llvm-gcc optimizer generates movaps instruction on non-aligned memory

[llvmbot]

Bugzilla Link	11330
Resolution	FIXED
Resolved on	Nov 08, 2011 10:41
Version	trunk
OS	MacOS X
Attachments	minimal source to repro + xcodeproj
Reporter	LLVM Bugzilla Contributor
CC	@efriedma-quic

Extended Description

Under certain conditions, llvm-gcc -O2 will generate a movaps instruction on an unaligned memory address. These are always a result of optimizing __builtin___memcpy_chk() emitted from memcpy() calls.

An Xcode project with minimal source to reproduce is attached.

clang 3.0 does not have this bug. I'm posting this bug so that you can (1) admire your progress and/or (2) see if this affects dragonegg.

[llvmbot]

This sounds like another instance of pr10514. What version of llvm-gcc are you using?

[llvmbot]

This sounds like another instance of pr10514. What version of llvm-gcc are you
using?

./xcode-4.1-4b110/usr/bin/llvm-gcc --version
i686-apple-darwin11-llvm-gcc-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)

I'll see if I can rustle up a newer version of llvm-gcc to see if http://llvm.org/pr10514 fixes it.

[llvmbot]

That's an old version of llvm-gcc, and even if it was from trunk, we're no longer interested in llvm-gcc bugs. Please move to clang or dragonegg.