diff mbox

[5/7] crypto: sha256-mb - Crypto computation (x8 AVX2)

Message ID 1466732448-27856-6-git-send-email-megha.dey@intel.com (mailing list archive)
State Accepted
Delegated to: Herbert Xu
Headers show

Commit Message

Dey, Megha June 24, 2016, 1:40 a.m. UTC
This patch introduces the assembly routines to do SHA256 computation
on buffers belonging to several jobs at once.  The assembly routines
are optimized with AVX2 instructions that have 8 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>
---
 arch/x86/crypto/sha256-mb/sha256_x8_avx2.S | 593 +++++++++++++++++++++++++++++
 1 file changed, 593 insertions(+)
 create mode 100644 arch/x86/crypto/sha256-mb/sha256_x8_avx2.S
diff mbox

Patch

diff --git a/arch/x86/crypto/sha256-mb/sha256_x8_avx2.S b/arch/x86/crypto/sha256-mb/sha256_x8_avx2.S
new file mode 100644
index 0000000..aa21aea
--- /dev/null
+++ b/arch/x86/crypto/sha256-mb/sha256_x8_avx2.S
@@ -0,0 +1,593 @@ 
+/*
+ * Multi-buffer SHA256 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.
+ */
+
+#include <linux/linkage.h>
+#include "sha256_mb_mgr_datastruct.S"
+
+## code to compute oct SHA256 using SSE-256
+## 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
+## Linux clobbers:    rax rbx rcx rdx rsi            r9 r10 r11 r12 r13 r14 r15
+## Linux preserves:                       rdi rbp r8
+##
+## clobbers %ymm0-15
+
+arg1 = %rdi
+arg2 = %rsi
+reg3 = %rcx
+reg4 = %rdx
+
+# Common definitions
+STATE = arg1
+INP_SIZE = arg2
+
+IDX = %rax
+ROUND = %rbx
+TBL = reg3
+
+inp0 = %r9
+inp1 = %r10
+inp2 = %r11
+inp3 = %r12
+inp4 = %r13
+inp5 = %r14
+inp6 = %r15
+inp7 = reg4
+
+a = %ymm0
+b = %ymm1
+c = %ymm2
+d = %ymm3
+e = %ymm4
+f = %ymm5
+g = %ymm6
+h = %ymm7
+
+T1 = %ymm8
+
+a0 = %ymm12
+a1 = %ymm13
+a2 = %ymm14
+TMP = %ymm15
+TMP0 = %ymm6
+TMP1 = %ymm7
+
+TT0 = %ymm8
+TT1 = %ymm9
+TT2 = %ymm10
+TT3 = %ymm11
+TT4 = %ymm12
+TT5 = %ymm13
+TT6 = %ymm14
+TT7 = %ymm15
+
+# Define stack usage
+
+# Assume stack aligned to 32 bytes before call
+# Therefore FRAMESZ mod 32 must be 32-8 = 24
+
+#define FRAMESZ	0x388
+
+#define VMOVPS	vmovups
+
+# TRANSPOSE8 r0, r1, r2, r3, r4, r5, r6, r7, t0, t1
+# "transpose" data in {r0...r7} using temps {t0...t1}
+# Input looks like: {r0 r1 r2 r3 r4 r5 r6 r7}
+# 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}
