[llvm-bugs] [Bug 169270] Suboptimal code for `std::isfinite`

[LLVM Bugs via llvm-bugs]

Issue	169270
Summary	Suboptimal code for `std::isfinite`
Labels	llvm:instcombine, missed-optimization
Assignees
Reporter	Kmeakin

    [C++ code](https://godbolt.org/z/Kv6nePEjW):
```c++
auto src(float f) { return std::fabs(f) < INFINITY; }

auto tgt(float f) { return (f - f) == (f - f); }
```

AArch64 assembly:
```asm
src(float):
        fmov    w9, s0
 mov     w8, #2139095040
        and     w9, w9, #0x7fffffff
        cmp w9, w8
        cset    w0, lt
        ret

tgt(float):
        fsub s0, s0, s0
        fcmp    s0, s0
        cset    w0, vc
 ret
```

x86_64 assembly:
```asm
src(float):
        movd    eax, xmm0
        and     eax, 2147483647
        cmp     eax, 2139095040
 setl    al
        ret

tgt(float):
        subss   xmm0, xmm0
 ucomiss xmm0, xmm0
        setnp   al
        ret
```

[LLVM IR](https://alive2.llvm.org/ce/z/SqJNMr):
```llvm
define dso_local noundef i1 @src(float)(float noundef %0) local_unnamed_addr #0 {
  %2 = tail call float @llvm.fabs.f32(float %0)
  %3 = fcmp one float %2, 0x7FF0000000000000
  ret i1 %3
}

define dso_local noundef i1 @tgt(float)(float noundef %0) local_unnamed_addr #0 {
  %2 = fsub float %0, %0
  %3 = fcmp ord float %2, 0.000000e+00
  ret i1 %3
}
```

This works because `inf - inf` gives `NaN` and `NaN != NaN`

_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs