1 // SwiftShader Software Renderer
\r
3 // Copyright(c) 2005-2013 TransGaming Inc.
\r
5 // All rights reserved. No part of this software may be copied, distributed, transmitted,
\r
6 // transcribed, stored in a retrieval system, translated into any human or computer
\r
7 // language by any means, or disclosed to third parties without the explicit written
\r
8 // agreement of TransGaming Inc. Without such an agreement, no rights or licenses, express
\r
9 // or implied, including but not limited to any patent rights, are granted to you.
\r
12 #include "compiler/OutputASM.h"
\r
14 #include "common/debug.h"
\r
15 #include "compiler/InfoSink.h"
\r
17 #include "libGLESv2/Shader.h"
\r
20 #include <GLES2/gl2.h>
\r
21 #include <GLES2/gl2ext.h>
\r
25 // Integer to TString conversion
\r
29 sprintf(buffer, "%d", i);
\r
33 class Temporary : public TIntermSymbol
\r
36 Temporary(OutputASM *assembler) : TIntermSymbol(0, "tmp", TType(EbtFloat, EbpHigh, EvqTemporary, 4, false, false)), assembler(assembler)
\r
42 assembler->freeTemporary(this);
\r
46 OutputASM *const assembler;
\r
49 class Constant : public TIntermConstantUnion
\r
52 Constant(float x, float y, float z, float w) : TIntermConstantUnion(constants, TType(EbtFloat, EbpHigh, EvqConst, 4, false, false))
\r
54 constants[0].setFConst(x);
\r
55 constants[1].setFConst(y);
\r
56 constants[2].setFConst(z);
\r
57 constants[3].setFConst(w);
\r
60 Constant(bool b) : TIntermConstantUnion(constants, TType(EbtBool, EbpHigh, EvqConst, 1, false, false))
\r
62 constants[0].setBConst(b);
\r
65 Constant(int i) : TIntermConstantUnion(constants, TType(EbtInt, EbpHigh, EvqConst, 1, false, false))
\r
67 constants[0].setIConst(i);
\r
75 ConstantUnion constants[4];
\r
78 Uniform::Uniform(GLenum type, GLenum precision, const std::string &name, int arraySize, int registerIndex)
\r
81 this->precision = precision;
\r
83 this->arraySize = arraySize;
\r
84 this->registerIndex = registerIndex;
\r
87 Attribute::Attribute()
\r
94 Attribute::Attribute(GLenum type, const std::string &name, int arraySize, int registerIndex)
\r
98 this->arraySize = arraySize;
\r
99 this->registerIndex = registerIndex;
\r
102 OutputASM::OutputASM(TParseContext &context, es2::Shader *shaderObject) : TIntermTraverser(true, true, true), mContext(context), shaderObject(shaderObject)
\r
110 shader = shaderObject->getShader();
\r
111 pixelShader = shaderObject->getPixelShader();
\r
112 vertexShader = shaderObject->getVertexShader();
\r
115 functionArray.push_back(Function(0, "main(", 0, 0));
\r
116 currentFunction = 0;
\r
117 outputQualifier = EvqOutput; // Set outputQualifier to any value other than EvqFragColor or EvqFragData
\r
120 OutputASM::~OutputASM()
\r
124 void OutputASM::output()
\r
128 emitShader(GLOBAL);
\r
130 if(functionArray.size() > 1) // Only call main() when there are other functions
\r
132 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
\r
133 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
\r
134 callMain->dst.index = 0; // main()
\r
136 emit(sw::Shader::OPCODE_RET);
\r
139 emitShader(FUNCTION);
\r
143 void OutputASM::emitShader(Scope scope)
\r
146 currentScope = GLOBAL;
\r
147 mContext.treeRoot->traverse(this);
\r
150 void OutputASM::freeTemporary(Temporary *temporary)
\r
152 free(temporaries, temporary);
\r
155 void OutputASM::visitSymbol(TIntermSymbol *symbol)
\r
157 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut)
\r
159 // Vertex varyings don't have to be actively used to successfully link
\r
160 // against pixel shaders that use them. So make sure they're declared.
\r
161 declareVarying(symbol, -1);
\r
165 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
\r
167 if(currentScope != emitScope)
\r
172 TIntermTyped *result = node;
\r
173 TIntermTyped *left = node->getLeft();
\r
174 TIntermTyped *right = node->getRight();
\r
175 const TType &leftType = left->getType();
\r
176 const TType &rightType = right->getType();
\r
177 const TType &resultType = node->getType();
\r
179 switch(node->getOp())
\r
182 if(visit == PostVisit)
\r
184 assignLvalue(left, right);
\r
185 copy(result, right);
\r
188 case EOpInitialize:
\r
189 if(visit == PostVisit)
\r
194 case EOpMatrixTimesScalarAssign:
\r
195 if(visit == PostVisit)
\r
197 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
199 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
200 mul->dst.index += i;
\r
201 argument(mul->src[0], left, i);
\r
204 assignLvalue(left, result);
\r
207 case EOpVectorTimesMatrixAssign:
\r
208 if(visit == PostVisit)
\r
210 int size = leftType.getNominalSize();
\r
212 for(int i = 0; i < size; i++)
\r
214 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
215 dot->dst.mask = 1 << i;
\r
216 argument(dot->src[1], right, i);
\r
219 assignLvalue(left, result);
\r
222 case EOpMatrixTimesMatrixAssign:
\r
223 if(visit == PostVisit)
\r
225 int dim = leftType.getNominalSize();
\r
227 for(int i = 0; i < dim; i++)
\r
229 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
230 mul->dst.index += i;
\r
231 argument(mul->src[1], right, i);
\r
232 mul->src[1].swizzle = 0x00;
\r
234 for(int j = 1; j < dim; j++)
\r
236 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
237 mad->dst.index += i;
\r
238 argument(mad->src[0], left, j);
\r
239 argument(mad->src[1], right, i);
\r
240 mad->src[1].swizzle = j * 0x55;
\r
241 argument(mad->src[2], result, i);
\r
245 assignLvalue(left, result);
\r
248 case EOpIndexDirect:
\r
249 if(visit == PostVisit)
\r
251 int index = right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
253 if(result->isMatrix() || result->isStruct())
\r
255 ASSERT(left->isArray());
\r
256 copy(result, left, index * left->elementRegisterCount());
\r
258 else if(result->isRegister())
\r
260 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
262 if(left->isRegister())
\r
264 mov->src[0].swizzle = index;
\r
266 else if(left->isArray())
\r
268 argument(mov->src[0], left, index * left->elementRegisterCount());
\r
270 else if(left->isMatrix())
\r
272 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
\r
273 argument(mov->src[0], left, index);
\r
275 else UNREACHABLE();
\r
277 else UNREACHABLE();
\r
280 case EOpIndexIndirect:
\r
281 if(visit == PostVisit)
\r
283 if(left->isArray() || left->isMatrix())
\r
285 for(int index = 0; index < result->totalRegisterCount(); index++)
\r
287 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
288 mov->dst.index += index;
\r
289 mov->dst.mask = writeMask(result, index);
\r
290 argument(mov->src[0], left, index);
\r
292 if(left->totalRegisterCount() > 1)
\r
294 sw::Shader::SourceParameter relativeRegister;
\r
295 argument(relativeRegister, right);
\r
297 mov->src[0].rel.type = relativeRegister.type;
\r
298 mov->src[0].rel.index = relativeRegister.index;
\r
299 mov->src[0].rel.scale = result->totalRegisterCount();
\r
300 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
304 else if(left->isRegister())
\r
306 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
\r
308 else UNREACHABLE();
\r
311 case EOpIndexDirectStruct:
\r
312 if(visit == PostVisit)
\r
314 ASSERT(leftType.isStruct());
\r
316 const TTypeList *structure = leftType.getStruct();
\r
317 const TString &fieldName = rightType.getFieldName();
\r
318 int fieldOffset = 0;
\r
320 for(size_t i = 0; i < structure->size(); i++)
\r
322 const TType &fieldType = *(*structure)[i].type;
\r
324 if(fieldType.getFieldName() == fieldName)
\r
329 fieldOffset += fieldType.totalRegisterCount();
\r
332 copy(result, left, fieldOffset);
\r
335 case EOpVectorSwizzle:
\r
336 if(visit == PostVisit)
\r
339 TIntermAggregate *components = right->getAsAggregate();
\r
343 TIntermSequence &sequence = components->getSequence();
\r
346 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
348 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
352 int i = element->getUnionArrayPointer()[0].getIConst();
\r
353 swizzle |= i << (component * 2);
\r
356 else UNREACHABLE();
\r
359 else UNREACHABLE();
\r
361 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
362 mov->src[0].swizzle = swizzle;
\r
365 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
366 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
367 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
368 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
369 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
370 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
371 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
372 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
374 if(visit == PostVisit)
\r
376 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
378 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
380 Temporary equal(this);
\r
381 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
382 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
387 if(visit == PostVisit)
\r
389 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
391 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
393 Temporary notEqual(this);
\r
394 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
395 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
399 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
400 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
401 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
402 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
403 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
404 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
405 case EOpMatrixTimesScalar:
\r
406 if(visit == PostVisit)
\r
408 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
410 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
411 mul->dst.index += i;
\r
412 argument(mul->src[0], left, i);
\r
416 case EOpVectorTimesMatrix:
\r
417 if(visit == PostVisit)
\r
419 int size = leftType.getNominalSize();
\r
421 for(int i = 0; i < size; i++)
\r
423 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
424 dot->dst.mask = 1 << i;
\r
425 argument(dot->src[1], right, i);
\r
429 case EOpMatrixTimesVector:
\r
430 if(visit == PostVisit)
\r
432 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
433 mul->src[1].swizzle = 0x00;
\r
435 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
437 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
438 argument(mad->src[0], left, i);
\r
439 mad->src[1].swizzle = i * 0x55;
\r
443 case EOpMatrixTimesMatrix:
\r
444 if(visit == PostVisit)
\r
446 int dim = leftType.getNominalSize();
\r
448 for(int i = 0; i < dim; i++)
\r
450 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
451 mul->dst.index += i;
\r
452 argument(mul->src[1], right, i);
\r
453 mul->src[1].swizzle = 0x00;
\r
455 for(int j = 1; j < dim; j++)
\r
457 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
458 mad->dst.index += i;
\r
459 argument(mad->src[0], left, j);
\r
460 argument(mad->src[1], right, i);
\r
461 mad->src[1].swizzle = j * 0x55;
\r
462 argument(mad->src[2], result, i);
\r
468 if(trivial(right, 6))
\r
470 if(visit == PostVisit)
\r
472 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
475 else // Short-circuit evaluation
\r
477 if(visit == InVisit)
\r
479 emit(sw::Shader::OPCODE_MOV, result, left);
\r
480 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
481 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
483 else if(visit == PostVisit)
\r
485 emit(sw::Shader::OPCODE_MOV, result, right);
\r
486 emit(sw::Shader::OPCODE_ENDIF);
\r
490 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
491 case EOpLogicalAnd:
\r
492 if(trivial(right, 6))
\r
494 if(visit == PostVisit)
\r
496 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
499 else // Short-circuit evaluation
\r
501 if(visit == InVisit)
\r
503 emit(sw::Shader::OPCODE_MOV, result, left);
\r
504 emit(sw::Shader::OPCODE_IF, 0, result);
\r
506 else if(visit == PostVisit)
\r
508 emit(sw::Shader::OPCODE_MOV, result, right);
\r
509 emit(sw::Shader::OPCODE_ENDIF);
\r
513 default: UNREACHABLE();
\r
519 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
521 if(currentScope != emitScope)
\r
526 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
527 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
528 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
530 TIntermTyped *result = node;
\r
531 TIntermTyped *arg = node->getOperand();
\r
533 switch(node->getOp())
\r
536 if(visit == PostVisit)
\r
538 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
540 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
541 neg->dst.index += index;
\r
542 argument(neg->src[0], arg, index);
\r
543 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
547 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
548 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
549 case EOpPostIncrement:
\r
550 if(visit == PostVisit)
\r
554 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
556 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
557 add->dst.index += index;
\r
558 argument(add->src[0], arg, index);
\r
561 assignLvalue(arg, arg);
\r
564 case EOpPostDecrement:
\r
565 if(visit == PostVisit)
\r
569 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
571 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
572 sub->dst.index += index;
\r
573 argument(sub->src[0], arg, index);
\r
576 assignLvalue(arg, arg);
\r
579 case EOpPreIncrement:
\r
580 if(visit == PostVisit)
\r
582 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
584 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
585 add->dst.index += index;
\r
586 argument(add->src[0], arg, index);
\r
589 assignLvalue(arg, result);
\r
592 case EOpPreDecrement:
\r
593 if(visit == PostVisit)
\r
595 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
597 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
598 sub->dst.index += index;
\r
599 argument(sub->src[0], arg, index);
\r
602 assignLvalue(arg, result);
\r
605 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
606 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
607 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
608 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
609 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
610 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
611 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
612 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
613 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
614 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