+# r4 = {e7 e6 e5 e4   e3 e2 e1 e0}
+# r5 = {f7 f6 f5 f4   f3 f2 f1 f0}
+# r6 = {g7 g6 g5 g4   g3 g2 g1 g0}
+# r7 = {h7 h6 h5 h4   h3 h2 h1 h0}
+#
+# Output looks like: {r0 r1 r2 r3 r4 r5 r6 r7}
+# r0 = {h0 g0 f0 e0   d0 c0 b0 a0}
+# r1 = {h1 g1 f1 e1   d1 c1 b1 a1}
+# r2 = {h2 g2 f2 e2   d2 c2 b2 a2}
+# r3 = {h3 g3 f3 e3   d3 c3 b3 a3}
+# r4 = {h4 g4 f4 e4   d4 c4 b4 a4}
+# r5 = {h5 g5 f5 e5   d5 c5 b5 a5}
+# r6 = {h6 g6 f6 e6   d6 c6 b6 a6}
+# r7 = {h7 g7 f7 e7   d7 c7 b7 a7}
+#
+
+.macro TRANSPOSE8 r0 r1 r2 r3 r4 r5 r6 r7 t0 t1
+	# process top half (r0..r3) {a...d}
+	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}
+	vshufps	$0xDD, \t1, \t0, \r3 # r3 = {d5 c5 b5 a5   d1 c1 b1 a1}
+	vshufps	$0x88, \r2, \r0, \r1 # r1 = {d6 c6 b6 a6   d2 c2 b2 a2}
+	vshufps	$0xDD, \r2, \r0, \r0 # r0 = {d7 c7 b7 a7   d3 c3 b3 a3}
+	vshufps	$0x88, \t1, \t0, \t0 # t0 = {d4 c4 b4 a4   d0 c0 b0 a0}
+
+	# use r2 in place of t0
+	# process bottom half (r4..r7) {e...h}
+	vshufps	$0x44, \r5, \r4, \r2 # r2 = {f5 f4 e5 e4   f1 f0 e1 e0}
+	vshufps	$0xEE, \r5, \r4, \r4 # r4 = {f7 f6 e7 e6   f3 f2 e3 e2}
+	vshufps	$0x44, \r7, \r6, \t1 # t1 = {h5 h4 g5 g4   h1 h0 g1 g0}
+	vshufps	$0xEE, \r7, \r6, \r6 # r6 = {h7 h6 g7 g6   h3 h2 g3 g2}
+	vshufps	$0xDD, \t1, \r2, \r7 # r7 = {h5 g5 f5 e5   h1 g1 f1 e1}
+	vshufps	$0x88, \r6, \r4, \r5 # r5 = {h6 g6 f6 e6   h2 g2 f2 e2}
+	vshufps	$0xDD, \r6, \r4, \r4 # r4 = {h7 g7 f7 e7   h3 g3 f3 e3}
+	vshufps	$0x88, \t1, \r2, \t1 # t1 = {h4 g4 f4 e4   h0 g0 f0 e0}
+
+	vperm2f128	$0x13, \r1, \r5, \r6  # h6...a6
+	vperm2f128	$0x02, \r1, \r5, \r2  # h2...a2
+	vperm2f128	$0x13, \r3, \r7, \r5  # h5...a5
+	vperm2f128	$0x02, \r3, \r7, \r1  # h1...a1
+	vperm2f128	$0x13, \r0, \r4, \r7  # h7...a7
+	vperm2f128	$0x02, \r0, \r4, \r3  # h3...a3
+	vperm2f128	$0x13, \t0, \t1, \r4  # h4...a4
+	vperm2f128	$0x02, \t0, \t1, \r0  # h0...a0
+
+.endm
+
+.macro ROTATE_ARGS
+TMP_ = h
+h = g
+g = f
+f = e
+e = d
+d = c
+c = b
+b = a
+a = TMP_
+.endm
+
+.macro _PRORD reg imm tmp
+	vpslld	$(32-\imm),\reg,\tmp
+	vpsrld	$\imm,\reg, \reg
+	vpor	\tmp,\reg, \reg
+.endm
+
+# PRORD_nd reg, imm, tmp, src
+.macro _PRORD_nd reg imm tmp src
+	vpslld	$(32-\imm), \src, \tmp
+	vpsrld	$\imm, \src, \reg
+	vpor	\tmp, \reg, \reg
+.endm
+
+# PRORD dst/src, amt
+.macro PRORD reg imm
