1 // SwiftShader Software Renderer
\r
3 // Copyright(c) 2005-2013 TransGaming Inc.
\r
5 // All rights reserved. No part of this software may be copied, distributed, transmitted,
\r
6 // transcribed, stored in a retrieval system, translated into any human or computer
\r
7 // language by any means, or disclosed to third parties without the explicit written
\r
8 // agreement of TransGaming Inc. Without such an agreement, no rights or licenses, express
\r
9 // or implied, including but not limited to any patent rights, are granted to you.
\r
12 #include "compiler/OutputASM.h"
\r
14 #include "common/debug.h"
\r
15 #include "compiler/InfoSink.h"
\r
17 #include "libGLESv2/Shader.h"
\r
20 #include <GLES2/gl2.h>
\r
21 #include <GLES2/gl2ext.h>
\r
25 // Integer to TString conversion
\r
29 sprintf(buffer, "%d", i);
\r
33 class Temporary : public TIntermSymbol
\r
36 Temporary(OutputASM *assembler) : TIntermSymbol(0, "tmp", TType(EbtFloat, EbpHigh, EvqTemporary, 4, false, false)), assembler(assembler)
\r
42 assembler->freeTemporary(this);
\r
46 OutputASM *const assembler;
\r
49 class Constant : public TIntermConstantUnion
\r
52 Constant(float x, float y, float z, float w) : TIntermConstantUnion(constants, TType(EbtFloat, EbpHigh, EvqConst, 4, false, false))
\r
54 constants[0].setFConst(x);
\r
55 constants[1].setFConst(y);
\r
56 constants[2].setFConst(z);
\r
57 constants[3].setFConst(w);
\r
60 Constant(bool b) : TIntermConstantUnion(constants, TType(EbtBool, EbpHigh, EvqConst, 1, false, false))
\r
62 constants[0].setBConst(b);
\r
65 Constant(int i) : TIntermConstantUnion(constants, TType(EbtInt, EbpHigh, EvqConst, 1, false, false))
\r
67 constants[0].setIConst(i);
\r
75 ConstantUnion constants[4];
\r
78 Uniform::Uniform(GLenum type, GLenum precision, const std::string &name, int arraySize, int registerIndex)
\r
81 this->precision = precision;
\r
83 this->arraySize = arraySize;
\r
84 this->registerIndex = registerIndex;
\r
87 Attribute::Attribute()
\r
94 Attribute::Attribute(GLenum type, const std::string &name, int arraySize, int registerIndex)
\r
98 this->arraySize = arraySize;
\r
99 this->registerIndex = registerIndex;
\r
102 OutputASM::OutputASM(TParseContext &context, es2::Shader *shaderObject) : TIntermTraverser(true, true, true), mContext(context), shaderObject(shaderObject)
\r
110 shader = shaderObject->getShader();
\r
111 pixelShader = shaderObject->getPixelShader();
\r
112 vertexShader = shaderObject->getVertexShader();
\r
115 functionArray.push_back(Function(0, "main(", 0, 0));
\r
116 currentFunction = 0;
\r
119 OutputASM::~OutputASM()
\r
123 void OutputASM::output()
\r
127 emitShader(GLOBAL);
\r
129 if(functionArray.size() > 1) // Only call main() when there are other functions
\r
131 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
\r
132 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
\r
133 callMain->dst.index = 0; // main()
\r
135 emit(sw::Shader::OPCODE_RET);
\r
138 emitShader(FUNCTION);
\r
142 void OutputASM::emitShader(Scope scope)
\r
145 currentScope = GLOBAL;
\r
146 mContext.treeRoot->traverse(this);
\r
149 void OutputASM::freeTemporary(Temporary *temporary)
\r
151 free(temporaries, temporary);
\r
154 void OutputASM::visitSymbol(TIntermSymbol *symbol)
\r
156 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut)
\r
158 // Vertex varyings don't have to be actively used to successfully link
\r
159 // against pixel shaders that use them. So make sure they're declared.
\r
160 declareVarying(symbol, -1);
\r
164 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
\r
166 if(currentScope != emitScope)
\r
171 TIntermTyped *result = node;
\r
172 TIntermTyped *left = node->getLeft();
\r
173 TIntermTyped *right = node->getRight();
\r
174 const TType &leftType = left->getType();
\r
175 const TType &rightType = right->getType();
\r
176 const TType &resultType = node->getType();
\r
178 switch(node->getOp())
\r
181 if(visit == PostVisit)
\r
183 assignLvalue(left, right);
\r
184 copy(result, right);
\r
187 case EOpInitialize:
\r
188 if(visit == PostVisit)
\r
193 case EOpMatrixTimesScalarAssign:
\r
194 if(visit == PostVisit)
\r
196 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
198 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
199 mul->dst.index += i;
\r
200 argument(mul->src[0], left, i);
\r
203 assignLvalue(left, result);
\r
206 case EOpVectorTimesMatrixAssign:
\r
207 if(visit == PostVisit)
\r
209 int size = leftType.getNominalSize();
\r
211 for(int i = 0; i < size; i++)
\r
213 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
214 dot->dst.mask = 1 << i;
\r
215 argument(dot->src[1], right, i);
\r
218 assignLvalue(left, result);
\r
221 case EOpMatrixTimesMatrixAssign:
\r
222 if(visit == PostVisit)
\r
224 int dim = leftType.getNominalSize();
\r
226 for(int i = 0; i < dim; i++)
\r
228 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
229 mul->dst.index += i;
\r
230 argument(mul->src[1], right, i);
\r
231 mul->src[1].swizzle = 0x00;
\r
233 for(int j = 1; j < dim; j++)
\r
235 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
236 mad->dst.index += i;
\r
237 argument(mad->src[0], left, j);
\r
238 argument(mad->src[1], right, i);
\r
239 mad->src[1].swizzle = j * 0x55;
\r
240 argument(mad->src[2], result, i);
\r
244 assignLvalue(left, result);
\r
247 case EOpIndexDirect:
\r
248 if(visit == PostVisit)
\r
250 int index = right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
252 if(result->isMatrix() || result->isStruct())
\r
254 ASSERT(left->isArray());
\r
255 copy(result, left, index * left->elementRegisterCount());
\r
257 else if(result->isRegister())
\r
259 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
261 if(left->isRegister())
\r
263 mov->src[0].swizzle = index;
\r
265 else if(left->isArray())
\r
267 argument(mov->src[0], left, index * left->elementRegisterCount());
\r
269 else if(left->isMatrix())
\r
271 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
\r
272 argument(mov->src[0], left, index);
\r
274 else UNREACHABLE();
\r
276 else UNREACHABLE();
\r
279 case EOpIndexIndirect:
\r
280 if(visit == PostVisit)
\r
282 if(left->isArray() || left->isMatrix())
\r
284 for(int index = 0; index < result->totalRegisterCount(); index++)
\r
286 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
287 mov->dst.index += index;
\r
288 mov->dst.mask = writeMask(result, index);
\r
289 argument(mov->src[0], left, index);
\r
291 if(left->totalRegisterCount() > 1)
\r
293 sw::Shader::SourceParameter relativeRegister;
\r
294 argument(relativeRegister, right);
\r
296 mov->src[0].rel.type = relativeRegister.type;
\r
297 mov->src[0].rel.index = relativeRegister.index;
\r
298 mov->src[0].rel.scale = result->totalRegisterCount();
\r
299 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
303 else if(left->isRegister())
\r
305 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
\r
307 else UNREACHABLE();
\r
310 case EOpIndexDirectStruct:
\r
311 if(visit == PostVisit)
\r
313 ASSERT(leftType.isStruct());
\r
315 const TTypeList *structure = leftType.getStruct();
\r
316 const TString &fieldName = rightType.getFieldName();
\r
317 int fieldOffset = 0;
\r
