[PATCH] D80542: [DAGCombiner] Require ninf for division estimation

[Qing Shan Zhang via Phabricator via llvm-commits]

steven.zhang added a comment.

I think, this is the right fix. Notice that, even with this fix, we still have issue when the input is denormal floating point. We might have to add some test before doing the software expansion as what we did for sqrt. But this check may hurt the performance.

> cat a.C

  #include <stdio.h>
  double __attribute__((noinline)) foo(double a, double b) {
  	return a/b;
  }
  
  int main() {
  	printf("%g\n", foo(1.0, 2.2250738585072009e-309));
  	return 0;
  }

> clang a.C -Ofast;./a.out

nan


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80542/new/

https://reviews.llvm.org/D80542