1 // Copyright 2016 The SwiftShader Authors. All Rights Reserved.
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
7 // http://www.apache.org/licenses/LICENSE-2.0
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
15 #include "OutputASM.h"
16 #include "Common/Math.hpp"
18 #include "common/debug.h"
21 #include "libGLESv2/Shader.h"
23 #include <GLES2/gl2.h>
24 #include <GLES2/gl2ext.h>
25 #include <GLES3/gl3.h>
29 // Integer to TString conversion
33 sprintf(buffer, "%d", i);
37 class Temporary : public TIntermSymbol
40 Temporary(OutputASM *assembler) : TIntermSymbol(TSymbolTableLevel::nextUniqueId(), "tmp", TType(EbtFloat, EbpHigh, EvqTemporary, 4, 1, false)), assembler(assembler)
46 assembler->freeTemporary(this);
50 OutputASM *const assembler;
53 class Constant : public TIntermConstantUnion
56 Constant(float x, float y, float z, float w) : TIntermConstantUnion(constants, TType(EbtFloat, EbpHigh, EvqConstExpr, 4, 1, false))
58 constants[0].setFConst(x);
59 constants[1].setFConst(y);
60 constants[2].setFConst(z);
61 constants[3].setFConst(w);
64 Constant(bool b) : TIntermConstantUnion(constants, TType(EbtBool, EbpHigh, EvqConstExpr, 1, 1, false))
66 constants[0].setBConst(b);
69 Constant(int i) : TIntermConstantUnion(constants, TType(EbtInt, EbpHigh, EvqConstExpr, 1, 1, false))
71 constants[0].setIConst(i);
79 ConstantUnion constants[4];
82 Uniform::Uniform(GLenum type, GLenum precision, const std::string &name, int arraySize, int registerIndex, int blockId, const BlockMemberInfo& blockMemberInfo) :
83 type(type), precision(precision), name(name), arraySize(arraySize), registerIndex(registerIndex), blockId(blockId), blockInfo(blockMemberInfo)
87 UniformBlock::UniformBlock(const std::string& name, unsigned int dataSize, unsigned int arraySize,
88 TLayoutBlockStorage layout, bool isRowMajorLayout, int registerIndex, int blockId) :
89 name(name), dataSize(dataSize), arraySize(arraySize), layout(layout),
90 isRowMajorLayout(isRowMajorLayout), registerIndex(registerIndex), blockId(blockId)
94 BlockLayoutEncoder::BlockLayoutEncoder(bool rowMajor)
95 : mCurrentOffset(0), isRowMajor(rowMajor)
99 BlockMemberInfo BlockLayoutEncoder::encodeType(const TType &type)
104 getBlockLayoutInfo(type, type.getArraySize(), isRowMajor, &arrayStride, &matrixStride);
106 const BlockMemberInfo memberInfo(static_cast<int>(mCurrentOffset * BytesPerComponent),
107 static_cast<int>(arrayStride * BytesPerComponent),
108 static_cast<int>(matrixStride * BytesPerComponent),
109 (matrixStride > 0) && isRowMajor);
111 advanceOffset(type, type.getArraySize(), isRowMajor, arrayStride, matrixStride);
117 size_t BlockLayoutEncoder::getBlockRegister(const BlockMemberInfo &info)
119 return (info.offset / BytesPerComponent) / ComponentsPerRegister;
123 size_t BlockLayoutEncoder::getBlockRegisterElement(const BlockMemberInfo &info)
125 return (info.offset / BytesPerComponent) % ComponentsPerRegister;
128 void BlockLayoutEncoder::nextRegister()
130 mCurrentOffset = sw::align(mCurrentOffset, ComponentsPerRegister);
133 Std140BlockEncoder::Std140BlockEncoder(bool rowMajor) : BlockLayoutEncoder(rowMajor)
137 void Std140BlockEncoder::enterAggregateType()
142 void Std140BlockEncoder::exitAggregateType()
147 void Std140BlockEncoder::getBlockLayoutInfo(const TType &type, unsigned int arraySize, bool isRowMajorMatrix, int *arrayStrideOut, int *matrixStrideOut)
149 size_t baseAlignment = 0;
150 int matrixStride = 0;
155 baseAlignment = ComponentsPerRegister;
156 matrixStride = ComponentsPerRegister;
160 const int numRegisters = isRowMajorMatrix ? type.getSecondarySize() : type.getNominalSize();
161 arrayStride = ComponentsPerRegister * numRegisters;
164 else if(arraySize > 0)
166 baseAlignment = ComponentsPerRegister;
167 arrayStride = ComponentsPerRegister;
171 const size_t numComponents = type.getElementSize();
172 baseAlignment = (numComponents == 3 ? 4u : numComponents);
175 mCurrentOffset = sw::align(mCurrentOffset, baseAlignment);
177 *matrixStrideOut = matrixStride;
178 *arrayStrideOut = arrayStride;
181 void Std140BlockEncoder::advanceOffset(const TType &type, unsigned int arraySize, bool isRowMajorMatrix, int arrayStride, int matrixStride)
185 mCurrentOffset += arrayStride * arraySize;
187 else if(type.isMatrix())
189 ASSERT(matrixStride == ComponentsPerRegister);
190 const int numRegisters = isRowMajorMatrix ? type.getSecondarySize() : type.getNominalSize();
191 mCurrentOffset += ComponentsPerRegister * numRegisters;
195 mCurrentOffset += type.getElementSize();
199 Attribute::Attribute()
206 Attribute::Attribute(GLenum type, const std::string &name, int arraySize, int location, int registerIndex)
210 this->arraySize = arraySize;
211 this->location = location;
212 this->registerIndex = registerIndex;
215 sw::PixelShader *Shader::getPixelShader() const
220 sw::VertexShader *Shader::getVertexShader() const
225 OutputASM::TextureFunction::TextureFunction(const TString& nodeName) : method(IMPLICIT), proj(false), offset(false)
227 TString name = TFunction::unmangleName(nodeName);
229 if(name == "texture2D" || name == "textureCube" || name == "texture" || name == "texture3D")
233 else if(name == "texture2DProj" || name == "textureProj")
238 else if(name == "texture2DLod" || name == "textureCubeLod" || name == "textureLod")
242 else if(name == "texture2DProjLod" || name == "textureProjLod")
247 else if(name == "textureSize")
251 else if(name == "textureOffset")
256 else if(name == "textureProjOffset")
262 else if(name == "textureLodOffset")
267 else if(name == "textureProjLodOffset")
273 else if(name == "texelFetch")
277 else if(name == "texelFetchOffset")
282 else if(name == "textureGrad")
286 else if(name == "textureGradOffset")
291 else if(name == "textureProjGrad")
296 else if(name == "textureProjGradOffset")
305 OutputASM::OutputASM(TParseContext &context, Shader *shaderObject) : TIntermTraverser(true, true, true), shaderObject(shaderObject), mContext(context)
313 shader = shaderObject->getShader();
314 pixelShader = shaderObject->getPixelShader();
315 vertexShader = shaderObject->getVertexShader();
318 functionArray.push_back(Function(0, "main(", 0, 0));
320 outputQualifier = EvqOutput; // Set outputQualifier to any value other than EvqFragColor or EvqFragData
323 OutputASM::~OutputASM()
327 void OutputASM::output()
333 if(functionArray.size() > 1) // Only call main() when there are other functions
335 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
336 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
337 callMain->dst.index = 0; // main()
339 emit(sw::Shader::OPCODE_RET);
342 emitShader(FUNCTION);
346 void OutputASM::emitShader(Scope scope)
349 currentScope = GLOBAL;
350 mContext.getTreeRoot()->traverse(this);
353 void OutputASM::freeTemporary(Temporary *temporary)
355 free(temporaries, temporary);
358 sw::Shader::Opcode OutputASM::getOpcode(sw::Shader::Opcode op, TIntermTyped *in) const
360 TBasicType baseType = in->getType().getBasicType();
364 case sw::Shader::OPCODE_NEG:
369 return sw::Shader::OPCODE_INEG;
374 case sw::Shader::OPCODE_ABS:
378 return sw::Shader::OPCODE_IABS;
383 case sw::Shader::OPCODE_SGN:
387 return sw::Shader::OPCODE_ISGN;
392 case sw::Shader::OPCODE_ADD:
397 return sw::Shader::OPCODE_IADD;
402 case sw::Shader::OPCODE_SUB:
407 return sw::Shader::OPCODE_ISUB;
412 case sw::Shader::OPCODE_MUL:
417 return sw::Shader::OPCODE_IMUL;
422 case sw::Shader::OPCODE_DIV:
426 return sw::Shader::OPCODE_IDIV;
428 return sw::Shader::OPCODE_UDIV;
433 case sw::Shader::OPCODE_IMOD:
434 return baseType == EbtUInt ? sw::Shader::OPCODE_UMOD : op;
435 case sw::Shader::OPCODE_ISHR:
436 return baseType == EbtUInt ? sw::Shader::OPCODE_USHR : op;
437 case sw::Shader::OPCODE_MIN:
441 return sw::Shader::OPCODE_IMIN;
443 return sw::Shader::OPCODE_UMIN;
448 case sw::Shader::OPCODE_MAX:
452 return sw::Shader::OPCODE_IMAX;
454 return sw::Shader::OPCODE_UMAX;
464 void OutputASM::visitSymbol(TIntermSymbol *symbol)
466 // Vertex varyings don't have to be actively used to successfully link
467 // against pixel shaders that use them. So make sure they're declared.
468 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut || symbol->getQualifier() == EvqVertexOut)
470 if(symbol->getBasicType() != EbtInvariant) // Typeless declarations are not new varyings
472 declareVarying(symbol, -1);
476 TInterfaceBlock* block = symbol->getType().getInterfaceBlock();
477 // OpenGL ES 3.0.4 spec, section 2.12.6 Uniform Variables:
478 // "All members of a named uniform block declared with a shared or std140 layout qualifier
479 // are considered active, even if they are not referenced in any shader in the program.
480 // The uniform block itself is also considered active, even if no member of the block is referenced."
481 if(block && ((block->blockStorage() == EbsShared) || (block->blockStorage() == EbsStd140)))
483 uniformRegister(symbol);
487 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
489 if(currentScope != emitScope)
494 TIntermTyped *result = node;
495 TIntermTyped *left = node->getLeft();
496 TIntermTyped *right = node->getRight();
497 const TType &leftType = left->getType();
498 const TType &rightType = right->getType();
500 if(isSamplerRegister(result))
502 return false; // Don't traverse, the register index is determined statically
505 switch(node->getOp())
508 if(visit == PostVisit)
510 assignLvalue(left, right);
515 if(visit == PostVisit)
520 case EOpMatrixTimesScalarAssign:
521 if(visit == PostVisit)
523 for(int i = 0; i < leftType.getNominalSize(); i++)
525 emit(sw::Shader::OPCODE_MUL, result, i, left, i, right);
528 assignLvalue(left, result);
531 case EOpVectorTimesMatrixAssign:
532 if(visit == PostVisit)
534 int size = leftType.getNominalSize();
536 for(int i = 0; i < size; i++)
538 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, 0, left, 0, right, i);
539 dot->dst.mask = 1 << i;
542 assignLvalue(left, result);
545 case EOpMatrixTimesMatrixAssign:
546 if(visit == PostVisit)
548 int dim = leftType.getNominalSize();
550 for(int i = 0; i < dim; i++)
552 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, i, left, 0, right, i);
553 mul->src[1].swizzle = 0x00;
555 for(int j = 1; j < dim; j++)
557 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, i, left, j, right, i, result, i);
558 mad->src[1].swizzle = j * 0x55;
562 assignLvalue(left, result);
566 if(visit == PostVisit)
568 int index = right->getAsConstantUnion()->getIConst(0);
570 if(result->isMatrix() || result->isStruct() || result->isInterfaceBlock())
572 ASSERT(left->isArray());
573 copy(result, left, index * left->elementRegisterCount());
575 else if(result->isRegister())
578 if(left->isRegister())
582 else if(left->isArray())
584 srcIndex = index * left->elementRegisterCount();
586 else if(left->isMatrix())
588 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
593 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, 0, left, srcIndex);
595 if(left->isRegister())
597 mov->src[0].swizzle = index;
603 case EOpIndexIndirect:
604 if(visit == PostVisit)
606 if(left->isArray() || left->isMatrix())
608 for(int index = 0; index < result->totalRegisterCount(); index++)
610 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, index, left, index);
611 mov->dst.mask = writeMask(result, index);
613 if(left->totalRegisterCount() > 1)
615 sw::Shader::SourceParameter relativeRegister;
616 argument(relativeRegister, right);
618 mov->src[0].rel.type = relativeRegister.type;
619 mov->src[0].rel.index = relativeRegister.index;
620 mov->src[0].rel.scale = result->totalRegisterCount();
621 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
625 else if(left->isRegister())
627 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
632 case EOpIndexDirectStruct:
633 case EOpIndexDirectInterfaceBlock:
634 if(visit == PostVisit)
636 ASSERT(leftType.isStruct() || (leftType.isInterfaceBlock()));
638 const TFieldList& fields = (node->getOp() == EOpIndexDirectStruct) ?
639 leftType.getStruct()->fields() :
640 leftType.getInterfaceBlock()->fields();
641 int index = right->getAsConstantUnion()->getIConst(0);
644 for(int i = 0; i < index; i++)
646 fieldOffset += fields[i]->type()->totalRegisterCount();
649 copy(result, left, fieldOffset);
652 case EOpVectorSwizzle:
653 if(visit == PostVisit)
656 TIntermAggregate *components = right->getAsAggregate();
660 TIntermSequence &sequence = components->getSequence();
663 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
665 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
669 int i = element->getUnionArrayPointer()[0].getIConst();
670 swizzle |= i << (component * 2);
678 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
679 mov->src[0].swizzle = swizzle;
682 case EOpAddAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_ADD, result), result, left, left, right); break;
683 case EOpAdd: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_ADD, result), result, left, right); break;
684 case EOpSubAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_SUB, result), result, left, left, right); break;
685 case EOpSub: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_SUB, result), result, left, right); break;
686 case EOpMulAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_MUL, result), result, left, left, right); break;
687 case EOpMul: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_MUL, result), result, left, right); break;
688 case EOpDivAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_DIV, result), result, left, left, right); break;
689 case EOpDiv: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_DIV, result), result, left, right); break;
690 case EOpIModAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_IMOD, result), result, left, left, right); break;
691 case EOpIMod: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_IMOD, result), result, left, right); break;
692 case EOpBitShiftLeftAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SHL, result, left, left, right); break;
693 case EOpBitShiftLeft: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SHL, result, left, right); break;
694 case EOpBitShiftRightAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_ISHR, result), result, left, left, right); break;
695 case EOpBitShiftRight: if(visit == PostVisit) emitBinary(getOpcode(sw::Shader::OPCODE_ISHR, result), result, left, right); break;
696 case EOpBitwiseAndAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_AND, result, left, left, right); break;
697 case EOpBitwiseAnd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_AND, result, left, right); break;
698 case EOpBitwiseXorAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_XOR, result, left, left, right); break;
699 case EOpBitwiseXor: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_XOR, result, left, right); break;
700 case EOpBitwiseOrAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_OR, result, left, left, right); break;
701 case EOpBitwiseOr: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_OR, result, left, right); break;
703 if(visit == PostVisit)
705 emitBinary(sw::Shader::OPCODE_EQ, result, left, right);
707 for(int index = 1; index < left->totalRegisterCount(); index++)
709 Temporary equal(this);
710 emit(sw::Shader::OPCODE_EQ, &equal, 0, left, index, right, index);
711 emit(sw::Shader::OPCODE_AND, result, result, &equal);
716 if(visit == PostVisit)
718 emitBinary(sw::Shader::OPCODE_NE, result, left, right);
720 for(int index = 1; index < left->totalRegisterCount(); index++)
722 Temporary notEqual(this);
723 emit(sw::Shader::OPCODE_NE, ¬Equal, 0, left, index, right, index);
724 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
728 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
729 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
730 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
731 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
732 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(getOpcode(sw::Shader::OPCODE_MUL, left), result, left, left, right); break;
733 case EOpVectorTimesScalar: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MUL, left), result, left, right); break;
734 case EOpMatrixTimesScalar:
735 if(visit == PostVisit)
739 for(int i = 0; i < leftType.getNominalSize(); i++)
741 emit(sw::Shader::OPCODE_MUL, result, i, left, i, right, 0);
744 else if(right->isMatrix())
746 for(int i = 0; i < rightType.getNominalSize(); i++)
748 emit(sw::Shader::OPCODE_MUL, result, i, left, 0, right, i);
754 case EOpVectorTimesMatrix:
755 if(visit == PostVisit)
757 sw::Shader::Opcode dpOpcode = sw::Shader::OPCODE_DP(leftType.getNominalSize());
759 int size = rightType.getNominalSize();
760 for(int i = 0; i < size; i++)
762 Instruction *dot = emit(dpOpcode, result, 0, left, 0, right, i);
763 dot->dst.mask = 1 << i;
767 case EOpMatrixTimesVector:
768 if(visit == PostVisit)
770 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
771 mul->src[1].swizzle = 0x00;
773 int size = rightType.getNominalSize();
774 for(int i = 1; i < size; i++)
776 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, 0, left, i, right, 0, result);
777 mad->src[1].swizzle = i * 0x55;
781 case EOpMatrixTimesMatrix:
782 if(visit == PostVisit)
784 int dim = leftType.getNominalSize();
786 int size = rightType.getNominalSize();
787 for(int i = 0; i < size; i++)
789 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, i, left, 0, right, i);
790 mul->src[1].swizzle = 0x00;
792 for(int j = 1; j < dim; j++)
794 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, i, left, j, right, i, result, i);
795 mad->src[1].swizzle = j * 0x55;
801 if(trivial(right, 6))
803 if(visit == PostVisit)
805 emit(sw::Shader::OPCODE_OR, result, left, right);
808 else // Short-circuit evaluation
812 emit(sw::Shader::OPCODE_MOV, result, left);
813 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
814 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
816 else if(visit == PostVisit)
818 emit(sw::Shader::OPCODE_MOV, result, right);
819 emit(sw::Shader::OPCODE_ENDIF);
823 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
825 if(trivial(right, 6))
827 if(visit == PostVisit)
829 emit(sw::Shader::OPCODE_AND, result, left, right);
832 else // Short-circuit evaluation
836 emit(sw::Shader::OPCODE_MOV, result, left);
837 emit(sw::Shader::OPCODE_IF, 0, result);
839 else if(visit == PostVisit)
841 emit(sw::Shader::OPCODE_MOV, result, right);
842 emit(sw::Shader::OPCODE_ENDIF);
846 default: UNREACHABLE(node->getOp());
852 void OutputASM::emitDeterminant(TIntermTyped *result, TIntermTyped *arg, int size, int col, int row, int outCol, int outRow)
856 case 1: // Used for cofactor computation only
858 // For a 2x2 matrix, the cofactor is simply a transposed move or negate
859 bool isMov = (row == col);
860 sw::Shader::Opcode op = isMov ? sw::Shader::OPCODE_MOV : sw::Shader::OPCODE_NEG;
861 Instruction *mov = emit(op, result, outCol, arg, isMov ? 1 - row : row);
862 mov->src[0].swizzle = 0x55 * (isMov ? 1 - col : col);
863 mov->dst.mask = 1 << outRow;
868 static const unsigned int swizzle[3] = { 0x99, 0x88, 0x44 }; // xy?? : yzyz, xzxz, xyxy
870 bool isCofactor = (col >= 0) && (row >= 0);
871 int col0 = (isCofactor && (col <= 0)) ? 1 : 0;
872 int col1 = (isCofactor && (col <= 1)) ? 2 : 1;
873 bool negate = isCofactor && ((col & 0x01) ^ (row & 0x01));
875 Instruction *det = emit(sw::Shader::OPCODE_DET2, result, outCol, arg, negate ? col1 : col0, arg, negate ? col0 : col1);
876 det->src[0].swizzle = det->src[1].swizzle = swizzle[isCofactor ? row : 2];
877 det->dst.mask = 1 << outRow;
882 static const unsigned int swizzle[4] = { 0xF9, 0xF8, 0xF4, 0xE4 }; // xyz? : yzww, xzww, xyww, xyzw
884 bool isCofactor = (col >= 0) && (row >= 0);
885 int col0 = (isCofactor && (col <= 0)) ? 1 : 0;
886 int col1 = (isCofactor && (col <= 1)) ? 2 : 1;
