When InstCombine simplifies a load -> extract element to gep -> load, place

the new load by the old load instead of by the extract element because
a store could have occurred between the load and extract element.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78891 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp
index b875420e5beb521ef015ecfc51f3fa07f7db0552..04c225f7460a85a5a80b4e343a34e5e001ef335a 100644 (file)
--- a/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -12465,12 +12465,14 @@ Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
         unsigned AS = 
           cast<PointerType>(I->getOperand(0)->getType())->getAddressSpace();
         Value *Ptr = InsertBitCastBefore(I->getOperand(0),
-                                  PointerType::get(EI.getType(), AS),EI);
+                                  PointerType::get(EI.getType(), AS),*I);
         GetElementPtrInst *GEP =
           GetElementPtrInst::Create(Ptr, EI.getOperand(1), I->getName()+".gep");
         cast<GEPOperator>(GEP)->setIsInBounds(true);
-        InsertNewInstBefore(GEP, EI);
-        return new LoadInst(GEP);
+        InsertNewInstBefore(GEP, *I);
+        LoadInst* Load = new LoadInst(GEP, "tmp");
+        InsertNewInstBefore(Load, *I);
+        return ReplaceInstUsesWith(EI, Load);
       }
     }
     if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {

diff --git a/test/Transforms/InstCombine/vec_extract_elt2.ll b/test/Transforms/InstCombine/vec_extract_elt2.ll
new file mode 100644 (file)
index 0000000..37463d2
--- /dev/null
+++ b/test/Transforms/InstCombine/vec_extract_elt2.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | FileCheck %s
+
+; The load replacing the extract element must occur before the call
+; that may modify local array a.
+
+declare void @mod_a_func(<4 x float>* %a);
+
+; CHECK: load float* %arraydecay1, align 16
+; CHECK: call void @mod_a_func
+
+define void @cl_jpegenc_k2(<4 x float> addrspace(1)* %src, float addrspace(1)* %dst) {
+  %a = alloca [2 x <4 x float>], align 16
+  %arraydecay = getelementptr [2 x <4 x float>]* %a, i32 0, i32 0
+  %arrayidx31 = getelementptr <4 x float> addrspace(1)* %src, i32 0
+  %tmp32 = load <4 x float> addrspace(1)* %arrayidx31
+  store <4 x float> %tmp32, <4 x float>* %arraydecay, align 16
+  %tmp86 = load <4 x float>* %arraydecay, align 16
+  call void @mod_a_func(<4 x float>* %arraydecay)
+  %arrayidx132 = getelementptr float addrspace(1)* %dst, i32 0
+  %tmp236 = extractelement <4 x float> %tmp86, i32 0
+  store float %tmp236, float addrspace(1)* %arrayidx132
+  ret void
+}
\ No newline at end of file