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 "OutputASM.h"
\r
14 #include "common/debug.h"
\r
15 #include "InfoSink.h"
\r
17 #include "libGLESv2/Shader.h"
\r
19 #include <GLES2/gl2.h>
\r
20 #include <GLES2/gl2ext.h>
\r
21 #include <GLES3/gl3.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, 1, 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, EvqConstExpr, 4, 1, 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, EvqConstExpr, 1, 1, false))
\r
62 constants[0].setBConst(b);
\r
65 Constant(int i) : TIntermConstantUnion(constants, TType(EbtInt, EbpHigh, EvqConstExpr, 1, 1, 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 sw::PixelShader *Shader::getPixelShader() const
\r
107 sw::VertexShader *Shader::getVertexShader() const
\r
112 OutputASM::OutputASM(TParseContext &context, Shader *shaderObject) : TIntermTraverser(true, true, true), mContext(context), shaderObject(shaderObject)
\r
120 shader = shaderObject->getShader();
\r
121 pixelShader = shaderObject->getPixelShader();
\r
122 vertexShader = shaderObject->getVertexShader();
\r
125 functionArray.push_back(Function(0, "main(", 0, 0));
\r
126 currentFunction = 0;
\r
127 outputQualifier = EvqOutput; // Set outputQualifier to any value other than EvqFragColor or EvqFragData
\r
130 OutputASM::~OutputASM()
\r
134 void OutputASM::output()
\r
138 emitShader(GLOBAL);
\r
140 if(functionArray.size() > 1) // Only call main() when there are other functions
\r
142 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
\r
143 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
\r
144 callMain->dst.index = 0; // main()
\r
146 emit(sw::Shader::OPCODE_RET);
\r
149 emitShader(FUNCTION);
\r
153 void OutputASM::emitShader(Scope scope)
\r
156 currentScope = GLOBAL;
\r
157 mContext.treeRoot->traverse(this);
\r
160 void OutputASM::freeTemporary(Temporary *temporary)
\r
162 free(temporaries, temporary);
\r
165 void OutputASM::visitSymbol(TIntermSymbol *symbol)
\r
167 // Vertex varyings don't have to be actively used to successfully link
\r
168 // against pixel shaders that use them. So make sure they're declared.
\r
169 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut)
\r
171 if(symbol->getBasicType() != EbtInvariant) // Typeless declarations are not new varyings
\r
173 declareVarying(symbol, -1);
\r
178 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
\r
180 if(currentScope != emitScope)
\r
185 TIntermTyped *result = node;
\r
186 TIntermTyped *left = node->getLeft();
\r
187 TIntermTyped *right = node->getRight();
\r
188 const TType &leftType = left->getType();
\r
189 const TType &rightType = right->getType();
\r
190 const TType &resultType = node->getType();
\r
192 switch(node->getOp())
\r
195 if(visit == PostVisit)
\r
197 assignLvalue(left, right);
\r
198 copy(result, right);
\r
201 case EOpInitialize:
\r
202 if(visit == PostVisit)
\r
207 case EOpMatrixTimesScalarAssign:
\r
208 if(visit == PostVisit)
\r
210 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
212 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
213 mul->dst.index += i;
\r
214 argument(mul->src[0], left, i);
\r
217 assignLvalue(left, result);
\r
220 case EOpVectorTimesMatrixAssign:
\r
221 if(visit == PostVisit)
\r
223 int size = leftType.getNominalSize();
\r
225 for(int i = 0; i < size; i++)
\r
227 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
228 dot->dst.mask = 1 << i;
\r
229 argument(dot->src[1], right, i);
\r
232 assignLvalue(left, result);
\r
235 case EOpMatrixTimesMatrixAssign:
\r
236 if(visit == PostVisit)
\r
238 int dim = leftType.getNominalSize();
\r
240 for(int i = 0; i < dim; i++)
\r
242 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
243 mul->dst.index += i;
\r
244 argument(mul->src[1], right, i);
\r
245 mul->src[1].swizzle = 0x00;
\r
247 for(int j = 1; j < dim; j++)
\r
249 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
250 mad->dst.index += i;
\r
251 argument(mad->src[0], left, j);
\r
252 argument(mad->src[1], right, i);
\r
253 mad->src[1].swizzle = j * 0x55;
\r
254 argument(mad->src[2], result, i);
\r
258 assignLvalue(left, result);
\r
261 case EOpIndexDirect:
\r
262 if(visit == PostVisit)
\r
264 int index = right->getAsConstantUnion()->getIConst(0);
\r
266 if(result->isMatrix() || result->isStruct())
\r
268 ASSERT(left->isArray());
\r
269 copy(result, left, index * left->elementRegisterCount());
\r
271 else if(result->isRegister())
\r
273 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
275 if(left->isRegister())
\r
277 mov->src[0].swizzle = index;
\r
279 else if(left->isArray())
\r
281 argument(mov->src[0], left, index * left->elementRegisterCount());
\r
283 else if(left->isMatrix())
\r
285 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
\r
286 argument(mov->src[0], left, index);
\r
288 else UNREACHABLE();
\r
290 else UNREACHABLE();
\r
293 case EOpIndexIndirect:
\r
294 if(visit == PostVisit)
\r
296 if(left->isArray() || left->isMatrix())
\r
298 for(int index = 0; index < result->totalRegisterCount(); index++)
\r
300 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
301 mov->dst.index += index;
\r
302 mov->dst.mask = writeMask(result, index);
\r
303 argument(mov->src[0], left, index);
\r
305 if(left->totalRegisterCount() > 1)
\r
307 sw::Shader::SourceParameter relativeRegister;
\r
308 argument(relativeRegister, right);
\r
310 mov->src[0].rel.type = relativeRegister.type;
\r
311 mov->src[0].rel.index = relativeRegister.index;
\r
312 mov->src[0].rel.scale = result->totalRegisterCount();
\r
313 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
317 else if(left->isRegister())
\r
319 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
\r
321 else UNREACHABLE();
\r
324 case EOpIndexDirectStruct:
\r
325 if(visit == PostVisit)
\r
327 ASSERT(leftType.isStruct());
\r
329 const TTypeList *structure = leftType.getStruct();
\r
330 const TString &fieldName = rightType.getFieldName();
\r
331 int fieldOffset = 0;
\r
333 for(size_t i = 0; i < structure->size(); i++)
\r
335 const TType &fieldType = *(*structure)[i].type;
\r
337 if(fieldType.getFieldName() == fieldName)
\r
342 fieldOffset += fieldType.totalRegisterCount();
\r
345 copy(result, left, fieldOffset);
\r
348 case EOpVectorSwizzle:
\r
349 if(visit == PostVisit)
\r
352 TIntermAggregate *components = right->getAsAggregate();
\r
356 TIntermSequence &sequence = components->getSequence();
\r
359 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
361 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
365 int i = element->getUnionArrayPointer()[0].getIConst();
\r
366 swizzle |= i << (component * 2);
\r
369 else UNREACHABLE();
\r
372 else UNREACHABLE();
\r
374 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
375 mov->src[0].swizzle = swizzle;
\r
378 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
379 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
380 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
381 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
382 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
383 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
384 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
385 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
387 if(visit == PostVisit)
\r
389 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
391 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
393 Temporary equal(this);
\r
394 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
