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.getTreeRoot()->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 case EOpIndexDirectInterfaceBlock:
\r
326 if(visit == PostVisit)
\r
328 ASSERT(leftType.isStruct() || (leftType.isInterfaceBlock()));
\r
330 const TFieldList& fields = (node->getOp() == EOpIndexDirectStruct) ?
\r
331 leftType.getStruct()->fields() :
\r
332 leftType.getInterfaceBlock()->fields();
\r
333 int index = right->getAsConstantUnion()->getIConst(0);
\r
334 int fieldOffset = 0;
\r
336 for(int i = 0; i < index; i++)
\r
338 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
341 copy(result, left, fieldOffset);
\r
344 case EOpVectorSwizzle:
\r
345 if(visit == PostVisit)
\r
348 TIntermAggregate *components = right->getAsAggregate();
\r
352 TIntermSequence &sequence = components->getSequence();
\r
355 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
357 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
361 int i = element->getUnionArrayPointer()[0].getIConst();
\r
362 swizzle |= i << (component * 2);
\r
365 else UNREACHABLE(0);
\r
368 else UNREACHABLE(0);
\r
370 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
371 mov->src[0].swizzle = swizzle;
\r
374 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
375 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
376 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
377 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
378 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
379 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
380 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
381 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
383 if(visit == PostVisit)
\r
385 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
387 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
389 Temporary equal(this);
\r
390 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
391 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
396 if(visit == PostVisit)
\r
398 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
400 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
402 Temporary notEqual(this);
\r
403 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
404 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
408 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
409 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
410 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
411 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
412 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
413 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
414 case EOpMatrixTimesScalar:
\r
415 if(visit == PostVisit)
\r
417 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
419 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
420 mul->dst.index += i;
\r
421 argument(mul->src[0], left, i);
\r
425 case EOpVectorTimesMatrix:
\r
426 if(visit == PostVisit)
\r
428 int size = leftType.getNominalSize();
\r
430 for(int i = 0; i < size; i++)
\r
432 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
433 dot->dst.mask = 1 << i;
\r
434 argument(dot->src[1], right, i);
\r
438 case EOpMatrixTimesVector:
\r
439 if(visit == PostVisit)
\r
441 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
442 mul->src[1].swizzle = 0x00;
\r
444 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
446 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
447 argument(mad->src[0], left, i);
\r
448 mad->src[1].swizzle = i * 0x55;
\r
452 case EOpMatrixTimesMatrix:
\r
453 if(visit == PostVisit)
\r
455 int dim = leftType.getNominalSize();
\r
457 for(int i = 0; i < dim; i++)
\r
459 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
460 mul->dst.index += i;
\r
461 argument(mul->src[1], right, i);
\r
462 mul->src[1].swizzle = 0x00;
\r
464 for(int j = 1; j < dim; j++)
\r
466 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
467 mad->dst.index += i;
\r
468 argument(mad->src[0], left, j);
\r
469 argument(mad->src[1], right, i);
\r
470 mad->src[1].swizzle = j * 0x55;
\r
471 argument(mad->src[2], result, i);
\r
477 if(trivial(right, 6))
\r
479 if(visit == PostVisit)
\r
481 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
484 else // Short-circuit evaluation
\r
486 if(visit == InVisit)
\r
488 emit(sw::Shader::OPCODE_MOV, result, left);
\r
489 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
490 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
492 else if(visit == PostVisit)
\r
494 emit(sw::Shader::OPCODE_MOV, result, right);
\r
495 emit(sw::Shader::OPCODE_ENDIF);
\r
499 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
500 case EOpLogicalAnd:
\r
501 if(trivial(right, 6))
\r
503 if(visit == PostVisit)
\r
505 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
508 else // Short-circuit evaluation
\r
510 if(visit == InVisit)
\r
512 emit(sw::Shader::OPCODE_MOV, result, left);
\r
513 emit(sw::Shader::OPCODE_IF, 0, result);
\r
515 else if(visit == PostVisit)
\r
517 emit(sw::Shader::OPCODE_MOV, result, right);
\r
518 emit(sw::Shader::OPCODE_ENDIF);
\r
522 default: UNREACHABLE(node->getOp());
\r
528 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
530 if(currentScope != emitScope)
\r
535 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
536 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
537 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
539 TIntermTyped *result = node;
\r
540 TIntermTyped *arg = node->getOperand();
\r
542 switch(node->getOp())
\r
545 if(visit == PostVisit)
\r
547 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
549 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
550 neg->dst.index += index;
\r
551 argument(neg->src[0], arg, index);
\r
552 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
556 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
557 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
558 case EOpPostIncrement:
\r
559 if(visit == PostVisit)
\r
563 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
565 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
566 add->dst.index += index;
\r
567 argument(add->src[0], arg, index);
\r
570 assignLvalue(arg, arg);
\r
573 case EOpPostDecrement:
\r
574 if(visit == PostVisit)
\r
578 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
580 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
581 sub->dst.index += index;
\r
582 argument(sub->src[0], arg, index);
\r
585 assignLvalue(arg, arg);
\r
588 case EOpPreIncrement:
\r
589 if(visit == PostVisit)
\r
591 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
593 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
594 add->dst.index += index;
\r
595 argument(add->src[0], arg, index);
\r
598 assignLvalue(arg, result);
\r
601 case EOpPreDecrement:
\r
602 if(visit == PostVisit)
\r
604 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
606 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
607 sub->dst.index += index;
\r
608 argument(sub->src[0], arg, index);
\r
611 assignLvalue(arg, result);
\r
614 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
615 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
616 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
617 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
618 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
619 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
620 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
621 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
622 case EOpSinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_SINH, result, arg); break;
\r
623 case EOpCosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_COSH, result, arg); break;
\r
624 case EOpTanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_TANH, result, arg); break;
\r
625 case EOpAsinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASINH, result, arg); break;
\r
626 case EOpAcosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOSH, result, arg); break;
\r
627 case EOpAtanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATANH, result, arg); break;
\r
628 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
629 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
630 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
631 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
632 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
633 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
634 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
635 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
636 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
637 case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
\r
638 case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
\r
639 case EOpRoundEven: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUNDEVEN, result, arg); break;
\r
640 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
641 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
642 case EOpIsNan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISNAN, result, arg); break;
\r
643 case EOpIsInf: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISINF, result, arg); break;
\r
644 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
645 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
646 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
647 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
648 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
649 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
650 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
651 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg); break;
\r
652 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg); break;
\r
653 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg); break;
\r
654 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg); break;
\r
655 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg); break;
\r
656 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg); break;
\r
657 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg); break;
\r
658 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg); break;
\r
659 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg); break;
\r
660 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg); break;
\r
662 if(visit == PostVisit)
\r
664 int numCols = arg->getNominalSize();
\r
665 int numRows = arg->getSecondarySize();
\r
666 for(int i = 0; i < numCols; ++i)
\r
668 for(int j = 0; j < numRows; ++j)
\r
670 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
671 mov->src[0].index += i;
\r
672 mov->src[0].swizzle = 0x55 * j;
\r
673 mov->dst.index += j;
\r
674 mov->dst.mask = 1 << i;
\r
679 default: UNREACHABLE(node->getOp());
\r
685 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
687 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
692 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
694 TIntermTyped *result = node;
\r
695 const TType &resultType = node->getType();
\r
696 TIntermSequence &arg = node->getSequence();
\r
697 int argumentCount = arg.size();
\r
699 switch(node->getOp())
\r
701 case EOpSequence: break;
\r
702 case EOpDeclaration: break;
\r
703 case EOpPrototype: break;
\r
705 if(visit == PostVisit)
\r
707 copy(result, arg[1]);
\r
711 if(visit == PreVisit)
\r
713 const TString &name = node->getName();
\r
715 if(emitScope == FUNCTION)
\r
717 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
719 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
720 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
722 const Function *function = findFunction(name);
\r
723 ASSERT(function); // Should have been added during global pass
\r
724 label->dst.index = function->label;
\r
725 currentFunction = function->label;
\r
728 else if(emitScope == GLOBAL)
\r
730 if(name != "main(")
\r
732 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
733 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
736 else UNREACHABLE(emitScope);
\r
738 currentScope = FUNCTION;
\r
740 else if(visit == PostVisit)
\r
742 if(emitScope == FUNCTION)
\r
744 if(functionArray.size() > 1) // No need to return when there's only main()
\r
746 emit(sw::Shader::OPCODE_RET);
\r
750 currentScope = GLOBAL;
\r
753 case EOpFunctionCall:
\r
754 if(visit == PostVisit)
\r
756 if(node->isUserDefined())
\r
758 const TString &name = node->getName();
\r
759 const Function *function = findFunction(name);
\r
763 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
767 TIntermSequence &arguments = *function->arg;
\r
769 for(int i = 0; i < argumentCount; i++)
\r
771 TIntermTyped *in = arguments[i]->getAsTyped();
\r
773 if(in->getQualifier() == EvqIn ||
\r
774 in->getQualifier() == EvqInOut ||
\r
775 in->getQualifier() == EvqConstReadOnly)
\r
781 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
782 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
783 call->dst.index = function->label;
\r
785 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
787 copy(result, function->ret);
\r
790 for(int i = 0; i < argumentCount; i++)
\r
792 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
793 TIntermTyped *out = arg[i]->getAsTyped();
\r
795 if(argument->getQualifier() == EvqOut ||
\r
796 argument->getQualifier() == EvqInOut)
\r
798 copy(out, argument);
\r
804 TString name = TFunction::unmangleName(node->getName());
\r
806 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
808 if(argumentCount == 2)
\r
810 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
812 else if(argumentCount == 3) // bias
\r
814 Temporary uvwb(this);
\r
815 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
816 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
817 bias->dst.mask = 0x8;
\r
819 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
822 else UNREACHABLE(argumentCount);
\r
824 else if(name == "texture2DProj")
\r
826 TIntermTyped *t = arg[1]->getAsTyped();
\r
828 if(argumentCount == 2)
\r
830 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
831 tex->project = true;
\r
833 if(t->getNominalSize() == 3)
\r
835 tex->src[0].swizzle = 0xA4;
\r
837 else ASSERT(t->getNominalSize() == 4);
\r
839 else if(argumentCount == 3) // bias
\r
841 Temporary proj(this);
\r
843 if(t->getNominalSize() == 3)
\r
845 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
846 div->src[1].swizzle = 0xAA;
\r
847 div->dst.mask = 0x3;
\r
849 else if(t->getNominalSize() == 4)
\r
