[PATCH] Branch folding causes different code generation at "-O2 -g" and "-O2"
chandlerc at google.com
Wed Mar 5 23:23:46 PST 2014
On Wed, Mar 5, 2014 at 5:36 PM, Robinson, Paul <
Paul_Robinson at playstation.sony.com> wrote:
> I skimmed CodeGen/X86 for tests containing 'dbg.value' (there are about
> two dozen) and none of them looked like they were verifying codegen. Some
> looked like they were making assumptions about codegen but were more
> interested in verifying correct debug location descriptions (e.g. a stack
> variable used the right base register, or that a range terminated at the
> point where a register should have been clobbered).
Nope, seems there isn't anything yet. I was just asking if there was, to
combine them. The answer of "nope, this is the first for x86" works great.
> There is *nothing* set up to generate code with and without debug info, to
> make sure they are the same.
While trust me, I agree with you that this is an area in need of much work,
I wasn't imagining a pattern which doesn't exist at all, I was thinking of
one that I touched quite recently in
test/Transforms/Inline/ignore-debug-info.ll. I wasn't sure if there was
something along these lines in the X86 tree, but seems not. =/
> If we produce two fixes along these lines and it makes sense to put them
> in the same file, sure we should try to put them in the same file.
Yea, I expect this will not be the last. =/
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits