RandomX: added SSE4.1-optimized Blake2b

+0.15% on `rx/0` +0.3% on `rx/wow`
2020-09-10 14:28:40 +02:00 · 2020-09-10 14:28:40 +02:00 · 4a9db89527
commit 4a9db89527
parent 0f09883429
7 changed files with 623 additions and 9 deletions
--- a/src/crypto/randomx/blake2/blake2b-round.h
+++ b/src/crypto/randomx/blake2/blake2b-round.h
@ -0,0 +1,119 @@
+/*
+   BLAKE2 reference source code package - optimized C implementations
+
+   Copyright 2012, Samuel Neves <sneves@dei.uc.pt>.  You may use this under the
+   terms of the CC0, the OpenSSL Licence, or the Apache Public License 2.0, at
+   your option.  The terms of these licenses can be found at:
+
+   - CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
+   - OpenSSL license   : https://www.openssl.org/source/license.html
+   - Apache 2.0        : http://www.apache.org/licenses/LICENSE-2.0
+
+   More information about the BLAKE2 hash function can be found at
+   https://blake2.net.
+*/
+#ifndef BLAKE2B_ROUND_H
+#define BLAKE2B_ROUND_H
+
+#define LOADU(p)  _mm_loadu_si128( (const __m128i *)(p) )
+#define STOREU(p,r) _mm_storeu_si128((__m128i *)(p), r)
+
+#define TOF(reg) _mm_castsi128_ps((reg))
+#define TOI(reg) _mm_castps_si128((reg))
+
+#define LIKELY(x) __builtin_expect((x),1)
+
+
+/* Microarchitecture-specific macros */
+#define _mm_roti_epi64(x, c) \
+    (-(c) == 32) ? _mm_shuffle_epi32((x), _MM_SHUFFLE(2,3,0,1))  \
+    : (-(c) == 24) ? _mm_shuffle_epi8((x), r24) \
+    : (-(c) == 16) ? _mm_shuffle_epi8((x), r16) \
+    : (-(c) == 63) ? _mm_xor_si128(_mm_srli_epi64((x), -(c)), _mm_add_epi64((x), (x)))  \
+    : _mm_xor_si128(_mm_srli_epi64((x), -(c)), _mm_slli_epi64((x), 64-(-(c))))
+
+
+
+#define G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1) \
+  row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \
+  row1h = _mm_add_epi64(_mm_add_epi64(row1h, b1), row2h); \
+  \
+  row4l = _mm_xor_si128(row4l, row1l); \
+  row4h = _mm_xor_si128(row4h, row1h); \
+  \
+  row4l = _mm_roti_epi64(row4l, -32); \
+  row4h = _mm_roti_epi64(row4h, -32); \
+  \
+  row3l = _mm_add_epi64(row3l, row4l); \
+  row3h = _mm_add_epi64(row3h, row4h); \
+  \
+  row2l = _mm_xor_si128(row2l, row3l); \
+  row2h = _mm_xor_si128(row2h, row3h); \
+  \
+  row2l = _mm_roti_epi64(row2l, -24); \
+  row2h = _mm_roti_epi64(row2h, -24); \
+
+#define G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1) \
+  row1l = _mm_add_epi64(_mm_add_epi64(row1l, b0), row2l); \
+  row1h = _mm_add_epi64(_mm_add_epi64(row1h, b1), row2h); \
+  \
+  row4l = _mm_xor_si128(row4l, row1l); \
+  row4h = _mm_xor_si128(row4h, row1h); \
+  \
+  row4l = _mm_roti_epi64(row4l, -16); \
+  row4h = _mm_roti_epi64(row4h, -16); \
+  \
+  row3l = _mm_add_epi64(row3l, row4l); \
+  row3h = _mm_add_epi64(row3h, row4h); \
+  \
+  row2l = _mm_xor_si128(row2l, row3l); \
+  row2h = _mm_xor_si128(row2h, row3h); \
+  \
+  row2l = _mm_roti_epi64(row2l, -63); \
+  row2h = _mm_roti_epi64(row2h, -63); \
+
+#define DIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \
+  t0 = _mm_alignr_epi8(row2h, row2l, 8); \
+  t1 = _mm_alignr_epi8(row2l, row2h, 8); \
+  row2l = t0; \
+  row2h = t1; \
+  \
+  t0 = row3l; \
+  row3l = row3h; \
+  row3h = t0;    \
+  \
+  t0 = _mm_alignr_epi8(row4h, row4l, 8); \
+  t1 = _mm_alignr_epi8(row4l, row4h, 8); \
+  row4l = t1; \
+  row4h = t0;
+
+#define UNDIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h) \
+  t0 = _mm_alignr_epi8(row2l, row2h, 8); \
+  t1 = _mm_alignr_epi8(row2h, row2l, 8); \
+  row2l = t0; \
+  row2h = t1; \
+  \
+  t0 = row3l; \
+  row3l = row3h; \
+  row3h = t0; \
+  \
+  t0 = _mm_alignr_epi8(row4l, row4h, 8); \
+  t1 = _mm_alignr_epi8(row4h, row4l, 8); \
+  row4l = t1; \
+  row4h = t0;
+
+#include "blake2b-load-sse41.h"
+
+#define ROUND(r) \
+  LOAD_MSG_ ##r ##_1(b0, b1); \
+  G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \
+  LOAD_MSG_ ##r ##_2(b0, b1); \
+  G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \
+  DIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h); \
+  LOAD_MSG_ ##r ##_3(b0, b1); \
+  G1(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \
+  LOAD_MSG_ ##r ##_4(b0, b1); \
+  G2(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h,b0,b1); \
+  UNDIAGONALIZE(row1l,row2l,row3l,row4l,row1h,row2h,row3h,row4h);
+
+#endif