851 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
852 div->src[1].swizzle = 0xFF;
\r
853 div->dst.mask = 0x3;
\r
855 else UNREACHABLE(t->getNominalSize());
\r
857 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
858 bias->dst.mask = 0x8;
\r
860 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
863 else UNREACHABLE(argumentCount);
\r
865 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
867 Temporary uvwb(this);
\r
868 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
869 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
870 lod->dst.mask = 0x8;
\r
872 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
874 else if(name == "texture2DProjLod")
\r
876 TIntermTyped *t = arg[1]->getAsTyped();
\r
877 Temporary proj(this);
\r
879 if(t->getNominalSize() == 3)
\r
881 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
882 div->src[1].swizzle = 0xAA;
\r
883 div->dst.mask = 0x3;
\r
885 else if(t->getNominalSize() == 4)
\r
887 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
888 div->src[1].swizzle = 0xFF;
\r
889 div->dst.mask = 0x3;
\r
891 else UNREACHABLE(t->getNominalSize());
\r
893 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
894 lod->dst.mask = 0x8;
\r
896 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
898 else UNREACHABLE(0);
\r
902 case EOpParameters:
\r
904 case EOpConstructFloat:
\r
905 case EOpConstructVec2:
\r
906 case EOpConstructVec3:
\r
907 case EOpConstructVec4:
\r
908 case EOpConstructBool:
\r
909 case EOpConstructBVec2:
\r
910 case EOpConstructBVec3:
\r
911 case EOpConstructBVec4:
\r
912 case EOpConstructInt:
\r
913 case EOpConstructIVec2:
\r
914 case EOpConstructIVec3:
\r
915 case EOpConstructIVec4:
\r
916 case EOpConstructUInt:
\r
917 case EOpConstructUVec2:
\r
918 case EOpConstructUVec3:
\r
919 case EOpConstructUVec4:
\r
920 if(visit == PostVisit)
\r
924 for(int i = 0; i < argumentCount; i++)
\r
926 TIntermTyped *argi = arg[i]->getAsTyped();
\r
927 int size = argi->getNominalSize();
\r
929 if(!argi->isMatrix())
\r
931 Instruction *mov = emitCast(result, argi);
\r
932 mov->dst.mask = (0xF << component) & 0xF;
\r
933 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
941 while(component < resultType.getNominalSize())
\r
943 Instruction *mov = emitCast(result, argi);
\r
944 mov->dst.mask = (0xF << component) & 0xF;
\r
945 mov->src[0].index += column;
\r
946 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
955 case EOpConstructMat2:
\r
956 case EOpConstructMat2x3:
\r
957 case EOpConstructMat2x4:
\r
958 case EOpConstructMat3x2:
\r
959 case EOpConstructMat3:
\r
960 case EOpConstructMat3x4:
\r
961 case EOpConstructMat4x2:
\r
962 case EOpConstructMat4x3:
\r
963 case EOpConstructMat4:
\r
964 if(visit == PostVisit)
\r
966 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
967 const int outCols = result->getNominalSize();
\r
968 const int outRows = result->getSecondarySize();
\r
970 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
972 for(int i = 0; i < outCols; i++)
\r
974 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
975 init->dst.index += i;
\r
976 Instruction *mov = emitCast(result, arg0);
\r
977 mov->dst.index += i;
\r
978 mov->dst.mask = 1 << i;
\r
979 ASSERT(mov->src[0].swizzle == 0x00);
\r
982 else if(arg0->isMatrix())
\r
984 const int inCols = arg0->getNominalSize();
\r
985 const int inRows = arg0->getSecondarySize();
\r
987 for(int i = 0; i < outCols; i++)
\r
989 if(i >= inCols || outRows > inRows)
\r
991 // Initialize to identity matrix
\r
992 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
993 Instruction *mov = emitCast(result, &col);
\r
994 mov->dst.index += i;
\r
999 Instruction *mov = emitCast(result, arg0);
\r
1000 mov->dst.index += i;
\r
1001 mov->dst.mask = 0xF >> (4 - inRows);
\r
1002 argument(mov->src[0], arg0, i);
\r
1011 for(int i = 0; i < argumentCount; i++)
\r
1013 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1014 int size = argi->getNominalSize();
\r
1017 while(element < size)
\r
1019 Instruction *mov = emitCast(result, argi);
\r
1020 mov->dst.index += column;
\r
1021 mov->dst.mask = (0xF << row) & 0xF;
\r
1022 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
1024 int end = row + size - element;
\r
1025 column = end >= outRows ? column + 1 : column;
\r
1026 element = element + outRows - row;
\r
1027 row = end >= outRows ? 0 : end;
\r
1033 case EOpConstructStruct:
\r
1034 if(visit == PostVisit)
\r
1037 for(int i = 0; i < argumentCount; i++)
\r
1039 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1040 int size = argi->totalRegisterCount();
\r
1042 for(int index = 0; index < size; index++)
\r
1044 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
1045 mov->dst.index += index + offset;
\r
1046 mov->dst.mask = writeMask(result, offset + index);
\r
1047 argument(mov->src[0], argi, index);
\r
1054 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1055 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1056 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1057 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1058 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1059 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1060 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1061 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1062 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1063 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1064 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1066 if(visit == PostVisit)
\r
1068 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1069 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1072 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1073 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1074 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); 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(node->getOp());
\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
1287 condition->traverse(this);
\r
1290 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1294 body->traverse(this);
\r
1297 emit(sw::Shader::OPCODE_TEST);
\r
1301 expression->traverse(this);
\r
1306 condition->traverse(this);
\r
1309 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1316 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1318 if(currentScope != emitScope)
\r
1323 switch(node->getFlowOp())
\r
1325 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1326 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1327 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1329 if(visit == PostVisit)
\r
1331 TIntermTyped *value = node->getExpression();
\r
1335 copy(functionArray[currentFunction].ret, value);
\r
1338 emit(sw::Shader::OPCODE_LEAVE);
\r
1341 default: UNREACHABLE(node->getFlowOp());
\r
1347 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1349 return operand && isSamplerRegister(operand->getType());
\r
1352 bool OutputASM::isSamplerRegister(const TType &type)
\r
1354 // A sampler register's qualifiers can be:
\r
1355 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1356 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1357 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1358 // the sampler has been copied and is no longer a sampler register.