+	_PRORD	\reg,\imm,TMP
+.endm
+
+# PRORD_nd dst, src, amt
+.macro PRORD_nd reg tmp imm
+	_PRORD_nd	\reg, \imm, TMP, \tmp
+.endm
+
+# arguments passed implicitly in preprocessor symbols i, a...h
+.macro ROUND_00_15 _T1 i
+	PRORD_nd	a0,e,5	# sig1: a0 = (e >> 5)
+
+	vpxor	g, f, a2	# ch: a2 = f^g
+	vpand	e,a2, a2	# ch: a2 = (f^g)&e
+	vpxor	g, a2, a2	# a2 = ch
+
+	PRORD_nd	a1,e,25	# sig1: a1 = (e >> 25)
+
+	vmovdqu	\_T1,(SZ8*(\i & 0xf))(%rsp)
+	vpaddd	(TBL,ROUND,1), \_T1, \_T1	# T1 = W + K
+	vpxor	e,a0, a0	# sig1: a0 = e ^ (e >> 5)
+	PRORD	a0, 6		# sig1: a0 = (e >> 6) ^ (e >> 11)
+	vpaddd	a2, h, h	# h = h + ch
+	PRORD_nd	a2,a,11	# sig0: a2 = (a >> 11)
+	vpaddd	\_T1,h, h 	# h = h + ch + W + K
+	vpxor	a1, a0, a0	# a0 = sigma1
+	PRORD_nd	a1,a,22	# sig0: a1 = (a >> 22)
+	vpxor	c, a, \_T1	# maj: T1 = a^c
+	add	$SZ8, ROUND	# ROUND++
+	vpand	b, \_T1, \_T1	# maj: T1 = (a^c)&b
+	vpaddd	a0, h, h
+	vpaddd	h, d, d
+	vpxor	a, a2, a2	# sig0: a2 = a ^ (a >> 11)
+	PRORD	a2,2		# 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
+	vpaddd	a1, h, h	# h = h + ch + W + K + maj
+	vpaddd	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	(SZ8*((\i-15)&0xf))(%rsp), \_T1
+	vmovdqu	(SZ8*((\i-2)&0xf))(%rsp), a1
+	vmovdqu	\_T1, a0
+	PRORD	\_T1,11
+	vmovdqu	a1, a2
+	PRORD	a1,2
+	vpxor	a0, \_T1, \_T1
+	PRORD	\_T1, 7
+	vpxor	a2, a1, a1
+	PRORD	a1, 17
+	vpsrld	$3, a0, a0
+	vpxor	a0, \_T1, \_T1
+	vpsrld	$10, a2, a2
+	vpxor	a2, a1, a1
+	vpaddd	(SZ8*((\i-16)&0xf))(%rsp), \_T1, \_T1
+	vpaddd	(SZ8*((\i-7)&0xf))(%rsp), a1, a1
+	vpaddd	a1, \_T1, \_T1
+
+	ROUND_00_15 \_T1,\i
+.endm
+
+# SHA256_ARGS:
+#   UINT128 digest[8];  // transposed digests
+#   UINT8  *data_ptr[4];
+
+# void sha256_x8_avx2(SHA256_ARGS *args, UINT64 bytes);
+# arg 1 : STATE : pointer to array of pointers to input data
+# arg 2 : INP_SIZE  : size of input in blocks
+	# general registers preserved in outer calling routine
+	# outer calling routine saves all the XMM registers
+	# save rsp, allocate 32-byte aligned for local variables
+ENTRY(sha256_x8_avx2)
+
+	# save callee-saved clobbered registers to comply with C function ABI
+	push    %r12
+	push    %r13
+	push    %r14
+	push    %r15
+
+	mov	%rsp, IDX
+	sub	$FRAMESZ, %rsp
+	and	$~0x1F, %rsp
+	mov	IDX, _rsp(%rsp)
+
+	# Load the pre-transposed incoming digest.
