Missed optimization in math expression: simplify calls std::exp functions

[zamazan4ik]

Bugzilla Link	35594
Resolution	FIXED
Resolved on	Jan 30, 2019 22:35
Version	trunk
OS	Linux
Blocks	#34959
CC	@hfinkel,@rotateright

Extended Description

clang(trunk) with '--std=c++17 -O3 -march=native -ffast-math' flags for this code:

#include <cmath>

double f(double a, double b)
{
    return std::exp(a) * std::exp(b);
}

generates this assembly:

f(double, double): # @f(double, double)
  sub rsp, 24
  vmovsd qword ptr [rsp + 16], xmm1 # 8-byte Spill
  call exp
  vmovsd qword ptr [rsp + 8], xmm0 # 8-byte Spill
  vmovsd xmm0, qword ptr [rsp + 16] # 8-byte Reload
  call exp
  vmulsd xmm0, xmm0, qword ptr [rsp + 8] # 8-byte Folded Reload
  add rsp, 24
  ret

GCC (trunk) with '--std=c++17 -O3 -march=native -ffast-math' flags generates more optimal code:

f(double, double):
        vaddsd  xmm0, xmm0, xmm1
        jmp     __exp_finite

[hfinkel]

This could also be in InstCombine.

[zamazan4ik]

See a lot of optimizations here: https://github.com/gcc-mirror/gcc/blob/07b69d3f1cd3dd8ebb0af1fbff95914daee477d2/gcc/match.pd

[llvmbot]

https://reviews.llvm.org/D41342

[llvmbot]

Commits: rL352613, rL352730

[RKSimon]

mentioned in issue #34959