319 for(size_t i = 0; i < structure->size(); i++)
\r
321 const TType &fieldType = *(*structure)[i].type;
\r
323 if(fieldType.getFieldName() == fieldName)
\r
328 fieldOffset += fieldType.totalRegisterCount();
\r
331 copy(result, left, fieldOffset);
\r
334 case EOpVectorSwizzle:
\r
335 if(visit == PostVisit)
\r
338 TIntermAggregate *components = right->getAsAggregate();
\r
342 TIntermSequence &sequence = components->getSequence();
\r
345 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
347 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
351 int i = element->getUnionArrayPointer()[0].getIConst();
\r
352 swizzle |= i << (component * 2);
\r
355 else UNREACHABLE();
\r
358 else UNREACHABLE();
\r
360 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
361 mov->src[0].swizzle = swizzle;
\r
364 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
365 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
366 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
367 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
368 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
369 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
370 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
371 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
373 if(visit == PostVisit)
\r
375 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
377 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
379 Temporary equal(this);
\r
380 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
381 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
386 if(visit == PostVisit)
\r
388 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
390 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
392 Temporary notEqual(this);
\r
393 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
394 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
398 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
399 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
400 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
401 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
402 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
403 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
404 case EOpMatrixTimesScalar:
\r
405 if(visit == PostVisit)
\r
407 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
409 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
410 mul->dst.index += i;
\r
411 argument(mul->src[0], left, i);
\r
415 case EOpVectorTimesMatrix:
\r
416 if(visit == PostVisit)
\r
418 int size = leftType.getNominalSize();
\r
420 for(int i = 0; i < size; i++)
\r
422 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
423 dot->dst.mask = 1 << i;
\r
424 argument(dot->src[1], right, i);
\r
428 case EOpMatrixTimesVector:
\r
429 if(visit == PostVisit)
\r
431 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
432 mul->src[1].swizzle = 0x00;
\r
434 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
436 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
437 argument(mad->src[0], left, i);
\r
438 mad->src[1].swizzle = i * 0x55;
\r
442 case EOpMatrixTimesMatrix:
\r
443 if(visit == PostVisit)
\r
445 int dim = leftType.getNominalSize();
\r
447 for(int i = 0; i < dim; i++)
\r
449 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
450 mul->dst.index += i;
\r
451 argument(mul->src[1], right, i);
\r
452 mul->src[1].swizzle = 0x00;
\r
454 for(int j = 1; j < dim; j++)
\r
456 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
457 mad->dst.index += i;
\r
458 argument(mad->src[0], left, j);
\r
459 argument(mad->src[1], right, i);
\r
460 mad->src[1].swizzle = j * 0x55;
\r
461 argument(mad->src[2], result, i);
\r
467 if(trivial(right, 6))
\r
469 if(visit == PostVisit)
\r
471 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
474 else // Short-circuit evaluation
\r
476 if(visit == InVisit)
\r
478 emit(sw::Shader::OPCODE_MOV, result, left);
\r
479 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
480 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
482 else if(visit == PostVisit)
\r
484 emit(sw::Shader::OPCODE_MOV, result, right);
\r
485 emit(sw::Shader::OPCODE_ENDIF);
\r
489 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
490 case EOpLogicalAnd:
\r
491 if(trivial(right, 6))
\r
493 if(visit == PostVisit)
\r
495 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
498 else // Short-circuit evaluation
\r
500 if(visit == InVisit)
\r
502 emit(sw::Shader::OPCODE_MOV, result, left);
\r
503 emit(sw::Shader::OPCODE_IF, 0, result);
\r
505 else if(visit == PostVisit)
\r
507 emit(sw::Shader::OPCODE_MOV, result, right);
\r
508 emit(sw::Shader::OPCODE_ENDIF);
\r
512 default: UNREACHABLE();
\r
518 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
520 if(currentScope != emitScope)
\r
525 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
526 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
527 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
529 TIntermTyped *result = node;
\r
530 TIntermTyped *arg = node->getOperand();
\r
532 switch(node->getOp())
\r
535 if(visit == PostVisit)
\r
537 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
539 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
540 neg->dst.index += index;
\r
541 argument(neg->src[0], arg, index);
\r
542 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
546 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
547 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
548 case EOpPostIncrement:
\r
549 if(visit == PostVisit)
\r
553 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
555 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
556 add->dst.index += index;
\r
557 argument(add->src[0], arg, index);
\r
560 assignLvalue(arg, arg);
\r
563 case EOpPostDecrement:
\r
564 if(visit == PostVisit)
\r
568 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
570 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
571 sub->dst.index += index;
\r
572 argument(sub->src[0], arg, index);
\r
575 assignLvalue(arg, arg);
\r
578 case EOpPreIncrement:
\r
579 if(visit == PostVisit)
\r
581 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
583 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
584 add->dst.index += index;
\r
585 argument(add->src[0], arg, index);
\r
588 assignLvalue(arg, result);
\r
591 case EOpPreDecrement:
\r
592 if(visit == PostVisit)
\r
594 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
596 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
597 sub->dst.index += index;
\r
598 argument(sub->src[0], arg, index);
\r
601 assignLvalue(arg, result);
\r
604 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
605 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
606 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
607 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
608 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
609 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
610 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
611 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
612 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
