public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug tree-optimization/110613] New: optimization about combined store of adjacent bitfields
@ 2023-07-10 14:47 lh_mouse at 126 dot com
2023-07-10 23:53 ` [Bug tree-optimization/110613] " pinskia at gcc dot gnu.org
` (2 more replies)
0 siblings, 3 replies; 4+ messages in thread
From: lh_mouse at 126 dot com @ 2023-07-10 14:47 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110613
Bug ID: 110613
Summary: optimization about combined store of adjacent
bitfields
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: lh_mouse at 126 dot com
Target Milestone: ---
This is a piece of code taken from a WebSocket frame parser:
```
#include <stdint.h>
struct Header
{
// 1st byte
uint8_t opcode : 4;
uint8_t rsv3 : 1;
uint8_t rsv2 : 1;
uint8_t rsv1 : 1;
uint8_t fin : 1;
// 2nd byte
uint8_t reserved_1 : 7;
uint8_t mask : 1;
// 3rd and 4th bytes
uint8_t reserved_2;
uint8_t reserved_3;
// 5th to 7th bytes
union {
char mask_key[4];
uint32_t mask_key_u32;
};
// 8th to 15th bytes
uint64_t payload_len;
};
void
set_header(Header* ph, const uint8_t* bptr)
{
uint8_t f = bptr[0];
uint8_t t = bptr[1];
ph->fin = f >> 7;
ph->rsv1 = f >> 6;
ph->rsv2 = f >> 5;
ph->rsv3 = f >> 4;
ph->opcode = f;
ph->mask = t >> 7;
ph->reserved_1 = t;
}
```
The structure is designed to match x86_64 ABI (little endian), so
```
ph->fin = f >> 7;
ph->rsv1 = f >> 6;
ph->rsv2 = f >> 5;
ph->rsv3 = f >> 4;
ph->opcode = f;
```
should be a simple move (https://gcc.godbolt.org/z/9vTqs7axj), and
```
ph->mask = t >> 7;
ph->reserved_1 = t;
```
should also be a simple move (https://gcc.godbolt.org/z/KdchWvEn1), but!
When put these two pieces of code together, guess what?:
(godbolt: https://gcc.godbolt.org/z/hbEaeb3MT)
```
set_header(Header*, unsigned char const*):
movzx edx, BYTE PTR [rsi]
movzx ecx, BYTE PTR [rsi+1]
mov eax, edx
mov esi, edx
shr al, 4
and esi, 15
and eax, 1
sal eax, 4
or eax, esi
mov esi, edx
shr sil, 5
and esi, 1
sal esi, 5
or eax, esi
mov esi, edx
shr dl, 7
shr sil, 6
movzx edx, dl
and esi, 1
sal edx, 7
sal esi, 6
or eax, esi
or eax, edx
mov edx, ecx
shr cl, 7
and edx, 127
sal ecx, 15
sal edx, 8
or eax, edx
or eax, ecx
mov WORD PTR [rdi], ax
ret
```
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/110613] optimization about combined store of adjacent bitfields
2023-07-10 14:47 [Bug tree-optimization/110613] New: optimization about combined store of adjacent bitfields lh_mouse at 126 dot com
@ 2023-07-10 23:53 ` pinskia at gcc dot gnu.org
2023-07-11 7:43 ` rguenth at gcc dot gnu.org
2023-07-11 8:38 ` lh_mouse at 126 dot com
2 siblings, 0 replies; 4+ messages in thread
From: pinskia at gcc dot gnu.org @ 2023-07-10 23:53 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110613
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Keywords| |missed-optimization
Last reconfirmed| |2023-07-10
--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.
Note neither independently is optimized at the tree level; only at the RTL
level.
Note moving the load from bptr[1] to right before the first usage fixes it at
the RTL level too (most likely due to aliasing of character types being special
in C/C++).
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/110613] optimization about combined store of adjacent bitfields
2023-07-10 14:47 [Bug tree-optimization/110613] New: optimization about combined store of adjacent bitfields lh_mouse at 126 dot com
2023-07-10 23:53 ` [Bug tree-optimization/110613] " pinskia at gcc dot gnu.org
@ 2023-07-11 7:43 ` rguenth at gcc dot gnu.org
2023-07-11 8:38 ` lh_mouse at 126 dot com
2 siblings, 0 replies; 4+ messages in thread
From: rguenth at gcc dot gnu.org @ 2023-07-11 7:43 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110613
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
bitfield lowering / store merging might be the enabling passes
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/110613] optimization about combined store of adjacent bitfields
2023-07-10 14:47 [Bug tree-optimization/110613] New: optimization about combined store of adjacent bitfields lh_mouse at 126 dot com
2023-07-10 23:53 ` [Bug tree-optimization/110613] " pinskia at gcc dot gnu.org
2023-07-11 7:43 ` rguenth at gcc dot gnu.org
@ 2023-07-11 8:38 ` lh_mouse at 126 dot com
2 siblings, 0 replies; 4+ messages in thread
From: lh_mouse at 126 dot com @ 2023-07-11 8:38 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110613
--- Comment #3 from LIU Hao <lh_mouse at 126 dot com> ---
There are some more cases about loading adjacent bitfields; not sure whether I
should create new PRs or paste them here. They seem highly related to the
aliasing characteristics of `unsigned char`; if I inject
`__atomic_signal_fence(__ATOMIC_RELAXED);` between the loads and the stores,
the issue is gone.
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2023-07-11 8:38 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-07-10 14:47 [Bug tree-optimization/110613] New: optimization about combined store of adjacent bitfields lh_mouse at 126 dot com
2023-07-10 23:53 ` [Bug tree-optimization/110613] " pinskia at gcc dot gnu.org
2023-07-11 7:43 ` rguenth at gcc dot gnu.org
2023-07-11 8:38 ` lh_mouse at 126 dot com
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).