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
656 if(visit == PostVisit)
\r
658 int numCols = arg->getNominalSize();
\r
659 int numRows = arg->getSecondarySize();
\r
660 for(int i = 0; i < numCols; ++i)
\r
662 for(int j = 0; j < numRows; ++j)
\r
664 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
665 mov->src[0].index += i;
\r
666 mov->src[0].swizzle = 0x55 * j;
\r
667 mov->dst.index += j;
\r
668 mov->dst.mask = 1 << i;
\r
673 default: UNREACHABLE();
\r
679 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
681 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
686 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
688 TIntermTyped *result = node;
\r
689 const TType &resultType = node->getType();
\r
690 TIntermSequence &arg = node->getSequence();
\r
691 int argumentCount = arg.size();
\r
693 switch(node->getOp())
\r
695 case EOpSequence: break;
\r
696 case EOpDeclaration: break;
\r
697 case EOpPrototype: break;
\r
699 if(visit == PostVisit)
\r
701 copy(result, arg[1]);
\r
705 if(visit == PreVisit)
\r
707 const TString &name = node->getName();
\r
709 if(emitScope == FUNCTION)
\r
711 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
713 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
714 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
716 const Function *function = findFunction(name);
\r
717 ASSERT(function); // Should have been added during global pass
\r
718 label->dst.index = function->label;
\r
719 currentFunction = function->label;
\r
722 else if(emitScope == GLOBAL)
\r
724 if(name != "main(")
\r
726 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
727 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
730 else UNREACHABLE();
\r
732 currentScope = FUNCTION;
\r
734 else if(visit == PostVisit)
\r
736 if(emitScope == FUNCTION)
\r
738 if(functionArray.size() > 1) // No need to return when there's only main()
\r
740 emit(sw::Shader::OPCODE_RET);
\r
744 currentScope = GLOBAL;
\r
747 case EOpFunctionCall:
\r
748 if(visit == PostVisit)
\r
750 if(node->isUserDefined())
\r
752 const TString &name = node->getName();
\r
753 const Function *function = findFunction(name);
\r
757 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
761 TIntermSequence &arguments = *function->arg;
\r
763 for(int i = 0; i < argumentCount; i++)
\r
765 TIntermTyped *in = arguments[i]->getAsTyped();
\r
767 if(in->getQualifier() == EvqIn ||
\r
768 in->getQualifier() == EvqInOut ||
\r
769 in->getQualifier() == EvqConstReadOnly)
\r
775 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
776 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
777 call->dst.index = function->label;
\r
779 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
781 copy(result, function->ret);
\r
784 for(int i = 0; i < argumentCount; i++)
\r
786 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
787 TIntermTyped *out = arg[i]->getAsTyped();
\r
789 if(argument->getQualifier() == EvqOut ||
\r
790 argument->getQualifier() == EvqInOut)
\r
792 copy(out, argument);
\r
798 TString name = TFunction::unmangleName(node->getName());
\r
800 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
802 if(argumentCount == 2)
\r
804 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
806 else if(argumentCount == 3) // bias
\r
808 Temporary uvwb(this);
\r
809 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
810 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
811 bias->dst.mask = 0x8;
\r
813 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
816 else UNREACHABLE();
\r
818 else if(name == "texture2DProj")
\r
820 TIntermTyped *t = arg[1]->getAsTyped();
\r
822 if(argumentCount == 2)
\r
824 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
825 tex->project = true;
\r
827 if(t->getNominalSize() == 3)
\r
829 tex->src[0].swizzle = 0xA4;
\r
831 else ASSERT(t->getNominalSize() == 4);
\r
833 else if(argumentCount == 3) // bias
\r
835 Temporary proj(this);
\r
837 if(t->getNominalSize() == 3)
\r
839 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
840 div->src[1].swizzle = 0xAA;
\r
841 div->dst.mask = 0x3;
\r
843 else if(t->getNominalSize() == 4)
\r
845 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
846 div->src[1].swizzle = 0xFF;
\r
847 div->dst.mask = 0x3;
\r
849 else UNREACHABLE();
\r
851 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
852 bias->dst.mask = 0x8;
\r
854 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
857 else UNREACHABLE();
\r
859 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
861 Temporary uvwb(this);
\r
862 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
863 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
864 lod->dst.mask = 0x8;
\r
866 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
868 else if(name == "texture2DProjLod")
\r
870 TIntermTyped *t = arg[1]->getAsTyped();
\r
871 Temporary proj(this);
\r
873 if(t->getNominalSize() == 3)
\r
875 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
876 div->src[1].swizzle = 0xAA;
\r
877 div->dst.mask = 0x3;
\r
879 else if(t->getNominalSize() == 4)
\r
881 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