\r
1359 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1362 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1364 if(isSamplerRegister(dst))
\r
1366 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1369 Instruction *instruction = new Instruction(op);
\r
1373 instruction->dst.type = registerType(dst);
\r
1374 instruction->dst.index = registerIndex(dst) + index;
\r
1375 instruction->dst.mask = writeMask(dst);
\r
1376 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1379 argument(instruction->src[0], src0, index);
\r
1380 argument(instruction->src[1], src1, index);
\r
1381 argument(instruction->src[2], src2, index);
\r
1383 shader->append(instruction);
\r
1385 return instruction;
\r
1388 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1390 // Integers are implemented as float
\r
1391 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1393 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1395 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1397 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1399 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1401 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1404 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1407 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1409 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1411 emit(op, dst, src0, src1, src2, index);
\r
1415 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1417 emitBinary(op, result, src0, src1);
\r
1418 assignLvalue(lhs, result);
\r
1421 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1423 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1424 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1426 Instruction *cmp = emit(opcode, dst, left, right);
\r
1427 cmp->control = cmpOp;
\r
1428 argument(cmp->src[0], left, index);
\r
1429 argument(cmp->src[1], right, index);
\r
1432 int componentCount(const TType &type, int registers)
\r
1434 if(registers == 0)
\r
1439 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1441 int index = registers / type.elementRegisterCount();
\r
1442 registers -= index * type.elementRegisterCount();
\r
1443 return index * type.getElementSize() + componentCount(type, registers);
\r
1446 if(type.isStruct() || type.isInterfaceBlock())
\r
1448 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
\r
1451 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1453 const TType &fieldType = *((*field)->type());
\r
1455 if(fieldType.totalRegisterCount() <= registers)
\r
1457 registers -= fieldType.totalRegisterCount();
\r
1458 elements += fieldType.getObjectSize();
\r
1460 else // Register within this field
\r
1462 return elements + componentCount(fieldType, registers);
\r
1466 else if(type.isMatrix())
\r
1468 return registers * type.registerSize();
\r
1475 int registerSize(const TType &type, int registers)
\r
1477 if(registers == 0)
\r
1479 if(type.isStruct())
\r
1481 return registerSize(*((*(type.getStruct()->fields().begin()))->type()), 0);
\r
1484 return type.registerSize();
\r
1487 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1489 int index = registers / type.elementRegisterCount();
\r
1490 registers -= index * type.elementRegisterCount();
\r
1491 return registerSize(type, registers);
\r
1494 if(type.isStruct() || type.isInterfaceBlock())
\r
1496 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
\r
1499 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1501 const TType &fieldType = *((*field)->type());
\r
1503 if(fieldType.totalRegisterCount() <= registers)
\r
1505 registers -= fieldType.totalRegisterCount();
\r
1506 elements += fieldType.getObjectSize();
\r
1508 else // Register within this field
\r
1510 return registerSize(fieldType, registers);
\r
1514 else if(type.isMatrix())
\r
1516 return registerSize(type, 0);
\r
1523 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1527 TIntermTyped *arg = argument->getAsTyped();
\r
1528 const TType &type = arg->getType();
\r
1529 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1531 int size = registerSize(type, index);
\r
1533 parameter.type = registerType(arg);
\r
1535 if(arg->getQualifier() == EvqConstExpr)
\r
1537 int component = componentCount(type, index);
\r
1538 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1540 for(int i = 0; i < 4; i++)
\r
1542 if(size == 1) // Replicate
\r
1544 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1548 parameter.value[i] = constants[component + i].getAsFloat();
\r
1552 parameter.value[i] = 0.0f;
\r
1558 parameter.index = registerIndex(arg) + index;
\r
1560 if(isSamplerRegister(arg))
\r
1562 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1566 TIntermTyped *left = binary->getLeft();
\r
1567 TIntermTyped *right = binary->getRight();
\r
1569 switch(binary->getOp())
\r
1571 case EOpIndexDirect:
\r
1572 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1574 case EOpIndexIndirect:
\r
1575 if(left->getArraySize() > 1)
\r
1577 parameter.rel.type = registerType(binary->getRight());
\r
1578 parameter.rel.index = registerIndex(binary->getRight());
\r
1579 parameter.rel.scale = 1;
\r
1580 parameter.rel.deterministic = true;
\r
1583 case EOpIndexDirectStruct:
\r
1584 case EOpIndexDirectInterfaceBlock:
\r
1585 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1588 UNREACHABLE(binary->getOp());
\r
1594 if(!IsSampler(arg->getBasicType()))
\r
1596 parameter.swizzle = readSwizzle(arg, size);
\r
1601 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1603 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1605 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1606 mov->dst.index += index;
\r
1607 mov->dst.mask = writeMask(dst, index);
\r
1608 argument(mov->src[0], src, offset + index);
\r
1612 int swizzleElement(int swizzle, int index)
\r
1614 return (swizzle >> (index * 2)) & 0x03;
\r
1617 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1619 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1620 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1621 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1622 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1625 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1628 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1629 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
\r
1631 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1634 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1636 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1638 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1640 Temporary address(this);
\r
1641 lvalue(insert->dst, address, dst);
\r
1643 insert->src[0].type = insert->dst.type;
\r
1644 insert->src[0].index = insert->dst.index;
\r
1645 insert->src[0].rel = insert->dst.rel;
\r
1646 argument(insert->src[1], src);
\r
1647 argument(insert->src[2], binary->getRight());
\r
1649 shader->append(insert);
\r
1653 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1655 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1657 Temporary address(this);
\r
1658 int swizzle = lvalue(mov->dst, address, dst);
\r
1659 mov->dst.index += offset;
\r
1663 mov->dst.mask = writeMask(dst, offset);
\r
1666 argument(mov->src[0], src, offset);
\r
1667 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1669 shader->append(mov);
\r
1674 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1676 TIntermTyped *result = node;
\r
1677 TIntermBinary *binary = node->getAsBinaryNode();
\r
1678 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1682 TIntermTyped *left = binary->getLeft();
\r
1683 TIntermTyped *right = binary->getRight();
\r
1685 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1687 switch(binary->getOp())
\r
1689 case EOpIndexDirect:
\r
1691 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1693 if(left->isRegister())
\r
1695 int leftMask = dst.mask;
\r
1698 while((leftMask & dst.mask) == 0)
\r
1700 dst.mask = dst.mask << 1;
\r
1703 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1704 dst.mask = 1 << element;
\r
1708 else if(left->isArray() || left->isMatrix())
\r
1710 dst.index += rightIndex * result->totalRegisterCount();
\r
1713 else UNREACHABLE(0);
\r
1716 case EOpIndexIndirect:
\r
1718 if(left->isRegister())
\r
1720 // Requires INSERT instruction (handled by calling function)
\r
1722 else if(left->isArray() || left->isMatrix())
\r
1724 int scale = result->totalRegisterCount();
\r
1726 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1728 if(left->totalRegisterCount() > 1)
\r
1730 sw::Shader::SourceParameter relativeRegister;
\r
1731 argument(relativeRegister, right);
\r
1733 dst.rel.index = relativeRegister.index;
\r
1734 dst.rel.type = relativeRegister.type;
\r
1735 dst.rel.scale = scale;
\r
1736 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1739 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1743 Constant oldScale((int)dst.rel.scale);
\r
1744 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1745 mad->src[0].index = dst.rel.index;
\r
1746 mad->src[0].type = dst.rel.type;
\r
1750 Constant oldScale((int)dst.rel.scale);
\r
1751 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1752 mul->src[0].index = dst.rel.index;
\r
1753 mul->src[0].type = dst.rel.type;
\r
1755 Constant newScale(scale);
\r
1756 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1759 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1760 dst.rel.index = registerIndex(&address);
\r
1761 dst.rel.scale = 1;
\r
1763 else // Just add the new index to the address register
\r
1767 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1771 Constant newScale(scale);
\r
1772 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1776 else UNREACHABLE(0);
\r
1779 case EOpIndexDirectStruct:
\r
1780 case EOpIndexDirectInterfaceBlock:
\r
1782 const TFieldList& fields = (binary->getOp() == EOpIndexDirectStruct) ?
\r
1783 left->getType().getStruct()->fields() :
\r
1784 left->getType().getInterfaceBlock()->fields();
\r
1785 int index = right->getAsConstantUnion()->getIConst(0);
\r
1786 int fieldOffset = 0;
\r
1788 for(int i = 0; i < index; i++)
\r
1790 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
1793 dst.type = registerType(left);
\r
1794 dst.index += fieldOffset;
\r
1795 dst.mask = writeMask(right);
\r
1800 case EOpVectorSwizzle:
\r
1802 ASSERT(left->isRegister());
\r
1804 int leftMask = dst.mask;
\r
1807 int rightMask = 0;
\r
1809 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1811 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1813 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
1815 int element = swizzleElement(leftSwizzle, index);
\r
1816 rightMask = rightMask | (1 << element);
\r
1817 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1820 dst.mask = leftMask & rightMask;
\r
1826 UNREACHABLE(binary->getOp()); // Not an l-value operator
\r
1832 dst.type = registerType(symbol);
\r
1833 dst.index = registerIndex(symbol);
\r
1834 dst.mask = writeMask(symbol);
\r
1841 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1843 if(isSamplerRegister(operand))
\r
1845 return sw::Shader::PARAMETER_SAMPLER;
\r
1848 const TQualifier qualifier = operand->getQualifier();
\r
1849 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1851 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1852 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1854 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1856 outputQualifier = qualifier;
\r
1861 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1862 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1863 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1864 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1865 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1866 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1867 case EvqVertexIn: return sw::Shader::PARAMETER_INPUT;
\r
1868 case EvqFragmentOut: return sw::Shader::PARAMETER_COLOROUT;
\r
1869 case EvqVertexOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1870 case EvqFragmentIn: return sw::Shader::PARAMETER_INPUT;
\r
1871 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1872 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1873 case EvqSmooth: return sw::Shader::PARAMETER_OUTPUT;
\r
1874 case EvqFlat: return sw::Shader::PARAMETER_OUTPUT;
\r
1875 case EvqCentroidOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1876 case EvqSmoothIn: return sw::Shader::PARAMETER_INPUT;
\r
1877 case EvqFlatIn: return sw::Shader::PARAMETER_INPUT;
\r
1878 case EvqCentroidIn: return sw::Shader::PARAMETER_INPUT;
\r
1879 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1880 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1881 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1882 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1883 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1884 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1885 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1886 case EvqInstanceID: return sw::Shader::PARAMETER_MISCTYPE;
\r
1887 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1888 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1889 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1890 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1891 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1892 default: UNREACHABLE(qualifier);
\r
1895 return sw::Shader::PARAMETER_VOID;
\r
1898 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1900 if(isSamplerRegister(operand))
\r
1902 return samplerRegister(operand);
\r
1905 switch(operand->getQualifier())
\r
1907 case EvqTemporary: return temporaryRegister(operand);
\r
1908 case EvqGlobal: return temporaryRegister(operand);
\r
1909 case EvqConstExpr: UNREACHABLE(EvqConstExpr);
\r
1910 case EvqAttribute: return attributeRegister(operand);
\r
1911 case EvqVaryingIn: return varyingRegister(operand);
\r
1912 case EvqVaryingOut: return varyingRegister(operand);
\r
1913 case EvqVertexIn: return attributeRegister(operand);
\r
1914 case EvqFragmentOut: return 0;
\r
1915 case EvqVertexOut: return varyingRegister(operand);
\r
1916 case EvqFragmentIn: return varyingRegister(operand);
\r
1917 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1918 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1919 case EvqSmooth: return varyingRegister(operand);
\r
1920 case EvqFlat: return varyingRegister(operand);
\r
1921 case EvqCentroidOut: return varyingRegister(operand);
\r
1922 case EvqSmoothIn: return varyingRegister(operand);
\r
1923 case EvqFlatIn: return varyingRegister(operand);
\r
1924 case EvqCentroidIn: return varyingRegister(operand);
\r
1925 case EvqUniform: return uniformRegister(operand);
\r
1926 case EvqIn: return temporaryRegister(operand);
\r
1927 case EvqOut: return temporaryRegister(operand);
\r
1928 case EvqInOut: return temporaryRegister(operand);
\r
1929 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1930 case EvqPosition: return varyingRegister(operand);
\r
1931 case EvqPointSize: return varyingRegister(operand);
\r
1932 case EvqInstanceID: vertexShader->instanceIdDeclared = true; return 0;
\r
1933 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1934 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1935 case EvqPointCoord: return varyingRegister(operand);
\r
1936 case EvqFragColor: return 0;
\r
1937 case EvqFragData: return 0;
\r
1938 default: UNREACHABLE(operand->getQualifier());
\r
1944 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1946 if(destination->getQualifier() == EvqPointSize)
\r
1948 return 0x2; // Point size stored in the y component
\r
1951 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1954 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1956 if(argument->getQualifier() == EvqPointSize)
\r
1958 return 0x55; // Point size stored in the y component
\r
1961 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1963 return swizzleSize[size];
\r
1966 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1967 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1969 if(!expression->isRegister())
\r
1974 return cost(expression, budget) >= 0;
\r
1977 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1978 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1985 if(expression->getAsSymbolNode())
\r
1989 else if(expression->getAsConstantUnion())
\r
1993 else if(expression->getAsBinaryNode())
\r
1995 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1997 switch(binary->getOp())
\r
1999 case EOpVectorSwizzle:
\r
2000 case EOpIndexDirect:
\r
2001 case EOpIndexDirectStruct:
\r
2002 case EOpIndexDirectInterfaceBlock:
\r
2003 return cost(binary->getLeft(), budget - 0);
\r
2007 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
2012 else if(expression->getAsUnaryNode())
\r
2014 TIntermUnary *unary = expression->getAsUnaryNode();
\r
2016 switch(unary->getOp())
\r
2020 return cost(unary->getOperand(), budget - 1);
\r
2025 else if(expression->getAsSelectionNode())
\r
2027 TIntermSelection *selection = expression->getAsSelectionNode();
\r
2029 if(selection->usesTernaryOperator())
\r
2031 TIntermTyped *condition = selection->getCondition();
\r
2032 TIntermNode *trueBlock = selection->getTrueBlock();
\r
2033 TIntermNode *falseBlock = selection->getFalseBlock();
\r
2034 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
2036 if(constantCondition)
\r
2038 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
2042 return cost(trueBlock, budget - 0);
\r
2046 return cost(falseBlock, budget - 0);
\r
2051 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
2059 const Function *OutputASM::findFunction(const TString &name)
\r
2061 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
2063 if(functionArray[f].name == name)
\r
2065 return &functionArray[f];
\r
2072 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
2074 return allocate(temporaries, temporary);
\r
2077 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
2079 int var = lookup(varyings, varying);
\r
2083 var = allocate(varyings, varying);
\r
2084 int componentCount = varying->registerSize();
\r
2085 int registerCount = varying->totalRegisterCount();
\r
2089 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2091 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2095 if(varying->getQualifier() == EvqPointCoord)
\r
2097 ASSERT(varying->isRegister());
\r
2098 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2099 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2100 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2101 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2105 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2107 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2108 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2109 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2110 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2114 else if(vertexShader)
\r
2116 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2118 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2122 if(varying->getQualifier() == EvqPosition)
\r
2124 ASSERT(varying->isRegister());
\r
2125 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2126 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2127 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2128 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2129 vertexShader->positionRegister = var;
\r
2131 else if(varying->getQualifier() == EvqPointSize)
\r
2133 ASSERT(varying->isRegister());
\r
2134 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2135 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2136 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2137 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2138 vertexShader->pointSizeRegister = var;
\r
2142 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2145 else UNREACHABLE(0);
\r
2147 declareVarying(varying, var);
\r
2153 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2155 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2157 const TType &type = varying->getType();
\r
2158 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2159 VaryingList &activeVaryings = shaderObject->varyings;
\r
2161 // Check if this varying has been declared before without having a register assigned
\r
2162 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2164 if(v->name == name)
\r
2168 ASSERT(v->reg < 0 || v->reg == reg);
\r
2176 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2180 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2182 const TType &type = uniform->getType();
\r
2183 ASSERT(!IsSampler(type.getBasicType()));
\r
2184 TInterfaceBlock *block = type.getAsInterfaceBlock();
\r
2185 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2186 ASSERT(symbol || block);
\r
2188 if(symbol || block)
\r
2190 int index = lookup(uniforms, uniform);
\r
2194 index = allocate(uniforms, uniform);
\r
2195 const TString &name = symbol ? symbol->getSymbol() : block->name();
\r
2197 declareUniform(type, name, index);
\r
2206 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2208 ASSERT(!attribute->isArray());
\r
2209 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2211 int index = lookup(attributes, attribute);
\r
2215 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2220 index = allocate(attributes, attribute);
\r
2221 const TType &type = attribute->getType();
\r
2222 int registerCount = attribute->totalRegisterCount();
\r
2224 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2226 for(int i = 0; i < registerCount; i++)
\r
2228 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2232 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2234 const char *name = symbol->getSymbol().c_str();
\r
