================ @@ -18901,6 +18901,150 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: { CGM.getHLSLRuntime().getRadiansIntrinsic(), ArrayRef<Value *>{Op0}, nullptr, "hlsl.radians"); } + // This should only be called when targeting DXIL + case Builtin::BI__builtin_hlsl_splitdouble: { + + assert((E->getArg(0)->getType()->hasFloatingRepresentation() && + E->getArg(1)->getType()->hasUnsignedIntegerRepresentation() && + E->getArg(2)->getType()->hasUnsignedIntegerRepresentation()) && + "asuint operands types mismatch"); + Value *Op0 = EmitScalarExpr(E->getArg(0)); + const HLSLOutArgExpr *OutArg1 = dyn_cast<HLSLOutArgExpr>(E->getArg(1)); + const HLSLOutArgExpr *OutArg2 = dyn_cast<HLSLOutArgExpr>(E->getArg(2)); + + CallArgList Args; + auto [Op1BaseLValue, Op1TmpLValue] = + EmitHLSLOutArgExpr(OutArg1, Args, OutArg1->getType()); + auto [Op2BaseLValue, Op2TmpLValue] = + EmitHLSLOutArgExpr(OutArg2, Args, OutArg2->getType()); + + if (CGM.getTarget().getTriple().getArch() == llvm::Triple::dxil) { + + llvm::StructType *retType = llvm::StructType::get(Int32Ty, Int32Ty); + + if (Op0->getType()->isVectorTy()) { + auto *Op0VecTy = E->getArg(0)->getType()->getAs<VectorType>(); + + llvm::VectorType *i32VecTy = llvm::VectorType::get( + Int32Ty, ElementCount::getFixed(Op0VecTy->getNumElements())); + retType = llvm::StructType::get(i32VecTy, i32VecTy); + } + + CallInst *CI = + Builder.CreateIntrinsic(retType, Intrinsic::dx_splitdouble, {Op0}, + nullptr, "hlsl.splitdouble"); + + Value *arg0 = Builder.CreateExtractValue(CI, 0); + Value *arg1 = Builder.CreateExtractValue(CI, 1); + + Builder.CreateStore(arg0, Op1TmpLValue.getAddress()); + auto *s = Builder.CreateStore(arg1, Op2TmpLValue.getAddress()); + + EmitWritebacks(*this, Args); + return s; + } + + if (!Op0->getType()->isVectorTy()) { + FixedVectorType *destTy = FixedVectorType::get(Int32Ty, 2); + Value *bitcast = Builder.CreateBitCast(Op0, destTy); + + Value *arg0 = Builder.CreateExtractElement(bitcast, 0.0); + Value *arg1 = Builder.CreateExtractElement(bitcast, 1.0); + + Builder.CreateStore(arg0, Op1TmpLValue.getAddress()); + auto *s = Builder.CreateStore(arg1, Op2TmpLValue.getAddress()); + + EmitWritebacks(*this, Args); + return s; + } + + auto emitVectorCode = + [](Value *Op, CGBuilderTy *Builder, + FixedVectorType *DestTy) -> std::pair<Value *, Value *> { + Value *bitcast = Builder->CreateBitCast(Op, DestTy); + + SmallVector<int> lowbitsIndex; + SmallVector<int> highbitsIndex; + + for (unsigned int idx = 0; idx < DestTy->getNumElements(); idx += 2) { + lowbitsIndex.push_back(idx); + highbitsIndex.push_back(idx + 1); + } + + Value *arg0 = Builder->CreateShuffleVector(bitcast, lowbitsIndex); + Value *arg1 = Builder->CreateShuffleVector(bitcast, highbitsIndex); + + return std::make_pair(arg0, arg1); + }; + + const VectorType *targTy = E->getArg(0)->getType()->getAs<VectorType>(); ---------------- llvm-beanz wrote:
nit: ```suggestion const auto *TargTy = E->getArg(0)->getType()->getAs<VectorType>(); ``` This is one of the few patterns LLVM encourages using auto for: https://llvm.org/docs/CodingStandards.html#use-auto-type-deduction-to-make-code-more-readable https://github.com/llvm/llvm-project/pull/109331 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits