https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90510
Bug ID: 90510
Summary: [10 Regression] Unnecessary permutation
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: hjl.tools at gmail dot com
CC: rguenther at suse dot de
Target Milestone: ---
On x86-64, r271153 caused:
[hjl@gnu-cfl-1 pr54855]$ cat 5.i
typedef double __v2df __attribute__ ((__vector_size__ (16)));
__v2df
_mm_add_sd (__v2df x, __v2df y)
{
__v2df z = { (x[0] + y[0]), x[1] };
return z;
}
[hjl@gnu-cfl-1 pr54855]$
/export/build/gnu/tools-build/gcc-debug/build-x86_64-linux/gcc/xgcc
-B/export/build/gnu/tools-build/gcc-debug/build-x86_64-linux/gcc/ -O2 -S 5.i
[hjl@gnu-cfl-1 pr54855]$ cat 5.s
.file "5.i"
.text
.p2align 4
.globl _mm_add_sd
.type _mm_add_sd, @function
_mm_add_sd:
.LFB0:
.cfi_startproc
movapd %xmm0, %xmm2
addsd %xmm1, %xmm2
unpcklpd %xmm2, %xmm2 <<<< This isn't needed.
movsd %xmm2, %xmm0
ret
.cfi_endproc
.LFE0:
.size _mm_add_sd, .-_mm_add_sd
.ident "GCC: (GNU) 10.0.0 20190514 (experimental)"
.section .note.GNU-stack,"",@progbits
[hjl@gnu-cfl-1 pr54855]$
instead of
_mm_add_sd:
.LFB0:
.cfi_startproc
movapd %xmm0, %xmm2
addsd %xmm1, %xmm2
movsd %xmm2, %xmm0
ret
.cfi_endproc
