Implement hardware accelerated CRC32 and SHA1, using them if possible (#3348)
* Implement hardware accelerated CRC32 and SHA1, use them if possible. CRC32's generic function is also replaced with zlib's as it is much more performant than our implementation Full hash of a ~731MB disc took only ~369 ms with this, and the generic CRC32 isn't so far behind at ~659 ms SHA1 should perform 4x faster if the user's CPU supports the SHA instructions. Co-authored-by: YoshiRulz <OSSYoshiRulz@gmail.com> Co-authored-by: Morilli <35152647+Morilli@users.noreply.github.com>
This commit is contained in:
CasualPokePlayer
GitHub
parent
0236a820ec
commit
32e8afcedc
Binary file not shown.
Binary file not shown.
|
|
@ -0,0 +1,17 @@
|
|||
CC = gcc
|
||||
|
||||
CFLAGS = -Wall -Wextra -O3 -flto -fvisibility=internal -fPIC -Icommon
|
||||
LFLAGS = -s -shared
|
||||
|
||||
SRCS = $(wildcard common/*.c) $(wildcard crc32/*.c) $(wildcard sha1/*.c) bizinterface.c
|
||||
|
||||
ifeq ($(OS),Windows_NT)
|
||||
EXT = dll
|
||||
else
|
||||
EXT = so
|
||||
endif
|
||||
|
||||
all: libbizhash
|
||||
|
||||
libbizhash: $(SRCS)
|
||||
$(CC) $(CFLAGS) $(SRCS) -o ../../Assets/dll/libbizhash.$(EXT) $(LFLAGS)
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
LibBizHash is the unmanaged side for BizHawk's hashing.
|
||||
|
||||
CRC32 code is taken from [zlib-ng](https://github.com/zlib-ng/zlib-ng) with massive slashing of code and various tweaks. This code is licensed under the zlib license.
|
||||
SHA1 is code is taken from [SHA-Intrinsics](https://github.com/noloader/SHA-Intrinsics) with some tweaks. This code is under the public domain.
|
||||
|
||||
To build, just do `make` in this directory. Note gcc 10 or later is required (due to missing intrinsics in older gcc versions)
|
||||
|
||||
zlib-ng's license:
|
||||
|
||||
```
|
||||
(C) 1995-2013 Jean-loup Gailly and Mark Adler
|
||||
|
||||
This software is provided 'as-is', without any express or implied warranty. In no event will the authors be held liable for any damages arising from the use of this software.
|
||||
|
||||
Permission is granted to anyone to use this software for any purpose, including commercial applications, and to alter it and redistribute it freely, subject to the following restrictions:
|
||||
|
||||
The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
|
||||
|
||||
Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
|
||||
|
||||
This notice may not be removed or altered from any source distribution.
|
||||
```
|
||||
|
|
@ -0,0 +1,54 @@
|
|||
#include <stdbool.h>
|
||||
#include "common.h"
|
||||
|
||||
__attribute__((visibility("default")))
|
||||
crc32_func BizCalcCrcFunc(void) {
|
||||
x86_check_features();
|
||||
return (x86_cpu_has_pclmulqdq && x86_cpu_has_sse41) ? &crc32_pclmulqdq : &crc32_braid;
|
||||
}
|
||||
|
||||
__attribute__((visibility("default")))
|
||||
bool BizSupportsShaInstructions(void) {
|
||||
x86_check_features();
|
||||
return x86_cpu_has_sha && x86_cpu_has_sse41;
|
||||
}
|
||||
|
||||
__attribute__((visibility("default")))
|
||||
void BizCalcSha1(uint32_t state[5], const uint8_t data[], uint32_t length) {
|
||||
x86_check_features();
|
||||
|
||||
uint64_t bit_length = length * 8ULL;
|
||||
|
||||
// hash most of the data, leaving at most 63 bytes left
|
||||
sha1_sha(state, data, length);
|
||||
data += length & ~0x3F;
|
||||
length &= 0x3F;
|
||||
|
||||
// copy all remaining data to a buffer
|
||||
uint8_t block[64] = {0};
|
||||
memcpy(block, data, length);
|
||||
|
||||
// pad data with '1' bit
|
||||
block[length++] = 0x80;
|
||||
|
||||
// the last 8 bytes in the last block contain the data length;
|
||||
// if the current block is too full hash it and start a new one (here the old one is cleared and re-used)
|
||||
if (__builtin_expect(length > 56, false)) {
|
||||
sha1_sha(state, block, 64);
|
||||
memset(block, 0, 56);
|
||||
}
|
||||
|
||||
// fill the last 8 bytes in the last block with the data length in bits (big endian)
|
||||
for (int i = 0; i != 8; i++) {
|
||||
block[63 - i] = bit_length >> i * 8;
|
||||
}
|
||||
// hash the last block
|
||||
sha1_sha(state, block, 64);
|
||||
|
||||
// byteswap state (to big endian format)
|
||||
state[0] = __builtin_bswap32(state[0]);
|
||||
state[1] = __builtin_bswap32(state[1]);
|
||||
state[2] = __builtin_bswap32(state[2]);
|
||||
state[3] = __builtin_bswap32(state[3]);
|
||||
state[4] = __builtin_bswap32(state[4]);
|
||||
}
|
||||
|
|
@ -0,0 +1,23 @@
|
|||
/* cpu_features.h -- CPU architecture feature check
|
||||
* Copyright (C) 2017 Hans Kristian Rosbach
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
|
||||
#ifndef COMMON_H_
|
||||
#define COMMON_H_
|
||||
|
||||
#include <stdint.h>
|
||||
#include <string.h>
|
||||
|
||||
#include "x86_features.h"
|
||||
|
||||
/* CRC32 */
|
||||
typedef uint32_t (*crc32_func)(uint32_t crc32, const uint8_t *buf, uint32_t len);
|
||||
|
||||
extern uint32_t crc32_braid(uint32_t crc, const uint8_t *buf, uint32_t len);
|
||||
extern uint32_t crc32_pclmulqdq(uint32_t crc32, const uint8_t *buf, uint32_t len);
|
||||
|
||||
/* SHA1 */
|
||||
void sha1_sha(uint32_t state[5], const uint8_t data[], uint32_t length);
|
||||
|
||||
#endif
|
||||
|
|
@ -0,0 +1,63 @@
|
|||
/* x86_features.c - x86 feature check
|
||||
*
|
||||
* Copyright (C) 2013 Intel Corporation. All rights reserved.
