X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fshader%2Fprogramopt.c;h=ac5fe0f691fadb460c03a4c84e32a062abca3652;hb=8f9ee069250fe65bc19c5859963ee85db96e24e1;hp=18da39c2d31f29db020ca99296af30d0c9ba9c5c;hpb=776bc9cf55b116e17dddde4d097985b51879c83f;p=mesa.git diff --git a/src/mesa/shader/programopt.c b/src/mesa/shader/programopt.c index 18da39c2d31..ac5fe0f691f 100644 --- a/src/mesa/shader/programopt.c +++ b/src/mesa/shader/programopt.c @@ -31,10 +31,11 @@ */ -#include "glheader.h" -#include "context.h" +#include "main/glheader.h" +#include "main/context.h" #include "prog_parameter.h" #include "prog_statevars.h" +#include "program.h" #include "programopt.h" #include "prog_instruction.h" @@ -44,8 +45,8 @@ * into a vertex program. * May be used to implement the position_invariant option. */ -void -_mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) +static void +_mesa_insert_mvp_dp4_code(GLcontext *ctx, struct gl_vertex_program *vprog) { struct prog_instruction *newInst; const GLuint origLen = vprog->Base.NumInstructions; @@ -56,7 +57,7 @@ _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) * Setup state references for the modelview/projection matrix. * XXX we should check if these state vars are already declared. */ - static const GLint mvpState[4][STATE_LENGTH] = { + static const gl_state_index mvpState[4][STATE_LENGTH] = { { STATE_MVP_MATRIX, 0, 0, 0, 0 }, /* state.matrix.mvp.row[0] */ { STATE_MVP_MATRIX, 0, 1, 1, 0 }, /* state.matrix.mvp.row[1] */ { STATE_MVP_MATRIX, 0, 2, 2, 0 }, /* state.matrix.mvp.row[2] */ @@ -99,11 +100,10 @@ _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) } /* Append original instructions after new instructions */ - _mesa_memcpy(newInst + 4, vprog->Base.Instructions, - origLen * sizeof(struct prog_instruction)); + _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); /* free old instructions */ - _mesa_free(vprog->Base.Instructions); + _mesa_free_instructions(vprog->Base.Instructions, origLen); /* install new instructions */ vprog->Base.Instructions = newInst; @@ -113,6 +113,121 @@ _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) } +static void +_mesa_insert_mvp_mad_code(GLcontext *ctx, struct gl_vertex_program *vprog) +{ + struct prog_instruction *newInst; + const GLuint origLen = vprog->Base.NumInstructions; + const GLuint newLen = origLen + 4; + GLuint hposTemp; + GLuint i; + + /* + * Setup state references for the modelview/projection matrix. + * XXX we should check if these state vars are already declared. + */ + static const gl_state_index mvpState[4][STATE_LENGTH] = { + { STATE_MVP_MATRIX, 0, 0, 0, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 1, 1, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 2, 2, STATE_MATRIX_TRANSPOSE }, + { STATE_MVP_MATRIX, 0, 3, 3, STATE_MATRIX_TRANSPOSE }, + }; + GLint mvpRef[4]; + + for (i = 0; i < 4; i++) { + mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters, + mvpState[i]); + } + + /* Alloc storage for new instructions */ + newInst = _mesa_alloc_instructions(newLen); + if (!newInst) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, + "glProgramString(inserting position_invariant code)"); + return; + } + + /* TEMP hposTemp; */ + hposTemp = vprog->Base.NumTemporaries++; + + /* + * Generated instructions: + * emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); + * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); + * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); + * emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); + */ + _mesa_init_instructions(newInst, 4); + + newInst[0].Opcode = OPCODE_MUL; + newInst[0].DstReg.File = PROGRAM_TEMPORARY; + newInst[0].DstReg.Index = hposTemp; + newInst[0].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[0].SrcReg[0].File = PROGRAM_INPUT; + newInst[0].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[0].SrcReg[0].Swizzle = SWIZZLE_XXXX; + newInst[0].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[0].SrcReg[1].Index = mvpRef[0]; + newInst[0].SrcReg[1].Swizzle = SWIZZLE_NOOP; + + for (i = 1; i <= 2; i++) { + newInst[i].Opcode = OPCODE_MAD; + newInst[i].DstReg.File = PROGRAM_TEMPORARY; + newInst[i].DstReg.Index = hposTemp; + newInst[i].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[i].SrcReg[0].File = PROGRAM_INPUT; + newInst[i].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(i,i,i,i); + newInst[i].