crypto: sha512-mb - Crypto computation (x4 AVX2)
authorMegha Dey <megha.dey@linux.intel.com>
Mon, 27 Jun 2016 17:20:08 +0000 (10:20 -0700)
committerHerbert Xu <herbert@gondor.apana.org.au>
Tue, 28 Jun 2016 08:06:40 +0000 (16:06 +0800)
This patch introduces the assembly routines to do SHA512 computation on
buffers belonging to several jobs at once. The assembly routines are
optimized with AVX2 instructions that have 4 data lanes and using AVX2
registers.

Signed-off-by: Megha Dey <megha.dey@linux.intel.com>
Reviewed-by: Fenghua Yu <fenghua.yu@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/sha512-mb/sha512_x4_avx2.S [new file with mode: 0644]

diff --git a/arch/x86/crypto/sha512-mb/sha512_x4_avx2.S b/arch/x86/crypto/sha512-mb/sha512_x4_avx2.S
new file mode 100644 (file)
index 0000000..31ab1ef
--- /dev/null
@@ -0,0 +1,529 @@
+/*
+ * Multi-buffer SHA512 algorithm hash compute routine
+ *
+ * This file is provided under a dual BSD/GPLv2 license.  When using or
+ * redistributing this file, you may do so under either license.
+ *
+ * GPL LICENSE SUMMARY
+ *
+ * Copyright(c) 2016 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * Contact Information:
+ *     Megha Dey <megha.dey@linux.intel.com>
+ *
+ * BSD LICENSE
+ *
+ * Copyright(c) 2016 Intel Corporation.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *   * Redistributions of source code must retain the above copyright
+ *     notice, this list of conditions and the following disclaimer.
+ *   * Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in
+ *     the documentation and/or other materials provided with the
+ *     distribution.
+ *   * Neither the name of Intel Corporation nor the names of its
+ *     contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+# code to compute quad SHA512 using AVX2
+# use YMMs to tackle the larger digest size
+# outer calling routine takes care of save and restore of XMM registers
+# Logic designed/laid out by JDG
+
+# Function clobbers: rax, rcx, rdx, rbx, rsi, rdi, r9-r15; ymm0-15
+# Stack must be aligned to 32 bytes before call
+# Linux clobbers: rax rbx rcx rsi r8 r9 r10 r11 r12
+# Linux preserves: rcx rdx rdi rbp r13 r14 r15
+# clobbers ymm0-15
+
+#include <linux/linkage.h>
+#include "sha512_mb_mgr_datastruct.S"
+
+arg1 = %rdi
+arg2 = %rsi
+
+# Common definitions
+STATE = arg1
+INP_SIZE = arg2
+
+IDX = %rax
+ROUND = %rbx
+TBL = %r8
+
+inp0 = %r9
+inp1 = %r10
+inp2 = %r11
+inp3 = %r12
+
+a = %ymm0
+b = %ymm1
+c = %ymm2
+d = %ymm3
+e = %ymm4
+f = %ymm5
+g = %ymm6
+h = %ymm7
+
+a0 = %ymm8
+a1 = %ymm9
+a2 = %ymm10
+
+TT0 = %ymm14
+TT1 = %ymm13
+TT2 = %ymm12
+TT3 = %ymm11
+TT4 = %ymm10
+TT5 = %ymm9
+
+T1 = %ymm14
+TMP = %ymm15
+
+# Define stack usage
+STACK_SPACE1 = SZ4*16 + NUM_SHA512_DIGEST_WORDS*SZ4 + 24
