe_format format) { - unsigned count = 0; - - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) { - count++; - } - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 1)) { - count++; - } - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 2)) { - count++; - } - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 3)) { - count++; - } - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 0)) { - count++; - } - if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) { - count++; - } - - return count; -} - /* Translate pipe_formats into PSC vertex types. */ static INLINE uint16_t r300_translate_vertex_data_type(enum pipe_format format) { uint32_t result = 0; const struct util_format_description *desc; - unsigned components = pf_component_count(format); + unsigned components = util_format_get_nr_components(format); desc = util_format_description(format); - if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH && - desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) { + if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) { debug_printf("r300: Bad format %s in %s:%d\n", util_format_name(format), __FUNCTION__, __LINE__); assert(0); @@ -565,36 +426,19 @@ r300_translate_vertex_data_type(enum pipe_format format) { static INLINE uint16_t r300_translate_vertex_data_swizzle(enum pipe_format format) { const struct util_format_description *desc = util_format_description(format); - unsigned swizzle[4], i; assert(format); - if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH && - desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) { + if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) { debug_printf("r300: Bad format %s in %s:%d\n", util_format_name(format), __FUNCTION__, __LINE__); return 0; } - /* Swizzles for 8bits formats are in the reversed order, not sure why. */ - if (desc->channel[0].size == 8) { - for (i = 0; i < 4; i++) { - if (desc->swizzle[i] <= 3) { - swizzle[i] = 3 - desc->swizzle[i]; - } else { - swizzle[i] = desc->swizzle[i]; - } - } - } else { - for (i = 0; i < 4; i++) { - swizzle[i] = desc->swizzle[i]; - } - } - - return ((swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) | - (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) | - (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) | - (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) | + return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) | + (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) | + (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) | + (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) | (0xf << R300_WRITE_ENA_SHIFT)); } diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c index 97927acf1b..4a2c68269b 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ b/src/gallium/drivers/r300/r300_state_invariant.c @@ -38,7 +38,8 @@ struct pipe_viewport_state r300_viewport_identity = { * * Note that eventually this should be empty, but it's useful for development * and general unduplication of code. */ -void r300_emit_invariant_state(struct r300_context* r300, void* state) +void r300_emit_invariant_state(struct r300_context* r300, + unsigned size, void* state) { struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps; CS_LOCALS(r300); diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h index 5d1a963654..83d031c7fe 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.h +++ b/src/gallium/drivers/r300/r300_state_invariant.h @@ -25,6 +25,7 @@ struct r300_context; -void r300_emit_invariant_state(struct r300_context* r300, void* state); +void r300_emit_invariant_state(struct r300_context* r300, + unsigned size, void* state); #endif /* R300_STATE_INVARIANT_H */ diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index 34a49bec34..04124afd4d 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2010 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -29,6 +30,7 @@ #include "r300_context.h" #include "r300_texture.h" #include "r300_screen.h" +#include "r300_state_inlines.h" #include "radeon_winsys.h" @@ -44,12 +46,465 @@ static const unsigned microblock_table[5][3][2] = { {{ 2, 1}, {0, 0}, {0, 0}} /* 128 bits per pixel */ }; +/* Translate a pipe_format into a useful texture format for sampling. + * + * Some special formats are translated directly using R300_EASY_TX_FORMAT, + * but the majority of them is translated in a generic way, automatically + * supporting all the formats hw can support. + * + * R300_EASY_TX_FORMAT swizzles the texture. + * Note the signature of R300_EASY_TX_FORMAT: + * R300_EASY_TX_FORMAT(B, G, R, A, FORMAT); + * + * The FORMAT specifies how the texture sampler will treat the texture, and + * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */ +static uint32_t r300_translate_texformat(enum pipe_format format) +{ + uint32_t result = 0; + const struct util_format_description *desc; + unsigned components = 0, i; + boolean uniform = TRUE; + const uint32_t swizzle_shift[4] = { + R300_TX_FORMAT_R_SHIFT, + R300_TX_FORMAT_G_SHIFT, + R300_TX_FORMAT_B_SHIFT, + R300_TX_FORMAT_A_SHIFT + }; + const uint32_t swizzle[4] = { + R300_TX_FORMAT_X, + R300_TX_FORMAT_Y, + R300_TX_FORMAT_Z, + R300_TX_FORMAT_W + }; + const uint32_t sign_bit[4] = { + R300_TX_FORMAT_SIGNED_X, + R300_TX_FORMAT_SIGNED_Y, + R300_TX_FORMAT_SIGNED_Z, + R300_TX_FORMAT_SIGNED_W, + }; + + desc = util_format_description(format); + + /* Colorspace (return non-RGB formats directly). */ + switch (desc->colorspace) { + /* Depth stencil formats. */ + case UTIL_FORMAT_COLORSPACE_ZS: + switch (format) { + case PIPE_FORMAT_Z16_UNORM: + return R300_EASY_TX_FORMAT(X, X, X, X, X16); + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP); + default: + return ~0; /* Unsupported. */ + } + + /* YUV formats. */ + case UTIL_FORMAT_COLORSPACE_YUV: + result |= R300_TX_FORMAT_YUV_TO_RGB; + + switch (format) { + case PIPE_FORMAT_UYVY: + return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result; + case PIPE_FORMAT_YUYV: + return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result; + default: + return ~0; /* Unsupported/unknown. */ + } + + /* Add gamma correction. */ + case UTIL_FORMAT_COLORSPACE_SRGB: + result |= R300_TX_FORMAT_GAMMA; + break; + + default:; + } + + /* Add