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 || symbol->getQualifier() == EvqVertexOut)
\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(0);
\r
290 else UNREACHABLE(0);
\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(0);
\r
324 case EOpIndexDirectStruct:
\r
325 if(visit == PostVisit)
\r
327 ASSERT(leftType.isStruct());
\r
329 const TFieldList& fields = leftType.getStruct()->fields();
\r
330 int index = right->getAsConstantUnion()->getIConst(0);
\r
331 int fieldOffset = 0;
\r
333 for(int i = 0; i < index; i++)
\r
335 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
338 copy(result, left, fieldOffset);
\r
341 case EOpVectorSwizzle:
\r
342 if(visit == PostVisit)
\r
345 TIntermAggregate *components = right->getAsAggregate();
\r
349 TIntermSequence &sequence = components->getSequence();
\r
352 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
354 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
358 int i = element->getUnionArrayPointer()[0].getIConst();
\r
359 swizzle |= i << (component * 2);
\r
362 else UNREACHABLE(0);
\r
365 else UNREACHABLE(0);
\r
367 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
368 mov->src[0].swizzle = swizzle;
\r
371 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
372 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
373 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
374 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
375 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
376 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
377 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
378 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
380 if(visit == PostVisit)
\r
382 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
384 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
386 Temporary equal(this);
\r
387 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
388 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
393 if(visit == PostVisit)
\r
395 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
397 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
399 Temporary notEqual(this);
\r
400 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
401 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
405 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
406 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
407 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
408 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
409 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
410 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
411 case EOpMatrixTimesScalar:
\r
412 if(visit == PostVisit)
\r
414 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
416 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
417 mul->dst.index += i;
\r
418 argument(mul->src[0], left, i);
\r
422 case EOpVectorTimesMatrix:
\r
423 if(visit == PostVisit)
\r
425 int size = leftType.getNominalSize();
\r
427 for(int i = 0; i < size; i++)
\r
429 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
430 dot->dst.mask = 1 << i;
\r
431 argument(dot->src[1], right, i);
\r
435 case EOpMatrixTimesVector:
\r
436 if(visit == PostVisit)
\r
438 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
439 mul->src[1].swizzle = 0x00;
\r
441 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
443 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
444 argument(mad->src[0], left, i);
\r
445 mad->src[1].swizzle = i * 0x55;
\r
449 case EOpMatrixTimesMatrix:
\r
450 if(visit == PostVisit)
\r
452 int dim = leftType.getNominalSize();
\r
454 for(int i = 0; i < dim; i++)
\r
456 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
457 mul->dst.index += i;
\r
458 argument(mul->src[1], right, i);
\r
459 mul->src[1].swizzle = 0x00;
\r
461 for(int j = 1; j < dim; j++)
\r
463 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
464 mad->dst.index += i;
\r
465 argument(mad->src[0], left, j);
\r
466 argument(mad->src[1], right, i);
\r
467 mad->src[1].swizzle = j * 0x55;
\r
468 argument(mad->src[2], result, i);
\r
474 if(trivial(right, 6))
\r
476 if(visit == PostVisit)
\r
478 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
481 else // Short-circuit evaluation
\r
483 if(visit == InVisit)
\r
485 emit(sw::Shader::OPCODE_MOV, result, left);
\r
486 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
487 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
489 else if(visit == PostVisit)
\r
491 emit(sw::Shader::OPCODE_MOV, result, right);
\r
492 emit(sw::Shader::OPCODE_ENDIF);
\r
496 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
497 case EOpLogicalAnd:
\r
498 if(trivial(right, 6))
\r
500 if(visit == PostVisit)
\r
502 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
505 else // Short-circuit evaluation
\r
507 if(visit == InVisit)
\r
509 emit(sw::Shader::OPCODE_MOV, result, left);
\r
510 emit(sw::Shader::OPCODE_IF, 0, result);
\r
512 else if(visit == PostVisit)
\r
514 emit(sw::Shader::OPCODE_MOV, result, right);
\r
515 emit(sw::Shader::OPCODE_ENDIF);
\r
519 default: UNREACHABLE(node->getOp());
\r
525 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
527 if(currentScope != emitScope)
\r
532 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
533 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
534 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
536 TIntermTyped *result = node;
\r
537 TIntermTyped *arg = node->getOperand();
\r
539 switch(node->getOp())
\r
542 if(visit == PostVisit)
\r
544 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
546 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
547 neg->dst.index += index;
\r
548 argument(neg->src[0], arg, index);
\r
549 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
553 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
554 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
555 case EOpPostIncrement:
\r
556 if(visit == PostVisit)
\r
560 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
562 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
563 add->dst.index += index;
\r
564 argument(add->src[0], arg, index);
\r
567 assignLvalue(arg, arg);
\r
570 case EOpPostDecrement:
\r
571 if(visit == PostVisit)
\r
575 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
577 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
578 sub->dst.index += index;
\r
579 argument(sub->src[0], arg, index);
\r
582 assignLvalue(arg, arg);
\r
585 case EOpPreIncrement:
\r
586 if(visit == PostVisit)
\r
588 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
590 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
591 add->dst.index += index;
\r
592 argument(add->src[0], arg, index);
\r
595 assignLvalue(arg, result);
\r
598 case EOpPreDecrement:
\r
599 if(visit == PostVisit)
\r
601 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
603 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
604 sub->dst.index += index;
\r
605 argument(sub->src[0], arg, index);
\r
608 assignLvalue(arg, result);
\r
611 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