882 div->src[1].swizzle = 0xFF;
\r
883 div->dst.mask = 0x3;
\r
885 else UNREACHABLE();
\r
887 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
888 lod->dst.mask = 0x8;
\r
890 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
892 else UNREACHABLE();
\r
896 case EOpParameters:
\r
898 case EOpConstructFloat:
\r
899 case EOpConstructVec2:
\r
900 case EOpConstructVec3:
\r
901 case EOpConstructVec4:
\r
902 case EOpConstructBool:
\r
903 case EOpConstructBVec2:
\r
904 case EOpConstructBVec3:
\r
905 case EOpConstructBVec4:
\r
906 case EOpConstructInt:
\r
907 case EOpConstructIVec2:
\r
908 case EOpConstructIVec3:
\r
909 case EOpConstructIVec4:
\r
910 case EOpConstructUInt:
\r
911 case EOpConstructUVec2:
\r
912 case EOpConstructUVec3:
\r
913 case EOpConstructUVec4:
\r
914 if(visit == PostVisit)
\r
918 for(int i = 0; i < argumentCount; i++)
\r
920 TIntermTyped *argi = arg[i]->getAsTyped();
\r
921 int size = argi->getNominalSize();
\r
923 if(!argi->isMatrix())
\r
925 Instruction *mov = emitCast(result, argi);
\r
926 mov->dst.mask = (0xF << component) & 0xF;
\r
927 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
935 while(component < resultType.getNominalSize())
\r
937 Instruction *mov = emitCast(result, argi);
\r
938 mov->dst.mask = (0xF << component) & 0xF;
\r
939 mov->src[0].index += column;
\r
940 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
949 case EOpConstructMat2:
\r
950 case EOpConstructMat2x3:
\r
951 case EOpConstructMat2x4:
\r
952 case EOpConstructMat3x2:
\r
953 case EOpConstructMat3:
\r
954 case EOpConstructMat3x4:
\r
955 case EOpConstructMat4x2:
\r
956 case EOpConstructMat4x3:
\r
957 case EOpConstructMat4:
\r
958 if(visit == PostVisit)
\r
960 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
961 const int dim = result->getNominalSize();
\r
963 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
965 for(int i = 0; i < dim; i++)
\r
967 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
968 init->dst.index += i;
\r
969 Instruction *mov = emitCast(result, arg0);
\r
970 mov->dst.index += i;
\r
971 mov->dst.mask = 1 << i;
\r
972 ASSERT(mov->src[0].swizzle == 0x00);
\r
975 else if(arg0->isMatrix())
\r
977 for(int i = 0; i < dim; i++)
\r
979 if(dim > dim2(arg0))
\r
981 // Initialize to identity matrix
\r
982 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
983 Instruction *mov = emitCast(result, &col);
\r
984 mov->dst.index += i;
\r
989 Instruction *mov = emitCast(result, arg0);
\r
990 mov->dst.index += i;
\r
991 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
992 argument(mov->src[0], arg0, i);
\r
1001 for(int i = 0; i < argumentCount; i++)
\r
1003 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1004 int size = argi->getNominalSize();
\r
1007 while(element < size)
\r
1009 Instruction *mov = emitCast(result, argi);
\r
1010 mov->dst.index += column;
\r
1011 mov->dst.mask = (0xF << row) & 0xF;
\r
1012 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
1014 int end = row + size - element;
\r
1015 column = end >= dim ? column + 1 : column;
\r
1016 element = element + dim - row;
\r
1017 row = end >= dim ? 0 : end;
\r
1023 case EOpConstructStruct:
\r
1024 if(visit == PostVisit)
\r
1027 for(int i = 0; i < argumentCount; i++)
\r
1029 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1030 int size = argi->totalRegisterCount();
\r
1032 for(int index = 0; index < size; index++)
\r
1034 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
1035 mov->dst.index += index + offset;
\r
1036 mov->dst.mask = writeMask(result, offset + index);
\r
1037 argument(mov->src[0], argi, index);
\r
1044 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1045 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1046 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1047 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1048 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1049 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1050 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1051 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1052 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1053 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1054 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1056 if(visit == PostVisit)
\r
1058 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1059 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1062 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1063 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1064 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1065 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg[0]); break;
\r
1066 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg[0]); break;
\r
1067 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg[0]); break;
\r
1068 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg[0]); break;
\r
1069 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg[0]); break;
\r
1070 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg[0]); break;
\r
1071 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg[0]); break;
\r
