diff options
Diffstat (limited to 'src/mesa/drivers/dri/r300/r300_fragprog_common.c')
-rw-r--r-- | src/mesa/drivers/dri/r300/r300_fragprog_common.c | 364 |
1 files changed, 114 insertions, 250 deletions
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c index f5c4c0f4a0..6674efc5bc 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_common.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c @@ -42,21 +42,51 @@ #include "shader/prog_parameter.h" #include "shader/prog_print.h" +#include "compiler/radeon_compiler.h" + #include "r300_state.h" -#include "r300_fragprog.h" -#include "r300_fragprog_swizzle.h" -#include "r500_fragprog.h" -#include "radeon_program.h" -#include "radeon_program_alu.h" -static void nqssadce_init(struct nqssadce_state* s) +static GLuint build_dtm(GLuint depthmode) +{ + switch(depthmode) { + default: + case GL_LUMINANCE: return 0; + case GL_INTENSITY: return 1; + case GL_ALPHA: return 2; + } +} + +static GLuint build_func(GLuint comparefunc) { - s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW; - s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W; + return comparefunc - GL_NEVER; } /** + * Collect all external state that is relevant for compiling the given + * fragment program. + */ +static void build_state( + r300ContextPtr r300, + struct gl_fragment_program *fp, + struct r300_fragment_program_external_state *state) +{ + int unit; + + _mesa_bzero(state, sizeof(*state)); + + for(unit = 0; unit < 16; ++unit) { + if (fp->Base.ShadowSamplers & (1 << unit)) { + struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; + + state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); + state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); + } + } +} + + +/** * Transform the program to support fragment.position. * * Introduce a small fragment at the start of the program that will be @@ -65,104 +95,26 @@ static void nqssadce_init(struct nqssadce_state* s) * to read from a newly allocated temporary. * */ -static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) +static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp) { - GLuint InputsRead = compiler->fp->Base->InputsRead; + int i; - if (!(InputsRead & FRAG_BIT_WPOS)) { - compiler->fp->wpos_attr = FRAG_ATTRIB_MAX; + if (!(compiler->Base.Program.InputsRead & FRAG_BIT_WPOS)) { + fp->wpos_attr = FRAG_ATTRIB_MAX; return; } - static gl_state_index tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0 - }; - struct prog_instruction *fpi; - GLuint window_index; - int i = 0; - for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i) { - if (!(InputsRead & (1 << i))) { - InputsRead &= ~(1 << FRAG_ATTRIB_WPOS); - InputsRead |= 1 << i; - compiler->fp->Base->InputsRead = InputsRead; - compiler->fp->wpos_attr = i; + if (!(compiler->Base.Program.InputsRead & (1 << i))) { + fp->wpos_attr = i; break; } } - GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY); - - _mesa_insert_instructions(compiler->program, 0, 3); - fpi = compiler->program->Instructions; - i = 0; - - /* perspective divide */ - fpi[i].Opcode = OPCODE_RCP; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_W; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW; - i++; - - fpi[i].Opcode = OPCODE_MUL; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW; - - fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[1].Index = tempregi; - fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW; - i++; - - /* viewport transformation */ - window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens); - - fpi[i].Opcode = OPCODE_MAD; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[0].Index = tempregi; - fpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[1].Index = window_index; - fpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[2].Index = window_index; - fpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - i++; - - for (; i < compiler->program->NumInstructions; ++i) { - int reg; - for (reg = 0; reg < 3; reg++) { - if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT && - fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) { - fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[reg].Index = tempregi; - } - } - } + rc_transform_fragment_wpos(&compiler->Base, FRAG_ATTRIB_WPOS, fp->wpos_attr); } - /** * Rewrite fragment.fogcoord to use a texture coordinate slot. * Note that fogcoord is forced into an X001 pattern, and this enforcement @@ -170,205 +122,117 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) * * See also the counterpart rewriting for vertex programs. */ -static void rewriteFog(struct r300_fragment_program_compiler *compiler) +static void rewriteFog(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp) { - struct r300_fragment_program *fp = compiler->fp; - GLuint InputsRead; + struct prog_src_register src; int i; - InputsRead = fp->Base->InputsRead; - - if (!(InputsRead & FRAG_BIT_FOGC)) { + if (!(compiler->Base.Program.InputsRead & FRAG_BIT_FOGC)) { fp->fog_attr = FRAG_ATTRIB_MAX; return; } for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i) { - if (!(InputsRead & (1 << i))) { - InputsRead &= ~(1 << FRAG_ATTRIB_FOGC); - InputsRead |= 1 << i; - fp->Base->InputsRead = InputsRead; + if (!(compiler->Base.Program.InputsRead & (1 << i))) { fp->fog_attr = i; break; } } - { - struct prog_instruction *inst; - - inst = compiler->program->Instructions; - while (inst->Opcode != OPCODE_END) { - const int src_regs = _mesa_num_inst_src_regs(inst->Opcode); - for (i = 0; i < src_regs; ++i) { - if (inst->SrcReg[i].File == PROGRAM_INPUT && inst->SrcReg[i].Index == FRAG_ATTRIB_FOGC) { - inst->SrcReg[i].Index = fp->fog_attr; - inst->SrcReg[i].Swizzle = combine_swizzles( - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE), - inst->SrcReg[i].Swizzle); - } - } - ++inst; - } - } -} - -static GLuint build_dtm(GLuint depthmode) -{ - switch(depthmode) { - default: - case GL_LUMINANCE: return 0; - case GL_INTENSITY: return 1; - case GL_ALPHA: return 2; - } + memset(&src, 0, sizeof(src)); + src.File = PROGRAM_INPUT; + src.Index = fp->fog_attr; + src.Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE); + rc_move_input(&compiler->Base, FRAG_ATTRIB_FOGC, src); } -static GLuint build_func(GLuint comparefunc) -{ - return comparefunc - GL_NEVER; -} /** - * Collect all external state that is relevant for compiling the given - * fragment program. + * Reserve hardware temporary registers for the program inputs. + * + * @note This allocation is performed explicitly, because the order of inputs + * is determined by the RS hardware. */ -static void build_state( - r300ContextPtr r300, - struct gl_fragment_program *fp, - struct r300_fragment_program_external_state *state) +static void allocate_hw_inputs( + struct r300_fragment_program_compiler * c, + void (*allocate)(void * data, unsigned input, unsigned hwreg), + void * mydata) { - int unit; - - _mesa_bzero(state, sizeof(*state)); - - for(unit = 0; unit < 16; ++unit) { - if (fp->Base.ShadowSamplers & (1 << unit)) { - struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; - - state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); - state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); - } + GLuint InputsRead = c->Base.Program.InputsRead; + int i; + GLuint hwindex = 0; + + /* Primary colour */ + if (InputsRead & FRAG_BIT_COL0) + allocate(mydata, FRAG_ATTRIB_COL0, hwindex++); + InputsRead &= ~FRAG_BIT_COL0; + + /* Secondary color */ + if (InputsRead & FRAG_BIT_COL1) + allocate(mydata, FRAG_ATTRIB_COL1, hwindex++); + InputsRead &= ~FRAG_BIT_COL1; + + /* Texcoords */ + for (i = 0; i < 8; i++) { + if (InputsRead & (FRAG_BIT_TEX0 << i)) + allocate(mydata, FRAG_ATTRIB_TEX0+i, hwindex++); } -} + InputsRead &= ~FRAG_BITS_TEX_ANY; -static void rewrite_depth_out(struct gl_program *prog) -{ - struct prog_instruction *inst; + /* Fogcoords treated as a texcoord */ + if (InputsRead & FRAG_BIT_FOGC) + allocate(mydata, FRAG_ATTRIB_FOGC, hwindex++); + InputsRead &= ~FRAG_BIT_FOGC; - for (inst = prog->Instructions; inst->Opcode != OPCODE_END; ++inst) { - if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != FRAG_RESULT_DEPTH) - continue; + /* fragment position treated as a texcoord */ + if (InputsRead & FRAG_BIT_WPOS) + allocate(mydata, FRAG_ATTRIB_WPOS, hwindex++); + InputsRead &= ~FRAG_BIT_WPOS; - if (inst->DstReg.WriteMask & WRITEMASK_Z) { - inst->DstReg.WriteMask = WRITEMASK_W; - } else { - inst->DstReg.WriteMask = 0; - continue; - } - - switch (inst->Opcode) { - case OPCODE_FRC: - case OPCODE_MOV: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - break; - case OPCODE_ADD: - case OPCODE_MAX: - case OPCODE_MIN: - case OPCODE_MUL: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); - break; - case OPCODE_CMP: - case OPCODE_MAD: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); - inst->SrcReg[2] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[2]); - break; - default: - // Scalar instructions needn't be reswizzled - break; - } - } + /* Anything else */ + if (InputsRead) + rc_error(&c->Base, "Don't know how to handle inputs 0x%x\n", InputsRead); } -void r300TranslateFragmentShader(GLcontext *ctx, struct r300_fragment_program *fp) + +static void translate_fragment_program(GLcontext *ctx, struct r300_fragment_program_cont *cont, struct r300_fragment_program *fp) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_fragment_program_compiler compiler; - compiler.r300 = r300; - compiler.fp = fp; + rc_init(&compiler.Base); + compiler.Base.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE; + compiler.code = &fp->code; - compiler.program = fp->Base; + compiler.state = fp->state; + compiler.is_r500 = (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) ? GL_TRUE : GL_FALSE; + compiler.OutputDepth = FRAG_RESULT_DEPTH; + compiler.OutputColor = FRAG_RESULT_COLOR; + compiler.AllocateHwInputs = &allocate_hw_inputs; - if (RADEON_DEBUG & DEBUG_PIXEL) { + if (compiler.Base.Debug) { fflush(stdout); _mesa_printf("Fragment Program: Initial program:\n"); - _mesa_print_program(compiler.program); + _mesa_print_program(&cont->Base.Base); fflush(stdout); } - insert_WPOS_trailer(&compiler); - - rewriteFog(&compiler); - - rewrite_depth_out(compiler.program); - - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - struct radeon_program_transformation transformations[] = { - { &r500_transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformDeriv, 0 }, - { &radeonTransformTrigScale, 0 } - }; - radeonLocalTransform(ctx, compiler.program, 4, transformations); - } else { - struct radeon_program_transformation transformations[] = { - { &r300_transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformTrigSimple, 0 } - }; - radeonLocalTransform(ctx, compiler.program, 3, transformations); - } + rc_mesa_to_rc_program(&compiler.Base, &cont->Base.Base); - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Fragment Program: After native rewrite:\n"); - _mesa_print_program(compiler.program); - fflush(stdout); - } - - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &r500FPIsNativeSwizzle, - .BuildSwizzle = &r500FPBuildSwizzle - }; - radeonNqssaDce(ctx, compiler.program, &nqssadce); - } else { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &r300FPIsNativeSwizzle, - .BuildSwizzle = &r300FPBuildSwizzle - }; - radeonNqssaDce(ctx, compiler.program, &nqssadce); - } + insert_WPOS_trailer(&compiler, fp); - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: after NqSSA-DCE:\n"); - _mesa_print_program(compiler.program); - fflush(stdout); - } + rewriteFog(&compiler, fp); - if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler)) - fp->error = GL_TRUE; + r3xx_compile_fragment_program(&compiler); + fp->error = compiler.Base.Error; - fp->translated = GL_TRUE; + fp->InputsRead = compiler.Base.Program.InputsRead; - if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL)) - r300->vtbl.FragmentProgramDump(&fp->code); + rc_destroy(&compiler.Base); } -struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx) +struct r300_fragment_program *r300SelectAndTranslateFragmentShader(GLcontext *ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_fragment_program_cont *fp_list; @@ -389,11 +253,11 @@ struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx) fp = _mesa_calloc(sizeof(struct r300_fragment_program)); fp->state = state; - fp->translated = GL_FALSE; - fp->Base = _mesa_clone_program(ctx, &ctx->FragmentProgram._Current->Base); fp->next = fp_list->progs; fp_list->progs = fp; + translate_fragment_program(ctx, fp_list, fp); + return r300->selected_fp = fp; } |