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[i].SrcReg[1].Index = mvpRef[i]; + newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP; + newInst[i].SrcReg[2].File = PROGRAM_TEMPORARY; + newInst[i].SrcReg[2].Index = hposTemp; + newInst[1].SrcReg[2].Swizzle = SWIZZLE_NOOP; + } + + newInst[3].Opcode = OPCODE_MAD; + newInst[3].DstReg.File = PROGRAM_OUTPUT; + newInst[3].DstReg.Index = VERT_RESULT_HPOS; + newInst[3].DstReg.WriteMask = WRITEMASK_XYZW; + newInst[3].SrcReg[0].File = PROGRAM_INPUT; + newInst[3].SrcReg[0].Index = VERT_ATTRIB_POS; + newInst[3].SrcReg[0].Swizzle = SWIZZLE_WWWW; + newInst[3].SrcReg[1].File = PROGRAM_STATE_VAR; + newInst[3].SrcReg[1].Index = mvpRef[3]; + newInst[3].SrcReg[1].Swizzle = SWIZZLE_NOOP; + newInst[3].SrcReg[2].File = PROGRAM_TEMPORARY; + newInst[3].SrcReg[2].Index = hposTemp; + newInst[3].SrcReg[2].Swizzle = SWIZZLE_NOOP; + + + /* Append original instructions after new instructions */ + _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen); + + /* free old instructions */ + _mesa_free_instructions(vprog->Base.Instructions, origLen); + + /* install new instructions */ + vprog->Base.Instructions = newInst; + vprog->Base.NumInstructions = newLen; + vprog->Base.InputsRead |= VERT_BIT_POS; + vprog->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS); +} + + +void +_mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) +{ + if (ctx->mvp_with_dp4) + _mesa_insert_mvp_dp4_code( ctx, vprog ); + else + _mesa_insert_mvp_mad_code( ctx, vprog ); +} + + + + + /** * Append extra instructions onto the given fragment program to implement @@ -125,9 +240,9 @@ _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog) void _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) { - static const GLint fogPStateOpt[STATE_LENGTH] + static const gl_state_index fogPStateOpt[STATE_LENGTH] = { STATE_INTERNAL, STATE_FOG_PARAMS_OPTIMIZED, 0, 0, 0 }; - static const GLint fogColorState[STATE_LENGTH] + static const gl_state_index fogColorState[STATE_LENGTH] = { STATE_FOG_COLOR, 0, 0, 0, 0}; struct prog_instruction *newInst, *inst; const GLuint origLen = fprog->Base.NumInstructions; @@ -151,8 +266,7 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) } /* Copy orig instructions into new instruction buffer */ - _mesa_memcpy(newInst, fprog->Base.Instructions, - origLen * sizeof(struct prog_instruction)); + _mesa_copy_instructions(newInst, fprog->Base.Instructions, origLen); /* PARAM fogParamsRefOpt = internal optimized fog params; */ fogPRefOpt @@ -172,7 +286,7 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) if (inst->Opcode == OPCODE_END) break; if (inst->DstReg.File == PROGRAM_OUTPUT && - inst->DstReg.Index == FRAG_RESULT_COLR) { + inst->DstReg.Index == FRAG_RESULT_COLOR) { /* change the instruction to write to colorTemp w/ clamping */ inst->DstReg.File = PROGRAM_TEMPORARY; inst->DstReg.Index = colorTemp; @@ -194,13 +308,14 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) inst->DstReg.WriteMask = WRITEMASK_X; inst->SrcReg[0].File = PROGRAM_INPUT; inst->SrcReg[0].Index = FRAG_ATTRIB_FOGC; - inst->SrcReg[0].Swizzle = SWIZZLE_X; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; inst->SrcReg[1].File = PROGRAM_STATE_VAR; inst->SrcReg[1].Index = fogPRefOpt; - inst->SrcReg[1].Swizzle = SWIZZLE_X; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; inst->SrcReg[2].File = PROGRAM_STATE_VAR; inst->SrcReg[2].Index = fogPRefOpt; - inst->SrcReg[2].Swizzle = SWIZZLE_Y; + inst->SrcReg[2].Swizzle = SWIZZLE_YYYY; + inst->SaturateMode = SATURATE_ZERO_ONE; inst++; } else { @@ -215,10 +330,10 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) inst->SrcReg[0].File = PROGRAM_STATE_VAR; inst->SrcReg[0].Index = fogPRefOpt; inst->SrcReg[0].Swizzle - = (fprog->FogOption == GL_EXP) ? SWIZZLE_Z : SWIZZLE_W; + = (fprog->FogOption == GL_EXP) ? SWIZZLE_ZZZZ : SWIZZLE_WWWW; inst->SrcReg[1].File = PROGRAM_INPUT; inst->SrcReg[1].Index = FRAG_ATTRIB_FOGC; - inst->SrcReg[1].Swizzle = SWIZZLE_X; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; inst++; if (fprog->FogOption == GL_EXP2) { /* MUL fogFactorTemp.x, fogFactorTemp.x, fogFactorTemp.x; */ @@ -228,10 +343,10 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) inst->DstReg.WriteMask = WRITEMASK_X; inst->SrcReg[0].File = PROGRAM_TEMPORARY; inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].Swizzle = SWIZZLE_X; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; inst->SrcReg[1].File = PROGRAM_TEMPORARY; inst->SrcReg[1].Index = fogFactorTemp; - inst->SrcReg[1].Swizzle = SWIZZLE_X; + inst->SrcReg[1].Swizzle = SWIZZLE_XXXX; inst++; } /* EX2_SAT