395 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
400 if(visit == PostVisit)
\r
402 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
404 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
406 Temporary notEqual(this);
\r
407 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
408 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
412 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
413 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
414 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
415 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
416 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
417 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
418 case EOpMatrixTimesScalar:
\r
419 if(visit == PostVisit)
\r
421 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
423 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
424 mul->dst.index += i;
\r
425 argument(mul->src[0], left, i);
\r
429 case EOpVectorTimesMatrix:
\r
430 if(visit == PostVisit)
\r
432 int size = leftType.getNominalSize();
\r
434 for(int i = 0; i < size; i++)
\r
436 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
437 dot->dst.mask = 1 << i;
\r
438 argument(dot->src[1], right, i);
\r
442 case EOpMatrixTimesVector:
\r
443 if(visit == PostVisit)
\r
445 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
446 mul->src[1].swizzle = 0x00;
\r
448 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
450 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
451 argument(mad->src[0], left, i);
\r
452 mad->src[1].swizzle = i * 0x55;
\r
456 case EOpMatrixTimesMatrix:
\r
457 if(visit == PostVisit)
\r
459 int dim = leftType.getNominalSize();
\r
461 for(int i = 0; i < dim; i++)
\r
463 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
464 mul->dst.index += i;
\r
465 argument(mul->src[1], right, i);
\r
466 mul->src[1].swizzle = 0x00;
\r
468 for(int j = 1; j < dim; j++)
\r
470 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
471 mad->dst.index += i;
\r
472 argument(mad->src[0], left, j);
\r
473 argument(mad->src[1], right, i);
\r
474 mad->src[1].swizzle = j * 0x55;
\r
475 argument(mad->src[2], result, i);
\r
481 if(trivial(right, 6))
\r
483 if(visit == PostVisit)
\r
485 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
488 else // Short-circuit evaluation
\r
490 if(visit == InVisit)
\r
492 emit(sw::Shader::OPCODE_MOV, result, left);
\r
493 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
494 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
496 else if(visit == PostVisit)
\r
498 emit(sw::Shader::OPCODE_MOV, result, right);
\r
499 emit(sw::Shader::OPCODE_ENDIF);
\r
503 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
504 case EOpLogicalAnd:
\r
505 if(trivial(right, 6))
\r
507 if(visit == PostVisit)
\r
509 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
512 else // Short-circuit evaluation
\r
514 if(visit == InVisit)
\r
516 emit(sw::Shader::OPCODE_MOV, result, left);
\r
517 emit(sw::Shader::OPCODE_IF, 0, result);
\r
519 else if(visit == PostVisit)
\r
521 emit(sw::Shader::OPCODE_MOV, result, right);
\r
522 emit(sw::Shader::OPCODE_ENDIF);
\r
526 default: UNREACHABLE();
\r
532 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
534 if(currentScope != emitScope)
\r
539 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
540 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
541 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
543 TIntermTyped *result = node;
\r
544 TIntermTyped *arg = node->getOperand();
\r
546 switch(node->getOp())
\r
549 if(visit == PostVisit)
\r
551 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
553 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
554 neg->dst.index += index;
\r
555 argument(neg->src[0], arg, index);
\r
556 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
560 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
561 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
562 case EOpPostIncrement:
\r
563 if(visit == PostVisit)
\r
567 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
569 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
570 add->dst.index += index;
\r
571 argument(add->src[0], arg, index);
\r
574 assignLvalue(arg, arg);
\r
577 case EOpPostDecrement:
\r
578 if(visit == PostVisit)
\r
582 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
584 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
585 sub->dst.index += index;
\r
586 argument(sub->src[0], arg, index);
\r
589 assignLvalue(arg, arg);
\r
592 case EOpPreIncrement:
\r
593 if(visit == PostVisit)
\r
595 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
597 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
598 add->dst.index += index;
\r
599 argument(add->src[0], arg, index);
\r
602 assignLvalue(arg, result);
\r
605 case EOpPreDecrement:
\r
606 if(visit == PostVisit)
\r
608 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
610 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
611 sub->dst.index += index;
\r
612 argument(sub->src[0], arg, index);
\r
615 assignLvalue(arg, result);
\r
618 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
619 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
620 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
621 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
622 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
623 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
624 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
625 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
626 case EOpSinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_SINH, result, arg); break;
\r
627 case EOpCosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_COSH, result, arg); break;
\r
628 case EOpTanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_TANH, result, arg); break;
\r
629 case EOpAsinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASINH, result, arg); break;
\r
630 case EOpAcosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOSH, result, arg); break;
\r
631 case EOpAtanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATANH, result, arg); break;
\r
632 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
633 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
634 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
635 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
636 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
637 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
638 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
639 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
640 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
641 case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
\r
642 case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
\r
643 case EOpRoundEven: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUNDEVEN, result, arg); break;
\r
644 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
645 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
646 case EOpIsNan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISNAN, result, arg); break;
\r
647 case EOpIsInf: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISINF, result, arg); break;
\r
648 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
649 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
650 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
651 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
652 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
653 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
654 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
655 default: UNREACHABLE();
\r
661 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
663 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
668 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
670 TIntermTyped *result = node;
\r
671 const TType &resultType = node->getType();
\r
672 TIntermSequence &arg = node->getSequence();
\r
673 int argumentCount = arg.size();
\r
675 switch(node->getOp())
\r
677 case EOpSequence: break;
\r
678 case EOpDeclaration: break;
\r
679 case EOpPrototype: break;
\r
681 if(visit == PostVisit)
\r
683 copy(result, arg[1]);
\r
687 if(visit == PreVisit)
\r
689 const TString &name = node->getName();
\r
691 if(emitScope == FUNCTION)
\r
693 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
695 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
696 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
698 const Function *function = findFunction(name);
\r
699 ASSERT(function); // Should have been added during global pass
\r
700 label->dst.index = function->label;
\r
701 currentFunction = function->label;
\r
704 else if(emitScope == GLOBAL)
\r
706 if(name != "main(")
\r
708 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
709 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
712 else UNREACHABLE();
\r
714 currentScope = FUNCTION;
\r
716 else if(visit == PostVisit)
\r
718 if(emitScope == FUNCTION)
\r
720 if(functionArray.size() > 1) // No need to return when there's only main()
\r
722 emit(sw::Shader::OPCODE_RET);
\r
726 currentScope = GLOBAL;
\r
729 case EOpFunctionCall:
\r
730 if(visit == PostVisit)
\r
732 if(node->isUserDefined())
\r
734 const TString &name = node->getName();
\r
735 const Function *function = findFunction(name);
\r
739 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
743 TIntermSequence &arguments = *function->arg;
\r
745 for(int i = 0; i < argumentCount; i++)
\r
747 TIntermTyped *in = arguments[i]->getAsTyped();
\r
749 if(in->getQualifier() == EvqIn ||
\r
750 in->getQualifier() == EvqInOut ||
\r
751 in->getQualifier() == EvqConstReadOnly)
\r
757 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
758 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
759 call->dst.index = function->label;
\r
761 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
763 copy(result, function->ret);
\r
766 for(int i = 0; i < argumentCount; i++)
\r
768 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
769 TIntermTyped *out = arg[i]->getAsTyped();
\r
771 if(argument->getQualifier() == EvqOut ||
\r
772 argument->getQualifier() == EvqInOut)
\r
774 copy(out, argument);
\r
780 TString name = TFunction::unmangleName(node->getName());
\r
782 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
784 if(argumentCount == 2)
\r
786 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
788 else if(argumentCount == 3) // bias
\r
790 Temporary uvwb(this);
\r
791 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
792 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
793 bias->dst.mask = 0x8;
\r
795 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
798 else UNREACHABLE();
\r
800 else if(name == "texture2DProj")
\r
802 TIntermTyped *t = arg[1]->getAsTyped();
\r
804 if(argumentCount == 2)
\r
806 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
807 tex->project = true;
\r
809 if(t->getNominalSize() == 3)
\r
811 tex->src[0].swizzle = 0xA4;
\r
813 else ASSERT(t->getNominalSize() == 4);
\r
815 else if(argumentCount == 3) // bias
\r
817 Temporary proj(this);
\r
819 if(t->getNominalSize() == 3)
\r
821 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
822 div->src[1].swizzle = 0xAA;
\r
823 div->dst.mask = 0x3;
\r
825 else if(t->getNominalSize() == 4)
\r
827 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
828 div->src[1].swizzle = 0xFF;
\r
829 div->dst.mask = 0x3;
\r
831 else UNREACHABLE();
\r
833 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
834 bias->dst.mask = 0x8;
\r
836 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
839 else UNREACHABLE();
\r
841 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
843 Temporary uvwb(this);
\r
844 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
845 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
846 lod->dst.mask = 0x8;
\r
848 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
850 else if(name == "texture2DProjLod")
\r
852 TIntermTyped *t = arg[1]->getAsTyped();
\r
853 Temporary proj(this);
\r
855 if(t->getNominalSize() == 3)
\r
857 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
858 div->src[1].swizzle = 0xAA;
\r
859 div->dst.mask = 0x3;
\r
861 else if(t->getNominalSize() == 4)
\r
863 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
864 div->src[1].swizzle = 0xFF;
\r
865 div->dst.mask = 0x3;
\r
867 else UNREACHABLE();
\r
869 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
870 lod->dst.mask = 0x8;
\r
872 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
874 else UNREACHABLE();
\r
878 case EOpParameters:
\r
880 case EOpConstructFloat:
\r
881 case EOpConstructVec2:
\r
882 case EOpConstructVec3:
\r
883 case EOpConstructVec4:
\r
884 case EOpConstructBool:
\r
885 case EOpConstructBVec2:
\r
886 case EOpConstructBVec3:
\r
887 case EOpConstructBVec4:
\r
888 case EOpConstructInt:
\r
889 case EOpConstructIVec2:
\r
890 case EOpConstructIVec3:
\r
891 case EOpConstructIVec4:
\r
892 case EOpConstructUInt:
\r
893 case EOpConstructUVec2:
\r
894 case EOpConstructUVec3:
\r
895 case EOpConstructUVec4:
\r
896 if(visit == PostVisit)
\r
900 for(int i = 0; i < argumentCount; i++)
\r
902 TIntermTyped *argi = arg[i]->getAsTyped();
\r
903 int size = argi->getNominalSize();
\r
905 if(!argi->isMatrix())
\r
907 Instruction *mov = emitCast(result, argi);
\r
908 mov->dst.mask = (0xF << component) & 0xF;
\r
909 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
917 while(component < resultType.getNominalSize())
\r
919 Instruction *mov = emitCast(result, argi);
\r
920 mov->dst.mask = (0xF << component) & 0xF;
\r
921 mov->src[0].index += column;
\r
922 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
931 case EOpConstructMat2:
\r
932 case EOpConstructMat2x3:
\r
933 case EOpConstructMat2x4:
\r
934 case EOpConstructMat3x2:
\r
935 case EOpConstructMat3:
\r
936 case EOpConstructMat3x4:
\r
937 case EOpConstructMat4x2:
\r
938 case EOpConstructMat4x3:
\r
939 case EOpConstructMat4:
\r
940 if(visit == PostVisit)
\r
942 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
943 const int dim = result->getNominalSize();
\r
945 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
947 for(int i = 0; i < dim; i++)
\r
949 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
950 init->dst.index += i;
\r
951 Instruction *mov = emitCast(result, arg0);
\r
952 mov->dst.index += i;
\r
953 mov->dst.mask = 1 << i;
\r
954 ASSERT(mov->src[0].swizzle == 0x00);
\r
957 else if(arg0->isMatrix())
\r
959 for(int i = 0; i < dim; i++)
\r
961 if(dim > dim2(arg0))
\r
963 // Initialize to identity matrix
\r
964 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
965 Instruction *mov = emitCast(result, &col);
\r
966 mov->dst.index += i;
\r
971 Instruction *mov = emitCast(result, arg0);
\r
972 mov->dst.index += i;
\r
973 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
974 argument(mov->src[0], arg0, i);
\r
983 for(int i = 0; i < argumentCount; i++)
\r
985 TIntermTyped *argi = arg[i]->getAsTyped();
\r
986 int size = argi->getNominalSize();
\r
989 while(element < size)
\r
991 Instruction *mov = emitCast(result, argi);
\r
992 mov->dst.index += column;
\r
993 mov->dst.mask = (0xF << row) & 0xF;
\r
994 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
996 int end = row + size - element;
\r
997 column = end >= dim ? column + 1 : column;
\r
998 element = element + dim - row;
\r
999 row = end >= dim ? 0 : end;
\r
1005 case EOpConstructStruct:
\r
1006 if(visit == PostVisit)
\r
1009 for(int i = 0; i < argumentCount; i++)
\r
1011 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1012 int size = argi->totalRegisterCount();
\r
1014 for(int index = 0; index < size; index++)
\r
1016 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
1017 mov->dst.index += index + offset;
\r
1018 mov->dst.mask = writeMask(result, offset + index);
\r
1019 argument(mov->src[0], argi, index);
\r
1026 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1027 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1028 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1029 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1030 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1031 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1032 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1033 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1034 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1035 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1036 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1038 if(visit == PostVisit)
\r
1040 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1041 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1044 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1045 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1046 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1047 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg[0]); break;
\r
1048 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg[0]); break;
\r
1049 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg[0]); break;
\r
1050 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg[0]); break;
\r
1051 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg[0]); break;
\r
1052 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg[0]); break;
\r
1053 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg[0]); break;
\r
1054 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg[0]); break;
\r
1055 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg[0]); break;
\r
1056 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg[0]); break;
\r
1057 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1058 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1059 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1060 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1061 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1062 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1064 if(visit == PostVisit)
\r
1066 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1068 for(int i = 0; i < dim2(arg[0]); i++)
\r
1070 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1071 mul->dst.index += i;
\r
1072 argument(mul->src[0], arg[0], i);
\r
1073 argument(mul->src[1], arg[1], i);
\r
1077 default: UNREACHABLE();
\r
1083 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1085 if(currentScope != emitScope)
\r
1090 TIntermTyped *condition = node->getCondition();
\r
1091 TIntermNode *trueBlock = node->getTrueBlock();
\r
1092 TIntermNode *falseBlock = node->getFalseBlock();
\r
1093 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1095 condition->traverse(this);
\r
1097 if(node->usesTernaryOperator())
\r
1099 if(constantCondition)
\r
1101 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1105 trueBlock->traverse(this);
\r
1106 copy(node, trueBlock);
\r
1110 falseBlock->traverse(this);
\r
1111 copy(node, falseBlock);
\r
1114 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1116 trueBlock->traverse(this);
\r
1117 falseBlock->traverse(this);
\r
1118 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1122 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1126 trueBlock->traverse(this);
\r
1127 copy(node, trueBlock);
\r
1132 emit(sw::Shader::OPCODE_ELSE);
\r
1133 falseBlock->traverse(this);
\r
1134 copy(node, falseBlock);
\r
1137 emit(sw::Shader::OPCODE_ENDIF);
\r
1140 else // if/else statement
\r
1142 if(constantCondition)
\r
1144 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1150 trueBlock->traverse(this);
\r
1157 falseBlock->traverse(this);
\r
1163 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1167 trueBlock->traverse(this);
\r
1172 emit(sw::Shader::OPCODE_ELSE);
\r
1173 falseBlock->traverse(this);
\r
1176 emit(sw::Shader::OPCODE_ENDIF);
\r
1183 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1185 if(currentScope != emitScope)
\r
1190 unsigned int iterations = loopCount(node);
\r
1192 if(iterations == 0)
\r
1197 bool unroll = (iterations <= 4);
\r
1201 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1202 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1205 TIntermNode *init = node->getInit();
\r
1206 TIntermTyped *condition = node->getCondition();
\r
1207 TIntermTyped *expression = node->getExpression();
\r
1208 TIntermNode *body = node->getBody();
\r
1210 if(node->getType() == ELoopDoWhile)
\r
1212 Temporary iterate(this);
\r
1213 Constant True(true);
\r
1214 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1216 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1220 body->traverse(this);
\r
1223 emit(sw::Shader::OPCODE_TEST);
\r
1225 condition->traverse(this);
\r
1226 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1228 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1234 init->traverse(this);
\r
1239 for(unsigned int i = 0; i < iterations; i++)
\r
1241 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1245 body->traverse(this);
\r
1250 expression->traverse(this);
\r
1256 condition->traverse(this);
\r
1258 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1262 body->traverse(this);
\r
1265 emit(sw::Shader::OPCODE_TEST);
\r
1269 expression->traverse(this);
\r
1272 condition->traverse(this);
\r
1274 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1281 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1283 if(currentScope != emitScope)
\r
1288 switch(node->getFlowOp())
\r
1290 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1291 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1292 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1294 if(visit == PostVisit)
\r
1296 TIntermTyped *value = node->getExpression();
\r
1300 copy(functionArray[currentFunction].ret, value);
\r
1303 emit(sw::Shader::OPCODE_LEAVE);
\r
1306 default: UNREACHABLE();
\r
1312 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1314 return operand && isSamplerRegister(operand->getType());
\r
1317 bool OutputASM::isSamplerRegister(const TType &type)
\r
1319 // A sampler register's qualifiers can be:
\r
1320 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1321 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1322 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1323 // the sampler has been copied and is no longer a sampler register.
