Do not drop fast-math flags in SimplifyBinOp

[Chandler Carruth]

Do we use the context instruction to extract fast math flags elsewhere?

I wouldn't expect that to be reliably at all. Nothing *requires* the
context instruction to be the one governing the semantics.

I think we need to pass the fast math flags directly into SimplifyBinOp for
this to work.

On Thu, Jan 29, 2015 at 4:12 PM, Michael Zolotukhin <mzolotukhin at apple.com>
wrote:

> Hi,
>
> Currently InlineCost pass fails to use fast-math flags when analyzing
> profitability of inlining. Here is a small example to demonstrate that:
>
> float foo(float *a, float b) {
>   float r;
>   if (a[1] * b)
>     r = a[1]+a[2]/a[3]-a[4]*a[5]+a[6]-a[7]+a[8]*a[9]+
>             a[11]+a[12]/a[13]-a[14]*a[15]+a[16]-a[17]+a[18]*a[19]+
>             a[21]+a[22]/a[23]-a[24]*a[25]+a[26]-a[27]+a[28]*a[29]+
>             a[31]+a[32]/a[33]-a[34]*a[35]+a[36]-a[37]+a[38]*a[39]+
>             a[41]+a[42]/a[43]-a[44]*a[45]+a[46]-a[47]+a[48]*a[49]+
>             a[51]+a[52]/a[53]-a[54]*a[55]+a[56]-a[57]+a[58]*a[59]+
>             a[61]+a[62]/a[63]-a[64]*a[65]+a[66]-a[67]+a[68]*a[69]+
>             a[0]*a[0]+b;
>   else
>     r = 1;
>   return r;
> }
> float boo(float *a) {
>   return foo(a, 0.0);
> }
>
> Here inliner can’t figure out that a[1]*b can be folded to 0 if the call
> is inlined, and decides not to inline.
>
> That happens because inliner drops fast-math flags in SimplifyBinOp. The
> attached patch fixes it by looking into the instruction math flags. Does it
> look good?
>
> Without patch:
> > clang inline.c -O3 -S -Rpass=inline -ffast-math
> // Empty output
>
> With patch:
> > clang inline.c -O3 -S -Rpass=inline -ffast-math
> inline.c:17:10: remark: foo inlined into boo [-Rpass=inline]
>   return foo(a, 0.0);
>          ^
>
>
>
>
> Thanks,
> Michael
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150129/cfca50e5/attachment.html>