Skip to content

Commit 16240f4

Browse files
laanwjlaanwj
committed
Merge #10821: Add SSE4 optimized SHA256
6b8d872 Protect SSE4 code behind a compile-time flag (Pieter Wuille) fa9be90 Add selftest for SHA256 transform (Pieter Wuille) c1ccb15 Add SSE4 based SHA256 (Pieter Wuille) 2991c91 Add SHA256 dispatcher (Pieter Wuille) 4d50f38 Support multi-block SHA256 transforms (Pieter Wuille) Pull request description: This adds an SSE4 assembly version of the SHA256 transform by Intel, and uses it at run time if SSE4 instructions are available, and use a fallback C++ implementation otherwise. Nearly every x86_64 CPU supports SSE4. The feature is only enabled when compiled with `--enable-experimental-asm`. In order to avoid build dependencies and other complications, the original Intel YASM code was translated to GCC extended asm syntax. This gives around a 50% speedup on the SHA256 benchmark for me. It is based on an earlier patch by @laanwj, though only includes a single assembly version (for now), and removes the YASM dependency. Tree-SHA512: d31c50695ceb45264291537b93c0d7497670be38edf021ca5402eaa7d4e1e0e1ae492326e28d4e93979d066168129e62d1825e0384b1b906d36f85d93dfcb43c
2 parents 7c2400c + 6b8d872 commit 16240f4

File tree

8 files changed

+1681
-88
lines changed

8 files changed

+1681
-88
lines changed

‎configure.ac

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -177,6 +177,16 @@ AC_ARG_ENABLE([glibc-back-compat],
177177
[use_glibc_compat=$enableval],
178178
[use_glibc_compat=no])
179179

180+
AC_ARG_ENABLE([experimental-asm],
181+
[AS_HELP_STRING([--enable-experimental-asm],
182+
[Enable experimental assembly routines (default is no)])],
183+
[experimental_asm=$enableval],
184+
[experimental_asm=no])
185+
186+
if test "x$experimental_asm" = xyes; then
187+
AC_DEFINE(EXPERIMENTAL_ASM, 1, [Define this symbol to build in experimental assembly routines])
188+
fi
189+
180190
AC_ARG_WITH([system-univalue],
181191
[AS_HELP_STRING([--with-system-univalue],
182192
[Build with system UniValue (default is no)])],
@@ -1162,6 +1172,7 @@ AM_CONDITIONAL([USE_LCOV],[test x$use_lcov = xyes])
11621172
AM_CONDITIONAL([GLIBC_BACK_COMPAT],[test x$use_glibc_compat = xyes])
11631173
AM_CONDITIONAL([HARDEN],[test x$use_hardening = xyes])
11641174
AM_CONDITIONAL([ENABLE_HWCRC32],[test x$enable_hwcrc32 = xyes])
1175+
AM_CONDITIONAL([EXPERIMENTAL_ASM],[test x$experimental_asm = xyes])
11651176

11661177
AC_DEFINE(CLIENT_VERSION_MAJOR, _CLIENT_VERSION_MAJOR, [Major version])
11671178
AC_DEFINE(CLIENT_VERSION_MINOR, _CLIENT_VERSION_MINOR, [Minor version])

‎src/Makefile.am

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -267,6 +267,10 @@ crypto_libbitcoin_crypto_a_SOURCES = \
267267
crypto/sha512.cpp \
268268
crypto/sha512.h
269269

270+
if EXPERIMENTAL_ASM
271+
crypto_libbitcoin_crypto_a_SOURCES += crypto/sha256_sse4.cpp
272+
endif
273+
270274
# consensus: shared between all executables that validate any consensus rules.
271275
libbitcoin_consensus_a_CPPFLAGS = $(AM_CPPFLAGS) $(BITCOIN_INCLUDES)
272276
libbitcoin_consensus_a_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS)

‎src/bench/bench_bitcoin.cpp

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,7 @@
44

55
#include "bench.h"
66

7+
#include "crypto/sha256.h"
78
#include "key.h"
89
#include "validation.h"
910
#include "util.h"
@@ -12,6 +13,7 @@
1213
int
1314
main(int argc, char** argv)
1415
{
16+
SHA256AutoDetect();
1517
RandomInit();
1618
ECC_Start();
1719
SetupEnvironment();