613 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
614 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
615 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
616 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
617 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
618 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
619 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
620 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
621 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
622 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
623 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
624 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
625 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
626 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
627 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
628 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
629 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
630 default: UNREACHABLE();
\r
636 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
638 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
643 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
645 TIntermTyped *result = node;
\r
646 const TType &resultType = node->getType();
\r
647 TIntermSequence &arg = node->getSequence();
\r
648 int argumentCount = arg.size();
\r
650 switch(node->getOp())
\r
652 case EOpSequence: break;
\r
653 case EOpDeclaration: break;
\r
654 case EOpPrototype: break;
\r
656 if(visit == PostVisit)
\r
658 copy(result, arg[1]);
\r
662 if(visit == PreVisit)
\r
664 const TString &name = node->getName();
\r
666 if(emitScope == FUNCTION)
\r
668 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
670 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
671 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
673 const Function *function = findFunction(name);
\r
674 ASSERT(function); // Should have been added during global pass
\r
675 label->dst.index = function->label;
\r
676 currentFunction = function->label;
\r
679 else if(emitScope == GLOBAL)
\r
681 if(name != "main(")
\r
683 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
684 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
687 else UNREACHABLE();
\r
689 currentScope = FUNCTION;
\r
691 else if(visit == PostVisit)
\r
693 if(emitScope == FUNCTION)
\r
695 if(functionArray.size() > 1) // No need to return when there's only main()
\r
697 emit(sw::Shader::OPCODE_RET);
\r
701 currentScope = GLOBAL;
\r
704 case EOpFunctionCall:
\r
705 if(visit == PostVisit)
\r
707 if(node->isUserDefined())
\r
709 const TString &name = node->getName();
\r
710 const Function *function = findFunction(name);
\r
714 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
718 TIntermSequence &arguments = *function->arg;
\r
720 for(int i = 0; i < argumentCount; i++)
\r
722 TIntermTyped *in = arguments[i]->getAsTyped();
\r
724 if(in->getQualifier() == EvqIn ||
\r
725 in->getQualifier() == EvqInOut ||
\r
726 in->getQualifier() == EvqConstReadOnly)
\r
732 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
733 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
734 call->dst.index = function->label;
\r
736 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
738 copy(result, function->ret);
\r
741 for(int i = 0; i < argumentCount; i++)
\r
743 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
744 TIntermTyped *out = arg[i]->getAsTyped();
\r
746 if(argument->getQualifier() == EvqOut ||
\r
747 argument->getQualifier() == EvqInOut)
\r
749 copy(out, argument);
\r
755 TString name = TFunction::unmangleName(node->getName());
\r
757 if(name == "texture2D" || name == "textureCube")
\r
759 if(argumentCount == 2)
\r
761 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
763 else if(argumentCount == 3) // bias
\r
765 Temporary uvwb(this);
\r
766 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
767 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
768 bias->dst.mask = 0x8;
\r
770 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
773 else UNREACHABLE();
\r
775 else if(name == "texture2DProj")
\r
777 TIntermTyped *t = arg[1]->getAsTyped();
\r
779 if(argumentCount == 2)
\r
781 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
782 tex->project = true;
\r
784 if(t->getNominalSize() == 3)
\r
786 tex->src[0].swizzle = 0xA4;
\r
788 else ASSERT(t->getNominalSize() == 4);
\r
790 else if(argumentCount == 3) // bias
\r
792 Temporary proj(this);
\r
794 if(t->getNominalSize() == 3)
\r
796 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
797 div->src[1].swizzle = 0xAA;
\r
798 div->dst.mask = 0x3;
\r
800 else if(t->getNominalSize() == 4)
\r
802 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
803 div->src[1].swizzle = 0xFF;
\r
804 div->dst.mask = 0x3;
\r
806 else UNREACHABLE();
\r
808 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
809 bias->dst.mask = 0x8;
\r
811 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
814 else UNREACHABLE();
\r
816 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
818 Temporary uvwb(this);
\r
819 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
820 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
821 lod->dst.mask = 0x8;
\r
823 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
825 else if(name == "texture2DProjLod")
\r
827 TIntermTyped *t = arg[1]->getAsTyped();
\r
828 Temporary proj(this);
\r
830 if(t->getNominalSize() == 3)
\r
832 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
833 div->src[1].swizzle = 0xAA;
\r
834 div->dst.mask = 0x3;
\r
836 else if(t->getNominalSize() == 4)
\r
838 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
839 div->src[1].swizzle = 0xFF;
\r
840 div->dst.mask = 0x3;
\r
842 else UNREACHABLE();
\r
844 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
845 lod->dst.mask = 0x8;
\r
847 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
849 else UNREACHABLE();
\r
853 case EOpParameters:
\r
855 case EOpConstructFloat:
\r
856 case EOpConstructVec2:
\r
857 case EOpConstructVec3:
\r
858 case EOpConstructVec4:
\r
859 case EOpConstructBool:
\r
860 case EOpConstructBVec2:
\r
861 case EOpConstructBVec3:
\r
862 case EOpConstructBVec4:
\r
863 case EOpConstructInt:
\r
864 case EOpConstructIVec2:
\r
865 case EOpConstructIVec3:
\r
866 case EOpConstructIVec4:
\r
867 if(visit == PostVisit)
\r
871 for(int i = 0; i < argumentCount; i++)
\r
873 TIntermTyped *argi = arg[i]->getAsTyped();
\r
874 int size = argi->getNominalSize();
\r
876 if(!argi->isMatrix())
\r
878 Instruction *mov = emitCast(result, argi);
\r
879 mov->dst.mask = (0xF << component) & 0xF;
\r
880 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
888 while(component < resultType.getNominalSize())
\r
890 Instruction *mov = emitCast(result, argi);
\r
891 mov->dst.mask = (0xF << component) & 0xF;
\r
892 mov->src[0].index += column;
\r
893 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
902 case EOpConstructMat2:
\r
903 case EOpConstructMat3:
\r
904 case EOpConstructMat4:
\r
905 if(visit == PostVisit)
\r
907 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
908 const int dim = result->getNominalSize();
\r