+
+#define VMOVPD vmovupd
+_digest = SZ4*16
+
+# transpose r0, r1, r2, r3, t0, t1
+# "transpose" data in {r0..r3} using temps {t0..t3}
+# Input looks like: {r0 r1 r2 r3}
+# r0 = {a7 a6 a5 a4 a3 a2 a1 a0}
+# r1 = {b7 b6 b5 b4 b3 b2 b1 b0}
+# r2 = {c7 c6 c5 c4 c3 c2 c1 c0}
+# r3 = {d7 d6 d5 d4 d3 d2 d1 d0}
+#
+# output looks like: {t0 r1 r0 r3}
+# t0 = {d1 d0 c1 c0 b1 b0 a1 a0}
+# r1 = {d3 d2 c3 c2 b3 b2 a3 a2}
+# r0 = {d5 d4 c5 c4 b5 b4 a5 a4}
+# r3 = {d7 d6 c7 c6 b7 b6 a7 a6}
+
+.macro TRANSPOSE r0 r1 r2 r3 t0 t1
+       vshufps  $0x44, \r1, \r0, \t0 # t0 = {b5 b4 a5 a4   b1 b0 a1 a0}
+        vshufps  $0xEE, \r1, \r0, \r0 # r0 = {b7 b6 a7 a6   b3 b2 a3 a2}
+        vshufps  $0x44, \r3, \r2, \t1 # t1 = {d5 d4 c5 c4   d1 d0 c1 c0}
+        vshufps  $0xEE, \r3, \r2, \r2 # r2 = {d7 d6 c7 c6   d3 d2 c3 c2}
+
+       vperm2f128      $0x20, \r2, \r0, \r1  # h6...a6
+        vperm2f128      $0x31, \r2, \r0, \r3  # h2...a2
+        vperm2f128      $0x31, \t1, \t0, \r0  # h5...a5
+        vperm2f128      $0x20, \t1, \t0, \t0  # h1...a1
+.endm
+
+.macro ROTATE_ARGS
+TMP_ = h
+h = g
+g = f
+f = e
+e = d
+d = c
+c = b
+b = a
+a = TMP_
+.endm
+
+# PRORQ reg, imm, tmp
+# packed-rotate-right-double
+# does a rotate by doing two shifts and an or
+.macro _PRORQ reg imm tmp
+       vpsllq  $(64-\imm),\reg,\tmp
+       vpsrlq  $\imm,\reg, \reg
+       vpor    \tmp,\reg, \reg
+.endm
+
+# non-destructive
+# PRORQ_nd reg, imm, tmp, src
+.macro _PRORQ_nd reg imm tmp src
+       vpsllq  $(64-\imm), \src, \tmp
+       vpsrlq  $\imm, \src, \reg
+       vpor    \tmp, \reg, \reg
+.endm
+
+# PRORQ dst/src, amt
+.macro PRORQ reg imm
+       _PRORQ  \reg, \imm, TMP
+.endm
+
+# PRORQ_nd dst, src, amt
+.macro PRORQ_nd reg tmp imm
+       _PRORQ_nd       \reg, \imm, TMP, \tmp
+.endm
+
+#; arguments passed implicitly in preprocessor symbols i, a...h
+.macro ROUND_00_15 _T1 i
+       PRORQ_nd a0, e, (18-14) # sig1: a0 = (e >> 4)
+
+       vpxor   g, f, a2        # ch: a2 = f^g
+        vpand   e,a2, a2                # ch: a2 = (f^g)&e
+        vpxor   g, a2, a2               # a2 = ch
+
+        PRORQ_nd        a1,e,41         # sig1: a1 = (e >> 25)
+
+        offset = SZ4*(\i & 0xf)
+        vmovdqu \_T1,offset(%rsp)
+        vpaddq  (TBL,ROUND,1), \_T1, \_T1       # T1 = W + K
+        vpxor   e,a0, a0        # sig1: a0 = e ^ (e >> 5)
+        PRORQ   a0, 14           # sig1: a0 = (e >> 6) ^ (e >> 11)
+        vpaddq  a2, h, h        # h = h + ch
+        PRORQ_nd        a2,a,6  # sig0: a2 = (a >> 11)
+        vpaddq  \_T1,h, h       # h = h + ch + W + K
+        vpxor   a1, a0, a0      # a0 = sigma1
+       vmovdqu a,\_T1
+        PRORQ_nd        a1,a,39 # sig0: a1 = (a >> 22)
+        vpxor   c, \_T1, \_T1      # maj: T1 = a^c