887 int col2 = (isCofactor && (col <= 2)) ? 3 : 2;
888 bool negate = isCofactor && ((col & 0x01) ^ (row & 0x01));
890 Instruction *det = emit(sw::Shader::OPCODE_DET3, result, outCol, arg, col0, arg, negate ? col2 : col1, arg, negate ? col1 : col2);
891 det->src[0].swizzle = det->src[1].swizzle = det->src[2].swizzle = swizzle[isCofactor ? row : 3];
892 det->dst.mask = 1 << outRow;
897 Instruction *det = emit(sw::Shader::OPCODE_DET4, result, outCol, arg, 0, arg, 1, arg, 2, arg, 3);
898 det->dst.mask = 1 << outRow;
907 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
909 if(currentScope != emitScope)
914 TIntermTyped *result = node;
915 TIntermTyped *arg = node->getOperand();
916 TBasicType basicType = arg->getType().getBasicType();
924 if(basicType == EbtInt || basicType == EbtUInt)
933 Constant one(one_value.f, one_value.f, one_value.f, one_value.f);
934 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
935 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
937 switch(node->getOp())
940 if(visit == PostVisit)
942 sw::Shader::Opcode negOpcode = getOpcode(sw::Shader::OPCODE_NEG, arg);
943 for(int index = 0; index < arg->totalRegisterCount(); index++)
945 emit(negOpcode, result, index, arg, index);
949 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
950 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
951 case EOpPostIncrement:
952 if(visit == PostVisit)
956 sw::Shader::Opcode addOpcode = getOpcode(sw::Shader::OPCODE_ADD, arg);
957 for(int index = 0; index < arg->totalRegisterCount(); index++)
959 emit(addOpcode, arg, index, arg, index, &one);
962 assignLvalue(arg, arg);
965 case EOpPostDecrement:
966 if(visit == PostVisit)
970 sw::Shader::Opcode subOpcode = getOpcode(sw::Shader::OPCODE_SUB, arg);
971 for(int index = 0; index < arg->totalRegisterCount(); index++)
973 emit(subOpcode, arg, index, arg, index, &one);
976 assignLvalue(arg, arg);
979 case EOpPreIncrement:
980 if(visit == PostVisit)
982 sw::Shader::Opcode addOpcode = getOpcode(sw::Shader::OPCODE_ADD, arg);
983 for(int index = 0; index < arg->totalRegisterCount(); index++)
985 emit(addOpcode, result, index, arg, index, &one);
988 assignLvalue(arg, result);
991 case EOpPreDecrement:
992 if(visit == PostVisit)
994 sw::Shader::Opcode subOpcode = getOpcode(sw::Shader::OPCODE_SUB, arg);
995 for(int index = 0; index < arg->totalRegisterCount(); index++)
997 emit(subOpcode, result, index, arg, index, &one);
1000 assignLvalue(arg, result);
1003 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
1004 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
1005 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
1006 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
1007 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
1008 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
1009 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
1010 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
1011 case EOpSinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_SINH, result, arg); break;
1012 case EOpCosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_COSH, result, arg); break;
1013 case EOpTanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_TANH, result, arg); break;
1014 case EOpAsinh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASINH, result, arg); break;
1015 case EOpAcosh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOSH, result, arg); break;
1016 case EOpAtanh: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATANH, result, arg); break;
1017 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
1018 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
1019 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
1020 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
1021 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
1022 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
1023 case EOpAbs: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_ABS, result), result, arg); break;
1024 case EOpSign: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_SGN, result), result, arg); break;
1025 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
1026 case EOpTrunc: if(visit == PostVisit) emit(sw::Shader::OPCODE_TRUNC, result, arg); break;
1027 case EOpRound: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUND, result, arg); break;
1028 case EOpRoundEven: if(visit == PostVisit) emit(sw::Shader::OPCODE_ROUNDEVEN, result, arg); break;
1029 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
1030 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
1031 case EOpIsNan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISNAN, result, arg); break;
1032 case EOpIsInf: if(visit == PostVisit) emit(sw::Shader::OPCODE_ISINF, result, arg); break;
1033 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
1034 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
1035 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
1036 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
1037 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
1038 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
1039 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
1040 case EOpFloatBitsToInt: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOINT, result, arg); break;
1041 case EOpFloatBitsToUint: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOATBITSTOUINT, result, arg); break;
1042 case EOpIntBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_INTBITSTOFLOAT, result, arg); break;
1043 case EOpUintBitsToFloat: if(visit == PostVisit) emit(sw::Shader::OPCODE_UINTBITSTOFLOAT, result, arg); break;
1044 case EOpPackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKSNORM2x16, result, arg); break;
1045 case EOpPackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKUNORM2x16, result, arg); break;
1046 case EOpPackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_PACKHALF2x16, result, arg); break;
1047 case EOpUnpackSnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKSNORM2x16, result, arg); break;
1048 case EOpUnpackUnorm2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKUNORM2x16, result, arg); break;
1049 case EOpUnpackHalf2x16: if(visit == PostVisit) emit(sw::Shader::OPCODE_UNPACKHALF2x16, result, arg); break;
1051 if(visit == PostVisit)
1053 int numCols = arg->getNominalSize();
1054 int numRows = arg->getSecondarySize();
1055 for(int i = 0; i < numCols; ++i)
1057 for(int j = 0; j < numRows; ++j)
1059 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, j, arg, i);
1060 mov->src[0].swizzle = 0x55 * j;
1061 mov->dst.mask = 1 << i;
1066 case EOpDeterminant:
1067 if(visit == PostVisit)
1069 int size = arg->getNominalSize();
1070 ASSERT(size == arg->getSecondarySize());
1072 emitDeterminant(result, arg, size);
1076 if(visit == PostVisit)
1078 int size = arg->getNominalSize();
1079 ASSERT(size == arg->getSecondarySize());
1081 // Compute transposed matrix of cofactors
1082 for(int i = 0; i < size; ++i)
1084 for(int j = 0; j < size; ++j)
1086 // For a 2x2 matrix, the cofactor is simply a transposed move or negate
1087 // For a 3x3 or 4x4 matrix, the cofactor is a transposed determinant
1088 emitDeterminant(result, arg, size - 1, j, i, i, j);
1092 // Compute 1 / determinant
1093 Temporary invDet(this);
1094 emitDeterminant(&invDet, arg, size);
1095 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
1096 Instruction *div = emit(sw::Shader::OPCODE_DIV, &invDet, &one, &invDet);
1097 div->src[1].swizzle = 0x00; // xxxx
1099 // Divide transposed matrix of cofactors by determinant
1100 for(int i = 0; i < size; ++i)
1102 emit(sw::Shader::OPCODE_MUL, result, i, result, i, &invDet);
1106 default: UNREACHABLE(node->getOp());
1112 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
1114 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
1119 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
1121 TIntermTyped *result = node;
1122 const TType &resultType = node->getType();
1123 TIntermSequence &arg = node->getSequence();
1124 size_t argumentCount = arg.size();
1126 switch(node->getOp())
1128 case EOpSequence: break;
1129 case EOpDeclaration: break;
1130 case EOpInvariantDeclaration: break;
1131 case EOpPrototype: break;
1133 if(visit == PostVisit)
1135 copy(result, arg[1]);
1139 if(visit == PreVisit)
1141 const TString &name = node->getName();
1143 if(emitScope == FUNCTION)
1145 if(functionArray.size() > 1) // No need for a label when there's only main()
1147 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
1148 label->dst.type = sw::Shader::PARAMETER_LABEL;
1150 const Function *function = findFunction(name);
1151 ASSERT(function); // Should have been added during global pass
1152 label->dst.index = function->label;
1153 currentFunction = function->label;
1156 else if(emitScope == GLOBAL)
1160 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
1161 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
1164 else UNREACHABLE(emitScope);
1166 currentScope = FUNCTION;
1168 else if(visit == PostVisit)
1170 if(emitScope == FUNCTION)
1172 if(functionArray.size() > 1) // No need to return when there's only main()
1174 emit(sw::Shader::OPCODE_RET);
1178 currentScope = GLOBAL;
1181 case EOpFunctionCall:
1182 if(visit == PostVisit)
1184 if(node->isUserDefined())
1186 const TString &name = node->getName();
1187 const Function *function = findFunction(name);
1191 mContext.error(node->getLine(), "function definition not found", name.c_str());
1195 TIntermSequence &arguments = *function->arg;
1197 for(size_t i = 0; i < argumentCount; i++)
1199 TIntermTyped *in = arguments[i]->getAsTyped();
1201 if(in->getQualifier() == EvqIn ||
1202 in->getQualifier() == EvqInOut ||
1203 in->getQualifier() == EvqConstReadOnly)
1209 Instruction *call = emit(sw::Shader::OPCODE_CALL);
1210 call->dst.type = sw::Shader::PARAMETER_LABEL;
1211 call->dst.index = function->label;
1213 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
1215 copy(result, function->ret);
1218 for(size_t i = 0; i < argumentCount; i++)
1220 TIntermTyped *argument = arguments[i]->getAsTyped();
1221 TIntermTyped *out = arg[i]->getAsTyped();
1223 if(argument->getQualifier() == EvqOut ||
1224 argument->getQualifier() == EvqInOut)
1226 assignLvalue(out, argument);
1232 const TextureFunction textureFunction(node->getName());
1233 TIntermTyped *t = arg[1]->getAsTyped();
1235 Temporary coord(this);
1237 if(textureFunction.proj)
1239 Instruction *rcp = emit(sw::Shader::OPCODE_RCPX, &coord, arg[1]);
1240 rcp->src[0].swizzle = 0x55 * (t->getNominalSize() - 1);
1241 rcp->dst.mask = 0x7;
1243 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &coord, arg[1], &coord);
1244 mul->dst.mask = 0x7;
1248 emit(sw::Shader::OPCODE_MOV, &coord, arg[1]);
1251 switch(textureFunction.method)
1253 case TextureFunction::IMPLICIT:
1255 TIntermNode* offset = textureFunction.offset ? arg[2] : 0;
1257 if(argumentCount == 2 || (textureFunction.offset && argumentCount == 3))
1259 emit(textureFunction.offset ? sw::Shader::OPCODE_TEXOFFSET : sw::Shader::OPCODE_TEX,
1260 result, &coord, arg[0], offset);