1072 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg[0]); break;
\r
1073 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg[0]); break;
\r
1074 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg[0]); break;
\r
1075 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1076 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1077 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1078 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1079 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1080 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1082 if(visit == PostVisit)
\r
1084 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1086 for(int i = 0; i < dim2(arg[0]); i++)
\r
1088 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1089 mul->dst.index += i;
\r
1090 argument(mul->src[0], arg[0], i);
\r
1091 argument(mul->src[1], arg[1], i);
\r
1095 case EOpOuterProduct:
\r
1096 if(visit == PostVisit)
\r
1098 for(int i = 0; i < dim(arg[1]); i++)
\r
1100 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1101 mul->dst.index += i;
\r
1102 mul->src[1].swizzle = 0x55 * i;
\r
1106 default: UNREACHABLE();
\r
1112 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1114 if(currentScope != emitScope)
\r
1119 TIntermTyped *condition = node->getCondition();
\r
1120 TIntermNode *trueBlock = node->getTrueBlock();
\r
1121 TIntermNode *falseBlock = node->getFalseBlock();
\r
1122 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1124 condition->traverse(this);
\r
1126 if(node->usesTernaryOperator())
\r
1128 if(constantCondition)
\r
1130 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1134 trueBlock->traverse(this);
\r
1135 copy(node, trueBlock);
\r
1139 falseBlock->traverse(this);
\r
1140 copy(node, falseBlock);
\r
1143 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1145 trueBlock->traverse(this);
\r
1146 falseBlock->traverse(this);
\r
1147 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1151 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1155 trueBlock->traverse(this);
\r
1156 copy(node, trueBlock);
\r
1161 emit(sw::Shader::OPCODE_ELSE);
\r
1162 falseBlock->traverse(this);
\r
1163 copy(node, falseBlock);
\r
1166 emit(sw::Shader::OPCODE_ENDIF);
\r
1169 else // if/else statement
\r
1171 if(constantCondition)
\r
1173 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1179 trueBlock->traverse(this);
\r
1186 falseBlock->traverse(this);
\r
1192 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1196 trueBlock->traverse(this);
\r
1201 emit(sw::Shader::OPCODE_ELSE);
\r
1202 falseBlock->traverse(this);
\r
1205 emit(sw::Shader::OPCODE_ENDIF);
\r
1212 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1214 if(currentScope != emitScope)
\r
1219 unsigned int iterations = loopCount(node);
\r
1221 if(iterations == 0)
\r
1226 bool unroll = (iterations <= 4);
\r
1230 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1231 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1234 TIntermNode *init = node->getInit();
\r
1235 TIntermTyped *condition = node->getCondition();
\r
1236 TIntermTyped *expression = node->getExpression();
\r
1237 TIntermNode *body = node->getBody();
\r
1239 if(node->getType() == ELoopDoWhile)
\r
1241 Temporary iterate(this);
\r
1242 Constant True(true);
\r
1243 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1245 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1249 body->traverse(this);
\r
1252 emit(sw::Shader::OPCODE_TEST);
\r
1254 condition->traverse(this);
\r
1255 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1257 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1263 init->traverse(this);
\r
1268 for(unsigned int i = 0; i < iterations; i++)
\r
1270 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1274 body->traverse(this);
\r
1279 expression->traverse(this);
\r
1285 condition->traverse(this);
\r
1287 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1291 body->traverse(this);
\r
1294 emit(sw::Shader::OPCODE_TEST);
\r
1298 expression->traverse(this);
\r
1301 condition->traverse(this);
\r
1303 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1310 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1312 if(currentScope != emitScope)
\r
1317 switch(node->getFlowOp())
\r
1319 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1320 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1321 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1323 if(visit == PostVisit)
\r
1325 TIntermTyped *value = node->getExpression();
\r
1329 copy(functionArray[currentFunction].ret, value);
\r
1332 emit(sw::Shader::OPCODE_LEAVE);
\r
1335 default: UNREACHABLE();
\r
1341 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1343 return operand && isSamplerRegister(operand->getType());
\r
1346 bool OutputASM::isSamplerRegister(const TType &type)
\r
1348 // A sampler register's qualifiers can be:
\r
1349 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1350 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1351 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1352 // the sampler has been copied and is no longer a sampler register.