‎src/crypto/sha256.cpp

Lines changed: 148 additions & 88 deletions
Original file line numberDiff line numberDiff line change
@@ -3,10 +3,21 @@
33
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
44

55
#include "crypto/sha256.h"
6-
76
#include "crypto/common.h"
87

8+
#include <assert.h>
99
#include <string.h>
10+
#include <atomic>
11+
12+
#if defined(__x86_64__) || defined(__amd64__)
13+
#if defined(EXPERIMENTAL_ASM)
14+
#include <cpuid.h>
15+
namespace sha256_sse4
16+
{
17+
void Transform(uint32_t* s, const unsigned char* chunk, size_t blocks);
18+
}
19+
#endif
20+
#endif
1021

1122
// Internal implementation code.
1223
namespace
@@ -43,93 +54,142 @@ void inline Initialize(uint32_t* s)
4354
s[7] = 0x5be0cd19ul;
4455
}
4556

46-
/** Perform one SHA-256 transformation, processing a 64-byte chunk. */
47-
void Transform(uint32_t* s, const unsigned char* chunk)
57+
/** Perform a number of SHA-256 transformations, processing 64-byte chunks. */
58+
void Transform(uint32_t* s, const unsigned char* chunk, size_t blocks)
4859
{
49-
uint32_t a = s[0], b = s[1], c = s[2], d = s[3], e = s[4], f = s[5], g = s[6], h = s[7];
50-
uint32_t w0, w1, w2, w3, w4, w5, w6, w7, w8, w9, w10, w11, w12, w13, w14, w15;
51-
52-
Round(a, b, c, d, e, f, g, h, 0x428a2f98, w0 = ReadBE32(chunk + 0));
53-
Round(h, a, b, c, d, e, f, g, 0x71374491, w1 = ReadBE32(chunk + 4));
54-
Round(g, h, a, b, c, d, e, f, 0xb5c0fbcf, w2 = ReadBE32(chunk + 8));
55-
Round(f, g, h, a, b, c, d, e, 0xe9b5dba5, w3 = ReadBE32(chunk + 12));
56-
Round(e, f, g, h, a, b, c, d, 0x3956c25b, w4 = ReadBE32(chunk + 16));
57-
Round(d, e, f, g, h, a, b, c, 0x59f111f1, w5 = ReadBE32(chunk + 20));
58-
Round(c, d, e, f, g, h, a, b, 0x923f82a4, w6 = ReadBE32(chunk + 24));
59-
Round(b, c, d, e, f, g, h, a, 0xab1c5ed5, w7 = ReadBE32(chunk + 28));
60-
Round(a, b, c, d, e, f, g, h, 0xd807aa98, w8 = ReadBE32(chunk + 32));
61-
Round(h, a, b, c, d, e, f, g, 0x12835b01, w9 = ReadBE32(chunk + 36));
62-
Round(g, h, a, b, c, d, e, f, 0x243185be, w10 = ReadBE32(chunk + 40));
63-
Round(f, g, h, a, b, c, d, e, 0x550c7dc3, w11 = ReadBE32(chunk + 44));
64-
Round(e, f, g, h, a, b, c, d, 0x72be5d74, w12 = ReadBE32(chunk + 48));
65-
Round(d, e, f, g, h, a, b, c, 0x80deb1fe, w13 = ReadBE32(chunk + 52));
66-
Round(c, d, e, f, g, h, a, b, 0x9bdc06a7, w14 = ReadBE32(chunk + 56));
67-
Round(b, c, d, e, f, g, h, a, 0xc19bf174, w15 = ReadBE32(chunk + 60));
68-
69-
Round(a, b, c, d, e, f, g, h, 0xe49b69c1, w0 += sigma1(w14) + w9 + sigma0(w1));
70-
Round(h, a, b, c, d, e, f, g, 0xefbe4786, w1 += sigma1(w15) + w10 + sigma0(w2));
71-
Round(g, h, a, b, c, d, e, f, 0x0fc19dc6, w2 += sigma1(w0) + w11 + sigma0(w3));
72-
Round(f, g, h, a, b, c, d, e, 0x240ca1cc, w3 += sigma1(w1) + w12 + sigma0(w4));
73-
Round(e, f, g, h, a, b, c, d, 0x2de92c6f, w4 += sigma1(w2) + w13 + sigma0(w5));
74-
Round(d, e, f, g, h, a, b, c, 0x4a7484aa, w5 += sigma1(w3) + w14 + sigma0(w6));
75-
Round(c, d, e, f, g, h, a, b, 0x5cb0a9dc, w6 += sigma1(w4) + w15 + sigma0(w7));
76-
Round(b, c, d, e, f, g, h, a, 0x76f988da, w7 += sigma1(w5) + w0 + sigma0(w8));
