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(TSymbolTableLevel::nextUniqueId(), "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, int offset, int blockId) :
\r
79 type(type), precision(precision), name(name), arraySize(arraySize), registerIndex(registerIndex), offset(offset), blockId(blockId)
\r
83 UniformBlock::UniformBlock(const std::string& name, const std::string& instanceName, unsigned int dataSize, unsigned int arraySize,
\r
84 TLayoutBlockStorage layout, bool isRowMajorLayout, int registerIndex, int blockId) :
\r
85 name(name), instanceName(instanceName), dataSize(dataSize), arraySize(arraySize), layout(layout),
\r
86 isRowMajorLayout(isRowMajorLayout), registerIndex(registerIndex), blockId(blockId)
\r
90 Attribute::Attribute()
\r
97 Attribute::Attribute(GLenum type, const std::string &name, int arraySize, int location, int registerIndex)
\r
101 this->arraySize = arraySize;
\r
102 this->location = location;
\r
103 this->registerIndex = registerIndex;
\r
106 sw::PixelShader *Shader::getPixelShader() const
\r
111 sw::VertexShader *Shader::getVertexShader() const
\r
116 OutputASM::OutputASM(TParseContext &context, Shader *shaderObject) : TIntermTraverser(true, true, true), mContext(context), shaderObject(shaderObject)
\r
124 shader = shaderObject->getShader();
\r
125 pixelShader = shaderObject->getPixelShader();
\r
126 vertexShader = shaderObject->getVertexShader();
\r
129 functionArray.push_back(Function(0, "main(", 0, 0));
\r
130 currentFunction = 0;
\r
131 outputQualifier = EvqOutput; // Set outputQualifier to any value other than EvqFragColor or EvqFragData
\r
134 OutputASM::~OutputASM()
\r
138 void OutputASM::output()
\r
142 emitShader(GLOBAL);
\r
144 if(functionArray.size() > 1) // Only call main() when there are other functions
\r
146 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
\r
147 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
\r
148 callMain->dst.index = 0; // main()
\r
150 emit(sw::Shader::OPCODE_RET);
\r
153 emitShader(FUNCTION);
\r
157 void OutputASM::emitShader(Scope scope)
\r
160 currentScope = GLOBAL;
\r
161 mContext.getTreeRoot()->traverse(this);
\r
164 void OutputASM::freeTemporary(Temporary *temporary)
\r
166 free(temporaries, temporary);
\r
169 sw::Shader::Opcode OutputASM::getOpcode(sw::Shader::Opcode op, TIntermTyped *in) const
\r
171 TBasicType baseType = in->getType().getBasicType();
\r
175 case sw::Shader::OPCODE_NEG:
\r
180 return sw::Shader::OPCODE_INEG;
\r
185 case sw::Shader::OPCODE_ADD:
\r
190 return sw::Shader::OPCODE_IADD;
\r
195 case sw::Shader::OPCODE_SUB:
\r
200 return sw::Shader::OPCODE_ISUB;
\r
205 case sw::Shader::OPCODE_MUL:
\r
210 return sw::Shader::OPCODE_IMUL;
\r
215 case sw::Shader::OPCODE_DIV:
\r
219 return sw::Shader::OPCODE_IDIV;
\r
221 return sw::Shader::OPCODE_UDIV;
\r
226 case sw::Shader::OPCODE_IMOD:
\r
227 return baseType == EbtUInt ? sw::Shader::OPCODE_UMOD : op;
\r
228 case sw::Shader::OPCODE_ISHR:
\r
229 return baseType == EbtUInt ? sw::Shader::OPCODE_USHR : op;
\r
230 case sw::Shader::OPCODE_MIN:
\r
234 return sw::Shader::OPCODE_IMIN;
\r
236 return sw::Shader::OPCODE_UMIN;
\r
241 case sw::Shader::OPCODE_MAX:
\r
245 return sw::Shader::OPCODE_IMAX;
\r
247 return sw::Shader::OPCODE_UMAX;
\r
257 void OutputASM::visitSymbol(TIntermSymbol *symbol)
\r
259 // Vertex varyings don't have to be actively used to successfully link
\r
260 // against pixel shaders that use them. So make sure they're declared.
\r
261 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut || symbol->getQualifier() == EvqVertexOut)
\r
263 if(symbol->getBasicType() != EbtInvariant) // Typeless declarations are not new varyings
\r
265 declareVarying(symbol, -1);
\r
270 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
\r
272 if(currentScope != emitScope)
\r
277 TIntermTyped *result = node;
\r
278 TIntermTyped *left = node->getLeft();
\r
279 TIntermTyped *right = node->getRight();
\r
280 const TType &leftType = left->getType();
\r
281 const TType &rightType = right->getType();
\r
282 const TType &resultType = node->getType();
\r
284 switch(node->getOp())
\r
287 if(visit == PostVisit)
\r
289 assignLvalue(left, right);
\r
290 copy(result, right);
\r
293 case EOpInitialize:
\r
294 if(visit == PostVisit)
\r
299 case EOpMatrixTimesScalarAssign:
\r
300 if(visit == PostVisit)
\r
302 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
304 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
305 mul->dst.index += i;
\r
306 argument(mul->src[0], left, i);
\r
309 assignLvalue(left, result);
\r
312 case EOpVectorTimesMatrixAssign:
\r
313 if(visit == PostVisit)
\r
315 int size = leftType.getNominalSize();
\r
317 for(int i = 0; i < size; i++)
\r
319 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
320 dot->dst.mask = 1 << i;
\r
321 argument(dot->src[1], right, i);
\r
324 assignLvalue(left, result);
\r
327 case EOpMatrixTimesMatrixAssign:
\r
328 if(visit == PostVisit)
\r
330 int dim = leftType.getNominalSize();
\r
332 for(int i = 0; i < dim; i++)
\r
334 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
335 mul->dst.index += i;
\r
336 argument(mul->src[1], right, i);
\r
337 mul->src[1].swizzle = 0x00;
\r
339 for(int j = 1; j < dim; j++)
\r
341 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
342 mad->dst.index += i;
\r
343 argument(mad->src[0], left, j);
\r
344 argument(mad->src[1], right, i);
\r
345 mad->src[1].swizzle = j * 0x55;
\r
346 argument(mad->src[2], result, i);
\r
350 assignLvalue(left, result);
\r
353 case EOpIndexDirect:
\r
354 if(visit == PostVisit)
\r
356 int index = right->getAsConstantUnion()->getIConst(0);
\r
358 if(result->isMatrix() || result->isStruct())
\r
360 ASSERT(left->isArray());
\r
361 copy(result, left, index * left->elementRegisterCount());
\r
363 else if(result->isRegister())
\r
365 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
367 if(left->isRegister())
\r
369 mov->src[0].swizzle = index;
\r
371 else if(left->isArray())
\r
373 argument(mov->src[0], left, index * left->elementRegisterCount());
\r
375 else if(left->isMatrix())
\r
377 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
\r
378 argument(mov->src[0], left, index);
\r
380 else UNREACHABLE(0);
\r
382 else UNREACHABLE(0);
\r
385 case EOpIndexIndirect:
\r
386 if(visit == PostVisit)
\r
388 if(left->isArray() || left->isMatrix())
\r
390 for(int index = 0; index < result->totalRegisterCount(); index++)
\r
392 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
393 mov->dst.index += index;
\r
394 mov->dst.mask = writeMask(result, index);
\r
395 argument(mov->src[0], left, index);
\r
397 if(left->totalRegisterCount() > 1)
\r
399 sw::Shader::SourceParameter relativeRegister;
\r
400 argument(relativeRegister, right);
\r
402 mov->src[0].rel.type = relativeRegister.type;
\r
403 mov->src[0].rel.index = relativeRegister.index;
\r
404 mov->src[0].rel.scale = result->totalRegisterCount();
\r
405 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
409 else if(left->isRegister())
\r
411 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
\r
413 else UNREACHABLE(0);
\r
416 case EOpIndexDirectStruct:
\r
417 case EOpIndexDirectInterfaceBlock:
\r
418 if(visit == PostVisit)
\r
420 ASSERT(leftType.isStruct() || (leftType.isInterfaceBlock()));
\r
422 const TFieldList& fields = (node->getOp() == EOpIndexDirectStruct) ?
