[PATCH] D60021: InstSimplify: Fold round intrinsics from sitofp

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In D60021#1448785 <https://reviews.llvm.org/D60021#1448785>, @lebedev.ri wrote:

> Do we not need any fast math flags here?
>
> https://llvm.org/docs/LangRef.html#id276
>
>   The ‘sitofp’ instruction interprets its operand as a signed integer quantity
>   and converts it to the corresponding floating-point value. If the value cannot
>   be exactly represented, it is rounded using the default rounding mode.
>


Right, *rounded*.

  #include <math.h>
  #include <limits>
  #include <cassert>
  static bool test(int x) {
    float f = x;
    float i;
    float fract = modff(f, &i);
    return fract == 0.0;
  }
  int main() {
    for(size_t i = 0; i <= std::numeric_limits<unsigned int>::max(); i++)
      assert(test(int(i)));
    return 0;
  }

agrees.

Might also want to fold

    %conv.i = sitofp i32 %conv1 to float
    %call.i = call float @modff(float %conv.i, float* nonnull %i.i) #4
    %cmp.i = fcmp oeq float %call.i, 0.000000e+00
  =>
    %cmp.i = true


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D60021/new/

https://reviews.llvm.org/D60021