910 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
912 for(int i = 0; i < dim; i++)
\r
914 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
915 init->dst.index += i;
\r
916 Instruction *mov = emitCast(result, arg0);
\r
917 mov->dst.index += i;
\r
918 mov->dst.mask = 1 << i;
\r
919 ASSERT(mov->src[0].swizzle == 0x00);
\r
922 else if(arg0->isMatrix())
\r
924 for(int i = 0; i < dim; i++)
\r
926 if(dim > dim2(arg0))
\r
928 // Initialize to identity matrix
\r
929 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
930 Instruction *mov = emitCast(result, &col);
\r
931 mov->dst.index += i;
\r
936 Instruction *mov = emitCast(result, arg0);
\r
937 mov->dst.index += i;
\r
938 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
939 argument(mov->src[0], arg0, i);
\r
948 for(int i = 0; i < argumentCount; i++)
\r
950 TIntermTyped *argi = arg[i]->getAsTyped();
\r
951 int size = argi->getNominalSize();
\r
954 while(element < size)
\r
956 Instruction *mov = emitCast(result, argi);
\r
957 mov->dst.index += column;
\r
958 mov->dst.mask = (0xF << row) & 0xF;
\r
959 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
961 int end = row + size - element;
\r
962 column = end >= dim ? column + 1 : column;
\r
963 element = element + dim - row;
\r
964 row = end >= dim ? 0 : end;
\r
970 case EOpConstructStruct:
\r
971 if(visit == PostVisit)
\r
974 for(int i = 0; i < argumentCount; i++)
\r
976 TIntermTyped *argi = arg[i]->getAsTyped();
\r
977 int size = argi->totalRegisterCount();
\r
979 for(int index = 0; index < size; index++)
\r
981 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
982 mov->dst.index += index + offset;
\r
983 mov->dst.mask = writeMask(result, offset + index);
\r
984 argument(mov->src[0], argi, index);
\r
991 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
992 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
993 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
994 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
995 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
996 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
997 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
998 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
999 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1000 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1001 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1003 if(visit == PostVisit)
\r
1005 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1006 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1009 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1010 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1011 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1012 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1013 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1014 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1015 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1016 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1017 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1019 if(visit == PostVisit)
\r
1021 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1023 for(int i = 0; i < dim2(arg[0]); i++)
\r
1025 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1026 mul->dst.index += i;
\r
1027 argument(mul->src[0], arg[0], i);
\r
1028 argument(mul->src[1], arg[1], i);
\r
1032 default: UNREACHABLE();
\r
1038 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1040 if(currentScope != emitScope)
\r
1045 TIntermTyped *condition = node->getCondition();
\r
1046 TIntermNode *trueBlock = node->getTrueBlock();
\r
1047 TIntermNode *falseBlock = node->getFalseBlock();
\r
1048 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1050 condition->traverse(this);
\r
1052 if(node->usesTernaryOperator())
\r
1054 if(constantCondition)
\r
1056 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1060 trueBlock->traverse(this);
\r
1061 copy(node, trueBlock);
\r
1065 falseBlock->traverse(this);
\r
1066 copy(node, falseBlock);
\r
1069 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1071 trueBlock->traverse(this);
\r
1072 falseBlock->traverse(this);
\r
1073 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1077 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1081 trueBlock->traverse(this);
\r
1082 copy(node, trueBlock);
\r
1087 emit(sw::Shader::OPCODE_ELSE);
\r
1088 falseBlock->traverse(this);
\r
1089 copy(node, falseBlock);
\r
1092 emit(sw::Shader::OPCODE_ENDIF);
\r
1095 else // if/else statement
\r
1097 if(constantCondition)
\r
1099 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1105 trueBlock->traverse(this);
\r
1112 falseBlock->traverse(this);
\r
1118 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1122 trueBlock->traverse(this);
\r
1127 emit(sw::Shader::OPCODE_ELSE);
\r
1128 falseBlock->traverse(this);
\r
1131 emit(sw::Shader::OPCODE_ENDIF);
\r
1138 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1140 if(currentScope != emitScope)
\r
1145 unsigned int iterations = loopCount(node);
\r
1147 if(iterations == 0)
\r
1152 bool unroll = (iterations <= 4);
\r
1156 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1157 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1160 TIntermNode *init = node->getInit();
\r
1161 TIntermTyped *condition = node->getCondition();
\r
1162 TIntermTyped *expression = node->getExpression();
\r
1163 TIntermNode *body = node->getBody();
\r
1165 if(node->getType() == ELoopDoWhile)
\r
1167 Temporary iterate(this);
\r
1168 Constant True(true);
\r
1169 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1171 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1175 body->traverse(this);
\r
1178 emit(sw::Shader::OPCODE_TEST);
\r
1180 condition->traverse(this);
\r
1181 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1183 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1189 init->traverse(this);
\r
1194 for(unsigned int i = 0; i < iterations; i++)
\r
1196 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1200 body->traverse(this);
\r
1205 expression->traverse(this);
\r
1211 condition->traverse(this);
\r
1213 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1217 body->traverse(this);
\r
1220 emit(sw::Shader::OPCODE_TEST);
\r
1224 expression->traverse(this);
\r
1227 condition->traverse(this);
\r
1229 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1236 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1238 if(currentScope != emitScope)
\r
1243 switch(node->getFlowOp())
\r
1245 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1246 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1247 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1249 if(visit == PostVisit)
\r
1251 TIntermTyped *value = node->getExpression();
\r
1255 copy(functionArray[currentFunction].ret, value);
\r
1258 emit(sw::Shader::OPCODE_LEAVE);
\r
1261 default: UNREACHABLE();
\r
1267 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1269 if(dst && registerType(dst) == sw::Shader::PARAMETER_SAMPLER)
\r
1271 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1274 Instruction *instruction = new Instruction(op);
\r