\r
423 leftType.getStruct()->fields() :
\r
424 leftType.getInterfaceBlock()->fields();
\r
425 int index = right->getAsConstantUnion()->getIConst(0);
\r
426 int fieldOffset = 0;
\r
428 for(int i = 0; i < index; i++)
\r
430 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
433 copy(result, left, fieldOffset);
\r
436 case EOpVectorSwizzle:
\r
437 if(visit == PostVisit)
\r
440 TIntermAggregate *components = right->getAsAggregate();
\r
444 TIntermSequence &sequence = components->getSequence();
\r
447 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
449 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
453 int i = element->getUnionArrayPointer()[0].getIConst();
\r
454 swizzle |= i << (component * 2);
\r
457 else UNREACHABLE(0);
\r
460 else UNREACHABLE(0);
\r
462 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
463 mov->src[0].swizzle = swizzle;
\r
466 case EOpAddAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_ADD, result), result, left, left, right); break;
\r
467 case EOpAdd: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_ADD, result), result, left, right); break;
\r
468 case EOpSubAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_SUB, result), result, left, left, right); break;
\r
469 case EOpSub: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_SUB, result), result, left, right); break;
\r
470 case EOpMulAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_MUL, result), result, left, left, right); break;
\r
471 case EOpMul: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_MUL, result), result, left, right); break;
\r
472 case EOpDivAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_DIV, result), result, left, left, right); break;
\r
473 case EOpDiv: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_DIV, result), result, left, right); break;
\r
474 case EOpIModAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_IMOD, result), result, left, left, right); break;
\r
475 case EOpIMod: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_IMOD, result), result, left, right); break;
\r
476 case EOpBitShiftLeftAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SHL, result, left, left, right); break;
\r
477 case EOpBitShiftLeft: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SHL, result, left, right); break;
\r
478 case EOpBitShiftRightAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_ISHR, result), result, left, left, right); break;
\r
479 case EOpBitShiftRight: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_ISHR, result), result, left, right); break;
\r
480 case EOpBitwiseAndAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_AND, result, left, left, right); break;
\r
481 case EOpBitwiseAnd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_AND, result, left, right); break;
\r
482 case EOpBitwiseXorAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_XOR, result, left, left, right); break;
\r
483 case EOpBitwiseXor: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
484 case EOpBitwiseOrAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_OR, result, left, left, right); break;
\r
485 case EOpBitwiseOr: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_OR, result, left, right); break;
\r
487 if(visit == PostVisit)
\r
489 emitBinary(sw::Shader::OPCODE_EQ, result, left, right);
\r
491 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
493 Temporary equal(this);
\r
494 Instruction *eq = emit(sw::Shader::OPCODE_EQ, &equal, left, right);
\r
495 argument(eq->src[0], left, index);
\r
496 argument(eq->src[1], right, index);
\r
497 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
502 if(visit == PostVisit)
\r
504 emitBinary(sw::Shader::OPCODE_NE, result, left, right);
\r
506 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
508 Temporary notEqual(this);
\r
509 Instruction *eq = emit(sw::Shader::OPCODE_NE, ¬Equal, left, right);
\r
510 argument(eq->src[0], left, index);
\r
511 argument(eq->src[1], right, index);
\r
512 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
516 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
517 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
518 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
519 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
520 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_MUL, left), result, left, left, right); break;
\r
521 case EOpVectorTimesScalar: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MUL, left), result, left, right); break;
\r
522 case EOpMatrixTimesScalar:
\r
523 if(visit == PostVisit)
\r
525 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
527 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
528 mul->dst.index += i;
\r
529 argument(mul->src[0], left, i);
\r
533 case EOpVectorTimesMatrix:
\r
534 if(visit == PostVisit)
\r
536 sw::Shader::Opcode dpOpcode = sw::Shader::OPCODE_DP(leftType.getNominalSize());
\r
538 int size = rightType.getNominalSize();
\r
539 for(int i = 0; i < size; i++)
\r
541 Instruction *dot = emit(dpOpcode, result, left, right);
\r
542 dot->dst.mask = 1 << i;
\r
543 argument(dot->src[1], right, i);
\r
547 case EOpMatrixTimesVector:
\r
548 if(visit == PostVisit)
\r
550 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
551 mul->src[1].swizzle = 0x00;
\r
553 int size = rightType.getNominalSize();
\r
554 for(int i = 1; i < size; i++)
\r
556 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
557 argument(mad->src[0], left, i);
\r
558 mad->src[1].swizzle = i * 0x55;
\r
562 case EOpMatrixTimesMatrix:
\r
563 if(visit == PostVisit)
\r
565 int dim = leftType.getNominalSize();
\r
567 int size = rightType.getNominalSize();
\r
568 for(int i = 0; i < size; i++)
\r
570 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
571 mul->dst.index += i;
\r
572 argument(mul->src[1], right, i);
\r
573 mul->src[1].swizzle = 0x00;
\r
575 for(int j = 1; j < dim; j++)
\r
577 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
578 mad->dst.index += i;
\r
579 argument(mad->src[0], left, j);
\r
580 argument(mad->src[1], right, i);
\r
581 mad->src[1].swizzle = j * 0x55;
\r
582 argument(mad->src[2], result, i);
\r
588 if(trivial(right, 6))
\r
590 if(visit == PostVisit)
\r
592 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
595 else // Short-circuit evaluation
\r
597 if(visit == InVisit)
\r
599 emit(sw::Shader::OPCODE_MOV, result, left);
\r
600 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
601 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
603 else if(visit == PostVisit)
\r
605 emit(sw::Shader::OPCODE_MOV, result, right);
\r
606 emit(sw::Shader::OPCODE_ENDIF);
\r
610 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
611 case EOpLogicalAnd:
\r
612 if(trivial(right, 6))
\r
614 if(visit == PostVisit)
\r
616 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
619 else // Short-circuit evaluation
\r
621 if(visit == InVisit)
\r
623 emit(sw::Shader::OPCODE_MOV, result, left);
\r
624 emit(sw::Shader::OPCODE_IF, 0, result);
\r
626 else if(visit == PostVisit)
\r
628 emit(sw::Shader::OPCODE_MOV, result, right);
\r
629 emit(sw::Shader::OPCODE_ENDIF);
\r
633 default: UNREACHABLE(node->getOp());
\r
639 void OutputASM::emitDeterminant(TIntermTyped *result, TIntermTyped *arg, int size, int col, int row, int outCol, int outRow)
\r
643 case 1: // Used for cofactor computation only
\r
645 // For a 2x2 matrix, the cofactor is simply a transposed move or negate
\r
646 bool isMov = (row == col);
\r
647 sw::Shader::Opcode op = isMov ? sw::Shader::OPCODE_MOV : sw::Shader::OPCODE_NEG;
\r
648 Instruction *mov = emit(op, result, arg);
\r
649 mov->src[0].index += isMov ? 1 - row : row;
\r
650 mov->src[0].swizzle = 0x55 * (isMov ? 1 - col : col);
\r
651 mov->dst.index += outCol;
\r
652 mov->dst.mask = 1 << outRow;
\r
657 static const unsigned int swizzle[3] = { 0x99, 0x88, 0x44 }; // xy?? : yzyz, xzxz, xyxy
\r
659 bool isCofactor = (col >= 0) && (row >= 0);
\r
660 int col0 = (isCofactor && (col <= 0)) ? 1 : 0;
\r
661 int col1 = (isCofactor && (col <= 1)) ? 2 : 1;
\r
662 bool negate = isCofactor && ((col & 0x01) ^ (row & 0x01));
\r
664 Instruction *det = emit(sw::Shader::OPCODE_DET2, result, arg, arg);
\r
665 det->src[0].index += negate ? col1 : col0;
\r
666 det->src[1].index += negate ? col0 : col1;
\r
667 det->src[0].swizzle = det->src[1].swizzle = swizzle[isCofactor ? row : 2];
\r
668 det->dst.index += outCol;
\r
669 det->dst.mask = 1 << outRow;
\r
674 static const unsigned int swizzle[4] = { 0xF9, 0xF8, 0xF4, 0xE4 }; // xyz? : yzww, xzww, xyww, xyzw
\r
676 bool isCofactor = (col >= 0) && (row >= 0);
\r
677 int col0 = (isCofactor && (col <= 0)) ? 1 : 0;
\r
678 int col1 = (isCofactor && (col <= 1)) ? 2 : 1;
\r
679 int col2 = (isCofactor && (col <= 2)) ? 3 : 2;
\r
680 bool negate = isCofactor && ((col & 0x01) ^ (row & 0x01));
\r
682 Instruction *det = emit(sw::Shader::OPCODE_DET3, result, arg, arg, arg);
\r
683 det->src[0].index += col0;
\r
684 det->src[1].index += negate ? col2 : col1;
\r
685 det->src[2].index += negate ? col1 : col2;
\r
686 det->src[0].swizzle = det->src[1].swizzle = det->src[2].swizzle = swizzle[isCofactor ? row : 3];
\r
687 det->dst.index += outCol;
\r
688 det->dst.mask = 1 << outRow;
\r
693 Instruction *det = emit(sw::Shader::OPCODE_DET4, result, arg, arg, arg, arg);
\r
694 det->src[1].index += 1;
\r
695 det->src[2].index += 2;
\r
696 det->src[3].index += 3;
\r
697 det->dst.index += outCol;
\r
698 det->dst.mask = 1 << outRow;
\r
707 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
709 if(currentScope != emitScope)
\r
714 TIntermTyped *result = node;
\r
715 TIntermTyped *arg = node->getOperand();
\r
716 TBasicType basicType = arg->getType().getBasicType();
\r
724 if(basicType == EbtInt || basicType == EbtUInt)
\r
730 one_value.f = 1.0f;
\r
733 Constant one(one_value.f, one_value.f, one_value.f, one_value.f);
\r
734 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
735 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
737 switch(node->getOp())
\r
740 if(visit == PostVisit)
\r
742 sw::Shader::Opcode negOpcode = getOpcode(sw::Shader::OPCODE_NEG, arg);
\r
743 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
745 Instruction *neg = emit(negOpcode, result, arg);
\r
746 neg->dst.index += index;
\r
747 argument(neg->src[0], arg, index);
\r
751 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
752 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
753 case EOpPostIncrement:
\r
754 if(visit == PostVisit)
\r
758 sw::Shader::Opcode addOpcode = getOpcode(sw::Shader::OPCODE_ADD, arg);
\r
759 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
761 Instruction *add = emit(addOpcode, arg, arg, &one);
\r
762 add->dst.index += index;
\r
763 argument(add->src[0], arg, index);
\r
766 assignLvalue(arg, arg);
\r
769 case EOpPostDecrement:
\r
770 if(visit == PostVisit)
\r
774 sw::Shader::Opcode subOpcode = getOpcode(sw::Shader::OPCODE_SUB, arg);
\r
775 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
777 Instruction *sub = emit(subOpcode, arg, arg, &one);
\r
778 sub->dst.index += index;
\r
779 argument(sub->src[0], arg, index);
\r
782 assignLvalue(arg, arg);
\r
785 case EOpPreIncrement:
\r
786 if(visit == PostVisit)
\r
788 sw::Shader::Opcode addOpcode = getOpcode(sw::Shader::OPCODE_ADD, arg);
\r
789 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
791 Instruction *add = emit(addOpcode, result, arg, &one);
\r
792 add->dst.index += index;
\r
793 argument(add->src[0], arg, index);
\r
796 assignLvalue(arg, result);
\r
799 case EOpPreDecrement:
\r
800 if(visit == PostVisit)
\r
802 sw::Shader::Opcode subOpcode = getOpcode(sw::Shader::OPCODE_SUB, arg);
\r
803 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
805 Instruction *sub = emit(subOpcode, result, arg, &one);
\r
806 sub->dst.index += index;
\r
807 argument(sub->src[0], arg, index);
\r
810 assignLvalue(arg, result);
\r
813 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
814 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
815 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
816 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
817 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
818 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
819 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
820 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
821 case EOpSinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_SINH, result, arg); break;
\r
822 case EOpCosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_COSH, result, arg); break;
\r
823 case EOpTanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_TANH, result, arg); break;
\r
824 case EOpAsinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASINH, result, arg); break;
\r
825 case EOpAcosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOSH, result, arg); break;
\r
826 case EOpAtanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATANH, result, arg); break;
\r
827 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
828 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
829 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
830 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
831 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
832 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
833 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
834 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
835 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
836 case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
\r
837 case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
\r
838 case EOpRoundEven: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUNDEVEN, result, arg); break;
\r
839 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
840 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
841 case EOpIsNan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISNAN, result, arg); break;
\r
842 case EOpIsInf: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISINF, result, arg); break;
\r
843 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
844 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
845 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
846 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
847 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
848 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
849 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
850 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg); break;
\r
851 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg); break;
\r
852 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg); break;
\r
853 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg); break;
\r
854 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg); break;
\r
855 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg); break;
\r
856 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg); break;
\r
857 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg); break;
\r
858 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg); break;
\r
859 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg); break;
\r
861 if(visit == PostVisit)
\r
863 int numCols = arg->getNominalSize();
\r
864 int numRows = arg->getSecondarySize();
\r
865 for(int i = 0; i < numCols; ++i)
\r
867 for(int j = 0; j < numRows; ++j)
\r
869 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
870 mov->src[0].index += i;
\r
871 mov->src[0].swizzle = 0x55 * j;
\r
872 mov->dst.index += j;
\r
873 mov->dst.mask = 1 << i;
\r
878 case EOpDeterminant:
\r
879 if(visit == PostVisit)
\r
881 int size = arg->getNominalSize();
\r
882 ASSERT(size == arg->getSecondarySize());
\r
884 emitDeterminant(result, arg, size);
\r
888 if(visit == PostVisit)
\r
890 int size = arg->getNominalSize();
\r
891 ASSERT(size == arg->getSecondarySize());
\r
893 // Compute transposed matrix of cofactors
\r
894 for(int i = 0; i < size; ++i)
\r
896 for(int j = 0; j < size; ++j)
\r
898 // For a 2x2 matrix, the cofactor is simply a transposed move or negate
\r
899 // For a 3x3 or 4x4 matrix, the cofactor is a transposed determinant
\r
900 emitDeterminant(result, arg, size - 1, j, i, i, j);
\r
904 // Compute 1 / determinant
\r
905 Temporary invDet(this);
\r
906 emitDeterminant(&invDet, arg, size);
\r
907 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
908 Instruction *div = emit(sw::Shader::OPCODE_DIV, &invDet, &one, &invDet);
\r
909 div->src[1].swizzle = 0x00; // xxxx
\r
911 // Divide transposed matrix of cofactors by determinant
\r
912 for(int i = 0; i < size; ++i)
\r
914 Instruction *div = emit(sw::Shader::OPCODE_MUL, result, result, &invDet);
\r
915 div->src[0].index += i;
\r
916 div->dst.index += i;
\r
920 default: UNREACHABLE(node->getOp());
\r
926 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
928 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
933 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
935 TIntermTyped *result = node;
\r
936 const TType &resultType = node->getType();
\r
937 TIntermSequence &arg = node->getSequence();
\r
938 int argumentCount = arg.size();
\r
940 switch(node->getOp())
\r
942 case EOpSequence: break;
\r
943 case EOpDeclaration: break;
\r
944 case EOpPrototype: break;
\r
946 if(visit == PostVisit)
\r
948 copy(result, arg[1]);
\r
952 if(visit == PreVisit)
\r
954 const TString &name = node->getName();
\r
956 if(emitScope == FUNCTION)
\r
958 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
960 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
961 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
963 const Function *function = findFunction(name);
\r
964 ASSERT(function); // Should have been added during global pass
\r
965 label->dst.index = function->label;
\r
966 currentFunction = function->label;
\r
969 else if(emitScope == GLOBAL)
\r
971 if(name != "main(")
\r
973 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
974 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
977 else UNREACHABLE(emitScope);
\r
979 currentScope = FUNCTION;
\r
981 else if(visit == PostVisit)
\r
983 if(emitScope == FUNCTION)
\r
985 if(functionArray.size() > 1) // No need to return when there's only main()
\r
987 emit(sw::Shader::OPCODE_RET);
\r
991 currentScope = GLOBAL;
\r
994 case EOpFunctionCall:
\r
995 if(visit == PostVisit)
\r
997 if(node->isUserDefined())
\r
999 const TString &name = node->getName();
\r
1000 const Function *function = findFunction(name);
\r
1004 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
1008 TIntermSequence &arguments = *function->arg;
\r
1010 for(int i = 0; i < argumentCount; i++)
\r
1012 TIntermTyped *in = arguments[i]->getAsTyped();
\r
1014 if(in->getQualifier() == EvqIn ||
\r
1015 in->getQualifier() == EvqInOut ||
\r
1016 in->getQualifier() == EvqConstReadOnly)
\r
1022 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
1023 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
1024 call->dst.index = function->label;
\r
1026 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
1028 copy(result, function->ret);
\r
1031 for(int i = 0; i < argumentCount; i++)
\r
1033 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
1034 TIntermTyped *out = arg[i]->getAsTyped();
\r
1036 if(argument->getQualifier() == EvqOut ||
\r
1037 argument->getQualifier() == EvqInOut)
\r
1039 copy(out, argument);
\r
1045 TString name = TFunction::unmangleName(node->getName());
\r
1047 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
1049 if(argumentCount == 2)
\r
1051 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
1053 else if(argumentCount == 3) // bias
\r
1055 Temporary uvwb(this);
\r
1056 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
1057 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
1058 bias->dst.mask = 0x8;
\r
1060 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
1063 else UNREACHABLE(argumentCount);
\r
1065 else if(name == "texture2DProj" || name == "textureProj")
\r
1067 TIntermTyped *t = arg[1]->getAsTyped();
\r
1069 if(argumentCount == 2)
\r
1071 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
1072 tex->project = true;
\r
1074 if(t->getNominalSize() == 3)
\r
1076 tex->src[0].swizzle = 0xA4;
\r
1078 else ASSERT(t->getNominalSize() == 4);
\r
1080 else if(argumentCount == 3) // bias
\r
1082 Temporary proj(this);
\r
1084 if(t->getNominalSize() == 3)
\r
1086 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
1087 div->src[1].swizzle = 0xAA;
\r
1088 div->dst.mask = 0x3;
\r
1090 else if(t->getNominalSize() == 4)
\r
1092 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
1093 div->src[1].swizzle = 0xFF;
\r
1094 div->dst.mask = 0x3;
\r
1096 else UNREACHABLE(t->getNominalSize());
\r
1098 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
1099 bias->dst.mask = 0x8;
\r
1101 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
1104 else UNREACHABLE(argumentCount);
\r
1106 else if(name == "texture2DLod" || name == "textureCubeLod" || name == "textureLod")
\r
1108 Temporary uvwb(this);
\r
1109 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
1110 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
1111 lod->dst.mask = 0x8;
\r
1113 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