\r
1353 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1356 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1358 if(isSamplerRegister(dst))
\r
1360 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1363 Instruction *instruction = new Instruction(op);
\r
1367 instruction->dst.type = registerType(dst);
\r
1368 instruction->dst.index = registerIndex(dst) + index;
\r
1369 instruction->dst.mask = writeMask(dst);
\r
1370 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1373 argument(instruction->src[0], src0, index);
\r
1374 argument(instruction->src[1], src1, index);
\r
1375 argument(instruction->src[2], src2, index);
\r
1377 shader->append(instruction);
\r
1379 return instruction;
\r
1382 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1384 // Integers are implemented as float
\r
1385 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1387 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1389 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1391 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1393 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1395 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1398 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1401 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1403 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1405 emit(op, dst, src0, src1, src2, index);
\r
1409 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1411 emitBinary(op, result, src0, src1);
\r
1412 assignLvalue(lhs, result);
\r
1415 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1417 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1418 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1420 Instruction *cmp = emit(opcode, dst, left, right);
\r
1421 cmp->control = cmpOp;
\r
1422 argument(cmp->src[0], left, index);
\r
1423 argument(cmp->src[1], right, index);
\r
1426 int componentCount(const TType &type, int registers)
\r
1428 if(registers == 0)
\r
1433 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1435 int index = registers / type.elementRegisterCount();
\r
1436 registers -= index * type.elementRegisterCount();
\r
1437 return index * type.getElementSize() + componentCount(type, registers);
\r
1440 if(type.isStruct())
\r
1442 TTypeList *structure = type.getStruct();
\r
1445 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1447 const TType &fieldType = *field->type;
\r
1449 if(fieldType.totalRegisterCount() <= registers)
\r
1451 registers -= fieldType.totalRegisterCount();
\r
1452 elements += fieldType.getObjectSize();
\r
1454 else // Register within this field
\r
1456 return elements + componentCount(fieldType, registers);
\r
1460 else if(type.isMatrix())
\r
1462 return registers * type.getSecondarySize();
\r
1469 int registerSize(const TType &type, int registers)
\r
1471 if(registers == 0)
\r
1473 if(type.isStruct())
\r
1475 return registerSize(*type.getStruct()->begin()->type, 0);
\r
1478 return type.isMatrix() ? type.getSecondarySize() : type.getNominalSize();
\r
1481 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1483 int index = registers / type.elementRegisterCount();
\r
1484 registers -= index * type.elementRegisterCount();
\r
1485 return registerSize(type, registers);
\r
1488 if(type.isStruct())
\r
1490 TTypeList *structure = type.getStruct();
\r
1493 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1495 const TType &fieldType = *field->type;
\r
1497 if(fieldType.totalRegisterCount() <= registers)
\r
1499 registers -= fieldType.totalRegisterCount();
\r
1500 elements += fieldType.getObjectSize();
\r
1502 else // Register within this field
\r
1504 return registerSize(fieldType, registers);
\r
1508 else if(type.isMatrix())
\r
1510 return registerSize(type, 0);
\r
1517 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1521 TIntermTyped *arg = argument->getAsTyped();
\r
1522 const TType &type = arg->getType();
\r
1523 const TTypeList *structure = type.getStruct();
\r
1524 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1526 int size = registerSize(type, index);
\r
1528 parameter.type = registerType(arg);
\r
1530 if(arg->getQualifier() == EvqConstExpr)
\r
1532 int component = componentCount(type, index);
\r
1533 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1535 for(int i = 0; i < 4; i++)
\r
1537 if(size == 1) // Replicate
\r
1539 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1543 parameter.value[i] = constants[component + i].getAsFloat();
\r
1547 parameter.value[i] = 0.0f;
\r
1553 parameter.index = registerIndex(arg) + index;
\r
1555 if(isSamplerRegister(arg))
\r
1557 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1561 TIntermTyped *left = binary->getLeft();
\r
1562 TIntermTyped *right = binary->getRight();
\r
1564 if(binary->getOp() == EOpIndexDirect)
\r
1566 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1568 else if(binary->getOp() == EOpIndexIndirect)
\r
1570 if(left->getArraySize() > 1)
\r
1572 parameter.rel.type = registerType(binary->getRight());
\r
1573 parameter.rel.index = registerIndex(binary->getRight());
\r
1574 parameter.rel.scale = 1;
\r
1575 parameter.rel.deterministic = true;
\r
1578 else if(binary->getOp() == EOpIndexDirectStruct)
\r
1580 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1582 else UNREACHABLE();
\r
1587 if(!IsSampler(arg->getBasicType()))
\r
1589 parameter.swizzle = readSwizzle(arg, size);
\r
1594 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1596 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1598 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1599 mov->dst.index += index;
\r
1600 mov->dst.mask = writeMask(dst, index);
\r
1601 argument(mov->src[0], src, offset + index);
\r
1605 int swizzleElement(int swizzle, int index)
\r
1607 return (swizzle >> (index * 2)) & 0x03;
\r
1610 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1612 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1613 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1614 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1615 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1618 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1621 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1622 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