1278 instruction->dst.type = registerType(dst);
\r
1279 instruction->dst.index = registerIndex(dst) + index;
\r
1280 instruction->dst.mask = writeMask(dst);
\r
1281 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1284 argument(instruction->src[0], src0, index);
\r
1285 argument(instruction->src[1], src1, index);
\r
1286 argument(instruction->src[2], src2, index);
\r
1288 shader->append(instruction);
\r
1290 return instruction;
\r
1293 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1295 // Integers are implemented as float
\r
1296 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1298 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1300 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1302 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1304 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1306 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1309 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1312 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1314 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1316 emit(op, dst, src0, src1, src2, index);
\r
1320 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1322 emitBinary(op, result, src0, src1);
\r
1323 assignLvalue(lhs, result);
\r
1326 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1328 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1329 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1331 Instruction *cmp = emit(opcode, dst, left, right);
\r
1332 cmp->control = cmpOp;
\r
1333 argument(cmp->src[0], left, index);
\r
1334 argument(cmp->src[1], right, index);
\r
1337 int componentCount(const TType &type, int registers)
\r
1339 if(registers == 0)
\r
1344 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1346 int index = registers / type.elementRegisterCount();
\r
1347 registers -= index * type.elementRegisterCount();
\r
1348 return index * type.getElementSize() + componentCount(type, registers);
\r
1351 if(type.isStruct())
\r
1353 TTypeList *structure = type.getStruct();
\r
1356 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1358 const TType &fieldType = *field->type;
\r
1360 if(fieldType.totalRegisterCount() <= registers)
\r
1362 registers -= fieldType.totalRegisterCount();
\r
1363 elements += fieldType.getObjectSize();
\r
1365 else // Register within this field
\r
1367 return elements + componentCount(fieldType, registers);
\r
1371 else if(type.isMatrix())
\r
1373 return registers * type.getNominalSize();
\r
1380 int registerSize(const TType &type, int registers)
\r
1382 if(registers == 0)
\r
1384 if(type.isStruct())
\r
1386 return registerSize(*type.getStruct()->begin()->type, 0);
\r
1389 return type.getNominalSize();
\r
1392 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1394 int index = registers / type.elementRegisterCount();
\r
1395 registers -= index * type.elementRegisterCount();
\r
1396 return registerSize(type, registers);
\r
1399 if(type.isStruct())
\r
1401 TTypeList *structure = type.getStruct();
\r
1404 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1406 const TType &fieldType = *field->type;
\r
1408 if(fieldType.totalRegisterCount() <= registers)
\r
1410 registers -= fieldType.totalRegisterCount();
\r
1411 elements += fieldType.getObjectSize();
\r
1413 else // Register within this field
\r
1415 return registerSize(fieldType, registers);
\r
1419 else if(type.isMatrix())
\r
1421 return registerSize(type, 0);
\r
1428 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1432 TIntermTyped *arg = argument->getAsTyped();
\r
1433 const TType &type = arg->getType();
\r
1434 const TTypeList *structure = type.getStruct();
\r
1435 ASSERT(index < arg->totalRegisterCount());
\r
1437 int size = registerSize(type, index);
\r
1439 parameter.type = registerType(arg);
\r
1441 if(arg->getQualifier() == EvqConst)
\r
1443 int component = componentCount(type, index);
\r
1444 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1446 for(int i = 0; i < 4; i++)
\r
1448 if(size == 1) // Replicate
\r
1450 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1454 parameter.value[i] = constants[component + i].getAsFloat();
\r
1458 parameter.value[i] = 0.0f;
\r
1464 parameter.index = registerIndex(arg) + index;
\r
1466 if(registerType(arg) == sw::Shader::PARAMETER_SAMPLER)
\r
1468 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1472 TIntermTyped *left = binary->getLeft();
\r
1473 TIntermTyped *right = binary->getRight();
\r
1475 if(binary->getOp() == EOpIndexDirect)
\r
1477 parameter.index += right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1479 else if(binary->getOp() == EOpIndexIndirect)
\r
1481 if(left->getArraySize() > 1)
\r
1483 parameter.rel.type = registerType(binary->getRight());
\r
1484 parameter.rel.index = registerIndex(binary->getRight());
\r
1485 parameter.rel.scale = 1;
\r
1486 parameter.rel.deterministic = true;
\r
1489 else UNREACHABLE();
\r
1494 if(!IsSampler(arg->getBasicType()))
\r
1496 parameter.swizzle = readSwizzle(arg, size);
\r
1501 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1503 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1505 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1506 mov->dst.index += index;
\r
1507 mov->dst.mask = writeMask(dst, index);
\r
1508 argument(mov->src[0], src, offset + index);
\r
1512 int swizzleElement(int swizzle, int index)
\r
1514 return (swizzle >> (index * 2)) & 0x03;
\r
1517 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1519 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1520 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1521 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1522 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1525 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermNode *src)
\r
1527 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1529 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1531 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1533 Temporary address(this);
\r
1534 lvalue(insert->dst, address, dst);
\r
1536 insert->src[0].type = insert->dst.type;
\r
1537 insert->src[0].index = insert->dst.index;
\r
1538 insert->src[0].rel = insert->dst.rel;
\r
1539 argument(insert->src[1], src);
\r
1540 argument(insert->src[2], binary->getRight());
\r
1542 shader->append(insert);
\r
1546 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1548 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1550 Temporary address(this);
\r
1551 int swizzle = lvalue(mov->dst, address, dst);
\r
1552 mov->dst.index += offset;
\r
1556 mov->dst.mask = writeMask(dst, offset);
\r
1559 argument(mov->src[0], src, offset);
\r
1560 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1562 shader->append(mov);
\r
1567 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1569 TIntermTyped *result = node;
\r
1570 TIntermBinary *binary = node->getAsBinaryNode();
\r
1571 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1575 TIntermTyped *left = binary->getLeft();
\r
1576 TIntermTyped *right = binary->getRight();
\r
1578 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1580 switch(binary->getOp())
\r
1582 case EOpIndexDirect:
\r
