public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index
@ 2022-11-14 3:38 ibuclaw at gdcproject dot org
2022-11-14 3:44 ` [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2) ibuclaw at gdcproject dot org
` (6 more replies)
0 siblings, 7 replies; 8+ messages in thread
From: ibuclaw at gdcproject dot org @ 2022-11-14 3:38 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
Bug ID: 107671
Summary: i386: Missed optimization: use of bt in bit test
pattern when LHS is an array index
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: ibuclaw at gdcproject dot org
Target Milestone: ---
int bt32_setb(const __UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
{
return ((p[bitnum >> 5] & (1 << (bitnum & 31)))) != 0;
}
int bt64_setb(const __UINT64_TYPE__ *p, __UINT64_TYPE__ bitnum)
{
return ((p[bitnum >> 6] & (1L << (bitnum & 63)))) != 0;
}
int bt32_setb2(const __UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
{
return (p[bitnum >> 5] >> (bitnum & 31)) & 1;
}
int bt64_setb2(const __UINT64_TYPE__ *p, __UINT64_TYPE__ bitnum)
{
return (p[bitnum >> 6] >> (bitnum & 63)) & 1;
}
int bt32_setae(const __UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
{
return ((p[bitnum >> 5] & (1 << (bitnum & 31)))) == 0;
}
int bt64_setae(const __UINT64_TYPE__ *p, __UINT64_TYPE__ bitnum)
{
return ((p[bitnum >> 6] & (1L << (bitnum & 63)))) == 0;
}
int bt32_setae2(const __UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
{
return !((p[bitnum >> 5] >> (bitnum & 31)) & 1);
}
int bt64_setae2(const __UINT64_TYPE__ *p, __UINT64_TYPE__ bitnum)
{
return !((p[bitnum >> 6] >> (bitnum & 63)) & 1);
}
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
@ 2022-11-14 3:44 ` ibuclaw at gdcproject dot org
2022-11-14 4:00 ` ibuclaw at gdcproject dot org
` (5 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: ibuclaw at gdcproject dot org @ 2022-11-14 3:44 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
--- Comment #1 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
Non-pointer variants also not detected.
---
int bt32v_setb(const __UINT32_TYPE__ v, __UINT32_TYPE__ bitnum)
{
return ((v & (1 << (bitnum & 31)))) != 0;
}
int bt64v_setb(const __UINT64_TYPE__ v, __UINT64_TYPE__ bitnum)
{
return ((v & (1L << (bitnum & 63)))) != 0;
}
int bt32v_setb2(const __UINT32_TYPE__ v, __UINT32_TYPE__ bitnum)
{
return (v >> (bitnum & 31)) & 1;
}
int bt64v_setb2(const __UINT64_TYPE__ v, __UINT64_TYPE__ bitnum)
{
return (v >> (bitnum & 63)) & 1;
}
int bt32v_setae(const __UINT32_TYPE__ v, __UINT32_TYPE__ bitnum)
{
return ((v & (1 << (bitnum & 31)))) == 0;
}
int bt64v_setae(const __UINT64_TYPE__ v, __UINT64_TYPE__ bitnum)
{
return ((v & (1L << (bitnum & 63)))) == 0;
}
int bt32v_setae2(const __UINT32_TYPE__ v, __UINT32_TYPE__ bitnum)
{
return !((v >> (bitnum & 31)) & 1);
}
int bt64v_setae2(const __UINT64_TYPE__ v, __UINT64_TYPE__ bitnum)
{
return !((v >> (bitnum & 63)) & 1);
}
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
2022-11-14 3:44 ` [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2) ibuclaw at gdcproject dot org
@ 2022-11-14 4:00 ` ibuclaw at gdcproject dot org
2022-11-15 1:27 ` crazylht at gmail dot com
` (4 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: ibuclaw at gdcproject dot org @ 2022-11-14 4:00 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
--- Comment #2 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
Expected generated code would be:
---
bt32_setb*:
...
shrl $5, %edx
movl (%eax,%edx,4), %edx
xorl %eax, %eax
btl %ecx, %edx
setb %al
...
