https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115147
Bug ID: 115147
Summary: exp2 with integer arguments could be translated into
ldexp
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: janschultke at googlemail dot com
Target Milestone: ---
https://godbolt.org/z/9141MdG6b
float e(int x) {
return __builtin_exp2f(x);
}
GCC (-O2) optimizes this to:
> e(int):
> pxor xmm0, xmm0
> cvtsi2ss xmm0, edi
> jmp exp2f
Clang (-O2) optimizes this to:
> .LCPI0_0:
> .long 0x3f800000
> e(int):
> movss xmm0, dword ptr [rip + .LCPI0_0]
> jmp ldexpf@PLT
I believe translating to ldexp is usually better, and this is a missed
optimization. ldexp takes an integer exponent. Converting x to float and then
going through exp2 seems wasteful in this case.
