/** * Scan/rewrite program to remove reads of custom (output) registers. * The passed type has to be PROGRAM_OUTPUT. * 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[VARYING_SLOT_MAX]; GLuint numVaryingReads = 0; GLboolean usedTemps[MAX_PROGRAM_TEMPS]; GLuint firstTemp = 0; _mesa_find_used_registers(prog, PROGRAM_TEMPORARY, usedTemps, MAX_PROGRAM_TEMPS); assert(type == PROGRAM_OUTPUT); for (i = 0; i < VARYING_SLOT_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(usedTemps, MAX_PROGRAM_TEMPS, firstTemp); firstTemp = outputMap[var] + 1; } 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; 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 < VARYING_SLOT_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++; } } } }
/** * Combine two programs into one. Fix instructions so the outputs of * the first program go to the inputs of the second program. */ struct gl_program * _mesa_combine_programs(struct gl_context *ctx, const struct gl_program *progA, const struct gl_program *progB) { struct prog_instruction *newInst; struct gl_program *newProg; const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */ const GLuint lenB = progB->NumInstructions; const GLuint numParamsA = _mesa_num_parameters(progA->Parameters); const GLuint newLength = lenA + lenB; GLboolean usedTemps[MAX_PROGRAM_TEMPS]; GLuint firstTemp = 0; GLbitfield inputsB; GLuint i; ASSERT(progA->Target == progB->Target); newInst = _mesa_alloc_instructions(newLength); if (!newInst) return GL_FALSE; _mesa_copy_instructions(newInst, progA->Instructions, lenA); _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB); /* adjust branch / instruction addresses for B's instructions */ for (i = 0; i < lenB; i++) { newInst[lenA + i].BranchTarget += lenA; } newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0); newProg->Instructions = newInst; newProg->NumInstructions = newLength; /* find used temp regs (we may need new temps below) */ _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY, usedTemps, MAX_PROGRAM_TEMPS); if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) { struct gl_fragment_program *fprogA, *fprogB, *newFprog; GLbitfield progB_inputsRead = progB->InputsRead; GLint progB_colorFile, progB_colorIndex; fprogA = (struct gl_fragment_program *) progA; fprogB = (struct gl_fragment_program *) progB; newFprog = (struct gl_fragment_program *) newProg; newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill; /* We'll do a search and replace for instances * of progB_colorFile/progB_colorIndex below... */ progB_colorFile = PROGRAM_INPUT; progB_colorIndex = FRAG_ATTRIB_COL0; /* * The fragment program may get color from a state var rather than * a fragment input (vertex output) if it's constant. * See the texenvprogram.c code. * So, search the program's parameter list now to see if the program * gets color from a state var instead of a conventional fragment * input register. */ for (i = 0; i < progB->Parameters->NumParameters; i++) { struct gl_program_parameter *p = &progB->Parameters->Parameters[i]; if (p->Type == PROGRAM_STATE_VAR && p->StateIndexes[0] == STATE_INTERNAL && p->StateIndexes[1] == STATE_CURRENT_ATTRIB && (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) { progB_inputsRead |= FRAG_BIT_COL0; progB_colorFile = PROGRAM_STATE_VAR; progB_colorIndex = i; break; } } /* Connect color outputs of fprogA to color inputs of fprogB, via a * new temporary register. */ if ((progA->OutputsWritten & (1 << FRAG_RESULT_COLOR)) && (progB_inputsRead & FRAG_BIT_COL0)) { GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS, firstTemp); if (tempReg < 0) { _mesa_problem(ctx, "No free temp regs found in " "_mesa_combine_programs(), using 31"); tempReg = 31; } firstTemp = tempReg + 1; /* replace writes to result.color[0] with tempReg */ replace_registers(newInst, lenA, PROGRAM_OUTPUT, FRAG_RESULT_COLOR, PROGRAM_TEMPORARY, tempReg); /* replace reads from the input color with tempReg */ replace_registers(newInst + lenA, lenB, progB_colorFile, progB_colorIndex, /* search for */ PROGRAM_TEMPORARY, tempReg /* replace with */ ); } /* compute combined program's InputsRead */ inputsB = progB_inputsRead; if (progA->OutputsWritten & (1 << FRAG_RESULT_COLOR)) { inputsB &= ~(1 << FRAG_ATTRIB_COL0); } newProg->InputsRead = progA->InputsRead | inputsB; newProg->OutputsWritten = progB->OutputsWritten; newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed; } else { /* vertex program */ assert(0); /* XXX todo */ } /* * Merge parameters (uniforms, constants, etc) */ newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters, progB->Parameters); adjust_param_indexes(newInst + lenA, lenB, numParamsA); return newProg; }