+	vmovdqu	0*SHA256_DIGEST_ROW_SIZE(STATE),a
+	vmovdqu	1*SHA256_DIGEST_ROW_SIZE(STATE),b
+	vmovdqu	2*SHA256_DIGEST_ROW_SIZE(STATE),c
+	vmovdqu	3*SHA256_DIGEST_ROW_SIZE(STATE),d
+	vmovdqu	4*SHA256_DIGEST_ROW_SIZE(STATE),e
+	vmovdqu	5*SHA256_DIGEST_ROW_SIZE(STATE),f
+	vmovdqu	6*SHA256_DIGEST_ROW_SIZE(STATE),g
+	vmovdqu	7*SHA256_DIGEST_ROW_SIZE(STATE),h
+
+	lea	K256_8(%rip),TBL
+
+	# load the address of each of the 4 message lanes
+	# getting ready to transpose input onto stack
+	mov	_args_data_ptr+0*PTR_SZ(STATE),inp0
+	mov	_args_data_ptr+1*PTR_SZ(STATE),inp1
+	mov	_args_data_ptr+2*PTR_SZ(STATE),inp2
+	mov	_args_data_ptr+3*PTR_SZ(STATE),inp3
+	mov	_args_data_ptr+4*PTR_SZ(STATE),inp4
+	mov	_args_data_ptr+5*PTR_SZ(STATE),inp5
+	mov	_args_data_ptr+6*PTR_SZ(STATE),inp6
+	mov	_args_data_ptr+7*PTR_SZ(STATE),inp7
+
+	xor	IDX, IDX
+lloop:
+	xor	ROUND, ROUND
+
+	# save old digest
+	vmovdqu	a, _digest(%rsp)
+	vmovdqu	b, _digest+1*SZ8(%rsp)
+	vmovdqu	c, _digest+2*SZ8(%rsp)
+	vmovdqu	d, _digest+3*SZ8(%rsp)
+	vmovdqu	e, _digest+4*SZ8(%rsp)
+	vmovdqu	f, _digest+5*SZ8(%rsp)
+	vmovdqu	g, _digest+6*SZ8(%rsp)
+	vmovdqu	h, _digest+7*SZ8(%rsp)
+	i = 0
+.rep 2
+	VMOVPS	i*32(inp0, IDX), TT0
+	VMOVPS	i*32(inp1, IDX), TT1
+	VMOVPS	i*32(inp2, IDX), TT2
+	VMOVPS	i*32(inp3, IDX), TT3
+	VMOVPS	i*32(inp4, IDX), TT4
+	VMOVPS	i*32(inp5, IDX), TT5
+	VMOVPS	i*32(inp6, IDX), TT6
+	VMOVPS	i*32(inp7, IDX), TT7
+	vmovdqu	g, _ytmp(%rsp)
+	vmovdqu	h, _ytmp+1*SZ8(%rsp)
+	TRANSPOSE8	TT0, TT1, TT2, TT3, TT4, TT5, TT6, TT7,   TMP0, TMP1
+	vmovdqu	PSHUFFLE_BYTE_FLIP_MASK(%rip), TMP1
+	vmovdqu	_ytmp(%rsp), g
+	vpshufb	TMP1, TT0, TT0
+	vpshufb	TMP1, TT1, TT1
+	vpshufb	TMP1, TT2, TT2
+	vpshufb	TMP1, TT3, TT3
+	vpshufb	TMP1, TT4, TT4
+	vpshufb	TMP1, TT5, TT5
+	vpshufb	TMP1, TT6, TT6
+	vpshufb	TMP1, TT7, TT7
+	vmovdqu	_ytmp+1*SZ8(%rsp), h
+	vmovdqu	TT4, _ytmp(%rsp)
+	vmovdqu	TT5, _ytmp+1*SZ8(%rsp)
+	vmovdqu	TT6, _ytmp+2*SZ8(%rsp)
+	vmovdqu	TT7, _ytmp+3*SZ8(%rsp)
+	ROUND_00_15	TT0,(i*8+0)
+	vmovdqu	_ytmp(%rsp), TT0
+	ROUND_00_15	TT1,(i*8+1)
+	vmovdqu	_ytmp+1*SZ8(%rsp), TT1
+	ROUND_00_15	TT2,(i*8+2)
+	vmovdqu	_ytmp+2*SZ8(%rsp), TT2
+	ROUND_00_15	TT3,(i*8+3)
+	vmovdqu	_ytmp+3*SZ8(%rsp), TT3
+	ROUND_00_15	TT0,(i*8+4)
+	ROUND_00_15	TT1,(i*8+5)
+	ROUND_00_15	TT2,(i*8+6)
+	ROUND_00_15	TT3,(i*8+7)
+	i = (i+1)
+.endr
+	add	$64, IDX
+	i = (i*8)
+
+	jmp	Lrounds_16_xx
+.align 16
+Lrounds_16_xx:
+.rep 16
+	ROUND_16_XX	T1, i
+	i = (i+1)
+.endr
+
+	cmp	$ROUNDS,ROUND
+	jb	Lrounds_16_xx
+
+	# add old digest
+	vpaddd	_digest+0*SZ8(%rsp), a, a
+	vpaddd	_digest+1*SZ8(%rsp), b, b
+	vpaddd	_digest+2*SZ8(%rsp), c, c
+	vpaddd	_digest+3*SZ8(%rsp), d, d
+	vpaddd	_digest+4*SZ8(%rsp), e, e