\r
1324 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1327 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1329 if(isSamplerRegister(dst))
\r
1331 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1334 Instruction *instruction = new Instruction(op);
\r
1338 instruction->dst.type = registerType(dst);
\r
1339 instruction->dst.index = registerIndex(dst) + index;
\r
1340 instruction->dst.mask = writeMask(dst);
\r
1341 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1344 argument(instruction->src[0], src0, index);
\r
1345 argument(instruction->src[1], src1, index);
\r
1346 argument(instruction->src[2], src2, index);
\r
1348 shader->append(instruction);
\r
1350 return instruction;
\r
1353 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1355 // Integers are implemented as float
\r
1356 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1358 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1360 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1362 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1364 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1366 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1369 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1372 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1374 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1376 emit(op, dst, src0, src1, src2, index);
\r
1380 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1382 emitBinary(op, result, src0, src1);
\r
1383 assignLvalue(lhs, result);
\r
1386 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1388 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1389 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1391 Instruction *cmp = emit(opcode, dst, left, right);
\r
1392 cmp->control = cmpOp;
\r
1393 argument(cmp->src[0], left, index);
\r
1394 argument(cmp->src[1], right, index);
\r
1397 int componentCount(const TType &type, int registers)
\r
1399 if(registers == 0)
\r
1404 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1406 int index = registers / type.elementRegisterCount();
\r
1407 registers -= index * type.elementRegisterCount();
\r
1408 return index * type.getElementSize() + componentCount(type, registers);
\r
1411 if(type.isStruct())
\r
1413 TTypeList *structure = type.getStruct();
\r
1416 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1418 const TType &fieldType = *field->type;
\r
1420 if(fieldType.totalRegisterCount() <= registers)
\r
1422 registers -= fieldType.totalRegisterCount();
\r
1423 elements += fieldType.getObjectSize();
\r
1425 else // Register within this field
\r
1427 return elements + componentCount(fieldType, registers);
\r
1431 else if(type.isMatrix())
\r
1433 return registers * type.getSecondarySize();
\r
1440 int registerSize(const TType &type, int registers)
\r
1442 if(registers == 0)
\r
1444 if(type.isStruct())
\r
1446 return registerSize(*type.getStruct()->begin()->type, 0);
\r
1449 return type.isMatrix() ? type.getSecondarySize() : type.getNominalSize();
\r
1452 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1454 int index = registers / type.elementRegisterCount();
\r
1455 registers -= index * type.elementRegisterCount();
\r
1456 return registerSize(type, registers);
\r
1459 if(type.isStruct())
\r
1461 TTypeList *structure = type.getStruct();
\r
1464 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1466 const TType &fieldType = *field->type;
\r
1468 if(fieldType.totalRegisterCount() <= registers)
\r
1470 registers -= fieldType.totalRegisterCount();
\r
1471 elements += fieldType.getObjectSize();
\r
1473 else // Register within this field
\r
1475 return registerSize(fieldType, registers);
\r
1479 else if(type.isMatrix())
\r
1481 return registerSize(type, 0);
\r
1488 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1492 TIntermTyped *arg = argument->getAsTyped();
\r
1493 const TType &type = arg->getType();
\r
1494 const TTypeList *structure = type.getStruct();
\r
1495 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1497 int size = registerSize(type, index);
\r
1499 parameter.type = registerType(arg);
\r
1501 if(arg->getQualifier() == EvqConstExpr)
\r
1503 int component = componentCount(type, index);
\r
1504 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1506 for(int i = 0; i < 4; i++)
\r
1508 if(size == 1) // Replicate
\r
1510 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1514 parameter.value[i] = constants[component + i].getAsFloat();
\r
1518 parameter.value[i] = 0.0f;
\r
1524 parameter.index = registerIndex(arg) + index;
\r
1526 if(isSamplerRegister(arg))
\r
1528 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1532 TIntermTyped *left = binary->getLeft();
\r
1533 TIntermTyped *right = binary->getRight();
\r
1535 if(binary->getOp() == EOpIndexDirect)
\r
1537 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1539 else if(binary->getOp() == EOpIndexIndirect)
\r
1541 if(left->getArraySize() > 1)
\r
1543 parameter.rel.type = registerType(binary->getRight());
\r
1544 parameter.rel.index = registerIndex(binary->getRight());
\r
1545 parameter.rel.scale = 1;
\r
1546 parameter.rel.deterministic = true;
\r
1549 else if(binary->getOp() == EOpIndexDirectStruct)
\r
1551 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1553 else UNREACHABLE();
\r
1558 if(!IsSampler(arg->getBasicType()))
\r
1560 parameter.swizzle = readSwizzle(arg, size);
\r
1565 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1567 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1569 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1570 mov->dst.index += index;
\r
1571 mov->dst.mask = writeMask(dst, index);
\r
1572 argument(mov->src[0], src, offset + index);
\r
1576 int swizzleElement(int swizzle, int index)
\r
1578 return (swizzle >> (index * 2)) & 0x03;
\r
1581 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1583 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1584 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1585 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1586 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1589 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1592 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1593 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
\r
1595 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1598 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1600 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1602 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1604 Temporary address(this);
\r
1605 lvalue(insert->dst, address, dst);
\r
1607 insert->src[0].type = insert->dst.type;
\r
1608 insert->src[0].index = insert->dst.index;
\r
1609 insert->src[0].rel = insert->dst.rel;
\r
1610 argument(insert->src[1], src);
\r
1611 argument(insert->src[2], binary->getRight());
\r
1613 shader->append(insert);
\r
1617 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1619 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1621 Temporary address(this);
\r
1622 int swizzle = lvalue(mov->dst, address, dst);
\r
1623 mov->dst.index += offset;
\r
1627 mov->dst.mask = writeMask(dst, offset);
\r
1630 argument(mov->src[0], src, offset);
\r
1631 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1633 shader->append(mov);
\r
1638 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1640 TIntermTyped *result = node;
\r
1641 TIntermBinary *binary = node->getAsBinaryNode();
\r
1642 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1646 TIntermTyped *left = binary->getLeft();
\r
1647 TIntermTyped *right = binary->getRight();
\r