612 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
613 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
614 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
615 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
616 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
617 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
618 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
619 case EOpSinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_SINH, result, arg); break;
\r
620 case EOpCosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_COSH, result, arg); break;
\r
621 case EOpTanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_TANH, result, arg); break;
\r
622 case EOpAsinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASINH, result, arg); break;
\r
623 case EOpAcosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOSH, result, arg); break;
\r
624 case EOpAtanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATANH, result, arg); break;
\r
625 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
626 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
627 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
628 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
629 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
630 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
631 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
632 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
633 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
634 case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
\r
635 case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
\r
636 case EOpRoundEven: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUNDEVEN, result, arg); break;
\r
637 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
638 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
639 case EOpIsNan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISNAN, result, arg); break;
\r
640 case EOpIsInf: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISINF, result, arg); break;
\r
641 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
642 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
643 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
644 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
645 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
646 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
647 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
649 if(visit == PostVisit)
\r
651 int numCols = arg->getNominalSize();
\r
652 int numRows = arg->getSecondarySize();
\r
653 for(int i = 0; i < numCols; ++i)
\r
655 for(int j = 0; j < numRows; ++j)
\r
657 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
658 mov->src[0].index += i;
\r
659 mov->src[0].swizzle = 0x55 * j;
\r
660 mov->dst.index += j;
\r
661 mov->dst.mask = 1 << i;
\r
666 default: UNREACHABLE(node->getOp());
\r
672 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
674 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
679 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
681 TIntermTyped *result = node;
\r
682 const TType &resultType = node->getType();
\r
683 TIntermSequence &arg = node->getSequence();
\r
684 int argumentCount = arg.size();
\r
686 switch(node->getOp())
\r
688 case EOpSequence: break;
\r
689 case EOpDeclaration: break;
\r
690 case EOpPrototype: break;
\r
692 if(visit == PostVisit)
\r
694 copy(result, arg[1]);
\r
698 if(visit == PreVisit)
\r
700 const TString &name = node->getName();
\r
702 if(emitScope == FUNCTION)
\r
704 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
706 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
707 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
709 const Function *function = findFunction(name);
\r
710 ASSERT(function); // Should have been added during global pass
\r
711 label->dst.index = function->label;
\r
712 currentFunction = function->label;
\r
715 else if(emitScope == GLOBAL)
\r
717 if(name != "main(")
\r
719 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
720 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
723 else UNREACHABLE(emitScope);
\r
725 currentScope = FUNCTION;
\r
727 else if(visit == PostVisit)
\r
729 if(emitScope == FUNCTION)
\r
731 if(functionArray.size() > 1) // No need to return when there's only main()
\r
733 emit(sw::Shader::OPCODE_RET);
\r
737 currentScope = GLOBAL;
\r
740 case EOpFunctionCall:
\r
741 if(visit == PostVisit)
\r
743 if(node->isUserDefined())
\r
745 const TString &name = node->getName();
\r
746 const Function *function = findFunction(name);
\r
750 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
754 TIntermSequence &arguments = *function->arg;
\r
756 for(int i = 0; i < argumentCount; i++)
\r
758 TIntermTyped *in = arguments[i]->getAsTyped();
\r
760 if(in->getQualifier() == EvqIn ||
\r
761 in->getQualifier() == EvqInOut ||
\r
762 in->getQualifier() == EvqConstReadOnly)
\r
768 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
769 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
770 call->dst.index = function->label;
\r
772 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
774 copy(result, function->ret);
\r
777 for(int i = 0; i < argumentCount; i++)
\r
779 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
780 TIntermTyped *out = arg[i]->getAsTyped();
\r
782 if(argument->getQualifier() == EvqOut ||
\r
783 argument->getQualifier() == EvqInOut)
\r
785 copy(out, argument);
\r
791 TString name = TFunction::unmangleName(node->getName());
\r
793 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
795 if(argumentCount == 2)
\r
797 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
799 else if(argumentCount == 3) // bias
\r
801 Temporary uvwb(this);
\r
802 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
803 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
804 bias->dst.mask = 0x8;
\r
806 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
809 else UNREACHABLE(argumentCount);
\r
811 else if(name == "texture2DProj")
\r
813 TIntermTyped *t = arg[1]->getAsTyped();
\r
815 if(argumentCount == 2)
\r
817 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
818 tex->project = true;
\r
820 if(t->getNominalSize() == 3)
\r
822 tex->src[0].swizzle = 0xA4;
\r
824 else ASSERT(t->getNominalSize() == 4);
\r
826 else if(argumentCount == 3) // bias
\r
828 Temporary proj(this);
\r
830 if(t->getNominalSize() == 3)
\r
832 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
833 div->src[1].swizzle = 0xAA;
\r
834 div->dst.mask = 0x3;
\r
836 else if(t->getNominalSize() == 4)
\r
838 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
839 div->src[1].swizzle = 0xFF;
\r
840 div->dst.mask = 0x3;
\r
842 else UNREACHABLE(t->getNominalSize());
\r
844 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
845 bias->dst.mask = 0x8;
\r
847 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
850 else UNREACHABLE(argumentCount);
\r
852 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