|
||||
* Author:
|
||||
* Jim Kukunas
|
||||
*
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
|
||||
#include <cpuid.h>
|
||||
#include <string.h>
|
||||
|
||||
int x86_cpu_has_avx2;
|
||||
int x86_cpu_has_avx512;
|
||||
int x86_cpu_has_avx512vnni;
|
||||
int x86_cpu_has_sse2;
|
||||
int x86_cpu_has_ssse3;
|
||||
int x86_cpu_has_sse41;
|
||||
int x86_cpu_has_sse42;
|
||||
int x86_cpu_has_pclmulqdq;
|
||||
int x86_cpu_has_vpclmulqdq;
|
||||
int x86_cpu_has_tzcnt;
|
||||
int x86_cpu_has_sha;
|
||||
|
||||
void x86_check_features(void) {
|
||||
static int features_checked = 0;
|
||||
if (features_checked)
|
||||
return;
|
||||
|
||||
unsigned eax, ebx, ecx, edx;
|
||||
unsigned maxbasic;
|
||||
|
||||
__cpuid(0, maxbasic, ebx, ecx, edx);
|
||||
__cpuid(1 /*CPU_PROCINFO_AND_FEATUREBITS*/, eax, ebx, ecx, edx);
|
||||
|
||||
x86_cpu_has_sse2 = edx & 0x4000000;
|
||||
x86_cpu_has_ssse3 = ecx & 0x200;
|
||||
x86_cpu_has_sse41 = ecx & 0x80000;
|
||||
x86_cpu_has_sse42 = ecx & 0x100000;
|
||||
x86_cpu_has_pclmulqdq = ecx & 0x2;
|
||||
|
||||
if (maxbasic >= 7) {
|
||||
__cpuid_count(7, 0, eax, ebx, ecx, edx);
|
||||
|
||||
// check BMI1 bit
|
||||
// Reference: https://software.intel.com/sites/default/files/article/405250/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family.pdf
|
||||
x86_cpu_has_tzcnt = ebx & 0x8;
|
||||
// check AVX2 bit
|
||||
x86_cpu_has_avx2 = ebx & 0x20;
|
||||
x86_cpu_has_avx512 = ebx & 0x00010000;
|
||||
x86_cpu_has_avx512vnni = ecx & 0x800;
|
||||
x86_cpu_has_vpclmulqdq = ecx & 0x400;
|
||||
// check SHA bit
|
||||
x86_cpu_has_sha = ebx & 0x20000000;
|
||||
} else {
|
||||
x86_cpu_has_tzcnt = 0;
|
||||
x86_cpu_has_avx2 = 0;
|
||||
x86_cpu_has_avx512 = 0;
|
||||
x86_cpu_has_avx512vnni = 0;
|
||||
x86_cpu_has_vpclmulqdq = 0;
|
||||
x86_cpu_has_sha = 0;
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,23 @@
|
|||
/* x86_features.h -- check for CPU features
|
||||
* Copyright (C) 2013 Intel Corporation Jim Kukunas
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
|
||||
#ifndef X86_FEATURES_H_
|
||||
#define X86_FEATURES_H_
|
||||
|
||||
extern int x86_cpu_has_avx2;
|
||||
extern int x86_cpu_has_avx512;
|
||||
extern int x86_cpu_has_avx512vnni;
|
||||
extern int x86_cpu_has_sse2;
|
||||
extern int x86_cpu_has_ssse3;
|
||||
extern int x86_cpu_has_sse41;
|
||||
extern int x86_cpu_has_sse42;
|
||||
extern int x86_cpu_has_pclmulqdq;
|
||||
extern int x86_cpu_has_vpclmulqdq;
|
||||
extern int x86_cpu_has_tzcnt;
|
||||
extern int x86_cpu_has_sha;
|
||||
|
||||
void x86_check_features(void);
|
||||
|
||||
#endif /* CPU_H_ */
|
||||
|
|
@ -0,0 +1,111 @@
|
|||
/* crc32_braid.c -- compute the CRC-32 of a data stream
|
||||
* Copyright (C) 1995-2022 Mark Adler
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*
|
||||
* This interleaved implementation of a CRC makes use of pipelined multiple
|
||||
* arithmetic-logic units, commonly found in modern CPU cores. It is due to
|
||||
* Kadatch and Jenkins (2010). See doc/crc-doc.1.0.pdf in upstream.
|
||||
*/
|
||||
|
||||
#include "common.h"
|
||||
#include "crc32_braid_tbl.h"
|
||||
|
||||
#define DO1 c = crc_table[(c ^ *buf++) & 0xff] ^ (c >> 8)
|
||||
#define DO8 DO1; DO1; DO1; DO1; DO1; DO1; DO1; DO1
|
||||
|
||||
static uint32_t crc_word(uint64_t data) {
|
||||
unsigned k;
|
||||
for (k = 0; k < sizeof(uint64_t); k++)
|
||||
data = (data >> 8) ^ crc_table[data & 0xff];
|
||||
return (uint32_t)data;
|
||||
}
|
||||
|
||||
__attribute__((visibility("hidden")))
|
||||
uint32_t crc32_braid(uint32_t crc, const uint8_t *buf, uint32_t len) {
|
||||
register uint32_t c;
|
||||
|
||||
c = crc;
|
||||
|
||||
/* If provided enough bytes, do a braided CRC calculation. */
|
||||
if (len >= 5 * sizeof(uint64_t) + sizeof(uint64_t) - 1) {
|
||||
uint64_t blks;
|
||||
uint64_t const *words;
|
||||
unsigned k;
|
||||
|
||||
/* Compute the CRC up to a uint64_t boundary. */
|
||||
while (len && ((uint64_t)buf & (sizeof(uint64_t) - 1)) != 0) {
|
||||
len--;
|
||||
DO1;
|
||||
}
|
||||
|
||||
/* Compute the CRC on as many 5 uint64_t blocks as are available. */
|
||||
blks = len / (5 * sizeof(uint64_t));
|
||||
len -= blks * 5 * sizeof(uint64_t);
|
||||
words = (uint64_t const *)buf;
|
||||
|
||||
uint64_t crc0, word0, comb;
|
||||
uint64_t crc1, word1;
|
||||
uint64_t crc2, word2;
|
||||
uint64_t crc3, word3;
|
||||
uint64_t crc4, word4;
|
||||
|
||||
/* Initialize the CRC for each braid. */
|
||||
crc0 = c;
|
||||
crc1 = 0;
|
||||
crc2 = 0;
|
||||
crc3 = 0;
|
||||
crc4 = 0;
|
||||
|
||||
/* Process the first blks-1 blocks, computing the CRCs on each braid independently. */
|
||||
while (--blks) {
|
||||
/* Load the word for each braid into registers. */
|
||||
word0 = crc0 ^ words[0];
|
||||
word1 = crc1 ^ words[1];
|
||||
word2 = crc2 ^ words[2];
|
||||
word3 = crc3 ^ words[3];
|
||||
word4 = crc4 ^ words[4];
|
||||
|
||||
words += 5;
|
||||
|
||||
/* Compute and update the CRC for each word. The loop should get unrolled. */
|
||||
crc0 = crc_braid_table[0][word0 & 0xff];
|
||||
crc1 = crc_braid_table[0][word1 & 0xff];
|
||||
crc2 = crc_braid_table[0][word2 & 0xff];
|
||||
crc3 = crc_braid_table[0][word3 & 0xff];
|
||||
crc4 = crc_braid_table[0][word4 & 0xff];
|
||||
|
||||
for (k = 1; k < sizeof(uint64_t); k++) {
|
||||
crc0 ^= crc_braid_table[k][(word0 >> (k << 3)) & 0xff];
|
||||
crc1 ^= crc_braid_table[k][(word1 >> (k << 3)) & 0xff];
|
||||
crc2 ^= crc_braid_table[k][(word2 >> (k << 3)) & 0xff];
|
||||
crc3 ^= crc_braid_table[k][(word3 >> (k << 3)) & 0xff];
|
||||
crc4 ^= crc_braid_table[k][(word4 >> (k << 3)) & 0xff];
|
||||
}
|
||||
}
|
||||
|
||||
/* Process the last block, combining the CRCs of the 5 braids at the same time. */
|
||||
comb = crc_word(crc0 ^ words[0]);
|
||||
comb = crc_word(crc1 ^ words[1] ^ comb);
|
||||
comb = crc_word(crc2 ^ words[2] ^ comb);
|
||||
comb = crc_word(crc3 ^ words[3] ^ comb);
|
||||
comb = crc_word(crc4 ^ words[4] ^ comb);
|
||||
|
||||
words += 5;
|
||||
c = comb;
|
||||
|
||||
/* Update the pointer to the remaining bytes to process. */
|
||||
buf = (const uint8_t *)words;
|
||||
}
|
||||
|
||||
/* Complete the computation of the CRC on any remaining bytes. */
|
||||
while (len >= 8) {
|
||||
len -= 8;
|
||||
DO8;
|
||||
}
|
||||
while (len) {
|
||||
len--;
|
||||
DO1;
|
||||
}
|
||||
|
||||
return c;
|
||||
}
|
||||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,16 @@
|
|||
/* crc32_fold.h -- crc32 folding interface
|
||||
* Copyright (C) 2021 Nathan Moinvaziri
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
#ifndef CRC32_FOLD_H_
|
||||
#define CRC32_FOLD_H_
|
||||
|
||||
#define CRC32_FOLD_BUFFER_SIZE (16 * 4)
|
||||
/* sizeof(__m128i) * (4 folds) */
|
||||
|
||||
typedef struct crc32_fold_s {
|
||||
uint8_t fold[CRC32_FOLD_BUFFER_SIZE];
|
||||
uint32_t value;
|
||||
} crc32_fold;
|
||||
|
||||
#endif
|
||||
|
|
@ -0,0 +1,449 @@
|
|||
/*
|
||||
* Compute the CRC32 using a parallelized folding approach with the PCLMULQDQ
|
||||
* instruction.
|
||||
*
|
||||
* A white paper describing this algorithm can be found at:
|
||||
* doc/crc-pclmulqdq.pdf
|
||||
*
|
||||
* Copyright (C) 2013 Intel Corporation. All rights reserved.