77-
Round(a, b, c, d, e, f, g, h, 0x983e5152, w8 += sigma1(w6) + w1 + sigma0(w9));
78-
Round(h, a, b, c, d, e, f, g, 0xa831c66d, w9 += sigma1(w7) + w2 + sigma0(w10));
79-
Round(g, h, a, b, c, d, e, f, 0xb00327c8, w10 += sigma1(w8) + w3 + sigma0(w11));
80-
Round(f, g, h, a, b, c, d, e, 0xbf597fc7, w11 += sigma1(w9) + w4 + sigma0(w12));
81-
Round(e, f, g, h, a, b, c, d, 0xc6e00bf3, w12 += sigma1(w10) + w5 + sigma0(w13));
82-
Round(d, e, f, g, h, a, b, c, 0xd5a79147, w13 += sigma1(w11) + w6 + sigma0(w14));
83-
Round(c, d, e, f, g, h, a, b, 0x06ca6351, w14 += sigma1(w12) + w7 + sigma0(w15));
84-
Round(b, c, d, e, f, g, h, a, 0x14292967, w15 += sigma1(w13) + w8 + sigma0(w0));
85-
86-
Round(a, b, c, d, e, f, g, h, 0x27b70a85, w0 += sigma1(w14) + w9 + sigma0(w1));
87-
Round(h, a, b, c, d, e, f, g, 0x2e1b2138, w1 += sigma1(w15) + w10 + sigma0(w2));
88-
Round(g, h, a, b, c, d, e, f, 0x4d2c6dfc, w2 += sigma1(w0) + w11 + sigma0(w3));
89-
Round(f, g, h, a, b, c, d, e, 0x53380d13, w3 += sigma1(w1) + w12 + sigma0(w4));
90-
Round(e, f, g, h, a, b, c, d, 0x650a7354, w4 += sigma1(w2) + w13 + sigma0(w5));
91-
Round(d, e, f, g, h, a, b, c, 0x766a0abb, w5 += sigma1(w3) + w14 + sigma0(w6));
92-
Round(c, d, e, f, g, h, a, b, 0x81c2c92e, w6 += sigma1(w4) + w15 + sigma0(w7));
93-
Round(b, c, d, e, f, g, h, a, 0x92722c85, w7 += sigma1(w5) + w0 + sigma0(w8));
94-
Round(a, b, c, d, e, f, g, h, 0xa2bfe8a1, w8 += sigma1(w6) + w1 + sigma0(w9));
95-
Round(h, a, b, c, d, e, f, g, 0xa81a664b, w9 += sigma1(w7) + w2 + sigma0(w10));
96-
Round(g, h, a, b, c, d, e, f, 0xc24b8b70, w10 += sigma1(w8) + w3 + sigma0(w11));
97-
Round(f, g, h, a, b, c, d, e, 0xc76c51a3, w11 += sigma1(w9) + w4 + sigma0(w12));
98-
Round(e, f, g, h, a, b, c, d, 0xd192e819, w12 += sigma1(w10) + w5 + sigma0(w13));
99-
Round(d, e, f, g, h, a, b, c, 0xd6990624, w13 += sigma1(w11) + w6 + sigma0(w14));
100-
Round(c, d, e, f, g, h, a, b, 0xf40e3585, w14 += sigma1(w12) + w7 + sigma0(w15));
101-
Round(b, c, d, e, f, g, h, a, 0x106aa070, w15 += sigma1(w13) + w8 + sigma0(w0));
102-
103-
Round(a, b, c, d, e, f, g, h, 0x19a4c116, w0 += sigma1(w14) + w9 + sigma0(w1));
104-
Round(h, a, b, c, d, e, f, g, 0x1e376c08, w1 += sigma1(w15) + w10 + sigma0(w2));
105-
Round(g, h, a, b, c, d, e, f, 0x2748774c, w2 += sigma1(w0) + w11 + sigma0(w3));
106-
Round(f, g, h, a, b, c, d, e, 0x34b0bcb5, w3 += sigma1(w1) + w12 + sigma0(w4));
107-
Round(e, f, g, h, a, b, c, d, 0x391c0cb3, w4 += sigma1(w2) + w13 + sigma0(w5));
108-
Round(d, e, f, g, h, a, b, c, 0x4ed8aa4a, w5 += sigma1(w3) + w14 + sigma0(w6));
109-
Round(c, d, e, f, g, h, a, b, 0x5b9cca4f, w6 += sigma1(w4) + w15 + sigma0(w7));
110-
Round(b, c, d, e, f, g, h, a, 0x682e6ff3, w7 += sigma1(w5) + w0 + sigma0(w8));
111-
Round(a, b, c, d, e, f, g, h, 0x748f82ee, w8 += sigma1(w6) + w1 + sigma0(w9));
112-
Round(h, a, b, c, d, e, f, g, 0x78a5636f, w9 += sigma1(w7) + w2 + sigma0(w10));
113-
Round(g, h, a, b, c, d, e, f, 0x84c87814, w10 += sigma1(w8) + w3 + sigma0(w11));
114-
Round(f, g, h, a, b, c, d, e, 0x8cc70208, w11 += sigma1(w9) + w4 + sigma0(w12));
