From 881d4a05e3b5045332a78d190cd9379ed18bc20e Mon Sep 17 00:00:00 2001 From: Philip Reames Date: Fri, 6 May 2016 18:46:45 +0000 Subject: [PATCH] [GVN] Handle unordered atomics in cross block FRE You'll note there are essentially no code changes here. Cross block FRE heavily reuses code from the block local FRE. All of the tricky parts were done as part of the previous patch and the refactoring that removed the original code duplication. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268775 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Transforms/Scalar/GVN.cpp | 10 +++---- test/Transforms/GVN/atomic.ll | 55 +++++++++++++++++++++++++++++++++++ 2 files changed, 60 insertions(+), 5 deletions(-) diff --git a/lib/Transforms/Scalar/GVN.cpp b/lib/Transforms/Scalar/GVN.cpp index 116dc69b372..7e937bf954a 100644 --- a/lib/Transforms/Scalar/GVN.cpp +++ b/lib/Transforms/Scalar/GVN.cpp @@ -1598,11 +1598,6 @@ bool GVN::processNonLocalLoad(LoadInst *LI) { if (LI->getParent()->getParent()->hasFnAttribute(Attribute::SanitizeAddress)) return false; - // This code hasn't been audited for atomic, ordered, or volatile memory - // access. - if (!LI->isSimple()) - return false; - // Step 1: Find the non-local dependencies of the load. LoadDepVect Deps; MD->getNonLocalPointerDependency(LI, Deps); @@ -1669,6 +1664,11 @@ bool GVN::processNonLocalLoad(LoadInst *LI) { return true; } + // This code hasn't been audited for atomic, ordered, or volatile memory + // access. + if (!LI->isSimple()) + return false; + // Step 4: Eliminate partial redundancy. if (!EnablePRE || !EnableLoadPRE) return false; diff --git a/test/Transforms/GVN/atomic.ll b/test/Transforms/GVN/atomic.ll index 4429e9bbad0..f8a70140cc3 100644 --- a/test/Transforms/GVN/atomic.ll +++ b/test/Transforms/GVN/atomic.ll @@ -337,3 +337,58 @@ define i64 @narrow2(i32* %P1) { ret i64 %res } +; Note: The cross block FRE testing is deliberately light. All of the tricky +; bits of legality are shared code with the block-local FRE above. These +; are here only to show that we haven't obviously broken anything. + +; unordered atomic to unordered atomic +define i32 @non_local_fre(i32* %P1) { +; CHECK-LABEL: @non_local_fre( +; CHECK: load atomic i32, i32* %P1 +; CHECK: ret i32 0 +; CHECK: ret i32 0 + %a = load atomic i32, i32* %P1 unordered, align 4 + %cmp = icmp eq i32 %a, 0 + br i1 %cmp, label %early, label %next +early: + ret i32 %a +next: + %b = load atomic i32, i32* %P1 unordered, align 4 + %res = sub i32 %a, %b + ret i32 %res +} + +; unordered atomic to non-atomic +define i32 @non_local_fre2(i32* %P1) { +; CHECK-LABEL: @non_local_fre2( +; CHECK: load atomic i32, i32* %P1 +; CHECK: ret i32 0 +; CHECK: ret i32 0 + %a = load atomic i32, i32* %P1 unordered, align 4 + %cmp = icmp eq i32 %a, 0 + br i1 %cmp, label %early, label %next +early: + ret i32 %a +next: + %b = load i32, i32* %P1 + %res = sub i32 %a, %b + ret i32 %res +} + +; Can't forward ordered atomics. +define i32 @non_local_fre3(i32* %P1) { +; CHECK-LABEL: @non_local_fre3( +; CHECK: load atomic i32, i32* %P1 acquire +; CHECK: ret i32 0 +; CHECK: load atomic i32, i32* %P1 acquire +; CHECK: ret i32 %res + %a = load atomic i32, i32* %P1 acquire, align 4 + %cmp = icmp eq i32 %a, 0 + br i1 %cmp, label %early, label %next +early: + ret i32 %a +next: + %b = load atomic i32, i32* %P1 acquire, align 4 + %res = sub i32 %a, %b + ret i32 %res +} -- 2.50.1