1262 else if(argumentCount == 3 || (textureFunction.offset && argumentCount == 4)) // bias
1264 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &coord, arg[textureFunction.offset ? 3 : 2]);
1265 bias->dst.mask = 0x8;
1267 Instruction *tex = emit(textureFunction.offset ? sw::Shader::OPCODE_TEXOFFSET : sw::Shader::OPCODE_TEX,
1268 result, &coord, arg[0], offset); // FIXME: Implement an efficient TEXLDB instruction
1271 else UNREACHABLE(argumentCount);
1274 case TextureFunction::LOD:
1276 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &coord, arg[2]);
1277 lod->dst.mask = 0x8;
1279 emit(textureFunction.offset ? sw::Shader::OPCODE_TEXLDLOFFSET : sw::Shader::OPCODE_TEXLDL,
1280 result, &coord, arg[0], textureFunction.offset ? arg[3] : nullptr);
1283 case TextureFunction::FETCH:
1285 if(argumentCount == 3 || (textureFunction.offset && argumentCount == 4))
1287 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &coord, arg[2]);
1288 lod->dst.mask = 0x8;
1290 TIntermNode *offset = textureFunction.offset ? arg[3] : nullptr;
1292 emit(textureFunction.offset ? sw::Shader::OPCODE_TEXELFETCHOFFSET : sw::Shader::OPCODE_TEXELFETCH,
1293 result, &coord, arg[0], offset);
1295 else UNREACHABLE(argumentCount);
1298 case TextureFunction::GRAD:
1300 if(argumentCount == 4 || (textureFunction.offset && argumentCount == 5))
1302 TIntermNode *offset = textureFunction.offset ? arg[4] : nullptr;
1304 emit(textureFunction.offset ? sw::Shader::OPCODE_TEXGRADOFFSET : sw::Shader::OPCODE_TEXGRAD,
1305 result, &coord, arg[0], arg[2], arg[3], offset);
1307 else UNREACHABLE(argumentCount);
1310 case TextureFunction::SIZE:
1311 emit(sw::Shader::OPCODE_TEXSIZE, result, arg[1], arg[0]);
1314 UNREACHABLE(textureFunction.method);
1321 case EOpConstructFloat:
1322 case EOpConstructVec2:
1323 case EOpConstructVec3:
1324 case EOpConstructVec4:
1325 case EOpConstructBool:
1326 case EOpConstructBVec2:
1327 case EOpConstructBVec3:
1328 case EOpConstructBVec4:
1329 case EOpConstructInt:
1330 case EOpConstructIVec2:
1331 case EOpConstructIVec3:
1332 case EOpConstructIVec4:
1333 case EOpConstructUInt:
1334 case EOpConstructUVec2:
1335 case EOpConstructUVec3:
1336 case EOpConstructUVec4:
1337 if(visit == PostVisit)
1341 for(size_t i = 0; i < argumentCount; i++)
1343 TIntermTyped *argi = arg[i]->getAsTyped();
1344 int size = argi->getNominalSize();
1346 if(!argi->isMatrix())
1348 Instruction *mov = emitCast(result, argi);
1349 mov->dst.mask = (0xF << component) & 0xF;
1350 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
1358 while(component < resultType.getNominalSize())
1360 Instruction *mov = emitCast(result, 0, argi, column);
1361 mov->dst.mask = (0xF << component) & 0xF;
1362 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
1371 case EOpConstructMat2:
1372 case EOpConstructMat2x3:
1373 case EOpConstructMat2x4:
1374 case EOpConstructMat3x2:
1375 case EOpConstructMat3:
1376 case EOpConstructMat3x4:
1377 case EOpConstructMat4x2:
1378 case EOpConstructMat4x3:
1379 case EOpConstructMat4:
1380 if(visit == PostVisit)
1382 TIntermTyped *arg0 = arg[0]->getAsTyped();
1383 const int outCols = result->getNominalSize();
1384 const int outRows = result->getSecondarySize();
1386 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
1388 for(int i = 0; i < outCols; i++)
1390 emit(sw::Shader::OPCODE_MOV, result, i, &zero);
1391 Instruction *mov = emitCast(result, i, arg0, 0);
1392 mov->dst.mask = 1 << i;
1393 ASSERT(mov->src[0].swizzle == 0x00);
1396 else if(arg0->isMatrix())
1398 const int inCols = arg0->getNominalSize();
1399 const int inRows = arg0->getSecondarySize();
1401 for(int i = 0; i < outCols; i++)
1403 if(i >= inCols || outRows > inRows)
1405 // Initialize to identity matrix
1406 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));
1407 emitCast(result, i, &col, 0);
1412 Instruction *mov = emitCast(result, i, arg0, i);
1413 mov->dst.mask = 0xF >> (4 - inRows);
1422 for(size_t i = 0; i < argumentCount; i++)
1424 TIntermTyped *argi = arg[i]->getAsTyped();
1425 int size = argi->getNominalSize();
1428 while(element < size)
1430 Instruction *mov = emitCast(result, column, argi, 0);
1431 mov->dst.mask = (0xF << row) & 0xF;
1432 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
1434 int end = row + size - element;
1435 column = end >= outRows ? column + 1 : column;
1436 element = element + outRows - row;
1437 row = end >= outRows ? 0 : end;
1443 case EOpConstructStruct:
1444 if(visit == PostVisit)
1447 for(size_t i = 0; i < argumentCount; i++)
1449 TIntermTyped *argi = arg[i]->getAsTyped();
1450 int size = argi->totalRegisterCount();
1452 for(int index = 0; index < size; index++)
1454 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, index + offset, argi, index);
1455 mov->dst.mask = writeMask(result, offset + index);
1462 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
1463 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
1464 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
1465 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
1466 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
1467 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
1468 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
1470 if(visit == PostVisit)
1472 TIntermTyped* arg1 = arg[1]->getAsTyped();
1473 emit(sw::Shader::OPCODE_TRUNC, arg1, arg[0]);
1474 assignLvalue(arg1, arg1);
1475 emitBinary(sw::Shader::OPCODE_SUB, result, arg[0], arg1);
1478 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
1479 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
1480 case EOpMin: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MIN, result), result, arg[0], arg[1]); break;
1481 case EOpMax: if(visit == PostVisit) emit(getOpcode(sw::Shader::OPCODE_MAX, result), result, arg[0], arg[1]); break;
1483 if(visit == PostVisit)
1485 emit(getOpcode(sw::Shader::OPCODE_MAX, result), result, arg[0], arg[1]);
1486 emit(getOpcode(sw::Shader::OPCODE_MIN, result), result, result, arg[2]);
1489 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
1490 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
1491 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
1492 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
1493 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
1494 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
1495 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
1496 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
1497 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
1499 if(visit == PostVisit)
1501 TIntermTyped *arg0 = arg[0]->getAsTyped();
1502 TIntermTyped *arg1 = arg[1]->getAsTyped();
1503 ASSERT((arg0->getNominalSize() == arg1->getNominalSize()) && (arg0->getSecondarySize() == arg1->getSecondarySize()));
1505 int size = arg0->getNominalSize();
1506 for(int i = 0; i < size; i++)
1508 emit(sw::Shader::OPCODE_MUL, result, i, arg[0], i, arg[1], i);
1512 case EOpOuterProduct:
1513 if(visit == PostVisit)
1515 for(int i = 0; i < dim(arg[1]); i++)
1517 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, i, arg[0], 0, arg[1]);
1518 mul->src[1].swizzle = 0x55 * i;
1522 default: UNREACHABLE(node->getOp());
1528 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
1530 if(currentScope != emitScope)
1535 TIntermTyped *condition = node->getCondition();
1536 TIntermNode *trueBlock = node->getTrueBlock();
1537 TIntermNode *falseBlock = node->getFalseBlock();
1538 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
1540 condition->traverse(this);
1542 if(node->usesTernaryOperator())
1544 if(constantCondition)
1546 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
1550 trueBlock->traverse(this);
1551 copy(node, trueBlock);
1555 falseBlock->traverse(this);
1556 copy(node, falseBlock);
1559 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
1561 trueBlock->traverse(this);
1562 falseBlock->traverse(this);
1563 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
1567 emit(sw::Shader::OPCODE_IF, 0, condition);
1571 trueBlock->traverse(this);
1572 copy(node, trueBlock);
1577 emit(sw::Shader::OPCODE_ELSE);
1578 falseBlock->traverse(this);
1579 copy(node, falseBlock);
1582 emit(sw::Shader::OPCODE_ENDIF);
1585 else // if/else statement
1587 if(constantCondition)
1589 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
1595 trueBlock->traverse(this);
1602 falseBlock->traverse(this);
1608 emit(sw::Shader::OPCODE_IF, 0, condition);
1612 trueBlock->traverse(this);
1617 emit(sw::Shader::OPCODE_ELSE);
1618 falseBlock->traverse(this);
1621 emit(sw::Shader::OPCODE_ENDIF);
1628 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
1630 if(currentScope != emitScope)
1635 unsigned int iterations = loopCount(node);
1642 bool unroll = (iterations <= 4);
1646 LoopUnrollable loopUnrollable;
1647 unroll = loopUnrollable.traverse(node);
1650 TIntermNode *init = node->getInit();
1651 TIntermTyped *condition = node->getCondition();
1652 TIntermTyped *expression = node->getExpression();
1653 TIntermNode *body = node->getBody();
1654 Constant True(true);
1656 if(node->getType() == ELoopDoWhile)
1658 Temporary iterate(this);
1659 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
1661 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
1665 body->traverse(this);
1668 emit(sw::Shader::OPCODE_TEST);
1670 condition->traverse(this);
1671 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
1673 emit(sw::Shader::OPCODE_ENDWHILE);
1679 init->traverse(this);
1684 for(unsigned int i = 0; i < iterations; i++)
1686 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
1690 body->traverse(this);
1695 expression->traverse(this);
1703 condition->traverse(this);
1710 emit(sw::Shader::OPCODE_WHILE, 0, condition);
1714 body->traverse(this);
1717 emit(sw::Shader::OPCODE_TEST);
1721 expression->traverse(this);
1726 condition->traverse(this);
1729 emit(sw::Shader::OPCODE_ENDWHILE);
1736 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
1738 if(currentScope != emitScope)
1743 switch(node->getFlowOp())
1745 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
1746 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
1747 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
1749 if(visit == PostVisit)
1751 TIntermTyped *value = node->getExpression();
1755 copy(functionArray[currentFunction].ret, value);
1758 emit(sw::Shader::OPCODE_LEAVE);
1761 default: UNREACHABLE(node->getFlowOp());
1767 bool OutputASM::visitSwitch(Visit visit, TIntermSwitch *node)
1769 if(currentScope != emitScope)
1774 TIntermTyped* switchValue = node->getInit();
1775 TIntermAggregate* opList = node->getStatementList();
1777 if(!switchValue || !opList)
1782 switchValue->traverse(this);
1784 emit(sw::Shader::OPCODE_SWITCH);
1786 TIntermSequence& sequence = opList->getSequence();
1787 TIntermSequence::iterator it = sequence.begin();
1788 TIntermSequence::iterator defaultIt = sequence.end();
1790 for(; it != sequence.end(); ++it)
1792 TIntermCase* currentCase = (*it)->getAsCaseNode();
1795 TIntermSequence::iterator caseIt = it;
1797 TIntermTyped* condition = currentCase->getCondition();
1798 if(condition) // non default case
