crypto: sha256 - Optimized sha256 x86_64 assembly routine with AVX instructions.

Provides SHA256 x86_64 assembly routine optimized with SSE and AVX instructions.
Speedup of 60% or more has been measured over the generic implementation.

Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/sha256-avx-asm.S

@@ -0,0 +1,496 @@
+########################################################################
+# Implement fast SHA-256 with AVX1 instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+#     James Guilford <james.guilford@intel.com>
+#     Kirk Yap <kirk.s.yap@intel.com>
+#     Tim Chen <tim.c.chen@linux.intel.com>
+#
+# This software is available to you under a choice of one of two
+# licenses.  You may choose to be licensed under the terms of the GNU
+# General Public License (GPL) Version 2, available from the file
+# COPYING in the main directory of this source tree, or the
+# OpenIB.org BSD license below:
+#
+#     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.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+# BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+# ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+########################################################################
+#
+# This code is described in an Intel White-Paper:
+# "Fast SHA-256 Implementations on Intel Architecture Processors"
+#
+# To find it, surf to http://www.intel.com/p/en_US/embedded
+# and search for that title.
+#
+########################################################################
+# This code schedules 1 block at a time, with 4 lanes per block
+########################################################################
+
+#ifdef CONFIG_AS_AVX
+#include <linux/linkage.h>
+
+## assume buffers not aligned
+#define    VMOVDQ vmovdqu
+
+################################ Define Macros
+
+# addm [mem], reg
+# Add reg to mem using reg-mem add and store
+.macro addm p1 p2
+	add     \p1, \p2
+	mov     \p2, \p1
+.endm
+
+
+.macro MY_ROR p1 p2
+	shld    $(32-(\p1)), \p2, \p2
+.endm
+
+################################
+
+# COPY_XMM_AND_BSWAP xmm, [mem], byte_flip_mask
+# Load xmm with mem and byte swap each dword
+.macro COPY_XMM_AND_BSWAP p1 p2 p3
+	VMOVDQ \p2, \p1
+	vpshufb \p3, \p1, \p1
+.endm
+
+################################
+
+X0 = %xmm4
+X1 = %xmm5
+X2 = %xmm6
+X3 = %xmm7
+
+XTMP0 = %xmm0
+XTMP1 = %xmm1
+XTMP2 = %xmm2
+XTMP3 = %xmm3
+XTMP4 = %xmm8
+XFER = %xmm9
+XTMP5 = %xmm11
+
+SHUF_00BA = %xmm10      # shuffle xBxA -> 00BA
+SHUF_DC00 = %xmm12      # shuffle xDxC -> DC00
+BYTE_FLIP_MASK = %xmm13
+
+NUM_BLKS = %rdx   # 3rd arg
+CTX = %rsi        # 2nd arg
+INP = %rdi        # 1st arg
+
+SRND = %rdi       # clobbers INP
+c = %ecx
+d = %r8d
+e = %edx
+TBL = %rbp
+a = %eax
+b = %ebx
+
+f = %r9d
+g = %r10d
+h = %r11d
+
+y0 = %r13d
+y1 = %r14d
+y2 = %r15d
+
+
+_INP_END_SIZE = 8
+_INP_SIZE = 8
+_XFER_SIZE = 8
+_XMM_SAVE_SIZE = 0
+
+_INP_END = 0
+_INP            = _INP_END  + _INP_END_SIZE
+_XFER           = _INP      + _INP_SIZE
+_XMM_SAVE       = _XFER     + _XFER_SIZE
+STACK_SIZE      = _XMM_SAVE + _XMM_SAVE_SIZE
+
+# rotate_Xs
+# Rotate values of symbols X0...X3
+.macro rotate_Xs
+X_ = X0
+X0 = X1
+X1 = X2
+X2 = X3
+X3 = X_
+.endm
+
+# ROTATE_ARGS
+# Rotate values of symbols a...h
+.macro ROTATE_ARGS
+TMP_ = h
+h = g
+g = f
+f = e
+e = d
+d = c
+c = b
+b = a
+a = TMP_
+.endm
+
+.macro FOUR_ROUNDS_AND_SCHED
+	## compute s0 four at a time and s1 two at a time
+	## compute W[-16] + W[-7] 4 at a time
+
+	mov     e, y0			# y0 = e
+	MY_ROR  (25-11), y0             # y0 = e >> (25-11)