|
||||
* Copyright (C) 2016 Marian Beermann (support for initial value)
|
||||
* Authors:
|
||||
* Wajdi Feghali <wajdi.k.feghali@intel.com>
|
||||
* Jim Guilford <james.guilford@intel.com>
|
||||
* Vinodh Gopal <vinodh.gopal@intel.com>
|
||||
* Erdinc Ozturk <erdinc.ozturk@intel.com>
|
||||
* Jim Kukunas <james.t.kukunas@linux.intel.com>
|
||||
*
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
|
||||
#include <immintrin.h>
|
||||
#include <wmmintrin.h>
|
||||
#include <smmintrin.h> // _mm_extract_epi32
|
||||
|
||||
#include <assert.h>
|
||||
|
||||
#include "common.h"
|
||||
#include "crc32_fold.h"
|
||||
|
||||
extern uint64_t fold_16_vpclmulqdq(__m128i *xmm_crc0, __m128i *xmm_crc1,
|
||||
__m128i *xmm_crc2, __m128i *xmm_crc3, const uint8_t *src, uint64_t len, __m128i init_crc,
|
||||
int32_t first);
|
||||
extern uint64_t fold_16_vpclmulqdq_copy(__m128i *xmm_crc0, __m128i *xmm_crc1,
|
||||
__m128i *xmm_crc2, __m128i *xmm_crc3, uint8_t *dst, const uint8_t *src, uint64_t len);
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static void fold_1(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
|
||||
const __m128i xmm_fold4 = _mm_set_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
__m128i x_tmp3;
|
||||
__m128 ps_crc0, ps_crc3, ps_res;
|
||||
|
||||
x_tmp3 = *xmm_crc3;
|
||||
|
||||
*xmm_crc3 = *xmm_crc0;
|
||||
*xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
|
||||
*xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
|
||||
ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
|
||||
ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
|
||||
ps_res = _mm_xor_ps(ps_crc0, ps_crc3);
|
||||
|
||||
*xmm_crc0 = *xmm_crc1;
|
||||
*xmm_crc1 = *xmm_crc2;
|
||||
*xmm_crc2 = x_tmp3;
|
||||
*xmm_crc3 = _mm_castps_si128(ps_res);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static void fold_2(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
|
||||
const __m128i xmm_fold4 = _mm_set_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
__m128i x_tmp3, x_tmp2;
|
||||
__m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3, ps_res31, ps_res20;
|
||||
|
||||
x_tmp3 = *xmm_crc3;
|
||||
x_tmp2 = *xmm_crc2;
|
||||
|
||||
*xmm_crc3 = *xmm_crc1;
|
||||
*xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
|
||||
*xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
|
||||
ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
|
||||
ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
|
||||
ps_res31 = _mm_xor_ps(ps_crc3, ps_crc1);
|
||||
|
||||
*xmm_crc2 = *xmm_crc0;
|
||||
*xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
|
||||
*xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x10);
|
||||
ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
|
||||
ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
|
||||
ps_res20 = _mm_xor_ps(ps_crc0, ps_crc2);
|
||||
|
||||
*xmm_crc0 = x_tmp2;
|
||||
*xmm_crc1 = x_tmp3;
|
||||
*xmm_crc2 = _mm_castps_si128(ps_res20);
|
||||
*xmm_crc3 = _mm_castps_si128(ps_res31);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static void fold_3(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
|
||||
const __m128i xmm_fold4 = _mm_set_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
__m128i x_tmp3;
|
||||
__m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3, ps_res32, ps_res21, ps_res10;
|
||||
|
||||
x_tmp3 = *xmm_crc3;
|
||||
|
||||
*xmm_crc3 = *xmm_crc2;
|
||||
*xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x01);
|
||||
*xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x10);
|
||||
ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
|
||||
ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
|
||||
ps_res32 = _mm_xor_ps(ps_crc2, ps_crc3);
|
||||
|
||||
*xmm_crc2 = *xmm_crc1;
|
||||
*xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
|
||||
*xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x10);
|
||||
ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
|
||||
ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
|
||||
ps_res21 = _mm_xor_ps(ps_crc1, ps_crc2);
|
||||
|
||||
*xmm_crc1 = *xmm_crc0;
|
||||
*xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
|
||||
*xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x10);
|
||||
ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
|
||||
ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
|
||||
ps_res10 = _mm_xor_ps(ps_crc0, ps_crc1);
|
||||
|
||||
*xmm_crc0 = x_tmp3;
|
||||
*xmm_crc1 = _mm_castps_si128(ps_res10);
|
||||
*xmm_crc2 = _mm_castps_si128(ps_res21);
|
||||
*xmm_crc3 = _mm_castps_si128(ps_res32);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static void fold_4(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2, __m128i *xmm_crc3) {
|
||||
const __m128i xmm_fold4 = _mm_set_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
__m128i x_tmp0, x_tmp1, x_tmp2, x_tmp3;
|
||||
__m128 ps_crc0, ps_crc1, ps_crc2, ps_crc3;
|
||||
__m128 ps_t0, ps_t1, ps_t2, ps_t3;
|
||||
__m128 ps_res0, ps_res1, ps_res2, ps_res3;
|
||||
|
||||
x_tmp0 = *xmm_crc0;
|
||||
x_tmp1 = *xmm_crc1;
|
||||
x_tmp2 = *xmm_crc2;
|
||||
x_tmp3 = *xmm_crc3;
|
||||
|
||||
*xmm_crc0 = _mm_clmulepi64_si128(*xmm_crc0, xmm_fold4, 0x01);
|
||||
x_tmp0 = _mm_clmulepi64_si128(x_tmp0, xmm_fold4, 0x10);
|
||||
ps_crc0 = _mm_castsi128_ps(*xmm_crc0);
|
||||
ps_t0 = _mm_castsi128_ps(x_tmp0);
|
||||
ps_res0 = _mm_xor_ps(ps_crc0, ps_t0);
|
||||
|
||||
*xmm_crc1 = _mm_clmulepi64_si128(*xmm_crc1, xmm_fold4, 0x01);
|
||||
x_tmp1 = _mm_clmulepi64_si128(x_tmp1, xmm_fold4, 0x10);
|
||||
ps_crc1 = _mm_castsi128_ps(*xmm_crc1);
|
||||
ps_t1 = _mm_castsi128_ps(x_tmp1);
|
||||
ps_res1 = _mm_xor_ps(ps_crc1, ps_t1);
|
||||
|
||||
*xmm_crc2 = _mm_clmulepi64_si128(*xmm_crc2, xmm_fold4, 0x01);
|
||||
x_tmp2 = _mm_clmulepi64_si128(x_tmp2, xmm_fold4, 0x10);
|
||||
ps_crc2 = _mm_castsi128_ps(*xmm_crc2);
|
||||
ps_t2 = _mm_castsi128_ps(x_tmp2);
|
||||
ps_res2 = _mm_xor_ps(ps_crc2, ps_t2);
|
||||
|
||||
*xmm_crc3 = _mm_clmulepi64_si128(*xmm_crc3, xmm_fold4, 0x01);
|
||||
x_tmp3 = _mm_clmulepi64_si128(x_tmp3, xmm_fold4, 0x10);
|
||||
ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
|