swizzle. */ + for (i = 0; i < 4; i++) { + switch (desc->swizzle[i]) { + case UTIL_FORMAT_SWIZZLE_X: + case UTIL_FORMAT_SWIZZLE_NONE: + result |= swizzle[0] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_Y: + result |= swizzle[1] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_Z: + result |= swizzle[2] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_W: + result |= swizzle[3] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_0: + result |= R300_TX_FORMAT_ZERO << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_1: + result |= R300_TX_FORMAT_ONE << swizzle_shift[i]; + break; + default: + return ~0; /* Unsupported. */ + } + } + + /* Compressed formats. */ + if (desc->layout == UTIL_FORMAT_LAYOUT_COMPRESSED) { + switch (format) { + case PIPE_FORMAT_DXT1_RGB: + case PIPE_FORMAT_DXT1_RGBA: + case PIPE_FORMAT_DXT1_SRGB: + case PIPE_FORMAT_DXT1_SRGBA: + return R300_TX_FORMAT_DXT1 | result; + case PIPE_FORMAT_DXT3_RGBA: + case PIPE_FORMAT_DXT3_SRGBA: + return R300_TX_FORMAT_DXT3 | result; + case PIPE_FORMAT_DXT5_RGBA: + case PIPE_FORMAT_DXT5_SRGBA: + return R300_TX_FORMAT_DXT5 | result; + default: + return ~0; /* Unsupported/unknown. */ + } + } + + /* Get the number of components. */ + for (i = 0; i < 4; i++) { + if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { + ++components; + } + } + + /* Add sign. */ + for (i = 0; i < components; i++) { + if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { + result |= sign_bit[i]; + } + } + + /* See whether the components are of the same size. */ + for (i = 1; i < components; i++) { + uniform = uniform && desc->channel[0].size == desc->channel[i].size; + } + + /* Non-uniform formats. */ + if (!uniform) { + switch (components) { + case 3: + if (desc->channel[0].size == 5 && + desc->channel[1].size == 6 && + desc->channel[2].size == 5) { + return R300_TX_FORMAT_Z5Y6X5 | result; + } + if (desc->channel[0].size == 5 && + desc->channel[1].size == 5 && + desc->channel[2].size == 6) { + return R300_TX_FORMAT_Z6Y5X5 | result; + } + return ~0; /* Unsupported/unknown. */ + + case 4: + if (desc->channel[0].size == 5 && + desc->channel[1].size == 5 && + desc->channel[2].size == 5 && + desc->channel[3].size == 1) { + return R300_TX_FORMAT_W1Z5Y5X5 | result; + } + if (desc->channel[0].size == 10 && + desc->channel[1].size == 10 && + desc->channel[2].size == 10 && + desc->channel[3].size == 2) { + return R300_TX_FORMAT_W2Z10Y10X10 | result; + } + } + return ~0; /* Unsupported/unknown. */ + } + + /* And finally, uniform formats. */ + switch (desc->channel[0].type) { + case UTIL_FORMAT_TYPE_UNSIGNED: + case UTIL_FORMAT_TYPE_SIGNED: + if (!desc->channel[0].normalized && + desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) { + return ~0; + } + + switch (desc->channel[0].size) { + case 4: + switch (components) { + case 2: + return R300_TX_FORMAT_Y4X4 | result; + case 4: + return R300_TX_FORMAT_W4Z4Y4X4 | result; + } + return ~0; + + case 8: + switch (components) { + case 1: + return R300_TX_FORMAT_X8 | result; + case 2: + return R300_TX_FORMAT_Y8X8 | result; + case 4: + return R300_TX_FORMAT_W8Z8Y8X8 | result; + } + return ~0; + + case 16: + switch (components) { + case 1: + return R300_TX_FORMAT_X16 | result; + case 2: + return R300_TX_FORMAT_Y16X16 | result; + case 4: + return R300_TX_FORMAT_W16Z16Y16X16 | result; + } + } + return ~0; + +/* XXX Enable float textures here. */ +#if 0 + case UTIL_FORMAT_TYPE_FLOAT: + switch (desc->channel[0].size) { + case 16: + switch (components) { + case 1: + return R300_TX_FORMAT_16F | result; + case 2: + return R300_TX_FORMAT_16F_16F | result; + case 4: + return R300_TX_FORMAT_16F_16F_16F_16F | result; + } + return ~0; + + case 32: + switch (components) { + case 1: + return R300_TX_FORMAT_32F | result; + case 2: + return R300_TX_FORMAT_32F_32F | result; + case 4: + return R300_TX_FORMAT_32F_32F_32F_32F | result; + } + } +#endif + } + + return ~0; /* Unsupported/unknown. */ +} + +/* Buffer formats. */ + +/* Colorbuffer formats. This is the unswizzled format of the RB3D block's + * output. For the swizzling of the targets, check the shader's format. */ +static uint32_t r300_translate_colorformat(enum pipe_format format) +{ + switch (format) { + /* 8-bit buffers. */ + case PIPE_FORMAT_A8_UNORM: + case PIPE_FORMAT_I8_UNORM: + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_L8_SRGB: + case PIPE_FORMAT_R8_UNORM: + case PIPE_FORMAT_R8_SNORM: + return R300_COLOR_FORMAT_I8; + + /* 16-bit buffers. */ + case PIPE_FORMAT_B5G6R5_UNORM: + return R300_COLOR_FORMAT_RGB565; + case PIPE_FORMAT_B5G5R5A1_UNORM: + return R300_COLOR_FORMAT_ARGB1555; + case PIPE_FORMAT_B4G4R4A4_UNORM: + return R300_COLOR_FORMAT_ARGB4444; + + /* 32-bit buffers. */ + case PIPE_FORMAT_B8G8R8A8_UNORM: + case PIPE_FORMAT_B8G8R8A8_SRGB: + case PIPE_FORMAT_B8G8R8X8_UNORM: + case PIPE_FORMAT_B8G8R8X8_SRGB: + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_A8R8G8B8_SRGB: + case PIPE_FORMAT_X8R8G8B8_UNORM: + case PIPE_FORMAT_X8R8G8B8_SRGB: + case PIPE_FORMAT_A8B8G8R8_UNORM: + case PIPE_FORMAT_R8G8B8A8_SNORM: + case PIPE_FORMAT_A8B8G8R8_SRGB: + case PIPE_FORMAT_X8B8G8R8_UNORM: + case PIPE_FORMAT_X8B8G8R8_SRGB: + case PIPE_FORMAT_R8SG8SB8UX8U_NORM: + return R300_COLOR_FORMAT_ARGB8888; + case PIPE_FORMAT_R10G10B10A2_UNORM: + return R500_COLOR_FORMAT_ARGB2101010; /* R5xx-only? */ + + /* 64-bit buffers. */ + case PIPE_FORMAT_R16G16B16A16_UNORM: + case PIPE_FORMAT_R16G16B16A16_SNORM: + //case PIPE_FORMAT_R16G16B16A16_FLOAT: /* not in pipe_format */ + return R300_COLOR_FORMAT_ARGB16161616; + +/* XXX Enable float textures here. */ +#if 0 + /* 128-bit buffers. */ + case PIPE_FORMAT_R32G32B32A32_FLOAT: + return R300_COLOR_FORMAT_ARGB32323232; +#endif + + /* YUV buffers. */ + case PIPE_FORMAT_UYVY: + return R300_COLOR_FORMAT_YVYU; + case PIPE_FORMAT_YUYV: + return R300_COLOR_FORMAT_VYUY; + default: + return ~0; /* Unsupported. */ + } +} + +/* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */ +static uint32_t r300_translate_zsformat(enum pipe_format format) +{ + switch (format) { + /* 16-bit depth, no stencil */ + case PIPE_FORMAT_Z16_UNORM: + return R300_DEPTHFORMAT_16BIT_INT_Z; + /* 24-bit depth, ignored