From 13623ad009ba13f7179900a7cfd5370eb85f692f Mon Sep 17 00:00:00 2001
From: Sanjoy Das <sanjoy@playingwithpointers.com>
Date: Thu, 14 Jul 2016 19:21:15 +0000
Subject: [PATCH] [JumpThreading] PRE unordered loads

Summary: Extend JumpThreading's PRE to unordered atomic loads.

Reviewers: hfinkel, reames

Subscribers: mcrosier, llvm-commits

Differential Revision: http://reviews.llvm.org/D22326

llvm-svn: 275456
---
 llvm/lib/Transforms/Scalar/JumpThreading.cpp       |  11 +-
 llvm/test/Transforms/JumpThreading/thread-loads.ll | 139 +++++++++++++++++++++
 2 files changed, 145 insertions(+), 5 deletions(-)

diff --git a/llvm/lib/Transforms/Scalar/JumpThreading.cpp b/llvm/lib/Transforms/Scalar/JumpThreading.cpp
index de1d7d5..b9e717c 100644
--- a/llvm/lib/Transforms/Scalar/JumpThreading.cpp
+++ b/llvm/lib/Transforms/Scalar/JumpThreading.cpp
@@ -924,8 +924,8 @@ bool JumpThreadingPass::ProcessImpliedCondition(BasicBlock *BB) {
 /// important optimization that encourages jump threading, and needs to be run
 /// interlaced with other jump threading tasks.
 bool JumpThreadingPass::SimplifyPartiallyRedundantLoad(LoadInst *LI) {
-  // Don't hack volatile/atomic loads.
-  if (!LI->isSimple()) return false;
+  // Don't hack volatile and ordered loads.
+  if (!LI->isUnordered()) return false;
 
   // If the load is defined in a block with exactly one predecessor, it can't be
   // partially redundant.
@@ -1055,9 +1055,10 @@ bool JumpThreadingPass::SimplifyPartiallyRedundantLoad(LoadInst *LI) {
   if (UnavailablePred) {
     assert(UnavailablePred->getTerminator()->getNumSuccessors() == 1 &&
            "Can't handle critical edge here!");
-    LoadInst *NewVal = new LoadInst(LoadedPtr, LI->getName()+".pr", false,
-                                 LI->getAlignment(),
-                                 UnavailablePred->getTerminator());
+    LoadInst *NewVal =
+        new LoadInst(LoadedPtr, LI->getName() + ".pr", false,
+                     LI->getAlignment(), LI->getOrdering(), LI->getSynchScope(),
+                     UnavailablePred->getTerminator());
     NewVal->setDebugLoc(LI->getDebugLoc());
     if (AATags)
       NewVal->setAAMetadata(AATags);
diff --git a/llvm/test/Transforms/JumpThreading/thread-loads.ll b/llvm/test/Transforms/JumpThreading/thread-loads.ll
index 0981e19..4b482cb 100644
--- a/llvm/test/Transforms/JumpThreading/thread-loads.ll
+++ b/llvm/test/Transforms/JumpThreading/thread-loads.ll
@@ -107,6 +107,145 @@ return:
   ret i32 13
 }
 