1115 else if(name == "texture2DProjLod" || name == "textureProjLod")
\r
1117 TIntermTyped *t = arg[1]->getAsTyped();
\r
1118 Temporary proj(this);
\r
1120 if(t->getNominalSize() == 3)
\r
1122 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
1123 div->src[1].swizzle = 0xAA;
\r
1124 div->dst.mask = 0x3;
\r
1126 else if(t->getNominalSize() == 4)
\r
1128 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
1129 div->src[1].swizzle = 0xFF;
\r
1130 div->dst.mask = 0x3;
\r
1132 else UNREACHABLE(t->getNominalSize());
\r
1134 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
1135 lod->dst.mask = 0x8;
\r
1137 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
1139 else UNREACHABLE(0);
\r
1143 case EOpParameters:
\r
1145 case EOpConstructFloat:
\r
1146 case EOpConstructVec2:
\r
1147 case EOpConstructVec3:
\r
1148 case EOpConstructVec4:
\r
1149 case EOpConstructBool:
\r
1150 case EOpConstructBVec2:
\r
1151 case EOpConstructBVec3:
\r
1152 case EOpConstructBVec4:
\r
1153 case EOpConstructInt:
\r
1154 case EOpConstructIVec2:
\r
1155 case EOpConstructIVec3:
\r
1156 case EOpConstructIVec4:
\r
1157 case EOpConstructUInt:
\r
1158 case EOpConstructUVec2:
\r
1159 case EOpConstructUVec3:
\r
1160 case EOpConstructUVec4:
\r
1161 if(visit == PostVisit)
\r
1163 int component = 0;
\r
1165 for(int i = 0; i < argumentCount; i++)
\r
1167 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1168 int size = argi->getNominalSize();
\r
1170 if(!argi->isMatrix())
\r
1172 Instruction *mov = emitCast(result, argi);
\r
1173 mov->dst.mask = (0xF << component) & 0xF;
\r
1174 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
1176 component += size;
\r
1182 while(component < resultType.getNominalSize())
\r
1184 Instruction *mov = emitCast(result, argi);
\r
1185 mov->dst.mask = (0xF << component) & 0xF;
\r
1186 mov->src[0].index += column;
\r
1187 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
1190 component += size;
\r
1196 case EOpConstructMat2:
\r
1197 case EOpConstructMat2x3:
\r
1198 case EOpConstructMat2x4:
\r
1199 case EOpConstructMat3x2:
\r
1200 case EOpConstructMat3:
\r
1201 case EOpConstructMat3x4:
\r
1202 case EOpConstructMat4x2:
\r
1203 case EOpConstructMat4x3:
\r
1204 case EOpConstructMat4:
\r
1205 if(visit == PostVisit)
\r
1207 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
1208 const int outCols = result->getNominalSize();
\r
1209 const int outRows = result->getSecondarySize();
\r
1211 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
1213 for(int i = 0; i < outCols; i++)
\r
1215 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
1216 init->dst.index += i;
\r
1217 Instruction *mov = emitCast(result, arg0);
\r
1218 mov->dst.index += i;
\r
1219 mov->dst.mask = 1 << i;
\r
1220 ASSERT(mov->src[0].swizzle == 0x00);
\r
1223 else if(arg0->isMatrix())
\r
1225 const int inCols = arg0->getNominalSize();
\r
1226 const int inRows = arg0->getSecondarySize();
\r
1228 for(int i = 0; i < outCols; i++)
\r
1230 if(i >= inCols || outRows > inRows)
\r
1232 // Initialize to identity matrix
\r
1233 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
1234 Instruction *mov = emitCast(result, &col);
\r
1235 mov->dst.index += i;
\r
1240 Instruction *mov = emitCast(result, arg0);
\r
1241 mov->dst.index += i;
\r
1242 mov->dst.mask = 0xF >> (4 - inRows);
\r
1243 argument(mov->src[0], arg0, i);
\r
1252 for(int i = 0; i < argumentCount; i++)
\r
1254 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1255 int size = argi->getNominalSize();
\r
1258 while(element < size)
\r
1260 Instruction *mov = emitCast(result, argi);
\r
1261 mov->dst.index += column;
\r
1262 mov->dst.mask = (0xF << row) & 0xF;
\r
1263 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
1265 int end = row + size - element;
\r
1266 column = end >= outRows ? column + 1 : column;
\r
1267 element = element + outRows - row;
\r
1268 row = end >= outRows ? 0 : end;
\r
1274 case EOpConstructStruct:
\r
1275 if(visit == PostVisit)
\r
1278 for(int i = 0; i < argumentCount; i++)
\r
1280 TIntermTyped *argi = arg[i]->getAsTyped();
\r
1281 int size = argi->totalRegisterCount();
\r
1283 for(int index = 0; index < size; index++)
\r
1285 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
1286 mov->dst.index += index + offset;
\r
1287 mov->dst.mask = writeMask(result, offset + index);
\r
1288 argument(mov->src[0], argi, index);
\r
1295 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1296 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1297 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1298 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1299 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1300 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1301 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1302 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1303 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1304 case EOpMin: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MIN, result), result, arg[0], arg[1]); break;
\r
1305 case EOpMax: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MAX, result), result, arg[0], arg[1]); break;
\r
1307 if(visit == PostVisit)
\r
1309 emit(getOpcode(sw::Shader::OPCODE_MAX, result), result, arg[0], arg[1]);
\r
1310 emit(getOpcode(sw::Shader::OPCODE_MIN, result), result, result, arg[2]);
\r
1313 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1314 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1315 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1316 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1317 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1318 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1319 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1320 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1321 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1323 if(visit == PostVisit)
\r
1325 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
1326 TIntermTyped *arg1 = arg[1]->getAsTyped();
\r
1327 ASSERT((arg0->getNominalSize() == arg1->getNominalSize()) && (arg0->getSecondarySize() == arg1->getSecondarySize()));
\r
1329 int size = arg0->getNominalSize();
\r
1330 for(int i = 0; i < size; i++)
\r
1332 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1333 mul->dst.index += i;
\r
1334 argument(mul->src[0], arg[0], i);
\r
1335 argument(mul->src[1], arg[1], i);
\r
1339 case EOpOuterProduct:
\r
1340 if(visit == PostVisit)
\r
1342 for(int i = 0; i < dim(arg[1]); i++)
\r
1344 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1345 mul->dst.index += i;
\r
1346 mul->src[1].swizzle = 0x55 * i;
\r
1350 default: UNREACHABLE(node->getOp());
\r
1356 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1358 if(currentScope != emitScope)
\r
1363 TIntermTyped *condition = node->getCondition();
\r
1364 TIntermNode *trueBlock = node->getTrueBlock();
\r
1365 TIntermNode *falseBlock = node->getFalseBlock();
\r
1366 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1368 condition->traverse(this);
\r
1370 if(node->usesTernaryOperator())
\r
1372 if(constantCondition)
\r
1374 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1378 trueBlock->traverse(this);
\r
1379 copy(node, trueBlock);
\r
1383 falseBlock->traverse(this);
\r
1384 copy(node, falseBlock);
\r
1387 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1389 trueBlock->traverse(this);
\r
1390 falseBlock->traverse(this);
\r
1391 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1395 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1399 trueBlock->traverse(this);
\r
1400 copy(node, trueBlock);
\r
1405 emit(sw::Shader::OPCODE_ELSE);
\r
1406 falseBlock->traverse(this);
\r
1407 copy(node, falseBlock);
\r
1410 emit(sw::Shader::OPCODE_ENDIF);
\r
1413 else // if/else statement
\r
1415 if(constantCondition)
\r
1417 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1423 trueBlock->traverse(this);
\r
1430 falseBlock->traverse(this);
\r
1436 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1440 trueBlock->traverse(this);
\r
1445 emit(sw::Shader::OPCODE_ELSE);
\r
1446 falseBlock->traverse(this);
\r
1449 emit(sw::Shader::OPCODE_ENDIF);
\r
1456 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1458 if(currentScope != emitScope)
\r
1463 unsigned int iterations = loopCount(node);
\r
1465 if(iterations == 0)
\r
1470 bool unroll = (iterations <= 4);
\r
1474 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1475 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1478 TIntermNode *init = node->getInit();
\r
1479 TIntermTyped *condition = node->getCondition();
\r
1480 TIntermTyped *expression = node->getExpression();
\r
1481 TIntermNode *body = node->getBody();
\r
1483 if(node->getType() == ELoopDoWhile)
\r
1485 Temporary iterate(this);
\r
1486 Constant True(true);
\r
1487 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1489 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1493 body->traverse(this);
\r
1496 emit(sw::Shader::OPCODE_TEST);
\r
1498 condition->traverse(this);
\r
1499 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1501 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1507 init->traverse(this);
\r
1512 for(unsigned int i = 0; i < iterations; i++)
\r
1514 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1518 body->traverse(this);
\r
1523 expression->traverse(this);
\r
1531 condition->traverse(this);
\r
1534 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1538 body->traverse(this);
\r
1541 emit(sw::Shader::OPCODE_TEST);
\r
1545 expression->traverse(this);
\r
1550 condition->traverse(this);
\r
1553 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1560 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1562 if(currentScope != emitScope)
\r
1567 switch(node->getFlowOp())
\r
1569 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1570 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1571 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1573 if(visit == PostVisit)
\r
1575 TIntermTyped *value = node->getExpression();
\r
1579 copy(functionArray[currentFunction].ret, value);
\r
1582 emit(sw::Shader::OPCODE_LEAVE);
\r
1585 default: UNREACHABLE(node->getFlowOp());
\r
1591 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1593 return operand && isSamplerRegister(operand->getType());
\r
1596 bool OutputASM::isSamplerRegister(const TType &type)
\r
1598 // A sampler register's qualifiers can be:
\r
1599 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1600 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1601 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1602 // the sampler has been copied and is no longer a sampler register.