+	vpaddd	_digest+5*SZ8(%rsp), f, f
+	vpaddd	_digest+6*SZ8(%rsp), g, g
+	vpaddd	_digest+7*SZ8(%rsp), h, h
+
+	sub	$1, INP_SIZE  # unit is blocks
+	jne	lloop
+
+	# write back to memory (state object) the transposed digest
+	vmovdqu	a, 0*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	b, 1*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	c, 2*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	d, 3*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	e, 4*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	f, 5*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	g, 6*SHA256_DIGEST_ROW_SIZE(STATE)
+	vmovdqu	h, 7*SHA256_DIGEST_ROW_SIZE(STATE)
+
+	# update input pointers
+	add	IDX, inp0
+	mov	inp0, _args_data_ptr+0*8(STATE)
+	add	IDX, inp1
+	mov	inp1, _args_data_ptr+1*8(STATE)
+	add	IDX, inp2
+	mov	inp2, _args_data_ptr+2*8(STATE)
+	add	IDX, inp3
+	mov	inp3, _args_data_ptr+3*8(STATE)
+	add	IDX, inp4
+	mov	inp4, _args_data_ptr+4*8(STATE)
+	add	IDX, inp5
+	mov	inp5, _args_data_ptr+5*8(STATE)
+	add	IDX, inp6
+	mov	inp6, _args_data_ptr+6*8(STATE)
+	add	IDX, inp7
+	mov	inp7, _args_data_ptr+7*8(STATE)
+
+	# Postamble
+	mov	_rsp(%rsp), %rsp
+
+	# restore callee-saved clobbered registers
+	pop     %r15
+	pop     %r14
+	pop     %r13
+	pop     %r12
+
+	ret
+ENDPROC(sha256_x8_avx2)
+.data
+.align 64
+K256_8:
+	.octa	0x428a2f98428a2f98428a2f98428a2f98
+	.octa	0x428a2f98428a2f98428a2f98428a2f98
+	.octa	0x71374491713744917137449171374491
+	.octa	0x71374491713744917137449171374491
+	.octa	0xb5c0fbcfb5c0fbcfb5c0fbcfb5c0fbcf
+	.octa	0xb5c0fbcfb5c0fbcfb5c0fbcfb5c0fbcf
+	.octa	0xe9b5dba5e9b5dba5e9b5dba5e9b5dba5
+	.octa	0xe9b5dba5e9b5dba5e9b5dba5e9b5dba5
+	.octa	0x3956c25b3956c25b3956c25b3956c25b
+	.octa	0x3956c25b3956c25b3956c25b3956c25b
+	.octa	0x59f111f159f111f159f111f159f111f1
+	.octa	0x59f111f159f111f159f111f159f111f1
+	.octa	0x923f82a4923f82a4923f82a4923f82a4
+	.octa	0x923f82a4923f82a4923f82a4923f82a4
+	.octa	0xab1c5ed5ab1c5ed5ab1c5ed5ab1c5ed5
+	.octa	0xab1c5ed5ab1c5ed5ab1c5ed5ab1c5ed5
+	.octa	0xd807aa98d807aa98d807aa98d807aa98
+	.octa	0xd807aa98d807aa98d807aa98d807aa98
+	.octa	0x12835b0112835b0112835b0112835b01
+	.octa	0x12835b0112835b0112835b0112835b01
+	.octa	0x243185be243185be243185be243185be
+	.octa	0x243185be243185be243185be243185be
+	.octa	0x550c7dc3550c7dc3550c7dc3550c7dc3
+	.octa	0x550c7dc3550c7dc3550c7dc3550c7dc3
+	.octa	0x72be5d7472be5d7472be5d7472be5d74
+	.octa	0x72be5d7472be5d7472be5d7472be5d74
+	.octa	0x80deb1fe80deb1fe80deb1fe80deb1fe
+	.octa	0x80deb1fe80deb1fe80deb1fe80deb1fe
+	.octa	0x9bdc06a79bdc06a79bdc06a79bdc06a7
+	.octa	0x9bdc06a79bdc06a79bdc06a79bdc06a7