+define i32 @test4(i32* %P) {
+; CHECK-LABEL: @test4(
+entry:
+  %v0 = tail call i32 (...) @f1()
+  %v1 = icmp eq i32 %v0, 0
+  br i1 %v1, label %bb1, label %bb
+
+bb:
+; CHECK: bb1.thread:
+; CHECK: store atomic
+; CHECK: br label %bb3
+  store atomic i32 42, i32* %P unordered, align 4
+  br label %bb1
+
+bb1:
+; CHECK: bb1:
+; CHECK-NOT: phi
+; CHECK: load atomic
+  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+  %v2 = load atomic i32, i32* %P unordered, align 4
+  %v3 = icmp sgt i32 %v2, 36
+  br i1 %v3, label %bb3, label %bb2
+
+bb2:
+  %v4 = tail call i32 (...) @f2()
+  ret i32 %res.0
+
+bb3:
+  ret i32 %res.0
+}
+
+define i32 @test5(i32* %P) {
+; Negative test
+
+; CHECK-LABEL: @test5(
+entry:
+  %v0 = tail call i32 (...) @f1()
+  %v1 = icmp eq i32 %v0, 0
+  br i1 %v1, label %bb1, label %bb
+
+bb:
+; CHECK: bb:
+; CHECK-NEXT:   store atomic i32 42, i32* %P release, align 4
+; CHECK-NEXT:   br label %bb1
+  store atomic i32 42, i32* %P release, align 4
+  br label %bb1
+
+bb1:
+; CHECK: bb1:
+; CHECK-NEXT:  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+; CHECK-NEXT:  %v2 = load atomic i32, i32* %P acquire, align 4
+; CHECK-NEXT:  %v3 = icmp sgt i32 %v2, 36
+; CHECK-NEXT:  br i1 %v3, label %bb3, label %bb2
+
+  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+  %v2 = load atomic i32, i32* %P acquire, align 4
+  %v3 = icmp sgt i32 %v2, 36
+  br i1 %v3, label %bb3, label %bb2
+
+bb2:
+  %v4 = tail call i32 (...) @f2()
+  ret i32 %res.0
+
+bb3:
+  ret i32 %res.0
+}
+
+define i32 @test6(i32* %P) {
+; Negative test
+
+; CHECK-LABEL: @test6(
+entry:
+  %v0 = tail call i32 (...) @f1()
+  %v1 = icmp eq i32 %v0, 0
+  br i1 %v1, label %bb1, label %bb
+
+bb:
+; CHECK: bb:
+; CHECK-NEXT:   store i32 42, i32* %P
+; CHECK-NEXT:   br label %bb1
+  store i32 42, i32* %P
+  br label %bb1
+
+bb1:
+; CHECK: bb1:
+; CHECK-NEXT:  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+; CHECK-NEXT:  %v2 = load atomic i32, i32* %P acquire, align 4
+; CHECK-NEXT:  %v3 = icmp sgt i32 %v2, 36
+; CHECK-NEXT:  br i1 %v3, label %bb3, label %bb2
+
+  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+  %v2 = load atomic i32, i32* %P acquire, align 4
+  %v3 = icmp sgt i32 %v2, 36
+  br i1 %v3, label %bb3, label %bb2
+
+bb2:
+  %v4 = tail call i32 (...) @f2()
+  ret i32 %res.0
+
+bb3:
+  ret i32 %res.0
+}
+
+define i32 @test7(i32* %P) {
+; Negative test
+
+; CHECK-LABEL: @test7(
+entry:
+  %v0 = tail call i32 (...) @f1()
+  %v1 = icmp eq i32 %v0, 0
+  br i1 %v1, label %bb1, label %bb
+
+bb:
+; CHECK: bb:
+; CHECK-NEXT:   %val = load i32, i32* %P
+; CHECK-NEXT:   br label %bb1
+  %val = load i32, i32* %P
+  br label %bb1
+
+bb1:
+; CHECK: bb1:
+; CHECK-NEXT:  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+; CHECK-NEXT:  %v2 = load atomic i32, i32* %P acquire, align 4
+; CHECK-NEXT:  %v3 = icmp sgt i32 %v2, 36
+; CHECK-NEXT:  br i1 %v3, label %bb3, label %bb2
+
+  %res.0 = phi i32 [ 1, %bb ], [ 0, %entry ]
+  %v2 = load atomic i32, i32* %P acquire, align 4
+  %v3 = icmp sgt i32 %v2, 36
+  br i1 %v3, label %bb3, label %bb2
+
+bb2:
+  %v4 = tail call i32 (...) @f2()
+  ret i32 %res.0
+
+bb3:
+  ret i32 %res.0
+}
+
 !0 = !{!3, !3, i64 0}
 !1 = !{!"omnipotent char", !2}
 !2 = !{!"Simple C/C++ TBAA", null}
-- 
2.7.4