+        add     $SZ4, ROUND     # ROUND++
+        vpand   b, \_T1, \_T1   # maj: T1 = (a^c)&b
+        vpaddq  a0, h, h
+        vpaddq  h, d, d
+        vpxor   a, a2, a2       # sig0: a2 = a ^ (a >> 11)
+        PRORQ   a2,28            # sig0: a2 = (a >> 2) ^ (a >> 13)
+        vpxor   a1, a2, a2      # a2 = sig0
+        vpand   c, a, a1        # maj: a1 = a&c
+        vpor    \_T1, a1, a1    # a1 = maj
+        vpaddq  a1, h, h        # h = h + ch + W + K + maj
+        vpaddq  a2, h, h        # h = h + ch + W + K + maj + sigma0
+        ROTATE_ARGS
+.endm
+
+
+#; arguments passed implicitly in preprocessor symbols i, a...h
+.macro ROUND_16_XX _T1 i
+       vmovdqu SZ4*((\i-15)&0xf)(%rsp), \_T1
+        vmovdqu SZ4*((\i-2)&0xf)(%rsp), a1
+        vmovdqu \_T1, a0
+        PRORQ   \_T1,7
+        vmovdqu a1, a2
+        PRORQ   a1,42
+        vpxor   a0, \_T1, \_T1
+        PRORQ   \_T1, 1
+        vpxor   a2, a1, a1
+        PRORQ   a1, 19
+        vpsrlq  $7, a0, a0
+        vpxor   a0, \_T1, \_T1
+        vpsrlq  $6, a2, a2
+        vpxor   a2, a1, a1
+        vpaddq  SZ4*((\i-16)&0xf)(%rsp), \_T1, \_T1
+        vpaddq  SZ4*((\i-7)&0xf)(%rsp), a1, a1
+        vpaddq  a1, \_T1, \_T1
+
+        ROUND_00_15 \_T1,\i
+.endm
+
+
+# void sha512_x4_avx2(void *STATE, const int INP_SIZE)
+# arg 1 : STATE    : pointer to input data
+# arg 2 : INP_SIZE : size of data in blocks (assumed >= 1)
+ENTRY(sha512_x4_avx2)
+       # general registers preserved in outer calling routine
+       # outer calling routine saves all the XMM registers
+       # save callee-saved clobbered registers to comply with C function ABI
+       push    %r12
+       push    %r13
+       push    %r14
+       push    %r15
+
+       sub     $STACK_SPACE1, %rsp
+
+        # Load the pre-transposed incoming digest.
+        vmovdqu 0*SHA512_DIGEST_ROW_SIZE(STATE),a
+        vmovdqu 1*SHA512_DIGEST_ROW_SIZE(STATE),b
+        vmovdqu 2*SHA512_DIGEST_ROW_SIZE(STATE),c
+        vmovdqu 3*SHA512_DIGEST_ROW_SIZE(STATE),d
+        vmovdqu 4*SHA512_DIGEST_ROW_SIZE(STATE),e
+        vmovdqu 5*SHA512_DIGEST_ROW_SIZE(STATE),f
+        vmovdqu 6*SHA512_DIGEST_ROW_SIZE(STATE),g
+        vmovdqu 7*SHA512_DIGEST_ROW_SIZE(STATE),h
+
+        lea     K512_4(%rip),TBL
+
+        # load the address of each of the 4 message lanes
+        # getting ready to transpose input onto stack
+        mov     _data_ptr+0*PTR_SZ(STATE),inp0
+        mov     _data_ptr+1*PTR_SZ(STATE),inp1
+        mov     _data_ptr+2*PTR_SZ(STATE),inp2
+        mov     _data_ptr+3*PTR_SZ(STATE),inp3
+
+        xor     IDX, IDX
+lloop:
+        xor     ROUND, ROUND
+
+       # save old digest
+        vmovdqu a, _digest(%rsp)
+        vmovdqu b, _digest+1*SZ4(%rsp)
+        vmovdqu c, _digest+2*SZ4(%rsp)
+        vmovdqu d, _digest+3*SZ4(%rsp)
+        vmovdqu e, _digest+4*SZ4(%rsp)
+        vmovdqu f, _digest+5*SZ4(%rsp)
+        vmovdqu g, _digest+6*SZ4(%rsp)
+        vmovdqu h, _digest+7*SZ4(%rsp)
+        i = 0
+.rep 4
+       vmovdqu PSHUFFLE_BYTE_FLIP_MASK(%rip), TMP
+        VMOVPD  i*32(inp0, IDX), TT2
+        VMOVPD  i*32(inp1, IDX), TT1
+        VMOVPD  i*32(inp2, IDX), TT4
+        VMOVPD  i*32(inp3, IDX), TT3
+       TRANSPOSE       TT2, TT1, TT4, TT3, TT0, TT5
+       vpshufb TMP, TT0, TT0
+       vpshufb TMP, TT1, TT1
+       vpshufb TMP, TT2, TT2
+       vpshufb TMP, TT3, TT3
+       ROUND_00_15     TT0,(i*4+0)
+       ROUND_00_15     TT1,(i*4+1)
+       ROUND_00_15     TT2,(i*4+2)
+       ROUND_00_15     TT3,(i*4+3)
+       i = (i+1)
+.endr
+        add     $128, IDX
+
+        i = (i*4)
+
+        jmp     Lrounds_16_xx
+.align 16
+Lrounds_16_xx:
+.rep 16
+        ROUND_16_XX     T1, i
+        i = (i+1)
+.endr
+        cmp     $0xa00,ROUND
+        jb      Lrounds_16_xx
+
+       # add old digest
+        vpaddq  _digest(%rsp), a, a
+        vpaddq  _digest+1*SZ4(%rsp), b, b
+        vpaddq  _digest+2*SZ4(%rsp), c, c
+        vpaddq  _digest+3*SZ4(%rsp), d, d
+        vpaddq  _digest+4*SZ4(%rsp), e, e
+        vpaddq  _digest+5*SZ4(%rsp), f, f
+        vpaddq  _digest+6*SZ4(%rsp), g, g
+        vpaddq  _digest+7*SZ4(%rsp), h, h
+
+        sub     $1, INP_SIZE  # unit is blocks
+        jne     lloop
+
+        # write back to memory (state object) the transposed digest
+        vmovdqu a, 0*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu b, 1*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu c, 2*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu d, 3*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu e, 4*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu f, 5*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu g, 6*SHA512_DIGEST_ROW_SIZE(STATE)
+        vmovdqu h, 7*SHA512_DIGEST_ROW_SIZE(STATE)
+
+       # update input data pointers
+       add     IDX, inp0
+        mov     inp0, _data_ptr+0*PTR_SZ(STATE)
+        add     IDX, inp1
+        mov     inp1, _data_ptr+1*PTR_SZ(STATE)
+        add     IDX, inp2
+        mov     inp2, _data_ptr+2*PTR_SZ(STATE)
+        add     IDX, inp3
+        mov     inp3, _data_ptr+3*PTR_SZ(STATE)
+
+       #;;;;;;;;;;;;;;;
+       #; Postamble
+       add $STACK_SPACE1, %rsp
+       # restore callee-saved clobbered registers
+
+       pop     %r15
+       pop     %r14
+       pop     %r13
+       pop     %r12
+
+       # outer calling routine restores XMM and other GP registers
+       ret
+ENDPROC(sha512_x4_avx2)
+
+.data
+.align 64
+K512_4:
+       .octa 0x428a2f98d728ae22428a2f98d728ae22,\