\r
1603 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1606 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, TIntermNode *src3, int index)
\r
1608 if(isSamplerRegister(dst))
\r
1610 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1613 Instruction *instruction = new Instruction(op);
\r
1617 instruction->dst.type = registerType(dst);
\r
1618 instruction->dst.index = registerIndex(dst) + index;
\r
1619 instruction->dst.mask = writeMask(dst);
\r
1620 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1623 argument(instruction->src[0], src0, index);
\r
1624 argument(instruction->src[1], src1, index);
\r
1625 argument(instruction->src[2], src2, index);
\r
1626 argument(instruction->src[3], src3, index);
\r
1628 shader->append(instruction);
\r
1630 return instruction;
\r
1633 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1635 switch(src->getBasicType())
\r
1638 switch(dst->getBasicType())
\r
1640 case EbtInt: return emit(sw::Shader::OPCODE_B2I, dst, src);
\r
1641 case EbtUInt: return emit(sw::Shader::OPCODE_B2U, dst, src);
\r
1642 case EbtFloat: return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1647 switch(dst->getBasicType())
\r
1649 case EbtBool: return emit(sw::Shader::OPCODE_I2B, dst, src);
\r
1650 case EbtFloat: return emit(sw::Shader::OPCODE_I2F, dst, src);
\r
1655 switch(dst->getBasicType())
\r
1657 case EbtBool: return emit(sw::Shader::OPCODE_U2B, dst, src);
\r
1658 case EbtFloat: return emit(sw::Shader::OPCODE_U2F, dst, src);
\r
1663 switch(dst->getBasicType())
\r
1665 case EbtBool: return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1666 case EbtInt: return emit(sw::Shader::OPCODE_F2I, dst, src);
\r
1667 case EbtUInt: return emit(sw::Shader::OPCODE_F2U, dst, src);
\r
1675 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1678 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1680 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1682 emit(op, dst, src0, src1, src2, 0, index);
\r
1686 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1688 emitBinary(op, result, src0, src1);
\r
1689 assignLvalue(lhs, result);
\r
1692 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1694 sw::Shader::Opcode opcode;
\r
1695 switch(left->getAsTyped()->getBasicType())
\r
1699 opcode = sw::Shader::OPCODE_ICMP;
\r
1702 opcode = sw::Shader::OPCODE_UCMP;
\r
1705 opcode = sw::Shader::OPCODE_CMP;
\r
1709 Instruction *cmp = emit(opcode, dst, left, right);
\r
1710 cmp->control = cmpOp;
\r
1711 argument(cmp->src[0], left, index);
\r
1712 argument(cmp->src[1], right, index);
\r
1715 int componentCount(const TType &type, int registers)
\r
1717 if(registers == 0)
\r
1722 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1724 int index = registers / type.elementRegisterCount();
\r
1725 registers -= index * type.elementRegisterCount();
\r
1726 return index * type.getElementSize() + componentCount(type, registers);
\r
1729 if(type.isStruct() || type.isInterfaceBlock())
\r
1731 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
\r
1734 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1736 const TType &fieldType = *((*field)->type());
\r
1738 if(fieldType.totalRegisterCount() <= registers)
\r
1740 registers -= fieldType.totalRegisterCount();
\r
1741 elements += fieldType.getObjectSize();
\r
1743 else // Register within this field
\r
1745 return elements + componentCount(fieldType, registers);
\r
1749 else if(type.isMatrix())
\r
1751 return registers * type.registerSize();
\r
1758 int registerSize(const TType &type, int registers)
\r
1760 if(registers == 0)
\r
1762 if(type.isStruct())
\r
1764 return registerSize(*((*(type.getStruct()->fields().begin()))->type()), 0);
\r
1767 return type.registerSize();
\r
1770 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1772 int index = registers / type.elementRegisterCount();
\r
1773 registers -= index * type.elementRegisterCount();
\r
1774 return registerSize(type, registers);
\r
1777 if(type.isStruct() || type.isInterfaceBlock())
\r
1779 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
\r
1782 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
\r
1784 const TType &fieldType = *((*field)->type());
\r
1786 if(fieldType.totalRegisterCount() <= registers)
\r
1788 registers -= fieldType.totalRegisterCount();
\r
1789 elements += fieldType.getObjectSize();
\r
1791 else // Register within this field
\r
1793 return registerSize(fieldType, registers);
\r
1797 else if(type.isMatrix())
\r
1799 return registerSize(type, 0);
\r
1806 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1810 TIntermTyped *arg = argument->getAsTyped();
\r
1811 const TType &type = arg->getType();
\r
1812 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1814 int size = registerSize(type, index);
\r
1816 parameter.type = registerType(arg);
\r
1818 if(arg->getQualifier() == EvqConstExpr)
\r
1820 int component = componentCount(type, index);
\r
1821 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1823 for(int i = 0; i < 4; i++)
\r
1825 if(size == 1) // Replicate
\r
1827 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1831 parameter.value[i] = constants[component + i].getAsFloat();
\r
1835 parameter.value[i] = 0.0f;
\r
1841 parameter.index = registerIndex(arg) + index;
\r
1843 if(isSamplerRegister(arg))
\r
1845 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1849 TIntermTyped *left = binary->getLeft();
\r
1850 TIntermTyped *right = binary->getRight();
\r
1852 switch(binary->getOp())
\r
1854 case EOpIndexDirect:
\r
1855 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1857 case EOpIndexIndirect:
\r
1858 if(left->getArraySize() > 1)
\r
1860 parameter.rel.type = registerType(binary->getRight());
\r
1861 parameter.rel.index = registerIndex(binary->getRight());
\r
1862 parameter.rel.scale = 1;
\r
1863 parameter.rel.deterministic = true;
\r
1866 case EOpIndexDirectStruct:
\r
1867 case EOpIndexDirectInterfaceBlock:
\r
1868 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1871 UNREACHABLE(binary->getOp());
\r
1877 if(!IsSampler(arg->getBasicType()))
\r
1879 parameter.swizzle = readSwizzle(arg, size);
\r
1884 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1886 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1888 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1889 mov->dst.index += index;
\r
1890 mov->dst.mask = writeMask(dst, index);
\r
1891 argument(mov->src[0], src, offset + index);
\r
1895 int swizzleElement(int swizzle, int index)
\r
1897 return (swizzle >> (index * 2)) & 0x03;
\r
1900 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1902 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1903 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1904 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1905 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1908 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1911 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1912 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
\r
1914 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1917 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1919 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1921 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1923 Temporary address(this);
\r
1924 lvalue(insert->dst, address, dst);
\r
1926 insert->src[0].type = insert->dst.type;
\r
1927 insert->src[0].index = insert->dst.index;
\r
1928 insert->src[0].rel = insert->dst.rel;
\r
1929 argument(insert->src[1], src);
\r
1930 argument(insert->src[2], binary->getRight());
\r
1932 shader->append(insert);
\r
1936 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1938 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1940 Temporary address(this);
\r
1941 int swizzle = lvalue(mov->dst, address, dst);
\r
1942 mov->dst.index += offset;
\r
1946 mov->dst.mask = writeMask(dst, offset);
\r
1949 argument(mov->src[0], src, offset);
\r
1950 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1952 shader->append(mov);
\r
1957 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1959 TIntermTyped *result = node;
\r
1960 TIntermBinary *binary = node->getAsBinaryNode();
\r
1961 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1965 TIntermTyped *left = binary->getLeft();
\r
1966 TIntermTyped *right = binary->getRight();
\r
1968 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1970 switch(binary->getOp())
\r
1972 case EOpIndexDirect:
\r
1974 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1976 if(left->isRegister())
\r
1978 int leftMask = dst.mask;
\r
1981 while((leftMask & dst.mask) == 0)
\r
1983 dst.mask = dst.mask << 1;
\r
1986 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1987 dst.mask = 1 << element;
\r
1991 else if(left->isArray() || left->isMatrix())
\r
1993 dst.index += rightIndex * result->totalRegisterCount();
\r
1996 else UNREACHABLE(0);
\r
1999 case EOpIndexIndirect:
\r
2001 if(left->isRegister())
\r
2003 // Requires INSERT instruction (handled by calling function)
\r
2005 else if(left->isArray() || left->isMatrix())
\r
2007 int scale = result->totalRegisterCount();
\r
2009 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
2011 if(left->totalRegisterCount() > 1)
\r
2013 sw::Shader::SourceParameter relativeRegister;
\r
2014 argument(relativeRegister, right);
\r
2016 dst.rel.index = relativeRegister.index;
\r
2017 dst.rel.type = relativeRegister.type;
\r
2018 dst.rel.scale = scale;
\r
2019 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
2022 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
2026 Constant oldScale((int)dst.rel.scale);
\r
2027 Instruction *mad = emit(sw::Shader::OPCODE_IMAD, &address, &address, &oldScale, right);
\r
2028 mad->src[0].index = dst.rel.index;
\r
2029 mad->src[0].type = dst.rel.type;
\r
2033 Constant oldScale((int)dst.rel.scale);
\r
2034 Instruction *mul = emit(sw::Shader::OPCODE_IMUL, &address, &address, &oldScale);
\r
2035 mul->src[0].index = dst.rel.index;
\r
2036 mul->src[0].type = dst.rel.type;
\r
2038 Constant newScale(scale);
\r
2039 emit(sw::Shader::OPCODE_IMAD, &address, right, &newScale, &address);
\r
2042 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
2043 dst.rel.index = registerIndex(&address);
\r
2044 dst.rel.scale = 1;
\r
2046 else // Just add the new index to the address register
\r
2050 emit(sw::Shader::OPCODE_IADD, &address, &address, right);
\r
2054 Constant newScale(scale);
\r
2055 emit(sw::Shader::OPCODE_IMAD, &address, right, &newScale, &address);
\r
2059 else UNREACHABLE(0);
\r
2062 case EOpIndexDirectStruct:
\r
2063 case EOpIndexDirectInterfaceBlock:
\r
2065 const TFieldList& fields = (binary->getOp() == EOpIndexDirectStruct) ?
