github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

fixes in shifting primitives

Browse Source
This commit is contained in:
Konstantinos Margaritis 2021-02-08 19:38:20 +02:00
parent f541f75400
commit be66cdb51d
1 changed files with 12 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
src/util/arch/arm/simd_utils.h
View File

@ -62,7 +62,7 @@ static really_inline int diff128(m128 a, m128 b) {
} }
static really_inline int isnonzero128(m128 a) { static really_inline int isnonzero128(m128 a) {
return !!diff128(a, zeroes128()); return diff128(a, zeroes128());
} }
/** /**
@ -121,7 +121,6 @@ static really_inline m128 eq64_m128(m128 a, m128 b) {
return (m128) vceqq_u64((int64x2_t)a, (int64x2_t)b); return (m128) vceqq_u64((int64x2_t)a, (int64x2_t)b);
} }
static really_inline u32 movemask128(m128 a) { static really_inline u32 movemask128(m128 a) {
static const uint8x16_t powers = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 }; static const uint8x16_t powers = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
@ -311,22 +310,28 @@ m128 palignr(m128 r, m128 l, int offset) {
static really_really_inline static really_really_inline
m128 rshiftbyte_m128(m128 a, unsigned b) { m128 rshiftbyte_m128(m128 a, unsigned b) {
if (b == 0) {
return a;
}
return palignr(zeroes128(), a, b); return palignr(zeroes128(), a, b);
} }
static really_really_inline static really_really_inline
m128 lshiftbyte_m128(m128 a, unsigned b) { m128 lshiftbyte_m128(m128 a, unsigned b) {
if (b == 0) {
return a;
}
return palignr(a, zeroes128(), 16 - b); return palignr(a, zeroes128(), 16 - b);
} }
static really_inline static really_inline
m128 variable_byte_shift_m128(m128 in, s32 amount) { m128 variable_byte_shift_m128(m128 in, s32 amount) {
assert(amount >= -16 && amount <= 16); assert(amount >= -16 && amount <= 16);
static const uint8x16_t vbs_mask = { 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f }; if (amount < 0) {
const uint8x16_t outside_mask = set1_16x8(0xf0); return palignr_imm(zeroes128(), in, -amount);
} else {
m128 shift_mask = palignr_imm(vbs_mask, outside_mask, 16 - amount); return palignr_imm(in, zeroes128(), 16 - amount);
return vqtbl1q_s8(in, shift_mask); }
} }
#ifdef __cplusplus #ifdef __cplusplus