854 Temporary uvwb(this);
\r
855 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
856 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
857 lod->dst.mask = 0x8;
\r
859 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
861 else if(name == "texture2DProjLod")
\r
863 TIntermTyped *t = arg[1]->getAsTyped();
\r
864 Temporary proj(this);
\r
866 if(t->getNominalSize() == 3)
\r
868 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
869 div->src[1].swizzle = 0xAA;
\r
870 div->dst.mask = 0x3;
\r
872 else if(t->getNominalSize() == 4)
\r
874 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
875 div->src[1].swizzle = 0xFF;
\r
876 div->dst.mask = 0x3;
\r
878 else UNREACHABLE(t->getNominalSize());
\r
880 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
881 lod->dst.mask = 0x8;
\r
883 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
885 else UNREACHABLE(0);
\r
889 case EOpParameters:
\r
891 case EOpConstructFloat:
\r
892 case EOpConstructVec2:
\r
893 case EOpConstructVec3:
\r
894 case EOpConstructVec4:
\r
895 case EOpConstructBool:
\r
896 case EOpConstructBVec2:
\r
897 case EOpConstructBVec3:
\r
898 case EOpConstructBVec4:
\r
899 case EOpConstructInt:
\r
900 case EOpConstructIVec2:
\r
901 case EOpConstructIVec3:
\r
902 case EOpConstructIVec4:
\r
903 case EOpConstructUInt:
\r
904 case EOpConstructUVec2:
\r
905 case EOpConstructUVec3:
\r
906 case EOpConstructUVec4:
\r
907 if(visit == PostVisit)
\r
911 for(int i = 0; i < argumentCount; i++)
\r
913 TIntermTyped *argi = arg[i]->getAsTyped();
\r
914 int size = argi->getNominalSize();
\r
916 if(!argi->isMatrix())
\r
918 Instruction *mov = emitCast(result, argi);
\r
919 mov->dst.mask = (0xF << component) & 0xF;
\r
920 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
928 while(component < resultType.getNominalSize())
\r
930 Instruction *mov = emitCast(result, argi);
\r
931 mov->dst.mask = (0xF << component) & 0xF;
\r
932 mov->src[0].index += column;
\r
933 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
942 case EOpConstructMat2:
\r
943 case EOpConstructMat2x3:
\r
944 case EOpConstructMat2x4:
\r
945 case EOpConstructMat3x2:
\r
946 case EOpConstructMat3:
\r
947 case EOpConstructMat3x4:
\r
948 case EOpConstructMat4x2:
\r
949 case EOpConstructMat4x3:
\r
950 case EOpConstructMat4:
\r
951 if(visit == PostVisit)
\r
953 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
954 const int dim = result->getNominalSize();
\r
956 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
958 for(int i = 0; i < dim; i++)
\r
960 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
961 init->dst.index += i;
\r
962 Instruction *mov = emitCast(result, arg0);
\r
963 mov->dst.index += i;
\r
964 mov->dst.mask = 1 << i;
\r
965 ASSERT(mov->src[0].swizzle == 0x00);
\r
968 else if(arg0->isMatrix())
\r
970 for(int i = 0; i < dim; i++)
\r
972 if(dim > dim2(arg0))
\r
974 // Initialize to identity matrix
\r
975 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
976 Instruction *mov = emitCast(result, &col);
\r
977 mov->dst.index += i;
\r
982 Instruction *mov = emitCast(result, arg0);
\r
983 mov->dst.index += i;
\r
984 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
985 argument(mov->src[0], arg0, i);
\r
994 for(int i = 0; i < argumentCount; i++)
\r
996 TIntermTyped *argi = arg[i]->getAsTyped();
\r
997 int size = argi->getNominalSize();
\r
1000 while(element < size)
\r
1002 Instruction *mov = emitCast(result, argi);
\r
1003 mov->dst.index += column;
\r
1004 mov->dst.mask = (0xF << row) & 0xF;
\r
1005 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
1007 int end = row + size - element;
\r
1008 column = end >= dim ? column + 1 : column;
\r
1009 element = element + dim - row;
\r
1010 row = end >= dim ? 0 : end;
\r
1016 case EOpConstructStruct:
\r
1017 if(visit == PostVisit)
\r
1020 for(int i = 0; i < argumentCount; i++)
\r
1022 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1023 int size = argi->totalRegisterCount();
\r
1025 for(int index = 0; index < size; index++)
\r
1027 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
1028 mov->dst.index += index + offset;
\r
1029 mov->dst.mask = writeMask(result, offset + index);
\r
1030 argument(mov->src[0], argi, index);
\r
1037 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1038 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1039 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1040 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1041 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1042 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1043 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1044 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1045 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1046 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1047 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1049 if(visit == PostVisit)
\r
1051 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1052 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1055 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1056 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1057 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1058 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg[0]); break;
\r
1059 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg[0]); break;
\r
1060 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg[0]); break;
\r
1061 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg[0]); break;
\r
1062 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg[0]); break;
\r
1063 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg[0]); break;
\r
1064 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg[0]); break;
\r
1065 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg[0]); break;
\r
1066 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg[0]); break;
\r
1067 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg[0]); break;
\r
1068 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1069 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1070 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1071 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1072 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1073 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1075 if(visit == PostVisit)
\r
1077 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1079 for(int i = 0; i < dim2(arg[0]); i++)
\r
1081 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1082 mul->dst.index += i;
\r
1083 argument(mul->src[0], arg[0], i);
\r
1084 argument(mul->src[1], arg[1], i);
\r
1088 case EOpOuterProduct:
\r
1089 if(visit == PostVisit)
\r
1091 for(int i = 0; i < dim(arg[1]); i++)
\r
1093 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1094 mul->dst.index += i;
\r
1095 mul->src[1].swizzle = 0x55 * i;
\r
1099 default: UNREACHABLE(node->getOp());
\r
1105 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1107 if(currentScope != emitScope)
\r
1112 TIntermTyped *condition = node->getCondition();