stencil */ + case PIPE_FORMAT_X8Z24_UNORM: + /* 24-bit depth, 8-bit stencil */ + case PIPE_FORMAT_S8Z24_UNORM: + return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; + default: + return ~0; /* Unsupported. */ + } +} + +/* Shader output formats. This is essentially the swizzle from the shader + * to the RB3D block. + * + * Note that formats are stored from C3 to C0. */ +static uint32_t r300_translate_out_fmt(enum pipe_format format) +{ + uint32_t modifier = 0; + unsigned i; + const struct util_format_description *desc; + static const uint32_t sign_bit[4] = { + R300_OUT_SIGN(0x1), + R300_OUT_SIGN(0x2), + R300_OUT_SIGN(0x4), + R300_OUT_SIGN(0x8), + }; + + desc = util_format_description(format); + + /* Specifies how the shader output is written to the fog unit. */ + if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) { + /* The gamma correction causes precision loss so we need + * higher precision to maintain reasonable quality. + * It has nothing to do with the colorbuffer format. */ + modifier |= R300_US_OUT_FMT_C4_10_GAMMA; + } else if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) { + if (desc->channel[0].size == 32) { + modifier |= R300_US_OUT_FMT_C4_32_FP; + } else { + modifier |= R300_US_OUT_FMT_C4_16_FP; + } + } else { + if (desc->channel[0].size == 16) { + modifier |= R300_US_OUT_FMT_C4_16; + } else { + /* C4_8 seems to be used for the formats whose pixel size + * is <= 32 bits. */ + modifier |= R300_US_OUT_FMT_C4_8; + } + } + + /* Add sign. */ + for (i = 0; i < 4; i++) + if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { + modifier |= sign_bit[i]; + } + + /* Add swizzles and return. */ + switch (format) { + /* 8-bit outputs. + * COLORFORMAT_I8 stores the C2 component. */ + case PIPE_FORMAT_A8_UNORM: + return modifier | R300_C2_SEL_A; + case PIPE_FORMAT_I8_UNORM: + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_L8_SRGB: + case PIPE_FORMAT_R8_UNORM: + case PIPE_FORMAT_R8_SNORM: + return modifier | R300_C2_SEL_R; + + /* ARGB 32-bit outputs. */ + case PIPE_FORMAT_B5G6R5_UNORM: + case PIPE_FORMAT_B5G5R5A1_UNORM: + case PIPE_FORMAT_B4G4R4A4_UNORM: + case PIPE_FORMAT_B8G8R8A8_UNORM: + case PIPE_FORMAT_B8G8R8A8_SRGB: + case PIPE_FORMAT_B8G8R8X8_UNORM: + case PIPE_FORMAT_B8G8R8X8_SRGB: + return modifier | + R300_C0_SEL_B | R300_C1_SEL_G | + R300_C2_SEL_R | R300_C3_SEL_A; + + /* BGRA 32-bit outputs. */ + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_A8R8G8B8_SRGB: + case PIPE_FORMAT_X8R8G8B8_UNORM: + case PIPE_FORMAT_X8R8G8B8_SRGB: + return modifier | + R300_C0_SEL_A | R300_C1_SEL_R | + R300_C2_SEL_G | R300_C3_SEL_B; + + /* RGBA 32-bit outputs. */ + case PIPE_FORMAT_A8B8G8R8_UNORM: + case PIPE_FORMAT_R8G8B8A8_SNORM: + case PIPE_FORMAT_A8B8G8R8_SRGB: + case PIPE_FORMAT_X8B8G8R8_UNORM: + case PIPE_FORMAT_X8B8G8R8_SRGB: + return modifier | + R300_C0_SEL_A | R300_C1_SEL_B | + R300_C2_SEL_G | R300_C3_SEL_R; + + /* ABGR 32-bit outputs. */ + case PIPE_FORMAT_R8SG8SB8UX8U_NORM: + case PIPE_FORMAT_R10G10B10A2_UNORM: + /* RGBA high precision outputs (same swizzles as ABGR low precision) */ + case PIPE_FORMAT_R16G16B16A16_UNORM: + case PIPE_FORMAT_R16G16B16A16_SNORM: + //case PIPE_FORMAT_R16G16B16A16_FLOAT: /* not in pipe_format */ + case PIPE_FORMAT_R32G32B32A32_FLOAT: + return modifier | + R300_C0_SEL_R | R300_C1_SEL_G | + R300_C2_SEL_B | R300_C3_SEL_A; + + default: + return ~0; /* Unsupported. */ + } +} + +boolean r300_is_colorbuffer_format_supported(enum pipe_format format) +{ + return r300_translate_colorformat(format) != ~0 && + r300_translate_out_fmt(format) != ~0; +} + +boolean r300_is_zs_format_supported(enum pipe_format format) +{ + return r300_translate_zsformat(format) != ~0; +} + +boolean r300_is_sampler_format_supported(enum pipe_format format) +{ + return r300_translate_texformat(format) != ~0; +} + static void r300_setup_texture_state(struct r300_screen* screen, struct r300_texture* tex) { - struct r300_texture_state* state = &tex->state; + struct r300_texture_format_state* state = &tex->state; struct pipe_texture *pt = &tex->tex; + unsigned i; boolean is_r500 = screen->caps->is_r500; + /* Set sampler state. */ state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) | R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff); @@ -83,6 +538,26 @@ static void r300_setup_texture_state(struct r300_screen* screen, struct r300_tex SCREEN_DBG(screen, DBG_TEX, "r300: Set texture state (%dx%d, %d levels)\n", pt->width0, pt->height0, pt->last_level); + + /* Set framebuffer state. */ + if (util_format_is_depth_or_stencil(tex->tex.format)) { + for (i = 0; i <= tex->tex.last_level; i++) { + tex->fb_state.depthpitch[i] = + tex->pitch[i] | + R300_DEPTHMACROTILE(tex->mip_macrotile[i]) | + R300_DEPTHMICROTILE(tex->microtile); + } + tex->fb_state.zb_format = r300_translate_zsformat(tex->tex.format); + } else { + for (i = 0; i <= tex->tex.last_level; i++) { + tex->fb_state.colorpitch[i] = + tex->pitch[i] | + r300_translate_colorformat(tex->tex.format) | + R300_COLOR_TILE(tex->mip_macrotile[i]) | + R300_COLOR_MICROTILE(tex->microtile); + } + tex->fb_state.us_out_fmt = r300_translate_out_fmt(tex->tex.format); + } } void r300_texture_reinterpret_format(struct pipe_screen *screen, @@ -142,18 +617,23 @@ static unsigned r300_texture_get_tile_size(struct r300_texture* tex, /* Return true if macrotiling should be enabled on the miplevel. */ static boolean r300_texture_macro_switch(struct r300_texture *tex, unsigned level, - boolean rv350_mode) + boolean rv350_mode, + int dim) { - unsigned tile_width, width; + unsigned tile, texdim; - tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH, TRUE); - width = u_minify(tex->tex.width0, level); + tile = r300_texture_get_tile_size(tex, dim, TRUE); + if (dim == TILE_WIDTH) { + texdim = u_minify(tex->tex.width0, level); + } else { + texdim = u_minify(tex->tex.height0, level); + } /* See TX_FILTER1_n.MACRO_SWITCH. */ if (rv350_mode) { - return width >= tile_width; + return texdim >= tile; } else { - return width > tile_width; + return texdim > tile; } } @@ -217,9 +697,10 @@ static void r300_setup_miptree(struct r300_screen* screen, for (i = 0; i <= base->last_level; i++) { /* Let's see if this miplevel