||||
ps_t3 = _mm_castsi128_ps(x_tmp3);
|
||||
ps_res3 = _mm_xor_ps(ps_crc3, ps_t3);
|
||||
|
||||
*xmm_crc0 = _mm_castps_si128(ps_res0);
|
||||
*xmm_crc1 = _mm_castps_si128(ps_res1);
|
||||
*xmm_crc2 = _mm_castps_si128(ps_res2);
|
||||
*xmm_crc3 = _mm_castps_si128(ps_res3);
|
||||
}
|
||||
|
||||
static const unsigned _Alignas(32) pshufb_shf_table[60] = {
|
||||
0x84838281, 0x88878685, 0x8c8b8a89, 0x008f8e8d, /* shl 15 (16 - 1)/shr1 */
|
||||
0x85848382, 0x89888786, 0x8d8c8b8a, 0x01008f8e, /* shl 14 (16 - 3)/shr2 */
|
||||
0x86858483, 0x8a898887, 0x8e8d8c8b, 0x0201008f, /* shl 13 (16 - 4)/shr3 */
|
||||
0x87868584, 0x8b8a8988, 0x8f8e8d8c, 0x03020100, /* shl 12 (16 - 4)/shr4 */
|
||||
0x88878685, 0x8c8b8a89, 0x008f8e8d, 0x04030201, /* shl 11 (16 - 5)/shr5 */
|
||||
0x89888786, 0x8d8c8b8a, 0x01008f8e, 0x05040302, /* shl 10 (16 - 6)/shr6 */
|
||||
0x8a898887, 0x8e8d8c8b, 0x0201008f, 0x06050403, /* shl 9 (16 - 7)/shr7 */
|
||||
0x8b8a8988, 0x8f8e8d8c, 0x03020100, 0x07060504, /* shl 8 (16 - 8)/shr8 */
|
||||
0x8c8b8a89, 0x008f8e8d, 0x04030201, 0x08070605, /* shl 7 (16 - 9)/shr9 */
|
||||
0x8d8c8b8a, 0x01008f8e, 0x05040302, 0x09080706, /* shl 6 (16 -10)/shr10*/
|
||||
0x8e8d8c8b, 0x0201008f, 0x06050403, 0x0a090807, /* shl 5 (16 -11)/shr11*/
|
||||
0x8f8e8d8c, 0x03020100, 0x07060504, 0x0b0a0908, /* shl 4 (16 -12)/shr12*/
|
||||
0x008f8e8d, 0x04030201, 0x08070605, 0x0c0b0a09, /* shl 3 (16 -13)/shr13*/
|
||||
0x01008f8e, 0x05040302, 0x09080706, 0x0d0c0b0a, /* shl 2 (16 -14)/shr14*/
|
||||
0x0201008f, 0x06050403, 0x0a090807, 0x0e0d0c0b /* shl 1 (16 -15)/shr15*/
|
||||
};
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static void partial_fold(const size_t len, __m128i *xmm_crc0, __m128i *xmm_crc1, __m128i *xmm_crc2,
|
||||
__m128i *xmm_crc3, __m128i *xmm_crc_part) {
|
||||
|
||||
const __m128i xmm_fold4 = _mm_set_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
const __m128i xmm_mask3 = _mm_set1_epi32((int32_t)0x80808080);
|
||||
|
||||
__m128i xmm_shl, xmm_shr, xmm_tmp1, xmm_tmp2, xmm_tmp3;
|
||||
__m128i xmm_a0_0, xmm_a0_1;
|
||||
__m128 ps_crc3, psa0_0, psa0_1, ps_res;
|
||||
|
||||
xmm_shl = _mm_load_si128((__m128i *)pshufb_shf_table + (len - 1));
|
||||
xmm_shr = xmm_shl;
|
||||
xmm_shr = _mm_xor_si128(xmm_shr, xmm_mask3);
|
||||
|
||||
xmm_a0_0 = _mm_shuffle_epi8(*xmm_crc0, xmm_shl);
|
||||
|
||||
*xmm_crc0 = _mm_shuffle_epi8(*xmm_crc0, xmm_shr);
|
||||
xmm_tmp1 = _mm_shuffle_epi8(*xmm_crc1, xmm_shl);
|
||||
*xmm_crc0 = _mm_or_si128(*xmm_crc0, xmm_tmp1);
|
||||
|
||||
*xmm_crc1 = _mm_shuffle_epi8(*xmm_crc1, xmm_shr);
|
||||
xmm_tmp2 = _mm_shuffle_epi8(*xmm_crc2, xmm_shl);
|
||||
*xmm_crc1 = _mm_or_si128(*xmm_crc1, xmm_tmp2);
|
||||
|
||||
*xmm_crc2 = _mm_shuffle_epi8(*xmm_crc2, xmm_shr);
|
||||
xmm_tmp3 = _mm_shuffle_epi8(*xmm_crc3, xmm_shl);
|
||||
*xmm_crc2 = _mm_or_si128(*xmm_crc2, xmm_tmp3);
|
||||
|
||||
*xmm_crc3 = _mm_shuffle_epi8(*xmm_crc3, xmm_shr);
|
||||
*xmm_crc_part = _mm_shuffle_epi8(*xmm_crc_part, xmm_shl);
|
||||
*xmm_crc3 = _mm_or_si128(*xmm_crc3, *xmm_crc_part);
|
||||
|
||||
xmm_a0_1 = _mm_clmulepi64_si128(xmm_a0_0, xmm_fold4, 0x10);
|
||||
xmm_a0_0 = _mm_clmulepi64_si128(xmm_a0_0, xmm_fold4, 0x01);
|
||||
|
||||
ps_crc3 = _mm_castsi128_ps(*xmm_crc3);
|
||||
psa0_0 = _mm_castsi128_ps(xmm_a0_0);
|
||||
psa0_1 = _mm_castsi128_ps(xmm_a0_1);
|
||||
|
||||
ps_res = _mm_xor_ps(ps_crc3, psa0_0);
|
||||
ps_res = _mm_xor_ps(ps_res, psa0_1);
|
||||
|
||||
*xmm_crc3 = _mm_castps_si128(ps_res);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static inline void crc32_fold_load(__m128i *fold, __m128i *fold0, __m128i *fold1, __m128i *fold2, __m128i *fold3) {
|
||||
*fold0 = _mm_load_si128(fold + 0);
|
||||
*fold1 = _mm_load_si128(fold + 1);
|
||||
*fold2 = _mm_load_si128(fold + 2);
|
||||
*fold3 = _mm_load_si128(fold + 3);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static inline void crc32_fold_save(__m128i *fold, __m128i fold0, __m128i fold1, __m128i fold2, __m128i fold3) {
|
||||
_mm_storeu_si128(fold + 0, fold0);
|
||||
_mm_storeu_si128(fold + 1, fold1);
|
||||
_mm_storeu_si128(fold + 2, fold2);
|
||||
_mm_storeu_si128(fold + 3, fold3);
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static uint32_t crc32_fold_pclmulqdq_reset(crc32_fold *crc) {
|
||||
__m128i xmm_crc0 = _mm_cvtsi32_si128(0x9db42487);
|
||||
__m128i xmm_zero = _mm_setzero_si128();
|
||||
crc32_fold_save((__m128i *)crc->fold, xmm_crc0, xmm_zero, xmm_zero, xmm_zero);
|
||||
return 0;
|
||||
}
|
||||
|
||||
#define ONCE(op) if (first) { first = 0; op; }
|
||||
#define XOR_INITIAL(where) ONCE(where = _mm_xor_si128(where, xmm_initial))
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
void crc32_fold_pclmulqdq(crc32_fold *crc, const uint8_t *src, uint64_t len, uint32_t init_crc) {
|
||||
size_t algn_diff;
|
||||
__m128i xmm_t0, xmm_t1, xmm_t2, xmm_t3;
|
||||
__m128i xmm_crc0, xmm_crc1, xmm_crc2, xmm_crc3;
|
||||
__m128i xmm_crc_part = _mm_setzero_si128();
|
||||
__m128i xmm_initial = _mm_cvtsi32_si128(init_crc);
|
||||
int32_t first = init_crc != 0;
|
||||
|
||||
/* Technically the CRC functions don't even call this for input < 64, but a bare minimum of 31
|
||||
* bytes of input is needed for the aligning load that occurs. If there's an initial CRC, to
|
||||
* carry it forward through the folded CRC there must be 16 - src % 16 + 16 bytes available, which
|
||||
* by definition can be up to 15 bytes + one full vector load. */
|
||||
assert(len >= 31 || first == 0);
|
||||
crc32_fold_load((__m128i *)crc->fold, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
|
||||
if (len < 16) {
|
||||
goto partial;
|
||||
}
|
||||
|
||||
algn_diff = ((uintptr_t)16 - ((uintptr_t)src & 0xF)) & 0xF;
|
||||
if (algn_diff) {
|
||||
xmm_crc_part = _mm_loadu_si128((__m128i *)src);
|
||||
XOR_INITIAL(xmm_crc_part);
|
||||
|
||||
if (algn_diff < 4 && init_crc != 0) {
|
||||
xmm_t0 = xmm_crc_part;
|
||||
xmm_crc_part = _mm_loadu_si128((__m128i*)src + 1);
|
||||
fold_1(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t0);
|
||||
src += 16;
|
||||
len -= 16;
|
||||
}
|
||||
|
||||