+	mov     a, y1                   # y1 = a
+	vpalignr $4, X2, X3, XTMP0      # XTMP0 = W[-7]
+	MY_ROR  (22-13), y1             # y1 = a >> (22-13)
+	xor     e, y0                   # y0 = e ^ (e >> (25-11))
+	mov     f, y2                   # y2 = f
+	MY_ROR  (11-6), y0              # y0 = (e >> (11-6)) ^ (e >> (25-6))
+	xor     a, y1                   # y1 = a ^ (a >> (22-13)
+	xor     g, y2                   # y2 = f^g
+	vpaddd  X0, XTMP0, XTMP0        # XTMP0 = W[-7] + W[-16]
+	xor     e, y0                   # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+	and     e, y2                   # y2 = (f^g)&e
+	MY_ROR  (13-2), y1              # y1 = (a >> (13-2)) ^ (a >> (22-2))
+	## compute s0
+	vpalignr $4, X0, X1, XTMP1      # XTMP1 = W[-15]
+	xor     a, y1                   # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+	MY_ROR  6, y0                   # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+	xor     g, y2                   # y2 = CH = ((f^g)&e)^g
+	MY_ROR  2, y1                   # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+	add     y0, y2                  # y2 = S1 + CH
+	add     _XFER(%rsp), y2         # y2 = k + w + S1 + CH
+	mov     a, y0                   # y0 = a
+	add     y2, h                   # h = h + S1 + CH + k + w
+	mov     a, y2                   # y2 = a
+	vpsrld  $7, XTMP1, XTMP2
+	or      c, y0                   # y0 = a|c
+	add     h, d                    # d = d + h + S1 + CH + k + w
+	and     c, y2                   # y2 = a&c
+	vpslld  $(32-7), XTMP1, XTMP3
+	and     b, y0                   # y0 = (a|c)&b
+	add     y1, h                   # h = h + S1 + CH + k + w + S0
+	vpor    XTMP2, XTMP3, XTMP3     # XTMP1 = W[-15] MY_ROR 7
+	or      y2, y0                  # y0 = MAJ = (a|c)&b)|(a&c)
+	add     y0, h                   # h = h + S1 + CH + k + w + S0 + MAJ
+	ROTATE_ARGS
+	mov     e, y0                   # y0 = e
+	mov     a, y1                   # y1 = a
+	MY_ROR  (25-11), y0             # y0 = e >> (25-11)
+	xor     e, y0                   # y0 = e ^ (e >> (25-11))
+	mov     f, y2                   # y2 = f
+	MY_ROR  (22-13), y1             # y1 = a >> (22-13)
+	vpsrld  $18, XTMP1, XTMP2       #
+	xor     a, y1                   # y1 = a ^ (a >> (22-13)
+	MY_ROR  (11-6), y0              # y0 = (e >> (11-6)) ^ (e >> (25-6))
+	xor     g, y2                   # y2 = f^g
+	vpsrld  $3, XTMP1, XTMP4        # XTMP4 = W[-15] >> 3
+	MY_ROR  (13-2), y1              # y1 = (a >> (13-2)) ^ (a >> (22-2))
+	xor     e, y0                   # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+	and     e, y2                   # y2 = (f^g)&e
+	MY_ROR  6, y0                   # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+	vpslld  $(32-18), XTMP1, XTMP1
+	xor     a, y1                   # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+	xor     g, y2                   # y2 = CH = ((f^g)&e)^g
+	vpxor   XTMP1, XTMP3, XTMP3     #
+	add     y0, y2                  # y2 = S1 + CH
+	add     (1*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+	MY_ROR  2, y1                   # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+	vpxor   XTMP2, XTMP3, XTMP3     # XTMP1 = W[-15] MY_ROR 7 ^ W[-15] MY_ROR
+	mov     a, y0                   # y0 = a
+	add     y2, h                   # h = h + S1 + CH + k + w
+	mov     a, y2                   # y2 = a
+	vpxor   XTMP4, XTMP3, XTMP1     # XTMP1 = s0
+	or      c, y0                   # y0 = a|c
+	add     h, d                    # d = d + h + S1 + CH + k + w
+	and     c, y2                   # y2 = a&c
