https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122662
--- Comment #4 from Mohamed <mohamed.selim at dxc dot com> ---
gcc-12 aarch64 outputs:
.arch armv8-a
.file "main.cpp"
.text
.align 2
.p2align 4,,11
.global test::is_std_nan(float)
.type test::is_std_nan(float), %function
test::is_std_nan(float):
.LFB2710:
fcmp s0, s0
cset w0, vs
ret
.LFE2710:
.size test::is_std_nan(float), .-test::is_std_nan(float)
.align 2
.p2align 4,,11
.global test::is_std_nan(double)
.type test::is_std_nan(double), %function
test::is_std_nan(double):
.LFB2711:
fcmp d0, d0
cset w0, vs
ret
.LFE2711:
One thing that was noticed also in gcc-12 generated assembly(above) built on
arm-v8, that the binary generated is triggering FE_INVALID, but due to
something else.
I believe it boils down to the fact that compare inst "fcmp d0, d0" or
"fcmp s0, s0" which when it compares nan with nan it ends up triggering
FE_INVALID.
Yet, also with clang using the flag "-ffp-exception-behavior=strict", generates
on arm-v8
test::is_std_nan(double): // @test::is_std_nan(double)
// %bb.0:
fmov x9, d0
mov x8, #9218868437227405312 // =0x7ff0000000000000
and x9, x9, #0x7fffffffffffffff
cmp x9, x8
cset w0, gt
ret
Which doesn't trigger FE_INVALID.