partial_fold(algn_diff, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3, &xmm_crc_part);
|
||||
|
||||
src += algn_diff;
|
||||
len -= algn_diff;
|
||||
}
|
||||
|
||||
if (x86_cpu_has_vpclmulqdq && x86_cpu_has_avx512 && (len >= 256)) {
|
||||
uint64_t n = fold_16_vpclmulqdq(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3, src, len, xmm_initial, first);
|
||||
first = 0;
|
||||
len -= n;
|
||||
src += n;
|
||||
}
|
||||
|
||||
while (len >= 64) {
|
||||
len -= 64;
|
||||
xmm_t0 = _mm_load_si128((__m128i *)src);
|
||||
xmm_t1 = _mm_load_si128((__m128i *)src + 1);
|
||||
xmm_t2 = _mm_load_si128((__m128i *)src + 2);
|
||||
xmm_t3 = _mm_load_si128((__m128i *)src + 3);
|
||||
src += 64;
|
||||
|
||||
fold_4(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
XOR_INITIAL(xmm_t0);
|
||||
|
||||
xmm_crc0 = _mm_xor_si128(xmm_crc0, xmm_t0);
|
||||
xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_t1);
|
||||
xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t2);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t3);
|
||||
}
|
||||
|
||||
// len = num bytes left - 64
|
||||
if (len >= 48) {
|
||||
len -= 48;
|
||||
|
||||
xmm_t0 = _mm_load_si128((__m128i *)src);
|
||||
xmm_t1 = _mm_load_si128((__m128i *)src + 1);
|
||||
xmm_t2 = _mm_load_si128((__m128i *)src + 2);
|
||||
src += 48;
|
||||
XOR_INITIAL(xmm_t0);
|
||||
fold_3(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
|
||||
xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_t0);
|
||||
xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t1);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t2);
|
||||
} else if (len >= 32) {
|
||||
len -= 32;
|
||||
|
||||
xmm_t0 = _mm_load_si128((__m128i *)src);
|
||||
xmm_t1 = _mm_load_si128((__m128i *)src + 1);
|
||||
src += 32;
|
||||
XOR_INITIAL(xmm_t0);
|
||||
fold_2(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
|
||||
xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_t0);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t1);
|
||||
} else if (len >= 16) {
|
||||
len -= 16;
|
||||
xmm_t0 = _mm_load_si128((__m128i *)src);
|
||||
src += 16;
|
||||
XOR_INITIAL(xmm_t0);
|
||||
fold_1(&xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_t0);
|
||||
}
|
||||
|
||||
partial:
|
||||
if (len) {
|
||||
memcpy(&xmm_crc_part, src, (size_t)len);
|
||||
partial_fold((size_t)len, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3, &xmm_crc_part);
|
||||
}
|
||||
|
||||
crc32_fold_save((__m128i *)crc->fold, xmm_crc0, xmm_crc1, xmm_crc2, xmm_crc3);
|
||||
}
|
||||
|
||||
static const unsigned _Alignas(16) crc_k[] = {
|
||||
0xccaa009e, 0x00000000, /* rk1 */
|
||||
0x751997d0, 0x00000001, /* rk2 */
|
||||
0xccaa009e, 0x00000000, /* rk5 */
|
||||
0x63cd6124, 0x00000001, /* rk6 */
|
||||
0xf7011640, 0x00000001, /* rk7 */
|
||||
0xdb710640, 0x00000001 /* rk8 */
|
||||
};
|
||||
|
||||
static const unsigned _Alignas(16) crc_mask[4] = {
|
||||
0xFFFFFFFF, 0xFFFFFFFF, 0x00000000, 0x00000000
|
||||
};
|
||||
|
||||
static const unsigned _Alignas(16) crc_mask2[4] = {
|
||||
0x00000000, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF
|
||||
};
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul")))
|
||||
static uint32_t crc32_fold_pclmulqdq_final(crc32_fold *crc) {
|
||||
const __m128i xmm_mask = _mm_load_si128((__m128i *)crc_mask);
|
||||
const __m128i xmm_mask2 = _mm_load_si128((__m128i *)crc_mask2);
|
||||
__m128i xmm_crc0, xmm_crc1, xmm_crc2, xmm_crc3;
|
||||
__m128i x_tmp0, x_tmp1, x_tmp2, crc_fold;
|
||||
|
||||
crc32_fold_load((__m128i *)crc->fold, &xmm_crc0, &xmm_crc1, &xmm_crc2, &xmm_crc3);
|
||||
|
||||
// k1
|
||||
crc_fold = _mm_load_si128((__m128i *)crc_k);
|
||||
|
||||
x_tmp0 = _mm_clmulepi64_si128(xmm_crc0, crc_fold, 0x10);
|
||||
xmm_crc0 = _mm_clmulepi64_si128(xmm_crc0, crc_fold, 0x01);
|
||||
xmm_crc1 = _mm_xor_si128(xmm_crc1, x_tmp0);
|
||||
xmm_crc1 = _mm_xor_si128(xmm_crc1, xmm_crc0);
|
||||
|
||||
x_tmp1 = _mm_clmulepi64_si128(xmm_crc1, crc_fold, 0x10);
|
||||
xmm_crc1 = _mm_clmulepi64_si128(xmm_crc1, crc_fold, 0x01);
|
||||
xmm_crc2 = _mm_xor_si128(xmm_crc2, x_tmp1);
|
||||
xmm_crc2 = _mm_xor_si128(xmm_crc2, xmm_crc1);
|
||||
|
||||
x_tmp2 = _mm_clmulepi64_si128(xmm_crc2, crc_fold, 0x10);
|
||||
xmm_crc2 = _mm_clmulepi64_si128(xmm_crc2, crc_fold, 0x01);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, x_tmp2);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
|
||||
|
||||
// k5
|
||||
crc_fold = _mm_load_si128((__m128i *)crc_k + 1);
|
||||
|
||||
xmm_crc0 = xmm_crc3;
|
||||
xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0);
|
||||
xmm_crc0 = _mm_srli_si128(xmm_crc0, 8);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc0);
|
||||
|
||||
xmm_crc0 = xmm_crc3;
|
||||
xmm_crc3 = _mm_slli_si128(xmm_crc3, 4);
|
||||
xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0x10);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc0);
|
||||
xmm_crc3 = _mm_and_si128(xmm_crc3, xmm_mask2);
|
||||
|
||||
// k7
|
||||
xmm_crc1 = xmm_crc3;
|
||||
xmm_crc2 = xmm_crc3;
|
||||
crc_fold = _mm_load_si128((__m128i *)crc_k + 2);
|
||||
|
||||
xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
|
||||
xmm_crc3 = _mm_and_si128(xmm_crc3, xmm_mask);
|
||||
|
||||
xmm_crc2 = xmm_crc3;
|
||||
xmm_crc3 = _mm_clmulepi64_si128(xmm_crc3, crc_fold, 0x10);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc2);
|
||||
xmm_crc3 = _mm_xor_si128(xmm_crc3, xmm_crc1);
|
||||
|
||||
crc->value = (uint32_t)_mm_extract_epi32(xmm_crc3, 2);
|
||||
|
||||
return crc->value;
|
||||
}
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul"))) __attribute__((visibility("hidden")))
|
||||
uint32_t crc32_pclmulqdq(uint32_t crc32, const uint8_t *buf, uint32_t len) {
|
||||
// For lens < 64, crc32_braid method is faster.
|
||||
if (len < 64)
|
||||
return crc32_braid(crc32, buf, len);
|
||||
|
||||
crc32_fold _Alignas(16) crc_state;
|
||||
crc32_fold_pclmulqdq_reset(&crc_state);
|
||||
crc32_fold_pclmulqdq(&crc_state, buf, len, crc32 ^ 0xFFFFFFFFU);
|
||||
return crc32_fold_pclmulqdq_final(&crc_state);
|
||||
}
|
||||
|
|
@ -0,0 +1,99 @@
|
|||
/* crc32_fold_vpclmulqdq.c -- VPCMULQDQ-based CRC32 folding implementation.
|
||||
* Copyright Wangyang Guo (wangyang.guo@intel.com)
|
||||
* For conditions of distribution and use, see copyright notice in README.md
|
||||
*/
|
||||
|
||||
#include <immintrin.h>
|
||||
#include <stdint.h>
|
||||
|
||||
#define ONCE(op) if (first) { first = 0; op; }
|
||||
#define XOR_INITIAL(where) ONCE(where = _mm512_xor_si512(where, zmm_initial))
|
||||
|
||||
__attribute__((target("sse4.1", "pclmul", "avx512f", "vpclmulqdq")))
|
||||
uint64_t fold_16_vpclmulqdq(__m128i *xmm_crc0, __m128i *xmm_crc1,
|
||||
__m128i *xmm_crc2, __m128i *xmm_crc3, const uint8_t *src, uint64_t len,
|
||||
__m128i init_crc, int32_t first) {
|
||||
__m512i zmm_initial = _mm512_zextsi128_si512(init_crc);
|
||||
__m512i zmm_t0, zmm_t1, zmm_t2, zmm_t3;
|
||||
__m512i zmm_crc0, zmm_crc1, zmm_crc2, zmm_crc3;
|
||||
__m512i z0, z1, z2, z3;
|
||||
uint64_t len_tmp = len;
|
||||
const __m512i zmm_fold4 = _mm512_set4_epi32(0x00000001, 0x54442bd4, 0x00000001, 0xc6e41596);
|
||||
const __m512i zmm_fold16 = _mm512_set4_epi32(0x00000001, 0x1542778a, 0x00000001, 0x322d1430);
|
||||
|
||||
// zmm register init
|
||||
zmm_crc0 = _mm512_setzero_si512();
|
||||
zmm_t0 = _mm512_loadu_si512((__m512i *)src);
|
||||
XOR_INITIAL(zmm_t0);
|
||||
zmm_crc1 = _mm512_loadu_si512((__m512i *)src + 1);
|
||||
zmm_crc2 = _mm512_loadu_si512((__m512i *)src + 2);
|
||||
zmm_crc3 = _mm512_loadu_si512((__m512i *)src + 3);
|
||||
|
||||
// already have intermediate CRC in xmm registers
|
||||
// fold4 with 4 xmm_crc to get zmm_crc0
|
||||
zmm_crc0 = _mm512_inserti32x4(zmm_crc0, *xmm_crc0, 0);
|
||||
zmm_crc0 = _mm512_inserti32x4(zmm_crc0, *xmm_crc1, 1);
|
||||
zmm_crc0 = _mm512_inserti32x4(zmm_crc0, *xmm_crc2, 2);
|
||||
zmm_crc0 = _mm512_inserti32x4(zmm_crc0, *xmm_crc3, 3);
|
||||
z0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x01);
|
||||
zmm_crc0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x10);
|
||||
zmm_crc0 = _mm512_xor_si512(z0, zmm_crc0);
|
||||
zmm_crc0 = _mm512_xor_si512(zmm_crc0, zmm_t0);
|
||||
|
||||
len -= 256;
|
||||
src += 256;
|
||||
|
||||
// fold-16 loops
|
||||
while (len >= 256) {
|
||||
zmm_t0 = _mm512_loadu_si512((__m512i *)src);
|
||||
zmm_t1 = _mm512_loadu_si512((__m512i *)src + 1);
|
||||
zmm_t2 = _mm512_loadu_si512((__m512i *)src + 2);
|
||||
zmm_t3 = _mm512_loadu_si512((__m512i *)src + 3);
|
||||
|
||||
z0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold16, 0x01);
|
||||
z1 = _mm512_clmulepi64_epi128(zmm_crc1, zmm_fold16, 0x01);
|
||||
z2 = _mm512_clmulepi64_epi128(zmm_crc2, zmm_fold16, 0x01);
|
||||
z3 = _mm512_clmulepi64_epi128(zmm_crc3, zmm_fold16, 0x01);
|
||||
|
||||
zmm_crc0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold16, 0x10);
|
||||
zmm_crc1 = _mm512_clmulepi64_epi128(zmm_crc1, zmm_fold16, 0x10);
|
||||
zmm_crc2 = _mm512_clmulepi64_epi128(zmm_crc2, zmm_fold16, 0x10);
|
||||
zmm_crc3 = _mm512_clmulepi64_epi128(zmm_crc3, zmm_fold16, 0x10);
|
||||
|
||||
zmm_crc0 = _mm512_xor_si512(z0, zmm_crc0);
|
||||
zmm_crc1 = _mm512_xor_si512(z1, zmm_crc1);
|
||||
zmm_crc2 = _mm512_xor_si512(z2, zmm_crc2);
|
||||
zmm_crc3 = _mm512_xor_si512(z3, zmm_crc3);
|
||||
|
||||
zmm_crc0 = _mm512_xor_si512(zmm_crc0, zmm_t0);
|
||||
zmm_crc1 = _mm512_xor_si512(zmm_crc1, zmm_t1);
|
||||
zmm_crc2 = _mm512_xor_si512(zmm_crc2, zmm_t2);
|
||||
zmm_crc3 = _mm512_xor_si512(zmm_crc3, zmm_t3);
|
||||
|
||||
len -= 256;
|
||||
src += 256;
|
||||
}
|
||||
// zmm_crc[0,1,2,3] -> zmm_crc0
|
||||
z0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x01);
|
||||
zmm_crc0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x10);
|
||||
zmm_crc0 = _mm512_xor_si512(z0, zmm_crc0);
|
||||
zmm_crc0 = _mm512_xor_si512(zmm_crc0, zmm_crc1);
|
||||
|
||||
z0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x01);
|
||||
zmm_crc0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x10);
|
||||
zmm_crc0 = _mm512_xor_si512(z0, zmm_crc0);
|
||||
zmm_crc0 = _mm512_xor_si512(zmm_crc0, zmm_crc2);
|
||||
|
||||
z0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x01);
|
||||
zmm_crc0 = _mm512_clmulepi64_epi128(zmm_crc0, zmm_fold4, 0x10);
|
||||
zmm_crc0 = _mm512_xor_si512(z0, zmm_crc0);
|
||||
zmm_crc0 = _mm512_xor_si512(zmm_crc0, zmm_crc3);
|
||||
|
||||
// zmm_crc0 -> xmm_crc[0, 1, 2, 3]
|
||||
*xmm_crc0 = _mm512_extracti32x4_epi32(zmm_crc0, 0);
|
||||
*xmm_crc1 = _mm512_extracti32x4_epi32(zmm_crc0, 1);
|
||||
*xmm_crc2 = _mm512_extracti32x4_epi32(zmm_crc0, 2);
|
||||
*xmm_crc3 = _mm512_extracti32x4_epi32(zmm_crc0, 3);
|
||||
|
||||
return (len_tmp - len); // return n bytes processed
|
||||
}
|
||||
|
|
@ -0,0 +1,193 @@
|
|||
/* Intel SHA extensions using C intrinsics */
|
||||
/* Written and place in public domain by Jeffrey Walton */
|
||||
/* Based on code from Intel, and by Sean Gulley for */
|
||||
/* the miTLS project. */
|
||||
|
||||
#include <immintrin.h>
|
||||
#include "common.h"
|
||||
|
||||
__attribute__((target("sha", "sse4.1")))
|
||||
void sha1_sha(uint32_t state[5], const uint8_t data[], uint32_t length) {
|
||||
__m128i ABCD, ABCD_SAVE, E0, E0_SAVE, E1;
|
||||
__m128i MSG0, MSG1, MSG2, MSG3;
|
||||
const __m128i MASK = _mm_set_epi64x(0x0001020304050607ULL, 0x08090a0b0c0d0e0fULL);
|
||||
|
||||
/* Load initial values */
|
||||
ABCD = _mm_loadu_si128((const __m128i*) state);
|
||||
E0 = _mm_set_epi32(state[4], 0, 0, 0);
|
||||
ABCD = _mm_shuffle_epi32(ABCD, 0x1B);
|
||||
|
||||
while (length >= 64) {
|
||||
/* Save current state */
|
||||
ABCD_SAVE = ABCD;
|
||||
E0_SAVE = E0;
|
||||
|
||||
/* Rounds 0-3 */
|
||||
MSG0 = _mm_loadu_si128((const __m128i*)(data + 0));
|
||||
MSG0 = _mm_shuffle_epi8(MSG0, MASK);
|
||||
E0 = _mm_add_epi32(E0, MSG0);
|
||||
E1 = ABCD;
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 0);
|
||||
|
||||
/* Rounds 4-7 */
|
||||
MSG1 = _mm_loadu_si128((const __m128i*)(data + 16));
|
||||
MSG1 = _mm_shuffle_epi8(MSG1, MASK);
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG1);
|
||||
E0 = ABCD;
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 0);
|
||||
MSG0 = _mm_sha1msg1_epu32(MSG0, MSG1);
|
||||
|
||||
/* Rounds 8-11 */
|
||||