\r
1624 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1627 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1629 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1631 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1633 Temporary address(this);
\r
1634 lvalue(insert->dst, address, dst);
\r
1636 insert->src[0].type = insert->dst.type;
\r
1637 insert->src[0].index = insert->dst.index;
\r
1638 insert->src[0].rel = insert->dst.rel;
\r
1639 argument(insert->src[1], src);
\r
1640 argument(insert->src[2], binary->getRight());
\r
1642 shader->append(insert);
\r
1646 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1648 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1650 Temporary address(this);
\r
1651 int swizzle = lvalue(mov->dst, address, dst);
\r
1652 mov->dst.index += offset;
\r
1656 mov->dst.mask = writeMask(dst, offset);
\r
1659 argument(mov->src[0], src, offset);
\r
1660 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1662 shader->append(mov);
\r
1667 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1669 TIntermTyped *result = node;
\r
1670 TIntermBinary *binary = node->getAsBinaryNode();
\r
1671 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1675 TIntermTyped *left = binary->getLeft();
\r
1676 TIntermTyped *right = binary->getRight();
\r
1678 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1680 switch(binary->getOp())
\r
1682 case EOpIndexDirect:
\r
1684 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1686 if(left->isRegister())
\r
1688 int leftMask = dst.mask;
\r
1691 while((leftMask & dst.mask) == 0)
\r
1693 dst.mask = dst.mask << 1;
\r
1696 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1697 dst.mask = 1 << element;
\r
1701 else if(left->isArray() || left->isMatrix())
\r
1703 dst.index += rightIndex * result->totalRegisterCount();
\r
1706 else UNREACHABLE();
\r
1709 case EOpIndexIndirect:
\r
1711 if(left->isRegister())
\r
1713 // Requires INSERT instruction (handled by calling function)
\r
1715 else if(left->isArray() || left->isMatrix())
\r
1717 int scale = result->totalRegisterCount();
\r
1719 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1721 if(left->totalRegisterCount() > 1)
\r
1723 sw::Shader::SourceParameter relativeRegister;
\r
1724 argument(relativeRegister, right);
\r
1726 dst.rel.index = relativeRegister.index;
\r
1727 dst.rel.type = relativeRegister.type;
\r
1728 dst.rel.scale = scale;
\r
1729 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1732 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1736 Constant oldScale((int)dst.rel.scale);
\r
1737 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1738 mad->src[0].index = dst.rel.index;
\r
1739 mad->src[0].type = dst.rel.type;
\r
1743 Constant oldScale((int)dst.rel.scale);
\r
1744 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1745 mul->src[0].index = dst.rel.index;
\r
1746 mul->src[0].type = dst.rel.type;
\r
1748 Constant newScale(scale);
\r
1749 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1752 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1753 dst.rel.index = registerIndex(&address);
\r
1754 dst.rel.scale = 1;
\r
1756 else // Just add the new index to the address register
\r
1760 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1764 Constant newScale(scale);
\r
1765 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1769 else UNREACHABLE();
\r
1772 case EOpIndexDirectStruct:
\r
1774 const TTypeList *structure = left->getType().getStruct();
\r
1775 const TString &fieldName = right->getType().getFieldName();
\r
1778 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1780 if(field->type->getFieldName() == fieldName)
\r
1782 dst.type = registerType(left);
\r
1783 dst.index += offset;
\r
1784 dst.mask = writeMask(right);
\r
1789 offset += field->type->totalRegisterCount();
\r
1793 case EOpVectorSwizzle:
\r
1795 ASSERT(left->isRegister());
\r
1797 int leftMask = dst.mask;
\r
1800 int rightMask = 0;
\r
1802 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1804 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1806 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
1808 int element = swizzleElement(leftSwizzle, index);
\r
1809 rightMask = rightMask | (1 << element);
\r
1810 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1813 dst.mask = leftMask & rightMask;
\r
1819 UNREACHABLE(); // Not an l-value operator
\r
1825 dst.type = registerType(symbol);
\r
1826 dst.index = registerIndex(symbol);
\r
1827 dst.mask = writeMask(symbol);
\r
1834 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1836 if(isSamplerRegister(operand))
\r
1838 return sw::Shader::PARAMETER_SAMPLER;
\r
1841 const TQualifier qualifier = operand->getQualifier();
\r
1842 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1844 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1845 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1847 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1849 outputQualifier = qualifier;
\r
1854 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1855 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1856 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1857 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1858 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1859 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1860 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1861 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1862 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1863 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1864 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1865 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1866 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1867 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1868 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1869 case EvqInstanceID: return sw::Shader::PARAMETER_MISCTYPE;
\r