615 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
616 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
617 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
618 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
619 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
620 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
621 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
622 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
623 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
624 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
625 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
626 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
627 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
628 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
629 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
630 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
631 default: UNREACHABLE();
\r
637 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
639 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
644 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
646 TIntermTyped *result = node;
\r
647 const TType &resultType = node->getType();
\r
648 TIntermSequence &arg = node->getSequence();
\r
649 int argumentCount = arg.size();
\r
651 switch(node->getOp())
\r
653 case EOpSequence: break;
\r
654 case EOpDeclaration: break;
\r
655 case EOpPrototype: break;
\r
657 if(visit == PostVisit)
\r
659 copy(result, arg[1]);
\r
663 if(visit == PreVisit)
\r
665 const TString &name = node->getName();
\r
667 if(emitScope == FUNCTION)
\r
669 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
671 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
672 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
674 const Function *function = findFunction(name);
\r
675 ASSERT(function); // Should have been added during global pass
\r
676 label->dst.index = function->label;
\r
677 currentFunction = function->label;
\r
680 else if(emitScope == GLOBAL)
\r
682 if(name != "main(")
\r
684 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
685 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
688 else UNREACHABLE();
\r
690 currentScope = FUNCTION;
\r
692 else if(visit == PostVisit)
\r
694 if(emitScope == FUNCTION)
\r
696 if(functionArray.size() > 1) // No need to return when there's only main()
\r
698 emit(sw::Shader::OPCODE_RET);
\r
702 currentScope = GLOBAL;
\r
705 case EOpFunctionCall:
\r
706 if(visit == PostVisit)
\r
708 if(node->isUserDefined())
\r
710 const TString &name = node->getName();
\r
711 const Function *function = findFunction(name);
\r
715 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
719 TIntermSequence &arguments = *function->arg;
\r
721 for(int i = 0; i < argumentCount; i++)
\r
723 TIntermTyped *in = arguments[i]->getAsTyped();
\r
725 if(in->getQualifier() == EvqIn ||
\r
726 in->getQualifier() == EvqInOut ||
\r
727 in->getQualifier() == EvqConstReadOnly)
\r
733 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
734 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
735 call->dst.index = function->label;
\r
737 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
739 copy(result, function->ret);
\r
742 for(int i = 0; i < argumentCount; i++)
\r
744 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
745 TIntermTyped *out = arg[i]->getAsTyped();
\r
747 if(argument->getQualifier() == EvqOut ||
\r
748 argument->getQualifier() == EvqInOut)
\r
750 copy(out, argument);
\r
756 TString name = TFunction::unmangleName(node->getName());
\r
758 if(name == "texture2D" || name == "textureCube")
\r
760 if(argumentCount == 2)
\r
762 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
764 else if(argumentCount == 3) // bias
\r
766 Temporary uvwb(this);
\r
767 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
768 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
769 bias->dst.mask = 0x8;
\r
771 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
774 else UNREACHABLE();
\r
776 else if(name == "texture2DProj")
\r
778 TIntermTyped *t = arg[1]->getAsTyped();
\r
780 if(argumentCount == 2)
\r
782 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
783 tex->project = true;
\r
785 if(t->getNominalSize() == 3)
\r
787 tex->src[0].swizzle = 0xA4;
\r
789 else ASSERT(t->getNominalSize() == 4);
\r
791 else if(argumentCount == 3) // bias
\r
793 Temporary proj(this);
\r
795 if(t->getNominalSize() == 3)
\r
797 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
798 div->src[1].swizzle = 0xAA;
\r
799 div->dst.mask = 0x3;
\r
801 else if(t->getNominalSize() == 4)
\r
803 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
804 div->src[1].swizzle = 0xFF;
\r
805 div->dst.mask = 0x3;
\r
807 else UNREACHABLE();
\r
809 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
810 bias->dst.mask = 0x8;
\r
812 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
815 else UNREACHABLE();
\r
817 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
819 Temporary uvwb(this);
\r
820 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
821 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
822 lod->dst.mask = 0x8;
\r
824 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
826 else if(name == "texture2DProjLod")
\r
828 TIntermTyped *t = arg[1]->getAsTyped();
\r
829 Temporary proj(this);
\r
831 if(t->getNominalSize() == 3)
\r
833 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
834 div->src[1].swizzle = 0xAA;
\r
835 div->dst.mask = 0x3;
\r
837 else if(t->getNominalSize() == 4)
\r
839 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
840 div->src[1].swizzle = 0xFF;
\r
841 div->dst.mask = 0x3;
\r
843 else UNREACHABLE();
\r
845 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
846 lod->dst.mask = 0x8;
\r
848 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
850 else UNREACHABLE();
\r
854 case EOpParameters:
\r
856 case EOpConstructFloat:
\r
857 case EOpConstructVec2:
\r
858 case EOpConstructVec3:
\r
859 case EOpConstructVec4:
\r
860 case EOpConstructBool:
\r
861 case EOpConstructBVec2:
\r
862 case EOpConstructBVec3:
\r
863 case EOpConstructBVec4:
\r
864 case EOpConstructInt:
\r
865 case EOpConstructIVec2:
\r
866 case EOpConstructIVec3:
\r
867 case EOpConstructIVec4:
\r
868 if(visit == PostVisit)
\r
872 for(int i = 0; i < argumentCount; i++)
\r
874 TIntermTyped *argi = arg[i]->getAsTyped();
\r
875 int size = argi->getNominalSize();
\r
877 if(!argi->isMatrix())
\r
879 Instruction *mov = emitCast(result, argi);
\r
880 mov->dst.mask = (0xF << component) & 0xF;
\r
881 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
889 while(component < resultType.getNominalSize())
\r
891 Instruction *mov = emitCast(result, argi);
\r
892 mov->dst.mask = (0xF << component) & 0xF;
\r
893 mov->src[0].index += column;
\r
894 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
903 case EOpConstructMat2:
\r
904 case EOpConstructMat3:
\r
905 case EOpConstructMat4:
\r
906 if(visit == PostVisit)
\r
908 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
909 const int dim = result->getNominalSize();
\r
911 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
913 for(int i = 0; i < dim; i++)
\r