+	.octa	0xc19bf174c19bf174c19bf174c19bf174
+	.octa	0xc19bf174c19bf174c19bf174c19bf174
+	.octa	0xe49b69c1e49b69c1e49b69c1e49b69c1
+	.octa	0xe49b69c1e49b69c1e49b69c1e49b69c1
+	.octa	0xefbe4786efbe4786efbe4786efbe4786
+	.octa	0xefbe4786efbe4786efbe4786efbe4786
+	.octa	0x0fc19dc60fc19dc60fc19dc60fc19dc6
+	.octa	0x0fc19dc60fc19dc60fc19dc60fc19dc6
+	.octa	0x240ca1cc240ca1cc240ca1cc240ca1cc
+	.octa	0x240ca1cc240ca1cc240ca1cc240ca1cc
+	.octa	0x2de92c6f2de92c6f2de92c6f2de92c6f
+	.octa	0x2de92c6f2de92c6f2de92c6f2de92c6f
+	.octa	0x4a7484aa4a7484aa4a7484aa4a7484aa
+	.octa	0x4a7484aa4a7484aa4a7484aa4a7484aa
+	.octa	0x5cb0a9dc5cb0a9dc5cb0a9dc5cb0a9dc
+	.octa	0x5cb0a9dc5cb0a9dc5cb0a9dc5cb0a9dc
+	.octa	0x76f988da76f988da76f988da76f988da
+	.octa	0x76f988da76f988da76f988da76f988da
+	.octa	0x983e5152983e5152983e5152983e5152
+	.octa	0x983e5152983e5152983e5152983e5152
+	.octa	0xa831c66da831c66da831c66da831c66d
+	.octa	0xa831c66da831c66da831c66da831c66d
+	.octa	0xb00327c8b00327c8b00327c8b00327c8
+	.octa	0xb00327c8b00327c8b00327c8b00327c8
+	.octa	0xbf597fc7bf597fc7bf597fc7bf597fc7
+	.octa	0xbf597fc7bf597fc7bf597fc7bf597fc7
+	.octa	0xc6e00bf3c6e00bf3c6e00bf3c6e00bf3
+	.octa	0xc6e00bf3c6e00bf3c6e00bf3c6e00bf3
+	.octa	0xd5a79147d5a79147d5a79147d5a79147
+	.octa	0xd5a79147d5a79147d5a79147d5a79147
+	.octa	0x06ca635106ca635106ca635106ca6351
+	.octa	0x06ca635106ca635106ca635106ca6351
+	.octa	0x14292967142929671429296714292967
+	.octa	0x14292967142929671429296714292967
+	.octa	0x27b70a8527b70a8527b70a8527b70a85
+	.octa	0x27b70a8527b70a8527b70a8527b70a85
+	.octa	0x2e1b21382e1b21382e1b21382e1b2138
+	.octa	0x2e1b21382e1b21382e1b21382e1b2138
+	.octa	0x4d2c6dfc4d2c6dfc4d2c6dfc4d2c6dfc
+	.octa	0x4d2c6dfc4d2c6dfc4d2c6dfc4d2c6dfc
+	.octa	0x53380d1353380d1353380d1353380d13
+	.octa	0x53380d1353380d1353380d1353380d13
+	.octa	0x650a7354650a7354650a7354650a7354
+	.octa	0x650a7354650a7354650a7354650a7354
+	.octa	0x766a0abb766a0abb766a0abb766a0abb
+	.octa	0x766a0abb766a0abb766a0abb766a0abb
+	.octa	0x81c2c92e81c2c92e81c2c92e81c2c92e
+	.octa	0x81c2c92e81c2c92e81c2c92e81c2c92e
+	.octa	0x92722c8592722c8592722c8592722c85
+	.octa	0x92722c8592722c8592722c8592722c85
+	.octa	0xa2bfe8a1a2bfe8a1a2bfe8a1a2bfe8a1
+	.octa	0xa2bfe8a1a2bfe8a1a2bfe8a1a2bfe8a1
+	.octa	0xa81a664ba81a664ba81a664ba81a664b
+	.octa	0xa81a664ba81a664ba81a664ba81a664b
+	.octa	0xc24b8b70c24b8b70c24b8b70c24b8b70
+	.octa	0xc24b8b70c24b8b70c24b8b70c24b8b70
+	.octa	0xc76c51a3c76c51a3c76c51a3c76c51a3
+	.octa	0xc76c51a3c76c51a3c76c51a3c76c51a3
+	.octa	0xd192e819d192e819d192e819d192e819
+	.octa	0xd192e819d192e819d192e819d192e819
+	.octa	0xd6990624d6990624d6990624d6990624
+	.octa	0xd6990624d6990624d6990624d6990624