+               0x428a2f98d728ae22428a2f98d728ae22
+       .octa 0x7137449123ef65cd7137449123ef65cd,\
+               0x7137449123ef65cd7137449123ef65cd
+       .octa 0xb5c0fbcfec4d3b2fb5c0fbcfec4d3b2f,\
+               0xb5c0fbcfec4d3b2fb5c0fbcfec4d3b2f
+       .octa 0xe9b5dba58189dbbce9b5dba58189dbbc,\
+               0xe9b5dba58189dbbce9b5dba58189dbbc
+       .octa 0x3956c25bf348b5383956c25bf348b538,\
+               0x3956c25bf348b5383956c25bf348b538
+       .octa 0x59f111f1b605d01959f111f1b605d019,\
+               0x59f111f1b605d01959f111f1b605d019
+       .octa 0x923f82a4af194f9b923f82a4af194f9b,\
+               0x923f82a4af194f9b923f82a4af194f9b
+       .octa 0xab1c5ed5da6d8118ab1c5ed5da6d8118,\
+               0xab1c5ed5da6d8118ab1c5ed5da6d8118
+       .octa 0xd807aa98a3030242d807aa98a3030242,\
+               0xd807aa98a3030242d807aa98a3030242
+       .octa 0x12835b0145706fbe12835b0145706fbe,\
+               0x12835b0145706fbe12835b0145706fbe
+       .octa 0x243185be4ee4b28c243185be4ee4b28c,\
+               0x243185be4ee4b28c243185be4ee4b28c
+       .octa 0x550c7dc3d5ffb4e2550c7dc3d5ffb4e2,\
+               0x550c7dc3d5ffb4e2550c7dc3d5ffb4e2
+       .octa 0x72be5d74f27b896f72be5d74f27b896f,\
+               0x72be5d74f27b896f72be5d74f27b896f
+       .octa 0x80deb1fe3b1696b180deb1fe3b1696b1,\
+               0x80deb1fe3b1696b180deb1fe3b1696b1
+       .octa 0x9bdc06a725c712359bdc06a725c71235,\
+               0x9bdc06a725c712359bdc06a725c71235
+       .octa 0xc19bf174cf692694c19bf174cf692694,\
+               0xc19bf174cf692694c19bf174cf692694
+       .octa 0xe49b69c19ef14ad2e49b69c19ef14ad2,\
+               0xe49b69c19ef14ad2e49b69c19ef14ad2
+       .octa 0xefbe4786384f25e3efbe4786384f25e3,\
+               0xefbe4786384f25e3efbe4786384f25e3
+       .octa 0x0fc19dc68b8cd5b50fc19dc68b8cd5b5,\
+               0x0fc19dc68b8cd5b50fc19dc68b8cd5b5
+       .octa 0x240ca1cc77ac9c65240ca1cc77ac9c65,\
+               0x240ca1cc77ac9c65240ca1cc77ac9c65
+       .octa 0x2de92c6f592b02752de92c6f592b0275,\
+               0x2de92c6f592b02752de92c6f592b0275
+       .octa 0x4a7484aa6ea6e4834a7484aa6ea6e483,\
+               0x4a7484aa6ea6e4834a7484aa6ea6e483
+       .octa 0x5cb0a9dcbd41fbd45cb0a9dcbd41fbd4,\
+               0x5cb0a9dcbd41fbd45cb0a9dcbd41fbd4
+       .octa 0x76f988da831153b576f988da831153b5,\
+               0x76f988da831153b576f988da831153b5
+       .octa 0x983e5152ee66dfab983e5152ee66dfab,\
+               0x983e5152ee66dfab983e5152ee66dfab
+       .octa 0xa831c66d2db43210a831c66d2db43210,\
+               0xa831c66d2db43210a831c66d2db43210
+       .octa 0xb00327c898fb213fb00327c898fb213f,\
+               0xb00327c898fb213fb00327c898fb213f
+       .octa 0xbf597fc7beef0ee4bf597fc7beef0ee4,\