1802 emit(sw::Shader::OPCODE_ELSE);
1805 condition->traverse(this);
1806 Temporary result(this);
1807 emitBinary(sw::Shader::OPCODE_EQ, &result, switchValue, condition);
1808 emit(sw::Shader::OPCODE_IF, 0, &result);
1811 for(++caseIt; caseIt != sequence.end(); ++caseIt)
1813 (*caseIt)->traverse(this);
1814 if((*caseIt)->getAsBranchNode()) // Kill, Break, Continue or Return
1822 defaultIt = it; // The default case might not be the last case, keep it for last
1827 // If there's a default case, traverse it here
1828 if(defaultIt != sequence.end())
1830 emit(sw::Shader::OPCODE_ELSE);
1831 for(++defaultIt; defaultIt != sequence.end(); ++defaultIt)
1833 (*defaultIt)->traverse(this);
1834 if((*defaultIt)->getAsBranchNode()) // Kill, Break, Continue or Return
1841 for(int i = 0; i < nbCases; ++i)
1843 emit(sw::Shader::OPCODE_ENDIF);
1846 emit(sw::Shader::OPCODE_ENDSWITCH);
1851 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, TIntermNode *src3, TIntermNode *src4)
1853 return emit(op, dst, 0, src0, 0, src1, 0, src2, 0, src3, 0, src4, 0);
1856 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, int dstIndex, TIntermNode *src0, int index0, TIntermNode *src1, int index1,
1857 TIntermNode *src2, int index2, TIntermNode *src3, int index3, TIntermNode *src4, int index4)
1859 Instruction *instruction = new Instruction(op);
1863 instruction->dst.type = registerType(dst);
1864 instruction->dst.index = registerIndex(dst) + dstIndex;
1865 instruction->dst.mask = writeMask(dst);
1866 instruction->dst.integer = (dst->getBasicType() == EbtInt);
1869 argument(instruction->src[0], src0, index0);
1870 argument(instruction->src[1], src1, index1);
1871 argument(instruction->src[2], src2, index2);
1872 argument(instruction->src[3], src3, index3);
1873 argument(instruction->src[4], src4, index4);
1875 shader->append(instruction);
1880 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
1882 return emitCast(dst, 0, src, 0);
1885 Instruction *OutputASM::emitCast(TIntermTyped *dst, int dstIndex, TIntermTyped *src, int srcIndex)
1887 switch(src->getBasicType())
1890 switch(dst->getBasicType())
1892 case EbtInt: return emit(sw::Shader::OPCODE_B2I, dst, dstIndex, src, srcIndex);
1893 case EbtUInt: return emit(sw::Shader::OPCODE_B2I, dst, dstIndex, src, srcIndex);
1894 case EbtFloat: return emit(sw::Shader::OPCODE_B2F, dst, dstIndex, src, srcIndex);
1899 switch(dst->getBasicType())
1901 case EbtBool: return emit(sw::Shader::OPCODE_I2B, dst, dstIndex, src, srcIndex);
1902 case EbtFloat: return emit(sw::Shader::OPCODE_I2F, dst, dstIndex, src, srcIndex);
1907 switch(dst->getBasicType())
1909 case EbtBool: return emit(sw::Shader::OPCODE_I2B, dst, dstIndex, src, srcIndex);
1910 case EbtFloat: return emit(sw::Shader::OPCODE_U2F, dst, dstIndex, src, srcIndex);
1915 switch(dst->getBasicType())
1917 case EbtBool: return emit(sw::Shader::OPCODE_F2B, dst, dstIndex, src, srcIndex);
1918 case EbtInt: return emit(sw::Shader::OPCODE_F2I, dst, dstIndex, src, srcIndex);
1919 case EbtUInt: return emit(sw::Shader::OPCODE_F2U, dst, dstIndex, src, srcIndex);
1927 ASSERT((src->getBasicType() == dst->getBasicType()) ||
1928 ((src->getBasicType() == EbtInt) && (dst->getBasicType() == EbtUInt)) ||
1929 ((src->getBasicType() == EbtUInt) && (dst->getBasicType() == EbtInt)));
1931 return emit(sw::Shader::OPCODE_MOV, dst, dstIndex, src, srcIndex);
1934 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
1936 for(int index = 0; index < dst->elementRegisterCount(); index++)
1938 emit(op, dst, index, src0, index, src1, index, src2, index);
1942 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
1944 emitBinary(op, result, src0, src1);
1945 assignLvalue(lhs, result);
1948 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
1950 sw::Shader::Opcode opcode;
1951 switch(left->getAsTyped()->getBasicType())
1955 opcode = sw::Shader::OPCODE_ICMP;
1958 opcode = sw::Shader::OPCODE_UCMP;
1961 opcode = sw::Shader::OPCODE_CMP;
1965 Instruction *cmp = emit(opcode, dst, 0, left, index, right, index);
1966 cmp->control = cmpOp;
1969 int componentCount(const TType &type, int registers)
1976 if(type.isArray() && registers >= type.elementRegisterCount())
1978 int index = registers / type.elementRegisterCount();
1979 registers -= index * type.elementRegisterCount();
1980 return index * type.getElementSize() + componentCount(type, registers);
1983 if(type.isStruct() || type.isInterfaceBlock())
1985 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
1988 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
1990 const TType &fieldType = *((*field)->type());
1992 if(fieldType.totalRegisterCount() <= registers)
1994 registers -= fieldType.totalRegisterCount();
1995 elements += fieldType.getObjectSize();
1997 else // Register within this field
1999 return elements + componentCount(fieldType, registers);
2003 else if(type.isMatrix())
2005 return registers * type.registerSize();
2012 int registerSize(const TType &type, int registers)
2018 return registerSize(*((*(type.getStruct()->fields().begin()))->type()), 0);
2020 else if(type.isInterfaceBlock())
2022 return registerSize(*((*(type.getInterfaceBlock()->fields().begin()))->type()), 0);
2025 return type.registerSize();
2028 if(type.isArray() && registers >= type.elementRegisterCount())
2030 int index = registers / type.elementRegisterCount();
2031 registers -= index * type.elementRegisterCount();
2032 return registerSize(type, registers);
2035 if(type.isStruct() || type.isInterfaceBlock())
2037 const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
2040 for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
2042 const TType &fieldType = *((*field)->type());
2044 if(fieldType.totalRegisterCount() <= registers)
2046 registers -= fieldType.totalRegisterCount();
2047 elements += fieldType.getObjectSize();
2049 else // Register within this field
2051 return registerSize(fieldType, registers);
2055 else if(type.isMatrix())
2057 return registerSize(type, 0);
2064 int OutputASM::getBlockId(TIntermTyped *arg)
2068 const TType &type = arg->getType();
2069 TInterfaceBlock* block = type.getInterfaceBlock();
2070 if(block && (type.getQualifier() == EvqUniform))
2072 // Make sure the uniform block is declared
2073 uniformRegister(arg);
2075 const char* blockName = block->name().c_str();
2077 // Fetch uniform block index from array of blocks
2078 for(ActiveUniformBlocks::const_iterator it = shaderObject->activeUniformBlocks.begin(); it != shaderObject->activeUniformBlocks.end(); ++it)
2080 if(blockName == it->name)
2093 OutputASM::ArgumentInfo OutputASM::getArgumentInfo(TIntermTyped *arg, int index)
2095 const TType &type = arg->getType();
2096 int blockId = getBlockId(arg);
2097 ArgumentInfo argumentInfo(BlockMemberInfo::getDefaultBlockInfo(), type, -1, -1);
2100 argumentInfo.bufferIndex = 0;
2101 for(int i = 0; i < blockId; ++i)
2103 int blockArraySize = shaderObject->activeUniformBlocks[i].arraySize;
2104 argumentInfo.bufferIndex += blockArraySize > 0 ? blockArraySize : 1;
2107 const BlockDefinitionIndexMap& blockDefinition = blockDefinitions[blockId];
2109 BlockDefinitionIndexMap::const_iterator itEnd = blockDefinition.end();
2110 BlockDefinitionIndexMap::const_iterator it = itEnd;
2112 argumentInfo.clampedIndex = index;
2113 if(type.isInterfaceBlock())
2115 // Offset index to the beginning of the selected instance
2116 int blockRegisters = type.elementRegisterCount();
2117 int bufferOffset = argumentInfo.clampedIndex / blockRegisters;
2118 argumentInfo.bufferIndex += bufferOffset;
2119 argumentInfo.clampedIndex -= bufferOffset * blockRegisters;
2122 int regIndex = registerIndex(arg);
2123 for(int i = regIndex + argumentInfo.clampedIndex; i >= regIndex; --i)
2125 it = blockDefinition.find(i);
2128 argumentInfo.clampedIndex -= (i - regIndex);
2132 ASSERT(it != itEnd);
2134 argumentInfo.typedMemberInfo = it->second;
2136 int registerCount = argumentInfo.typedMemberInfo.type.totalRegisterCount();
2137 argumentInfo.clampedIndex = (argumentInfo.clampedIndex >= registerCount) ? registerCount - 1 : argumentInfo.clampedIndex;
2141 argumentInfo.clampedIndex = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
2144 return argumentInfo;
2147 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
2151 TIntermTyped *arg = argument->getAsTyped();
2152 Temporary unpackedUniform(this);
2154 const TType& srcType = arg->getType();
2155 TInterfaceBlock* srcBlock = srcType.getInterfaceBlock();
2156 if(srcBlock && (srcType.getQualifier() == EvqUniform))
2158 const ArgumentInfo argumentInfo = getArgumentInfo(arg, index);
2159 const TType &memberType = argumentInfo.typedMemberInfo.type;
2161 if(memberType.getBasicType() == EbtBool)
2163 int arraySize = (memberType.isArray() ? memberType.getArraySize() : 1);
2164 ASSERT(argumentInfo.clampedIndex < arraySize);
2166 // Convert the packed bool, which is currently an int, to a true bool
2167 Instruction *instruction = new Instruction(sw::Shader::OPCODE_I2B);
2168 instruction->dst.type = sw::Shader::PARAMETER_TEMP;
2169 instruction->dst.index = registerIndex(&unpackedUniform);
2170 instruction->src[0].type = sw::Shader::PARAMETER_CONST;
2171 instruction->src[0].bufferIndex = argumentInfo.bufferIndex;
2172 instruction->src[0].index = argumentInfo.typedMemberInfo.offset + argumentInfo.clampedIndex * argumentInfo.typedMemberInfo.arrayStride;
2174 shader->append(instruction);
2176 arg = &unpackedUniform;
2179 else if((srcBlock->matrixPacking() == EmpRowMajor) && memberType.isMatrix())
2181 int numCols = memberType.getNominalSize();
2182 int numRows = memberType.getSecondarySize();
2183 int arraySize = (memberType.isArray() ? memberType.getArraySize() : 1);
2185 ASSERT(argumentInfo.clampedIndex < (numCols * arraySize));
2187 unsigned int dstIndex = registerIndex(&unpackedUniform);
2188 unsigned int srcSwizzle = (argumentInfo.clampedIndex % numCols) * 0x55;
2189 int arrayIndex = argumentInfo.clampedIndex / numCols;
2190 int matrixStartOffset = argumentInfo.typedMemberInfo.offset + arrayIndex * argumentInfo.typedMemberInfo.arrayStride;
2192 for(int j = 0; j < numRows; ++j)
2194 // Transpose the row major matrix
2195 Instruction *instruction = new Instruction(sw::Shader::OPCODE_MOV);
2196 instruction->dst.type = sw::Shader::PARAMETER_TEMP;
2197 instruction->dst.index = dstIndex;
2198 instruction->dst.mask = 1 << j;
2199 instruction->src[0].type = sw::Shader::PARAMETER_CONST;
2200 instruction->src[0].bufferIndex = argumentInfo.bufferIndex;
2201 instruction->src[0].index = matrixStartOffset + j * argumentInfo.typedMemberInfo.matrixStride;
2202 instruction->src[0].swizzle = srcSwizzle;
2204 shader->append(instruction);
2207 arg = &unpackedUniform;
2212 const ArgumentInfo argumentInfo = getArgumentInfo(arg, index);
2213 const TType &type = argumentInfo.typedMemberInfo.type;
2215 int size = registerSize(type, argumentInfo.clampedIndex);
2217 parameter.type = registerType(arg);
2218 parameter.bufferIndex = argumentInfo.bufferIndex;
2220 if(arg->getAsConstantUnion() && arg->getAsConstantUnion()->getUnionArrayPointer())
2222 int component = componentCount(type, argumentInfo.clampedIndex);
