X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fr300%2Fr300_render.c;h=db6b77a0c22824fe6b02cbece66510c9329caa81;hb=f1d785188b170f1d3e79ba257cb8634f13685bca;hp=63b21b9379abf5056bb47d24e0468441d8196770;hpb=d619cceea47dc3070ebb7f7ea4f8b6b31a672d38;p=mesa.git diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c index 63b21b9379a..db6b77a0c22 100644 --- a/src/mesa/drivers/dri/r300/r300_render.c +++ b/src/mesa/drivers/dri/r300/r300_render.c @@ -24,12 +24,22 @@ OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. **************************************************************************/ - /* - * Authors: - * Nicolai Haehnle + * \file + * + * \brief R300 Render (Vertex Buffer Implementation) + * + * The immediate implementation has been removed from CVS in favor of the vertex + * buffer implementation. + * + * When falling back to software TCL still attempt to use hardware + * rasterization. + * + * I am not sure that the cache related registers are setup correctly, but + * obviously this does work... Further investigation is needed. + * + * \author Nicolai Haehnle */ - #include "glheader.h" #include "state.h" #include "imports.h" @@ -38,14 +48,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "context.h" #include "dd.h" #include "simple_list.h" - #include "api_arrayelt.h" #include "swrast/swrast.h" #include "swrast_setup/swrast_setup.h" #include "vbo/vbo.h" #include "tnl/tnl.h" #include "tnl/t_vp_build.h" - #include "radeon_reg.h" #include "radeon_macros.h" #include "radeon_ioctl.h" @@ -58,147 +66,129 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_tex.h" #include "r300_maos.h" #include "r300_emit.h" - extern int future_hw_tcl_on; -/********************************************************************** -* Hardware rasterization -* -* When we fell back to software TCL, we still try to use the -* rasterization hardware for rendering. -**********************************************************************/ - -static int r300_get_primitive_type(r300ContextPtr rmesa, GLcontext *ctx, int prim) +static int r300PrimitiveType(r300ContextPtr rmesa, GLcontext * ctx, + int prim) { - int type=-1; + int type = -1; switch (prim & PRIM_MODE_MASK) { case GL_POINTS: - type=R300_VAP_VF_CNTL__PRIM_POINTS; - break; + type = R300_VAP_VF_CNTL__PRIM_POINTS; + break; case GL_LINES: - type=R300_VAP_VF_CNTL__PRIM_LINES; - break; + type = R300_VAP_VF_CNTL__PRIM_LINES; + break; case GL_LINE_STRIP: - type=R300_VAP_VF_CNTL__PRIM_LINE_STRIP; - break; + type = R300_VAP_VF_CNTL__PRIM_LINE_STRIP; + break; case GL_LINE_LOOP: - type=R300_VAP_VF_CNTL__PRIM_LINE_LOOP; - break; - case GL_TRIANGLES: - type=R300_VAP_VF_CNTL__PRIM_TRIANGLES; - break; - case GL_TRIANGLE_STRIP: - type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP; - break; - case GL_TRIANGLE_FAN: - type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN; - break; + type = R300_VAP_VF_CNTL__PRIM_LINE_LOOP; + break; + case GL_TRIANGLES: + type = R300_VAP_VF_CNTL__PRIM_TRIANGLES; + break; + case GL_TRIANGLE_STRIP: + type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP; + break; + case GL_TRIANGLE_FAN: + type = R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN; + break; case GL_QUADS: - type=R300_VAP_VF_CNTL__PRIM_QUADS; - break; + type = R300_VAP_VF_CNTL__PRIM_QUADS; + break; case GL_QUAD_STRIP: - type=R300_VAP_VF_CNTL__PRIM_QUAD_STRIP; - break; + type = R300_VAP_VF_CNTL__PRIM_QUAD_STRIP; + break; case GL_POLYGON: - type=R300_VAP_VF_CNTL__PRIM_POLYGON; + type = R300_VAP_VF_CNTL__PRIM_POLYGON; break; - default: - fprintf(stderr, "%s:%s Do not know how to handle primitive %02x - help me !