aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h
diff options
context:
space:
mode:
authorAlexSm <alex@ydb.tech>2024-03-05 10:40:59 +0100
committerGitHub <noreply@github.com>2024-03-05 12:40:59 +0300
commit1ac13c847b5358faba44dbb638a828e24369467b (patch)
tree07672b4dd3604ad3dee540a02c6494cb7d10dc3d /contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h
parentffcca3e7f7958ddc6487b91d3df8c01054bd0638 (diff)
downloadydb-1ac13c847b5358faba44dbb638a828e24369467b.tar.gz
Library import 16 (#2433)
Co-authored-by: robot-piglet <robot-piglet@yandex-team.com> Co-authored-by: deshevoy <deshevoy@yandex-team.com> Co-authored-by: robot-contrib <robot-contrib@yandex-team.com> Co-authored-by: thegeorg <thegeorg@yandex-team.com> Co-authored-by: robot-ya-builder <robot-ya-builder@yandex-team.com> Co-authored-by: svidyuk <svidyuk@yandex-team.com> Co-authored-by: shadchin <shadchin@yandex-team.com> Co-authored-by: robot-ratatosk <robot-ratatosk@yandex-team.com> Co-authored-by: innokentii <innokentii@yandex-team.com> Co-authored-by: arkady-e1ppa <arkady-e1ppa@yandex-team.com> Co-authored-by: snermolaev <snermolaev@yandex-team.com> Co-authored-by: dimdim11 <dimdim11@yandex-team.com> Co-authored-by: kickbutt <kickbutt@yandex-team.com> Co-authored-by: abdullinsaid <abdullinsaid@yandex-team.com> Co-authored-by: korsunandrei <korsunandrei@yandex-team.com> Co-authored-by: petrk <petrk@yandex-team.com> Co-authored-by: miroslav2 <miroslav2@yandex-team.com> Co-authored-by: serjflint <serjflint@yandex-team.com> Co-authored-by: akhropov <akhropov@yandex-team.com> Co-authored-by: prettyboy <prettyboy@yandex-team.com> Co-authored-by: ilikepugs <ilikepugs@yandex-team.com> Co-authored-by: hiddenpath <hiddenpath@yandex-team.com> Co-authored-by: mikhnenko <mikhnenko@yandex-team.com> Co-authored-by: spreis <spreis@yandex-team.com> Co-authored-by: andreyshspb <andreyshspb@yandex-team.com> Co-authored-by: dimaandreev <dimaandreev@yandex-team.com> Co-authored-by: rashid <rashid@yandex-team.com> Co-authored-by: robot-ydb-importer <robot-ydb-importer@yandex-team.com> Co-authored-by: r-vetrov <r-vetrov@yandex-team.com> Co-authored-by: ypodlesov <ypodlesov@yandex-team.com> Co-authored-by: zaverden <zaverden@yandex-team.com> Co-authored-by: vpozdyayev <vpozdyayev@yandex-team.com> Co-authored-by: robot-cozmo <robot-cozmo@yandex-team.com> Co-authored-by: v-korovin <v-korovin@yandex-team.com> Co-authored-by: arikon <arikon@yandex-team.com> Co-authored-by: khoden <khoden@yandex-team.com> Co-authored-by: psydmm <psydmm@yandex-team.com> Co-authored-by: robot-javacom <robot-javacom@yandex-team.com> Co-authored-by: dtorilov <dtorilov@yandex-team.com> Co-authored-by: sennikovmv <sennikovmv@yandex-team.com> Co-authored-by: hcpp <hcpp@ydb.tech>
Diffstat (limited to 'contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h')
-rw-r--r--contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h91
1 files changed, 91 insertions, 0 deletions
diff --git a/contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h b/contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h
new file mode 100644
index 0000000000..3af4be35be
--- /dev/null
+++ b/contrib/tools/python3/Modules/_blake2/impl/blake2s-round.h
@@ -0,0 +1,91 @@
+/*
+ BLAKE2 reference source code package - optimized C implementations
+
+ Written in 2012 by Samuel Neves <sneves@dei.uc.pt>
+
+ To the extent possible under law, the author(s) have dedicated all copyright
+ and related and neighboring rights to this software to the public domain
+ worldwide. This software is distributed without any warranty.