can be macrotiled. */ - tex->mip_macrotile[i] = (tex->macrotile == R300_BUFFER_TILED && - r300_texture_macro_switch(tex, i, rv350_mode)) ? - R300_BUFFER_TILED : R300_BUFFER_LINEAR; + tex->mip_macrotile[i] = + (tex->macrotile == R300_BUFFER_TILED && + r300_texture_macro_switch(tex, i, rv350_mode, TILE_WIDTH)) ? + R300_BUFFER_TILED : R300_BUFFER_LINEAR; stride = r300_texture_get_stride(screen, tex, i); nblocksy = r300_texture_get_nblocksy(tex, i); @@ -249,10 +730,46 @@ static void r300_setup_flags(struct r300_texture* tex) !util_is_power_of_two(tex->tex.height0); } +static void r300_setup_tiling(struct pipe_screen *screen, + struct r300_texture *tex) +{ + enum pipe_format format = tex->tex.format; + boolean rv350_mode = r300_screen(screen)->caps->family >= CHIP_FAMILY_RV350; + + if (util_format_is_compressed(format)) { + return; + } + + if (tex->tex.width0 == 1 || + tex->tex.height0 == 1) { + return; + } + + /* Set microtiling. */ + switch (util_format_get_blocksize(format)) { + case 1: + case 4: + tex->microtile = R300_BUFFER_TILED; + break; + + /* XXX Square-tiling doesn't work with kernel older than 2.6.34, + * XXX need to check the DRM version */ + /*case 2: + case 8: + tex->microtile = R300_BUFFER_SQUARETILED; + break;*/ + } + + /* Set macrotiling. */ + if (r300_texture_macro_switch(tex, 0, rv350_mode, TILE_WIDTH) && + r300_texture_macro_switch(tex, 0, rv350_mode, TILE_HEIGHT)) { + tex->macrotile = R300_BUFFER_TILED; + } +} + /* Create a new texture. */ -static struct pipe_texture* - r300_texture_create(struct pipe_screen* screen, - const struct pipe_texture* template) +static struct pipe_texture* r300_texture_create(struct pipe_screen* screen, + const struct pipe_texture* template) { struct r300_texture* tex = CALLOC_STRUCT(r300_texture); struct r300_screen* rscreen = r300_screen(screen); @@ -267,6 +784,9 @@ static struct pipe_texture* tex->tex.screen = screen; r300_setup_flags(tex); + if (!(template->tex_usage & R300_TEXTURE_USAGE_TRANSFER)) { + r300_setup_tiling(screen, tex); + } r300_setup_miptree(rscreen, tex); r300_setup_texture_state(rscreen, tex); @@ -331,14 +851,17 @@ static void r300_tex_surface_destroy(struct pipe_surface* s) FREE(s); } + static struct pipe_texture* - r300_texture_blanket(struct pipe_screen* screen, - const struct pipe_texture* base, - const unsigned* stride, - struct pipe_buffer* buffer) + r300_texture_from_handle(struct pipe_screen* screen, + const struct pipe_texture* base, + struct winsys_handle *whandle) { - struct r300_texture* tex; + struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys; struct r300_screen* rscreen = r300_screen(screen); + struct pipe_buffer *buffer; + struct r300_texture* tex; + unsigned stride; /* Support only 2D textures without mipmaps */ if (base->target != PIPE_TEXTURE_2D || @@ -347,6 +870,11 @@ static struct pipe_texture* return NULL; } + buffer = winsys->buffer_from_handle(winsys, screen, whandle, &stride); + if (!buffer) { + return NULL; + } + tex = CALLOC_STRUCT(r300_texture); if (!tex) { return NULL; @@ -356,17 +884,38 @@ static struct pipe_texture* pipe_reference_init(&tex->tex.reference, 1); tex->tex.screen = screen; - tex->stride_override = *stride; - tex->pitch[0] = *stride / util_format_get_blocksize(base->format); + tex->stride_override = stride; + tex->pitch[0] = stride / util_format_get_blocksize(base->format); r300_setup_flags(tex); r300_setup_texture_state(rscreen, tex); - pipe_buffer_reference(&tex->buffer, buffer); + /* one ref already taken */ + tex->buffer = buffer; return (struct pipe_texture*)tex; } +static boolean + r300_texture_get_handle(struct pipe_screen* screen, + struct pipe_texture *texture, + struct winsys_handle *whandle) +{ + struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys; + struct r300_texture* tex = (struct r300_texture*)texture; + unsigned stride; + + if (!tex) { + return FALSE; + } + + stride = r300_texture_get_stride(r300_screen(screen), tex, 0); + + winsys->buffer_get_handle(winsys, tex->buffer, stride, whandle); + + return TRUE; +} + static struct pipe_video_surface * r300_video_surface_create(struct pipe_screen *screen, enum pipe_video_chroma_format chroma_format, @@ -390,7 +939,7 @@ r300_video_surface_create(struct pipe_screen *screen, memset(&template, 0, sizeof(struct pipe_texture)); template.target = PIPE_TEXTURE_2D; - template.format = PIPE_FORMAT_X8R8G8B8_UNORM; + template.format = PIPE_FORMAT_B8G8R8X8_UNORM; template.last_level = 0; template.width0 = util_next_power_of_two(width); template.height0 = util_next_power_of_two(height); @@ -418,30 +967,13 @@ static void r300_video_surface_destroy(struct pipe_video_surface *vsfc) void r300_init_screen_texture_functions(struct pipe_screen* screen) { screen->texture_create = r300_texture_create; + screen->texture_from_handle = r300_texture_from_handle; + screen->texture_get_handle = r300_texture_get_handle; screen->texture_destroy = r300_texture_destroy; screen->get_tex_surface = r300_get_tex_surface; screen->tex_surface_destroy = r300_tex_surface_destroy; - screen->texture_blanket = r300_texture_blanket; screen->video_surface_create = r300_video_surface_create; screen->video_surface_destroy= r300_video_surface_destroy; } -boolean r300_get_texture_buffer(struct pipe_screen* screen, - struct pipe_texture* texture, - struct pipe_buffer** buffer, - unsigned* stride) -{ - struct r300_texture* tex = (struct r300_texture*)texture; - if (!tex) { - return FALSE; - } - - pipe_buffer_reference(buffer, tex->buffer); - - if (stride) { - *stride = r300_texture_get_stride(r300_screen(screen), tex, 0); - } - - return TRUE; -} diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h index 153f4eebda..138b62784e 100644 --- a/src/gallium/drivers/r300/r300_texture.h +++ b/src/gallium/drivers/r300/r300_texture.h @@ -42,239 +42,11 @@ void r300_texture_reinterpret_format(struct pipe_screen *screen, struct pipe_texture *tex, enum pipe_format new_format); -/* Translate a pipe_format into a useful texture format for sampling. - * - * R300_EASY_TX_FORMAT swizzles the texture. - * Note the signature of R300_EASY_TX_FORMAT: - * R300_EASY_TX_FORMAT(B, G, R, A, FORMAT); - * - * The FORMAT specifies how the texture sampler will