public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug middle-end/103870] New: ARM: Wrong branch instruction with optimization O2 and higher
@ 2021-12-30 22:03 petro.karashchenko at gmail dot com
2021-12-30 22:06 ` [Bug middle-end/103870] " petro.karashchenko at gmail dot com
2021-12-30 22:07 ` pinskia at gcc dot gnu.org
0 siblings, 2 replies; 3+ messages in thread
From: petro.karashchenko at gmail dot com @ 2021-12-30 22:03 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103870
Bug ID: 103870
Summary: ARM: Wrong branch instruction with optimization O2 and
higher
Product: gcc
Version: 10.3.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: petro.karashchenko at gmail dot com
Target Milestone: ---
Code test.c:
void f(void)
{
double a = 2.2204460492503131e-16;
int b = 1;
int c = 0;
while (a > 0)
{
if (++c >= 10)
{
b *= 2;
c = 0;
}
if (b > 1)
{
a *= b;
}
}
}
----------
arm-none-eabi-gcc -save-temps -Wall -Wextra -c -march=armv7e-m -mtune=cortex-m7
-mthumb -mfpu=fpv5-d16 -mfloat-abi=hard test.c -O0
----------
Generated assembly:
.arch armv7e-m
.fpu fpv5-d16
.eabi_attribute 28, 1
.eabi_attribute 20, 1
.eabi_attribute 21, 1
.eabi_attribute 23, 3
.eabi_attribute 24, 1
.eabi_attribute 25, 1
.eabi_attribute 26, 1
.eabi_attribute 30, 6
.eabi_attribute 34, 1
.eabi_attribute 18, 4
.file "test2.c"
.text
.align 1
.global f
.syntax unified
.thumb
.thumb_func
.type f, %function
f:
@ args = 0, pretend = 0, frame = 16
@ frame_needed = 1, uses_anonymous_args = 0
@ link register save eliminated.
push {r7}
sub sp, sp, #20
add r7, sp, #0
mov r2, #0
ldr r3, .L5
strd r2, [r7, #8]
movs r3, #1
str r3, [r7, #4]
movs r3, #0
str r3, [r7]
b .L2
.L4:
ldr r3, [r7]
adds r3, r3, #1
str r3, [r7]
ldr r3, [r7]
cmp r3, #9
ble .L3
ldr r3, [r7, #4]
lsls r3, r3, #1
str r3, [r7, #4]
movs r3, #0
str r3, [r7]
.L3:
ldr r3, [r7, #4]
cmp r3, #1
ble .L2
ldr r3, [r7, #4]
vmov s15, r3 @ int
vcvt.f64.s32 d7, s15
vldr.64 d6, [r7, #8]
vmul.f64 d7, d6, d7
vstr.64 d7, [r7, #8]
.L2:
vldr.64 d7, [r7, #8]
vcmpe.f64 d7, #0
vmrs APSR_nzcv, FPSCR
bgt .L4
nop
nop
adds r7, r7, #20
mov sp, r7
@ sp needed
ldr r7, [sp], #4
bx lr
.L6:
.align 2
.L5:
.word 1018167296
.size f, .-f
.ident "GCC: (GNU Arm Embedded Toolchain 10.3-2021.10) 10.3.1 20210824
(release)"
----------
arm-none-eabi-gcc -save-temps -Wall -Wextra -c -march=armv7e-m -mtune=cortex-m7
-mthumb -mfpu=fpv5-d16 -mfloat-abi=hard test.c -Os
----------
Generated assembly:
.arch armv7e-m
.fpu fpv5-d16
.eabi_attribute 28, 1
.eabi_attribute 20, 1
.eabi_attribute 21, 1
.eabi_attribute 23, 3
.eabi_attribute 24, 1
.eabi_attribute 25, 1
.eabi_attribute 26, 1
.eabi_attribute 30, 4
.eabi_attribute 34, 1
.eabi_attribute 18, 4
.file "test2.c"
.text
.align 1
.global f
.syntax unified
.thumb
.thumb_func
.type f, %function
f:
@ args = 0, pretend = 0, frame = 0
@ frame_needed = 0, uses_anonymous_args = 0
@ link register save eliminated.
vldr.64 d7, .L10
movs r3, #0
movs r2, #1
.L5:
adds r3, r3, #1
cmp r3, #9
ble .L2
lsls r2, r2, #1
movs r3, #0
.L3:
vmov s13, r2 @ int
vcvt.f64.s32 d6, s13
vmul.f64 d7, d7, d6
b .L4
.L2:
cmp r2, #1
bne .L3
.L4:
vcmpe.f64 d7, #0
vmrs APSR_nzcv, FPSCR
bgt .L5
bx lr
.L11:
.align 3
.L10:
.word 0
.word 1018167296
.size f, .-f
.ident "GCC: (GNU Arm Embedded Toolchain 10.3-2021.10) 10.3.1 20210824
(release)"
----------
The problem appears with condition "if (b > 1)". If -O0 is applied then code
ldr r3, [r7, #4]
cmp r3, #1
ble .L2
is generate that "ble" instruction code is used, however with -Os we are
getting
cmp r2, #1
bne .L3
code generated that use "bne" instruction code. The case is that with code
above "b *= 2;" at some point "b" overflows and becomes equal to zero. The
"ble" instruction code will catch this however "bne" will not. So usage of
"bne" is inappropriate for "if (b > 1)" condition that is intended to truncate
negative values and zero.
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug middle-end/103870] ARM: Wrong branch instruction with optimization O2 and higher
2021-12-30 22:03 [Bug middle-end/103870] New: ARM: Wrong branch instruction with optimization O2 and higher petro.karashchenko at gmail dot com
@ 2021-12-30 22:06 ` petro.karashchenko at gmail dot com
2021-12-30 22:07 ` pinskia at gcc dot gnu.org
1 sibling, 0 replies; 3+ messages in thread
From: petro.karashchenko at gmail dot com @ 2021-12-30 22:06 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103870
--- Comment #1 from Petro Karashchenko <petro.karashchenko at gmail dot com> ---
I would expect "bgt" instruction instead of "bne" in such case.
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Bug middle-end/103870] ARM: Wrong branch instruction with optimization O2 and higher
2021-12-30 22:03 [Bug middle-end/103870] New: ARM: Wrong branch instruction with optimization O2 and higher petro.karashchenko at gmail dot com
2021-12-30 22:06 ` [Bug middle-end/103870] " petro.karashchenko at gmail dot com
@ 2021-12-30 22:07 ` pinskia at gcc dot gnu.org
1 sibling, 0 replies; 3+ messages in thread
From: pinskia at gcc dot gnu.org @ 2021-12-30 22:07 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103870
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution|--- |INVALID
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Use either -fwrapv or have the math done as unsigned int so you can detect
signed integer overflow (which is undefined). You could also use builtins too.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2021-12-30 22:07 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-30 22:03 [Bug middle-end/103870] New: ARM: Wrong branch instruction with optimization O2 and higher petro.karashchenko at gmail dot com
2021-12-30 22:06 ` [Bug middle-end/103870] " petro.karashchenko at gmail dot com
2021-12-30 22:07 ` 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).