915 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
916 init->dst.index += i;
\r
917 Instruction *mov = emitCast(result, arg0);
\r
918 mov->dst.index += i;
\r
919 mov->dst.mask = 1 << i;
\r
920 ASSERT(mov->src[0].swizzle == 0x00);
\r
923 else if(arg0->isMatrix())
\r
925 for(int i = 0; i < dim; i++)
\r
927 if(dim > dim2(arg0))
\r
929 // Initialize to identity matrix
\r
930 Constant col((i == 0 ? 1.0f : 0.0f), (i == 1 ? 1.0f : 0.0f), (i == 2 ? 1.0f : 0.0f), (i == 3 ? 1.0f : 0.0f));
\r
931 Instruction *mov = emitCast(result, &col);
\r
932 mov->dst.index += i;
\r
937 Instruction *mov = emitCast(result, arg0);
\r
938 mov->dst.index += i;
\r
939 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
940 argument(mov->src[0], arg0, i);
\r
949 for(int i = 0; i < argumentCount; i++)
\r
951 TIntermTyped *argi = arg[i]->getAsTyped();
\r
952 int size = argi->getNominalSize();
\r
955 while(element < size)
\r
957 Instruction *mov = emitCast(result, argi);
\r
958 mov->dst.index += column;
\r
959 mov->dst.mask = (0xF << row) & 0xF;
\r
960 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
962 int end = row + size - element;
\r
963 column = end >= dim ? column + 1 : column;
\r
964 element = element + dim - row;
\r
965 row = end >= dim ? 0 : end;
\r
971 case EOpConstructStruct:
\r
972 if(visit == PostVisit)
\r
975 for(int i = 0; i < argumentCount; i++)
\r
977 TIntermTyped *argi = arg[i]->getAsTyped();
\r
978 int size = argi->totalRegisterCount();
\r
980 for(int index = 0; index < size; index++)
\r
982 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
983 mov->dst.index += index + offset;
\r
984 mov->dst.mask = writeMask(result, offset + index);
\r
985 argument(mov->src[0], argi, index);
\r
992 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
993 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
994 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
995 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
996 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
997 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
998 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
999 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1000 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1001 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1002 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1004 if(visit == PostVisit)
\r
1006 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1007 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1010 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1011 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1012 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1013 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1014 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1015 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1016 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1017 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1018 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1020 if(visit == PostVisit)
\r
1022 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1024 for(int i = 0; i < dim2(arg[0]); i++)
\r
1026 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1027 mul->dst.index += i;
\r
1028 argument(mul->src[0], arg[0], i);
\r
1029 argument(mul->src[1], arg[1], i);
\r
1033 default: UNREACHABLE();
\r
1039 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1041 if(currentScope != emitScope)
\r
1046 TIntermTyped *condition = node->getCondition();
\r
1047 TIntermNode *trueBlock = node->getTrueBlock();
\r
1048 TIntermNode *falseBlock = node->getFalseBlock();
\r
1049 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1051 condition->traverse(this);
\r
1053 if(node->usesTernaryOperator())
\r
1055 if(constantCondition)
\r
1057 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1061 trueBlock->traverse(this);
\r
1062 copy(node, trueBlock);
\r
1066 falseBlock->traverse(this);
\r
1067 copy(node, falseBlock);
\r
1070 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1072 trueBlock->traverse(this);
\r
1073 falseBlock->traverse(this);
\r
1074 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1078 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1082 trueBlock->traverse(this);
\r
1083 copy(node, trueBlock);
\r
1088 emit(sw::Shader::OPCODE_ELSE);
\r
1089 falseBlock->traverse(this);
\r
1090 copy(node, falseBlock);
\r
1093 emit(sw::Shader::OPCODE_ENDIF);
\r
1096 else // if/else statement
\r
1098 if(constantCondition)
\r
1100 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1106 trueBlock->traverse(this);
\r
1113 falseBlock->traverse(this);
\r
1119 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1123 trueBlock->traverse(this);
\r
1128 emit(sw::Shader::OPCODE_ELSE);
\r
1129 falseBlock->traverse(this);
\r
1132 emit(sw::Shader::OPCODE_ENDIF);
\r
1139 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1141 if(currentScope != emitScope)
\r
1146 unsigned int iterations = loopCount(node);
\r
1148 if(iterations == 0)
\r
1153 bool unroll = (iterations <= 4);
\r
1157 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1158 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1161 TIntermNode *init = node->getInit();
\r
1162 TIntermTyped *condition = node->getCondition();
\r
1163 TIntermTyped *expression = node->getExpression();
\r
1164 TIntermNode *body = node->getBody();
\r
1166 if(node->getType() == ELoopDoWhile)
\r
1168 Temporary iterate(this);
\r
1169 Constant True(true);
\r
1170 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1172 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1176 body->traverse(this);
\r
1179 emit(sw::Shader::OPCODE_TEST);
\r
1181 condition->traverse(this);
\r
1182 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1184 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1190 init->traverse(this);
\r
1195 for(unsigned int i = 0; i < iterations; i++)
\r
1197 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1201 body->traverse(this);
\r
1206 expression->traverse(this);
\r
1212 condition->traverse(this);
\r
1214 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1218 body->traverse(this);
\r
1221 emit(sw::Shader::OPCODE_TEST);
\r
1225 expression->traverse(this);
\r
1228 condition->traverse(this);
\r
1230 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1237 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1239 if(currentScope != emitScope)
\r
1244 switch(node->getFlowOp())
\r
1246 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1247 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1248 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1250 if(visit == PostVisit)
\r
1252 TIntermTyped *value = node->getExpression();
\r
1256 copy(functionArray[currentFunction].ret, value);
\r
1259 emit(sw::Shader::OPCODE_LEAVE);
\r
1262 default: UNREACHABLE();
\r
1268 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1270 // A sampler register's qualifiers can be:
\r
1271 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1272 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1273 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1274 // the sampler has been copied and is no longer a sampler register.