\r
2066 left->getType().getStruct()->fields() :
\r
2067 left->getType().getInterfaceBlock()->fields();
\r
2068 int index = right->getAsConstantUnion()->getIConst(0);
\r
2069 int fieldOffset = 0;
\r
2071 for(int i = 0; i < index; i++)
\r
2073 fieldOffset += fields[i]->type()->totalRegisterCount();
\r
2076 dst.type = registerType(left);
\r
2077 dst.index += fieldOffset;
\r
2078 dst.mask = writeMask(right);
\r
2083 case EOpVectorSwizzle:
\r
2085 ASSERT(left->isRegister());
\r
2087 int leftMask = dst.mask;
\r
2090 int rightMask = 0;
\r
2092 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
2094 for(unsigned int i = 0; i < sequence.size(); i++)
\r
2096 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
2098 int element = swizzleElement(leftSwizzle, index);
\r
2099 rightMask = rightMask | (1 << element);
\r
2100 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
2103 dst.mask = leftMask & rightMask;
\r
2109 UNREACHABLE(binary->getOp()); // Not an l-value operator
\r
2115 dst.type = registerType(symbol);
\r
2116 dst.index = registerIndex(symbol);
\r
2117 dst.mask = writeMask(symbol);
\r
2124 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
2126 if(isSamplerRegister(operand))
\r
2128 return sw::Shader::PARAMETER_SAMPLER;
\r
2131 const TQualifier qualifier = operand->getQualifier();
\r
2132 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
2134 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
2135 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
2137 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
2139 outputQualifier = qualifier;
\r
2144 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
2145 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
2146 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
2147 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
2148 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
2149 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
2150 case EvqVertexIn: return sw::Shader::PARAMETER_INPUT;
\r
2151 case EvqFragmentOut: return sw::Shader::PARAMETER_COLOROUT;
\r
2152 case EvqVertexOut: return sw::Shader::PARAMETER_OUTPUT;
\r
2153 case EvqFragmentIn: return sw::Shader::PARAMETER_INPUT;
\r
2154 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
2155 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
2156 case EvqSmooth: return sw::Shader::PARAMETER_OUTPUT;
\r
2157 case EvqFlat: return sw::Shader::PARAMETER_OUTPUT;
\r
2158 case EvqCentroidOut: return sw::Shader::PARAMETER_OUTPUT;
\r
2159 case EvqSmoothIn: return sw::Shader::PARAMETER_INPUT;
\r
2160 case EvqFlatIn: return sw::Shader::PARAMETER_INPUT;
\r
2161 case EvqCentroidIn: return sw::Shader::PARAMETER_INPUT;
\r
2162 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
2163 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
2164 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
2165 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
2166 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
2167 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
2168 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
2169 case EvqInstanceID: return sw::Shader::PARAMETER_MISCTYPE;
\r
2170 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
2171 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
2172 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
2173 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
2174 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
2175 case EvqFragDepth: return sw::Shader::PARAMETER_DEPTHOUT;
\r
2176 default: UNREACHABLE(qualifier);
\r
2179 return sw::Shader::PARAMETER_VOID;
\r
2182 int OutputASM::registerIndex(TIntermTyped *operand)
\r
2184 if(isSamplerRegister(operand))
\r
2186 return samplerRegister(operand);
\r
2189 switch(operand->getQualifier())
\r
2191 case EvqTemporary: return temporaryRegister(operand);
\r
2192 case EvqGlobal: return temporaryRegister(operand);
\r
2193 case EvqConstExpr: UNREACHABLE(EvqConstExpr);
\r
2194 case EvqAttribute: return attributeRegister(operand);
\r
2195 case EvqVaryingIn: return varyingRegister(operand);
\r
2196 case EvqVaryingOut: return varyingRegister(operand);
\r
2197 case EvqVertexIn: return attributeRegister(operand);
\r
2198 case EvqFragmentOut: return 0;
\r
2199 case EvqVertexOut: return varyingRegister(operand);
\r
2200 case EvqFragmentIn: return varyingRegister(operand);
\r
2201 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
2202 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
2203 case EvqSmooth: return varyingRegister(operand);
\r
2204 case EvqFlat: return varyingRegister(operand);
\r
2205 case EvqCentroidOut: return varyingRegister(operand);
\r
2206 case EvqSmoothIn: return varyingRegister(operand);
\r
2207 case EvqFlatIn: return varyingRegister(operand);
\r
2208 case EvqCentroidIn: return varyingRegister(operand);
\r
2209 case EvqUniform: return uniformRegister(operand);
\r
2210 case EvqIn: return temporaryRegister(operand);
\r
2211 case EvqOut: return temporaryRegister(operand);
\r
2212 case EvqInOut: return temporaryRegister(operand);
\r
2213 case EvqConstReadOnly: return temporaryRegister(operand);
\r
2214 case EvqPosition: return varyingRegister(operand);
\r
2215 case EvqPointSize: return varyingRegister(operand);
\r
2216 case EvqInstanceID: vertexShader->instanceIdDeclared = true; return 0;
\r
2217 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
2218 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
2219 case EvqPointCoord: return varyingRegister(operand);
\r
2220 case EvqFragColor: return 0;
\r
2221 case EvqFragData: return 0;
\r
2222 case EvqFragDepth: return 0;
\r
2223 default: UNREACHABLE(operand->getQualifier());
\r
2229 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
2231 if(destination->getQualifier() == EvqPointSize)
\r
2233 return 0x2; // Point size stored in the y component
\r
2236 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
2239 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
2241 if(argument->getQualifier() == EvqPointSize)
\r
2243 return 0x55; // Point size stored in the y component
\r
2246 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
2248 return swizzleSize[size];
\r
2251 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
2252 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
2254 if(!expression->isRegister())
\r
2259 return cost(expression, budget) >= 0;
\r
2262 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
2263 int OutputASM::cost(TIntermNode *expression, int budget)
\r
2270 if(expression->getAsSymbolNode())
\r
2274 else if(expression->getAsConstantUnion())
\r
2278 else if(expression->getAsBinaryNode())
\r
2280 TIntermBinary *binary = expression->getAsBinaryNode();
\r
2282 switch(binary->getOp())
\r
2284 case EOpVectorSwizzle:
\r
2285 case EOpIndexDirect:
\r
2286 case EOpIndexDirectStruct:
\r
2287 case EOpIndexDirectInterfaceBlock:
\r
2288 return cost(binary->getLeft(), budget - 0);
\r
2292 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
2297 else if(expression->getAsUnaryNode())
\r
2299 TIntermUnary *unary = expression->getAsUnaryNode();
\r
2301 switch(unary->getOp())
\r
2305 return cost(unary->getOperand(), budget - 1);
\r
2310 else if(expression->getAsSelectionNode())
\r
2312 TIntermSelection *selection = expression->getAsSelectionNode();
\r
2314 if(selection->usesTernaryOperator())
\r
2316 TIntermTyped *condition = selection->getCondition();
\r
2317 TIntermNode *trueBlock = selection->getTrueBlock();
\r
2318 TIntermNode *falseBlock = selection->getFalseBlock();
\r
2319 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
2321 if(constantCondition)
\r
2323 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
2327 return cost(trueBlock, budget - 0);
\r
2331 return cost(falseBlock, budget - 0);
\r
2336 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
2344 const Function *OutputASM::findFunction(const TString &name)
\r
2346 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
2348 if(functionArray[f].name == name)
\r
2350 return &functionArray[f];
\r
2357 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
2359 return allocate(temporaries, temporary);
\r
2362 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
2364 int var = lookup(varyings, varying);
\r
2368 var = allocate(varyings, varying);
\r
2369 int componentCount = varying->registerSize();
\r
2370 int registerCount = varying->totalRegisterCount();
\r
2374 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2376 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2380 if(varying->getQualifier() == EvqPointCoord)
\r
2382 ASSERT(varying->isRegister());
\r
2383 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2384 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2385 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2386 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2390 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2392 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2393 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2394 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2395 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2399 else if(vertexShader)
\r
2401 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2403 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2407 if(varying->getQualifier() == EvqPosition)
\r
2409 ASSERT(varying->isRegister());
\r
2410 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2411 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2412 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2413 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2414 vertexShader->positionRegister = var;
\r
2416 else if(varying->getQualifier() == EvqPointSize)
\r
2418 ASSERT(varying->isRegister());
\r
2419 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2420 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2421 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2422 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2423 vertexShader->pointSizeRegister = var;
\r
2427 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2430 else UNREACHABLE(0);
\r
2432 declareVarying(varying, var);
\r
2438 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2440 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2442 const TType &type = varying->getType();
\r
2443 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2444 VaryingList &activeVaryings = shaderObject->varyings;
\r
2446 // Check if this varying has been declared before without having a register assigned
\r
2447 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2449 if(v->name == name)
\r
2453 ASSERT(v->reg < 0 || v->reg == reg);
\r
2461 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2465 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2467 const TType &type = uniform->getType();
\r
2468 ASSERT(!IsSampler(type.getBasicType()));
\r
2469 TInterfaceBlock *block = type.getAsInterfaceBlock();
\r
2470 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2471 ASSERT(symbol || block);
\r
2473 if(symbol || block)
\r
2475 int index = lookup(uniforms, uniform);
\r
2479 index = allocate(uniforms, uniform);
\r
2480 const TString &name = symbol ? symbol->getSymbol() : block->name();
\r
2482 declareUniform(type, name, index);
\r
2491 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2493 ASSERT(!attribute->isArray());
\r
2495 int index = lookup(attributes, attribute);
\r
2499 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2504 index = allocate(attributes, attribute);
\r
2505 const TType &type = attribute->getType();
\r
2506 int registerCount = attribute->totalRegisterCount();
\r
2508 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2510 for(int i = 0; i < registerCount; i++)
\r
2512 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2516 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2518 const char *name = symbol->getSymbol().c_str();
\r
2519 activeAttributes.push_back(Attribute(glVariableType(type), name, type.getArraySize(), type.getLayoutQualifier().location, index));
\r
2526 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2528 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2529 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2530 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2534 return samplerRegister(symbol);
\r
2538 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect ||
\r
2539 binary->getOp() == EOpIndexDirectStruct || binary->getOp() == EOpIndexDirectInterfaceBlock);
\r
2541 return samplerRegister(binary->getLeft()); // Index added later
\r
2543 else UNREACHABLE(0);
\r
2548 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2550 const TType &type = sampler->getType();
\r
2551 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2553 int index = lookup(samplers, sampler);
\r
2557 index = allocate(samplers, sampler);
\r
2559 if(sampler->getQualifier() == EvqUniform)
\r
2561 const char *name = sampler->getSymbol().c_str();
\r
2562 declareUniform(type, name, index);
\r
2569 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2571 for(unsigned int i = 0; i < list.size(); i++)
\r
2573 if(list[i] == variable)
\r
2575 return i; // Pointer match
\r
2579 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2580 TInterfaceBlock *varBlock = variable->getType().getAsInterfaceBlock();
\r
2584 for(unsigned int i = 0; i < list.size(); i++)
\r
2588 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2592 if(listSymbol->getId() == varSymbol->getId())
\r
2594 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2595 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2596 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2606 for(unsigned int i = 0; i < list.size(); i++)
\r
2610 TInterfaceBlock *listBlock = list[i]->getType().getAsInterfaceBlock();
\r
2614 if(listBlock->name() == varBlock->name())
\r
2616 ASSERT(listBlock->arraySize() == varBlock->arraySize());
\r
2617 ASSERT(listBlock->fields() == varBlock->fields());
\r
2618 ASSERT(listBlock->blockStorage() == varBlock->blockStorage());
\r
2619 ASSERT(listBlock->matrixPacking() == varBlock->matrixPacking());
\r
2631 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2633 int index = lookup(list, variable);
\r
2637 unsigned int registerCount = variable->totalRegisterCount();
\r
2639 for(unsigned int i = 0; i < list.size(); i++)
\r
2643 unsigned int j = 1;
\r
2644 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2646 if(list[i + j] != 0)
\r
2652 if(j == registerCount) // Found free slots
\r
2654 for(unsigned int j = 0; j < registerCount; j++)
\r
2656 list[i + j] = variable;
\r
2664 index = list.size();
\r
2666 for(unsigned int i = 0; i < registerCount; i++)
\r
2668 list.push_back(variable);
\r
2675 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2677 int index = lookup(list, variable);
\r
2685 void OutputASM::declareUniform(const TType &type, const TString &name, int registerIndex, int offset, int blockId)
\r
2687 const TStructure *structure = type.getStruct();
\r
2688 const TInterfaceBlock *block = (type.isInterfaceBlock() || (blockId == -1)) ? type.getInterfaceBlock() : nullptr;
\r
2689 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2693 ActiveUniformBlocks &activeUniformBlocks = shaderObject->activeUniformBlocks;
\r
2694 blockId = activeUniformBlocks.size();
\r
2695 unsigned int dataSize = block->objectSize() * 4; // FIXME: assuming 4 bytes per element
\r
2696 activeUniformBlocks.push_back(UniformBlock(block->name().c_str(), block->hasInstanceName() ? block->instanceName().c_str() : std::string(), dataSize,
\r
2697 block->arraySize(), block->blockStorage(), block->matrixPacking() == EmpRowMajor, registerIndex, blockId));
\r
2700 if(!structure && !block)
\r
2704 shaderObject->activeUniformBlocks[blockId].fields.push_back(activeUniforms.size());
\r
2706 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), registerIndex, offset, blockId));
\r
2708 if(isSamplerRegister(type))
\r
2710 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2712 shader->declareSampler(registerIndex + i);
\r
2718 const TFieldList& fields = structure ? structure->fields() : block->fields();
\r
2719 const bool containerHasName = structure || block->hasInstanceName();
\r
2720 const TString &containerName = structure ? name : (containerHasName ? block->instanceName() : TString());
\r
2721 if(type.isArray() && (structure || type.isInterfaceBlock()))
\r
2723 int fieldRegisterIndex = (blockId == -1) ? registerIndex : 0;
\r
2724 int fieldOffset = 0;
\r
2726 for(int i = 0; i < type.getArraySize(); i++)
\r
2728 for(size_t j = 0; j < fields.size(); j++)
\r
2730 const TType &fieldType = *(fields[j]->type());
\r
2731 const TString &fieldName = fields[j]->name();
\r
2733 const TString uniformName = containerHasName ? containerName + "[" + str(i) + "]." + fieldName : fieldName;
\r
2734 declareUniform(fieldType, uniformName, fieldRegisterIndex, fieldOffset, blockId);
\r
2735 int registerCount = fieldType.totalRegisterCount();
\r
2736 fieldRegisterIndex += registerCount;
