[PATCH] Allow constant folding of fma and fmuladd
Matthew.Arsenault at amd.com
Tue Mar 4 15:32:37 PST 2014
Comment at: lib/Analysis/ConstantFolding.cpp:1640
@@ +1639,3 @@
+ default: break;
+ case Intrinsic::fma:
hfinkel at anl.gov wrote:
> I'm confused as to why you're doing this? Can't APFloat compute the FMA properly regardless of the host capabilities?
> Furthermore, if you do it this way, it looks like the answer might be wrong depending on how the compiler was configured?
I forgot APFloat had this already, and for fmuladd it's not supposed to matter. I'll fix it to use APFloat
More information about the llvm-commits