1870 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1871 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1872 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1873 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1874 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1875 default: UNREACHABLE();
\r
1878 return sw::Shader::PARAMETER_VOID;
\r
1881 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1883 if(isSamplerRegister(operand))
\r
1885 return samplerRegister(operand);
\r
1888 switch(operand->getQualifier())
\r
1890 case EvqTemporary: return temporaryRegister(operand);
\r
1891 case EvqGlobal: return temporaryRegister(operand);
\r
1892 case EvqConstExpr: UNREACHABLE();
\r
1893 case EvqAttribute: return attributeRegister(operand);
\r
1894 case EvqVaryingIn: return varyingRegister(operand);
\r
1895 case EvqVaryingOut: return varyingRegister(operand);
\r
1896 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1897 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1898 case EvqUniform: return uniformRegister(operand);
\r
1899 case EvqIn: return temporaryRegister(operand);
\r
1900 case EvqOut: return temporaryRegister(operand);
\r
1901 case EvqInOut: return temporaryRegister(operand);
\r
1902 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1903 case EvqPosition: return varyingRegister(operand);
\r
1904 case EvqPointSize: return varyingRegister(operand);
\r
1905 case EvqInstanceID: vertexShader->instanceIdDeclared = true; return 0;
\r
1906 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1907 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1908 case EvqPointCoord: return varyingRegister(operand);
\r
1909 case EvqFragColor: return 0;
\r
1910 case EvqFragData: return 0;
\r
1911 default: UNREACHABLE();
\r
1917 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1919 if(destination->getQualifier() == EvqPointSize)
\r
1921 return 0x2; // Point size stored in the y component
\r
1924 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1927 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1929 if(argument->getQualifier() == EvqPointSize)
\r
1931 return 0x55; // Point size stored in the y component
\r
1934 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1936 return swizzleSize[size];
\r
1939 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1940 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1942 if(!expression->isRegister())
\r
1947 return cost(expression, budget) >= 0;
\r
1950 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1951 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1958 if(expression->getAsSymbolNode())
\r
1962 else if(expression->getAsConstantUnion())
\r
1966 else if(expression->getAsBinaryNode())
\r
1968 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1970 switch(binary->getOp())
\r
1972 case EOpVectorSwizzle:
\r
1973 case EOpIndexDirect:
\r
1974 case EOpIndexDirectStruct:
\r
1975 return cost(binary->getLeft(), budget - 0);
\r
1979 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1984 else if(expression->getAsUnaryNode())
\r
1986 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1988 switch(unary->getOp())
\r
1992 return cost(unary->getOperand(), budget - 1);
\r
1997 else if(expression->getAsSelectionNode())
\r
1999 TIntermSelection *selection = expression->getAsSelectionNode();
\r
2001 if(selection->usesTernaryOperator())
\r
2003 TIntermTyped *condition = selection->getCondition();
\r
2004 TIntermNode *trueBlock = selection->getTrueBlock();
\r
2005 TIntermNode *falseBlock = selection->getFalseBlock();
\r
2006 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
2008 if(constantCondition)
\r
2010 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
2014 return cost(trueBlock, budget - 0);
\r
2018 return cost(falseBlock, budget - 0);
\r
2023 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
2031 const Function *OutputASM::findFunction(const TString &name)
\r
2033 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
2035 if(functionArray[f].name == name)
\r
2037 return &functionArray[f];
\r
2044 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
2046 return allocate(temporaries, temporary);
\r
2049 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
2051 int var = lookup(varyings, varying);
\r
2055 var = allocate(varyings, varying);
\r
2056 int componentCount = varying->getNominalSize();
\r
2057 int registerCount = varying->totalRegisterCount();
\r
2061 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2063 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2067 if(varying->getQualifier() == EvqPointCoord)
\r
2069 ASSERT(varying->isRegister());
\r
2070 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2071 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2072 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2073 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2077 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2079 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2080 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2081 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2082 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2086 else if(vertexShader)
\r
2088 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2090 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2094 if(varying->getQualifier() == EvqPosition)
\r
2096 ASSERT(varying->isRegister());
\r
2097 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2098 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2099 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2100 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2101 vertexShader->positionRegister = var;
\r
2103 else if(varying->getQualifier() == EvqPointSize)
\r
2105 ASSERT(varying->isRegister());
\r
2106 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2107 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2108 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2109 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2110 vertexShader->pointSizeRegister = var;
\r