115-
Round(e, f, g, h, a, b, c, d, 0x90befffa, w12 += sigma1(w10) + w5 + sigma0(w13));
116-
Round(d, e, f, g, h, a, b, c, 0xa4506ceb, w13 += sigma1(w11) + w6 + sigma0(w14));
117-
Round(c, d, e, f, g, h, a, b, 0xbef9a3f7, w14 + sigma1(w12) + w7 + sigma0(w15));
118-
Round(b, c, d, e, f, g, h, a, 0xc67178f2, w15 + sigma1(w13) + w8 + sigma0(w0));
119-
120-
s[0] += a;
121-
s[1] += b;
122-
s[2] += c;
123-
s[3] += d;
124-
s[4] += e;
125-
s[5] += f;
126-
s[6] += g;
127-
s[7] += h;
60+
while (blocks--) {
61+
uint32_t a = s[0], b = s[1], c = s[2], d = s[3], e = s[4], f = s[5], g = s[6], h = s[7];
62+
uint32_t w0, w1, w2, w3, w4, w5, w6, w7, w8, w9, w10, w11, w12, w13, w14, w15;
63+
64+
Round(a, b, c, d, e, f, g, h, 0x428a2f98, w0 = ReadBE32(chunk + 0));
65+
Round(h, a, b, c, d, e, f, g, 0x71374491, w1 = ReadBE32(chunk + 4));
66+
Round(g, h, a, b, c, d, e, f, 0xb5c0fbcf, w2 = ReadBE32(chunk + 8));
67+
Round(f, g, h, a, b, c, d, e, 0xe9b5dba5, w3 = ReadBE32(chunk + 12));
68+
Round(e, f, g, h, a, b, c, d, 0x3956c25b, w4 = ReadBE32(chunk + 16));
69+
Round(d, e, f, g, h, a, b, c, 0x59f111f1, w5 = ReadBE32(chunk + 20));
70+
Round(c, d, e, f, g, h, a, b, 0x923f82a4, w6 = ReadBE32(chunk + 24));
71+
Round(b, c, d, e, f, g, h, a, 0xab1c5ed5, w7 = ReadBE32(chunk + 28));
72+
Round(a, b, c, d, e, f, g, h, 0xd807aa98, w8 = ReadBE32(chunk + 32));
73+
Round(h, a, b, c, d, e, f, g, 0x12835b01, w9 = ReadBE32(chunk + 36));
74+
Round(g, h, a, b, c, d, e, f, 0x243185be, w10 = ReadBE32(chunk + 40));
75+
Round(f, g, h, a, b, c, d, e, 0x550c7dc3, w11 = ReadBE32(chunk + 44));
76+
Round(e, f, g, h, a, b, c, d, 0x72be5d74, w12 = ReadBE32(chunk + 48));
77+
Round(d, e, f, g, h, a, b, c, 0x80deb1fe, w13 = ReadBE32(chunk + 52));
78+
Round(c, d, e, f, g, h, a, b, 0x9bdc06a7, w14 = ReadBE32(chunk + 56));
79+
Round(b, c, d, e, f, g, h, a, 0xc19bf174, w15 = ReadBE32(chunk + 60));
80+
81+
Round(a, b, c, d, e, f, g, h, 0xe49b69c1, w0 += sigma1(w14) + w9 + sigma0(w1));
82+
Round(h, a, b, c, d, e, f, g, 0xefbe4786, w1 += sigma1(w15) + w10 + sigma0(w2));
83+
Round(g, h, a, b, c, d, e, f, 0x0fc19dc6, w2 += sigma1(w0) + w11 + sigma0(w3));
84+
Round(f, g, h, a, b, c, d, e, 0x240ca1cc, w3 += sigma1(w1) + w12 + sigma0(w4));
85+
Round(e, f, g, h, a, b, c, d, 0x2de92c6f, w4 += sigma1(w2) + w13 + sigma0(w5));
86+
Round(d, e, f, g, h, a, b, c, 0x4a7484aa, w5 += sigma1(w3) + w14 + sigma0(w6));
87+
Round(c, d, e, f, g, h, a, b, 0x5cb0a9dc, w6 += sigma1(w4) + w15 + sigma0(w7));
88+
Round(b, c, d, e, f, g, h, a, 0x76f988da, w7 += sigma1(w5) + w0 + sigma0(w8));
89+
Round(a, b, c, d, e, f, g, h, 0x983e5152, w8 += sigma1(w6) + w1 + sigma0(w9));
90+
Round(h, a, b, c, d, e, f, g, 0xa831c66d, w9 += sigma1(w7) + w2 + sigma0(w10));
91+
Round(g, h, a, b, c, d, e, f, 0xb00327c8, w10 += sigma1(w8) + w3 + sigma0(w11));
92+
Round(f, g, h, a, b, c, d, e, 0xbf597fc7, w11 += sigma1(w9) + w4 + sigma0(w12));
93+
Round(e, f, g, h, a, b, c, d, 0xc6e00bf3, w12 += sigma1(w10) + w5 + sigma0(w13));
94+
Round(d, e, f, g, h, a, b, c, 0xd5a79147, w13 += sigma1(w11) + w6 + sigma0(w14));
95+