+               0xbf597fc7beef0ee4bf597fc7beef0ee4
+       .octa 0xc6e00bf33da88fc2c6e00bf33da88fc2,\
+               0xc6e00bf33da88fc2c6e00bf33da88fc2
+       .octa 0xd5a79147930aa725d5a79147930aa725,\
+               0xd5a79147930aa725d5a79147930aa725
+       .octa 0x06ca6351e003826f06ca6351e003826f,\
+               0x06ca6351e003826f06ca6351e003826f
+       .octa 0x142929670a0e6e70142929670a0e6e70,\
+               0x142929670a0e6e70142929670a0e6e70
+       .octa 0x27b70a8546d22ffc27b70a8546d22ffc,\
+               0x27b70a8546d22ffc27b70a8546d22ffc
+       .octa 0x2e1b21385c26c9262e1b21385c26c926,\
+               0x2e1b21385c26c9262e1b21385c26c926
+       .octa 0x4d2c6dfc5ac42aed4d2c6dfc5ac42aed,\
+               0x4d2c6dfc5ac42aed4d2c6dfc5ac42aed
+       .octa 0x53380d139d95b3df53380d139d95b3df,\
+               0x53380d139d95b3df53380d139d95b3df
+       .octa 0x650a73548baf63de650a73548baf63de,\
+               0x650a73548baf63de650a73548baf63de
+       .octa 0x766a0abb3c77b2a8766a0abb3c77b2a8,\
+               0x766a0abb3c77b2a8766a0abb3c77b2a8
+       .octa 0x81c2c92e47edaee681c2c92e47edaee6,\
+               0x81c2c92e47edaee681c2c92e47edaee6
+       .octa 0x92722c851482353b92722c851482353b,\
+               0x92722c851482353b92722c851482353b
+       .octa 0xa2bfe8a14cf10364a2bfe8a14cf10364,\
+               0xa2bfe8a14cf10364a2bfe8a14cf10364
+       .octa 0xa81a664bbc423001a81a664bbc423001,\
+               0xa81a664bbc423001a81a664bbc423001
+       .octa 0xc24b8b70d0f89791c24b8b70d0f89791,\
+               0xc24b8b70d0f89791c24b8b70d0f89791
+       .octa 0xc76c51a30654be30c76c51a30654be30,\
+               0xc76c51a30654be30c76c51a30654be30
+       .octa 0xd192e819d6ef5218d192e819d6ef5218,\
+               0xd192e819d6ef5218d192e819d6ef5218
+       .octa 0xd69906245565a910d69906245565a910,\
+               0xd69906245565a910d69906245565a910
+       .octa 0xf40e35855771202af40e35855771202a,\
+               0xf40e35855771202af40e35855771202a
+       .octa 0x106aa07032bbd1b8106aa07032bbd1b8,\
+               0x106aa07032bbd1b8106aa07032bbd1b8
+       .octa 0x19a4c116b8d2d0c819a4c116b8d2d0c8,\
+               0x19a4c116b8d2d0c819a4c116b8d2d0c8
+       .octa 0x1e376c085141ab531e376c085141ab53,\
+               0x1e376c085141ab531e376c085141ab53
+       .octa 0x2748774cdf8eeb992748774cdf8eeb99,\
+               0x2748774cdf8eeb992748774cdf8eeb99
+       .octa 0x34b0bcb5e19b48a834b0bcb5e19b48a8,\
+               0x34b0bcb5e19b48a834b0bcb5e19b48a8
+       .octa 0x391c0cb3c5c95a63391c0cb3c5c95a63,\
+               0x391c0cb3c5c95a63391c0cb3c5c95a63
+       .octa 0x4ed8aa4ae3418acb4ed8aa4ae3418acb,\
+               0x4ed8aa4ae3418acb4ed8aa4ae3418acb
+       .octa 0x5b9cca4f7763e3735b9cca4f7763e373,\