\r
1275 return IsSampler(operand->getBasicType()) && (operand->getQualifier() == EvqUniform || operand->getQualifier() == EvqTemporary);
\r
1278 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1280 if(dst && isSamplerRegister(dst))
\r
1282 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1285 Instruction *instruction = new Instruction(op);
\r
1289 instruction->dst.type = registerType(dst);
\r
1290 instruction->dst.index = registerIndex(dst) + index;
\r
1291 instruction->dst.mask = writeMask(dst);
\r
1292 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1295 argument(instruction->src[0], src0, index);
\r
1296 argument(instruction->src[1], src1, index);
\r
1297 argument(instruction->src[2], src2, index);
\r
1299 shader->append(instruction);
\r
1301 return instruction;
\r
1304 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1306 // Integers are implemented as float
\r
1307 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1309 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1311 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1313 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1315 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1317 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1320 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1323 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1325 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1327 emit(op, dst, src0, src1, src2, index);
\r
1331 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1333 emitBinary(op, result, src0, src1);
\r
1334 assignLvalue(lhs, result);
\r
1337 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1339 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1340 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1342 Instruction *cmp = emit(opcode, dst, left, right);
\r
1343 cmp->control = cmpOp;
\r
1344 argument(cmp->src[0], left, index);
\r
1345 argument(cmp->src[1], right, index);
\r
1348 int componentCount(const TType &type, int registers)
\r
1350 if(registers == 0)
\r
1355 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1357 int index = registers / type.elementRegisterCount();
\r
1358 registers -= index * type.elementRegisterCount();
\r
1359 return index * type.getElementSize() + componentCount(type, registers);
\r
1362 if(type.isStruct())
\r
1364 TTypeList *structure = type.getStruct();
\r
1367 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1369 const TType &fieldType = *field->type;
\r
1371 if(fieldType.totalRegisterCount() <= registers)
\r
1373 registers -= fieldType.totalRegisterCount();
\r
1374 elements += fieldType.getObjectSize();
\r
1376 else // Register within this field
\r
1378 return elements + componentCount(fieldType, registers);
\r
1382 else if(type.isMatrix())
\r
1384 return registers * type.getNominalSize();
\r
1391 int registerSize(const TType &type, int registers)
\r
1393 if(registers == 0)
\r
1395 if(type.isStruct())
\r
1397 return registerSize(*type.getStruct()->begin()->type, 0);
\r
1400 return type.getNominalSize();
\r
1403 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1405 int index = registers / type.elementRegisterCount();
\r
1406 registers -= index * type.elementRegisterCount();
\r
1407 return registerSize(type, registers);
\r
1410 if(type.isStruct())
\r
1412 TTypeList *structure = type.getStruct();
\r
1415 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1417 const TType &fieldType = *field->type;
\r
1419 if(fieldType.totalRegisterCount() <= registers)
\r
1421 registers -= fieldType.totalRegisterCount();
\r
1422 elements += fieldType.getObjectSize();
\r
1424 else // Register within this field
\r
1426 return registerSize(fieldType, registers);
\r
1430 else if(type.isMatrix())
\r
1432 return registerSize(type, 0);
\r
1439 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1443 TIntermTyped *arg = argument->getAsTyped();
\r
1444 const TType &type = arg->getType();
\r
1445 const TTypeList *structure = type.getStruct();
\r
1446 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1448 int size = registerSize(type, index);
\r
1450 parameter.type = registerType(arg);
\r
1452 if(arg->getQualifier() == EvqConst)
\r
1454 int component = componentCount(type, index);
\r
1455 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1457 for(int i = 0; i < 4; i++)
\r
1459 if(size == 1) // Replicate
\r
1461 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1465 parameter.value[i] = constants[component + i].getAsFloat();
\r
1469 parameter.value[i] = 0.0f;
\r
1475 parameter.index = registerIndex(arg) + index;
\r
1477 if(isSamplerRegister(arg))
\r
1479 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1483 TIntermTyped *left = binary->getLeft();
\r
1484 TIntermTyped *right = binary->getRight();
\r
1486 if(binary->getOp() == EOpIndexDirect)
\r
1488 parameter.index += right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1490 else if(binary->getOp() == EOpIndexIndirect)
\r
1492 if(left->getArraySize() > 1)
\r
1494 parameter.rel.type = registerType(binary->getRight());
\r
1495 parameter.rel.index = registerIndex(binary->getRight());
\r
1496 parameter.rel.scale = 1;
\r
1497 parameter.rel.deterministic = true;
\r
1500 else UNREACHABLE();
\r
1505 if(!IsSampler(arg->getBasicType()))
\r
1507 parameter.swizzle = readSwizzle(arg, size);
\r
1512 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1514 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1516 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1517 mov->dst.index += index;
\r
1518 mov->dst.mask = writeMask(dst, index);
\r
1519 argument(mov->src[0], src, offset + index);
\r
1523 int swizzleElement(int swizzle, int index)
\r
1525 return (swizzle >> (index * 2)) & 0x03;
\r
1528 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1530 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1531 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1532 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1533 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1536 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1539 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1540 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize())))))
\r
1542 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1545 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1547 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1549 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1551 Temporary address(this);
\r
1552 lvalue(insert->dst, address, dst);
\r
1554 insert->src[0].type = insert->dst.type;
\r
1555 insert->src[0].index = insert->dst.index;
\r
1556 insert->src[0].rel = insert->dst.rel;
\r
1557 argument(insert->src[1], src);
\r
1558 argument(insert->src[2], binary->getRight());
\r
1560 shader->append(insert);
\r
1564 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1566 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1568 Temporary address(this);
\r
1569 int swizzle = lvalue(mov->dst, address, dst);
\r
1570 mov->dst.index += offset;
\r
1574 mov->dst.mask = writeMask(dst, offset);
\r
1577 argument(mov->src[0], src, offset);
\r
1578 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1580 shader->append(mov);
\r
1585 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1587 TIntermTyped *result = node;
\r
1588 TIntermBinary *binary = node->getAsBinaryNode();
\r
1589 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1593 TIntermTyped *left = binary->getLeft();
\r
1594 TIntermTyped *right = binary->getRight();
\r
1596 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1598 switch(binary->getOp())
\r
1600 case EOpIndexDirect:
\r
1602 int rightIndex = right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1604 if(left->isRegister())
\r
1606 int leftMask = dst.mask;
\r
1609 while((leftMask & dst.mask) == 0)
\r
1611 dst.mask = dst.mask << 1;
\r
1614 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1615 dst.mask = 1 << element;
\r
1619 else if(left->isArray() || left->isMatrix())
\r
1621 dst.index += rightIndex * result->totalRegisterCount();
\r
1624 else UNREACHABLE();
\r
1627 case EOpIndexIndirect:
\r
1629 if(left->isRegister())
\r
1631 // Requires INSERT instruction (handled by calling function)
\r
1633 else if(left->isArray() || left->isMatrix())
\r
1635 int scale = result->totalRegisterCount();
\r
1637 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1639 if(left->totalRegisterCount() > 1)
\r
1641 sw::Shader::SourceParameter relativeRegister;
\r
1642 argument(relativeRegister, right);
\r
1644 dst.rel.index = relativeRegister.index;
\r
1645 dst.rel.type = relativeRegister.type;
\r
1646 dst.rel.scale = scale;
\r
1647 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1650 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1654 Constant oldScale((int)dst.rel.scale);
\r
1655 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1656 mad->src[0].index = dst.rel.index;
\r
1657 mad->src[0].type = dst.rel.type;
\r
1661 Constant oldScale((int)dst.rel.scale);
\r
1662 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1663 mul->src[0].index = dst.rel.index;
\r
1664 mul->src[0].type = dst.rel.type;
\r
1666 Constant newScale(scale);
\r
1667 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1670 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1671 dst.rel.index = registerIndex(&address);
\r
1672 dst.rel.scale = 1;
\r
1674 else // Just add the new index to the address register
\r
1678 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1682 Constant newScale(scale);
\r
1683 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1687 else UNREACHABLE();
\r
1690 case EOpIndexDirectStruct:
\r
1692 const TTypeList *structure = left->getType().getStruct();
\r
1693 const TString &fieldName = right->getType().getFieldName();
\r
1696 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1698 if(field->type->getFieldName() == fieldName)
\r
1700 dst.type = registerType(left);
\r
1701 dst.index += offset;
\r
1702 dst.mask = writeMask(right);
\r
1707 offset += field->type->totalRegisterCount();
\r
1711 case EOpVectorSwizzle:
\r
1713 ASSERT(left->isRegister());
\r
1715 int leftMask = dst.mask;
\r
1718 int rightMask = 0;
\r
1720 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1722 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1724 int index = sequence[i]->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1726 int element = swizzleElement(leftSwizzle, index);
\r
1727 rightMask = rightMask | (1 << element);
\r
1728 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1731 dst.mask = leftMask & rightMask;
\r
1737 UNREACHABLE(); // Not an l-value operator
\r
1743 dst.type = registerType(symbol);
\r
1744 dst.index = registerIndex(symbol);
\r
1745 dst.mask = writeMask(symbol);
\r
1752 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1754 if(isSamplerRegister(operand))
\r
1756 return sw::Shader::PARAMETER_SAMPLER;
\r
1759 const TQualifier qualifier = operand->getQualifier();
\r
1760 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1762 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1763 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1765 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1767 outputQualifier = qualifier;
\r
1772 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1773 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1774 case EvqConst: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1775 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1776 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1777 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1778 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1779 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1780 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1781 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1782 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1783 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1784 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1785 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1786 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1787 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1788 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1789 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1790 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1791 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1792 default: UNREACHABLE();
\r
1795 return sw::Shader::PARAMETER_VOID;
\r
1798 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1800 if(isSamplerRegister(operand))
\r
1802 return samplerRegister(operand);
\r
1805 switch(operand->getQualifier())
\r
1807 case EvqTemporary: return temporaryRegister(operand);
\r
1808 case EvqGlobal: return temporaryRegister(operand);
\r
1809 case EvqConst: UNREACHABLE();
\r
1810 case EvqAttribute: return attributeRegister(operand);
\r
1811 case EvqVaryingIn: return varyingRegister(operand);
\r
1812 case EvqVaryingOut: return varyingRegister(operand);
\r
1813 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1814 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1815 case EvqUniform: return uniformRegister(operand);
\r
1816 case EvqIn: return temporaryRegister(operand);
\r
1817 case EvqOut: return temporaryRegister(operand);
\r
1818 case EvqInOut: return temporaryRegister(operand);
\r
1819 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1820 case EvqPosition: return varyingRegister(operand);
\r
1821 case EvqPointSize: return varyingRegister(operand);
\r
1822 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1823 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1824 case EvqPointCoord: return varyingRegister(operand);
\r
1825 case EvqFragColor: return 0;
\r
1826 case EvqFragData: return 0;
\r
1827 default: UNREACHABLE();
\r
1833 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1835 if(destination->getQualifier() == EvqPointSize)
\r
1837 return 0x2; // Point size stored in the y component
\r
1840 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1843 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1845 if(argument->getQualifier() == EvqPointSize)
\r
1847 return 0x55; // Point size stored in the y component
\r
1850 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1852 return swizzleSize[size];
\r
1855 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1856 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1858 if(!expression->isRegister())
\r
1863 return cost(expression, budget) >= 0;
\r
1866 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1867 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1874 if(expression->getAsSymbolNode())
