From 6e5b564f917786b9967b2a0e5a8cb2c898dac03e Mon Sep 17 00:00:00 2001 From: Luqman Aden Date: Sat, 20 Apr 2019 13:14:18 +0000 Subject: [PATCH] [CorrelatedValuePropagation] Mark subs that we know not to wrap with nuw/nsw. Summary: Teach CorrelatedValuePropagation to also handle sub instructions in addition to add. Relatively simple since makeGuaranteedNoWrapRegion already understood sub instructions. Only subtle change is which range is passed as "Other" to that function, since sub isn't commutative. Note that CorrelatedValuePropagation::processAddSub is still hidden behind a default-off flag as IndVarSimplify hasn't yet been fixed to strip the added nsw/nuw flags and causes a miscompile. (PR31181) Reviewers: sanjoy, apilipenko, nikic Reviewed By: nikic Subscribers: hiraditya, jfb, jdoerfert, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D60036 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358816 91177308-0d34-0410-b5e6-96231b3b80d8 --- .../Scalar/CorrelatedValuePropagation.cpp | 51 +++++++++++----------- test/Transforms/CorrelatedValuePropagation/add.ll | 2 +- test/Transforms/CorrelatedValuePropagation/sub.ll | 36 +++++++-------- 3 files changed, 45 insertions(+), 44 deletions(-) diff --git a/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp b/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp index a56a37a618b..2c31e4aa6cd 100644 --- a/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp +++ b/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp @@ -64,7 +64,7 @@ STATISTIC(NumAShrs, "Number of ashr converted to lshr"); STATISTIC(NumSRems, "Number of srem converted to urem"); STATISTIC(NumOverflows, "Number of overflow checks removed"); -static cl::opt DontProcessAdds("cvp-dont-process-adds", cl::init(true)); +static cl::opt DontAddNoWrapFlags("cvp-dont-add-nowrap-flags", cl::init(true)); namespace { @@ -607,53 +607,53 @@ static bool processAShr(BinaryOperator *SDI, LazyValueInfo *LVI) { return true; } -static bool processAdd(BinaryOperator *AddOp, LazyValueInfo *LVI) { +static bool processBinOp(BinaryOperator *BinOp, LazyValueInfo *LVI) { using OBO = OverflowingBinaryOperator; - if (DontProcessAdds) + if (DontAddNoWrapFlags) return false; - if (AddOp->getType()->isVectorTy()) + if (BinOp->getType()->isVectorTy()) return false; - bool NSW = AddOp->hasNoSignedWrap(); - bool NUW = AddOp->hasNoUnsignedWrap(); + bool NSW = BinOp->hasNoSignedWrap(); + bool NUW = BinOp->hasNoUnsignedWrap(); if (NSW && NUW) return false; - BasicBlock *BB = AddOp->getParent(); + BasicBlock *BB = BinOp->getParent(); - Value *LHS = AddOp->getOperand(0); - Value *RHS = AddOp->getOperand(1); + Value *LHS = BinOp->getOperand(0); + Value *RHS = BinOp->getOperand(1); - ConstantRange LRange = LVI->getConstantRange(LHS, BB, AddOp); + ConstantRange RRange = LVI->getConstantRange(RHS, BB, BinOp); - // Initialize RRange only if we need it. If we know that guaranteed no wrap - // range for the given LHS range is empty don't spend time calculating the - // range for the RHS. - Optional RRange; - auto LazyRRange = [&] () { - if (!RRange) - RRange = LVI->getConstantRange(RHS, BB, AddOp); - return RRange.getValue(); + // Initialize LRange only if we need it. If we know that guaranteed no wrap + // range for the given RHS range is empty don't spend time calculating the + // range for the LHS. + Optional LRange; + auto LazyLRange = [&] () { + if (!LRange) + LRange = LVI->getConstantRange(LHS, BB, BinOp); + return LRange.getValue(); }; bool Changed = false; if (!NUW) { ConstantRange NUWRange = ConstantRange::makeGuaranteedNoWrapRegion( - BinaryOperator::Add, LRange, OBO::NoUnsignedWrap); + BinOp->getOpcode(), RRange, OBO::NoUnsignedWrap); if (!NUWRange.isEmptySet()) { - bool NewNUW = NUWRange.contains(LazyRRange()); - AddOp->setHasNoUnsignedWrap(NewNUW); + bool NewNUW = NUWRange.contains(LazyLRange()); + BinOp->setHasNoUnsignedWrap(NewNUW); Changed |= NewNUW; } } if (!NSW) { ConstantRange NSWRange = ConstantRange::makeGuaranteedNoWrapRegion( - BinaryOperator::Add, LRange, OBO::NoSignedWrap); + BinOp->getOpcode(), RRange, OBO::NoSignedWrap); if (!NSWRange.isEmptySet()) { - bool NewNSW = NSWRange.contains(LazyRRange()); - AddOp->setHasNoSignedWrap(NewNSW); + bool NewNSW = NSWRange.contains(LazyLRange()); + BinOp->setHasNoSignedWrap(NewNSW); Changed |= NewNSW; } } @@ -729,7 +729,8 @@ static bool runImpl(Function &F, LazyValueInfo *LVI, DominatorTree *DT, BBChanged |= processAShr(cast(II), LVI); break; case Instruction::Add: - BBChanged |= processAdd(cast(II), LVI); + case Instruction::Sub: + BBChanged |= processBinOp(cast(II), LVI); break; } } diff --git a/test/Transforms/CorrelatedValuePropagation/add.ll b/test/Transforms/CorrelatedValuePropagation/add.ll index 4001f511f94..026af75220d 100644 --- a/test/Transforms/CorrelatedValuePropagation/add.ll +++ b/test/Transforms/CorrelatedValuePropagation/add.ll @@ -1,4 +1,4 @@ -; RUN: opt < %s -correlated-propagation -cvp-dont-process-adds=false -S | FileCheck %s +; RUN: opt < %s -correlated-propagation -cvp-dont-add-nowrap-flags=false -S | FileCheck %s ; CHECK-LABEL: @test0( define void @test0(i32 %a) { diff --git a/test/Transforms/CorrelatedValuePropagation/sub.ll b/test/Transforms/CorrelatedValuePropagation/sub.ll index 83f6ff15d1a..6e954e8ff67 100644 --- a/test/Transforms/CorrelatedValuePropagation/sub.ll +++ b/test/Transforms/CorrelatedValuePropagation/sub.ll @@ -1,5 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py -; RUN: opt < %s -correlated-propagation -cvp-dont-process-adds=false -S | FileCheck %s +; RUN: opt < %s -correlated-propagation -cvp-dont-add-nowrap-flags=false -S | FileCheck %s define void @test0(i32 %a) { ; CHECK-LABEL: @test0( @@ -7,7 +7,7 @@ define void @test0(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A:%.*]], 100 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -30,7 +30,7 @@ define void @test1(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[A:%.*]], 100 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -76,7 +76,7 @@ define void @test3(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A:%.*]], -1 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -99,7 +99,7 @@ define void @test4(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[A:%.*]], 2147483647 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -163,7 +163,7 @@ define void @test7(i32 %a, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = and i1 [[CMP_1]], [[FLAG:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -191,7 +191,7 @@ define void @test8(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = and i1 [[CMP_1]], [[CMP_2]] ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -250,7 +250,7 @@ define void @test9(i32 %a, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = and i1 [[CMP_1]], [[CMP_3]] ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -279,7 +279,7 @@ define void @test10(i32 %a, i32 %b, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = and i1 [[CMP_1]], [[FLAG:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -307,7 +307,7 @@ define i32 @test11(i32* %p, i32 %i) { ; CHECK-NEXT: [[WITHIN:%.*]] = and i1 [[WITHIN_1]], [[WITHIN_2]] ; CHECK-NEXT: br i1 [[WITHIN]], label [[THEN:%.*]], label [[ELSE:%.*]] ; CHECK: then: -; CHECK-NEXT: [[I_MINUS_6:%.*]] = sub i32 [[I]], 6 +; CHECK-NEXT: [[I_MINUS_6:%.*]] = sub nuw nsw i32 [[I]], 6 ; CHECK-NEXT: ret i32 [[I_MINUS_6]] ; CHECK: else: ; CHECK-NEXT: ret i32 0 @@ -336,7 +336,7 @@ define void @test12(i32 %a, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = or i1 [[CMP_1]], [[FLAG:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[BB:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -364,7 +364,7 @@ define void @test13(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = or i1 [[CMP_1]], [[CMP_2]] ; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[BB:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -423,7 +423,7 @@ define void @test14(i32 %a, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = or i1 [[CMP_1]], [[CMP_3]] ; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[BB:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -452,7 +452,7 @@ define void @test15(i32 %a, i32 %b, i1 %flag) { ; CHECK-NEXT: [[CMP:%.*]] = or i1 [[CMP_1]], [[FLAG:%.*]] ; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[BB:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[A]], 1 +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 [[A]], 1 ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -483,7 +483,7 @@ define i32 @test16(i32* %n, i32* %a) { ; CHECK-NEXT: [[X:%.*]] = load atomic i32, i32* [[A:%.*]] unordered, align 8 ; CHECK-NEXT: fence acquire ; CHECK-NEXT: [[ACC_CURR]] = sub i32 [[ACC]], [[X]] -; CHECK-NEXT: [[IV_NEXT]] = sub i32 [[IV]], -1 +; CHECK-NEXT: [[IV_NEXT]] = sub nsw i32 [[IV]], -1 ; CHECK-NEXT: [[NVAL:%.*]] = load atomic i32, i32* [[N:%.*]] unordered, align 8 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[IV_NEXT]], [[NVAL]] ; CHECK-NEXT: br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]] @@ -514,7 +514,7 @@ define void @test17(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A:%.*]], 100 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 1, [[A]] +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 1, [[A]] ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -537,7 +537,7 @@ define void @test18(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A:%.*]], 10000 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 -2, [[A]] +; CHECK-NEXT: [[SUB:%.*]] = sub nuw i32 -2, [[A]] ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void @@ -560,7 +560,7 @@ define void @test19(i32 %a) { ; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[A:%.*]], 100 ; CHECK-NEXT: br i1 [[CMP]], label [[BB:%.*]], label [[EXIT:%.*]] ; CHECK: bb: -; CHECK-NEXT: [[SUB:%.*]] = sub i32 -1, [[A]] +; CHECK-NEXT: [[SUB:%.*]] = sub nuw nsw i32 -1, [[A]] ; CHECK-NEXT: br label [[EXIT]] ; CHECK: exit: ; CHECK-NEXT: ret void -- 2.11.0