From: Keno Fischer Date: Thu, 26 Jul 2018 21:55:06 +0000 (+0000) Subject: [SCEV] Don't expand Wrap predicate using inttoptr in ni addrspaces X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=864fbd8e9ad3e865c6d51d159b094badc6a14b76;p=platform%2Fupstream%2Fllvm.git [SCEV] Don't expand Wrap predicate using inttoptr in ni addrspaces Summary: In non-integral address spaces, we're not allowed to introduce inttoptr/ptrtoint intrinsics. Instead, we need to expand any pointer arithmetic as geps on the base pointer. Luckily this is a common task for SCEV, so all we have to do here is hook up the corresponding helper function and add test case. Fixes PR38290 Reviewers: sanjoy Differential Revision: https://reviews.llvm.org/D49832 llvm-svn: 338073 --- diff --git a/llvm/lib/Analysis/ScalarEvolutionExpander.cpp b/llvm/lib/Analysis/ScalarEvolutionExpander.cpp index b49a5cd..8f89389 100644 --- a/llvm/lib/Analysis/ScalarEvolutionExpander.cpp +++ b/llvm/lib/Analysis/ScalarEvolutionExpander.cpp @@ -2159,8 +2159,9 @@ Value *SCEVExpander::generateOverflowCheck(const SCEVAddRecExpr *AR, const SCEV *Step = AR->getStepRecurrence(SE); const SCEV *Start = AR->getStart(); + Type *ARTy = AR->getType(); unsigned SrcBits = SE.getTypeSizeInBits(ExitCount->getType()); - unsigned DstBits = SE.getTypeSizeInBits(AR->getType()); + unsigned DstBits = SE.getTypeSizeInBits(ARTy); // The expression {Start,+,Step} has nusw/nssw if // Step < 0, Start - |Step| * Backedge <= Start @@ -2172,11 +2173,12 @@ Value *SCEVExpander::generateOverflowCheck(const SCEVAddRecExpr *AR, Value *TripCountVal = expandCodeFor(ExitCount, CountTy, Loc); IntegerType *Ty = - IntegerType::get(Loc->getContext(), SE.getTypeSizeInBits(AR->getType())); + IntegerType::get(Loc->getContext(), SE.getTypeSizeInBits(ARTy)); + Type *ARExpandTy = DL.isNonIntegralPointerType(ARTy) ? ARTy : Ty; Value *StepValue = expandCodeFor(Step, Ty, Loc); Value *NegStepValue = expandCodeFor(SE.getNegativeSCEV(Step), Ty, Loc); - Value *StartValue = expandCodeFor(Start, Ty, Loc); + Value *StartValue = expandCodeFor(Start, ARExpandTy, Loc); ConstantInt *Zero = ConstantInt::get(Loc->getContext(), APInt::getNullValue(DstBits)); @@ -2199,8 +2201,18 @@ Value *SCEVExpander::generateOverflowCheck(const SCEVAddRecExpr *AR, // Compute: // Start + |Step| * Backedge < Start // Start - |Step| * Backedge > Start - Value *Add = Builder.CreateAdd(StartValue, MulV); - Value *Sub = Builder.CreateSub(StartValue, MulV); + Value *Add = nullptr, *Sub = nullptr; + if (PointerType *ARPtrTy = dyn_cast(ARExpandTy)) { + const SCEV *MulS = SE.getSCEV(MulV); + const SCEV *NegMulS = SE.getNegativeSCEV(MulS); + Add = Builder.CreateBitCast(expandAddToGEP(MulS, ARPtrTy, Ty, StartValue), + ARPtrTy); + Sub = Builder.CreateBitCast( + expandAddToGEP(NegMulS, ARPtrTy, Ty, StartValue), ARPtrTy); + } else { + Add = Builder.CreateAdd(StartValue, MulV); + Sub = Builder.CreateSub(StartValue, MulV); + } Value *EndCompareGT = Builder.CreateICmp( Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT, Sub, StartValue); diff --git a/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-ni.ll b/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-ni.ll new file mode 100644 index 0000000..8bceaa2 --- /dev/null +++ b/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-ni.ll @@ -0,0 +1,54 @@ +; RUN: opt -loop-versioning -S < %s | FileCheck %s -check-prefix=LV + +; NB: addrspaces 10-13 are non-integral +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:10:11:12:13" + +; This matches the test case from PR38290 +; Check that we expand the SCEV predicate check using GEP, rather +; than ptrtoint. + +%jl_value_t = type opaque +%jl_array_t = type { i8 addrspace(13)*, i64, i16, i16, i32 } + +declare i64 @julia_steprange_last_4949() + +define void @"japi1_align!_9477"(%jl_value_t addrspace(10)** %arg) { +; LV-LAVEL: L26.lver.check +; LV: [[OFMul:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[Step:%[^ ]*]]) +; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul]], 0 +; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul]], 1 +; LV-NEXT: [[PosGEP:%[^ ]*]] = getelementptr i32, i32 addrspace(13)* [[Base:%[^ ]*]], i64 [[Step]] +; LV-NEXT: [[NegGEP:%[^ ]*]] = getelementptr i32, i32 addrspace(13)* [[Base]], i64 [[NegStep:%[^ ]*]] +; LV-NEXT: icmp ugt i32 addrspace(13)* [[NegGEP]], [[Base]] +; LV-NEXT: icmp ult i32 addrspace(13)* [[PosGEP]], [[Base]] +; LV-NOT: inttoptr +; LV-NOT: ptrtoint +top: + %tmp = load %jl_value_t addrspace(10)*, %jl_value_t addrspace(10)** %arg, align 8 + %tmp1 = load i32, i32* inttoptr (i64 12 to i32*), align 4 + %tmp2 = sub i32 0, %tmp1 + %tmp3 = call i64 @julia_steprange_last_4949() + %tmp4 = addrspacecast %jl_value_t addrspace(10)* %tmp to %jl_value_t addrspace(11)* + %tmp5 = bitcast %jl_value_t addrspace(11)* %tmp4 to %jl_value_t addrspace(10)* addrspace(11)* + %tmp6 = load %jl_value_t addrspace(10)*, %jl_value_t addrspace(10)* addrspace(11)* %tmp5, align 8 + %tmp7 = addrspacecast %jl_value_t addrspace(10)* %tmp6 to %jl_value_t addrspace(11)* + %tmp8 = bitcast %jl_value_t addrspace(11)* %tmp7 to i32 addrspace(13)* addrspace(11)* + %tmp9 = load i32 addrspace(13)*, i32 addrspace(13)* addrspace(11)* %tmp8, align 8 + %tmp10 = sext i32 %tmp2 to i64 + br label %L26 + +L26: + %value_phi3 = phi i64 [ 0, %top ], [ %tmp11, %L26 ] + %tmp11 = add i64 %value_phi3, -1 + %tmp12 = getelementptr inbounds i32, i32 addrspace(13)* %tmp9, i64 %tmp11 + %tmp13 = load i32, i32 addrspace(13)* %tmp12, align 4 + %tmp14 = add i64 %tmp11, %tmp10 + %tmp15 = getelementptr inbounds i32, i32 addrspace(13)* %tmp9, i64 %tmp14 + store i32 %tmp13, i32 addrspace(13)* %tmp15, align 4 + %tmp16 = icmp eq i64 %value_phi3, %tmp3 + br i1 %tmp16, label %L45, label %L26 + +L45: + ret void +} + diff --git a/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-versioning.ll b/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-versioning.ll index a7e5bce..fa6fcce 100644 --- a/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-versioning.ll +++ b/llvm/test/Analysis/LoopAccessAnalysis/wrapping-pointer-versioning.ll @@ -58,10 +58,10 @@ target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" ; LV-NEXT: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE]]) ; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0 ; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1 -; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 %a2, [[OFMulResult1]] -; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 %a2, [[OFMulResult1]] -; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], %a2 -; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], %a2 +; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 [[A0:%[^ ]*]], [[OFMulResult1]] +; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 [[A0]], [[OFMulResult1]] +; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], [[A0]] +; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], [[A0]] ; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg1]], i1 [[CmpPos1]] ; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]] @@ -233,10 +233,10 @@ for.end: ; preds = %for.body ; LV: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE:%[^ ]*]]) ; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0 ; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1 -; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 %a2, [[OFMulResult1]] -; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 %a2, [[OFMulResult1]] -; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], %a2 -; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], %a2 +; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 [[A0:%[^ ]*]], [[OFMulResult1]] +; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 [[A0]], [[OFMulResult1]] +; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], [[A0]] +; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], [[A0]] ; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg1]], i1 [[CmpPos1]] ; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]