\r
1878 else if(expression->getAsConstantUnion())
\r
1882 else if(expression->getAsBinaryNode())
\r
1884 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1886 switch(binary->getOp())
\r
1888 case EOpVectorSwizzle:
\r
1889 case EOpIndexDirect:
\r
1890 case EOpIndexDirectStruct:
\r
1891 return cost(binary->getLeft(), budget - 0);
\r
1895 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1900 else if(expression->getAsUnaryNode())
\r
1902 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1904 switch(unary->getOp())
\r
1908 return cost(unary->getOperand(), budget - 1);
\r
1913 else if(expression->getAsSelectionNode())
\r
1915 TIntermSelection *selection = expression->getAsSelectionNode();
\r
1917 if(selection->usesTernaryOperator())
\r
1919 TIntermTyped *condition = selection->getCondition();
\r
1920 TIntermNode *trueBlock = selection->getTrueBlock();
\r
1921 TIntermNode *falseBlock = selection->getFalseBlock();
\r
1922 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1924 if(constantCondition)
\r
1926 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1930 return cost(trueBlock, budget - 0);
\r
1934 return cost(falseBlock, budget - 0);
\r
1939 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
1947 const Function *OutputASM::findFunction(const TString &name)
\r
1949 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
1951 if(functionArray[f].name == name)
\r
1953 return &functionArray[f];
\r
1960 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
1962 return allocate(temporaries, temporary);
\r
1965 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
1967 int var = lookup(varyings, varying);
\r
1971 var = allocate(varyings, varying);
\r
1972 int componentCount = varying->getNominalSize();
\r
1973 int registerCount = varying->totalRegisterCount();
\r
1977 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
1979 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
1983 if(varying->getQualifier() == EvqPointCoord)
\r
1985 ASSERT(varying->isRegister());
\r
1986 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1987 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1988 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1989 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1993 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
1995 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1996 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1997 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1998 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2002 else if(vertexShader)
\r
2004 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2006 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2010 if(varying->getQualifier() == EvqPosition)
\r
2012 ASSERT(varying->isRegister());
\r
2013 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2014 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2015 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2016 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2017 vertexShader->positionRegister = var;
\r
2019 else if(varying->getQualifier() == EvqPointSize)
\r
2021 ASSERT(varying->isRegister());
\r
2022 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2023 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2024 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2025 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2026 vertexShader->pointSizeRegister = var;
\r
2030 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2033 else UNREACHABLE();
\r
2035 declareVarying(varying, var);
\r
2041 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2043 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2045 const TType &type = varying->getType();
\r
2046 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2047 es2::VaryingList &activeVaryings = shaderObject->varyings;
\r
2049 // Check if this varying has been declared before without having a register assigned
\r
2050 for(es2::VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2052 if(v->name == name)
\r
2056 ASSERT(v->reg < 0 || v->reg == reg);
\r
2064 activeVaryings.push_back(es2::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2068 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2070 const TType &type = uniform->getType();
\r
2071 ASSERT(!IsSampler(type.getBasicType()));
\r
2072 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2077 int index = lookup(uniforms, uniform);
\r
2081 index = allocate(uniforms, uniform);
\r
2082 const TString &name = symbol->getSymbol().c_str();
\r
2084 declareUniform(type, name, index);
\r
2093 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2095 ASSERT(!attribute->isArray());
\r
2096 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2098 int index = lookup(attributes, attribute);
\r
2102 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2107 index = allocate(attributes, attribute);
\r
2108 const TType &type = attribute->getType();
\r
2109 int registerCount = attribute->totalRegisterCount();
\r
2111 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2113 for(int i = 0; i < registerCount; i++)
\r
2115 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2119 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2121 const char *name = symbol->getSymbol().c_str();
\r
2122 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2129 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2131 const TType &type = sampler->getType();
\r
2132 ASSERT(IsSampler(type.getBasicType()));
\r
2133 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2134 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2138 int index = lookup(samplers, sampler);
\r
2142 index = allocate(samplers, sampler);
\r
2143 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2144 const char *name = symbol->getSymbol().c_str();
\r
2145 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name, sampler->getArraySize(), index));
\r
2147 for(int i = 0; i < sampler->totalRegisterCount(); i++)
\r
2149 shader->declareSampler(index + i);
\r
2157 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect);
\r
2159 return samplerRegister(binary->getLeft()); // Index added later
\r
2161 else UNREACHABLE();
\r
2166 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2168 for(unsigned int i = 0; i < list.size(); i++)
\r
2170 if(list[i] == variable)
\r
2172 return i; // Pointer match
\r
2176 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2180 for(unsigned int i = 0; i < list.size(); i++)
\r
2184 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2188 if(listSymbol->getId() == varSymbol->getId())
\r
2190 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2191 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2192 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2204 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2206 int index = lookup(list, variable);
\r
2210 unsigned int registerCount = variable->totalRegisterCount();
\r
2212 for(unsigned int i = 0; i < list.size(); i++)
\r
2216 unsigned int j = 1;
\r
2217 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2219 if(list[i + j] != 0)
\r
2225 if(j == registerCount) // Found free slots
\r
2227 for(unsigned int j = 0; j < registerCount; j++)
\r
2229 list[i + j] = variable;
\r
2237 index = list.size();
\r
2239 for(unsigned int i = 0; i < registerCount; i++)
\r
2241 list.push_back(variable);
\r
2248 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2250 int index = lookup(list, variable);
\r
2258 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2260 const TTypeList *structure = type.getStruct();
\r
2261 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2265 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2269 if(type.isArray())
\r
2271 int elementIndex = index;
\r
2273 for(int i = 0; i < type.getArraySize(); i++)
\r
2275 for(size_t j = 0; j < structure->size(); j++)
\r
2277 const TType &fieldType = *(*structure)[j].type;
\r
2278 const TString &fieldName = fieldType.getFieldName();
\r
2280 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2281 declareUniform(fieldType, uniformName, elementIndex);
\r
2282 elementIndex += fieldType.totalRegisterCount();
\r
2288 int fieldIndex = index;
\r
2290 for(size_t i = 0; i < structure->size(); i++)
\r
2292 const TType &fieldType = *(*structure)[i].type;
\r
2293 const TString &fieldName = fieldType.getFieldName();
\r
2295 const TString uniformName = name + "." + fieldName;
\r
2296 declareUniform(fieldType, uniformName, fieldIndex);
\r
2297 fieldIndex += fieldType.totalRegisterCount();
\r
2303 GLenum OutputASM::glVariableType(const TType &type)
\r
2305 if(type.getBasicType() == EbtFloat)
\r
2307 if(type.isScalar())
\r
2311 else if(type.isVector())
\r
2313 switch(type.getNominalSize())
\r
2315 case 2: return GL_FLOAT_VEC2;
\r
2316 case 3: return GL_FLOAT_VEC3;
\r
2317 case 4: return GL_FLOAT_VEC4;
\r
2318 default: UNREACHABLE();
\r
2321 else if(type.isMatrix())
\r
2323 switch(type.getNominalSize())
\r
2325 case 2: return GL_FLOAT_MAT2;
\r
2326 case 3: return GL_FLOAT_MAT3;
\r
2327 case 4: return GL_FLOAT_MAT4;
\r
2328 default: UNREACHABLE();
\r
2331 else UNREACHABLE();
\r
2333 else if(type.getBasicType() == EbtInt)
\r
2335 if(type.isScalar())
\r
2339 else if(type.isVector())
\r
2341 switch(type.getNominalSize())
\r
2343 case 2: return GL_INT_VEC2;
\r
2344 case 3: return GL_INT_VEC3;
\r
2345 case 4: return GL_INT_VEC4;
\r
2346 default: UNREACHABLE();
\r
2349 else UNREACHABLE();
\r
2351 else if(type.getBasicType() == EbtBool)
\r
2353 if(type.isScalar())
\r
2357 else if(type.isVector())
\r
2359 switch(type.getNominalSize())
\r
2361 case 2: return GL_BOOL_VEC2;
\r
2362 case 3: return GL_BOOL_VEC3;
\r
2363 case 4: return GL_BOOL_VEC4;
\r
2364 default: UNREACHABLE();
\r
2367 else UNREACHABLE();
\r
2369 else if(type.getBasicType() == EbtSampler2D)
\r
2371 return GL_SAMPLER_2D;
\r
2373 else if(type.getBasicType() == EbtSamplerCube)
\r
2375 return GL_SAMPLER_CUBE;
\r
2377 else if(type.getBasicType() == EbtSamplerExternalOES)
\r
2379 return GL_SAMPLER_EXTERNAL_OES;
\r
2381 else UNREACHABLE();
\r
2386 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2388 if(type.getBasicType() == EbtFloat)
\r
2390 switch(type.getPrecision())
\r
2392 case EbpHigh: return GL_HIGH_FLOAT;
\r
2393 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2394 case EbpLow: return GL_LOW_FLOAT;
\r
2395 case EbpUndefined:
\r
2396 // Should be defined as the default precision by the parser
\r
2397 default: UNREACHABLE();
\r
2400 else if(type.getBasicType() == EbtInt)
\r
2402 switch (type.getPrecision())
\r
2404 case EbpHigh: return GL_HIGH_INT;
\r
2405 case EbpMedium: return GL_MEDIUM_INT;
\r
2406 case EbpLow: return GL_LOW_INT;
\r
2407 case EbpUndefined:
\r
2408 // Should be defined as the default precision by the parser
\r
2409 default: UNREACHABLE();
\r
2413 // Other types (boolean, sampler) don't have a precision
\r
2417 int OutputASM::dim(TIntermNode *v)
\r
2419 TIntermTyped *vector = v->getAsTyped();
\r
2420 ASSERT(vector && vector->isRegister());
\r
2421 return vector->getNominalSize();
\r
2424 int OutputASM::dim2(TIntermNode *m)
\r
2426 TIntermTyped *matrix = m->getAsTyped();
\r
2427 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2428 return matrix->getNominalSize();
\r
2431 // Returns ~0 if no loop count could be determined
\r
2432 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2434 // Parse loops of the form:
\r
2435 // for(int index = initial; index [comparator] limit; index += increment)
\r
2436 TIntermSymbol *index = 0;
\r
2437 TOperator comparator = EOpNull;
\r
2440 int increment = 0;
\r
2442 // Parse index name and intial value
\r
2443 if(node->getInit())
\r
2445 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2449 TIntermSequence &sequence = init->getSequence();
\r
2450 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2452 if(variable && variable->getQualifier() == EvqTemporary)
\r
2454 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2456 if(assign->getOp() == EOpInitialize)
\r
2458 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2459 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2461 if(symbol && constant)
\r
2463 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2466 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2474 // Parse comparator and limit value
\r
2475 if(index && node->getCondition())
\r
2477 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2479 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2481 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2485 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2487 comparator = test->getOp();
\r
2488 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2494 // Parse increment
\r
2495 if(index && comparator != EOpNull && node->getExpression())
\r
2497 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2498 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2500 if(binaryTerminal)
\r
2502 TOperator op = binaryTerminal->getOp();
\r
2503 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2507 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2509 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2513 case EOpAddAssign: increment = value; break;
\r
2514 case EOpSubAssign: increment = -value; break;
\r
2515 default: UNIMPLEMENTED();
\r
2520 else if(unaryTerminal)
\r
2522 TOperator op = unaryTerminal->getOp();
\r
2526 case EOpPostIncrement: increment = 1; break;
\r
2527 case EOpPostDecrement: increment = -1; break;
\r
2528 case EOpPreIncrement: increment = 1; break;
\r
2529 case EOpPreDecrement: increment = -1; break;
\r
2530 default: UNIMPLEMENTED();
\r
2535 if(index && comparator != EOpNull && increment != 0)
\r
2537 if(comparator == EOpLessThanEqual)
\r
2539 comparator = EOpLessThan;
\r
2543 if(comparator == EOpLessThan)
\r
2545 int iterations = (limit - initial) / increment;
\r
2547 if(iterations <= 0)
\r
2552 return iterations;
\r
2554 else UNIMPLEMENTED(); // Falls through
\r
2560 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2563 loopDiscontinuity = false;
\r
2565 node->traverse(this);
\r
2567 return loopDiscontinuity;
\r
2570 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2572 if(visit == PreVisit)
\r
2576 else if(visit == PostVisit)
\r
2584 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2586 if(loopDiscontinuity)
\r
2596 switch(node->getFlowOp())
\r
2603 loopDiscontinuity = true;
\r
2605 default: UNREACHABLE();
\r
2608 return !loopDiscontinuity;
\r
2611 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2613 return !loopDiscontinuity;
\r