The inst combining of inttoptr into GEP with one index was using the bit size of
the type instead of the byte size. This was causing troublesome mis-compilations.
True to form, this took 2 days to find and is a one-line fix. :-P
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@48354 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp
index ec51508..9b423a4 100644
--- a/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -7416,7 +7416,7 @@
// is a single-index GEP.
if (X->getType() == CI.getType()) {
// Get the size of the pointee type.
- uint64_t Size = TD->getABITypeSizeInBits(DestPointee);
+ uint64_t Size = TD->getABITypeSize(DestPointee);
// Convert the constant to intptr type.
APInt Offset = Cst->getValue();
diff --git a/test/Transforms/InstCombine/2008-03-13-IntToPtr.ll b/test/Transforms/InstCombine/2008-03-13-IntToPtr.ll
new file mode 100644
index 0000000..31e7e35
--- /dev/null
+++ b/test/Transforms/InstCombine/2008-03-13-IntToPtr.ll
@@ -0,0 +1,9 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep {16} | count 1
+
+define i8* @bork(i8** %qux) {
+ %tmp275 = load i8** %qux, align 1
+ %tmp275276 = ptrtoint i8* %tmp275 to i32
+ %tmp277 = add i32 %tmp275276, 16
+ %tmp277278 = inttoptr i32 %tmp277 to i8*
+ ret i8* %tmp277278
+}