2235 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2242 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2244 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2245 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2246 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2250 return samplerRegister(symbol);
\r
2254 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect ||
\r
2255 binary->getOp() == EOpIndexDirectStruct || binary->getOp() == EOpIndexDirectInterfaceBlock);
\r
2257 return samplerRegister(binary->getLeft()); // Index added later
\r
2259 else UNREACHABLE(0);
\r
2264 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2266 const TType &type = sampler->getType();
\r
2267 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2269 int index = lookup(samplers, sampler);
\r
2273 index = allocate(samplers, sampler);
\r
2275 if(sampler->getQualifier() == EvqUniform)
\r
2277 const char *name = sampler->getSymbol().c_str();
\r
2278 declareUniform(type, name, index);
\r
2285 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2287 for(unsigned int i = 0; i < list.size(); i++)
\r
2289 if(list[i] == variable)
\r
2291 return i; // Pointer match
\r
2295 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2296 TInterfaceBlock *varBlock = variable->getType().getAsInterfaceBlock();
\r
2300 for(unsigned int i = 0; i < list.size(); i++)
\r
2304 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2308 if(listSymbol->getId() == varSymbol->getId())
\r
2310 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2311 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2312 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2322 for(unsigned int i = 0; i < list.size(); i++)
\r
2326 TInterfaceBlock *listBlock = list[i]->getType().getAsInterfaceBlock();
\r
2330 if(listBlock->name() == varBlock->name())
\r
2332 ASSERT(listBlock->arraySize() == varBlock->arraySize());
\r
2333 ASSERT(listBlock->fields() == varBlock->fields());
\r
2334 ASSERT(listBlock->blockStorage() == varBlock->blockStorage());
\r
2335 ASSERT(listBlock->matrixPacking() == varBlock->matrixPacking());
\r
2347 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2349 int index = lookup(list, variable);
\r
2353 unsigned int registerCount = variable->totalRegisterCount();
\r
2355 for(unsigned int i = 0; i < list.size(); i++)
\r
2359 unsigned int j = 1;
\r
2360 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2362 if(list[i + j] != 0)
\r
2368 if(j == registerCount) // Found free slots
\r
2370 for(unsigned int j = 0; j < registerCount; j++)
\r
2372 list[i + j] = variable;
\r
2380 index = list.size();
\r
2382 for(unsigned int i = 0; i < registerCount; i++)
\r
2384 list.push_back(variable);
\r
2391 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2393 int index = lookup(list, variable);
\r
2401 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2403 const TStructure *structure = type.getStruct();
\r
2404 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2408 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2410 if(isSamplerRegister(type))
\r
2412 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2414 shader->declareSampler(index + i);
\r
2420 const TFieldList& fields = structure->fields();
\r
2421 if(type.isArray())
\r
2423 int elementIndex = index;
\r
2425 for(int i = 0; i < type.getArraySize(); i++)
\r
2427 for(size_t j = 0; j < fields.size(); j++)
\r
2429 const TType &fieldType = *(fields[j]->type());
\r
2430 const TString &fieldName = fields[j]->name();
\r
2432 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2433 declareUniform(fieldType, uniformName, elementIndex);
\r
2434 elementIndex += fieldType.totalRegisterCount();
\r
2440 int fieldIndex = index;
\r
2442 for(size_t i = 0; i < fields.size(); i++)
\r
2444 const TType &fieldType = *(fields[i]->type());
\r
2445 const TString &fieldName = fields[i]->name();
\r
2447 const TString uniformName = name + "." + fieldName;
\r
2448 declareUniform(fieldType, uniformName, fieldIndex);
\r
2449 fieldIndex += fieldType.totalRegisterCount();
\r
2455 GLenum OutputASM::glVariableType(const TType &type)
\r
2457 switch(type.getBasicType())
\r
2460 if(type.isScalar())
\r
2464 else if(type.isVector())
\r
2466 switch(type.getNominalSize())
\r
2468 case 2: return GL_FLOAT_VEC2;
\r
2469 case 3: return GL_FLOAT_VEC3;
\r
2470 case 4: return GL_FLOAT_VEC4;
\r
2471 default: UNREACHABLE(type.getNominalSize());
\r
2474 else if(type.isMatrix())
\r
2476 switch(type.getNominalSize())
\r
2479 switch(type.getSecondarySize())
\r
2481 case 2: return GL_FLOAT_MAT2;
\r
2482 case 3: return GL_FLOAT_MAT2x3;
\r
2483 case 4: return GL_FLOAT_MAT2x4;
\r
2484 default: UNREACHABLE(type.getSecondarySize());
\r
2487 switch(type.getSecondarySize())
\r
2489 case 2: return GL_FLOAT_MAT3x2;
\r
2490 case 3: return GL_FLOAT_MAT3;
\r
2491 case 4: return GL_FLOAT_MAT3x4;
\r
2492 default: UNREACHABLE(type.getSecondarySize());
\r
2495 switch(type.getSecondarySize())
\r
2497 case 2: return GL_FLOAT_MAT4x2;
\r
2498 case 3: return GL_FLOAT_MAT4x3;
\r
2499 case 4: return GL_FLOAT_MAT4;
\r
2500 default: UNREACHABLE(type.getSecondarySize());
\r
2502 default: UNREACHABLE(type.getNominalSize());
\r
2505 else UNREACHABLE(0);
\r
2508 if(type.isScalar())
\r
2512 else if(type.isVector())
\r
2514 switch(type.getNominalSize())
\r
2516 case 2: return GL_INT_VEC2;
\r
2517 case 3: return GL_INT_VEC3;
\r
2518 case 4: return GL_INT_VEC4;
\r
2519 default: UNREACHABLE(type.getNominalSize());
\r
2522 else UNREACHABLE(0);
\r
2525 if(type.isScalar())
\r
2527 return GL_UNSIGNED_INT;
\r
2529 else if(type.isVector())
\r
2531 switch(type.getNominalSize())
\r
2533 case 2: return GL_UNSIGNED_INT_VEC2;
\r
2534 case 3: return GL_UNSIGNED_INT_VEC3;
\r
2535 case 4: return GL_UNSIGNED_INT_VEC4;
\r
2536 default: UNREACHABLE(type.getNominalSize());
\r
2539 else UNREACHABLE(0);
\r
2542 if(type.isScalar())
\r
2546 else if(type.isVector())
\r