\r
1113 TIntermNode *trueBlock = node->getTrueBlock();
\r
1114 TIntermNode *falseBlock = node->getFalseBlock();
\r
1115 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1117 condition->traverse(this);
\r
1119 if(node->usesTernaryOperator())
\r
1121 if(constantCondition)
\r
1123 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1127 trueBlock->traverse(this);
\r
1128 copy(node, trueBlock);
\r
1132 falseBlock->traverse(this);
\r
1133 copy(node, falseBlock);
\r
1136 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1138 trueBlock->traverse(this);
\r
1139 falseBlock->traverse(this);
\r
1140 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1144 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1148 trueBlock->traverse(this);
\r
1149 copy(node, trueBlock);
\r
1154 emit(sw::Shader::OPCODE_ELSE);
\r
1155 falseBlock->traverse(this);
\r
1156 copy(node, falseBlock);
\r
1159 emit(sw::Shader::OPCODE_ENDIF);
\r
1162 else // if/else statement
\r
1164 if(constantCondition)
\r
1166 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1172 trueBlock->traverse(this);
\r
1179 falseBlock->traverse(this);
\r
1185 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1189 trueBlock->traverse(this);
\r
1194 emit(sw::Shader::OPCODE_ELSE);
\r
1195 falseBlock->traverse(this);
\r
1198 emit(sw::Shader::OPCODE_ENDIF);
\r
1205 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1207 if(currentScope != emitScope)
\r
1212 unsigned int iterations = loopCount(node);
\r
1214 if(iterations == 0)
\r
1219 bool unroll = (iterations <= 4);
\r
1223 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1224 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1227 TIntermNode *init = node->getInit();
\r
1228 TIntermTyped *condition = node->getCondition();
\r
1229 TIntermTyped *expression = node->getExpression();
\r
1230 TIntermNode *body = node->getBody();
\r
1232 if(node->getType() == ELoopDoWhile)
\r
1234 Temporary iterate(this);
\r
1235 Constant True(true);
\r
1236 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1238 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1242 body->traverse(this);
\r
1245 emit(sw::Shader::OPCODE_TEST);
\r
1247 condition->traverse(this);
\r
1248 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1250 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1256 init->traverse(this);
\r
1261 for(unsigned int i = 0; i < iterations; i++)
\r
1263 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1267 body->traverse(this);
\r
1272 expression->traverse(this);
\r
1280 condition->traverse(this);
\r
1283 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1287 body->traverse(this);
\r
1290 emit(sw::Shader::OPCODE_TEST);
\r
1294 expression->traverse(this);
\r
1299 condition->traverse(this);
\r
1302 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1309 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1311 if(currentScope != emitScope)
\r
1316 switch(node->getFlowOp())
\r
1318 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1319 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1320 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1322 if(visit == PostVisit)
\r
1324 TIntermTyped *value = node->getExpression();
\r
1328 copy(functionArray[currentFunction].ret, value);
\r
1331 emit(sw::Shader::OPCODE_LEAVE);
\r
1334 default: UNREACHABLE(node->getFlowOp());
\r
1340 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1342 return operand && isSamplerRegister(operand->getType());
\r
1345 bool OutputASM::isSamplerRegister(const TType &type)
\r
1347 // A sampler register's qualifiers can be:
\r
1348 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1349 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1350 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1351 // the sampler has been copied and is no longer a sampler register.
\r
1352 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1355 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1357 if(isSamplerRegister(dst))
\r
1359 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1362 Instruction *instruction = new Instruction(op);
\r
1366 instruction->dst.type = registerType(dst);
\r
1367 instruction->dst.index = registerIndex(dst) + index;
\r
1368 instruction->dst.mask = writeMask(dst);
\r
1369 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1372 argument(instruction->src[0], src0, index);
\r
1373 argument(instruction->src[1], src1, index);
\r
1374 argument(instruction->src[2], src2, index);
\r
1376 shader->append(instruction);
\r
1378 return instruction;
\r
1381 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1383 // Integers are implemented as float
\r
1384 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1386 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1388 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1390 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1392 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1394 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1397 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1400 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1402 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1404 emit(op, dst, src0, src1, src2, index);
\r
1408 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1410 emitBinary(op, result, src0, src1);
\r
1411 assignLvalue(lhs, result);
\r
1414 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1416 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1417 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1419 Instruction *cmp = emit(opcode, dst, left, right);
\r
1420 cmp->control = cmpOp;
\r
1421 argument(cmp->src[0], left, index);
\r
1422 argument(cmp->src[1], right, index);
\r
1425 int componentCount(const TType &type, int registers)
\r
1427 if(registers == 0)
\r
1432 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1434 int index = registers / type.elementRegisterCount();
\r
1435 registers -= index * type.elementRegisterCount();
\r
1436 return index * type.getElementSize() + componentCount(type, registers);
\r
1439 if(type.isStruct())
\r
1441 const TFieldList& fields = type.getStruct()->fields();
\r
1444 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1446 const TType &fieldType = *((*field)->type());
\r
1448 if(fieldType.totalRegisterCount() <= registers)
\r
1450 registers -= fieldType.totalRegisterCount();
\r
1451 elements += fieldType.getObjectSize();
\r
1453 else // Register within this field
\r
1455 return elements + componentCount(fieldType, registers);
\r
1459 else if(type.isMatrix())
\r
1461 return registers * type.getSecondarySize();
\r
1468 int registerSize(const TType &type, int registers)
\r
1470 if(registers == 0)
\r
1472 if(type.isStruct())
\r
1474 return registerSize(*((*(type.getStruct()->fields().begin()))->type()), 0);
\r
1477 return type.isMatrix() ? type.getSecondarySize() : type.getNominalSize();
\r
1480 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1482 int index = registers / type.elementRegisterCount();
\r
