aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/tools/python3/Modules/_blake2/impl/blake2b-load-sse2.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/blake2b-load-sse2.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/blake2b-load-sse2.h')
-rw-r--r--contrib/tools/python3/Modules/_blake2/impl/blake2b-load-sse2.h68
1 files changed, 68 insertions, 0 deletions
diff --git a/contrib/tools/python3/Modules/_blake2/impl/blake2b-load-sse2.h b/contrib/tools/python3/Modules/_blake2/impl/blake2b-load-sse2.h
new file mode 100644
index 0000000000..1ba153c87d
--- /dev/null
+++ b/contrib/tools/python3/Modules/_blake2/impl/blake2b-load-sse2.h
@@ -0,0 +1,68 @@
+/*
+ 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 __BLAKE2B_LOAD_SSE2_H__
+#define __BLAKE2B_LOAD_SSE2_H__
+
+#define LOAD_MSG_0_1(b0, b1) b0 = _mm_set_epi64x(m2, m0); b1 = _mm_set_epi64x(m6, m4)
+#define LOAD_MSG_0_2(b0, b1) b0 = _mm_set_epi64x(m3, m1); b1 = _mm_set_epi64x(m7, m5)
+#define LOAD_MSG_0_3(b0, b1) b0 = _mm_set_epi64x(m10, m8); b1 = _mm_set_epi64x(m14, m12)
+#define LOAD_MSG_0_4(b0, b1) b0 = _mm_set_epi64x(m11, m9); b1 = _mm_set_epi64x(m15, m13)
+#define LOAD_MSG_1_1(b0, b1) b0 = _mm_set_epi64x(m4, m14); b1 = _mm_set_epi64x(m13, m9)
+#define LOAD_MSG_1_2(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m6, m15)
+#define LOAD_MSG_1_3(b0, b1) b0 = _mm_set_epi64x(m0, m1); b1 = _mm_set_epi64x(m5, m11)
+#define LOAD_MSG_1_4(b0, b1) b0 = _mm_set_epi64x(m2, m12); b1 = _mm_set_epi64x(m3, m7)
+#define LOAD_MSG_2_1(b0, b1) b0 = _mm_set_epi64x(m12, m11); b1 = _mm_set_epi64x(m15, m5)
+#define LOAD_MSG_2_2(b0, b1) b0 = _mm_set_epi64x(m0, m8); b1 = _mm_set_epi64x(m13, m2)
+#define LOAD_MSG_2_3(b0, b1) b0 = _mm_set_epi64x(m3, m10); b1 = _mm_set_epi64x(m9, m7)
+#define LOAD_MSG_2_4(b0, b1) b0 = _mm_set_epi64x(m6, m14); b1 = _mm_set_epi64x(m4, m1)
+#define LOAD_MSG_3_1(b0, b1) b0 = _mm_set_epi64x(m3, m7); b1 = _mm_set_epi64x(m11, m13)
+#define LOAD_MSG_3_2(b0, b1) b0 = _mm_set_epi64x(m1, m9); b1 = _mm_set_epi64x(m14, m12)
+#define LOAD_MSG_3_3(b0, b1) b0 = _mm_set_epi64x(m5, m2); b1 = _mm_set_epi64x(m15, m4)
+#define LOAD_MSG_3_4(b0, b1) b0 = _mm_set_epi64x(m10, m6); b1 = _mm_set_epi64x(m8, m0)
+#define LOAD_MSG_4_1(b0, b1) b0 = _mm_set_epi64x(m5, m9); b1 = _mm_set_epi64x(m10, m2)
+#define LOAD_MSG_4_2(b0, b1) b0 = _mm_set_epi64x(m7, m0); b1 = _mm_set_epi64x(m15, m4)
+#define LOAD_MSG_4_3(b0, b1) b0 = _mm_set_epi64x(m11, m14); b1 = _mm_set_epi64x(m3, m6)
+#define LOAD_MSG_4_4(b0, b1) b0 = _mm_set_epi64x(m12, m1); b1 = _mm_set_epi64x(m13, m8)