Round(c, d, e, f, g, h, a, b, 0x06ca6351, w14 += sigma1(w12) + w7 + sigma0(w15));
96+
Round(b, c, d, e, f, g, h, a, 0x14292967, w15 += sigma1(w13) + w8 + sigma0(w0));
97+
98+
Round(a, b, c, d, e, f, g, h, 0x27b70a85, w0 += sigma1(w14) + w9 + sigma0(w1));
99+
Round(h, a, b, c, d, e, f, g, 0x2e1b2138, w1 += sigma1(w15) + w10 + sigma0(w2));
100+
Round(g, h, a, b, c, d, e, f, 0x4d2c6dfc, w2 += sigma1(w0) + w11 + sigma0(w3));
101+
Round(f, g, h, a, b, c, d, e, 0x53380d13, w3 += sigma1(w1) + w12 + sigma0(w4));
102+
Round(e, f, g, h, a, b, c, d, 0x650a7354, w4 += sigma1(w2) + w13 + sigma0(w5));
103+
Round(d, e, f, g, h, a, b, c, 0x766a0abb, w5 += sigma1(w3) + w14 + sigma0(w6));
104+
Round(c, d, e, f, g, h, a, b, 0x81c2c92e, w6 += sigma1(w4) + w15 + sigma0(w7));
105+
Round(b, c, d, e, f, g, h, a, 0x92722c85, w7 += sigma1(w5) + w0 + sigma0(w8));
106+
Round(a, b, c, d, e, f, g, h, 0xa2bfe8a1, w8 += sigma1(w6) + w1 + sigma0(w9));
107+
Round(h, a, b, c, d, e, f, g, 0xa81a664b, w9 += sigma1(w7) + w2 + sigma0(w10));
108+
Round(g, h, a, b, c, d, e, f, 0xc24b8b70, w10 += sigma1(w8) + w3 + sigma0(w11));
109+
Round(f, g, h, a, b, c, d, e, 0xc76c51a3, w11 += sigma1(w9) + w4 + sigma0(w12));
110+
Round(e, f, g, h, a, b, c, d, 0xd192e819, w12 += sigma1(w10) + w5 + sigma0(w13));
111+
Round(d, e, f, g, h, a, b, c, 0xd6990624, w13 += sigma1(w11) + w6 + sigma0(w14));
112+
Round(c, d, e, f, g, h, a, b, 0xf40e3585, w14 += sigma1(w12) + w7 + sigma0(w15));
113+
Round(b, c, d, e, f, g, h, a, 0x106aa070, w15 += sigma1(w13) + w8 + sigma0(w0));
114+
115+
Round(a, b, c, d, e, f, g, h, 0x19a4c116, w0 += sigma1(w14) + w9 + sigma0(w1));
116+
Round(h, a, b, c, d, e, f, g, 0x1e376c08, w1 += sigma1(w15) + w10 + sigma0(w2));
117+
Round(g, h, a, b, c, d, e, f, 0x2748774c, w2 += sigma1(w0) + w11 + sigma0(w3));
118+
Round(f, g, h, a, b, c, d, e, 0x34b0bcb5, w3 += sigma1(w1) + w12 + sigma0(w4));
119+
Round(e, f, g, h, a, b, c, d, 0x391c0cb3, w4 += sigma1(w2) + w13 + sigma0(w5));
120+
Round(d, e, f, g, h, a, b, c, 0x4ed8aa4a, w5 += sigma1(w3) + w14 + sigma0(w6));
121+
Round(c, d, e, f, g, h, a, b, 0x5b9cca4f, w6 += sigma1(w4) + w15 + sigma0(w7));
122+
Round(b, c, d, e, f, g, h, a, 0x682e6ff3, w7 += sigma1(w5) + w0 + sigma0(w8));
123+
Round(a, b, c, d, e, f, g, h, 0x748f82ee, w8 += sigma1(w6) + w1 + sigma0(w9));
124+
Round(h, a, b, c, d, e, f, g, 0x78a5636f, w9 += sigma1(w7) + w2 + sigma0(w10));
125+
Round(g, h, a, b, c, d, e, f, 0x84c87814, w10 += sigma1(w8) + w3 + sigma0(w11));
126+
Round(f, g, h, a, b, c, d, e, 0x8cc70208, w11 += sigma1(w9) + w4 + sigma0(w12));
127+
Round(e, f, g, h, a, b, c, d, 0x90befffa, w12 += sigma1(w10) + w5 + sigma0(w13));
128+
Round(d, e, f, g, h, a, b, c, 0xa4506ceb, w13 += sigma1(w11) + w6 + sigma0(w14));
129+
Round(c, d, e, f, g, h, a, b, 0xbef9a3f7, w14 + sigma1(w12) + w7 + sigma0(w15));
130+
Round(b, c, d, e, f, g, h, a, 0xc67178f2, w15 + sigma1(w13) + w8 + sigma0(w0));
131+
132+
s[0] += a;
133+
s[1] += b;
134+
s[2] += c;
135+
s[3] += d;
136+
s[4] += e;
137+
s[5] += f;
138+
s[6] += g;
139+
s[7] += h;
140+
chunk += 64;
141+
}
128142
}
129143