+	.octa	0xf40e3585f40e3585f40e3585f40e3585
+	.octa	0xf40e3585f40e3585f40e3585f40e3585
+	.octa	0x106aa070106aa070106aa070106aa070
+	.octa	0x106aa070106aa070106aa070106aa070
+	.octa	0x19a4c11619a4c11619a4c11619a4c116
+	.octa	0x19a4c11619a4c11619a4c11619a4c116
+	.octa	0x1e376c081e376c081e376c081e376c08
+	.octa	0x1e376c081e376c081e376c081e376c08
+	.octa	0x2748774c2748774c2748774c2748774c
+	.octa	0x2748774c2748774c2748774c2748774c
+	.octa	0x34b0bcb534b0bcb534b0bcb534b0bcb5
+	.octa	0x34b0bcb534b0bcb534b0bcb534b0bcb5
+	.octa	0x391c0cb3391c0cb3391c0cb3391c0cb3
+	.octa	0x391c0cb3391c0cb3391c0cb3391c0cb3
+	.octa	0x4ed8aa4a4ed8aa4a4ed8aa4a4ed8aa4a
+	.octa	0x4ed8aa4a4ed8aa4a4ed8aa4a4ed8aa4a
+	.octa	0x5b9cca4f5b9cca4f5b9cca4f5b9cca4f
+	.octa	0x5b9cca4f5b9cca4f5b9cca4f5b9cca4f
+	.octa	0x682e6ff3682e6ff3682e6ff3682e6ff3
+	.octa	0x682e6ff3682e6ff3682e6ff3682e6ff3
+	.octa	0x748f82ee748f82ee748f82ee748f82ee
+	.octa	0x748f82ee748f82ee748f82ee748f82ee
+	.octa	0x78a5636f78a5636f78a5636f78a5636f
+	.octa	0x78a5636f78a5636f78a5636f78a5636f
+	.octa	0x84c8781484c8781484c8781484c87814
+	.octa	0x84c8781484c8781484c8781484c87814
+	.octa	0x8cc702088cc702088cc702088cc70208
+	.octa	0x8cc702088cc702088cc702088cc70208
+	.octa	0x90befffa90befffa90befffa90befffa
+	.octa	0x90befffa90befffa90befffa90befffa
+	.octa	0xa4506ceba4506ceba4506ceba4506ceb
+	.octa	0xa4506ceba4506ceba4506ceba4506ceb
+	.octa	0xbef9a3f7bef9a3f7bef9a3f7bef9a3f7
+	.octa	0xbef9a3f7bef9a3f7bef9a3f7bef9a3f7
+	.octa	0xc67178f2c67178f2c67178f2c67178f2
+	.octa	0xc67178f2c67178f2c67178f2c67178f2
+PSHUFFLE_BYTE_FLIP_MASK:
+.octa 0x0c0d0e0f08090a0b0405060700010203
+.octa 0x0c0d0e0f08090a0b0405060700010203
+
+.align 64
+.global K256
+K256:
+	.int	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+	.int	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+	.int	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+	.int	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+	.int	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+	.int	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+	.int	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+	.int	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+	.int	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+	.int	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+	.int	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+	.int	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+	.int	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+	.int	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+	.int	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+	.int	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2