llvmbot
commented
5 months ago @llvm/issue-subscribers-backend-aarch64
Author: Niles Salter (Validark)
This movemask routine ([Godbolt link](https://godbolt.org/z/feq485549)):
```c
#include <arm_neon.h>
uint64_t vmovmaskq_u8_(const uint8x16_t p0, const uint8x16_t p1, const uint8x16_t p2, const uint8x16_t p3) {
const uint8x16_t bitmask1 = { 0x01, 0x10, 0x01, 0x10, 0x01, 0x10, 0x01, 0x10,
0x01, 0x10, 0x01, 0x10, 0x01, 0x10, 0x01, 0x10};
const uint8x16_t bitmask2 = { 0x02, 0x20, 0x02, 0x20, 0x02, 0x20, 0x02, 0x20,
0x02, 0x20, 0x02, 0x20, 0x02, 0x20, 0x02, 0x20};
const uint8x16_t bitmask3 = { 0x04, 0x40, 0x04, 0x40, 0x04, 0x40, 0x04, 0x40,
0x04, 0x40, 0x04, 0x40, 0x04, 0x40, 0x04, 0x40};
const uint8x16_t bitmask4 = { 0x08, 0x80, 0x08, 0x80, 0x08, 0x80, 0x08, 0x80,
0x08, 0x80, 0x08, 0x80, 0x08, 0x80, 0x08, 0x80};
uint8x16_t t0 = vandq_u8(p0, bitmask1);
uint8x16_t t1 = vbslq_u8(bitmask2, p1, t0);
uint8x16_t t2 = vbslq_u8(bitmask3, p2, t1);
uint8x16_t tmp = vbslq_u8(bitmask4, p3, t2);
uint8x16_t sum = vpaddq_u8(tmp, tmp);
return vgetq_lane_u64(vreinterpretq_u64_u8(sum), 0);
}
```
Results in this emit (excluding table data):
```asm
vmovmaskq_u8_: // @vmovmaskq_u8_
adrp x8, .LCPI0_0
ldr q4, [x8, :lo12:.LCPI0_0]
adrp x8, .LCPI0_1
and v0.16b, v0.16b, v4.16b
ldr q4, [x8, :lo12:.LCPI0_1]
adrp x8, .LCPI0_2
and v1.16b, v1.16b, v4.16b
orr v0.16b, v1.16b, v0.16b
ldr q1, [x8, :lo12:.LCPI0_2]
and v1.16b, v2.16b, v1.16b
movi v2.8h, #8
fneg v2.8h, v2.8h
and v2.16b, v3.16b, v2.16b
orr v1.16b, v1.16b, v2.16b
orr v0.16b, v0.16b, v1.16b
addp v0.16b, v0.16b, v0.16b
fmov x0, d0
ret
```
Using inline assembly, one of these should be right (I don't know which order I am supposed to use):
```asm
vmovmaskq_u8_: // @vmovmaskq_u8_
adrp x8, .LCPI0_0
ldr q4, [x8, :lo12:.LCPI0_0]
adrp x8, .LCPI0_1
and v0.16b, v0.16b, v4.16b
ldr q4, [x8, :lo12:.LCPI0_1]
adrp x8, .LCPI0_2
bsl v0.16b, v1.16b, v4.16b
ldr q1, [x8, :lo12:.LCPI0_2]
adrp x8, .LCPI0_3
bsl v0.16b, v2.16b, v1.16b
ldr q1, [x8, :lo12:.LCPI0_3]
bsl v0.16b, v3.16b, v1.16b
addp v0.16b, v0.16b, v0.16b
fmov x0, d0
ret
vmovmaskq_u8_: // @vmovmaskq_u8_
adrp x8, .LCPI0_0
ldr q4, [x8, :lo12:.LCPI0_0]
adrp x8, .LCPI0_1
and v0.16b, v0.16b, v4.16b
ldr q4, [x8, :lo12:.LCPI0_1]
adrp x8, .LCPI0_2
bsl v4.16b, v1.16b, v0.16b
ldr q1, [x8, :lo12:.LCPI0_2]
adrp x8, .LCPI0_3
bsl v1.16b, v2.16b, v0.16b
ldr q1, [x8, :lo12:.LCPI0_3]
bsl v1.16b, v3.16b, v0.16b
addp v0.16b, v0.16b, v0.16b
fmov x0, d0
ret
```
This movemask routine (Godbolt link):
Results in this emit (excluding table data):
Using inline assembly, one of these should be right (I don't know which order I am supposed to use):