130144
} // namespace sha256
145+
146+
typedef void (*TransformType)(uint32_t*, const unsigned char*, size_t);
147+
148+
bool SelfTest(TransformType tr) {
149+
static const unsigned char in1[65] = {0, 0x80};
150+
static const unsigned char in2[129] = {
151+
0,
152+
32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
153+
32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
154+
0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
155+
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0
156+
};
157+
static const uint32_t init[8] = {0x6a09e667ul, 0xbb67ae85ul, 0x3c6ef372ul, 0xa54ff53aul, 0x510e527ful, 0x9b05688cul, 0x1f83d9abul, 0x5be0cd19ul};
158+
static const uint32_t out1[8] = {0xe3b0c442ul, 0x98fc1c14ul, 0x9afbf4c8ul, 0x996fb924ul, 0x27ae41e4ul, 0x649b934cul, 0xa495991bul, 0x7852b855ul};
159+
static const uint32_t out2[8] = {0xce4153b0ul, 0x147c2a86ul, 0x3ed4298eul, 0xe0676bc8ul, 0x79fc77a1ul, 0x2abe1f49ul, 0xb2b055dful, 0x1069523eul};
160+
uint32_t buf[8];
161+
memcpy(buf, init, sizeof(buf));
162+
// Process nothing, and check we remain in the initial state.
163+
tr(buf, nullptr, 0);
164+
if (memcmp(buf, init, sizeof(buf))) return false;
165+
// Process the padded empty string (unaligned)
166+
tr(buf, in1 + 1, 1);
167+
if (memcmp(buf, out1, sizeof(buf))) return false;
168+
// Process 64 spaces (unaligned)
169+
memcpy(buf, init, sizeof(buf));
170+
tr(buf, in2 + 1, 2);
171+
if (memcmp(buf, out2, sizeof(buf))) return false;
172+
return true;
173+
}
174+
175+
TransformType Transform = sha256::Transform;
176+
131177
} // namespace
132178