\n", - __FILE__, __FUNCTION__, - prim & PRIM_MODE_MASK); + default: + fprintf(stderr, + "%s:%s Do not know how to handle primitive 0x%04x - help me !\n", + __FILE__, __FUNCTION__, prim & PRIM_MODE_MASK); return -1; - break; - } - return type; + break; + } + return type; } -int r300_get_num_verts(r300ContextPtr rmesa, int num_verts, int prim) +static int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim) { - int verts_off=0; - char *name="UNKNOWN"; + int verts_off = 0; switch (prim & PRIM_MODE_MASK) { case GL_POINTS: - name="P"; verts_off = 0; - break; + break; case GL_LINES: - name="L"; verts_off = num_verts % 2; - break; + break; case GL_LINE_STRIP: - name="LS"; - if(num_verts < 2) + if (num_verts < 2) verts_off = num_verts; - break; + break; case GL_LINE_LOOP: - name="LL"; - if(num_verts < 2) + if (num_verts < 2) verts_off = num_verts; - break; - case GL_TRIANGLES: - name="T"; + break; + case GL_TRIANGLES: verts_off = num_verts % 3; - break; - case GL_TRIANGLE_STRIP: - name="TS"; - if(num_verts < 3) + break; + case GL_TRIANGLE_STRIP: + if (num_verts < 3) verts_off = num_verts; - break; - case GL_TRIANGLE_FAN: - name="TF"; - if(num_verts < 3) + break; + case GL_TRIANGLE_FAN: + if (num_verts < 3) verts_off = num_verts; - break; + break; case GL_QUADS: - name="Q"; verts_off = num_verts % 4; - break; + break; case GL_QUAD_STRIP: - name="QS"; - if(num_verts < 4) + if (num_verts < 4) verts_off = num_verts; else verts_off = num_verts % 2; - break; + break; case GL_POLYGON: - name="P"; - if(num_verts < 3) + if (num_verts < 3) verts_off = num_verts; break; - default: - fprintf(stderr, "%s:%s Do not know how to handle primitive %02x - help me !\n", - __FILE__, __FUNCTION__, - prim & PRIM_MODE_MASK); + default: + fprintf(stderr, + "%s:%s Do not know how to handle primitive 0x%04x - help me !\n", + __FILE__, __FUNCTION__, prim & PRIM_MODE_MASK); return -1; - break; - } + break; + } if (RADEON_DEBUG & DEBUG_VERTS) { if (num_verts - verts_off == 0) { - WARN_ONCE("user error: Need more than %d vertices to draw primitive %s !\n", num_verts, name); + WARN_ONCE + ("user error: Need more than %d vertices to draw primitive 0x%04x !\n", + num_verts, prim & PRIM_MODE_MASK); return 0; } if (verts_off > 0) { - WARN_ONCE("user error: %d is not a valid number of vertices for primitive %s !\n", num_verts, name); + WARN_ONCE + ("user error: %d is not a valid number of vertices for primitive 0x%04x !\n", + num_verts, prim & PRIM_MODE_MASK); } } return num_verts - verts_off; } -/* Immediate implementation has been removed from CVS. */ - -/* vertex buffer implementation */ - -static void inline fire_EB(r300ContextPtr rmesa, unsigned long addr, int vertex_count, int type, int elt_size) +static void inline r300FireEB(r300ContextPtr rmesa, unsigned long addr, + int vertex_count, int type, int elt_size) { int cmd_reserved = 0; int cmd_written = 0; drm_radeon_cmd_header_t *cmd = NULL; - unsigned long addr_a; unsigned long t_addr; unsigned long magic_1, magic_2; GLcontext *ctx; @@ -206,12 +196,10 @@ static void inline fire_EB(r300ContextPtr rmesa, unsigned long addr, int vertex_ assert(elt_size == 2 || elt_size == 4); - if(addr & (elt_size-1)){ + if (addr & (elt_size - 1)) { WARN_ONCE("Badly aligned buffer\n"); - return ; + return; } -#ifdef OPTIMIZE_ELTS - addr_a = 0; magic_1 = (addr % 32) / 4; t_addr = addr & (~0x1d); @@ -220,106 +208,69 @@ static void inline fire_EB(r300ContextPtr rmesa, unsigned long addr, int vertex_ check_space(6); start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0); - if(elt_size == 4){ - e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit); + if (elt_size == 4) { + e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + (vertex_count << 16) | type | + R300_VAP_VF_CNTL__INDEX_SIZE_32bit); } else { - e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type); + e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + (vertex_count << 16) | type); } start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2); - if(elt_size == 4){ +#ifdef OPTIMIZE_ELTS + if (elt_size == 4) { e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2); - e32(addr /*& 0xffffffe3*/); + e32(addr); } else { e32(R300_EB_UNK1 | (magic_1 << 16) | R300_EB_UNK2); e32(t_addr); } - - if(elt_size == 4){ - e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */ - } else { - e32(magic_2); /* Total number of dwords needed? */ - } - //cp_delay(rmesa, 1); -#if 0 - fprintf(stderr, "magic_1 %d\n", magic_1); - fprintf(stderr, "t_addr %x\n", t_addr); - fprintf(stderr, "magic_2 %d\n", magic_2); - exit(1); -#endif #else - (void)magic_2, (void)magic_1, (void)t_addr; - - addr_a = 0; - - check_space(6); - - start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0); - if(elt_size == 4){ - e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit); - } else { - e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type); - } - - start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2); e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2); - e32(addr /*& 0xffffffe3*/); + e32(addr); +#endif - if(elt_size == 4){ - e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */ + if (elt_size == 4) { + e32(vertex_count); } else { - e32((vertex_count+1)/2 /*+ addr_a/4*/); /* Total number of dwords needed? */ - } - //cp_delay(rmesa, 1); +#ifdef OPTIMIZE_ELTS + e32(magic_2); +#else + e32((vertex_count + 1) / 2); #endif + } } -static void r300_render_vb_primitive(r300ContextPtr rmesa, - GLcontext *ctx, - int start, - int end, - int prim) +static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx, + int start, int end, int prim) { - int type, num_verts; - - type=r300_get_primitive_type(rmesa, ctx, prim); - num_verts=r300_get_num_verts(rmesa, end-start, prim); + int type, num_verts; - if(type<0 || num_verts <= 0)return; + type = r300PrimitiveType(rmesa, ctx, prim); + num_verts = r300NumVerts(rmesa, end - start, prim); - if(rmesa->state.VB.Elts){ - r300EmitAOS(rmesa, rmesa->state.aos_count, /*0*/start); -#if 0 - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; - int i; - start_index32_packet(num_verts, type); - for(i=0; i < num_verts; i++) - e32(((unsigned long *)rmesa->state.VB.Elts)[i]/*rmesa->state.Elts[start+i]*/); /* start ? */ -#else - if(num_verts == 1){ - //start_index32_packet(num_verts, type); - //e32(rmesa->state.Elts[start]); + if (type < 0 || num_verts <= 0) return; - } - if(num_verts > 65535){ /* not implemented yet */ - WARN_ONCE("Too many elts\n"); - return; + if (rmesa->state.VB.Elts) { + r300EmitAOS(rmesa, rmesa->state.aos_count, /*0 */ start); + if (num_verts > 65535) { /* not implemented yet */ + WARN_ONCE("Too many elts\n"); + return; + } + r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts, + rmesa->state.VB.elt_size); + r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset, + num_verts, type, rmesa->state.VB.elt_size); + } else { + r300EmitAOS(rmesa, rmesa->state.aos_count, start); + fire_AOS(rmesa, num_verts, type); } - - r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts, rmesa->state.VB.elt_size); - fire_EB(rmesa, rmesa->state.elt_dma.aos_offset, num_verts, type, rmesa->state.VB.elt_size); -#endif - }else{ - r300EmitAOS(rmesa, rmesa->state.aos_count, start); - fire_AOS(rmesa, num_verts, type); - } } -GLboolean