+	## compute low s1
+	vpshufd $0b11111010, X3, XTMP2  # XTMP2 = W[-2] {BBAA}
+	and     b, y0                   # y0 = (a|c)&b
+	add     y1, h                   # h = h + S1 + CH + k + w + S0
+	vpaddd  XTMP1, XTMP0, XTMP0     # XTMP0 = W[-16] + W[-7] + s0
+	or      y2, y0                  # y0 = MAJ = (a|c)&b)|(a&c)
+	add     y0, h                   # h = h + S1 + CH + k + w + S0 + MAJ
+	ROTATE_ARGS
+	mov     e, y0                   # y0 = e
+	mov     a, y1                   # y1 = a
+	MY_ROR  (25-11), y0             # y0 = e >> (25-11)
+	xor     e, y0                   # y0 = e ^ (e >> (25-11))
+	MY_ROR  (22-13), y1             # y1 = a >> (22-13)
+	mov     f, y2                   # y2 = f
+	xor     a, y1                   # y1 = a ^ (a >> (22-13)
+	MY_ROR  (11-6), y0              # y0 = (e >> (11-6)) ^ (e >> (25-6))
+	vpsrld  $10, XTMP2, XTMP4       # XTMP4 = W[-2] >> 10 {BBAA}
+	xor     g, y2                   # y2 = f^g
+	vpsrlq  $19, XTMP2, XTMP3       # XTMP3 = W[-2] MY_ROR 19 {xBxA}
+	xor     e, y0                   # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+	and     e, y2                   # y2 = (f^g)&e
+	vpsrlq  $17, XTMP2, XTMP2       # XTMP2 = W[-2] MY_ROR 17 {xBxA}
+	MY_ROR  (13-2), y1              # y1 = (a >> (13-2)) ^ (a >> (22-2))
+	xor     a, y1                   # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+	xor     g, y2                   # y2 = CH = ((f^g)&e)^g
+	MY_ROR  6, y0                   # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+	vpxor   XTMP3, XTMP2, XTMP2     #
+	add     y0, y2                  # y2 = S1 + CH
+	MY_ROR  2, y1                   # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+	add     (2*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+	vpxor   XTMP2, XTMP4, XTMP4     # XTMP4 = s1 {xBxA}
+	mov     a, y0                   # y0 = a
+	add     y2, h                   # h = h + S1 + CH + k + w
+	mov     a, y2                   # y2 = a
+	vpshufb SHUF_00BA, XTMP4, XTMP4 # XTMP4 = s1 {00BA}
+	or      c, y0                   # y0 = a|c
+	add     h, d                    # d = d + h + S1 + CH + k + w
+	and     c, y2                   # y2 = a&c
+	vpaddd  XTMP4, XTMP0, XTMP0     # XTMP0 = {..., ..., W[1], W[0]}
+	and     b, y0                   # y0 = (a|c)&b
+	add     y1, h                   # h = h + S1 + CH + k + w + S0
+	## compute high s1
+	vpshufd $0b01010000, XTMP0, XTMP2 # XTMP2 = W[-2] {DDCC}
+	or      y2, y0                  # y0 = MAJ = (a|c)&b)|(a&c)
+	add     y0, h                   # h = h + S1 + CH + k + w + S0 + MAJ
+	ROTATE_ARGS
+	mov     e, y0                   # y0 = e
+	MY_ROR  (25-11), y0             # y0 = e >> (25-11)
+	mov     a, y1                   # y1 = a
+	MY_ROR  (22-13), y1             # y1 = a >> (22-13)
+	xor     e, y0                   # y0 = e ^ (e >> (25-11))
+	mov     f, y2                   # y2 = f
+	MY_ROR  (11-6), y0              # y0 = (e >> (11-6)) ^ (e >> (25-6))
+	vpsrld  $10, XTMP2, XTMP5       # XTMP5 = W[-2] >> 10 {DDCC}
+	xor     a, y1                   # y1 = a ^ (a >> (22-13)
+	xor     g, y2                   # y2 = f^g
+	vpsrlq  $19, XTMP2, XTMP3       # XTMP3 = W[-2] MY_ROR 19 {xDxC}
+	xor     e, y0                   # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+	and     e, y2                   # y2 = (f^g)&e
+	MY_ROR  (13-2), y1              # y1 = (a >> (13-2)) ^ (a >> (22-2))
+	vpsrlq  $17, XTMP2, XTMP2       # XTMP2 = W[-2] MY_ROR 17 {xDxC}
