public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics
@ 2021-11-05 9:50 tnfchris at gcc dot gnu.org
2021-11-05 9:50 ` [Bug target/103094] " tnfchris at gcc dot gnu.org
` (8 more replies)
0 siblings, 9 replies; 10+ messages in thread
From: tnfchris at gcc dot gnu.org @ 2021-11-05 9:50 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Bug ID: 103094
Summary: [12 Regression] Incorrect codegen from AArch64
intrinsics
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: tnfchris at gcc dot gnu.org
Target Milestone: ---
Target: aarch64-*
The following testcase
---
#include <arm_neon.h>
void foo (uint8x8x2_t cols_01_23, uint8x8x2_t cols_45_67, uint16_t* outptr0)
{
uint16x4x4_t cols_01_23_45_67 = { {
vreinterpret_u16_u8(cols_01_23.val[0]),
vreinterpret_u16_u8(cols_01_23.val[1]),
vreinterpret_u16_u8(cols_45_67.val[0]),
vreinterpret_u16_u8(cols_45_67.val[1])
} };
vst4_lane_u16(outptr0, cols_01_23_45_67, 0);
}
---
Causes incorrect codegen
```
foo:
mov v4.8b, v0.8b
mov v5.8b, v1.8b
mov v6.8b, v1.8b
mov v7.8b, v2.8b
st4 {v4.h - v7.h}[0], [x0]
ret
```
as the same register `v1` is used as source for two inputs.
This happens in cprop_hardreg where a correct RTL
```
(insn 20 37 26 2 (set (reg:V4HI 61 v29 [ __val+8 ])
(reg:V4HI 36 v4 [ cols_01_23+8 ])) 1158 {*aarch64_simd_movv4hi}
(nil))
```
is turned into
```
(insn 20 37 26 2 (set (reg:V4HI 61 v29 [ __val+8 ])
(reg:V4HI 33 v1 [orig:36 cols_01_23+8 ] [36])) 1158
{*aarch64_simd_movv4hi}
(nil))
```
which is a bogus transform as `v1` is still live and pointing to `cols_45_67`.
started somewhere between g:4096eb50d108a9ccef134f2e891c751b1cc9f6b2 and
g:29a1af24eface3620e348be9429e7c2e872accbc
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from AArch64 intrinsics
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
@ 2021-11-05 9:50 ` tnfchris at gcc dot gnu.org
2021-11-05 12:18 ` tnfchris at gcc dot gnu.org
` (7 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: tnfchris at gcc dot gnu.org @ 2021-11-05 9:50 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Tamar Christina <tnfchris at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |12.0
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from AArch64 intrinsics
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
2021-11-05 9:50 ` [Bug target/103094] " tnfchris at gcc dot gnu.org
@ 2021-11-05 12:18 ` tnfchris at gcc dot gnu.org
2021-11-05 12:22 ` pinskia at gcc dot gnu.org
` (6 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: tnfchris at gcc dot gnu.org @ 2021-11-05 12:18 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
--- Comment #1 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
Looks like it's wrong from expand already, it's expanding into overlapping
registers.
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from AArch64 intrinsics
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
2021-11-05 9:50 ` [Bug target/103094] " tnfchris at gcc dot gnu.org
2021-11-05 12:18 ` tnfchris at gcc dot gnu.org
@ 2021-11-05 12:22 ` pinskia at gcc dot gnu.org
2021-11-05 12:29 ` tnfchris at gcc dot gnu.org
` (5 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-11-05 12:22 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
See Also| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=101529
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Tamar Christina from comment #1)
> Looks like it's wrong from expand already, it's expanding into overlapping
> registers.
Maybe a dup of bug 101529?
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from AArch64 intrinsics
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (2 preceding siblings ...)
2021-11-05 12:22 ` pinskia at gcc dot gnu.org
@ 2021-11-05 12:29 ` tnfchris at gcc dot gnu.org
2021-11-06 5:50 ` pinskia at gcc dot gnu.org
` (4 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: tnfchris at gcc dot gnu.org @ 2021-11-05 12:29 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
--- Comment #3 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> (In reply to Tamar Christina from comment #1)
> > Looks like it's wrong from expand already, it's expanding into overlapping
> > registers.
>
> Maybe a dup of bug 101529?
hmm didn't know about that ticket, will have a look at stage 3, but this is
likely unrelated. I think this is related to the new packed vector modes added
last night to handle the structure intrinsics (st[1,2,3,4] etc).
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from AArch64 intrinsics
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (3 preceding siblings ...)
2021-11-05 12:29 ` tnfchris at gcc dot gnu.org
@ 2021-11-06 5:50 ` pinskia at gcc dot gnu.org
2021-11-06 22:33 ` [Bug target/103094] [12 Regression] Incorrect codegen from uint8x8x2_t function arguments pinskia at gcc dot gnu.org
` (3 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-11-06 5:50 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2021-11-06
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
See Also|https://gcc.gnu.org/bugzill |
|a/show_bug.cgi?id=101529 |
CC| |pinskia at gcc dot gnu.org
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
What is wrong rather not understanding the incoming arguments is passed as q0
rather than d0/d1.
(insn 2 6 3 2 (set (reg/v:V2x8QI 101 [ cols_01_23 ])
(reg:V2x8QI 32 v0 [ cols_01_23 ])) "t66.c":5:1 -1
(nil))
(insn 3 2 4 2 (set (reg/v:V2x8QI 102 [ cols_45_67 ])
(reg:V2x8QI 33 v1 [ cols_45_67 ])) "t66.c":5:1 -1
(nil))
Caused by r12-4912-geb04ccf4bfd6586 .
