https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127885
Bug ID: 127885
Summary: r17-4622-gbbf9808d787eac disabled vectorizer
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: hjl.tools at gmail dot com
CC: rguenther at suse dot de
Target Milestone: ---
Target: x86-64
Created attachment 65833
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65833&action=edit
A testcase
When compiled with -O2 -march=x86-64-v4, r17-4622-gbbf9808d787eac disabled
vectorizer. Before:
.globl _Z3foov
.type _Z3foov, @function
_Z3foov:
.LFB0:
.cfi_startproc
movq foo_ptr(%rip), %rcx
vpcmpeqd %xmm3, %xmm3, %xmm3
movabsq $47244640257, %rdx
vpsrld $31, %xmm3, %xmm3
vmovdqu (%rcx), %ymm2
vmovdqu (%rcx), %xmm1
movq 24(%rcx), %rax
vpermq $0, %ymm2, %ymm0
vpermilpd $0, %xmm1, %xmm1
vmovd %xmm2, x_param_align_loop_iterations(%rip)
vpsrlvq .LC0(%rip), %ymm0, %ymm2
movl %eax, x_param_phiopt_factor_max_stmts_live(%rip)
vpsrlvq .LC2(%rip), %xmm1, %xmm1
vpsrlvq .LC4(%rip), %ymm0, %ymm0
vpmovqd %ymm2, %xmm2
vpandd %xmm3, %xmm2, %xmm2
vpmovqd %xmm1, %xmm1
vpmovqd %ymm0, %xmm0
vmovdqu %xmm2, foo_opts_set(%rip)
vmovq %rdx, %xmm2
vpandd %xmm3, %xmm0, %xmm0
vmovdqu %xmm0, foo_opts_set+28(%rip)
vpand %xmm1, %xmm2, %xmm1
vmovq %xmm1, foo_opts_set+16(%rip)
vzeroupper
ret
.cfi_endproc
.LFE0:
.size _Z3foov, .-_Z3foov
.section .rodata.cst32,"aM",@progbits,32
.align 32
.LC0:
.quad 1
.quad 1
.quad 1
.quad 2
.section .rodata.cst16,"aM",@progbits,16
.align 16
.LC2:
.quad 2
.quad 3
.section .rodata.cst32
.align 32
.LC4:
.quad 3
.quad 4
.quad 5
.quad 6
After:
_Z3foov:
.LFB0:
.cfi_startproc
movq foo_ptr(%rip), %rcx
movq (%rcx), %rax
movq %rax, %rdx
movl %eax, x_param_align_loop_iterations(%rip)
shrq $1, %rdx
andl $1, %edx
movl %edx, foo_opts_set+8(%rip)
movl %edx, foo_opts_set+4(%rip)
movl %edx, foo_opts_set(%rip)
movq %rax, %rdx
shrq $2, %rdx
andl $1, %edx
movl %edx, foo_opts_set+16(%rip)
movl %edx, foo_opts_set+12(%rip)
movq %rax, %rdx
shrq $3, %rdx
movl %edx, %esi
andl $1, %edx
movl %edx, foo_opts_set+28(%rip)
movq %rax, %rdx
andl $11, %esi
shrq $4, %rdx
movl %esi, foo_opts_set+20(%rip)
andl $1, %edx
movl %edx, foo_opts_set+32(%rip)
movq %rax, %rdx
shrq $6, %rax
andl $1, %eax
shrq $5, %rdx
movl %eax, foo_opts_set+40(%rip)
movq 24(%rcx), %rax
andl $1, %edx
movl %edx, foo_opts_set+36(%rip)
movl %eax, x_param_phiopt_factor_max_stmts_live(%rip)
ret
.cfi_endproc
.LFE0:
.size _Z3foov, .-_Z3foov