1649 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1651 switch(binary->getOp())
\r
1653 case EOpIndexDirect:
\r
1655 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1657 if(left->isRegister())
\r
1659 int leftMask = dst.mask;
\r
1662 while((leftMask & dst.mask) == 0)
\r
1664 dst.mask = dst.mask << 1;
\r
1667 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1668 dst.mask = 1 << element;
\r
1672 else if(left->isArray() || left->isMatrix())
\r
1674 dst.index += rightIndex * result->totalRegisterCount();
\r
1677 else UNREACHABLE();
\r
1680 case EOpIndexIndirect:
\r
1682 if(left->isRegister())
\r
1684 // Requires INSERT instruction (handled by calling function)
\r
1686 else if(left->isArray() || left->isMatrix())
\r
1688 int scale = result->totalRegisterCount();
\r
1690 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1692 if(left->totalRegisterCount() > 1)
\r
1694 sw::Shader::SourceParameter relativeRegister;
\r
1695 argument(relativeRegister, right);
\r
1697 dst.rel.index = relativeRegister.index;
\r
1698 dst.rel.type = relativeRegister.type;
\r
1699 dst.rel.scale = scale;
\r
1700 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1703 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1707 Constant oldScale((int)dst.rel.scale);
\r
1708 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1709 mad->src[0].index = dst.rel.index;
\r
1710 mad->src[0].type = dst.rel.type;
\r
1714 Constant oldScale((int)dst.rel.scale);
\r
1715 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1716 mul->src[0].index = dst.rel.index;
\r
1717 mul->src[0].type = dst.rel.type;
\r
1719 Constant newScale(scale);
\r
1720 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1723 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1724 dst.rel.index = registerIndex(&address);
\r
1725 dst.rel.scale = 1;
\r
1727 else // Just add the new index to the address register
\r
1731 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1735 Constant newScale(scale);
\r
1736 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1740 else UNREACHABLE();
\r
1743 case EOpIndexDirectStruct:
\r
1745 const TTypeList *structure = left->getType().getStruct();
\r
1746 const TString &fieldName = right->getType().getFieldName();
\r
1749 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1751 if(field->type->getFieldName() == fieldName)
\r
1753 dst.type = registerType(left);
\r
1754 dst.index += offset;
\r
1755 dst.mask = writeMask(right);
\r
1760 offset += field->type->totalRegisterCount();
\r
1764 case EOpVectorSwizzle:
\r
1766 ASSERT(left->isRegister());
\r
1768 int leftMask = dst.mask;
\r
1771 int rightMask = 0;
\r
1773 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1775 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1777 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
1779 int element = swizzleElement(leftSwizzle, index);
\r
1780 rightMask = rightMask | (1 << element);
\r
1781 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1784 dst.mask = leftMask & rightMask;
\r
1790 UNREACHABLE(); // Not an l-value operator
\r
1796 dst.type = registerType(symbol);
\r
1797 dst.index = registerIndex(symbol);
\r
1798 dst.mask = writeMask(symbol);
\r
1805 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1807 if(isSamplerRegister(operand))
\r
1809 return sw::Shader::PARAMETER_SAMPLER;
\r
1812 const TQualifier qualifier = operand->getQualifier();
\r
1813 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1815 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1816 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1818 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1820 outputQualifier = qualifier;
\r
1825 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1826 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1827 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1828 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1829 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1830 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1831 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1832 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1833 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1834 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1835 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1836 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1837 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1838 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1839 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1840 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1841 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1842 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1843 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1844 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1845 default: UNREACHABLE();
\r
1848 return sw::Shader::PARAMETER_VOID;
\r
1851 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1853 if(isSamplerRegister(operand))
\r
1855 return samplerRegister(operand);
\r
1858 switch(operand->getQualifier())
\r
1860 case EvqTemporary: return temporaryRegister(operand);
\r
1861 case EvqGlobal: return temporaryRegister(operand);
\r
1862 case EvqConstExpr: UNREACHABLE();
\r
1863 case EvqAttribute: return attributeRegister(operand);
\r
1864 case EvqVaryingIn: return varyingRegister(operand);
\r
1865 case EvqVaryingOut: return varyingRegister(operand);
\r
1866 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1867 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1868 case EvqUniform: return uniformRegister(operand);
\r
1869 case EvqIn: return temporaryRegister(operand);
\r
1870 case EvqOut: return temporaryRegister(operand);
\r
1871 case EvqInOut: return temporaryRegister(operand);
\r
1872 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1873 case EvqPosition: return varyingRegister(operand);
\r
1874 case EvqPointSize: return varyingRegister(operand);
\r
1875 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1876 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1877 case EvqPointCoord: return varyingRegister(operand);
\r
1878 case EvqFragColor: return 0;
\r
1879 case EvqFragData: return 0;
\r
1880 default: UNREACHABLE();
\r
1886 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1888 if(destination->getQualifier() == EvqPointSize)
\r
1890 return 0x2; // Point size stored in the y component
\r
1893 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1896 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1898 if(argument->getQualifier() == EvqPointSize)
\r
1900 return 0x55; // Point size stored in the y component
\r
1903 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1905 return swizzleSize[size];
\r
1908 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1909 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1911 if(!expression->isRegister())
\r
1916 return cost(expression, budget) >= 0;
\r
1919 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1920 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1927 if(expression->getAsSymbolNode())
\r
1931 else if(expression->getAsConstantUnion())
\r
1935 else if(expression->getAsBinaryNode())
\r
1937 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1939 switch(binary->getOp())
\r
1941 case EOpVectorSwizzle:
\r
1942 case EOpIndexDirect:
\r
1943 case EOpIndexDirectStruct:
\r
1944 return cost(binary->getLeft(), budget - 0);
\r
1948 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1953 else if(expression->getAsUnaryNode())
\r