fogFactorTemp.x, -fogFactorTemp.x; */ @@ -241,20 +356,19 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) inst->DstReg.WriteMask = WRITEMASK_X; inst->SrcReg[0].File = PROGRAM_TEMPORARY; inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].NegateBase = GL_TRUE; - inst->SrcReg[0].Swizzle = SWIZZLE_X; + inst->SrcReg[0].Negate = NEGATE_XYZW; + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; inst->SaturateMode = SATURATE_ZERO_ONE; inst++; } /* LRP result.color.xyz, fogFactorTemp.xxxx, colorTemp, fogColorRef; */ inst->Opcode = OPCODE_LRP; inst->DstReg.File = PROGRAM_OUTPUT; - inst->DstReg.Index = FRAG_RESULT_COLR; + inst->DstReg.Index = FRAG_RESULT_COLOR; inst->DstReg.WriteMask = WRITEMASK_XYZ; inst->SrcReg[0].File = PROGRAM_TEMPORARY; inst->SrcReg[0].Index = fogFactorTemp; - inst->SrcReg[0].Swizzle - = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X); + inst->SrcReg[0].Swizzle = SWIZZLE_XXXX; inst->SrcReg[1].File = PROGRAM_TEMPORARY; inst->SrcReg[1].Index = colorTemp; inst->SrcReg[1].Swizzle = SWIZZLE_NOOP; @@ -265,7 +379,7 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) /* MOV result.color.w, colorTemp.x; # copy alpha */ inst->Opcode = OPCODE_MOV; inst->DstReg.File = PROGRAM_OUTPUT; - inst->DstReg.Index = FRAG_RESULT_COLR; + inst->DstReg.Index = FRAG_RESULT_COLOR; inst->DstReg.WriteMask = WRITEMASK_W; inst->SrcReg[0].File = PROGRAM_TEMPORARY; inst->SrcReg[0].Index = colorTemp; @@ -276,12 +390,13 @@ _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog) inst++; /* free old instructions */ - _mesa_free(fprog->Base.Instructions); + _mesa_free_instructions(fprog->Base.Instructions, origLen); /* install new instructions */ fprog->Base.Instructions = newInst; fprog->Base.NumInstructions = inst - newInst; fprog->Base.InputsRead |= FRAG_BIT_FOGC; + fprog->UsesFogFragCoord = GL_TRUE; /* XXX do this? fprog->FogOption = GL_NONE; */ } @@ -366,3 +481,96 @@ _mesa_count_texture_instructions(struct gl_program *prog) } } + +/** + * Scan/rewrite program to remove reads of custom (output) registers. + * The passed type has to be either PROGRAM_OUTPUT or PROGRAM_VARYING + * (for vertex shaders). + * In GLSL shaders, varying vars can be read and written. + * On some hardware, trying to read an output register causes trouble. + * So, rewrite the program to use a temporary register in this case. + */ +void +_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type) +{ + GLuint i; + GLint outputMap[VERT_RESULT_MAX]; + GLuint numVaryingReads = 0; + + assert(type == PROGRAM_VARYING || type == PROGRAM_OUTPUT); + assert(prog->Target == GL_VERTEX_PROGRAM_ARB || type != PROGRAM_VARYING); + + for (i = 0; i < VERT_RESULT_MAX; i++) + outputMap[i] = -1; + + /* look for instructions which read from varying vars */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->SrcReg[j].File == type) { + /* replace the read with a temp reg */ + const GLuint var = inst->SrcReg[j].Index; + if (outputMap[var] == -1) { + numVaryingReads++; + outputMap[var] = _mesa_find_free_register(prog, + PROGRAM_TEMPORARY); + } + inst->SrcReg[j].File = PROGRAM_TEMPORARY; + inst->SrcReg[j].Index = outputMap[var]; + } + } + } + + if (numVaryingReads == 0) + return; /* nothing to be done */ + + /* look for instructions which write to the varying vars identified above */ + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode); + GLuint j; + for (j = 0; j < numSrc; j++) { + if (inst->DstReg.File == type && + outputMap[inst->DstReg.Index] >= 0) { + /* change inst to write to the temp reg, instead of the varying */ + inst->DstReg.File = PROGRAM_TEMPORARY; + inst->DstReg.Index = outputMap[inst->DstReg.Index]; + } + } + } + + /* insert new instructions to copy the temp vars to the varying vars */ + { + struct prog_instruction *inst; + GLint endPos, var; + + /* Look for END instruction and insert the new varying writes */ + endPos = -1; + for (i = 0; i < prog->NumInstructions; i++) { + struct prog_instruction *inst = prog->Instructions + i; + if (inst->Opcode == OPCODE_END) { + endPos = i; + _mesa_insert_instructions(prog, i, numVaryingReads); + break; + } + } + + assert(endPos >= 0); + + /* insert new MOV instructions here */ + inst = prog->Instructions + endPos; + for (var = 0; var < VERT_RESULT_MAX; var++) { + if (outputMap[var] >= 0) { + /* MOV VAR[var], TEMP[tmp]; */ + inst->Opcode = OPCODE_MOV; + inst->DstReg.File = type; + inst->DstReg.Index = var; + inst->SrcReg[0].File = PROGRAM_TEMPORARY; + inst->SrcReg[0].Index = outputMap[var]; + inst++; + } + } + } +}