---
bt32_setae*:
...
shrl $5, %edx
movl (%eax,%edx,4), %edx
xorl %eax, %eax
btl %ecx, %edx
setae %al
...
---
bt32v_setb*:
...
xorl %eax, %eax
btl %esi, %edi
setb %al
...
---
bt32v_setae*:
...
xorl %eax, %eax
btl %esi, %edi
setae %al
...
---
bt64_setb*:
...
shrq $6, %rax
movq (%rdi,%rax,8), %rcx
xorl %eax, %eax
btq %rsi, %rcx
setb %al
...
---
bt64_setae*:
...
shrq $6, %rax
movq (%rdi,%rax,8), %rcx
xorl %eax, %eax
btq %rsi, %rcx
setae %al
...
---
bt64v_setb*:
...
xorl %eax, %eax
btq %rsi, %rdi
setb %al
...
---
bt64v_setae*:
...
xorl %eax, %eax
btq %rsi, %rdi
setae %al
...
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
2022-11-14 3:44 ` [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2) ibuclaw at gdcproject dot org
2022-11-14 4:00 ` ibuclaw at gdcproject dot org
@ 2022-11-15 1:27 ` crazylht at gmail dot com
2022-11-15 13:15 ` ubizjak at gmail dot com
` (3 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: crazylht at gmail dot com @ 2022-11-15 1:27 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
--- Comment #3 from Hongtao.liu <crazylht at gmail dot com> ---
We already have
----------cut from i386.md--------
15204;; Help combine recognize bt followed by setc
15205(define_insn_and_split "*bt<mode>_setcqi"
15206 [(set (subreg:SWI48 (match_operand:QI 0 "register_operand") 0)
15207 (zero_extract:SWI48
15208 (match_operand:SWI48 1 "register_operand")
15209 (const_int 1)
15210 (zero_extend:SI (match_operand:QI 2 "register_operand"))))
15211 (clobber (reg:CC FLAGS_REG))]
15212 "TARGET_USE_BT && ix86_pre_reload_split ()"
....
15262;; Help combine recognize bt followed by setnc
15263(define_insn_and_split "*bt<mode>_setncqi"
15264 [(set (match_operand:QI 0 "register_operand")
15265 (and:QI
15266 (not:QI
15267 (subreg:QI
15268 (lshiftrt:SWI48 (match_operand:SWI48 1 "register_operand")
15269 (match_operand:QI 2 "register_operand")) 0))
15270 (const_int 1)))
15271 (clobber (reg:CC FLAGS_REG))]
15272 "TARGET_USE_BT && ix86_pre_reload_split ()"
15273 "#"
15274 "&& 1"
-----------cut end-------------
Guess we need more variants for that.
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
` (2 preceding siblings ...)
2022-11-15 1:27 ` crazylht at gmail dot com
@ 2022-11-15 13:15 ` ubizjak at gmail dot com
2022-11-15 16:19 ` ibuclaw at gdcproject dot org
` (2 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: ubizjak at gmail dot com @ 2022-11-15 13:15 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
--- Comment #4 from Uroš Bizjak <ubizjak at gmail dot com> ---
Created attachment 53901
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53901&action=edit
Patch that adds relevant zero_extract patterns
This patch adds relevant zero_extract patterns that optimize:
return ((v & (1 << (bitnum & 31)))) != 0;
return ((v & (1L << (bitnum & 63)))) != 0;
return (v >> (bitnum & 31)) & 1;
return (v >> (bitnum & 63)) & 1;
from:
movl %esi, %ecx
movl $1, %eax
sall %cl, %eax
testl %edi, %eax
setne %al
movzbl %al, %eax
to:
xorl %eax, %eax
btl %esi, %edi
setc %al
The patch adapts *jcc_bt patterns to similar *scc_bt patterns.
Please note that the patch optimizes only to SETB conditional set instruction,
since it builds on a combine transformation to the ZERO_EXTRACT rtx.
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
` (3 preceding siblings ...)
