From: Chris Lattner <sabre@nondot.org>
Date: Fri, 5 Mar 2010 18:53:28 +0000 (+0000)
Subject: fix PR6512, a case where instcombine would incorrectly merge loads
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=4c5fb1a311a7a5e3ceb802f5b26329d30f2713b1;p=oota-llvm.git

fix PR6512, a case where instcombine would incorrectly merge loads
from different addr spaces.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97813 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Transforms/InstCombine/InstCombinePHI.cpp b/lib/Transforms/InstCombine/InstCombinePHI.cpp
index fba83542cdb..65f03936870 100644
--- a/lib/Transforms/InstCombine/InstCombinePHI.cpp
+++ b/lib/Transforms/InstCombine/InstCombinePHI.cpp
@@ -266,6 +266,7 @@ Instruction *InstCombiner::FoldPHIArgLoadIntoPHI(PHINode &PN) {
   // and if TD isn't around, we can't handle the mixed case.
   bool isVolatile = FirstLI->isVolatile();
   unsigned LoadAlignment = FirstLI->getAlignment();
+  unsigned LoadAddrSpace = FirstLI->getPointerAddressSpace();
   
   // We can't sink the load if the loaded value could be modified between the
   // load and the PHI.
@@ -290,6 +291,7 @@ Instruction *InstCombiner::FoldPHIArgLoadIntoPHI(PHINode &PN) {
     // the load and the PHI.
     if (LI->isVolatile() != isVolatile ||
         LI->getParent() != PN.getIncomingBlock(i) ||
+        LI->getPointerAddressSpace() != LoadAddrSpace ||
         !isSafeAndProfitableToSinkLoad(LI))
       return 0;
       
diff --git a/test/Transforms/InstCombine/phi.ll b/test/Transforms/InstCombine/phi.ll
index f0343e44c2b..fc321e96822 100644
--- a/test/Transforms/InstCombine/phi.ll
+++ b/test/Transforms/InstCombine/phi.ll
@@ -362,3 +362,43 @@ end:
 ; CHECK-NEXT: ret i64
 }
 
+; PR6512 - Shouldn't merge loads from different addr spaces.
+define i32 @test16(i32 addrspace(1)* %pointer1, i32 %flag, i32* %pointer2)
+nounwind {
+entry:
+  %retval = alloca i32, align 4                   ; <i32*> [#uses=2]
+  %pointer1.addr = alloca i32 addrspace(1)*, align 4 ; <i32 addrspace(1)**>
+  %flag.addr = alloca i32, align 4                ; <i32*> [#uses=2]
+  %pointer2.addr = alloca i32*, align 4           ; <i32**> [#uses=2]
+  %res = alloca i32, align 4                      ; <i32*> [#uses=4]
+  store i32 addrspace(1)* %pointer1, i32 addrspace(1)** %pointer1.addr
+  store i32 %flag, i32* %flag.addr
+  store i32* %pointer2, i32** %pointer2.addr
+  store i32 10, i32* %res
+  %tmp = load i32* %flag.addr                     ; <i32> [#uses=1]
+  %tobool = icmp ne i32 %tmp, 0                   ; <i1> [#uses=1]
+  br i1 %tobool, label %if.then, label %if.else
+
+return:                                           ; preds = %if.end
+  %tmp7 = load i32* %retval                       ; <i32> [#uses=1]
+  ret i32 %tmp7
+
+if.end:                                           ; preds = %if.else, %if.then
+  %tmp6 = load i32* %res                          ; <i32> [#uses=1]
+  store i32 %tmp6, i32* %retval
+  br label %return
+
+if.then:                                          ; preds = %entry
+  %tmp1 = load i32 addrspace(1)** %pointer1.addr  ; <i32 addrspace(1)*>
+  %arrayidx = getelementptr i32 addrspace(1)* %tmp1, i32 0 ; <i32 addrspace(1)*> [#uses=1]
+  %tmp2 = load i32 addrspace(1)* %arrayidx        ; <i32> [#uses=1]
+  store i32 %tmp2, i32* %res
+  br label %if.end
+
+if.else:                                          ; preds = %entry
+  %tmp3 = load i32** %pointer2.addr               ; <i32*> [#uses=1]
+  %arrayidx4 = getelementptr i32* %tmp3, i32 0    ; <i32*> [#uses=1]
+  %tmp5 = load i32* %arrayidx4                    ; <i32> [#uses=1]
+  store i32 %tmp5, i32* %res
+  br label %if.end
+}