2114 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2117 else UNREACHABLE();
\r
2119 declareVarying(varying, var);
\r
2125 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2127 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2129 const TType &type = varying->getType();
\r
2130 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2131 VaryingList &activeVaryings = shaderObject->varyings;
\r
2133 // Check if this varying has been declared before without having a register assigned
\r
2134 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2136 if(v->name == name)
\r
2140 ASSERT(v->reg < 0 || v->reg == reg);
\r
2148 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2152 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2154 const TType &type = uniform->getType();
\r
2155 ASSERT(!IsSampler(type.getBasicType()));
\r
2156 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2161 int index = lookup(uniforms, uniform);
\r
2165 index = allocate(uniforms, uniform);
\r
2166 const TString &name = symbol->getSymbol().c_str();
\r
2168 declareUniform(type, name, index);
\r
2177 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2179 ASSERT(!attribute->isArray());
\r
2180 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2182 int index = lookup(attributes, attribute);
\r
2186 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2191 index = allocate(attributes, attribute);
\r
2192 const TType &type = attribute->getType();
\r
2193 int registerCount = attribute->totalRegisterCount();
\r
2195 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2197 for(int i = 0; i < registerCount; i++)
\r
2199 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2203 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2205 const char *name = symbol->getSymbol().c_str();
\r
2206 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2213 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2215 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2216 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2217 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2221 return samplerRegister(symbol);
\r
2225 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect || binary->getOp() == EOpIndexDirectStruct);
\r
2227 return samplerRegister(binary->getLeft()); // Index added later
\r
2229 else UNREACHABLE();
\r
2234 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2236 const TType &type = sampler->getType();
\r
2237 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2239 int index = lookup(samplers, sampler);
\r
2243 index = allocate(samplers, sampler);
\r
2245 if(sampler->getQualifier() == EvqUniform)
\r
2247 const char *name = sampler->getSymbol().c_str();
\r
2248 declareUniform(type, name, index);
\r
2255 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2257 for(unsigned int i = 0; i < list.size(); i++)
\r
2259 if(list[i] == variable)
\r
2261 return i; // Pointer match
\r
2265 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2269 for(unsigned int i = 0; i < list.size(); i++)
\r
2273 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2277 if(listSymbol->getId() == varSymbol->getId())
\r
2279 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2280 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2281 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2293 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2295 int index = lookup(list, variable);
\r
2299 unsigned int registerCount = variable->totalRegisterCount();
\r
2301 for(unsigned int i = 0; i < list.size(); i++)
\r
2305 unsigned int j = 1;
\r
2306 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2308 if(list[i + j] != 0)
\r
2314 if(j == registerCount) // Found free slots
\r
2316 for(unsigned int j = 0; j < registerCount; j++)
\r
2318 list[i + j] = variable;
\r
2326 index = list.size();
\r
2328 for(unsigned int i = 0; i < registerCount; i++)
\r
2330 list.push_back(variable);
\r
2337 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2339 int index = lookup(list, variable);
\r
2347 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2349 const TTypeList *structure = type.getStruct();
\r
2350 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2354 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2356 if(isSamplerRegister(type))
\r
2358 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2360 shader->declareSampler(index + i);
\r
2366 if(type.isArray())
\r
2368 int elementIndex = index;
\r
2370 for(int i = 0; i < type.getArraySize(); i++)
\r
2372 for(size_t j = 0; j < structure->size(); j++)
\r
2374 const TType &fieldType = *(*structure)[j].type;
\r
2375 const TString &fieldName = fieldType.getFieldName();
\r
2377 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2378 declareUniform(fieldType, uniformName, elementIndex);
\r
2379 elementIndex += fieldType.totalRegisterCount();
\r
2385 int fieldIndex = index;
\r
2387 for(size_t i = 0; i < structure->size(); i++)
\r
2389 const TType &fieldType = *(*structure)[i].type;
\r
2390 const TString &fieldName = fieldType.getFieldName();
\r
2392 const TString uniformName = name + "." + fieldName;
\r
2393 declareUniform(fieldType, uniformName, fieldIndex);
\r
2394 fieldIndex += fieldType.totalRegisterCount();
\r
2400 GLenum OutputASM::glVariableType(const TType &type)
\r
2402 switch(type.getBasicType())
\r
2405 if(type.isScalar())
\r
2409 else if(type.isVector())
\r
2411 switch(type.getNominalSize())
\r
2413 case 2: return GL_FLOAT_VEC2;
\r
2414 case 3: return GL_FLOAT_VEC3;
\r
2415 case 4: return GL_FLOAT_VEC4;
\r
2416 default: UNREACHABLE();
\r
2419 else if(type.isMatrix())
\r
2421 switch(type.getNominalSize())
\r
2424 switch(type.getSecondarySize())
\r
2426 case 2: return GL_FLOAT_MAT2;
\r
2427 case 3: return GL_FLOAT_MAT2x3;
\r
2428 case 4: return GL_FLOAT_MAT2x4;
\r
2429 default: UNREACHABLE();
\r
2432 switch(type.getSecondarySize())
\r
2434 case 2: return GL_FLOAT_MAT3x2;
\r
2435 case 3: return GL_FLOAT_MAT3;
\r
2436 case 4: return GL_FLOAT_MAT3x4;
\r
2437 default: UNREACHABLE();
\r
2440 switch(type.getSecondarySize())
\r
2442 case 2: return GL_FLOAT_MAT4x2;