1483 registers -= index * type.elementRegisterCount();
\r
1484 return registerSize(type, registers);
\r
1487 if(type.isStruct())
\r
1489 const TFieldList& fields = type.getStruct()->fields();
\r
1492 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1494 const TType &fieldType = *((*field)->type());
\r
1496 if(fieldType.totalRegisterCount() <= registers)
\r
1498 registers -= fieldType.totalRegisterCount();
\r
1499 elements += fieldType.getObjectSize();
\r
1501 else // Register within this field
\r
1503 return registerSize(fieldType, registers);
\r
1507 else if(type.isMatrix())
\r
1509 return registerSize(type, 0);
\r
1516 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1520 TIntermTyped *arg = argument->getAsTyped();
\r
1521 const TType &type = arg->getType();
\r
1522 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1524 int size = registerSize(type, index);
\r
1526 parameter.type = registerType(arg);
\r
1528 if(arg->getQualifier() == EvqConstExpr)
\r
1530 int component = componentCount(type, index);
\r
1531 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1533 for(int i = 0; i < 4; i++)
\r
1535 if(size == 1) // Replicate
\r
1537 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1541 parameter.value[i] = constants[component + i].getAsFloat();
\r
1545 parameter.value[i] = 0.0f;
\r
1551 parameter.index = registerIndex(arg) + index;
\r
1553 if(isSamplerRegister(arg))
\r
1555 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1559 TIntermTyped *left = binary->getLeft();
\r
1560 TIntermTyped *right = binary->getRight();
\r
1562 if(binary->getOp() == EOpIndexDirect)
\r
1564 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1566 else if(binary->getOp() == EOpIndexIndirect)
\r
1568 if(left->getArraySize() > 1)
\r
1570 parameter.rel.type = registerType(binary->getRight());
\r
1571 parameter.rel.index = registerIndex(binary->getRight());
\r
1572 parameter.rel.scale = 1;
\r
1573 parameter.rel.deterministic = true;
\r
1576 else if(binary->getOp() == EOpIndexDirectStruct)
\r
1578 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1580 else UNREACHABLE(binary->getOp());
\r
1585 if(!IsSampler(arg->getBasicType()))
\r
1587 parameter.swizzle = readSwizzle(arg, size);
\r
1592 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1594 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1596 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1597 mov->dst.index += index;
\r
1598 mov->dst.mask = writeMask(dst, index);
\r
1599 argument(mov->src[0], src, offset + index);
\r
1603 int swizzleElement(int swizzle, int index)
\r
1605 return (swizzle >> (index * 2)) & 0x03;
\r
1608 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1610 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1611 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1612 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1613 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1616 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1619 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1620 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
\r
1622 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1625 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1627 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1629 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1631 Temporary address(this);
\r
1632 lvalue(insert->dst, address, dst);
\r
1634 insert->src[0].type = insert->dst.type;
\r
1635 insert->src[0].index = insert->dst.index;
\r
1636 insert->src[0].rel = insert->dst.rel;
\r
1637 argument(insert->src[1], src);
\r
1638 argument(insert->src[2], binary->getRight());
\r
1640 shader->append(insert);
\r
1644 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1646 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1648 Temporary address(this);
\r
1649 int swizzle = lvalue(mov->dst, address, dst);
\r
1650 mov->dst.index += offset;
\r
1654 mov->dst.mask = writeMask(dst, offset);
\r
1657 argument(mov->src[0], src, offset);
\r
1658 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1660 shader->append(mov);
\r
1665 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1667 TIntermTyped *result = node;
\r
1668 TIntermBinary *binary = node->getAsBinaryNode();
\r
1669 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1673 TIntermTyped *left = binary->getLeft();
\r
1674 TIntermTyped *right = binary->getRight();
\r
1676 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1678 switch(binary->getOp())
\r
1680 case EOpIndexDirect:
\r
1682 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1684 if(left->isRegister())
\r
1686 int leftMask = dst.mask;
\r
1689 while((leftMask & dst.mask) == 0)
\r
1691 dst.mask = dst.mask << 1;
\r
1694 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1695 dst.mask = 1 << element;
\r
1699 else if(left->isArray() || left->isMatrix())
\r
1701 dst.index += rightIndex * result->totalRegisterCount();
\r
1704 else UNREACHABLE(0);
\r
1707 case EOpIndexIndirect:
\r
1709 if(left->isRegister())
\r
1711 // Requires INSERT instruction (handled by calling function)
\r
1713 else if(left->isArray() || left->isMatrix())
\r
1715 int scale = result->totalRegisterCount();
\r
1717 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1719 if(left->totalRegisterCount() > 1)
\r
1721 sw::Shader::SourceParameter relativeRegister;
\r
1722 argument(relativeRegister, right);
\r
1724 dst.rel.index = relativeRegister.index;
\r
1725 dst.rel.type = relativeRegister.type;
\r
1726 dst.rel.scale = scale;
\r
1727 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1730 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1734 Constant oldScale((int)dst.rel.scale);
\r
1735 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1736 mad->src[0].index = dst.rel.index;
\r
1737 mad->src[0].type = dst.rel.type;
\r
1741 Constant oldScale((int)dst.rel.scale);
\r
1742 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1743 mul->src[0].index = dst.rel.index;
\r
1744 mul->src[0].type = dst.rel.type;
\r
1746 Constant newScale(scale);
\r
1747 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1750 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1751 dst.rel.index = registerIndex(&address);
\r
1752 dst.rel.scale = 1;
\r
1754 else // Just add the new index to the address register
\r
1758 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1762 Constant newScale(scale);
\r
1763 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1767 else UNREACHABLE(0);
\r
1770 case EOpIndexDirectStruct:
\r
1772 const TFieldList& fields = left->getType().getStruct()->fields();
\r
1773 int index = right->getAsConstantUnion()->getIConst(0);
\r
1774 int fieldOffset = 0;
\r
1776 for(int i = 0; i < index; i++)
\r
1778 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
1781 dst.type = registerType(left);
\r
1782 dst.index += fieldOffset;
\r
1783 dst.mask = writeMask(right);
\r
1788 case EOpVectorSwizzle:
\r
1790 ASSERT(left->isRegister());
\r
1792 int leftMask = dst.mask;
\r
1795 int rightMask = 0;
\r