treat the texture, and - * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */ -static INLINE uint32_t r300_translate_texformat(enum pipe_format format) -{ - uint32_t result = 0; - const struct util_format_description *desc; - unsigned components = 0, i; - boolean uniform = TRUE; - const uint32_t swizzle_shift[4] = { - R300_TX_FORMAT_R_SHIFT, - R300_TX_FORMAT_G_SHIFT, - R300_TX_FORMAT_B_SHIFT, - R300_TX_FORMAT_A_SHIFT - }; - const uint32_t sign_bit[4] = { - R300_TX_FORMAT_SIGNED_X, - R300_TX_FORMAT_SIGNED_Y, - R300_TX_FORMAT_SIGNED_Z, - R300_TX_FORMAT_SIGNED_W, - }; - - desc = util_format_description(format); - - /* Colorspace (return non-RGB formats directly). */ - switch (desc->colorspace) { - /* Depth stencil formats. */ - case UTIL_FORMAT_COLORSPACE_ZS: - switch (format) { - case PIPE_FORMAT_Z16_UNORM: - return R300_EASY_TX_FORMAT(X, X, X, X, X16); - case PIPE_FORMAT_Z24X8_UNORM: - case PIPE_FORMAT_Z24S8_UNORM: - return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP); - default: - return ~0; /* Unsupported. */ - } - - /* YUV formats. */ - case UTIL_FORMAT_COLORSPACE_YUV: - result |= R300_TX_FORMAT_YUV_TO_RGB; - - switch (format) { - case PIPE_FORMAT_YCBCR: - return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result; - case PIPE_FORMAT_YCBCR_REV: - return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result; - default: - return ~0; /* Unsupported/unknown. */ - } - - /* Add gamma correction. */ - case UTIL_FORMAT_COLORSPACE_SRGB: - result |= R300_TX_FORMAT_GAMMA; - break; - - default:; - } - - /* Add swizzle. */ - for (i = 0; i < 4; i++) { - switch (desc->swizzle[i]) { - case UTIL_FORMAT_SWIZZLE_X: - case UTIL_FORMAT_SWIZZLE_NONE: - result |= R300_TX_FORMAT_X << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_Y: - result |= R300_TX_FORMAT_Y << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_Z: - result |= R300_TX_FORMAT_Z << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_W: - result |= R300_TX_FORMAT_W << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_0: - result |= R300_TX_FORMAT_ZERO << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_1: - result |= R300_TX_FORMAT_ONE << swizzle_shift[i]; - break; - default: - return ~0; /* Unsupported. */ - } - } - - /* Compressed formats. */ - if (desc->layout == UTIL_FORMAT_LAYOUT_DXT) { - switch (format) { - case PIPE_FORMAT_DXT1_RGB: - case PIPE_FORMAT_DXT1_RGBA: - case PIPE_FORMAT_DXT1_SRGB: - case PIPE_FORMAT_DXT1_SRGBA: - return R300_TX_FORMAT_DXT1 | result; - case PIPE_FORMAT_DXT3_RGBA: - case PIPE_FORMAT_DXT3_SRGBA: - return R300_TX_FORMAT_DXT3 | result; - case PIPE_FORMAT_DXT5_RGBA: - case PIPE_FORMAT_DXT5_SRGBA: - return R300_TX_FORMAT_DXT5 | result; - default: - return ~0; /* Unsupported/unknown. */ - } - } - - /* Get the number of components. */ - for (i = 0; i < 4; i++) { - if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { - ++components; - } - } +boolean r300_is_colorbuffer_format_supported(enum pipe_format format); - /* Add sign. */ - for (i = 0; i < components; i++) { - if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { - result |= sign_bit[i]; - } - } +boolean r300_is_zs_format_supported(enum pipe_format format); - /* See whether the components are of the same size. */ - for (i = 1; i < components; i++) { - uniform = uniform && desc->channel[0].size == desc->channel[i].size; - } - - /* Non-uniform formats. */ - if (!uniform) { - switch (components) { - case 3: - if (desc->channel[0].size == 5 && - desc->channel[1].size == 6 && - desc->channel[2].size == 5) { - return R300_TX_FORMAT_Z5Y6X5 | result; - } - if (desc->channel[0].size == 5 && - desc->channel[1].size == 5 && - desc->channel[2].size == 6) { - return R300_TX_FORMAT_Z6Y5X5 | result; - } - return ~0; /* Unsupported/unknown. */ - - case 4: - if (desc->channel[0].size == 5 && - desc->channel[1].size == 5 && - desc->channel[2].size == 5 && - desc->channel[3].size == 1) { - return R300_TX_FORMAT_W1Z5Y5X5 | result; - } - if (desc->channel[0].size == 10 && - desc->channel[1].size == 10 && - desc->channel[2].size == 10 && - desc->channel[3].size == 2) { - return R300_TX_FORMAT_W2Z10Y10X10 | result; - } - } - return ~0; /* Unsupported/unknown. */ - } - - /* And finally, uniform formats. */ - switch (desc->channel[0].type) { - case UTIL_FORMAT_TYPE_UNSIGNED: - case UTIL_FORMAT_TYPE_SIGNED: - if (!desc->channel[0].normalized) { - return ~0; - } - - switch (desc->channel[0].size) { - case 4: - switch (components) { - case 2: - return R300_TX_FORMAT_Y4X4 | result; - case 4: - return R300_TX_FORMAT_W4Z4Y4X4 | result; - } - return ~0; - - case 8: - switch (components) { - case 1: - return R300_TX_FORMAT_X8 | result; - case 2: - return R300_TX_FORMAT_Y8X8 | result; - case 4: - return R300_TX_FORMAT_W8Z8Y8X8 | result; - } - return ~0; - - case 16: - switch (components) { - case 1: - return R300_TX_FORMAT_X16 | result; - case 2: - return R300_TX_FORMAT_Y16X16 | result; - case 4: - return R300_TX_FORMAT_W16Z16Y16X16 | result; - } - } - return ~0; - -/* XXX Enable float textures here. */ -#if 0 - case UTIL_FORMAT_TYPE_FLOAT: - switch (desc->channel[0].size) { - case 16: - switch (components) { - case 1: - return R300_TX_FORMAT_16F | result; - case 2: - return R300_TX_FORMAT_16F_16F | result; - case 4: - return R300_TX_FORMAT_16F_16F_16F_16F | result; - } - return ~0; - - case 32: - switch (components) { - case 1: - return R300_TX_FORMAT_32F | result; - case 2: - return R300_TX_FORMAT_32F_32F | result; - case 4: - return R300_TX_FORMAT_32F_32F_32F_32F | result; - } - } -#endif - } - - return ~0; /* Unsupported/unknown. */ -} +boolean r300_is_sampler_format_supported(enum pipe_format format); struct r300_video_surface { @@ -288,13 +60,11 @@ r300_video_surface(struct pipe_video_surface *pvs) return (struct r300_video_surface *)pvs; } -#ifndef R300_WINSYS_H - +/* Used internally for texture_is_referenced() + */ boolean r300_get_texture_buffer(struct pipe_screen* screen, struct pipe_texture* texture, struct pipe_buffer** buffer, unsigned* stride); -#endif /* R300_WINSYS_H */ - #endif /* R300_TEXTURE_H */ diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c new file mode 100644 index 0000000000..ec89681a3c --- /dev/null +++ b/src/gallium/drivers/r300/r300_transfer.c @@ -0,0 +1,270 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2010 Marek Olšák <maraeo@gmail.com> + * + * 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 + * on 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 + * THE AUTHOR(S) AND/OR THEIR 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 "r300_context.h" +#include "r300_transfer.h" +#include "r300_texture.h" +#include "r300_screen.h" + +#include "util/u_memory.h" +#include "util/u_format.h" + +struct r300_transfer { + /* Parent class */ + struct pipe_transfer transfer; + + /* Pipe context. */ + struct pipe_context *ctx; + + /* Parameters of get_tex_transfer. */ + unsigned x, y, level, zslice, face; + + /* Offset from start of buffer. */ + unsigned offset; + + /* Detiled texture. */ + struct r300_texture *detiled_texture; + + /* Transfer and format flags. */ + unsigned buffer_usage, render_target_usage; +}; + +/* Convenience cast wrapper. */ +static INLINE struct r300_transfer* +r300_transfer(struct pipe_transfer* transfer) +{ + return (struct r300_transfer*)transfer; +} + +/* Copy from a tiled texture to a detiled one. */ +static void r300_copy_from_tiled_texture(struct pipe_context *ctx, + struct r300_transfer *r300transfer) +{ + struct pipe_screen *screen = ctx->screen; + struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer; + struct pipe_texture *tex = transfer->texture; + struct pipe_surface *src, *dst; + + src = screen->get_tex_surface(screen, tex, r300transfer->face, + r300transfer->level, r300transfer->zslice, + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_PIXEL); + + dst = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_WRITE | + PIPE_BUFFER_USAGE_PIXEL | + r300transfer->buffer_usage); + + ctx->surface_copy(ctx, dst, 0, 0, src, r300transfer->x, r300transfer->y, + transfer->width, transfer->height); + + pipe_surface_reference(&src, NULL); + pipe_surface_reference(&dst, NULL); +} + +/* Copy a detiled texture to a tiled one. */ +static void r300_copy_into_tiled_texture(struct pipe_context *ctx, + struct r300_transfer *r300transfer) +{ + struct pipe_screen *screen = ctx->screen; + struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer; + struct pipe_texture *tex = transfer->texture; + struct pipe_surface *src, *dst; + + src = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex, + 0, 0, 0, + PIPE_BUFFER_USAGE_GPU_READ | + PIPE_BUFFER_USAGE_PIXEL); + + dst = screen->get_tex_surface(screen, tex, r300transfer->face, + r300transfer->level, r300transfer->zslice, + PIPE_BUFFER_USAGE_GPU_WRITE | + PIPE_BUFFER_USAGE_PIXEL); + + /* XXX this flush prevents the following DRM error from occuring: + * [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation ! + * Reproducible with perf/copytex. */ + ctx->flush(ctx, 0, NULL); + + ctx->surface_copy(ctx, dst, r300transfer->x, r300transfer->y, src, 0, 0, + transfer->width, transfer->height); + + /* XXX this flush fixes a few piglit tests (e.g. glean/pixelFormats). */ + ctx->flush(ctx, 0, NULL); + + pipe_surface_reference(&src, NULL); + pipe_surface_reference(&dst, NULL); +} + +static struct pipe_transfer* +r300_get_tex_transfer(struct pipe_screen *screen, + struct pipe_texture *texture, + unsigned face, unsigned level, unsigned zslice, + enum pipe_transfer_usage usage, unsigned x, unsigned y, + unsigned w, unsigned h) +{ + struct r300_texture *tex = (struct r300_texture *)texture; + struct r300_transfer *trans; + struct r300_screen *r300screen = r300_screen(screen); + struct pipe_texture template; + + trans = CALLOC_STRUCT(r300_transfer); + if (trans) { + /* Initialize the transfer object. */ + pipe_texture_reference(&trans->transfer.texture, texture); + trans->transfer.usage = usage; + trans->transfer.width = w; + trans->transfer.height = h; + trans->ctx = r300screen->ctx; + trans->x = x; + trans->y = y; + trans->level = level; + trans->zslice = zslice; + trans->face = face; + + /* If the texture is tiled, we must create a temporary detiled texture + * for this transfer. */ + if (tex->microtile || tex->macrotile) { + trans->buffer_usage = pipe_transfer_buffer_flags(&trans->transfer); + trans->render_target_usage = + util_format_is_depth_or_stencil(texture->format) ? + PIPE_TEXTURE_USAGE_DEPTH_STENCIL : + PIPE_TEXTURE_USAGE_RENDER_TARGET; + + template.target = PIPE_TEXTURE_2D; + template.format = texture->format; + template.width0 = w; + template.height0 = h; + template.depth0 = 0; + template.last_level = 0; + template.nr_samples = 0; + template.tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC | + R300_TEXTURE_USAGE_TRANSFER; + + /* For texture reading, the temporary (detiled) texture is used as + * a render target when blitting from a tiled texture. */ + if (usage & PIPE_TRANSFER_READ) { + template.tex_usage |= trans->render_target_usage; + } + /* For texture writing, the temporary texture is used as a sampler + * when blitting into a tiled texture. */ + if (usage & PIPE_TRANSFER_WRITE) { + template.tex_usage |= PIPE_TEXTURE_USAGE_SAMPLER; + } + + /* Create the temporary texture. */ + trans->detiled_texture = + (struct r300_texture*)screen->texture_create(screen, &template); + assert(!trans->detiled_texture->microtile && + !trans->detiled_texture->macrotile); + + /* Set the stride. + * Parameters x, y, level, zslice, and face remain zero. */ + trans->transfer.stride = + r300_texture_get_stride(r300screen, trans->detiled_texture, 0); + + if (usage & PIPE_TRANSFER_READ) { + /* We cannot map a tiled texture directly because the data is + * in a different order, therefore we do detiling using a blit. */ + r300_copy_from_tiled_texture(r300screen->ctx, trans); + } + } else { + trans->transfer.x = x; + trans->transfer.y = y; + trans->transfer.stride = + r300_texture_get_stride(r300screen, tex, level); + trans->transfer.level = level; + trans->transfer.zslice = zslice; + trans->transfer.face = face; + trans->offset = r300_texture_get_offset(tex, level, zslice, face); + } + } + return &trans->transfer; +} + +static void r300_tex_transfer_destroy(struct pipe_transfer *trans) +{ + struct r300_transfer *r300transfer = r300_transfer(trans); + + if (r300transfer->detiled_texture) { + if (trans->usage & PIPE_TRANSFER_WRITE) { + r300_copy_into_tiled_texture(r300transfer->ctx, r300transfer); + } + + pipe_texture_reference( + (struct pipe_texture**)&r300transfer->detiled_texture, NULL); + } + pipe_texture_reference(&trans->texture, NULL); + FREE(trans); +} + +static void* r300_transfer_map(struct pipe_screen *screen, + struct pipe_transfer *transfer) +{ + struct r300_transfer *r300transfer = r300_transfer(transfer); + struct r300_texture *tex = (struct r300_texture*)transfer->texture; + char *map; + enum pipe_format format = tex->tex.format; + + if (r300transfer->detiled_texture) { + /* The detiled texture is of the same size as the region being mapped + * (no offset needed). */ + return pipe_buffer_map(screen, + r300transfer->detiled_texture->buffer, + pipe_transfer_buffer_flags(transfer)); + } else { + /* Tiling is disabled. */ + map = pipe_buffer_map(screen, tex->buffer, + pipe_transfer_buffer_flags(transfer)); + + if (!map) { + return NULL; + } + + return map + r300_transfer(transfer)->offset + + transfer->y / util_format_get_blockheight(format) * transfer->stride + + transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format); + } +} + +static void r300_transfer_unmap(struct pipe_screen *screen, + struct pipe_transfer *transfer) +{ + struct r300_transfer *r300transfer = r300_transfer(transfer); + struct r300_texture *tex = (struct r300_texture*)transfer->texture; + + if (r300transfer->detiled_texture) { + pipe_buffer_unmap(screen, r300transfer->detiled_texture->buffer); + } else { + pipe_buffer_unmap(screen, tex->buffer); + } +} + +void r300_init_screen_transfer_functions(struct pipe_screen *screen) +{ + screen->get_tex_transfer = r300_get_tex_transfer; + screen->tex_transfer_destroy = r300_tex_transfer_destroy; + screen->transfer_map = r300_transfer_map; + screen->transfer_unmap = r300_transfer_unmap; +} diff --git a/src/gallium/drivers/r300/r300_transfer.h b/src/gallium/drivers/r300/r300_transfer.h new file mode 100644 index 0000000000..60d1d3dc85 --- /dev/null +++ b/src/gallium/drivers/r300/r300_transfer.h @@ -0,0 +1,31 @@ +/* + * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2010 Marek Olšák <maraeo@gmail.com> + * + * 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 + * on 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 + * THE AUTHOR(S) AND/OR THEIR 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 R300_TRANSFER +#define R300_TRANSFER + +#include "pipe/p_screen.h" + +void r300_init_screen_transfer_functions(struct pipe_screen *screen); + +#endif diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c index a6786c321c..379939ac75 100644 --- a/src/gallium/drivers/r300/r300_vs.c +++ b/src/gallium/drivers/r300/r300_vs.c @@ -89,95 +89,41 @@ static void r300_shader_read_vs_outputs( assert(0); } } + + /* WPOS is a straight copy of POSITION and it's always emitted. */ + vs_outputs->wpos = i; } -static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs) +/* This function sets up: + * - VAP mapping, which maps VS registers to output semantics and + * at the same time it indicates which attributes are enabled and should + * be rasterized. + * - Stream mapping to VS outputs if TCL is not present. */ +static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs) { struct r300_shader_semantics* vs_outputs = &vs->outputs; - uint32_t* hwfmt = vs->hwfmt; - int i, gen_count; + struct r300_vap_output_state *vap_out = &vs->vap_out; + int *stream_loc = vs->stream_loc_notcl; + int i, gen_count, tabi = 0; boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || vs_outputs->bcolor[1] != ATTR_UNUSED; - /* Do the actual vertex_info setup. - * - * vertex_info has four uints of hardware-specific data in it. - * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL - * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM - * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0 - * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */ - - hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */ + vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */ /* Position. */ if (vs_outputs->pos != ATTR_UNUSED) { - hwfmt[1] |= R300_INPUT_CNTL_POS; - hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; + vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS; + vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; + + stream_loc[tabi++] = 0; } else { assert(0); } /* Point size. */ if (vs_outputs->psize != ATTR_UNUSED) { - hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; - } - - /* Colors. */ - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used || - vs_outputs->color[1] != ATTR_UNUSED) { - hwfmt[1] |= R300_INPUT_CNTL_COLOR; - hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i; - } - } - - /* Back-face colors. */ - if (any_bcolor_used) { - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - hwfmt[1] |= R300_INPUT_CNTL_COLOR; - hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i); - } - } + vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; - /* Texture coordinates. */ - gen_count = 0; - for (i = 0; i < ATTR_GENERIC_COUNT; i++) { - if (vs_outputs->generic[i] != ATTR_UNUSED) { - hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count); - hwfmt[3] |= (4 << (3 * gen_count)); - gen_count++; - } - } - - /* Fog coordinates. */ - if (vs_outputs->fog != ATTR_UNUSED) { - hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count); - hwfmt[3] |= (4 << (3 * gen_count)); - gen_count++; - } - - /* XXX magic */ - assert(gen_count <= 8); - - /* WPOS. */ - vs->wpos_tex_output = gen_count; -} - -/* Sets up stream mapping to equivalent VS outputs if TCL is bypassed - * or isn't present. */ -static void r300_stream_locations_notcl( - struct r300_shader_semantics* vs_outputs, - int* stream_loc) -{ - int i, tabi = 0, gen_count; - boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || - vs_outputs->bcolor[1] != ATTR_UNUSED; - - /* Position. */ - stream_loc[tabi++] = 0; - - /* Point size. */ - if (vs_outputs->psize != ATTR_UNUSED) { stream_loc[tabi++] = 1; } @@ -185,6 +131,9 @@ static void r300_stream_locations_notcl( for (i = 0; i < ATTR_COLOR_COUNT; i++) { if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used || vs_outputs->color[1] != ATTR_UNUSED) { + vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; + vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i; + stream_loc[tabi++] = 2 + i; } } @@ -192,6 +141,9 @@ static void r300_stream_locations_notcl( /* Back-face colors. */ if (any_bcolor_used) { for (i = 0; i < ATTR_COLOR_COUNT; i++) { + vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; + vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i); + stream_loc[tabi++] = 4 + i; } } @@ -200,6 +152,9 @@ static void r300_stream_locations_notcl( gen_count = 0; for (i = 0; i < ATTR_GENERIC_COUNT; i++) { if (vs_outputs->generic[i] != ATTR_UNUSED) { + vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count); + vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count)); + assert(tabi < 16); stream_loc[tabi++] = 6 + gen_count; gen_count++; @@ -208,17 +163,22 @@ static void r300_stream_locations_notcl( /* Fog coordinates. */ if (vs_outputs->fog != ATTR_UNUSED) { + vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count); + vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count)); + assert(tabi < 16); stream_loc[tabi++] = 6 + gen_count; gen_count++; } + /* XXX magic */ + assert(gen_count <= 8); + /* WPOS. */ - if (vs_outputs->wpos != ATTR_UNUSED) { - assert(tabi < 16); - stream_loc[tabi++] = 6 + gen_count; - gen_count++; - } + vs->wpos_tex_output = gen_count; + + assert(tabi < 16); + stream_loc[tabi++] = 6 + gen_count; for (; tabi < 16;) { stream_loc[tabi++] = -1; @@ -294,26 +254,16 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) } } -static void r300_insert_wpos(struct r300_vertex_program_compiler* c, - struct r300_shader_semantics* outputs) +void r300_vertex_shader_common_init(struct r300_vertex_shader *vs, + const struct pipe_shader_state *shader) { - int i, lastOutput = 0; - - /* Find the max output index. */ - lastOutput = MAX2(lastOutput, outputs->psize); - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - lastOutput = MAX2(lastOutput, outputs->color[i]); - lastOutput = MAX2(lastOutput, outputs->bcolor[i]); - } - for (i = 0; i < ATTR_GENERIC_COUNT; i++) { - lastOutput = MAX2(lastOutput, outputs->generic[i]); - } - lastOutput = MAX2(lastOutput, outputs->fog); + /* Copy state directly into shader. */ + vs->state = *shader; + vs->state.tokens = tgsi_dup_tokens(shader->tokens); + tgsi_scan_shader(shader->tokens, &vs->info); - /* Set WPOS after the last output. */ - lastOutput++; - rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */ - outputs->wpos = lastOutput; + r300_shader_read_vs_outputs(&vs->info, &vs->outputs); + r300_init_vs_output_mapping(vs); } void r300_translate_vertex_shader(struct r300_context* r300, @@ -322,9 +272,6 @@ void r300_translate_vertex_shader(struct r300_context* r300, struct r300_vertex_program_compiler compiler; struct tgsi_to_rc ttr; - /* Initialize. */ - r300_shader_read_vs_outputs(&vs->info, &vs->outputs); - /* Setup the compiler */ rc_init(&compiler.Base); @@ -348,10 +295,7 @@ void r300_translate_vertex_shader(struct r300_context* r300, compiler.SetHwInputOutput = &set_vertex_inputs_outputs; /* Insert the WPOS output. */ - r300_insert_wpos(&compiler, &vs->outputs); - - r300_shader_vap_output_fmt(vs); - r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl); + rc_copy_output(&compiler.Base, 0, vs->outputs.wpos); /* Invoke the compiler */ r3xx_compile_vertex_program(&compiler); @@ -363,30 +307,29 @@ void r300_translate_vertex_shader(struct r300_context* r300, /* And, finally... */ rc_destroy(&compiler.Base); - vs->translated = TRUE; } boolean r300_vertex_shader_setup_wpos(struct r300_context* r300) { - struct r300_vertex_shader* vs = r300->vs; - int tex_output = r300->vs->wpos_tex_output; + struct r300_vertex_shader* vs = r300->vs_state.state; + struct r300_vap_output_state *vap_out = &vs->vap_out; + int tex_output = vs->wpos_tex_output; uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output; - uint32_t* hwfmt = vs->hwfmt; if (r300->fs->inputs.wpos != ATTR_UNUSED) { /* Enable WPOS in VAP. */ - if (!(hwfmt[1] & tex_fmt)) { - hwfmt[1] |= tex_fmt; - hwfmt[3] |= (4 << (3 * tex_output)); + if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) { + vap_out->vap_vsm_vtx_assm |= tex_fmt; + vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output)); assert(tex_output < 8); return TRUE; } } else { /* Disable WPOS in VAP. */ - if (hwfmt[1] & tex_fmt) { - hwfmt[1] &= ~tex_fmt; - hwfmt[3] &= ~(4 << (3 * tex_output)); + if (vap_out->vap_vsm_vtx_assm & tex_fmt) { + vap_out->vap_vsm_vtx_assm &= ~tex_fmt; + vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output)); return TRUE; } } diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h index 18cfeee3cd..f6f0b86b68 100644 --- a/src/gallium/drivers/r300/r300_vs.h +++ b/src/gallium/drivers/r300/r300_vs.h @@ -28,6 +28,7 @@ #include "tgsi/tgsi_scan.h" #include "radeon_code.h" +#include "r300_context.h" #include "r300_shader_semantics.h" struct r300_context; @@ -38,7 +39,7 @@ struct r300_vertex_shader { struct tgsi_shader_info info; struct r300_shader_semantics outputs; - uint hwfmt[4]; + struct r300_vap_output_state vap_out; /* Stream locations for SWTCL or if TCL is bypassed. */ int stream_loc_notcl[16]; @@ -46,13 +47,17 @@ struct r300_vertex_shader { /* Output stream location for WPOS. */ int wpos_tex_output; - /* Has this shader been translated yet? */ - boolean translated; - + /* HWTCL-specific. */ /* Machine code (if translated) */ struct r300_vertex_program_code code; + + /* SWTCL-specific. */ + void *draw_vs; }; +void r300_vertex_shader_common_init(struct r300_vertex_shader *vs, + const struct pipe_shader_state *shader); + void r300_translate_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs); diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h index 40fb8a95ca..ddf2b79003 100644 --- a/src/gallium/drivers/r300/r300_winsys.h +++ b/src/gallium/drivers/r300/r300_winsys.h @@ -40,11 +40,6 @@ struct radeon_winsys; struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys); -boolean r300_get_texture_buffer(struct pipe_screen* screen, - struct pipe_texture* texture, - struct pipe_buffer** buffer, - unsigned* stride); - #ifdef __cplusplus } #endif |