+               0x5b9cca4f7763e3735b9cca4f7763e373
+       .octa 0x682e6ff3d6b2b8a3682e6ff3d6b2b8a3,\
+               0x682e6ff3d6b2b8a3682e6ff3d6b2b8a3
+       .octa 0x748f82ee5defb2fc748f82ee5defb2fc,\
+               0x748f82ee5defb2fc748f82ee5defb2fc
+       .octa 0x78a5636f43172f6078a5636f43172f60,\
+               0x78a5636f43172f6078a5636f43172f60
+       .octa 0x84c87814a1f0ab7284c87814a1f0ab72,\
+               0x84c87814a1f0ab7284c87814a1f0ab72
+       .octa 0x8cc702081a6439ec8cc702081a6439ec,\
+               0x8cc702081a6439ec8cc702081a6439ec
+       .octa 0x90befffa23631e2890befffa23631e28,\
+               0x90befffa23631e2890befffa23631e28
+       .octa 0xa4506cebde82bde9a4506cebde82bde9,\
+               0xa4506cebde82bde9a4506cebde82bde9
+       .octa 0xbef9a3f7b2c67915bef9a3f7b2c67915,\
+               0xbef9a3f7b2c67915bef9a3f7b2c67915
+       .octa 0xc67178f2e372532bc67178f2e372532b,\
+               0xc67178f2e372532bc67178f2e372532b
+       .octa 0xca273eceea26619cca273eceea26619c,\
+               0xca273eceea26619cca273eceea26619c
+       .octa 0xd186b8c721c0c207d186b8c721c0c207,\
+               0xd186b8c721c0c207d186b8c721c0c207
+       .octa 0xeada7dd6cde0eb1eeada7dd6cde0eb1e,\
+               0xeada7dd6cde0eb1eeada7dd6cde0eb1e
+       .octa 0xf57d4f7fee6ed178f57d4f7fee6ed178,\
+               0xf57d4f7fee6ed178f57d4f7fee6ed178
+       .octa 0x06f067aa72176fba06f067aa72176fba,\
+               0x06f067aa72176fba06f067aa72176fba
+       .octa 0x0a637dc5a2c898a60a637dc5a2c898a6,\
+               0x0a637dc5a2c898a60a637dc5a2c898a6
+       .octa 0x113f9804bef90dae113f9804bef90dae,\
+               0x113f9804bef90dae113f9804bef90dae
+       .octa 0x1b710b35131c471b1b710b35131c471b,\
+               0x1b710b35131c471b1b710b35131c471b
+       .octa 0x28db77f523047d8428db77f523047d84,\
+               0x28db77f523047d8428db77f523047d84
+       .octa 0x32caab7b40c7249332caab7b40c72493,\
+               0x32caab7b40c7249332caab7b40c72493
+       .octa 0x3c9ebe0a15c9bebc3c9ebe0a15c9bebc,\
+               0x3c9ebe0a15c9bebc3c9ebe0a15c9bebc
+       .octa 0x431d67c49c100d4c431d67c49c100d4c,\
+               0x431d67c49c100d4c431d67c49c100d4c
+       .octa 0x4cc5d4becb3e42b64cc5d4becb3e42b6,\
+               0x4cc5d4becb3e42b64cc5d4becb3e42b6
+       .octa 0x597f299cfc657e2a597f299cfc657e2a,\
+               0x597f299cfc657e2a597f299cfc657e2a
+       .octa 0x5fcb6fab3ad6faec5fcb6fab3ad6faec,\
+               0x5fcb6fab3ad6faec5fcb6fab3ad6faec
+       .octa 0x6c44198c4a4758176c44198c4a475817,\
+               0x6c44198c4a4758176c44198c4a475817
+
+PSHUFFLE_BYTE_FLIP_MASK: .octa 0x08090a0b0c0d0e0f0001020304050607
+                         .octa 0x18191a1b1c1d1e1f1011121314151617