2022-11-15 13:15 ` ubizjak at gmail dot com
@ 2022-11-15 16:19 ` ibuclaw at gdcproject dot org
2023-11-12 22:41 ` pinskia at gcc dot gnu.org
2023-11-12 22:46 ` pinskia at gcc dot gnu.org
6 siblings, 0 replies; 8+ messages in thread
From: ibuclaw at gdcproject dot org @ 2022-11-15 16:19 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
--- Comment #5 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
(In reply to Uroš Bizjak from comment #4)
> from:
> movl %esi, %ecx
> movl $1, %eax
> sall %cl, %eax
> testl %edi, %eax
> setne %al
> movzbl %al, %eax
>
> to:
> xorl %eax, %eax
> btl %esi, %edi
> setc %al
>
> The patch adapts *jcc_bt patterns to similar *scc_bt patterns.
Thanks, have been tinkering with the pointer index version, and found that I
can coax scc_bt_mask to match if I cast the lhs to a signed type. It's not
obvious to me why there would be a difference.
return ((__INT32_TYPE__)p[bitnum >> 5] & (1 << (bitnum & 31))) != 0;
movl %esi, %eax
shrl $5, %eax
movl (%rdi,%rax,4), %eax
btl %esi, %eax
setc %al
Things get even stranger once I expand to "bit test and op" variants though
(better to put in another PR though)
__INT32_TYPE__ btc32(__UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
{
__INT32_TYPE__ result = ((__INT32_TYPE__)p[bitnum >> 5]
& (1 << (bitnum & 31))) != 0;
p[bitnum >> 5] ^= 1 << (bitnum & 31);
return result;
}
Patch changes code-gen in the following way.
from:
movl %esi, %eax
movl %esi, %ecx
shrl $5, %eax
leaq (%rdi,%rax,4), %rdx
movl (%rdx), %eax
movl %eax, %esi
sarl %cl, %eax
btcl %ecx, %esi
andl $1, %eax
movl %esi, (%rdx)
to:
movl %esi, %eax
shrl $5, %eax
leaq (%rdi,%rax,4), %rdx
movl (%rdx), %eax
movl %eax, %ecx
btcl %esi, %ecx
btl %esi, %eax
setc %al
movl %ecx, (%rdx)
movzbl %al, %eax
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
` (4 preceding siblings ...)
2022-11-15 16:19 ` ibuclaw at gdcproject dot org
@ 2023-11-12 22:41 ` pinskia at gcc dot gnu.org
2023-11-12 22:46 ` pinskia at gcc dot gnu.org
6 siblings, 0 replies; 8+ messages in thread
From: pinskia at gcc dot gnu.org @ 2023-11-12 22:41 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
^ permalink raw reply [flat|nested] 8+ messages in thread
* [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2)
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
` (5 preceding siblings ...)
2023-11-12 22:41 ` pinskia at gcc dot gnu.org
@ 2023-11-12 22:46 ` pinskia at gcc dot gnu.org
6 siblings, 0 replies; 8+ messages in thread
From: pinskia at gcc dot gnu.org @ 2023-11-12 22:46 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107671
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Last reconfirmed| |2023-11-12
Status|UNCONFIRMED |NEW
--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.
bt32_setb2, bt32v_setb, bt32v_setb2 are the only ones done for 64bit.
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2023-11-12 22:46 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-14 3:38 [Bug target/107671] New: i386: Missed optimization: use of bt in bit test pattern when LHS is an array index ibuclaw at gdcproject dot org
2022-11-14 3:44 ` [Bug target/107671] i386: Missed optimization: use of bt in bit test pattern (using -O2 -mtune=core2) ibuclaw at gdcproject dot org
2022-11-14 4:00 ` ibuclaw at gdcproject dot org
2022-11-15 1:27 ` crazylht at gmail dot com
2022-11-15 13:15 ` ubizjak at gmail dot com
2022-11-15 16:19 ` ibuclaw at gdcproject dot org
2023-11-12 22:41 ` pinskia at gcc dot gnu.org
2023-11-12 22:46 ` pinskia 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).