https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97194
Bug ID: 97194
Summary: optimize vector element set/extract at variable
position
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: rguenth at gcc dot gnu.org
Target Milestone: ---
#define N 32
typedef int T;
typedef T V __attribute__((vector_size(N)));
V set (V v, int idx, T val)
{
v[idx] = val;
return v;
}
T get (V v, int idx)
{
return v[idx];
}
generates with -mavx2
set:
.LFB0:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movslq %edi, %rdi
movq %rsp, %rbp
.cfi_def_cfa_register 6
andq $-32, %rsp
vmovdqa %ymm0, -32(%rsp)
movl %esi, -32(%rsp,%rdi,4)
vmovdqa -32(%rsp), %ymm0
^^^ store forwarding fail
leave
.cfi_def_cfa 7, 8
ret
get:
.LFB1:
.cfi_startproc
pushq %rbp
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movslq %edi, %rdi
movq %rsp, %rbp
.cfi_def_cfa_register 6
andq $-32, %rsp
vmovdqa %ymm0, -32(%rsp)
movl -32(%rsp,%rdi,4), %eax
leave
.cfi_def_cfa 7, 8
ret
maybe not too bad.
Vary N and T to cover all types and vector sizes.
It should be possible to do the 'get' case via variable permutes
and the 'set' case via a splat of the value and a blend using
a mask generated from 'idx'.