+
+ You should have received a copy of the CC0 Public Domain Dedication along with
+ this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
+*/
+#pragma once
+#ifndef __BLAKE2S_ROUND_H__
+#define __BLAKE2S_ROUND_H__
+
+#define LOAD(p) _mm_load_si128( (__m128i *)(p) )
+#define STORE(p,r) _mm_store_si128((__m128i *)(p), r)
+
+#define LOADU(p) _mm_loadu_si128( (__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 */
+#ifndef HAVE_XOP
+#ifdef HAVE_SSSE3
+#define _mm_roti_epi32(r, c) ( \
+ (8==-(c)) ? _mm_shuffle_epi8(r,r8) \
+ : (16==-(c)) ? _mm_shuffle_epi8(r,r16) \
+ : _mm_xor_si128(_mm_srli_epi32( (r), -(c) ),_mm_slli_epi32( (r), 32-(-(c)) )) )
+#else
+#define _mm_roti_epi32(r, c) _mm_xor_si128(_mm_srli_epi32( (r), -(c) ),_mm_slli_epi32( (r), 32-(-(c)) ))
+#endif
+#else
+/* ... */
+#endif
+
+
+#define G1(row1,row2,row3,row4,buf) \
+ row1 = _mm_add_epi32( _mm_add_epi32( row1, buf), row2 ); \
+ row4 = _mm_xor_si128( row4, row1 ); \
+ row4 = _mm_roti_epi32(row4, -16); \
+ row3 = _mm_add_epi32( row3, row4 ); \
+ row2 = _mm_xor_si128( row2, row3 ); \
+ row2 = _mm_roti_epi32(row2, -12);
+
+#define G2(row1,row2,row3,row4,buf) \
+ row1 = _mm_add_epi32( _mm_add_epi32( row1, buf), row2 ); \
+ row4 = _mm_xor_si128( row4, row1 ); \
+ row4 = _mm_roti_epi32(row4, -8); \
+ row3 = _mm_add_epi32( row3, row4 ); \
+ row2 = _mm_xor_si128( row2, row3 ); \
+ row2 = _mm_roti_epi32(row2, -7);
+
+#define DIAGONALIZE(row1,row2,row3,row4) \
+ row4 = _mm_shuffle_epi32( row4, _MM_SHUFFLE(2,1,0,3) ); \
+ row3 = _mm_shuffle_epi32( row3, _MM_SHUFFLE(1,0,3,2) ); \
+ row2 = _mm_shuffle_epi32( row2, _MM_SHUFFLE(0,3,2,1) );
+
+#define UNDIAGONALIZE(row1,row2,row3,row4) \
+ row4 = _mm_shuffle_epi32( row4, _MM_SHUFFLE(0,3,2,1) ); \
+ row3 = _mm_shuffle_epi32( row3, _MM_SHUFFLE(1,0,3,2) ); \
+ row2 = _mm_shuffle_epi32( row2, _MM_SHUFFLE(2,1,0,3) );
+
+#if defined(HAVE_XOP)
+#include "blake2s-load-xop.h"
+#elif defined(HAVE_SSE4_1)
+#include "blake2s-load-sse41.h"
+#else
+#include "blake2s-load-sse2.h"
+#endif
+
+#define ROUND(r) \
+ LOAD_MSG_ ##r ##_1(buf1); \
+ G1(row1,row2,row3,row4,buf1); \
+ LOAD_MSG_ ##r ##_2(buf2); \
+ G2(row1,row2,row3,row4,buf2); \
+ DIAGONALIZE(row1,row2,row3,row4); \
+ LOAD_MSG_ ##r ##_3(buf3); \
+ G1(row1,row2,row3,row4,buf3); \
+ LOAD_MSG_ ##r ##_4(buf4); \
+ G2(row1,row2,row3,row4,buf4); \
+ UNDIAGONALIZE(row1,row2,row3,row4); \
+
+#endif
+