MSG2 = _mm_loadu_si128((const __m128i*)(data + 32));
|
||||
MSG2 = _mm_shuffle_epi8(MSG2, MASK);
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG2);
|
||||
E1 = ABCD;
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 0);
|
||||
MSG1 = _mm_sha1msg1_epu32(MSG1, MSG2);
|
||||
MSG0 = _mm_xor_si128(MSG0, MSG2);
|
||||
|
||||
/* Rounds 12-15 */
|
||||
MSG3 = _mm_loadu_si128((const __m128i*)(data + 48));
|
||||
MSG3 = _mm_shuffle_epi8(MSG3, MASK);
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG3);
|
||||
E0 = ABCD;
|
||||
MSG0 = _mm_sha1msg2_epu32(MSG0, MSG3);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 0);
|
||||
MSG2 = _mm_sha1msg1_epu32(MSG2, MSG3);
|
||||
MSG1 = _mm_xor_si128(MSG1, MSG3);
|
||||
|
||||
/* Rounds 16-19 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG0);
|
||||
E1 = ABCD;
|
||||
MSG1 = _mm_sha1msg2_epu32(MSG1, MSG0);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 0);
|
||||
MSG3 = _mm_sha1msg1_epu32(MSG3, MSG0);
|
||||
MSG2 = _mm_xor_si128(MSG2, MSG0);
|
||||
|
||||
/* Rounds 20-23 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG1);
|
||||
E0 = ABCD;
|
||||
MSG2 = _mm_sha1msg2_epu32(MSG2, MSG1);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 1);
|
||||
MSG0 = _mm_sha1msg1_epu32(MSG0, MSG1);
|
||||
MSG3 = _mm_xor_si128(MSG3, MSG1);
|
||||
|
||||
/* Rounds 24-27 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG2);
|
||||
E1 = ABCD;
|
||||
MSG3 = _mm_sha1msg2_epu32(MSG3, MSG2);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 1);
|
||||
MSG1 = _mm_sha1msg1_epu32(MSG1, MSG2);
|
||||
MSG0 = _mm_xor_si128(MSG0, MSG2);
|
||||
|
||||
/* Rounds 28-31 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG3);
|
||||
E0 = ABCD;
|
||||
MSG0 = _mm_sha1msg2_epu32(MSG0, MSG3);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 1);
|
||||
MSG2 = _mm_sha1msg1_epu32(MSG2, MSG3);
|
||||
MSG1 = _mm_xor_si128(MSG1, MSG3);
|
||||
|
||||
/* Rounds 32-35 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG0);
|
||||
E1 = ABCD;
|
||||
MSG1 = _mm_sha1msg2_epu32(MSG1, MSG0);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 1);
|
||||
MSG3 = _mm_sha1msg1_epu32(MSG3, MSG0);
|
||||
MSG2 = _mm_xor_si128(MSG2, MSG0);
|
||||
|
||||
/* Rounds 36-39 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG1);
|
||||
E0 = ABCD;
|
||||
MSG2 = _mm_sha1msg2_epu32(MSG2, MSG1);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 1);
|
||||
MSG0 = _mm_sha1msg1_epu32(MSG0, MSG1);
|
||||
MSG3 = _mm_xor_si128(MSG3, MSG1);
|
||||
|
||||
/* Rounds 40-43 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG2);
|
||||
E1 = ABCD;
|
||||
MSG3 = _mm_sha1msg2_epu32(MSG3, MSG2);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 2);
|
||||
MSG1 = _mm_sha1msg1_epu32(MSG1, MSG2);
|
||||
MSG0 = _mm_xor_si128(MSG0, MSG2);
|
||||
|
||||
/* Rounds 44-47 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG3);
|
||||
E0 = ABCD;
|
||||
MSG0 = _mm_sha1msg2_epu32(MSG0, MSG3);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 2);
|
||||
MSG2 = _mm_sha1msg1_epu32(MSG2, MSG3);
|
||||
MSG1 = _mm_xor_si128(MSG1, MSG3);
|
||||
|
||||
/* Rounds 48-51 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG0);
|
||||
E1 = ABCD;
|
||||
MSG1 = _mm_sha1msg2_epu32(MSG1, MSG0);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 2);
|
||||
MSG3 = _mm_sha1msg1_epu32(MSG3, MSG0);
|
||||
MSG2 = _mm_xor_si128(MSG2, MSG0);
|
||||
|
||||
/* Rounds 52-55 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG1);
|
||||
E0 = ABCD;
|
||||
MSG2 = _mm_sha1msg2_epu32(MSG2, MSG1);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 2);
|
||||
MSG0 = _mm_sha1msg1_epu32(MSG0, MSG1);
|
||||
MSG3 = _mm_xor_si128(MSG3, MSG1);
|
||||
|
||||
/* Rounds 56-59 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG2);
|
||||
E1 = ABCD;
|
||||
MSG3 = _mm_sha1msg2_epu32(MSG3, MSG2);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 2);
|
||||
MSG1 = _mm_sha1msg1_epu32(MSG1, MSG2);
|
||||
MSG0 = _mm_xor_si128(MSG0, MSG2);
|
||||
|
||||
/* Rounds 60-63 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG3);
|
||||
E0 = ABCD;
|
||||
MSG0 = _mm_sha1msg2_epu32(MSG0, MSG3);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 3);
|
||||
MSG2 = _mm_sha1msg1_epu32(MSG2, MSG3);
|
||||
MSG1 = _mm_xor_si128(MSG1, MSG3);
|
||||
|
||||
/* Rounds 64-67 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG0);
|
||||
E1 = ABCD;
|
||||
MSG1 = _mm_sha1msg2_epu32(MSG1, MSG0);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 3);
|
||||
MSG3 = _mm_sha1msg1_epu32(MSG3, MSG0);
|
||||
MSG2 = _mm_xor_si128(MSG2, MSG0);
|
||||
|
||||
/* Rounds 68-71 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG1);
|
||||
E0 = ABCD;
|
||||
MSG2 = _mm_sha1msg2_epu32(MSG2, MSG1);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 3);
|
||||
MSG3 = _mm_xor_si128(MSG3, MSG1);
|
||||
|
||||
/* Rounds 72-75 */
|
||||
E0 = _mm_sha1nexte_epu32(E0, MSG2);
|
||||
E1 = ABCD;
|
||||
MSG3 = _mm_sha1msg2_epu32(MSG3, MSG2);
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E0, 3);
|
||||
|
||||
/* Rounds 76-79 */
|
||||
E1 = _mm_sha1nexte_epu32(E1, MSG3);
|
||||
E0 = ABCD;
|
||||
ABCD = _mm_sha1rnds4_epu32(ABCD, E1, 3);
|
||||
|
||||
/* Combine state */
|
||||
E0 = _mm_sha1nexte_epu32(E0, E0_SAVE);
|
||||
ABCD = _mm_add_epi32(ABCD, ABCD_SAVE);
|
||||
|
||||
data += 64;
|
||||
length -= 64;
|
||||
}
|
||||
|
||||
/* Save state */
|
||||
ABCD = _mm_shuffle_epi32(ABCD, 0x1B);
|
||||
_mm_storeu_si128((__m128i*) state, ABCD);
|
||||
state[4] = _mm_extract_epi32(E0, 3);
|
||||
}
|
||||
|
|
@ -1,4 +1,5 @@
|
|||
using System;
|
||||
using System.Runtime.InteropServices;
|
||||
|
||||
namespace BizHawk.Common
|
||||
{
|
||||
|
|
@ -8,25 +9,18 @@ namespace BizHawk.Common
|
|||
/// <remarks>coefficients of the polynomial, in the format Wikipedia calls "reversed"</remarks>
|
||||
public const uint POLYNOMIAL_CONST = 0xEDB88320U;
|
||||
|
||||
private static readonly uint[] COMBINER_INIT_STATE;
|
||||
/// <summary>
|
||||
/// Delegate to unmanaged code that actually does the calculation.
|
||||
/// This may be hardware accelerated, if the CPU supports such.