1797 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1799 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1801 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
1803 int element = swizzleElement(leftSwizzle, index);
\r
1804 rightMask = rightMask | (1 << element);
\r
1805 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1808 dst.mask = leftMask & rightMask;
\r
1814 UNREACHABLE(binary->getOp()); // Not an l-value operator
\r
1820 dst.type = registerType(symbol);
\r
1821 dst.index = registerIndex(symbol);
\r
1822 dst.mask = writeMask(symbol);
\r
1829 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1831 if(isSamplerRegister(operand))
\r
1833 return sw::Shader::PARAMETER_SAMPLER;
\r
1836 const TQualifier qualifier = operand->getQualifier();
\r
1837 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1839 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1840 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1842 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1844 outputQualifier = qualifier;
\r
1849 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1850 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1851 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1852 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1853 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1854 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1855 case EvqVertexIn: return sw::Shader::PARAMETER_INPUT;
\r
1856 case EvqFragmentOut: return sw::Shader::PARAMETER_COLOROUT;
\r
1857 case EvqVertexOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1858 case EvqFragmentIn: return sw::Shader::PARAMETER_INPUT;
\r
1859 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1860 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1861 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1862 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1863 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1864 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1865 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1866 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1867 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1868 case EvqInstanceID: return sw::Shader::PARAMETER_MISCTYPE;
\r
1869 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1870 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1871 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1872 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1873 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1874 default: UNREACHABLE(qualifier);
\r
1877 return sw::Shader::PARAMETER_VOID;
\r
1880 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1882 if(isSamplerRegister(operand))
\r
1884 return samplerRegister(operand);
\r
1887 switch(operand->getQualifier())
\r
1889 case EvqTemporary: return temporaryRegister(operand);
\r
1890 case EvqGlobal: return temporaryRegister(operand);
\r
1891 case EvqConstExpr: UNREACHABLE(EvqConstExpr);
\r
1892 case EvqAttribute: return attributeRegister(operand);
\r
1893 case EvqVaryingIn: return varyingRegister(operand);
\r
1894 case EvqVaryingOut: return varyingRegister(operand);
\r
1895 case EvqVertexIn: return attributeRegister(operand);
\r
1896 case EvqFragmentOut: return 0;
\r
1897 case EvqVertexOut: return varyingRegister(operand);
\r
1898 case EvqFragmentIn: return varyingRegister(operand);
\r
1899 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1900 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1901 case EvqUniform: return uniformRegister(operand);
\r
1902 case EvqIn: return temporaryRegister(operand);
\r
1903 case EvqOut: return temporaryRegister(operand);
\r
1904 case EvqInOut: return temporaryRegister(operand);
\r
1905 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1906 case EvqPosition: return varyingRegister(operand);
\r
1907 case EvqPointSize: return varyingRegister(operand);
\r
1908 case EvqInstanceID: vertexShader->instanceIdDeclared = true; return 0;
\r
1909 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1910 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1911 case EvqPointCoord: return varyingRegister(operand);
\r
1912 case EvqFragColor: return 0;
\r
1913 case EvqFragData: return 0;
\r
1914 default: UNREACHABLE(operand->getQualifier());
\r
1920 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1922 if(destination->getQualifier() == EvqPointSize)
\r
1924 return 0x2; // Point size stored in the y component
\r
1927 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1930 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1932 if(argument->getQualifier() == EvqPointSize)
\r
1934 return 0x55; // Point size stored in the y component
\r
1937 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1939 return swizzleSize[size];
\r
1942 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1943 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1945 if(!expression->isRegister())
\r
1950 return cost(expression, budget) >= 0;
\r
1953 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1954 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1961 if(expression->getAsSymbolNode())
\r
1965 else if(expression->getAsConstantUnion())
\r
1969 else if(expression->getAsBinaryNode())
\r
1971 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1973 switch(binary->getOp())
\r
1975 case EOpVectorSwizzle:
\r
1976 case EOpIndexDirect:
\r
1977 case EOpIndexDirectStruct:
\r
1978 return cost(binary->getLeft(), budget - 0);
\r
1982 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1987 else if(expression->getAsUnaryNode())
\r
1989 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1991 switch(unary->getOp())
\r
1995 return cost(unary->getOperand(), budget - 1);
\r
2000 else if(expression->getAsSelectionNode())
\r
2002 TIntermSelection *selection = expression->getAsSelectionNode();
\r
2004 if(selection->usesTernaryOperator())
\r
2006 TIntermTyped *condition = selection->getCondition();
\r
2007 TIntermNode *trueBlock = selection->getTrueBlock();
\r
2008 TIntermNode *falseBlock = selection->getFalseBlock();
\r
2009 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
2011 if(constantCondition)
\r
2013 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
2017 return cost(trueBlock, budget - 0);
\r
2021 return cost(falseBlock, budget - 0);
\r
2026 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
2034 const Function *OutputASM::findFunction(const TString &name)
\r
2036 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
2038 if(functionArray[f].name == name)
\r
2040 return &functionArray[f];
\r
2047 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
2049 return allocate(temporaries, temporary);
\r
2052 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
2054 int var = lookup(varyings, varying);
\r
2058 var = allocate(varyings, varying);
\r
2059 int componentCount = varying->getNominalSize();
\r
2060 int registerCount = varying->totalRegisterCount();
\r
