https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64681
Bug ID: 64681
Summary: gcc assign wrong register for arm inline assembly
Product: gcc
Version: 4.9.0
Status: UNCONFIRMED
Severity: blocker
Priority: P3
Component: inline-asm
Assignee: unassigned at gcc dot gnu.org
Reporter: zhongwei.yao at arm dot com
Created attachment 34492
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34492&action=edit
all related file.
I compile following code on linux by arm-linux-androideabi-gcc (gcc version 4.9
20140827) with following command:
arm-linux-androideabi-g++ -mfloat-abi=softfp -mfpu=neon -mthumb -Wall -Wextra
-march=armv7-a --sysroot=$ndk/platforms/android-21/arch-arm -O2 -Wall main.cpp
==c code start==
#include
int main(void) {
return 0;
}
char buffer[32];
void bar(int n) {
int j = 0;
int64x1_t s = vdup_n_s64(0);
int64x1_t onev = vdup_n_s64(1);
int64_t sum = 0;
for (j = 0; j <= n; j++) {
asm ("vsub.s64 %0, %0, d1"
: "+w" (s): : "memory");
}
sum = (int64_t)j;
sum >>= vget_lane_s64(s, 0);
if(sum >0)
s = vsub_s64(s, onev);
vst1_s64((int64_t*)buffer, s);
}
==code end==
It returns:
/tmp/ccE0j9sL.s: Assembler messages:
/tmp/ccE0j9sL.s:55: Error: invalid instruction shape -- `vsub.s64 s14,s14,d1'
I think it is the bug in gcc that assign wrong register for variable s. It
should be d14 here, while it assignes s14.
The generated assembly file is:
==asm code start==
.syntax unified
.arch armv7-a
.eabi_attribute 27, 3
.fpu neon
.eabi_attribute 20, 1
.eabi_attribute 21, 1
.eabi_attribute 23, 3
.eabi_attribute 24, 1
.eabi_attribute 25, 1
.eabi_attribute 26, 2
.eabi_attribute 30, 2
.eabi_attribute 34, 1
.eabi_attribute 18, 4
.thumb
.file"main.cpp"
.section.text.startup,"ax",%progbits
.align2
.globalmain
.thumb
.thumb_func
.typemain, %function
main:
.fnstart
.LFB1870:
@ args = 0, pretend = 0, frame = 0
@ frame_needed = 0, uses_anonymous_args = 0
@ link register save eliminated.
movsr0, #0
bxlr
.cantunwind
.fnend
.sizemain, .-main
.text
.align2
.global_Z3bari
.thumb
.thumb_func
.type_Z3bari, %function
_Z3bari:
.fnstart
.LFB1871:
@ args = 0, pretend = 0, frame = 0
@ frame_needed = 0, uses_anonymous_args = 0
cmpr0, #0
vmov.i32d7, #0 @ di
push{lr}
blt.L5
addsr0, r0, #1
movsr2, #0
.L4:
addsr2, r2, #1
cmpr2, r0
#APP
@ 17 "src/main.cpp" 1
vsub.s64 s14, s14, d1
@ 0 "" 2
.thumb
bne.L4
fmrsr1, s14@ int
asrsr3, r2, #31
rsbr0, r1, #32
subslr, r1, #32
lslr0, r3, r0
lsrr2, r2, r1
itpl
asrpllr, r3, lr
orrr2, r2, r0
itpl
orrplr2, r2, lr
asrsr3, r3, r1
cmpr2, #1
sbcsr3, r3, #0
blt.L5
vmov.i32d16, #0x @ di
vadd.i64d7, d7, d16
.L5:
ldrr3, .L10
.LPIC1:
addr3, pc
ldrr3, [r3]
vst1.64{d7}, [r3:64]
ldrpc, [sp], #4
.L11:
.align2
.L10:
.wordbuffer(GOT_PREL)+(.-(.LPIC1+4))
.cantunwind
.fnend
.size_Z3bari, .-_Z3bari
.globalbuffer
.bss
.align3
.typebuffer, %object
.sizebuffer, 32
buffer:
.space32
.ident"GCC: (GNU) 4.9 20140827 (prerelease)"
.section.note.GNU-stack,"",%progbits
==asm code end==