[PATCH] D106058: [DAG] Fold select(cond,binop(x,y),binop(x,z)) -> binop(x,select(cond,y,z))

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: llvm/test/CodeGen/NVPTX/fast-math.ll:150
 ; CHECK: selp.f32
+; CHECK: div.rn.f32
   %x = fdiv arcp float %a, %divisor
----------------
lebedev.ri wrote:
> tra wrote:
> > tra wrote:
> > > This looks like it may be a performance regression. Judging by the name of the tests, we do expect to see reciprocal and multiplies, not `div`.
> > > 
> > Never mind. These changes are not in the final version of the patch.
> Then you were already missing some kind of an inverse transform,
> because middle-end would have already done this fold:
> https://godbolt.org/z/cb38zEj9Y
I made a kuldge fix in rG3cc38703d5ab05be0b01c31f829d19b47f183c5f so it still tests the combineRepeatedFPDivisors code path which appeared to be their purpose. But as @lebedev.ri said, we're unlikely to see the original test code patterns from real world code as InstCombine will have already broken the pattern.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106058/new/

https://reviews.llvm.org/D106058