2064 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2066 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2070 if(varying->getQualifier() == EvqPointCoord)
\r
2072 ASSERT(varying->isRegister());
\r
2073 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2074 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2075 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2076 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2080 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2082 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2083 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2084 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2085 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2089 else if(vertexShader)
\r
2091 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2093 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2097 if(varying->getQualifier() == EvqPosition)
\r
2099 ASSERT(varying->isRegister());
\r
2100 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2101 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2102 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2103 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2104 vertexShader->positionRegister = var;
\r
2106 else if(varying->getQualifier() == EvqPointSize)
\r
2108 ASSERT(varying->isRegister());
\r
2109 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2110 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2111 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2112 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2113 vertexShader->pointSizeRegister = var;
\r
2117 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2120 else UNREACHABLE(0);
\r
2122 declareVarying(varying, var);
\r
2128 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2130 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2132 const TType &type = varying->getType();
\r
2133 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2134 VaryingList &activeVaryings = shaderObject->varyings;
\r
2136 // Check if this varying has been declared before without having a register assigned
\r
2137 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2139 if(v->name == name)
\r
2143 ASSERT(v->reg < 0 || v->reg == reg);
\r
2151 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2155 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2157 const TType &type = uniform->getType();
\r
2158 ASSERT(!IsSampler(type.getBasicType()));
\r
2159 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2164 int index = lookup(uniforms, uniform);
\r
2168 index = allocate(uniforms, uniform);
\r
2169 const TString &name = symbol->getSymbol().c_str();
\r
2171 declareUniform(type, name, index);
\r
2180 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2182 ASSERT(!attribute->isArray());
\r
2183 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2185 int index = lookup(attributes, attribute);
\r
2189 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2194 index = allocate(attributes, attribute);
\r
2195 const TType &type = attribute->getType();
\r
2196 int registerCount = attribute->totalRegisterCount();
\r
2198 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2200 for(int i = 0; i < registerCount; i++)
\r
2202 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2206 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2208 const char *name = symbol->getSymbol().c_str();
\r
2209 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2216 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2218 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2219 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2220 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2224 return samplerRegister(symbol);
\r
2228 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect || binary->getOp() == EOpIndexDirectStruct);
\r
2230 return samplerRegister(binary->getLeft()); // Index added later
\r
2232 else UNREACHABLE(0);
\r
2237 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2239 const TType &type = sampler->getType();
\r
2240 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2242 int index = lookup(samplers, sampler);
\r
2246 index = allocate(samplers, sampler);
\r
2248 if(sampler->getQualifier() == EvqUniform)
\r
2250 const char *name = sampler->getSymbol().c_str();
\r
2251 declareUniform(type, name, index);
\r
2258 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2260 for(unsigned int i = 0; i < list.size(); i++)
\r
2262 if(list[i] == variable)
\r
2264 return i; // Pointer match
\r
2268 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2272 for(unsigned int i = 0; i < list.size(); i++)
\r
2276 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2280 if(listSymbol->getId() == varSymbol->getId())
\r
2282 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2283 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2284 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2296 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2298 int index = lookup(list, variable);
\r
2302 unsigned int registerCount = variable->totalRegisterCount();
\r
2304 for(unsigned int i = 0; i < list.size(); i++)
\r
2308 unsigned int j = 1;
\r
2309 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2311 if(list[i + j] != 0)
\r
2317 if(j == registerCount) // Found free slots
\r
2319 for(unsigned int j = 0; j < registerCount; j++)
\r
2321 list[i + j] = variable;
\r
2329 index = list.size();
\r
2331 for(unsigned int i = 0; i < registerCount; i++)
\r
2333 list.push_back(variable);
\r
2340 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2342 int index = lookup(list, variable);
\r
2350 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2352 const TStructure *structure = type.getStruct();
\r
2353 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2357 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2359 if(isSamplerRegister(type))
\r
2361 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2363 shader->declareSampler(index + i);
\r
2369 const TFieldList& fields = structure->fields();
\r
2370 if(type.isArray())
\r
2372 int elementIndex = index;
\r
2374 for(int i = 0; i < type.getArraySize(); i++)
\r
2376 for(size_t j = 0; j < fields.size(); j++)
\r
2378 const TType &fieldType = *(fields[j]->type());
\r
2379 const TString &fieldName = fields[j]->name();
\r
2381 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2382 declareUniform(fieldType, uniformName, elementIndex);
\r
2383 elementIndex += fieldType.totalRegisterCount();
\r
2389 int fieldIndex = index;
\r
2391 for(size_t i = 0; i < fields.size(); i++)
\r
2393 const TType &fieldType = *(fields[i]->type());
\r
2394 const TString &fieldName = fields[i]->name();
\r
2396 const TString uniformName = name + "." + fieldName;
\r
2397 declareUniform(fieldType, uniformName, fieldIndex);
\r
2398 fieldIndex += fieldType.totalRegisterCount();
\r
2404 GLenum OutputASM::glVariableType(const TType &type)
\r
2406 switch(type.getBasicType())
\r
2409 if(type.isScalar())
\r
2413 else if(type.isVector())
\r
2415 switch(type.getNominalSize())
\r