+#define LOAD_MSG_5_1(b0, b1) b0 = _mm_set_epi64x(m6, m2); b1 = _mm_set_epi64x(m8, m0)
+#define LOAD_MSG_5_2(b0, b1) b0 = _mm_set_epi64x(m10, m12); b1 = _mm_set_epi64x(m3, m11)
+#define LOAD_MSG_5_3(b0, b1) b0 = _mm_set_epi64x(m7, m4); b1 = _mm_set_epi64x(m1, m15)
+#define LOAD_MSG_5_4(b0, b1) b0 = _mm_set_epi64x(m5, m13); b1 = _mm_set_epi64x(m9, m14)
+#define LOAD_MSG_6_1(b0, b1) b0 = _mm_set_epi64x(m1, m12); b1 = _mm_set_epi64x(m4, m14)
+#define LOAD_MSG_6_2(b0, b1) b0 = _mm_set_epi64x(m15, m5); b1 = _mm_set_epi64x(m10, m13)
+#define LOAD_MSG_6_3(b0, b1) b0 = _mm_set_epi64x(m6, m0); b1 = _mm_set_epi64x(m8, m9)
+#define LOAD_MSG_6_4(b0, b1) b0 = _mm_set_epi64x(m3, m7); b1 = _mm_set_epi64x(m11, m2)
+#define LOAD_MSG_7_1(b0, b1) b0 = _mm_set_epi64x(m7, m13); b1 = _mm_set_epi64x(m3, m12)
+#define LOAD_MSG_7_2(b0, b1) b0 = _mm_set_epi64x(m14, m11); b1 = _mm_set_epi64x(m9, m1)
+#define LOAD_MSG_7_3(b0, b1) b0 = _mm_set_epi64x(m15, m5); b1 = _mm_set_epi64x(m2, m8)
+#define LOAD_MSG_7_4(b0, b1) b0 = _mm_set_epi64x(m4, m0); b1 = _mm_set_epi64x(m10, m6)
+#define LOAD_MSG_8_1(b0, b1) b0 = _mm_set_epi64x(m14, m6); b1 = _mm_set_epi64x(m0, m11)
+#define LOAD_MSG_8_2(b0, b1) b0 = _mm_set_epi64x(m9, m15); b1 = _mm_set_epi64x(m8, m3)
+#define LOAD_MSG_8_3(b0, b1) b0 = _mm_set_epi64x(m13, m12); b1 = _mm_set_epi64x(m10, m1)
+#define LOAD_MSG_8_4(b0, b1) b0 = _mm_set_epi64x(m7, m2); b1 = _mm_set_epi64x(m5, m4)
+#define LOAD_MSG_9_1(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m1, m7)
+#define LOAD_MSG_9_2(b0, b1) b0 = _mm_set_epi64x(m4, m2); b1 = _mm_set_epi64x(m5, m6)
+#define LOAD_MSG_9_3(b0, b1) b0 = _mm_set_epi64x(m9, m15); b1 = _mm_set_epi64x(m13, m3)
+#define LOAD_MSG_9_4(b0, b1) b0 = _mm_set_epi64x(m14, m11); b1 = _mm_set_epi64x(m0, m12)
+#define LOAD_MSG_10_1(b0, b1) b0 = _mm_set_epi64x(m2, m0); b1 = _mm_set_epi64x(m6, m4)
+#define LOAD_MSG_10_2(b0, b1) b0 = _mm_set_epi64x(m3, m1); b1 = _mm_set_epi64x(m7, m5)
+#define LOAD_MSG_10_3(b0, b1) b0 = _mm_set_epi64x(m10, m8); b1 = _mm_set_epi64x(m14, m12)
+#define LOAD_MSG_10_4(b0, b1) b0 = _mm_set_epi64x(m11, m9); b1 = _mm_set_epi64x(m15, m13)
+#define LOAD_MSG_11_1(b0, b1) b0 = _mm_set_epi64x(m4, m14); b1 = _mm_set_epi64x(m13, m9)
+#define LOAD_MSG_11_2(b0, b1) b0 = _mm_set_epi64x(m8, m10); b1 = _mm_set_epi64x(m6, m15)
+#define LOAD_MSG_11_3(b0, b1) b0 = _mm_set_epi64x(m0, m1); b1 = _mm_set_epi64x(m5, m11)
+#define LOAD_MSG_11_4(b0, b1) b0 = _mm_set_epi64x(m2, m12); b1 = _mm_set_epi64x(m3, m7)
+
+
+#endif
+