+	xor     a, y1                   # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+	MY_ROR  6, y0                   # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+	xor     g, y2                   # y2 = CH = ((f^g)&e)^g
+	vpxor   XTMP3, XTMP2, XTMP2
+	MY_ROR  2, y1                   # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+	add     y0, y2                  # y2 = S1 + CH
+	add     (3*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+	vpxor   XTMP2, XTMP5, XTMP5     # XTMP5 = s1 {xDxC}
+	mov     a, y0                   # y0 = a
+	add     y2, h                   # h = h + S1 + CH + k + w
+	mov     a, y2                   # y2 = a
+	vpshufb SHUF_DC00, XTMP5, XTMP5 # XTMP5 = s1 {DC00}
+	or      c, y0                   # y0 = a|c
+	add     h, d                    # d = d + h + S1 + CH + k + w
+	and     c, y2                   # y2 = a&c
+	vpaddd  XTMP0, XTMP5, X0        # X0 = {W[3], W[2], W[1], W[0]}
+	and     b, y0                   # y0 = (a|c)&b
+	add     y1, h                   # h = h + S1 + CH + k + w + S0
+	or      y2, y0                  # y0 = MAJ = (a|c)&b)|(a&c)
+	add     y0, h                   # h = h + S1 + CH + k + w + S0 + MAJ
+	ROTATE_ARGS
+	rotate_Xs
+.endm
+
+## input is [rsp + _XFER + %1 * 4]
+.macro DO_ROUND round
+	mov	e, y0			# y0 = e
+        MY_ROR  (25-11), y0             # y0 = e >> (25-11)
+        mov     a, y1                   # y1 = a
+        xor     e, y0                   # y0 = e ^ (e >> (25-11))
+        MY_ROR  (22-13), y1             # y1 = a >> (22-13)
+        mov     f, y2                   # y2 = f
+        xor     a, y1                   # y1 = a ^ (a >> (22-13)
+        MY_ROR  (11-6), y0              # y0 = (e >> (11-6)) ^ (e >> (25-6))
+        xor     g, y2                   # y2 = f^g
+        xor     e, y0                   # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+        MY_ROR  (13-2), y1              # y1 = (a >> (13-2)) ^ (a >> (22-2))
+        and     e, y2                   # y2 = (f^g)&e
+        xor     a, y1                   # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+        MY_ROR  6, y0                   # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+        xor     g, y2                   # y2 = CH = ((f^g)&e)^g
+        add     y0, y2                  # y2 = S1 + CH
+        MY_ROR  2, y1                   # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+        offset = \round * 4 + _XFER     #
+        add     offset(%rsp), y2	# y2 = k + w + S1 + CH
+        mov     a, y0			# y0 = a
+        add     y2, h                   # h = h + S1 + CH + k + w
+        mov     a, y2                   # y2 = a
+        or      c, y0                   # y0 = a|c
+        add     h, d                    # d = d + h + S1 + CH + k + w
+        and     c, y2                   # y2 = a&c
+        and     b, y0                   # y0 = (a|c)&b
+        add     y1, h                   # h = h + S1 + CH + k + w + S0
+        or      y2, y0                  # y0 = MAJ = (a|c)&b)|(a&c)
+        add     y0, h                   # h = h + S1 + CH + k + w + S0 + MAJ
+        ROTATE_ARGS
+.endm
+
+########################################################################
+## void sha256_transform_avx(void *input_data, UINT32 digest[8], UINT64 num_blks)
+## arg 1 : pointer to input data
+## arg 2 : pointer to digest
+## arg 3 : Num blocks
+########################################################################
+.text
+ENTRY(sha256_transform_avx)
+.align 32
+	pushq   %rbx
+	pushq   %rbp
+	pushq   %r13
+	pushq   %r14
+	pushq   %r15
+	pushq   %r12
+
+	mov	%rsp, %r12
+	subq    $STACK_SIZE, %rsp	# allocate stack space
