lib/zstd
1
0
Fork 0
mirror of https://github.com/facebook/zstd.git synced 2025-09-08 13:32:28 +03:00
Code Activity

AVX2 version of ZSTD_get1BlockSummary()

Browse Source
This commit is contained in:
Yann Collet
2025-01-07 23:32:05 -08:00
committed by Yann Collet
parent b6a4d5a8ba
commit ed0a8b8be1
1 changed files with 39 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
lib/compress/zstd_compress.c
View File

@@ -7395,56 +7395,65 @@ size_t ZSTD_convertBlockSequences(ZSTD_CCtx* cctx,
return ZSTD_convertBlockSequences_internal(cctx, inSeqs, nbSequences, 0);
}
#if 0 && defined(__AVX2__)
#if defined(__AVX2__)
/* C90-compatible alignment macro (GCC/Clang). Adjust for other compilers if needed. */
#if defined(__GNUC__)
# define ALIGNED32 __attribute__((aligned(32)))
#elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */
# define ALIGNED32 alignas(32)
#else
/* this compiler will require its own alignment instruction */
# define ALIGNED32
#endif
BlockSummary ZSTD_get1BlockSummary(const ZSTD_Sequence* seqs, size_t nbSeqs)
{
size_t i;
__m256i sumVec; /* accumulates match+lit in 32-bit lanes */
__m256i mask; /* shuffling control */
ALIGNED32 int tmp[8]; /* temporary buffer for reduction */
uint64_t sum;
int k;
sumVec = _mm256_setzero_si256();
mask = _mm256_setr_epi32(
1,5, /* match(0), match(1) */
2,6, /* lit(0), lit(1) */
1,5, /* match(0), match(1) */
2,6 /* lit(0), lit(1) */
);
__m256i const zeroVec = _mm256_setzero_si256();
__m256i sumVec = zeroVec; /* accumulates match+lit in 32-bit lanes */
__m256i shuffle32; /* shuffling control */
ALIGNED32 U32 tmp[8]; /* temporary buffer for reduction */
size_t mSum = 0, lSum = 0;
/* Process 2 structs (32 bytes) at a time */
for (i = 0; i + 2 <= count; i += 2) {
/* Load two consecutive MyStructs (8×4 = 32 bytes) */
__m256i data = _mm256_loadu_si256((const __m256i*)&arr[i]);
/* Shuffle out lanes 1,2,5,6 => match(0), match(1), lit(0), lit(1), repeated */
__m256i selected = _mm256_permutevar8x32_epi32(data, mask);
for (i = 0; i + 2 <= nbSeqs; i += 2) {
/* Load two consecutive ZSTD_Sequence (8×4 = 32 bytes) */
__m256i data = _mm256_loadu_si256((const __m256i*)&seqs[i]);
/* check end of block signal */
__m256i cmp = _mm256_cmpeq_epi32(data, zeroVec);
int cmp_res = _mm256_movemask_epi8(cmp);
/* indices for match lengths correspond to bits [8..11], [24..27]
* => combined mask = 0x0F000F00 */
if (cmp_res & 0x0F000F00) break;
/* Accumulate in sumVec */
sumVec = _mm256_add_epi32(sumVec, selected);
sumVec = _mm256_add_epi32(sumVec, data);
}
/* Horizontal reduction of sumVec */
/* Horizontal reduction */
_mm256_store_si256((__m256i*)tmp, sumVec);
sum = 0;
for (k = 0; k < 8; k++) {
sum += (uint64_t)tmp[k]; /* each lane is match+lit from pairs, repeated twice */
lSum = tmp[1] + tmp[5];
mSum = tmp[2] + tmp[6];
/* Handle the leftover */
for (; i < nbSeqs; i++) {
lSum += seqs[i].litLength;
mSum += seqs[i].matchLength;
if (seqs[i].matchLength == 0) break; /* end of block */
}
/* Handle the leftover (if count is odd) */
for (; i < count; i++) {
sum += arr[i].matchLength;
sum += arr[i].litLength;
if (i==nbSeqs) {
/* reaching end of sequences: end of block signal was not present */
BlockSummary bs;
bs.nbSequences = ERROR(externalSequences_invalid);
return bs;
}
{ BlockSummary bs;
bs.nbSequences = i+1;
bs.blockSize = lSum + mSum;
bs.litSize = lSum;
return bs;
}
return sum;
}
#else