2417 case 2: return GL_FLOAT_VEC2;
\r
2418 case 3: return GL_FLOAT_VEC3;
\r
2419 case 4: return GL_FLOAT_VEC4;
\r
2420 default: UNREACHABLE(type.getNominalSize());
\r
2423 else if(type.isMatrix())
\r
2425 switch(type.getNominalSize())
\r
2428 switch(type.getSecondarySize())
\r
2430 case 2: return GL_FLOAT_MAT2;
\r
2431 case 3: return GL_FLOAT_MAT2x3;
\r
2432 case 4: return GL_FLOAT_MAT2x4;
\r
2433 default: UNREACHABLE(type.getSecondarySize());
\r
2436 switch(type.getSecondarySize())
\r
2438 case 2: return GL_FLOAT_MAT3x2;
\r
2439 case 3: return GL_FLOAT_MAT3;
\r
2440 case 4: return GL_FLOAT_MAT3x4;
\r
2441 default: UNREACHABLE(type.getSecondarySize());
\r
2444 switch(type.getSecondarySize())
\r
2446 case 2: return GL_FLOAT_MAT4x2;
\r
2447 case 3: return GL_FLOAT_MAT4x3;
\r
2448 case 4: return GL_FLOAT_MAT4;
\r
2449 default: UNREACHABLE(type.getSecondarySize());
\r
2451 default: UNREACHABLE(type.getNominalSize());
\r
2454 else UNREACHABLE(0);
\r
2457 if(type.isScalar())
\r
2461 else if(type.isVector())
\r
2463 switch(type.getNominalSize())
\r
2465 case 2: return GL_INT_VEC2;
\r
2466 case 3: return GL_INT_VEC3;
\r
2467 case 4: return GL_INT_VEC4;
\r
2468 default: UNREACHABLE(type.getNominalSize());
\r
2471 else UNREACHABLE(0);
\r
2474 if(type.isScalar())
\r
2476 return GL_UNSIGNED_INT;
\r
2478 else if(type.isVector())
\r
2480 switch(type.getNominalSize())
\r
2482 case 2: return GL_UNSIGNED_INT_VEC2;
\r
2483 case 3: return GL_UNSIGNED_INT_VEC3;
\r
2484 case 4: return GL_UNSIGNED_INT_VEC4;
\r
2485 default: UNREACHABLE(type.getNominalSize());
\r
2488 else UNREACHABLE(0);
\r
2491 if(type.isScalar())
\r
2495 else if(type.isVector())
\r
2497 switch(type.getNominalSize())
\r
2499 case 2: return GL_BOOL_VEC2;
\r
2500 case 3: return GL_BOOL_VEC3;
\r
2501 case 4: return GL_BOOL_VEC4;
\r
2502 default: UNREACHABLE(type.getNominalSize());
\r
2505 else UNREACHABLE(0);
\r
2507 case EbtSampler2D:
\r
2508 case EbtISampler2D:
\r
2509 case EbtUSampler2D:
\r
2510 return GL_SAMPLER_2D;
\r
2511 case EbtSamplerCube:
\r
2512 case EbtISamplerCube:
\r
2513 case EbtUSamplerCube:
\r
2514 return GL_SAMPLER_CUBE;
\r
2515 case EbtSamplerExternalOES:
\r
2516 return GL_SAMPLER_EXTERNAL_OES;
\r
2517 case EbtSampler3D:
\r
2518 case EbtISampler3D:
\r
2519 case EbtUSampler3D:
\r
2520 return GL_SAMPLER_3D_OES;
\r
2521 case EbtSampler2DArray:
\r
2522 case EbtISampler2DArray:
\r
2523 case EbtUSampler2DArray:
\r
2524 return GL_SAMPLER_2D_ARRAY;
\r
2526 UNREACHABLE(type.getBasicType());
\r
2533 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2535 if(type.getBasicType() == EbtFloat)
\r
2537 switch(type.getPrecision())
\r
2539 case EbpHigh: return GL_HIGH_FLOAT;
\r
2540 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2541 case EbpLow: return GL_LOW_FLOAT;
\r
2542 case EbpUndefined:
\r
2543 // Should be defined as the default precision by the parser
\r
2544 default: UNREACHABLE(type.getPrecision());
\r
2547 else if(type.getBasicType() == EbtInt)
\r
2549 switch(type.getPrecision())
\r
2551 case EbpHigh: return GL_HIGH_INT;
\r
2552 case EbpMedium: return GL_MEDIUM_INT;
\r
2553 case EbpLow: return GL_LOW_INT;
\r
2554 case EbpUndefined:
\r
2555 // Should be defined as the default precision by the parser
\r
2556 default: UNREACHABLE(type.getPrecision());
\r
2560 // Other types (boolean, sampler) don't have a precision
\r
2564 int OutputASM::dim(TIntermNode *v)
\r
2566 TIntermTyped *vector = v->getAsTyped();
\r
2567 ASSERT(vector && vector->isRegister());
\r
2568 return vector->getNominalSize();
\r
2571 int OutputASM::dim2(TIntermNode *m)
\r
2573 TIntermTyped *matrix = m->getAsTyped();
\r
2574 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2575 return matrix->getSecondarySize();
\r
2578 // Returns ~0 if no loop count could be determined
\r
2579 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2581 // Parse loops of the form:
\r
2582 // for(int index = initial; index [comparator] limit; index += increment)
\r
2583 TIntermSymbol *index = 0;
\r
2584 TOperator comparator = EOpNull;
\r
2587 int increment = 0;
\r
2589 // Parse index name and intial value
\r
2590 if(node->getInit())
\r
2592 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2596 TIntermSequence &sequence = init->getSequence();
\r
2597 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2599 if(variable && variable->getQualifier() == EvqTemporary)
\r
2601 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2603 if(assign->getOp() == EOpInitialize)
\r
2605 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2606 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2608 if(symbol && constant)
\r
2610 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2613 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2621 // Parse comparator and limit value
\r
2622 if(index && node->getCondition())
\r
2624 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2626 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2628 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2632 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2634 comparator = test->getOp();
\r
2635 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2641 // Parse increment
\r
2642 if(index && comparator != EOpNull && node->getExpression())
\r
2644 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2645 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2647 if(binaryTerminal)
\r
2649 TOperator op = binaryTerminal->getOp();
\r
2650 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2654 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2656 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2660 case EOpAddAssign: increment = value; break;
\r
2661 case EOpSubAssign: increment = -value; break;
\r
2662 default: UNIMPLEMENTED();
\r
2667 else if(unaryTerminal)
\r
2669 TOperator op = unaryTerminal->getOp();
\r
2673 case EOpPostIncrement: increment = 1; break;
\r
2674 case EOpPostDecrement: increment = -1; break;
\r
2675 case EOpPreIncrement: increment = 1; break;
\r
2676 case EOpPreDecrement: increment = -1; break;
\r
2677 default: UNIMPLEMENTED();
\r
2682 if(index && comparator != EOpNull && increment != 0)
\r
2684 if(comparator == EOpLessThanEqual)
\r
2686 comparator = EOpLessThan;
\r
2690 if(comparator == EOpLessThan)
\r
2692 int iterations = (limit - initial) / increment;
\r
2694 if(iterations <= 0)
\r
2699 return iterations;
\r
2701 else UNIMPLEMENTED(); // Falls through
\r
2707 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2710 loopDiscontinuity = false;
\r
2712 node->traverse(this);
\r
2714 return loopDiscontinuity;
\r
2717 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2719 if(visit == PreVisit)
\r
2723 else if(visit == PostVisit)
\r
2731 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2733 if(loopDiscontinuity)
\r
2743 switch(node->getFlowOp())
\r
2750 loopDiscontinuity = true;
\r
2752 default: UNREACHABLE(node->getFlowOp());
\r
2755 return !loopDiscontinuity;
\r
2758 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2760 return !loopDiscontinuity;
\r