1584 int rightIndex = right->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1586 if(left->isRegister())
\r
1588 int leftMask = dst.mask;
\r
1591 while((leftMask & dst.mask) == 0)
\r
1593 dst.mask = dst.mask << 1;
\r
1596 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1597 dst.mask = 1 << element;
\r
1601 else if(left->isArray() || left->isMatrix())
\r
1603 dst.index += rightIndex * result->totalRegisterCount();
\r
1606 else UNREACHABLE();
\r
1609 case EOpIndexIndirect:
\r
1611 if(left->isRegister())
\r
1613 // Requires INSERT instruction (handled by calling function)
\r
1615 else if(left->isArray() || left->isMatrix())
\r
1617 int scale = result->totalRegisterCount();
\r
1619 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1621 if(left->totalRegisterCount() > 1)
\r
1623 sw::Shader::SourceParameter relativeRegister;
\r
1624 argument(relativeRegister, right);
\r
1626 dst.rel.index = relativeRegister.index;
\r
1627 dst.rel.type = relativeRegister.type;
\r
1628 dst.rel.scale = scale;
\r
1629 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1632 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1636 Constant oldScale((int)dst.rel.scale);
\r
1637 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1638 mad->src[0].index = dst.rel.index;
\r
1639 mad->src[0].type = dst.rel.type;
\r
1643 Constant oldScale((int)dst.rel.scale);
\r
1644 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1645 mul->src[0].index = dst.rel.index;
\r
1646 mul->src[0].type = dst.rel.type;
\r
1648 Constant newScale(scale);
\r
1649 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1652 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1653 dst.rel.index = registerIndex(&address);
\r
1654 dst.rel.scale = 1;
\r
1656 else // Just add the new index to the address register
\r
1660 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1664 Constant newScale(scale);
\r
1665 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1669 else UNREACHABLE();
\r
1672 case EOpIndexDirectStruct:
\r
1674 const TTypeList *structure = left->getType().getStruct();
\r
1675 const TString &fieldName = right->getType().getFieldName();
\r
1678 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1680 if(field->type->getFieldName() == fieldName)
\r
1682 dst.type = registerType(left);
\r
1683 dst.index += offset;
\r
1684 dst.mask = writeMask(right);
\r
1689 offset += field->type->totalRegisterCount();
\r
1693 case EOpVectorSwizzle:
\r
1695 ASSERT(left->isRegister());
\r
1697 int leftMask = dst.mask;
\r
1700 int rightMask = 0;
\r
1702 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1704 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1706 int index = sequence[i]->getAsConstantUnion()->getUnionArrayPointer()->getIConst();
\r
1708 int element = swizzleElement(leftSwizzle, index);
\r
1709 rightMask = rightMask | (1 << element);
\r
1710 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1713 dst.mask = leftMask & rightMask;
\r
1719 UNREACHABLE(); // Not an l-value operator
\r
1725 dst.type = registerType(symbol);
\r
1726 dst.index = registerIndex(symbol);
\r
1727 dst.mask = writeMask(symbol);
\r
1734 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1736 if(IsSampler(operand->getBasicType()) && (operand->getQualifier() == EvqUniform || operand->getQualifier() == EvqTemporary)) // Function parameters are temporaries
\r
1738 return sw::Shader::PARAMETER_SAMPLER;
\r
1741 switch(operand->getQualifier())
\r
1743 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1744 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1745 case EvqConst: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1746 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1747 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1748 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1749 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1750 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1751 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1752 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1753 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1754 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1755 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1756 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1757 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1758 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1759 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1760 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1761 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1762 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1763 default: UNREACHABLE();
\r
1766 return sw::Shader::PARAMETER_VOID;
\r
1769 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1771 if(registerType(operand) == sw::Shader::PARAMETER_SAMPLER)
\r
1773 return samplerRegister(operand);
\r
1776 switch(operand->getQualifier())
\r
1778 case EvqTemporary: return temporaryRegister(operand);
\r
1779 case EvqGlobal: return temporaryRegister(operand);
\r
1780 case EvqConst: UNREACHABLE();
\r
1781 case EvqAttribute: return attributeRegister(operand);
\r
1782 case EvqVaryingIn: return varyingRegister(operand);
\r
1783 case EvqVaryingOut: return varyingRegister(operand);
\r
1784 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1785 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1786 case EvqUniform: return uniformRegister(operand);
\r
1787 case EvqIn: return temporaryRegister(operand);
\r
1788 case EvqOut: return temporaryRegister(operand);
\r
1789 case EvqInOut: return temporaryRegister(operand);
\r
1790 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1791 case EvqPosition: return varyingRegister(operand);
\r
1792 case EvqPointSize: return varyingRegister(operand);
\r
1793 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1794 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1795 case EvqPointCoord: return varyingRegister(operand);
\r
1796 case EvqFragColor: return 0;
\r
1797 case EvqFragData: return 0;
\r
1798 default: UNREACHABLE();
\r
1804 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1806 if(destination->getQualifier() == EvqPointSize)
\r
1808 return 0x2; // Point size stored in the y component
\r
1811 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1814 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1816 if(argument->getQualifier() == EvqPointSize)
\r
1818 return 0x55; // Point size stored in the y component
\r
1821 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1823 return swizzleSize[size];
\r
1826 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1827 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1829 if(!expression->isRegister())
\r
1834 return cost(expression, budget) >= 0;
\r
1837 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1838 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1845 if(expression->getAsSymbolNode())
\r
1849 else if(expression->getAsConstantUnion())
\r
1853 else if(expression->getAsBinaryNode())
\r
1855 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1857 switch(binary->getOp())
\r
1859 case EOpVectorSwizzle:
\r
1860 case EOpIndexDirect:
\r
1861 case EOpIndexDirectStruct:
\r
1862 return cost(binary->getLeft(), budget - 0);
\r
1866 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1871 else if(expression->getAsUnaryNode())
\r
1873 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1875 switch(unary->getOp())
\r
1879 return cost(unary->getOperand(), budget - 1);
\r
1884 else if(expression->getAsSelectionNode())
\r
1886 TIntermSelection *selection = expression->getAsSelectionNode();
\r
1888 if(selection->usesTernaryOperator())
\r
1890 TIntermTyped *condition = selection->getCondition();
\r
1891 TIntermNode *trueBlock = selection->getTrueBlock();
\r
1892 TIntermNode *falseBlock = selection->getFalseBlock();
\r
1893 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1895 if(constantCondition)
\r
1897 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1901 return cost(trueBlock, budget - 0);
\r
1905 return cost(falseBlock, budget - 0);
\r
1910 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
1918 const Function *OutputASM::findFunction(const TString &name)
\r
1920 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
1922 if(functionArray[f].name == name)
\r
1924 return &functionArray[f];
\r
1931 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
1933 return allocate(temporaries, temporary);
\r
1936 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
1938 int var = lookup(varyings, varying);
\r
1942 var = allocate(varyings, varying);
\r
1943 int componentCount = varying->getNominalSize();
\r
1944 int registerCount = varying->totalRegisterCount();
\r
1946 if(pixelShader && (var + registerCount) <= sw::PixelShader::MAX_INPUT_VARYINGS)
\r
1948 if(varying->getQualifier() == EvqPointCoord)
\r
1950 ASSERT(varying->isRegister());
\r
1951 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1952 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1953 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1954 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
1958 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
1960 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1961 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1962 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1963 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
1967 else if(vertexShader && (var + registerCount) <= sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
1969 if(varying->getQualifier() == EvqPosition)
\r
1971 ASSERT(varying->isRegister());
\r
1972 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
1973 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
1974 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
1975 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
1976 vertexShader->positionRegister = var;
\r
1978 else if(varying->getQualifier() == EvqPointSize)
\r
1980 ASSERT(varying->isRegister());
\r
1981 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
1982 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
1983 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
1984 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
1985 vertexShader->pointSizeRegister = var;
\r
1989 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
1992 else UNREACHABLE();
\r
1994 declareVarying(varying, var);
\r
2000 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2002 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2004 const TType &type = varying->getType();
\r
2005 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2006 es2::VaryingList &activeVaryings = shaderObject->varyings;
\r
2008 // Check if this varying has been declared before without having a register assigned
\r
2009 for(es2::VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2011 if(v->name == name)
\r
2015 ASSERT(v->reg < 0 || v->reg == reg);
\r
2023 activeVaryings.push_back(es2::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2027 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2029 const TType &type = uniform->getType();
\r
2030 ASSERT(!IsSampler(type.getBasicType()));
\r
2031 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2036 int index = lookup(uniforms, uniform);
\r
2040 index = allocate(uniforms, uniform);
\r
2041 const TString &name = symbol->getSymbol().c_str();
\r
2043 declareUniform(type, name, index);
\r
2052 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2054 ASSERT(!attribute->isArray());
\r
2055 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2057 int index = lookup(attributes, attribute);
\r
2061 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2066 index = allocate(attributes, attribute);
\r
2067 const TType &type = attribute->getType();
\r
2068 int registerCount = attribute->totalRegisterCount();
\r
2070 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2072 for(int i = 0; i < registerCount; i++)
\r
2074 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2078 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2080 const char *name = symbol->getSymbol().c_str();
\r
2081 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2088 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2090 const TType &type = sampler->getType();
\r
2091 ASSERT(IsSampler(type.getBasicType()));
\r
2092 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2093 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2097 int index = lookup(samplers, sampler);
\r
2101 index = allocate(samplers, sampler);
\r
2102 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2103 const char *name = symbol->getSymbol().c_str();
\r
2104 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name, sampler->getArraySize(), index));
\r
2106 for(int i = 0; i < sampler->totalRegisterCount(); i++)
\r
2108 shader->declareSampler(index + i);
\r
2116 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect);
\r
2118 return samplerRegister(binary->getLeft()); // Index added later
\r
2120 else UNREACHABLE();
\r
2125 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2127 for(unsigned int i = 0; i < list.size(); i++)
\r
2129 if(list[i] == variable)
\r
2131 return i; // Pointer match
\r
2135 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2139 for(unsigned int i = 0; i < list.size(); i++)
\r
2143 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2147 if(listSymbol->getId() == varSymbol->getId())
\r
2149 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2150 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2151 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2163 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2165 int index = lookup(list, variable);
\r
2169 unsigned int registerCount = variable->totalRegisterCount();
\r
2171 for(unsigned int i = 0; i < list.size(); i++)
\r
2175 unsigned int j = 1;
\r
2176 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2178 if(list[i + j] != 0)
\r
2184 if(j == registerCount) // Found free slots
\r
2186 for(unsigned int j = 0; j < registerCount; j++)
\r
2188 list[i + j] = variable;
\r
2196 index = list.size();
\r
2198 for(unsigned int i = 0; i < registerCount; i++)
\r
2200 list.push_back(variable);
\r
2207 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2209 int index = lookup(list, variable);
\r
2217 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2219 const TTypeList *structure = type.getStruct();
\r
2220 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2224 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2228 if(type.isArray())
\r
2230 int elementIndex = index;
\r
2232 for(int i = 0; i < type.getArraySize(); i++)
\r
2234 for(size_t j = 0; j < structure->size(); j++)
\r
2236 const TType &fieldType = *(*structure)[j].type;
\r
2237 const TString &fieldName = fieldType.getFieldName();
\r
2239 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2240 declareUniform(fieldType, uniformName, elementIndex);
\r
2241 elementIndex += fieldType.totalRegisterCount();
\r
2247 int fieldIndex = index;
\r
2249 for(size_t i = 0; i < structure->size(); i++)
\r
2251 const TType &fieldType = *(*structure)[i].type;
\r
2252 const TString &fieldName = fieldType.getFieldName();
\r
2254 const TString uniformName = name + "." + fieldName;
\r
2255 declareUniform(fieldType, uniformName, fieldIndex);
\r
2256 fieldIndex += fieldType.totalRegisterCount();
\r
2262 GLenum OutputASM::glVariableType(const TType &type)
\r
2264 if(type.getBasicType() == EbtFloat)
\r
2266 if(type.isScalar())
\r
2270 else if(type.isVector())
\r
2272 switch(type.getNominalSize())
\r
2274 case 2: return GL_FLOAT_VEC2;
\r
2275 case 3: return GL_FLOAT_VEC3;
\r
2276 case 4: return GL_FLOAT_VEC4;
\r
2277 default: UNREACHABLE();
\r
2280 else if(type.isMatrix())
\r
2282 switch(type.getNominalSize())
\r
2284 case 2: return GL_FLOAT_MAT2;
\r
2285 case 3: return GL_FLOAT_MAT3;
\r
2286 case 4: return GL_FLOAT_MAT4;
\r
2287 default: UNREACHABLE();
\r
2290 else UNREACHABLE();
\r
2292 else if(type.getBasicType() == EbtInt)
\r
2294 if(type.isScalar())
\r
2298 else if(type.isVector())
\r
2300 switch(type.getNominalSize())
\r
2302 case 2: return GL_INT_VEC2;
\r
2303 case 3: return GL_INT_VEC3;
\r
2304 case 4: return GL_INT_VEC4;
\r
2305 default: UNREACHABLE();
\r
2308 else UNREACHABLE();
\r
2310 else if(type.getBasicType() == EbtBool)
\r
2312 if(type.isScalar())
\r
2316 else if(type.isVector())
\r
2318 switch(type.getNominalSize())
\r
2320 case 2: return GL_BOOL_VEC2;
\r
2321 case 3: return GL_BOOL_VEC3;
\r
2322 case 4: return GL_BOOL_VEC4;
\r
2323 default: UNREACHABLE();
\r
2326 else UNREACHABLE();
\r
2328 else if(type.getBasicType() == EbtSampler2D)
\r
2330 return GL_SAMPLER_2D;
\r
2332 else if(type.getBasicType() == EbtSamplerCube)
\r
2334 return GL_SAMPLER_CUBE;
\r
2336 else if(type.getBasicType() == EbtSamplerExternalOES)
\r
2338 return GL_SAMPLER_EXTERNAL_OES;
\r
2340 else UNREACHABLE();
\r
2345 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2347 if(type.getBasicType() == EbtFloat)
\r
2349 switch(type.getPrecision())
\r
2351 case EbpHigh: return GL_HIGH_FLOAT;
\r
2352 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2353 case EbpLow: return GL_LOW_FLOAT;
\r
2354 case EbpUndefined:
\r
2355 // Should be defined as the default precision by the parser
\r
2356 default: UNREACHABLE();
\r
2359 else if (type.getBasicType() == EbtInt)
\r
2361 switch (type.getPrecision())
\r
2363 case EbpHigh: return GL_HIGH_INT;
\r
2364 case EbpMedium: return GL_MEDIUM_INT;
\r
2365 case EbpLow: return GL_LOW_INT;
\r
2366 case EbpUndefined:
\r
2367 // Should be defined as the default precision by the parser
\r
2368 default: UNREACHABLE();
\r
2372 // Other types (boolean, sampler) don't have a precision
\r
2376 int OutputASM::dim(TIntermNode *v)
\r
2378 TIntermTyped *vector = v->getAsTyped();
\r
2379 ASSERT(vector && vector->isRegister());
\r
2380 return vector->getNominalSize();
\r
2383 int OutputASM::dim2(TIntermNode *m)
\r
2385 TIntermTyped *matrix = m->getAsTyped();
\r
2386 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2387 return matrix->getNominalSize();
\r
2390 // Returns ~0 if no loop count could be determined
\r
2391 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2393 // Parse loops of the form:
\r
2394 // for(int index = initial; index [comparator] limit; index += increment)
\r
2395 TIntermSymbol *index = 0;
\r
2396 TOperator comparator = EOpNull;
\r
2399 int increment = 0;
\r
2401 // Parse index name and intial value
\r
2402 if(node->getInit())
\r
2404 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2408 TIntermSequence &sequence = init->getSequence();
\r
2409 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2411 if(variable && variable->getQualifier() == EvqTemporary)
\r
2413 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2415 if(assign->getOp() == EOpInitialize)
\r
2417 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2418 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2420 if(symbol && constant)
\r
2422 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2425 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2433 // Parse comparator and limit value
\r
2434 if(index && node->getCondition())
\r
2436 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2438 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2440 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2444 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2446 comparator = test->getOp();
\r
2447 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2453 // Parse increment
\r
2454 if(index && comparator != EOpNull && node->getExpression())
\r
2456 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2457 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2459 if(binaryTerminal)
\r
2461 TOperator op = binaryTerminal->getOp();
\r
2462 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2466 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2468 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2472 case EOpAddAssign: increment = value; break;
\r
2473 case EOpSubAssign: increment = -value; break;
\r
2474 default: UNIMPLEMENTED();
\r
2479 else if(unaryTerminal)
\r
2481 TOperator op = unaryTerminal->getOp();
\r
2485 case EOpPostIncrement: increment = 1; break;
\r
2486 case EOpPostDecrement: increment = -1; break;
\r
2487 case EOpPreIncrement: increment = 1; break;
\r
2488 case EOpPreDecrement: increment = -1; break;
\r
2489 default: UNIMPLEMENTED();
\r
2494 if(index && comparator != EOpNull && increment != 0)
\r
2496 if(comparator == EOpLessThanEqual)
\r
2498 comparator = EOpLessThan;
\r
2502 if(comparator == EOpLessThan)
\r
2504 int iterations = (limit - initial) / increment;
\r
2506 if(iterations <= 0)
\r
2511 return iterations;
\r
2513 else UNIMPLEMENTED(); // Falls through
\r
2519 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2522 loopDiscontinuity = false;
\r
2524 node->traverse(this);
\r
2526 return loopDiscontinuity;
\r
2529 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2531 if(visit == PreVisit)
\r
2535 else if(visit == PostVisit)
\r
2543 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2545 if(loopDiscontinuity)
\r
2555 switch(node->getFlowOp())
\r
2562 loopDiscontinuity = true;
\r
2564 default: UNREACHABLE();
\r
2567 return !loopDiscontinuity;
\r
2570 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2572 return !loopDiscontinuity;
\r