| Issue |
208731
|
| Summary |
[clang][armv8] Wrong compilation when using neon intrinsics
|
| Labels |
clang
|
| Assignees |
|
| Reporter |
M-HT
|
I'm using LLVM 22.1.8 on Linux (LLVM-22.1.8-Linux-X64.tar.xz from releases).
I have following test file with a function which should fill the buffer with zeroes:
```c
#include <arm_neon.h>
#include <stdint.h>
void test(int32_t counter, uint8_t * restrict dst)
{
int16x4_t d0;
d0 = vmov_n_s16(0);
do
{
vst1_s16((int16_t *)dst, d0);
dst += 8;
counter -= 4;
} while (counter > 0);
}
```
When I compile it with command `clang -c -O2 -Wall -target arm64-unknown-linux-gnu test.c` and disassemble the result with command `llvm-objdump -d test.o`, I get following:
```
test.o: file format elf64-littleaarch64
Disassembly of section .text:
0000000000000000 <test>:
0: 52800088 mov w8, #0x4 // =4
4: 7100101f cmp w0, #0x4
8: 1a88b008 csel w8, w0, w8, lt
c: 4b080008 sub w8, w0, w8
10: aa0103e0 mov x0, x1
14: 2a1f03e1 mov w1, wzr
18: 11000d08 add w8, w8, #0x3
1c: 53027d08 lsr w8, w8, #2
20: d37d7d08 ubfiz x8, x8, #3, #32
24: 91002102 add x2, x8, #0x8
28: 14000000 b 0x28 <test+0x28>
```
There is no instruction for writing to memory and the last instruction is infinite cycle, so that's clearly not the correct result.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