|
||||
/// </summary>
|
||||
private static readonly LibBizHash.CalcCRC _calcCRC;
|
||||
|
||||
private static readonly uint[] CRC32Table;
|
||||
private static readonly uint[] COMBINER_INIT_STATE;
|
||||
|
||||
static CRC32()
|
||||
{
|
||||
// for Add (CRC32 computation):
|
||||
CRC32Table = new uint[256];
|
||||
for (var i = 0U; i < 256U; i++)
|
||||
{
|
||||
var crc = i;
|
||||
for (var j = 0; j < 8; j++)
|
||||
{
|
||||
var xor = (crc & 1U) == 1U;
|
||||
crc >>= 1;
|
||||
if (xor) crc ^= POLYNOMIAL_CONST;
|
||||
}
|
||||
CRC32Table[i] = crc;
|
||||
}
|
||||
_calcCRC = Marshal.GetDelegateForFunctionPointer<LibBizHash.CalcCRC>(LibBizHash.BizCalcCrcFunc());
|
||||
|
||||
// for Incorporate:
|
||||
var combinerState = (COMBINER_INIT_STATE = new uint[64]).AsSpan();
|
||||
|
|
@ -80,17 +74,11 @@ namespace BizHawk.Common
|
|||
/// <summary>The negated output (the typical result of the CRC calculation)</summary>
|
||||
public uint Result => ~_current;
|
||||
|
||||
public void Add(byte datum)
|
||||
public unsafe void Add(ReadOnlySpan<byte> data)
|
||||
{
|
||||
_current = CRC32Table[(_current ^ datum) & 0xFF] ^ (_current >> 8);
|
||||
}
|
||||
|
||||
public void Add(ReadOnlySpan<byte> data)
|
||||
{
|
||||
foreach (var b in data)
|
||||
fixed (byte* d = &data.GetPinnableReference())
|
||||
{
|
||||
// Add(b); // I assume this would be slower
|
||||
_current = CRC32Table[(_current ^ b) & 0xFF] ^ (_current >> 8);
|
||||
_current = _calcCRC(_current, (IntPtr) d, data.Length);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,22 @@
|
|||
using System;
|
||||
using System.Runtime.InteropServices;
|
||||
|
||||
namespace BizHawk.Common
|
||||
{
|
||||
public static class LibBizHash
|
||||
{
|
||||
private const CallingConvention cc = CallingConvention.Cdecl;
|
||||
|
||||
[UnmanagedFunctionPointer(cc)]
|
||||
public delegate uint CalcCRC(uint current, IntPtr buffer, int len);
|
||||
|
||||
[DllImport("libbizhash", CallingConvention = cc)]
|
||||
public static extern IntPtr BizCalcCrcFunc();
|
||||
|
||||
[DllImport("libbizhash", CallingConvention = cc)]
|
||||
public static extern bool BizSupportsShaInstructions();
|
||||
|
||||
[DllImport("libbizhash", CallingConvention = cc)]
|
||||
public static extern void BizCalcSha1(IntPtr state, byte[] data, int len);
|
||||
}
|
||||
}
|
||||
|
|
@ -1,11 +1,46 @@
|
|||
using System;
|
||||
using System.Diagnostics;
|
||||
using System.Runtime.InteropServices;
|
||||
using System.Security.Cryptography;
|
||||
|
||||
using BizHawk.Common.BufferExtensions;
|
||||
|
||||
namespace BizHawk.Common
|
||||
{
|
||||
public interface ISHA1
|
||||
{
|
||||
byte[] ComputeHash(byte[] buffer);
|
||||
}
|
||||
|
||||
public sealed class NETSHA1 : ISHA1
|
||||
{
|
||||
private readonly SHA1 _sha1Impl;
|
||||
|
||||
public NETSHA1()
|
||||
{
|
||||
_sha1Impl = SHA1.Create();
|
||||
Debug.Assert(_sha1Impl.CanReuseTransform && _sha1Impl.HashSize is SHA1Checksum.EXPECTED_LENGTH);
|
||||
}
|
||||
|
||||
public byte[] ComputeHash(byte[] buffer)
|
||||
=> _sha1Impl.ComputeHash(buffer);
|
||||
}
|
||||
|
||||
public sealed class FastSHA1 : ISHA1
|
||||
{
|
||||
public unsafe byte[] ComputeHash(byte[] buffer)
|
||||
{
|
||||
// Set SHA1 start state
|
||||
var state = stackalloc uint[] { 0x67452301, 0xEFCDAB89, 0x98BADCFE, 0x10325476, 0xC3D2E1F0 };
|
||||
// This will use dedicated SHA instructions, which perform 4x faster than a generic implementation
|
||||
LibBizHash.BizCalcSha1((IntPtr)state, buffer, buffer.Length);
|
||||
// The copy seems wasteful, but pinning the state down actually has a bigger performance impact
|
||||
var ret = new byte[20];
|
||||
Marshal.Copy((IntPtr)state, ret, 0, 20);
|
||||
return ret;
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>uses <see cref="SHA1"/> implementation from BCL</summary>
|
||||
/// <seealso cref="CRC32Checksum"/>
|
||||
/// <seealso cref="MD5Checksum"/>
|
||||
|
|
@ -35,16 +70,17 @@ namespace BizHawk.Common
|
|||
return impl.GetHashAndReset();
|
||||
}
|
||||
#else
|
||||
private static SHA1? _sha1Impl;
|
||||
private static ISHA1? _sha1Impl;
|
||||
|
||||
private static SHA1 SHA1Impl
|
||||
private static ISHA1 SHA1Impl
|
||||
{
|
||||
get
|
||||
{
|
||||
if (_sha1Impl == null)
|
||||
{
|
||||
_sha1Impl = SHA1.Create();
|
||||
Debug.Assert(_sha1Impl.CanReuseTransform && _sha1Impl.HashSize is EXPECTED_LENGTH);
|
||||
_sha1Impl = LibBizHash.BizSupportsShaInstructions()
|
||||
? new FastSHA1()
|
||||
: new NETSHA1();
|
||||
}
|
||||
return _sha1Impl;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -0,0 +1,57 @@
|
|||
using System;
|
||||
using System.Linq;
|
||||
using System.Text;
|
||||
using BizHawk.Common;
|
||||
|
||||
using Microsoft.VisualStudio.TestTools.UnitTesting;
|
||||
|
||||
namespace BizHawk.Tests.Common.checksums
|
||||
{
|
||||
[TestClass]
|
||||
public sealed class SHA1Tests
|
||||
{
|
||||
[TestMethod]
|
||||
public void TestSHA1Empty()
|
||||
{
|
||||
byte[] data = Array.Empty<byte>(); // empty data
|
||||
byte[] expectedSha = { 0xda, 0x39, 0xa3, 0xee, 0x5e, 0x6b, 0x4b, 0x0d, 0x32, 0x55, 0xbf, 0xef, 0x95, 0x60, 0x18, 0x90, 0xaf, 0xd8, 0x07, 0x09 };
|
||||
|
||||
Assert.IsTrue(expectedSha.SequenceEqual(SHA1Checksum.Compute(data)));
|
||||
}
|
||||
|
||||
[TestMethod]
|
||||
public void TestSHA1Simple()
|
||||
{
|
||||
byte[] data = { (byte)'h', (byte)'a', (byte)'s', (byte)'h' }; // random short data
|
||||
byte[] expectedSha = { 0x23, 0x46, 0xad, 0x27, 0xd7, 0x56, 0x8b, 0xa9, 0x89, 0x6f, 0x1b, 0x7d, 0xa6, 0xb5, 0x99, 0x12, 0x51, 0xde, 0xbd, 0xf2 };
|
||||
|
||||
Assert.IsTrue(expectedSha.SequenceEqual(SHA1Checksum.Compute(data)));
|
||||
Assert.IsTrue(expectedSha.SequenceEqual(SHA1Checksum.ComputeConcat(Array.Empty<byte>(), data)));
|
||||
Assert.IsTrue(expectedSha.SequenceEqual(SHA1Checksum.ComputeConcat(data, Array.Empty<byte>())));
|
||||
|
||||
data = new[] { (byte)'h', (byte)'a' };
|
||||
byte[] data2 = { (byte)'s', (byte)'h' };
|
||||
|
||||
Assert.IsTrue(expectedSha.SequenceEqual(SHA1Checksum.ComputeConcat(data, data2)));
|
||||
}
|
||||
|
||||
[TestMethod]
|
||||
public void TestSHA1LessSimple()
|
||||
{
|
||||
const string testString = "The quick brown fox jumps over the lazy dog.";
|
||||
byte[] data = Encoding.ASCII.GetBytes(testString);
|
||||
byte[] expectedSha1 = { 0x40, 0x8d, 0x94, 0x38, 0x42, 0x16, 0xf8, 0x90, 0xff, 0x7a, 0x0c, 0x35, 0x28, 0xe8, 0xbe, 0xd1, 0xe0, 0xb0, 0x16, 0x21 };
|
||||
|
||||
Assert.IsTrue(expectedSha1.SequenceEqual(SHA1Checksum.Compute(data)));
|
||||
|
||||
data = new byte[65];
|
||||
Encoding.ASCII.GetBytes(testString).CopyTo(data, 0);
|
||||
|
||||
byte[] expectedSha2 = { 0x65, 0x87, 0x84, 0xE2, 0x68, 0xBF, 0xB1, 0x67, 0x94, 0x7B, 0xB7, 0xF3, 0xFB, 0x76, 0x69, 0x62, 0x79, 0x3E, 0x8C, 0x46 };
|
||||
Assert.IsTrue(expectedSha2.SequenceEqual(SHA1Checksum.Compute(new Span<byte>(data, 0, 64))));
|
||||
|
||||
byte[] expectedSha3 = { 0x34, 0xF3, 0xA2, 0x57, 0xBD, 0x12, 0x5E, 0x6E, 0x0E, 0x28, 0xD0, 0xE5, 0xDA, 0xBE, 0x22, 0x28, 0x97, 0xFA, 0x69, 0x55 };
|
||||
Assert.IsTrue(expectedSha3.SequenceEqual(SHA1Checksum.Compute(data)));
|
||||
}
|
||||
}
|
||||
}
|
||||
Loading…
Reference in New Issue