https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107546
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|simd, redundant pcmpeqb and |[10/11/12/13 Regression]
|pxor |simd, redundant pcmpeqb and
| |pxor
Target Milestone|--- |10.5
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
For GNU C++ vectors produced
GCC 4.8 until GCC 11 produced:
movdqa xmm0, XMMWORD PTR [rdi]
pcmpeqd xmm1, xmm1
pcmpgtb xmm0, XMMWORD PTR .LC0[rip]
pandn xmm0, xmm1
ret
GCC 11+ produces:
movdqa xmm0, XMMWORD PTR [rdi]
pxor xmm1, xmm1
pcmpgtb xmm0, XMMWORD PTR .LC0[rip]
pcmpeqb xmm0, xmm1
ret
But the intrinics produced the expected thing until GCC 9.
in GCC 8 the intrinsics produces:
_3 = VEC_COND_EXPR <_4 < { 48, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48, 48,
48, 48, 48, 48 }, { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
-1 }, { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }>;
even. Notice the < vs <= there.
I suspect the <= expansion part of the x86_64 backend needs to be fixed up to
produce better code.
So this is a regression for the intrinsics and marking it as such.