\r
2737 fieldOffset += registerCount * fieldType.registerSize();
\r
2743 int fieldRegisterIndex = (blockId == -1) ? registerIndex : 0;
\r
2744 int fieldOffset = 0;
\r
2746 for(size_t i = 0; i < fields.size(); i++)
\r
2748 const TType &fieldType = *(fields[i]->type());
\r
2749 const TString &fieldName = fields[i]->name();
\r
2751 const TString uniformName = containerHasName ? containerName + "." + fieldName : fieldName;
\r
2752 declareUniform(fieldType, uniformName, fieldRegisterIndex, fieldOffset, blockId);
\r
2753 int registerCount = fieldType.totalRegisterCount();
\r
2754 fieldRegisterIndex += registerCount;
\r
2755 fieldOffset += registerCount * fieldType.registerSize();
\r
2761 GLenum OutputASM::glVariableType(const TType &type)
\r
2763 switch(type.getBasicType())
\r
2766 if(type.isScalar())
\r
2770 else if(type.isVector())
\r
2772 switch(type.getNominalSize())
\r
2774 case 2: return GL_FLOAT_VEC2;
\r
2775 case 3: return GL_FLOAT_VEC3;
\r
2776 case 4: return GL_FLOAT_VEC4;
\r
2777 default: UNREACHABLE(type.getNominalSize());
\r
2780 else if(type.isMatrix())
\r
2782 switch(type.getNominalSize())
\r
2785 switch(type.getSecondarySize())
\r
2787 case 2: return GL_FLOAT_MAT2;
\r
2788 case 3: return GL_FLOAT_MAT2x3;
\r
2789 case 4: return GL_FLOAT_MAT2x4;
\r
2790 default: UNREACHABLE(type.getSecondarySize());
\r
2793 switch(type.getSecondarySize())
\r
2795 case 2: return GL_FLOAT_MAT3x2;
\r
2796 case 3: return GL_FLOAT_MAT3;
\r
2797 case 4: return GL_FLOAT_MAT3x4;
\r
2798 default: UNREACHABLE(type.getSecondarySize());
\r
2801 switch(type.getSecondarySize())
\r
2803 case 2: return GL_FLOAT_MAT4x2;
\r
2804 case 3: return GL_FLOAT_MAT4x3;
\r
2805 case 4: return GL_FLOAT_MAT4;
\r
2806 default: UNREACHABLE(type.getSecondarySize());
\r
2808 default: UNREACHABLE(type.getNominalSize());
\r
2811 else UNREACHABLE(0);
\r
2814 if(type.isScalar())
\r
2818 else if(type.isVector())
\r
2820 switch(type.getNominalSize())
\r
2822 case 2: return GL_INT_VEC2;
\r
2823 case 3: return GL_INT_VEC3;
\r
2824 case 4: return GL_INT_VEC4;
\r
2825 default: UNREACHABLE(type.getNominalSize());
\r
2828 else UNREACHABLE(0);
\r
2831 if(type.isScalar())
\r
2833 return GL_UNSIGNED_INT;
\r
2835 else if(type.isVector())
\r
2837 switch(type.getNominalSize())
\r
2839 case 2: return GL_UNSIGNED_INT_VEC2;
\r
2840 case 3: return GL_UNSIGNED_INT_VEC3;
\r
2841 case 4: return GL_UNSIGNED_INT_VEC4;
\r
2842 default: UNREACHABLE(type.getNominalSize());
\r
2845 else UNREACHABLE(0);
\r
2848 if(type.isScalar())
\r
2852 else if(type.isVector())
\r
2854 switch(type.getNominalSize())
\r
2856 case 2: return GL_BOOL_VEC2;
\r
2857 case 3: return GL_BOOL_VEC3;
\r
2858 case 4: return GL_BOOL_VEC4;
\r
2859 default: UNREACHABLE(type.getNominalSize());
\r
2862 else UNREACHABLE(0);
\r
2864 case EbtSampler2D:
\r
2865 return GL_SAMPLER_2D;
\r
2866 case EbtISampler2D:
\r
2867 return GL_INT_SAMPLER_2D;
\r
2868 case EbtUSampler2D:
\r
2869 return GL_UNSIGNED_INT_SAMPLER_2D;
\r
2870 case EbtSamplerCube:
\r
2871 return GL_SAMPLER_CUBE;
\r
2872 case EbtISamplerCube:
\r
2873 return GL_INT_SAMPLER_CUBE;
\r
2874 case EbtUSamplerCube:
\r
2875 return GL_UNSIGNED_INT_SAMPLER_CUBE;
\r
2876 case EbtSamplerExternalOES:
\r
2877 return GL_SAMPLER_EXTERNAL_OES;
\r
2878 case EbtSampler3D:
\r
2879 return GL_SAMPLER_3D_OES;
\r
2880 case EbtISampler3D:
\r
2881 return GL_INT_SAMPLER_3D;
\r
2882 case EbtUSampler3D:
\r
2883 return GL_UNSIGNED_INT_SAMPLER_3D;
\r
2884 case EbtSampler2DArray:
\r
2885 return GL_SAMPLER_2D_ARRAY;
\r
2886 case EbtISampler2DArray:
\r
2887 return GL_INT_SAMPLER_2D_ARRAY;
\r
2888 case EbtUSampler2DArray:
\r
2889 return GL_UNSIGNED_INT_SAMPLER_2D_ARRAY;
\r
2890 case EbtSampler2DShadow:
\r
2891 return GL_SAMPLER_2D_SHADOW;
\r
2892 case EbtSamplerCubeShadow:
\r
2893 return GL_SAMPLER_CUBE_SHADOW;
\r
2894 case EbtSampler2DArrayShadow:
\r
2895 return GL_SAMPLER_2D_ARRAY_SHADOW;
\r
2897 UNREACHABLE(type.getBasicType());
\r
2904 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2906 if(type.getBasicType() == EbtFloat)
\r
2908 switch(type.getPrecision())
\r
2910 case EbpHigh: return GL_HIGH_FLOAT;
\r
2911 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2912 case EbpLow: return GL_LOW_FLOAT;
\r
2913 case EbpUndefined:
\r
2914 // Should be defined as the default precision by the parser
\r
2915 default: UNREACHABLE(type.getPrecision());
\r
2918 else if(type.getBasicType() == EbtInt)
\r
2920 switch(type.getPrecision())
\r
2922 case EbpHigh: return GL_HIGH_INT;
\r
2923 case EbpMedium: return GL_MEDIUM_INT;
\r
2924 case EbpLow: return GL_LOW_INT;
\r
2925 case EbpUndefined:
\r
2926 // Should be defined as the default precision by the parser
\r
2927 default: UNREACHABLE(type.getPrecision());
\r
2931 // Other types (boolean, sampler) don't have a precision
\r
2935 int OutputASM::dim(TIntermNode *v)
\r
2937 TIntermTyped *vector = v->getAsTyped();
\r
2938 ASSERT(vector && vector->isRegister());
\r
2939 return vector->getNominalSize();
\r
2942 int OutputASM::dim2(TIntermNode *m)
\r
2944 TIntermTyped *matrix = m->getAsTyped();
\r
2945 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2946 return matrix->getSecondarySize();
\r
2949 // Returns ~0 if no loop count could be determined
\r
2950 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2952 // Parse loops of the form:
\r
2953 // for(int index = initial; index [comparator] limit; index += increment)
\r
2954 TIntermSymbol *index = 0;
\r
2955 TOperator comparator = EOpNull;
\r
2958 int increment = 0;
\r
2960 // Parse index name and intial value
\r
2961 if(node->getInit())
\r
2963 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2967 TIntermSequence &sequence = init->getSequence();
\r
2968 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2970 if(variable && variable->getQualifier() == EvqTemporary)
\r
2972 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2974 if(assign->getOp() == EOpInitialize)
\r
2976 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2977 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2979 if(symbol && constant)
\r
2981 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2984 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2992 // Parse comparator and limit value
\r
2993 if(index && node->getCondition())
\r
2995 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2997 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2999 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
3003 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
3005 comparator = test->getOp();
\r
3006 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
3012 // Parse increment
\r
3013 if(index && comparator != EOpNull && node->getExpression())
\r
3015 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
3016 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
3018 if(binaryTerminal)
\r
3020 TOperator op = binaryTerminal->getOp();
\r
3021 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
3025 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
3027 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
3031 case EOpAddAssign: increment = value; break;
\r
3032 case EOpSubAssign: increment = -value; break;
\r
3033 default: UNIMPLEMENTED();
\r
3038 else if(unaryTerminal)
\r
3040 TOperator op = unaryTerminal->getOp();
\r
3044 case EOpPostIncrement: increment = 1; break;
\r
3045 case EOpPostDecrement: increment = -1; break;
\r
3046 case EOpPreIncrement: increment = 1; break;
\r
3047 case EOpPreDecrement: increment = -1; break;
\r
3048 default: UNIMPLEMENTED();
\r
3053 if(index && comparator != EOpNull && increment != 0)
\r
3055 if(comparator == EOpLessThanEqual)
\r
3057 comparator = EOpLessThan;
\r
3061 if(comparator == EOpLessThan)
\r
3063 int iterations = (limit - initial) / increment;
\r
3065 if(iterations <= 0)
\r
3070 return iterations;
\r
3072 else UNIMPLEMENTED(); // Falls through
\r
3078 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
3081 loopDiscontinuity = false;
\r
3083 node->traverse(this);
\r
3085 return loopDiscontinuity;
\r
3088 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
3090 if(visit == PreVisit)
\r
3094 else if(visit == PostVisit)
\r
3102 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
3104 if(loopDiscontinuity)
\r
3114 switch(node->getFlowOp())
\r
3121 loopDiscontinuity = true;
\r
3123 default: UNREACHABLE(node->getFlowOp());
\r
3126 return !loopDiscontinuity;
\r
3129 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
3131 return !loopDiscontinuity;
\r
OSDN Git Service