179+
std::string SHA256AutoDetect()
180+
{
181+
#if defined(EXPERIMENTAL_ASM) && (defined(__x86_64__) || defined(__amd64__))
182+
uint32_t eax, ebx, ecx, edx;
183+
if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) && (ecx >> 19) & 1) {
184+
Transform = sha256_sse4::Transform;
185+
assert(SelfTest(Transform));
186+
return "sse4";
187+
}
188+
#endif
189+
190+
assert(SelfTest(Transform));
191+
return "standard";
192+
}
133193

134194
////// SHA-256
135195

@@ -147,14 +207,14 @@ CSHA256& CSHA256::Write(const unsigned char* data, size_t len)
147207
memcpy(buf + bufsize, data, 64 - bufsize);
148208
bytes += 64 - bufsize;
149209
data += 64 - bufsize;
150-
sha256::Transform(s, buf);
210+
Transform(s, buf, 1);
151211
bufsize = 0;
152212
}
153-
while (end >= data + 64) {
154-
// Process full chunks directly from the source.
155-
sha256::Transform(s, data);
156-
bytes += 64;
157-
data += 64;
213+
if (end - data >= 64) {
214+
size_t blocks = (end - data) / 64;
215+
Transform(s, data, blocks);
216+
data += 64 * blocks;
217+
bytes += 64 * blocks;
158218
}
159219
if (end > data) {
160220
// Fill the buffer with what remains.

‎src/crypto/sha256.h

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77

88
#include <stdint.h>
99
#include <stdlib.h>
10+
#include <string>
1011

1112
/** A hasher class for SHA-256. */
1213
class CSHA256
@@ -25,4 +26,9 @@ class CSHA256
2526
CSHA256& Reset();
2627
};
2728

29+
/** Autodetect the best available SHA256 implementation.
30+
* Returns the name of the implementation.
31+
*/
32+
std::string SHA256AutoDetect();
33+
2834
#endif // BITCOIN_CRYPTO_SHA256_H

0 commit comments

Comments
 (0)