OLD | NEW |
1 //===- subzero/src/IceTargetLoweringX86BaseImpl.h - x86 lowering -*- C++ -*-==// | 1 //===- subzero/src/IceTargetLoweringX86BaseImpl.h - x86 lowering -*- C++ -*-==// |
2 // | 2 // |
3 // The Subzero Code Generator | 3 // The Subzero Code Generator |
4 // | 4 // |
5 // This file is distributed under the University of Illinois Open Source | 5 // This file is distributed under the University of Illinois Open Source |
6 // License. See LICENSE.TXT for details. | 6 // License. See LICENSE.TXT for details. |
7 // | 7 // |
8 //===----------------------------------------------------------------------===// | 8 //===----------------------------------------------------------------------===// |
9 /// | 9 /// |
10 /// \file | 10 /// \file |
(...skipping 6043 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
6054 SearchSpanStack.emplace(Span.Begin, Span.Size / 2, Label); | 6054 SearchSpanStack.emplace(Span.Begin, Span.Size / 2, Label); |
6055 SearchSpanStack.emplace(PivotIndex, Span.Size - (Span.Size / 2), nullptr); | 6055 SearchSpanStack.emplace(PivotIndex, Span.Size - (Span.Size / 2), nullptr); |
6056 DoneCmp = true; | 6056 DoneCmp = true; |
6057 break; | 6057 break; |
6058 } | 6058 } |
6059 } | 6059 } |
6060 | 6060 |
6061 _br(DefaultTarget); | 6061 _br(DefaultTarget); |
6062 } | 6062 } |
6063 | 6063 |
6064 template <typename TraitsType> | |
6065 void TargetX86Base<TraitsType>::scalarizeArithmetic(InstArithmetic::OpKind Kind, | |
6066 Variable *Dest, | |
6067 Operand *Src0, | |
6068 Operand *Src1) { | |
6069 assert(isVectorType(Dest->getType())); | |
6070 Type Ty = Dest->getType(); | |
6071 Type ElementTy = typeElementType(Ty); | |
6072 SizeT NumElements = typeNumElements(Ty); | |
6073 | |
6074 Operand *T = Ctx->getConstantUndef(Ty); | |
6075 for (SizeT I = 0; I < NumElements; ++I) { | |
6076 Constant *Index = Ctx->getConstantInt32(I); | |
6077 | |
6078 // Extract the next two inputs. | |
6079 Variable *Op0 = Func->makeVariable(ElementTy); | |
6080 Context.insert<InstExtractElement>(Op0, Src0, Index); | |
6081 Variable *Op1 = Func->makeVariable(ElementTy); | |
6082 Context.insert<InstExtractElement>(Op1, Src1, Index); | |
6083 | |
6084 // Perform the arithmetic as a scalar operation. | |
6085 Variable *Res = Func->makeVariable(ElementTy); | |
6086 auto *Arith = Context.insert<InstArithmetic>(Kind, Res, Op0, Op1); | |
6087 // We might have created an operation that needed a helper call. | |
6088 genTargetHelperCallFor(Arith); | |
6089 | |
6090 // Insert the result into position. | |
6091 Variable *DestT = Func->makeVariable(Ty); | |
6092 Context.insert<InstInsertElement>(DestT, T, Res, Index); | |
6093 T = DestT; | |
6094 } | |
6095 | |
6096 Context.insert<InstAssign>(Dest, T); | |
6097 } | |
6098 | |
6099 /// The following pattern occurs often in lowered C and C++ code: | 6064 /// The following pattern occurs often in lowered C and C++ code: |
6100 /// | 6065 /// |
6101 /// %cmp = fcmp/icmp pred <n x ty> %src0, %src1 | 6066 /// %cmp = fcmp/icmp pred <n x ty> %src0, %src1 |
6102 /// %cmp.ext = sext <n x i1> %cmp to <n x ty> | 6067 /// %cmp.ext = sext <n x i1> %cmp to <n x ty> |
6103 /// | 6068 /// |
6104 /// We can eliminate the sext operation by copying the result of pcmpeqd, | 6069 /// We can eliminate the sext operation by copying the result of pcmpeqd, |
6105 /// pcmpgtd, or cmpps (which produce sign extended results) to the result of the | 6070 /// pcmpgtd, or cmpps (which produce sign extended results) to the result of the |
6106 /// sext operation. | 6071 /// sext operation. |
6107 template <typename TraitsType> | 6072 template <typename TraitsType> |
6108 void TargetX86Base<TraitsType>::eliminateNextVectorSextInstruction( | 6073 void TargetX86Base<TraitsType>::eliminateNextVectorSextInstruction( |
(...skipping 1344 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
7453 emitGlobal(*Var, SectionSuffix); | 7418 emitGlobal(*Var, SectionSuffix); |
7454 } | 7419 } |
7455 } | 7420 } |
7456 } break; | 7421 } break; |
7457 } | 7422 } |
7458 } | 7423 } |
7459 } // end of namespace X86NAMESPACE | 7424 } // end of namespace X86NAMESPACE |
7460 } // end of namespace Ice | 7425 } // end of namespace Ice |
7461 | 7426 |
7462 #endif // SUBZERO_SRC_ICETARGETLOWERINGX86BASEIMPL_H | 7427 #endif // SUBZERO_SRC_ICETARGETLOWERINGX86BASEIMPL_H |
OLD | NEW |