x86_64 code generation defects when SSE instructions are disabled

Hi all,

Recently, I discovered a defect in GCC’s code generation for x86_64 targets when SSE is disabled. Upon testing with LLVM 3.2, I found that it also has the same issues so I’m linking a GCC bug report I made. The proposed solution for the second issue listed there also applies to LLVM:

As I noted in the bug report, I’m not sure how a fix should be handled/whether it should be handled simply because I believe such fixes would not necessarily adhere to the ABI specification.

In any case, I just wanted to throw the issue this way.


Do you know if this is still a problem for recent versions of LLVM ?