1955 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1957 switch(unary->getOp())
\r
1961 return cost(unary->getOperand(), budget - 1);
\r
1966 else if(expression->getAsSelectionNode())
\r
1968 TIntermSelection *selection = expression->getAsSelectionNode();
\r
1970 if(selection->usesTernaryOperator())
\r
1972 TIntermTyped *condition = selection->getCondition();
\r
1973 TIntermNode *trueBlock = selection->getTrueBlock();
\r
1974 TIntermNode *falseBlock = selection->getFalseBlock();
\r
1975 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1977 if(constantCondition)
\r
1979 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1983 return cost(trueBlock, budget - 0);
\r
1987 return cost(falseBlock, budget - 0);
\r
1992 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
2000 const Function *OutputASM::findFunction(const TString &name)
\r
2002 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
2004 if(functionArray[f].name == name)
\r
2006 return &functionArray[f];
\r
2013 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
2015 return allocate(temporaries, temporary);
\r
2018 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
2020 int var = lookup(varyings, varying);
\r
2024 var = allocate(varyings, varying);
\r
2025 int componentCount = varying->getNominalSize();
\r
2026 int registerCount = varying->totalRegisterCount();
\r
2030 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2032 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2036 if(varying->getQualifier() == EvqPointCoord)
\r
2038 ASSERT(varying->isRegister());
\r
2039 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2040 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2041 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2042 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2046 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2048 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2049 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2050 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2051 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2055 else if(vertexShader)
\r
2057 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2059 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2063 if(varying->getQualifier() == EvqPosition)
\r
2065 ASSERT(varying->isRegister());
\r
2066 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2067 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2068 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2069 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2070 vertexShader->positionRegister = var;
\r
2072 else if(varying->getQualifier() == EvqPointSize)
\r
2074 ASSERT(varying->isRegister());
\r
2075 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2076 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2077 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2078 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2079 vertexShader->pointSizeRegister = var;
\r
2083 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2086 else UNREACHABLE();
\r
2088 declareVarying(varying, var);
\r
2094 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2096 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2098 const TType &type = varying->getType();
\r
2099 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2100 VaryingList &activeVaryings = shaderObject->varyings;
\r
2102 // Check if this varying has been declared before without having a register assigned
\r
2103 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2105 if(v->name == name)
\r
2109 ASSERT(v->reg < 0 || v->reg == reg);
\r
2117 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2121 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2123 const TType &type = uniform->getType();
\r
2124 ASSERT(!IsSampler(type.getBasicType()));
\r
2125 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2130 int index = lookup(uniforms, uniform);
\r
2134 index = allocate(uniforms, uniform);
\r
2135 const TString &name = symbol->getSymbol().c_str();
\r
2137 declareUniform(type, name, index);
\r
2146 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2148 ASSERT(!attribute->isArray());
\r
2149 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2151 int index = lookup(attributes, attribute);
\r
2155 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2160 index = allocate(attributes, attribute);
\r
2161 const TType &type = attribute->getType();
\r
2162 int registerCount = attribute->totalRegisterCount();
\r
2164 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2166 for(int i = 0; i < registerCount; i++)
\r
2168 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2172 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2174 const char *name = symbol->getSymbol().c_str();
\r
2175 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2182 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2184 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2185 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2186 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2190 return samplerRegister(symbol);
\r
2194 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect || binary->getOp() == EOpIndexDirectStruct);
\r
2196 return samplerRegister(binary->getLeft()); // Index added later
\r
2198 else UNREACHABLE();
\r
2203 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2205 const TType &type = sampler->getType();
\r
2206 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2208 int index = lookup(samplers, sampler);
\r
2212 index = allocate(samplers, sampler);
\r
2214 if(sampler->getQualifier() == EvqUniform)
\r
2216 const char *name = sampler->getSymbol().c_str();
\r
2217 declareUniform(type, name, index);
\r
2224 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2226 for(unsigned int i = 0; i < list.size(); i++)
\r
2228 if(list[i] == variable)
\r
2230 return i; // Pointer match
\r
2234 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2238 for(unsigned int i = 0; i < list.size(); i++)
\r
2242 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2246 if(listSymbol->getId() == varSymbol->getId())
\r
2248 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2249 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2250 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2262 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2264 int index = lookup(list, variable);
\r
2268 unsigned int registerCount = variable->totalRegisterCount();
\r
2270 for(unsigned int i = 0; i < list.size(); i++)
\r
2274 unsigned int j = 1;
\r
2275 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2277 if(list[i + j] != 0)
\r
2283 if(j == registerCount) // Found free slots
\r
2285 for(unsigned int j = 0; j < registerCount; j++)
\r
2287 list[i + j] = variable;
\r
2295 index = list.size();
\r
2297 for(unsigned int i = 0; i < registerCount; i++)
\r
2299 list.push_back(variable);
\r
2306 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2308 int index = lookup(list, variable);
\r
2316 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2318 const TTypeList *structure = type.getStruct();
\r
2319 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2323 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2325 if(isSamplerRegister(type))
\r
2327 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2329 shader->declareSampler(index + i);
\r
2335 if(type.isArray())
\r
2337 int elementIndex = index;
\r
2339 for(int i = 0; i < type.getArraySize(); i++)
\r
2341 for(size_t j = 0; j < structure->size(); j++)
\r
2343 const TType &fieldType = *(*structure)[j].type;
\r