r300_run_vb_render(GLcontext *ctx, - struct tnl_pipeline_stage *stage) +static GLboolean r300RunRender(GLcontext * ctx, + struct tnl_pipeline_stage *stage) { r300ContextPtr rmesa = R300_CONTEXT(ctx); struct radeon_vertex_buffer *VB = &rmesa->state.VB; @@ -328,12 +279,11 @@ GLboolean r300_run_vb_render(GLcontext *ctx, int cmd_written = 0; drm_radeon_cmd_header_t *cmd = NULL; - if (RADEON_DEBUG & DEBUG_PRIMS) fprintf(stderr, "%s\n", __FUNCTION__); if (stage) { - TNLcontext *tnl = TNL_CONTEXT(ctx); + TNLcontext *tnl = TNL_CONTEXT(ctx); radeon_vb_to_rvb(rmesa, VB, &tnl->vb); } @@ -343,27 +293,29 @@ GLboolean r300_run_vb_render(GLcontext *ctx, r300UpdateShaderStates(rmesa); - reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0); + reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0); e32(R300_RB3D_DSTCACHE_UNKNOWN_0A); - reg_start(R300_RB3D_ZCACHE_CTLSTAT,0); + reg_start(R300_RB3D_ZCACHE_CTLSTAT, 0); e32(R300_RB3D_ZCACHE_UNKNOWN_03); r300EmitState(rmesa); - for(i=0; i < VB->PrimitiveCount; i++){ + for (i = 0; i < VB->PrimitiveCount; i++) { GLuint prim = _tnl_translate_prim(&VB->Primitive[i]); GLuint start = VB->Primitive[i].start; GLuint length = VB->Primitive[i].count; - r300_render_vb_primitive(rmesa, ctx, start, start + length, prim); + r300RunRenderPrimitive(rmesa, ctx, start, start + length, + prim); } - reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0); - e32(R300_RB3D_DSTCACHE_UNKNOWN_0A /*R300_RB3D_DSTCACHE_UNKNOWN_02*/); + reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0); + e32(R300_RB3D_DSTCACHE_UNKNOWN_0A + /*R300_RB3D_DSTCACHE_UNKNOWN_02 */ ); - reg_start(R300_RB3D_ZCACHE_CTLSTAT,0); - e32(R300_RB3D_ZCACHE_UNKNOWN_03 /*R300_RB3D_ZCACHE_UNKNOWN_01*/); + reg_start(R300_RB3D_ZCACHE_CTLSTAT, 0); + e32(R300_RB3D_ZCACHE_UNKNOWN_03 /*R300_RB3D_ZCACHE_UNKNOWN_01 */ ); #ifdef USER_BUFFERS r300UseArrays(ctx); @@ -382,61 +334,45 @@ GLboolean r300_run_vb_render(GLcontext *ctx, } \ } while(0) -int r300Fallback(GLcontext *ctx) +int r300Fallback(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); - struct r300_fragment_program *rp = - (struct r300_fragment_program *) - (char *)ctx->FragmentProgram._Current; - int i; - - if (rp) { - if (!rp->translated) - r300_translate_fragment_shader(r300, rp); + struct r300_fragment_program *fp = (struct r300_fragment_program *) + (char *)ctx->FragmentProgram._Current; - FALLBACK_IF(!rp->translated); + if (fp) { + if (!fp->translated) + r300_translate_fragment_shader(r300, fp); + FALLBACK_IF(!fp->translated); } - /* We do not do SELECT or FEEDBACK (yet ?) - * Is it worth doing them ? - */ FALLBACK_IF(ctx->RenderMode != GL_RENDER); -#if 0 - /* These should work now.. */ - FALLBACK_IF(ctx->Color.DitherFlag); - /* GL_ALPHA_TEST */ - FALLBACK_IF(ctx->Color.AlphaEnabled); - /* GL_BLEND */ - FALLBACK_IF(ctx->Color.BlendEnabled); - /* GL_POLYGON_OFFSET_FILL */ - FALLBACK_IF(ctx->Polygon.OffsetFill); - /* FOG seems to trigger an unknown output - * in vertex program. - */ - FALLBACK_IF(ctx->Fog.Enabled); -#endif - FALLBACK_IF(ctx->Stencil._TestTwoSide && - (ctx->Stencil.Ref[0] != ctx->Stencil.Ref[1] || - ctx->Stencil.ValueMask[0] != ctx->Stencil.ValueMask[1] || - ctx->Stencil.WriteMask[0] != ctx->Stencil.WriteMask[1])); + FALLBACK_IF(ctx->Stencil._TestTwoSide + && (ctx->Stencil.Ref[0] != ctx->Stencil.Ref[1] + || ctx->Stencil.ValueMask[0] != + ctx->Stencil.ValueMask[1] + || ctx->Stencil.WriteMask[0] != + ctx->Stencil.WriteMask[1])); + + /* GL_COLOR_LOGIC_OP */ + FALLBACK_IF(ctx->Color.ColorLogicOpEnabled); - if(!r300->disable_lowimpact_fallback){ + /* GL_POINT_SPRITE_ARB, GL_POINT_SPRITE_NV */ + if (ctx->Extensions.NV_point_sprite + || ctx->Extensions.ARB_point_sprite) + FALLBACK_IF(ctx->Point.PointSprite); + + if (!r300->disable_lowimpact_fallback) { /* GL_POLYGON_OFFSET_POINT */ FALLBACK_IF(ctx->Polygon.OffsetPoint); /* GL_POLYGON_OFFSET_LINE */ FALLBACK_IF(ctx->Polygon.OffsetLine); -#if 0 - /* GL_STENCIL_TEST */ - FALLBACK_IF(ctx->Stencil.Enabled); - /* GL_POLYGON_SMOOTH disabling to get blender going */ - FALLBACK_IF(ctx->Polygon.SmoothFlag); -#endif /* GL_POLYGON_STIPPLE */ FALLBACK_IF(ctx->Polygon.StippleFlag); - /* GL_MULTISAMPLE_ARB */ + /* GL_MULTISAMPLE */ FALLBACK_IF(ctx->Multisample.Enabled); - /* blender ? */ + /* GL_LINE_STIPPLE */ FALLBACK_IF(ctx->Line.StippleFlag); /* GL_LINE_SMOOTH */ FALLBACK_IF(ctx->Line.SmoothFlag); @@ -444,15 +380,6 @@ int r300Fallback(GLcontext *ctx) FALLBACK_IF(ctx->Point.SmoothFlag); } - /* Fallback for LOGICOP */ - FALLBACK_IF(ctx->Color.ColorLogicOpEnabled); - - /* Rest could be done with vertex fragments */ - if (ctx->Extensions.NV_point_sprite || - ctx->Extensions.ARB_point_sprite) - /* GL_POINT_SPRITE_NV */ - FALLBACK_IF(ctx->Point.PointSprite); - return R300_FALLBACK_NONE; } @@ -462,39 +389,30 @@ int r300Fallback(GLcontext *ctx) * rasterization) or false to indicate that the pipeline has finished * after we render something. */ -static GLboolean r300_run_render(GLcontext *ctx, - struct tnl_pipeline_stage *stage) +static GLboolean r300RunNonTNLRender(GLcontext * ctx, + struct tnl_pipeline_stage *stage) { - if (RADEON_DEBUG & DEBUG_PRIMS) fprintf(stderr, "%s\n", __FUNCTION__); if (r300Fallback(ctx) >= R300_FALLBACK_RAST) return GL_TRUE; - return r300_run_vb_render(ctx, stage); + return r300RunRender(ctx, stage); } -const struct tnl_pipeline_stage _r300_render_stage = { - "r300 hw rasterize", - NULL, - NULL, - NULL, - NULL, - r300_run_render /* run */ -}; - -static GLboolean r300_run_tcl_render(GLcontext *ctx, - struct tnl_pipeline_stage *stage) +static GLboolean r300RunTCLRender(GLcontext * ctx, + struct tnl_pipeline_stage *stage) { r300ContextPtr rmesa = R300_CONTEXT(ctx); struct r300_vertex_program *vp; - hw_tcl_on=future_hw_tcl_on; + hw_tcl_on = future_hw_tcl_on; if (RADEON_DEBUG & DEBUG_PRIMS) fprintf(stderr, "%s\n", __FUNCTION__); - if(hw_tcl_on == GL_FALSE) + + if (hw_tcl_on == GL_FALSE) return GL_TRUE; if (r300Fallback(ctx) >= R300_FALLBACK_TCL) { @@ -505,42 +423,28 @@ static GLboolean r300_run_tcl_render(GLcontext *ctx, r300UpdateShaders(rmesa); vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx); -#if 0 /* Draw every second request with software arb vp */ - vp->native++; - vp->native &= 1; - //vp->native = GL_FALSE; -#endif - -#if 0 /* You dont want to know what this does... */ - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct tnl_cache *cache; - struct tnl_cache_item *c; - - cache = tnl->vp_cache; - c = cache->items[0xc000cc0e % cache->size]; - - if(c && c->data == vp) - vp->native = GL_FALSE; - -#endif -#if 0 - vp->native = GL_FALSE; -#endif if (vp->native == GL_FALSE) { hw_tcl_on = GL_FALSE; return GL_TRUE; } - //r300UpdateShaderStates(rmesa); - return r300_run_vb_render(ctx, stage); + return r300RunRender(ctx, stage); } -const struct tnl_pipeline_stage _r300_tcl_stage = { - "r300 tcl", +const struct tnl_pipeline_stage _r300_render_stage = { + "r300 Hardware Rasterization", NULL, NULL, NULL, NULL, - r300_run_tcl_render /* run */ + r300RunNonTNLRender }; +const struct tnl_pipeline_stage _r300_tcl_stage = { + "r300 Hardware Transform, Clipping and Lighting", + NULL, + NULL, + NULL, + NULL, + r300RunTCLRender +};