aaru-dps/Aaru.Checksums.Native
1
0
Fork 0
mirror of https://github.com/aaru-dps/Aaru.Checksums.Native.git synced 2025-12-16 19:24:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

General refactor and cleanup.

Browse Source
This commit is contained in:
Natalia Portillo
2024-04-30 15:12:48 +01:00
parent 552aa9da02
commit bd5051ce18
48 changed files with 1157 additions and 1290 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
fletcher16_avx2.c
View File

@@ -42,8 +42,7 @@
* @param data Pointer to the data buffer.
* @param len Length of the data buffer in bytes.
*/
AARU_EXPORT TARGET_WITH_AVX2 void AARU_CALL
fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
AARU_EXPORT TARGET_WITH_AVX2 void AARU_CALL fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
{
uint32_t s1 = *sum1;
uint32_t s2 = *sum2;
@@ -54,7 +53,7 @@ fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
const unsigned BLOCK_SIZE = 1 << 5;
if(len >= BLOCK_SIZE)
{
long blocks = len / BLOCK_SIZE;
long blocks = len / BLOCK_SIZE;
len -= blocks * BLOCK_SIZE;
while(blocks)
@@ -64,38 +63,8 @@ fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
if(n > blocks) n = (unsigned)blocks;
blocks -= n;
const __m256i tap = _mm256_set_epi8(1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32);
const __m256i tap = _mm256_set_epi8(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20,
21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32);
const __m256i zero = _mm256_setzero_si256();
const __m256i ones = _mm256_set1_epi16(1);
@@ -106,8 +75,7 @@ fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
__m256i v_ps = _mm256_set_epi32(0, 0, 0, 0, 0, 0, 0, (s1 * n));
__m256i v_s2 = _mm256_set_epi32(0, 0, 0, 0, 0, 0, 0, s2);
__m256i v_s1 = _mm256_setzero_si256();
do
{
do {
/*
* Load 32 input bytes.
*/
@@ -116,24 +84,23 @@ fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
/*
* Add previous block byte sum to v_ps.
*/
v_ps = _mm256_add_epi32(v_ps, v_s1);
v_ps = _mm256_add_epi32(v_ps, v_s1);
/*
* Horizontally add the bytes for s1, multiply-adds the
* bytes by [ 32, 31, 30, ... ] for s2.
*/
v_s1 = _mm256_add_epi32(v_s1, _mm256_sad_epu8(bytes, zero));
v_s1 = _mm256_add_epi32(v_s1, _mm256_sad_epu8(bytes, zero));
const __m256i mad = _mm256_maddubs_epi16(bytes, tap);
v_s2 = _mm256_add_epi32(v_s2, _mm256_madd_epi16(mad, ones));
v_s2 = _mm256_add_epi32(v_s2, _mm256_madd_epi16(mad, ones));
data += BLOCK_SIZE;
}
while(--n);
} while(--n);
__m128i sum = _mm_add_epi32(_mm256_castsi256_si128(v_s1), _mm256_extracti128_si256(v_s1, 1));
__m128i hi = _mm_unpackhi_epi64(sum, sum);
sum = _mm_add_epi32(hi, sum);
hi = _mm_shuffle_epi32(sum, 177);
sum = _mm_add_epi32(sum, hi);
sum = _mm_add_epi32(hi, sum);
hi = _mm_shuffle_epi32(sum, 177);
sum = _mm_add_epi32(sum, hi);
s1 += _mm_cvtsi128_si32(sum);
v_s2 = _mm256_add_epi32(v_s2, _mm256_slli_epi32(v_ps, 5));
@@ -177,8 +144,7 @@ fletcher16_avx2(uint8_t *sum1, uint8_t *sum2, const uint8_t *data, long len)
s2 += (s1 += *data++);
len -= 16;
}
while(len--)
{ s2 += (s1 += *data++); }
while(len--) { s2 += (s1 += *data++); }
s1 %= FLETCHER16_MODULE;
s2 %= FLETCHER16_MODULE;
}