2344 const TString &fieldName = fieldType.getFieldName();
\r
2346 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2347 declareUniform(fieldType, uniformName, elementIndex);
\r
2348 elementIndex += fieldType.totalRegisterCount();
\r
2354 int fieldIndex = index;
\r
2356 for(size_t i = 0; i < structure->size(); i++)
\r
2358 const TType &fieldType = *(*structure)[i].type;
\r
2359 const TString &fieldName = fieldType.getFieldName();
\r
2361 const TString uniformName = name + "." + fieldName;
\r
2362 declareUniform(fieldType, uniformName, fieldIndex);
\r
2363 fieldIndex += fieldType.totalRegisterCount();
\r
2369 GLenum OutputASM::glVariableType(const TType &type)
\r
2371 switch(type.getBasicType())
\r
2374 if(type.isScalar())
\r
2378 else if(type.isVector())
\r
2380 switch(type.getNominalSize())
\r
2382 case 2: return GL_FLOAT_VEC2;
\r
2383 case 3: return GL_FLOAT_VEC3;
\r
2384 case 4: return GL_FLOAT_VEC4;
\r
2385 default: UNREACHABLE();
\r
2388 else if(type.isMatrix())
\r
2390 switch(type.getNominalSize())
\r
2393 switch(type.getSecondarySize())
\r
2395 case 2: return GL_FLOAT_MAT2;
\r
2396 case 3: return GL_FLOAT_MAT2x3;
\r
2397 case 4: return GL_FLOAT_MAT2x4;
\r
2398 default: UNREACHABLE();
\r
2401 switch(type.getSecondarySize())
\r
2403 case 2: return GL_FLOAT_MAT3x2;
\r
2404 case 3: return GL_FLOAT_MAT3;
\r
2405 case 4: return GL_FLOAT_MAT3x4;
\r
2406 default: UNREACHABLE();
\r
2409 switch(type.getSecondarySize())
\r
2411 case 2: return GL_FLOAT_MAT4x2;
\r
2412 case 3: return GL_FLOAT_MAT4x3;
\r
2413 case 4: return GL_FLOAT_MAT4;
\r
2414 default: UNREACHABLE();
\r
2416 default: UNREACHABLE();
\r
2419 else UNREACHABLE();
\r
2422 if(type.isScalar())
\r
2426 else if(type.isVector())
\r
2428 switch(type.getNominalSize())
\r
2430 case 2: return GL_INT_VEC2;
\r
2431 case 3: return GL_INT_VEC3;
\r
2432 case 4: return GL_INT_VEC4;
\r
2433 default: UNREACHABLE();
\r
2436 else UNREACHABLE();
\r
2439 if(type.isScalar())
\r
2441 return GL_UNSIGNED_INT;
\r
2443 else if(type.isVector())
\r
2445 switch(type.getNominalSize())
\r
2447 case 2: return GL_UNSIGNED_INT_VEC2;
\r
2448 case 3: return GL_UNSIGNED_INT_VEC3;
\r
2449 case 4: return GL_UNSIGNED_INT_VEC4;
\r
2450 default: UNREACHABLE();
\r
2453 else UNREACHABLE();
\r
2456 if(type.isScalar())
\r
2460 else if(type.isVector())
\r
2462 switch(type.getNominalSize())
\r
2464 case 2: return GL_BOOL_VEC2;
\r
2465 case 3: return GL_BOOL_VEC3;
\r
2466 case 4: return GL_BOOL_VEC4;
\r
2467 default: UNREACHABLE();
\r
2470 else UNREACHABLE();
\r
2472 case EbtSampler2D:
\r
2473 case EbtISampler2D:
\r
2474 case EbtUSampler2D:
\r
2475 return GL_SAMPLER_2D;
\r
2476 case EbtSamplerCube:
\r
2477 case EbtISamplerCube:
\r
2478 case EbtUSamplerCube:
\r
2479 return GL_SAMPLER_CUBE;
\r
2480 case EbtSamplerExternalOES:
\r
2481 return GL_SAMPLER_EXTERNAL_OES;
\r
2482 case EbtSampler3D:
\r
2483 case EbtISampler3D:
\r
2484 case EbtUSampler3D:
\r
2485 return GL_SAMPLER_3D_OES;
\r
2486 case EbtSampler2DArray:
\r
2487 case EbtISampler2DArray:
\r
2488 case EbtUSampler2DArray:
\r
2489 return GL_SAMPLER_2D_ARRAY;
\r
2498 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2500 if(type.getBasicType() == EbtFloat)
\r
2502 switch(type.getPrecision())
\r
2504 case EbpHigh: return GL_HIGH_FLOAT;
\r
2505 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2506 case EbpLow: return GL_LOW_FLOAT;
\r
2507 case EbpUndefined:
\r
2508 // Should be defined as the default precision by the parser
\r
2509 default: UNREACHABLE();
\r
2512 else if(type.getBasicType() == EbtInt)
\r
2514 switch (type.getPrecision())
\r
2516 case EbpHigh: return GL_HIGH_INT;
\r
2517 case EbpMedium: return GL_MEDIUM_INT;
\r
2518 case EbpLow: return GL_LOW_INT;
\r
2519 case EbpUndefined:
\r
2520 // Should be defined as the default precision by the parser
\r
2521 default: UNREACHABLE();
\r
2525 // Other types (boolean, sampler) don't have a precision
\r
2529 int OutputASM::dim(TIntermNode *v)
\r
2531 TIntermTyped *vector = v->getAsTyped();
\r
2532 ASSERT(vector && vector->isRegister());
\r
2533 return vector->getNominalSize();
\r
2536 int OutputASM::dim2(TIntermNode *m)
\r
2538 TIntermTyped *matrix = m->getAsTyped();
\r
2539 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2540 return matrix->getSecondarySize();
\r
2543 // Returns ~0 if no loop count could be determined
\r
2544 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2546 // Parse loops of the form:
\r
2547 // for(int index = initial; index [comparator] limit; index += increment)
\r
2548 TIntermSymbol *index = 0;
\r
2549 TOperator comparator = EOpNull;
\r
2552 int increment = 0;
\r
2554 // Parse index name and intial value
\r
2555 if(node->getInit())
\r
2557 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2561 TIntermSequence &sequence = init->getSequence();
\r
2562 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2564 if(variable && variable->getQualifier() == EvqTemporary)
\r
2566 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2568 if(assign->getOp() == EOpInitialize)
\r
2570 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2571 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2573 if(symbol && constant)
\r
2575 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2578 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2586 // Parse comparator and limit value
\r
2587 if(index && node->getCondition())
\r
2589 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2591 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2593 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2597 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2599 comparator = test->getOp();
\r
2600 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2606 // Parse increment
\r
2607 if(index && comparator != EOpNull && node->getExpression())
\r
2609 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2610 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2612 if(binaryTerminal)
\r
2614 TOperator op = binaryTerminal->getOp();
\r
2615 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2619 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2621 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2625 case EOpAddAssign: increment = value; break;
\r
2626 case EOpSubAssign: increment = -value; break;
\r
2627 default: UNIMPLEMENTED();
\r
2632 else if(unaryTerminal)
\r
2634 TOperator op = unaryTerminal->getOp();
\r
2638 case EOpPostIncrement: increment = 1; break;
\r
2639 case EOpPostDecrement: increment = -1; break;
\r
2640 case EOpPreIncrement: increment = 1; break;
\r
2641 case EOpPreDecrement: increment = -1; break;
\r
2642 default: UNIMPLEMENTED();
\r
2647 if(index && comparator != EOpNull && increment != 0)
\r
2649 if(comparator == EOpLessThanEqual)
\r
2651 comparator = EOpLessThan;
\r
2655 if(comparator == EOpLessThan)
\r
2657 int iterations = (limit - initial) / increment;
\r
2659 if(iterations <= 0)
\r
2664 return iterations;
\r
2666 else UNIMPLEMENTED(); // Falls through
\r
2672 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2675 loopDiscontinuity = false;
\r
2677 node->traverse(this);
\r
2679 return loopDiscontinuity;
\r
2682 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2684 if(visit == PreVisit)
\r
2688 else if(visit == PostVisit)
\r
2696 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2698 if(loopDiscontinuity)
\r
2708 switch(node->getFlowOp())
\r
2715 loopDiscontinuity = true;
\r
2717 default: UNREACHABLE();
\r
2720 return !loopDiscontinuity;
\r
2723 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2725 return !loopDiscontinuity;
\r