public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug target/62128] New: Use vpalignr for AVX2 rotation
@ 2014-08-13 20:21 glisse at gcc dot gnu.org
2014-08-14 8:16 ` [Bug target/62128] " evstupac at gmail dot com
` (4 more replies)
0 siblings, 5 replies; 6+ messages in thread
From: glisse at gcc dot gnu.org @ 2014-08-13 20:21 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
Bug ID: 62128
Summary: Use vpalignr for AVX2 rotation
Product: gcc
Version: 5.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: glisse at gcc dot gnu.org
Target: x86_64-linux-gnu
typedef unsigned char vec __attribute__((vector_size(32)));
vec f(vec x){
vec
m={1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,0};
return __builtin_shuffle(x,m);
}
We generate, with -O3 -mavx2:
vpshufb .LC0(%rip), %ymm0, %ymm1
vpshufb .LC1(%rip), %ymm0, %ymm0
vpermq $78, %ymm1, %ymm1
vpor %ymm1, %ymm0, %ymm0
But unless I am mistaken, a lane swap and vpalignr should do it in 2
instructions and without reading constants from memory. There is a function
expand_vec_perm_palignr but it only handles some 128 bit cases. Even for
permutations that can be done with a single 256 bit vpalignr instruction, we
never seem to generate it.
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Bug target/62128] Use vpalignr for AVX2 rotation
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
@ 2014-08-14 8:16 ` evstupac at gmail dot com
2014-08-26 13:50 ` evstupac at gmail dot com
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: evstupac at gmail dot com @ 2014-08-14 8:16 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
Stupachenko Evgeny <evstupac at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |evstupac at gmail dot com
--- Comment #1 from Stupachenko Evgeny <evstupac at gmail dot com> ---
Confirm.
A part of the patch fixing this discussed at:
https://gcc.gnu.org/ml/gcc-patches/2014-08/msg01434.html
The other part is generation of corresponding pattern for rotation. I'll create
corresponding patch.
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Bug target/62128] Use vpalignr for AVX2 rotation
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
2014-08-14 8:16 ` [Bug target/62128] " evstupac at gmail dot com
@ 2014-08-26 13:50 ` evstupac at gmail dot com
2014-10-02 7:30 ` jakub at gcc dot gnu.org
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: evstupac at gmail dot com @ 2014-08-26 13:50 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
--- Comment #2 from Stupachenko Evgeny <evstupac at gmail dot com> ---
The patch fixing this submitted for review.
Code generated when patch applied:
vperm2i128 $33, %ymm0, %ymm0, %ymm1
vpalignr $1, %ymm0, %ymm1, %ymm1
vmovdqa %ymm1, %ymm0
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Bug target/62128] Use vpalignr for AVX2 rotation
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
2014-08-14 8:16 ` [Bug target/62128] " evstupac at gmail dot com
2014-08-26 13:50 ` evstupac at gmail dot com
@ 2014-10-02 7:30 ` jakub at gcc dot gnu.org
2014-10-03 7:28 ` jakub at gcc dot gnu.org
2014-11-28 9:08 ` jakub at gcc dot gnu.org
4 siblings, 0 replies; 6+ messages in thread
From: jakub at gcc dot gnu.org @ 2014-10-02 7:30 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Thu Oct 2 07:29:49 2014
New Revision: 215796
URL: https://gcc.gnu.org/viewcvs?rev=215796&root=gcc&view=rev
Log:
PR target/62128
* config/i386/i386.c (expand_vec_perm_1): Try expand_vec_perm_palignr
if it expands to a single insn only.
(expand_vec_perm_palignr): Add SINGLE_INSN_ONLY_P argument. If true,
fail unless in_order is true. Add forward declaration.
(expand_vec_perm_vperm2f128): Fix up comment about which permutation
is useful for one_operand_p.
(ix86_expand_vec_perm_const_1): Adjust expand_vec_perm_palignr caller.
Modified:
trunk/gcc/ChangeLog
trunk/gcc/config/i386/i386.c
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Bug target/62128] Use vpalignr for AVX2 rotation
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
` (2 preceding siblings ...)
2014-10-02 7:30 ` jakub at gcc dot gnu.org
@ 2014-10-03 7:28 ` jakub at gcc dot gnu.org
2014-11-28 9:08 ` jakub at gcc dot gnu.org
4 siblings, 0 replies; 6+ messages in thread
From: jakub at gcc dot gnu.org @ 2014-10-03 7:28 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Fri Oct 3 07:27:50 2014
New Revision: 215834
URL: https://gcc.gnu.org/viewcvs?rev=215834&root=gcc&view=rev
Log:
PR target/62128
* config/i386/i386.c (expand_vec_perm_palignr): If op1, op0 order
of palignr arguments can't be used due to min 0 or max - min
too high, try also op0, op1 order of palignr arguments.
* gcc.dg/torture/vshuf-16.inc (TESTS): Add 2 new permutations.
* gcc.dg/torture/vshuf-32.inc (TESTS): Add 5 new permutations.
Modified:
trunk/gcc/ChangeLog
trunk/gcc/config/i386/i386.c
trunk/gcc/testsuite/ChangeLog
trunk/gcc/testsuite/gcc.dg/torture/vshuf-16.inc
trunk/gcc/testsuite/gcc.dg/torture/vshuf-32.inc
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Bug target/62128] Use vpalignr for AVX2 rotation
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
` (3 preceding siblings ...)
2014-10-03 7:28 ` jakub at gcc dot gnu.org
@ 2014-11-28 9:08 ` jakub at gcc dot gnu.org
4 siblings, 0 replies; 6+ messages in thread
From: jakub at gcc dot gnu.org @ 2014-11-28 9:08 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62128
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
CC| |jakub at gcc dot gnu.org
Resolution|--- |FIXED
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Should be fixed now.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2014-11-28 9:08 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-08-13 20:21 [Bug target/62128] New: Use vpalignr for AVX2 rotation glisse at gcc dot gnu.org
2014-08-14 8:16 ` [Bug target/62128] " evstupac at gmail dot com
2014-08-26 13:50 ` evstupac at gmail dot com
2014-10-02 7:30 ` jakub at gcc dot gnu.org
2014-10-03 7:28 ` jakub at gcc dot gnu.org
2014-11-28 9:08 ` jakub at gcc dot gnu.org
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).