2223 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
2225 for(int i = 0; i < 4; i++)
2227 if(size == 1) // Replicate
2229 parameter.value[i] = constants[component + 0].getAsFloat();
2233 parameter.value[i] = constants[component + i].getAsFloat();
2237 parameter.value[i] = 0.0f;
2243 parameter.index = registerIndex(arg) + argumentInfo.clampedIndex;
2245 if(parameter.bufferIndex != -1)
2247 int stride = (argumentInfo.typedMemberInfo.matrixStride > 0) ? argumentInfo.typedMemberInfo.matrixStride : argumentInfo.typedMemberInfo.arrayStride;
2248 parameter.index = argumentInfo.typedMemberInfo.offset + argumentInfo.clampedIndex * stride;
2252 if(!IsSampler(arg->getBasicType()))
2254 parameter.swizzle = readSwizzle(arg, size);
2259 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
2261 for(int index = 0; index < dst->totalRegisterCount(); index++)
2263 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, index, src, offset + index);
2264 mov->dst.mask = writeMask(dst, index);
2268 int swizzleElement(int swizzle, int index)
2270 return (swizzle >> (index * 2)) & 0x03;
2273 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
2275 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
2276 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
2277 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
2278 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
2281 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
2284 ((src->isVector() && (!dst->isVector() || (src->getNominalSize() != dst->getNominalSize()))) ||
2285 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize()) || (src->getSecondarySize() != dst->getSecondarySize())))))
2287 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
2290 TIntermBinary *binary = dst->getAsBinaryNode();
2292 if(binary && binary->getOp() == EOpIndexIndirect && binary->getLeft()->isVector() && dst->isScalar())
2294 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
2296 Temporary address(this);
2297 lvalue(insert->dst, address, dst);
2299 insert->src[0].type = insert->dst.type;
2300 insert->src[0].index = insert->dst.index;
2301 insert->src[0].rel = insert->dst.rel;
2302 argument(insert->src[1], src);
2303 argument(insert->src[2], binary->getRight());
2305 shader->append(insert);
2309 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
2311 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
2313 Temporary address(this);
2314 int swizzle = lvalue(mov->dst, address, dst);
2315 mov->dst.index += offset;
2319 mov->dst.mask = writeMask(dst, offset);
2322 argument(mov->src[0], src, offset);
2323 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
2325 shader->append(mov);
2330 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
2332 TIntermTyped *result = node;
2333 TIntermBinary *binary = node->getAsBinaryNode();
2334 TIntermSymbol *symbol = node->getAsSymbolNode();
2338 TIntermTyped *left = binary->getLeft();
2339 TIntermTyped *right = binary->getRight();
2341 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
2343 switch(binary->getOp())
2345 case EOpIndexDirect:
2347 int rightIndex = right->getAsConstantUnion()->getIConst(0);
2349 if(left->isRegister())
2351 int leftMask = dst.mask;
2354 while((leftMask & dst.mask) == 0)
2356 dst.mask = dst.mask << 1;
2359 int element = swizzleElement(leftSwizzle, rightIndex);
2360 dst.mask = 1 << element;
2364 else if(left->isArray() || left->isMatrix())
2366 dst.index += rightIndex * result->totalRegisterCount();
2369 else UNREACHABLE(0);
2372 case EOpIndexIndirect:
2374 if(left->isRegister())
2376 // Requires INSERT instruction (handled by calling function)
2378 else if(left->isArray() || left->isMatrix())
2380 int scale = result->totalRegisterCount();
2382 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
2384 if(left->totalRegisterCount() > 1)
2386 sw::Shader::SourceParameter relativeRegister;
2387 argument(relativeRegister, right);
2389 dst.rel.index = relativeRegister.index;
2390 dst.rel.type = relativeRegister.type;
2391 dst.rel.scale = scale;
2392 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
2395 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
2399 Constant oldScale((int)dst.rel.scale);
2400 Instruction *mad = emit(sw::Shader::OPCODE_IMAD, &address, &address, &oldScale, right);
2401 mad->src[0].index = dst.rel.index;
2402 mad->src[0].type = dst.rel.type;
2406 Constant oldScale((int)dst.rel.scale);
2407 Instruction *mul = emit(sw::Shader::OPCODE_IMUL, &address, &address, &oldScale);
2408 mul->src[0].index = dst.rel.index;
2409 mul->src[0].type = dst.rel.type;
2411 Constant newScale(scale);
2412 emit(sw::Shader::OPCODE_IMAD, &address, right, &newScale, &address);
2415 dst.rel.type = sw::Shader::PARAMETER_TEMP;
2416 dst.rel.index = registerIndex(&address);
2419 else // Just add the new index to the address register
2423 emit(sw::Shader::OPCODE_IADD, &address, &address, right);
2427 Constant newScale(scale);
2428 emit(sw::Shader::OPCODE_IMAD, &address, right, &newScale, &address);
2432 else UNREACHABLE(0);
2435 case EOpIndexDirectStruct:
2436 case EOpIndexDirectInterfaceBlock:
2438 const TFieldList& fields = (binary->getOp() == EOpIndexDirectStruct) ?
2439 left->getType().getStruct()->fields() :
2440 left->getType().getInterfaceBlock()->fields();
2441 int index = right->getAsConstantUnion()->getIConst(0);
2442 int fieldOffset = 0;
2444 for(int i = 0; i < index; i++)
2446 fieldOffset += fields[i]->type()->totalRegisterCount();
2449 dst.type = registerType(left);
2450 dst.index += fieldOffset;
2451 dst.mask = writeMask(right);
2456 case EOpVectorSwizzle:
2458 ASSERT(left->isRegister());
2460 int leftMask = dst.mask;
2465 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
2467 for(unsigned int i = 0; i < sequence.size(); i++)
2469 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
2471 int element = swizzleElement(leftSwizzle, index);
2472 rightMask = rightMask | (1 << element);
2473 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
2476 dst.mask = leftMask & rightMask;
2482 UNREACHABLE(binary->getOp()); // Not an l-value operator
2488 dst.type = registerType(symbol);
2489 dst.index = registerIndex(symbol);
2490 dst.mask = writeMask(symbol);
2497 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
2499 if(isSamplerRegister(operand))
2501 return sw::Shader::PARAMETER_SAMPLER;
2504 const TQualifier qualifier = operand->getQualifier();
2505 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
2507 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
2508 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
2510 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
2512 outputQualifier = qualifier;
2515 if(qualifier == EvqConstExpr && (!operand->getAsConstantUnion() || !operand->getAsConstantUnion()->getUnionArrayPointer()))
2517 return sw::Shader::PARAMETER_TEMP;
2522 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
2523 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
2524 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
2525 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
2526 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
2527 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
2528 case EvqVertexIn: return sw::Shader::PARAMETER_INPUT;
2529 case EvqFragmentOut: return sw::Shader::PARAMETER_COLOROUT;
2530 case EvqVertexOut: return sw::Shader::PARAMETER_OUTPUT;
2531 case EvqFragmentIn: return sw::Shader::PARAMETER_INPUT;
2532 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
2533 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
2534 case EvqSmooth: return sw::Shader::PARAMETER_OUTPUT;
2535 case EvqFlat: return sw::Shader::PARAMETER_OUTPUT;
2536 case EvqCentroidOut: return sw::Shader::PARAMETER_OUTPUT;
2537 case EvqSmoothIn: return sw::Shader::PARAMETER_INPUT;
2538 case EvqFlatIn: return sw::Shader::PARAMETER_INPUT;
2539 case EvqCentroidIn: return sw::Shader::PARAMETER_INPUT;
2540 case EvqUniform: return sw::Shader::PARAMETER_CONST;
2541 case EvqIn: return sw::Shader::PARAMETER_TEMP;
2542 case EvqOut: return sw::Shader::PARAMETER_TEMP;
2543 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
2544 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
2545 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
2546 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
2547 case EvqInstanceID: return sw::Shader::PARAMETER_MISCTYPE;
2548 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
2549 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
2550 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
2551 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
2552 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
2553 case EvqFragDepth: return sw::Shader::PARAMETER_DEPTHOUT;
2554 default: UNREACHABLE(qualifier);
2557 return sw::Shader::PARAMETER_VOID;
2560 bool OutputASM::hasFlatQualifier(TIntermTyped *operand)
2562 const TQualifier qualifier = operand->getQualifier();
2563 return qualifier == EvqFlat || qualifier == EvqFlatOut || qualifier == EvqFlatIn;
2566 unsigned int OutputASM::registerIndex(TIntermTyped *operand)
2568 if(isSamplerRegister(operand))
2570 return samplerRegister(operand);
2573 switch(operand->getQualifier())
2575 case EvqTemporary: return temporaryRegister(operand);
2576 case EvqGlobal: return temporaryRegister(operand);
2577 case EvqConstExpr: return temporaryRegister(operand); // Unevaluated constant expression
2578 case EvqAttribute: return attributeRegister(operand);
2579 case EvqVaryingIn: return varyingRegister(operand);
2580 case EvqVaryingOut: return varyingRegister(operand);
2581 case EvqVertexIn: return attributeRegister(operand);
2582 case EvqFragmentOut: return fragmentOutputRegister(operand);
2583 case EvqVertexOut: return varyingRegister(operand);
2584 case EvqFragmentIn: return varyingRegister(operand);
2585 case EvqInvariantVaryingIn: return varyingRegister(operand);
2586 case EvqInvariantVaryingOut: return varyingRegister(operand);
2587 case EvqSmooth: return varyingRegister(operand);
2588 case EvqFlat: return varyingRegister(operand);
2589 case EvqCentroidOut: return varyingRegister(operand);
2590 case EvqSmoothIn: return varyingRegister(operand);
2591 case EvqFlatIn: return varyingRegister(operand);
2592 case EvqCentroidIn: return varyingRegister(operand);
2593 case EvqUniform: return uniformRegister(operand);
2594 case EvqIn: return temporaryRegister(operand);
2595 case EvqOut: return temporaryRegister(operand);
2596 case EvqInOut: return temporaryRegister(operand);
2597 case EvqConstReadOnly: return temporaryRegister(operand);
2598 case EvqPosition: return varyingRegister(operand);
2599 case EvqPointSize: return varyingRegister(operand);
2600 case EvqInstanceID: vertexShader->instanceIdDeclared = true; return 0;
2601 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
2602 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
2603 case EvqPointCoord: return varyingRegister(operand);
2604 case EvqFragColor: return 0;
2605 case EvqFragData: return fragmentOutputRegister(operand);
2606 case EvqFragDepth: return 0;
2607 default: UNREACHABLE(operand->getQualifier());
2613 int OutputASM::writeMask(TIntermTyped *destination, int index)
2615 if(destination->getQualifier() == EvqPointSize)
2617 return 0x2; // Point size stored in the y component
2620 return 0xF >> (4 - registerSize(destination->getType(), index));
2623 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
2625 if(argument->getQualifier() == EvqPointSize)
2627 return 0x55; // Point size stored in the y component
2630 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
2632 return swizzleSize[size];
2635 // Conservatively checks whether an expression is fast to compute and has no side effects
2636 bool OutputASM::trivial(TIntermTyped *expression, int budget)
2638 if(!expression->isRegister())
2643 return cost(expression, budget) >= 0;
2646 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
2647 int OutputASM::cost(TIntermNode *expression, int budget)
2654 if(expression->getAsSymbolNode())
2658 else if(expression->getAsConstantUnion())
2662 else if(expression->getAsBinaryNode())
2664 TIntermBinary *binary = expression->getAsBinaryNode();
2666 switch(binary->getOp())
2668 case EOpVectorSwizzle:
2669 case EOpIndexDirect:
2670 case EOpIndexDirectStruct:
2671 case EOpIndexDirectInterfaceBlock:
2672 return cost(binary->getLeft(), budget - 0);
2676 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
2681 else if(expression->getAsUnaryNode())
2683 TIntermUnary *unary = expression->getAsUnaryNode();
2685 switch(unary->getOp())
2689 return cost(unary->getOperand(), budget - 1);
2694 else if(expression->getAsSelectionNode())
2696 TIntermSelection *selection = expression->getAsSelectionNode();
2698 if(selection->usesTernaryOperator())
2700 TIntermTyped *condition = selection->getCondition();
2701 TIntermNode *trueBlock = selection->getTrueBlock();
2702 TIntermNode *falseBlock = selection->getFalseBlock();
2703 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
2705 if(constantCondition)
2707 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
2711 return cost(trueBlock, budget - 0);
2715 return cost(falseBlock, budget - 0);
2720 return cost(trueBlock, cost(falseBlock, budget - 2));
2728 const Function *OutputASM::findFunction(const TString &name)
2730 for(unsigned int f = 0; f < functionArray.size(); f++)
2732 if(functionArray[f].name == name)
2734 return &functionArray[f];
2741 int OutputASM::temporaryRegister(TIntermTyped *temporary)
2743 return allocate(temporaries, temporary);
2746 int OutputASM::varyingRegister(TIntermTyped *varying)
2748 int var = lookup(varyings, varying);
2752 var = allocate(varyings, varying);
2753 int componentCount = varying->registerSize();
2754 int registerCount = varying->totalRegisterCount();
2758 if((var + registerCount) > sw::MAX_FRAGMENT_INPUTS)
2760 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
2764 if(varying->getQualifier() == EvqPointCoord)
2766 ASSERT(varying->isRegister());
2767 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
2768 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
2769 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
2770 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
2774 for(int i = 0; i < varying->totalRegisterCount(); i++)
2776 bool flat = hasFlatQualifier(varying);
2778 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i, flat);
2779 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i, flat);
2780 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i, flat);
2781 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i, flat);
2785 else if(vertexShader)
2787 if((var + registerCount) > sw::MAX_VERTEX_OUTPUTS)
2789 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
2793 if(varying->getQualifier() == EvqPosition)
2795 ASSERT(varying->isRegister());
2796 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
2797 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
2798 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
2799 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
2800 vertexShader->positionRegister = var;
2802 else if(varying->getQualifier() == EvqPointSize)
2804 ASSERT(varying->isRegister());
2805 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
2806 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
2807 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
2808 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
2809 vertexShader->pointSizeRegister = var;
2813 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
2816 else UNREACHABLE(0);
2818 declareVarying(varying, var);
2824 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
2826 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
2828 const TType &type = varying->getType();
2829 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
2830 VaryingList &activeVaryings = shaderObject->varyings;
2832 // Check if this varying has been declared before without having a register assigned
2833 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
2839 ASSERT(v->reg < 0 || v->reg == reg);
2847 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
2851 int OutputASM::uniformRegister(TIntermTyped *uniform)
2853 const TType &type = uniform->getType();
2854 ASSERT(!IsSampler(type.getBasicType()));
2855 TInterfaceBlock *block = type.getAsInterfaceBlock();
2856 TIntermSymbol *symbol = uniform->getAsSymbolNode();
2857 ASSERT(symbol || block);
2861 TInterfaceBlock* parentBlock = type.getInterfaceBlock();
2862 bool isBlockMember = (!block && parentBlock);
2863 int index = isBlockMember ? lookup(uniforms, parentBlock) : lookup(uniforms, uniform);
2865 if(index == -1 || isBlockMember)
2869 index = allocate(uniforms, uniform);
2872 // Verify if the current uniform is a member of an already declared block
2873 const TString &name = symbol ? symbol->getSymbol() : block->name();
2874 int blockMemberIndex = blockMemberLookup(type, name, index);
2875 if(blockMemberIndex == -1)
2877 declareUniform(type, name, index);
2881 index = blockMemberIndex;
2891 int OutputASM::attributeRegister(TIntermTyped *attribute)
2893 ASSERT(!attribute->isArray());
2895 int index = lookup(attributes, attribute);
2899 TIntermSymbol *symbol = attribute->getAsSymbolNode();
2904 index = allocate(attributes, attribute);
2905 const TType &type = attribute->getType();
2906 int registerCount = attribute->totalRegisterCount();
2908 if(vertexShader && (index + registerCount) <= sw::MAX_VERTEX_INPUTS)
2910 for(int i = 0; i < registerCount; i++)
2912 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
2916 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
2918 const char *name = symbol->getSymbol().c_str();
2919 activeAttributes.push_back(Attribute(glVariableType(type), name, type.getArraySize(), type.getLayoutQualifier().location, index));
2926 int OutputASM::fragmentOutputRegister(TIntermTyped *fragmentOutput)
2928 return allocate(fragmentOutputs, fragmentOutput);
2931 int OutputASM::samplerRegister(TIntermTyped *sampler)
2933 const TType &type = sampler->getType();
2934 ASSERT(IsSampler(type.getBasicType()) || type.isStruct()); // Structures can contain samplers
2936 TIntermSymbol *symbol = sampler->getAsSymbolNode();
2937 TIntermBinary *binary = sampler->getAsBinaryNode();
2939 if(symbol && type.getQualifier() == EvqUniform)
2941 return samplerRegister(symbol);
2945 TIntermTyped *left = binary->getLeft();
2946 TIntermTyped *right = binary->getRight();
2947 const TType &leftType = left->getType();
2948 int index = right->getAsConstantUnion() ? right->getAsConstantUnion()->getIConst(0) : 0;
2951 switch(binary->getOp())
2953 case EOpIndexDirect:
2954 ASSERT(left->isArray());
2955 offset = index * leftType.elementRegisterCount();
2957 case EOpIndexDirectStruct:
2958 ASSERT(leftType.isStruct());
2960 const TFieldList &fields = leftType.getStruct()->fields();
2962 for(int i = 0; i < index; i++)
2964 offset += fields[i]->type()->totalRegisterCount();
2968 case EOpIndexIndirect: // Indirect indexing produces a temporary, not a sampler register
2970 case EOpIndexDirectInterfaceBlock: // Interface blocks can't contain samplers
2972 UNREACHABLE(binary->getOp());
2976 int base = samplerRegister(left);
2983 return base + offset;
2987 return -1; // Not a sampler register
2990 int OutputASM::samplerRegister(TIntermSymbol *sampler)
2992 const TType &type = sampler->getType();
2993 ASSERT(IsSampler(type.getBasicType()) || type.isStruct()); // Structures can contain samplers
2995 int index = lookup(samplers, sampler);
2999 index = allocate(samplers, sampler);
3001 if(sampler->getQualifier() == EvqUniform)
3003 const char *name = sampler->getSymbol().c_str();
3004 declareUniform(type, name, index);
3011 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
3013 return operand && IsSampler(operand->getBasicType()) && samplerRegister(operand) >= 0;
3016 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
3018 for(unsigned int i = 0; i < list.size(); i++)
3020 if(list[i] == variable)
3022 return i; // Pointer match
3026 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
3027 TInterfaceBlock *varBlock = variable->getType().getAsInterfaceBlock();
3031 for(unsigned int i = 0; i < list.size(); i++)
3035 TInterfaceBlock *listBlock = list[i]->getType().getAsInterfaceBlock();
3039 if(listBlock->name() == varBlock->name())
3041 ASSERT(listBlock->arraySize() == varBlock->arraySize());
3042 ASSERT(listBlock->fields() == varBlock->fields());
3043 ASSERT(listBlock->blockStorage() == varBlock->blockStorage());
3044 ASSERT(listBlock->matrixPacking() == varBlock->matrixPacking());
3054 for(unsigned int i = 0; i < list.size(); i++)
3058 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
3062 if(listSymbol->getId() == varSymbol->getId())
3064 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
3065 ASSERT(listSymbol->getType() == varSymbol->getType());
3066 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
3078 int OutputASM::lookup(VariableArray &list, TInterfaceBlock *block)
3080 for(unsigned int i = 0; i < list.size(); i++)
3082 if(list[i] && (list[i]->getType().getInterfaceBlock() == block))
3084 return i; // Pointer match
3090 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
3092 int index = lookup(list, variable);
3096 unsigned int registerCount = variable->blockRegisterCount();
3098 for(unsigned int i = 0; i < list.size(); i++)
3103 for( ; j < registerCount && (i + j) < list.size(); j++)
3105 if(list[i + j] != 0)
3111 if(j == registerCount) // Found free slots
3113 for(unsigned int j = 0; j < registerCount; j++)
3115 list[i + j] = variable;
3123 index = list.size();
3125 for(unsigned int i = 0; i < registerCount; i++)
3127 list.push_back(variable);
3134 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
3136 int index = lookup(list, variable);
3144 int OutputASM::blockMemberLookup(const TType &type, const TString &name, int registerIndex)
3146 const TInterfaceBlock *block = type.getInterfaceBlock();
3150 ActiveUniformBlocks &activeUniformBlocks = shaderObject->activeUniformBlocks;
3151 const TFieldList& fields = block->fields();
3152 const TString &blockName = block->name();
