68b2e7f763
Some checks failed
sm-rpc / build (Debug, arm-linux-gnueabihf) (push) Successful in 1m34s
sm-rpc / build (Debug, aarch64-linux-gnu) (push) Successful in 2m46s
sm-rpc / build (Debug, host.gcc) (push) Failing after 1m28s
sm-rpc / build (Release, aarch64-linux-gnu) (push) Successful in 2m14s
sm-rpc / build (Release, arm-linux-gnueabihf) (push) Successful in 2m8s
sm-rpc / build (Debug, mipsel-linux-gnu) (push) Successful in 5m35s
sm-rpc / build (Release, host.gcc) (push) Failing after 1m55s
sm-rpc / build (Release, mipsel-linux-gnu) (push) Successful in 7m21s
33 lines
1.3 KiB
C
33 lines
1.3 KiB
C
/* adler32_avx2_p.h -- adler32 avx2 utility functions
|
|
* Copyright (C) 2022 Adam Stylinski
|
|
* For conditions of distribution and use, see copyright notice in zlib.h
|
|
*/
|
|
|
|
#ifndef ADLER32_AVX2_P_H_
|
|
#define ADLER32_AVX2_P_H_
|
|
|
|
#if defined(X86_AVX2) || defined(X86_AVX512VNNI)
|
|
|
|
/* 32 bit horizontal sum, adapted from Agner Fog's vector library. */
|
|
static inline uint32_t hsum256(__m256i x) {
|
|
__m128i sum1 = _mm_add_epi32(_mm256_extracti128_si256(x, 1),
|
|
_mm256_castsi256_si128(x));
|
|
__m128i sum2 = _mm_add_epi32(sum1, _mm_unpackhi_epi64(sum1, sum1));
|
|
__m128i sum3 = _mm_add_epi32(sum2, _mm_shuffle_epi32(sum2, 1));
|
|
return (uint32_t)_mm_cvtsi128_si32(sum3);
|
|
}
|
|
|
|
static inline uint32_t partial_hsum256(__m256i x) {
|
|
/* We need a permutation vector to extract every other integer. The
|
|
* rest are going to be zeros */
|
|
const __m256i perm_vec = _mm256_setr_epi32(0, 2, 4, 6, 1, 1, 1, 1);
|
|
__m256i non_zero = _mm256_permutevar8x32_epi32(x, perm_vec);
|
|
__m128i non_zero_sse = _mm256_castsi256_si128(non_zero);
|
|
__m128i sum2 = _mm_add_epi32(non_zero_sse,_mm_unpackhi_epi64(non_zero_sse, non_zero_sse));
|
|
__m128i sum3 = _mm_add_epi32(sum2, _mm_shuffle_epi32(sum2, 1));
|
|
return (uint32_t)_mm_cvtsi128_si32(sum3);
|
|
}
|
|
#endif
|
|
|
|
#endif
|