\r
2443 case 3: return GL_FLOAT_MAT4x3;
\r
2444 case 4: return GL_FLOAT_MAT4;
\r
2445 default: UNREACHABLE();
\r
2447 default: UNREACHABLE();
\r
2450 else UNREACHABLE();
\r
2453 if(type.isScalar())
\r
2457 else if(type.isVector())
\r
2459 switch(type.getNominalSize())
\r
2461 case 2: return GL_INT_VEC2;
\r
2462 case 3: return GL_INT_VEC3;
\r
2463 case 4: return GL_INT_VEC4;
\r
2464 default: UNREACHABLE();
\r
2467 else UNREACHABLE();
\r
2470 if(type.isScalar())
\r
2472 return GL_UNSIGNED_INT;
\r
2474 else if(type.isVector())
\r
2476 switch(type.getNominalSize())
\r
2478 case 2: return GL_UNSIGNED_INT_VEC2;
\r
2479 case 3: return GL_UNSIGNED_INT_VEC3;
\r
2480 case 4: return GL_UNSIGNED_INT_VEC4;
\r
2481 default: UNREACHABLE();
\r
2484 else UNREACHABLE();
\r
2487 if(type.isScalar())
\r
2491 else if(type.isVector())
\r
2493 switch(type.getNominalSize())
\r
2495 case 2: return GL_BOOL_VEC2;
\r
2496 case 3: return GL_BOOL_VEC3;
\r
2497 case 4: return GL_BOOL_VEC4;
\r
2498 default: UNREACHABLE();
\r
2501 else UNREACHABLE();
\r
2503 case EbtSampler2D:
\r
2504 case EbtISampler2D:
\r
2505 case EbtUSampler2D:
\r
2506 return GL_SAMPLER_2D;
\r
2507 case EbtSamplerCube:
\r
2508 case EbtISamplerCube:
\r
2509 case EbtUSamplerCube:
\r
2510 return GL_SAMPLER_CUBE;
\r
2511 case EbtSamplerExternalOES:
\r
2512 return GL_SAMPLER_EXTERNAL_OES;
\r
2513 case EbtSampler3D:
\r
2514 case EbtISampler3D:
\r
2515 case EbtUSampler3D:
\r
2516 return GL_SAMPLER_3D_OES;
\r
2517 case EbtSampler2DArray:
\r
2518 case EbtISampler2DArray:
\r
2519 case EbtUSampler2DArray:
\r
2520 return GL_SAMPLER_2D_ARRAY;
\r
2529 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2531 if(type.getBasicType() == EbtFloat)
\r
2533 switch(type.getPrecision())
\r
2535 case EbpHigh: return GL_HIGH_FLOAT;
\r
2536 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2537 case EbpLow: return GL_LOW_FLOAT;
\r
2538 case EbpUndefined:
\r
2539 // Should be defined as the default precision by the parser
\r
2540 default: UNREACHABLE();
\r
2543 else if(type.getBasicType() == EbtInt)
\r
2545 switch (type.getPrecision())
\r
2547 case EbpHigh: return GL_HIGH_INT;
\r
2548 case EbpMedium: return GL_MEDIUM_INT;
\r
2549 case EbpLow: return GL_LOW_INT;
\r
2550 case EbpUndefined:
\r
2551 // Should be defined as the default precision by the parser
\r
2552 default: UNREACHABLE();
\r
2556 // Other types (boolean, sampler) don't have a precision
\r
2560 int OutputASM::dim(TIntermNode *v)
\r
2562 TIntermTyped *vector = v->getAsTyped();
\r
2563 ASSERT(vector && vector->isRegister());
\r
2564 return vector->getNominalSize();
\r
2567 int OutputASM::dim2(TIntermNode *m)
\r
2569 TIntermTyped *matrix = m->getAsTyped();
\r
2570 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2571 return matrix->getSecondarySize();
\r
2574 // Returns ~0 if no loop count could be determined
\r
2575 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2577 // Parse loops of the form:
\r
2578 // for(int index = initial; index [comparator] limit; index += increment)
\r
2579 TIntermSymbol *index = 0;
\r
2580 TOperator comparator = EOpNull;
\r
2583 int increment = 0;
\r
2585 // Parse index name and intial value
\r
2586 if(node->getInit())
\r
2588 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2592 TIntermSequence &sequence = init->getSequence();
\r
2593 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2595 if(variable && variable->getQualifier() == EvqTemporary)
\r
2597 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2599 if(assign->getOp() == EOpInitialize)
\r
2601 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2602 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2604 if(symbol && constant)
\r
2606 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2609 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2617 // Parse comparator and limit value
\r
2618 if(index && node->getCondition())
\r
2620 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2622 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2624 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2628 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2630 comparator = test->getOp();
\r
2631 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2637 // Parse increment
\r
2638 if(index && comparator != EOpNull && node->getExpression())
\r
2640 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2641 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2643 if(binaryTerminal)
\r
2645 TOperator op = binaryTerminal->getOp();
\r
2646 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2650 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2652 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2656 case EOpAddAssign: increment = value; break;
\r
2657 case EOpSubAssign: increment = -value; break;
\r
2658 default: UNIMPLEMENTED();
\r
2663 else if(unaryTerminal)
\r
2665 TOperator op = unaryTerminal->getOp();
\r
2669 case EOpPostIncrement: increment = 1; break;
\r
2670 case EOpPostDecrement: increment = -1; break;
\r
2671 case EOpPreIncrement: increment = 1; break;
\r
2672 case EOpPreDecrement: increment = -1; break;
\r
2673 default: UNIMPLEMENTED();
\r
2678 if(index && comparator != EOpNull && increment != 0)
\r
2680 if(comparator == EOpLessThanEqual)
\r
2682 comparator = EOpLessThan;
\r
2686 if(comparator == EOpLessThan)
\r
2688 int iterations = (limit - initial) / increment;
\r
2690 if(iterations <= 0)
\r
2695 return iterations;
\r
2697 else UNIMPLEMENTED(); // Falls through
\r
2703 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2706 loopDiscontinuity = false;
\r
2708 node->traverse(this);
\r
2710 return loopDiscontinuity;
\r
2713 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2715 if(visit == PreVisit)
\r
2719 else if(visit == PostVisit)
\r
2727 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2729 if(loopDiscontinuity)
\r
2739 switch(node->getFlowOp())
\r
2746 loopDiscontinuity = true;
\r
2748 default: UNREACHABLE();
\r
2751 return !loopDiscontinuity;
\r
2754 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2756 return !loopDiscontinuity;
\r