3153 int fieldRegisterIndex = registerIndex;
3155 if(!type.isInterfaceBlock())
3157 // This is a uniform that's part of a block, let's see if the block is already defined
3158 for(size_t i = 0; i < activeUniformBlocks.size(); ++i)
3160 if(activeUniformBlocks[i].name == blockName.c_str())
3162 // The block is already defined, find the register for the current uniform and return it
3163 for(size_t j = 0; j < fields.size(); j++)
3165 const TString &fieldName = fields[j]->name();
3166 if(fieldName == name)
3168 return fieldRegisterIndex;
3171 fieldRegisterIndex += fields[j]->type()->totalRegisterCount();
3175 return fieldRegisterIndex;
3184 void OutputASM::declareUniform(const TType &type, const TString &name, int registerIndex, int blockId, BlockLayoutEncoder* encoder)
3186 const TStructure *structure = type.getStruct();
3187 const TInterfaceBlock *block = (type.isInterfaceBlock() || (blockId == -1)) ? type.getInterfaceBlock() : nullptr;
3189 if(!structure && !block)
3191 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
3192 const BlockMemberInfo blockInfo = encoder ? encoder->encodeType(type) : BlockMemberInfo::getDefaultBlockInfo();
3195 blockDefinitions[blockId][registerIndex] = TypedMemberInfo(blockInfo, type);
3196 shaderObject->activeUniformBlocks[blockId].fields.push_back(activeUniforms.size());
3198 int fieldRegisterIndex = encoder ? shaderObject->activeUniformBlocks[blockId].registerIndex + BlockLayoutEncoder::getBlockRegister(blockInfo) : registerIndex;
3199 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(),
3200 fieldRegisterIndex, blockId, blockInfo));
3201 if(IsSampler(type.getBasicType()))
3203 for(int i = 0; i < type.totalRegisterCount(); i++)
3205 shader->declareSampler(fieldRegisterIndex + i);
3211 ActiveUniformBlocks &activeUniformBlocks = shaderObject->activeUniformBlocks;
3212 const TFieldList& fields = block->fields();
3213 const TString &blockName = block->name();
3214 int fieldRegisterIndex = registerIndex;
3215 bool isUniformBlockMember = !type.isInterfaceBlock() && (blockId == -1);
3217 blockId = activeUniformBlocks.size();
3218 bool isRowMajor = block->matrixPacking() == EmpRowMajor;
3219 activeUniformBlocks.push_back(UniformBlock(blockName.c_str(), 0, block->arraySize(),
3220 block->blockStorage(), isRowMajor, registerIndex, blockId));
3221 blockDefinitions.push_back(BlockDefinitionIndexMap());
3223 Std140BlockEncoder currentBlockEncoder(isRowMajor);
3224 currentBlockEncoder.enterAggregateType();
3225 for(size_t i = 0; i < fields.size(); i++)
3227 const TType &fieldType = *(fields[i]->type());
3228 const TString &fieldName = fields[i]->name();
3229 if(isUniformBlockMember && (fieldName == name))
3231 registerIndex = fieldRegisterIndex;
3234 const TString uniformName = block->hasInstanceName() ? blockName + "." + fieldName : fieldName;
3236 declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, ¤tBlockEncoder);
3237 fieldRegisterIndex += fieldType.totalRegisterCount();
3239 currentBlockEncoder.exitAggregateType();
3240 activeUniformBlocks[blockId].dataSize = currentBlockEncoder.getBlockSize();
3244 int fieldRegisterIndex = registerIndex;
3246 const TFieldList& fields = structure->fields();
3247 if(type.isArray() && (structure || type.isInterfaceBlock()))
3249 for(int i = 0; i < type.getArraySize(); i++)
3253 encoder->enterAggregateType();
3255 for(size_t j = 0; j < fields.size(); j++)
3257 const TType &fieldType = *(fields[j]->type());
3258 const TString &fieldName = fields[j]->name();
3259 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
3261 declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, encoder);
3262 fieldRegisterIndex += fieldType.totalRegisterCount();
3266 encoder->exitAggregateType();
3274 encoder->enterAggregateType();
3276 for(size_t i = 0; i < fields.size(); i++)
3278 const TType &fieldType = *(fields[i]->type());
3279 const TString &fieldName = fields[i]->name();
3280 const TString uniformName = name + "." + fieldName;
3282 declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, encoder);
3283 fieldRegisterIndex += fieldType.totalRegisterCount();
3287 encoder->exitAggregateType();
3293 GLenum OutputASM::glVariableType(const TType &type)
3295 switch(type.getBasicType())
3302 else if(type.isVector())
3304 switch(type.getNominalSize())
3306 case 2: return GL_FLOAT_VEC2;
3307 case 3: return GL_FLOAT_VEC3;
3308 case 4: return GL_FLOAT_VEC4;
3309 default: UNREACHABLE(type.getNominalSize());
3312 else if(type.isMatrix())
3314 switch(type.getNominalSize())
3317 switch(type.getSecondarySize())
3319 case 2: return GL_FLOAT_MAT2;
3320 case 3: return GL_FLOAT_MAT2x3;
3321 case 4: return GL_FLOAT_MAT2x4;
3322 default: UNREACHABLE(type.getSecondarySize());
3325 switch(type.getSecondarySize())
3327 case 2: return GL_FLOAT_MAT3x2;
3328 case 3: return GL_FLOAT_MAT3;
3329 case 4: return GL_FLOAT_MAT3x4;
3330 default: UNREACHABLE(type.getSecondarySize());
3333 switch(type.getSecondarySize())
3335 case 2: return GL_FLOAT_MAT4x2;
3336 case 3: return GL_FLOAT_MAT4x3;
3337 case 4: return GL_FLOAT_MAT4;
3338 default: UNREACHABLE(type.getSecondarySize());
3340 default: UNREACHABLE(type.getNominalSize());
3343 else UNREACHABLE(0);
3350 else if(type.isVector())
3352 switch(type.getNominalSize())
3354 case 2: return GL_INT_VEC2;
3355 case 3: return GL_INT_VEC3;
3356 case 4: return GL_INT_VEC4;
3357 default: UNREACHABLE(type.getNominalSize());
3360 else UNREACHABLE(0);
3365 return GL_UNSIGNED_INT;
3367 else if(type.isVector())
3369 switch(type.getNominalSize())
3371 case 2: return GL_UNSIGNED_INT_VEC2;
3372 case 3: return GL_UNSIGNED_INT_VEC3;
3373 case 4: return GL_UNSIGNED_INT_VEC4;
3374 default: UNREACHABLE(type.getNominalSize());
3377 else UNREACHABLE(0);
3384 else if(type.isVector())
3386 switch(type.getNominalSize())
3388 case 2: return GL_BOOL_VEC2;
3389 case 3: return GL_BOOL_VEC3;
3390 case 4: return GL_BOOL_VEC4;
3391 default: UNREACHABLE(type.getNominalSize());
3394 else UNREACHABLE(0);
3397 return GL_SAMPLER_2D;
3399 return GL_INT_SAMPLER_2D;
3401 return GL_UNSIGNED_INT_SAMPLER_2D;
3402 case EbtSamplerCube:
3403 return GL_SAMPLER_CUBE;
3404 case EbtISamplerCube:
3405 return GL_INT_SAMPLER_CUBE;
3406 case EbtUSamplerCube:
3407 return GL_UNSIGNED_INT_SAMPLER_CUBE;
3408 case EbtSamplerExternalOES:
3409 return GL_SAMPLER_EXTERNAL_OES;
3411 return GL_SAMPLER_3D_OES;
3413 return GL_INT_SAMPLER_3D;
3415 return GL_UNSIGNED_INT_SAMPLER_3D;
3416 case EbtSampler2DArray:
3417 return GL_SAMPLER_2D_ARRAY;
3418 case EbtISampler2DArray:
3419 return GL_INT_SAMPLER_2D_ARRAY;
3420 case EbtUSampler2DArray:
3421 return GL_UNSIGNED_INT_SAMPLER_2D_ARRAY;
3422 case EbtSampler2DShadow:
3423 return GL_SAMPLER_2D_SHADOW;
3424 case EbtSamplerCubeShadow:
3425 return GL_SAMPLER_CUBE_SHADOW;
3426 case EbtSampler2DArrayShadow:
3427 return GL_SAMPLER_2D_ARRAY_SHADOW;
3429 UNREACHABLE(type.getBasicType());
3436 GLenum OutputASM::glVariablePrecision(const TType &type)
3438 if(type.getBasicType() == EbtFloat)
3440 switch(type.getPrecision())
3442 case EbpHigh: return GL_HIGH_FLOAT;
3443 case EbpMedium: return GL_MEDIUM_FLOAT;
3444 case EbpLow: return GL_LOW_FLOAT;
3446 // Should be defined as the default precision by the parser
3447 default: UNREACHABLE(type.getPrecision());
3450 else if(type.getBasicType() == EbtInt)
3452 switch(type.getPrecision())
3454 case EbpHigh: return GL_HIGH_INT;
3455 case EbpMedium: return GL_MEDIUM_INT;
3456 case EbpLow: return GL_LOW_INT;
3458 // Should be defined as the default precision by the parser
3459 default: UNREACHABLE(type.getPrecision());
3463 // Other types (boolean, sampler) don't have a precision
3467 int OutputASM::dim(TIntermNode *v)
3469 TIntermTyped *vector = v->getAsTyped();
3470 ASSERT(vector && vector->isRegister());
3471 return vector->getNominalSize();
3474 int OutputASM::dim2(TIntermNode *m)
3476 TIntermTyped *matrix = m->getAsTyped();
3477 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
3478 return matrix->getSecondarySize();
3481 // Returns ~0u if no loop count could be determined
3482 unsigned int OutputASM::loopCount(TIntermLoop *node)
3484 // Parse loops of the form:
3485 // for(int index = initial; index [comparator] limit; index += increment)
3486 TIntermSymbol *index = 0;
3487 TOperator comparator = EOpNull;
3492 // Parse index name and intial value
3495 TIntermAggregate *init = node->getInit()->getAsAggregate();
3499 TIntermSequence &sequence = init->getSequence();
3500 TIntermTyped *variable = sequence[0]->getAsTyped();
3502 if(variable && variable->getQualifier() == EvqTemporary)
3504 TIntermBinary *assign = variable->getAsBinaryNode();
3506 if(assign->getOp() == EOpInitialize)
3508 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
3509 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
3511 if(symbol && constant)
3513 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
3516 initial = constant->getUnionArrayPointer()[0].getIConst();
3524 // Parse comparator and limit value
3525 if(index && node->getCondition())
3527 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
3528 TIntermSymbol *left = test ? test->getLeft()->getAsSymbolNode() : nullptr;
3530 if(left && (left->getId() == index->getId()))
3532 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
3536 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
3538 comparator = test->getOp();
3539 limit = constant->getUnionArrayPointer()[0].getIConst();
3546 if(index && comparator != EOpNull && node->getExpression())
3548 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
3549 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
3553 TOperator op = binaryTerminal->getOp();
3554 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
3558 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
3560 int value = constant->getUnionArrayPointer()[0].getIConst();
3564 case EOpAddAssign: increment = value; break;
3565 case EOpSubAssign: increment = -value; break;
3566 default: UNIMPLEMENTED();
3571 else if(unaryTerminal)
3573 TOperator op = unaryTerminal->getOp();
3577 case EOpPostIncrement: increment = 1; break;
3578 case EOpPostDecrement: increment = -1; break;
3579 case EOpPreIncrement: increment = 1; break;
3580 case EOpPreDecrement: increment = -1; break;
3581 default: UNIMPLEMENTED();
3586 if(index && comparator != EOpNull && increment != 0)
3588 if(comparator == EOpLessThanEqual)
3590 comparator = EOpLessThan;
3594 if(comparator == EOpLessThan)
3596 int iterations = (limit - initial) / increment;
3605 else UNIMPLEMENTED(); // Falls through
3611 bool LoopUnrollable::traverse(TIntermNode *node)
3614 loopUnrollable = true;
3616 node->traverse(this);
3618 return loopUnrollable;
3621 bool LoopUnrollable::visitLoop(Visit visit, TIntermLoop *loop)
3623 if(visit == PreVisit)
3627 else if(visit == PostVisit)
3635 bool LoopUnrollable::visitBranch(Visit visit, TIntermBranch *node)
3647 switch(node->getFlowOp())
3654 loopUnrollable = false;
3656 default: UNREACHABLE(node->getFlowOp());
3659 return loopUnrollable;
3662 bool LoopUnrollable::visitAggregate(Visit visit, TIntermAggregate *node)
3664 return loopUnrollable;