2548 switch(type.getNominalSize())
\r
2550 case 2: return GL_BOOL_VEC2;
\r
2551 case 3: return GL_BOOL_VEC3;
\r
2552 case 4: return GL_BOOL_VEC4;
\r
2553 default: UNREACHABLE(type.getNominalSize());
\r
2556 else UNREACHABLE(0);
\r
2558 case EbtSampler2D:
\r
2559 return GL_SAMPLER_2D;
\r
2560 case EbtISampler2D:
\r
2561 return GL_INT_SAMPLER_2D;
\r
2562 case EbtUSampler2D:
\r
2563 return GL_UNSIGNED_INT_SAMPLER_2D;
\r
2564 case EbtSamplerCube:
\r
2565 return GL_SAMPLER_CUBE;
\r
2566 case EbtISamplerCube:
\r
2567 return GL_INT_SAMPLER_CUBE;
\r
2568 case EbtUSamplerCube:
\r
2569 return GL_UNSIGNED_INT_SAMPLER_CUBE;
\r
2570 case EbtSamplerExternalOES:
\r
2571 return GL_SAMPLER_EXTERNAL_OES;
\r
2572 case EbtSampler3D:
\r
2573 return GL_SAMPLER_3D_OES;
\r
2574 case EbtISampler3D:
\r
2575 return GL_INT_SAMPLER_3D;
\r
2576 case EbtUSampler3D:
\r
2577 return GL_UNSIGNED_INT_SAMPLER_3D;
\r
2578 case EbtSampler2DArray:
\r
2579 return GL_SAMPLER_2D_ARRAY;
\r
2580 case EbtISampler2DArray:
\r
2581 return GL_INT_SAMPLER_2D_ARRAY;
\r
2582 case EbtUSampler2DArray:
\r
2583 return GL_UNSIGNED_INT_SAMPLER_2D_ARRAY;
\r
2584 case EbtSampler2DShadow:
\r
2585 return GL_SAMPLER_2D_SHADOW;
\r
2586 case EbtSamplerCubeShadow:
\r
2587 return GL_SAMPLER_CUBE_SHADOW;
\r
2588 case EbtSampler2DArrayShadow:
\r
2589 return GL_SAMPLER_2D_ARRAY_SHADOW;
\r
2591 UNREACHABLE(type.getBasicType());
\r
2598 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2600 if(type.getBasicType() == EbtFloat)
\r
2602 switch(type.getPrecision())
\r
2604 case EbpHigh: return GL_HIGH_FLOAT;
\r
2605 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2606 case EbpLow: return GL_LOW_FLOAT;
\r
2607 case EbpUndefined:
\r
2608 // Should be defined as the default precision by the parser
\r
2609 default: UNREACHABLE(type.getPrecision());
\r
2612 else if(type.getBasicType() == EbtInt)
\r
2614 switch(type.getPrecision())
\r
2616 case EbpHigh: return GL_HIGH_INT;
\r
2617 case EbpMedium: return GL_MEDIUM_INT;
\r
2618 case EbpLow: return GL_LOW_INT;
\r
2619 case EbpUndefined:
\r
2620 // Should be defined as the default precision by the parser
\r
2621 default: UNREACHABLE(type.getPrecision());
\r
2625 // Other types (boolean, sampler) don't have a precision
\r
2629 int OutputASM::dim(TIntermNode *v)
\r
2631 TIntermTyped *vector = v->getAsTyped();
\r
2632 ASSERT(vector && vector->isRegister());
\r
2633 return vector->getNominalSize();
\r
2636 int OutputASM::dim2(TIntermNode *m)
\r
2638 TIntermTyped *matrix = m->getAsTyped();
\r
2639 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2640 return matrix->getSecondarySize();
\r
2643 // Returns ~0 if no loop count could be determined
\r
2644 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2646 // Parse loops of the form:
\r
2647 // for(int index = initial; index [comparator] limit; index += increment)
\r
2648 TIntermSymbol *index = 0;
\r
2649 TOperator comparator = EOpNull;
\r
2652 int increment = 0;
\r
2654 // Parse index name and intial value
\r
2655 if(node->getInit())
\r
2657 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2661 TIntermSequence &sequence = init->getSequence();
\r
2662 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2664 if(variable && variable->getQualifier() == EvqTemporary)
\r
2666 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2668 if(assign->getOp() == EOpInitialize)
\r
2670 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2671 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2673 if(symbol && constant)
\r
2675 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2678 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2686 // Parse comparator and limit value
\r
2687 if(index && node->getCondition())
\r
2689 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2691 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2693 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2697 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2699 comparator = test->getOp();
\r
2700 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2706 // Parse increment
\r
2707 if(index && comparator != EOpNull && node->getExpression())
\r
2709 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2710 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2712 if(binaryTerminal)
\r
2714 TOperator op = binaryTerminal->getOp();
\r
2715 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2719 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2721 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2725 case EOpAddAssign: increment = value; break;
\r
2726 case EOpSubAssign: increment = -value; break;
\r
2727 default: UNIMPLEMENTED();
\r
2732 else if(unaryTerminal)
\r
2734 TOperator op = unaryTerminal->getOp();
\r
2738 case EOpPostIncrement: increment = 1; break;
\r
2739 case EOpPostDecrement: increment = -1; break;
\r
2740 case EOpPreIncrement: increment = 1; break;
\r
2741 case EOpPreDecrement: increment = -1; break;
\r
2742 default: UNIMPLEMENTED();
\r
2747 if(index && comparator != EOpNull && increment != 0)
\r
2749 if(comparator == EOpLessThanEqual)
\r
2751 comparator = EOpLessThan;
\r
2755 if(comparator == EOpLessThan)
\r
2757 int iterations = (limit - initial) / increment;
\r
2759 if(iterations <= 0)
\r
2764 return iterations;
\r
2766 else UNIMPLEMENTED(); // Falls through
\r
2772 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2775 loopDiscontinuity = false;
\r
2777 node->traverse(this);
\r
2779 return loopDiscontinuity;
\r
2782 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2784 if(visit == PreVisit)
\r
2788 else if(visit == PostVisit)
\r
2796 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2798 if(loopDiscontinuity)
\r
2808 switch(node->getFlowOp())
\r
2815 loopDiscontinuity = true;
\r
2817 default: UNREACHABLE(node->getFlowOp());
\r
2820 return !loopDiscontinuity;
\r
2823 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2825 return !loopDiscontinuity;
\r