+	and	$~15, %rsp		# align stack pointer
+
+	shl     $6, NUM_BLKS		# convert to bytes
+	jz      done_hash
+	add     INP, NUM_BLKS		# pointer to end of data
+	mov     NUM_BLKS, _INP_END(%rsp)
+
+	## load initial digest
+	mov     4*0(CTX), a
+	mov     4*1(CTX), b
+	mov     4*2(CTX), c
+	mov     4*3(CTX), d
+	mov     4*4(CTX), e
+	mov     4*5(CTX), f
+	mov     4*6(CTX), g
+	mov     4*7(CTX), h
+
+	vmovdqa  PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
+	vmovdqa  _SHUF_00BA(%rip), SHUF_00BA
+	vmovdqa  _SHUF_DC00(%rip), SHUF_DC00
+loop0:
+	lea     K256(%rip), TBL
+
+	## byte swap first 16 dwords
+	COPY_XMM_AND_BSWAP      X0, 0*16(INP), BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP      X1, 1*16(INP), BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP      X2, 2*16(INP), BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP      X3, 3*16(INP), BYTE_FLIP_MASK
+
+	mov     INP, _INP(%rsp)
+
+	## schedule 48 input dwords, by doing 3 rounds of 16 each
+	mov     $3, SRND
+.align 16
+loop1:
+	vpaddd  (TBL), X0, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	FOUR_ROUNDS_AND_SCHED
+
+	vpaddd  1*16(TBL), X0, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	FOUR_ROUNDS_AND_SCHED
+
+	vpaddd  2*16(TBL), X0, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	FOUR_ROUNDS_AND_SCHED
+
+	vpaddd  3*16(TBL), X0, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	add	$4*16, TBL
+	FOUR_ROUNDS_AND_SCHED
+
+	sub     $1, SRND
+	jne     loop1
+
+	mov     $2, SRND
+loop2:
+	vpaddd  (TBL), X0, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	DO_ROUND        0
+	DO_ROUND        1
+	DO_ROUND        2
+	DO_ROUND        3
+
+	vpaddd  1*16(TBL), X1, XFER
+	vmovdqa XFER, _XFER(%rsp)
+	add     $2*16, TBL
+	DO_ROUND        0
+	DO_ROUND        1
+	DO_ROUND        2
+	DO_ROUND        3
+
+	vmovdqa X2, X0
+	vmovdqa X3, X1
+
+	sub     $1, SRND
+	jne     loop2
+
+	addm    (4*0)(CTX),a
+	addm    (4*1)(CTX),b
+	addm    (4*2)(CTX),c
+	addm    (4*3)(CTX),d
+	addm    (4*4)(CTX),e
+	addm    (4*5)(CTX),f
+	addm    (4*6)(CTX),g
+	addm    (4*7)(CTX),h
+
+	mov     _INP(%rsp), INP
+	add     $64, INP
+	cmp     _INP_END(%rsp), INP
+	jne     loop0
+
+done_hash:
+
+	mov	%r12, %rsp
+
+	popq	%r12
+	popq    %r15
+	popq    %r14
+	popq    %r13
+	popq    %rbp
+	popq    %rbx
+	ret
+ENDPROC(sha256_transform_avx)
+
+.data
+.align 64
+K256:
+	.long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+	.long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+	.long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+	.long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+	.long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+	.long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+	.long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+	.long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+	.long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+	.long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+	.long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+	.long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+	.long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+	.long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+	.long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+	.long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+
+PSHUFFLE_BYTE_FLIP_MASK:
+	.octa 0x0c0d0e0f08090a0b0405060700010203
+
+# shuffle xBxA -> 00BA
+_SHUF_00BA:
+	.octa 0xFFFFFFFFFFFFFFFF0b0a090803020100
+
+# shuffle xDxC -> DC00
+_SHUF_DC00:
+	.octa 0x0b0a090803020100FFFFFFFFFFFFFFFF
+#endif