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] Incorrect codegen from uint8x8x2_t function arguments
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (4 preceding siblings ...)
2021-11-06 5:50 ` pinskia at gcc dot gnu.org
@ 2021-11-06 22:33 ` pinskia at gcc dot gnu.org
2021-12-13 9:12 ` [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect tnfchris at gcc dot gnu.org
` (2 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-11-06 22:33 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|[12 Regression] Incorrect |[12 Regression] Incorrect
|codegen from AArch64 |codegen from uint8x8x2_t
|intrinsics |function arguments
Severity|normal |blocker
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect.
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (5 preceding siblings ...)
2021-11-06 22:33 ` [Bug target/103094] [12 Regression] Incorrect codegen from uint8x8x2_t function arguments pinskia at gcc dot gnu.org
@ 2021-12-13 9:12 ` tnfchris at gcc dot gnu.org
2021-12-15 12:19 ` cvs-commit at gcc dot gnu.org
2021-12-15 12:24 ` rsandifo at gcc dot gnu.org
8 siblings, 0 replies; 10+ messages in thread
From: tnfchris at gcc dot gnu.org @ 2021-12-13 9:12 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
Tamar Christina <tnfchris at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jonathan.wright at arm dot com
Priority|P3 |P1
Summary|[12 Regression] Incorrect |[12 Regression] AAPCS for
|codegen from uint8x8x2_t |new partial vector mode
|function arguments |types (e.g. V2x8QI) are
| |incorrect.
Assignee|unassigned at gcc dot gnu.org |tnfchris at gcc dot gnu.org
--- Comment #5 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
The frame layout and argument layout code for the new structs are incorrect, it
leads to the compiler thinking that every pair of values only use a single
register.
i.e. V2x8QI is 16 bytes long, so the compiler thinks the values are packed in a
single register. This causes any code that returns or passes these new types
as function arguments to be incorrect.
Mine.
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect.
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (6 preceding siblings ...)
2021-12-13 9:12 ` [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect tnfchris at gcc dot gnu.org
@ 2021-12-15 12:19 ` cvs-commit at gcc dot gnu.org
2021-12-15 12:24 ` rsandifo at gcc dot gnu.org
8 siblings, 0 replies; 10+ messages in thread
From: cvs-commit at gcc dot gnu.org @ 2021-12-15 12:19 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Sandiford <rsandifo@gcc.gnu.org>:
https://gcc.gnu.org/g:73c3dacef9a30d7d66918606a97c496c71289f1b
commit r12-5998-g73c3dacef9a30d7d66918606a97c496c71289f1b
Author: Richard Sandiford <richard.sandiford@arm.com>
Date: Wed Dec 15 12:19:00 2021 +0000
aarch64: Don't classify vector pairs as short vectors [PR103094]
In this PR we were wrongly classifying a pair of 8-byte vectors
as a 16-byte âshort vectorâ (in the AAPCS64 sense). As the
comment in the patch says, this stems from an old condition
in aarch64_short_vector_p that is too loose, but that would
be difficult to tighten now.
We can still do the right thing for the newly-added modes though,
since there are no backwards compatibility concerns there.
Co-authored-by: Tamar Christina <tamar.christina@arm.com>
gcc/
PR target/103094
* config/aarch64/aarch64.c (aarch64_short_vector_p): Return false
for structure modes, rather than ignoring the type in that case.
gcc/testsuite/
PR target/103094
* gcc.target/aarch64/pr103094.c: New test.
^ permalink raw reply [flat|nested] 10+ messages in thread
* [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect.
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
` (7 preceding siblings ...)
2021-12-15 12:19 ` cvs-commit at gcc dot gnu.org
@ 2021-12-15 12:24 ` rsandifo at gcc dot gnu.org
8 siblings, 0 replies; 10+ messages in thread
From: rsandifo at gcc dot gnu.org @ 2021-12-15 12:24 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103094
rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rsandifo at gcc dot gnu.org
Status|NEW |RESOLVED
Resolution|--- |FIXED
--- Comment #7 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> ---
Fixed.
^ permalink raw reply [flat|nested] 10+ messages in thread
end of thread, other threads:[~2021-12-15 12:24 UTC | newest]
Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-11-05 9:50 [Bug target/103094] New: [12 Regression] Incorrect codegen from AArch64 intrinsics tnfchris at gcc dot gnu.org
2021-11-05 9:50 ` [Bug target/103094] " tnfchris at gcc dot gnu.org
2021-11-05 12:18 ` tnfchris at gcc dot gnu.org
2021-11-05 12:22 ` pinskia at gcc dot gnu.org
2021-11-05 12:29 ` tnfchris at gcc dot gnu.org
2021-11-06 5:50 ` pinskia at gcc dot gnu.org
2021-11-06 22:33 ` [Bug target/103094] [12 Regression] Incorrect codegen from uint8x8x2_t function arguments pinskia at gcc dot gnu.org
2021-12-13 9:12 ` [Bug target/103094] [12 Regression] AAPCS for new partial vector mode types (e.g. V2x8QI) are incorrect tnfchris at gcc dot gnu.org
2021-12-15 12:19 ` cvs-commit at gcc dot gnu.org
2021-12-15 12:24 ` rsandifo 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).