[PATCH] SHA-256: Add SSSE3 implementation for x86-64

Jussi Kivilinna jussi.kivilinna at iki.fi
Wed Dec 11 20:49:14 CET 2013


* cipher/Makefile.am: Add 'sha256-ssse3-amd64.S'.
* cipher/sha256-ssse3-amd64.S: New.
* cipher/sha256.c (USE_SSSE3): New.
(SHA256_CONTEXT) [USE_SSSE3]: Add 'use_ssse3'.
(sha256_init, sha224_init) [USE_SSSE3]: Initialize 'use_ssse3'.
(transform): Rename to...
(_transform): This.
[USE_SSSE3] (_gcry_sha256_transform_amd64_ssse3): New.
(transform): New.
* configure.ac (HAVE_INTEL_SYNTAX_PLATFORM_AS): New check.
(sha256): Add 'sha256-ssse3-amd64.lo'.
* src/g10lib.h (HWF_INTEL_SSSE3): New.
* src/global.c (hwflist): Add "intel-ssse3".
* src/hwf-x86.c (detect_x86_gnuc): Test for SSSE3.
--

Patch adds fast SSSE3 implementation of SHA-256 by Intel Corporation. The
assembly source is licensed under 3-clause BSD license, thus compatible
with LGPL2.1+. Original source can be accessed at:
 http://www.intel.com/p/en_US/embedded/hwsw/technology/packet-processing#docs

Implementation is described in white paper
 "Fast SHA - 256 Implementations on Intel® Architecture Processors"
 http://www.intel.com/content/www/us/en/intelligent-systems/intel-technology/sha-256-implementations-paper.html

Benchmarks:

cpu                 Old         New         Diff
Intel i5-4570       13.99 c/B   10.66 c/B   1.31x
Intel i5-2450M      21.53 c/B   15.79 c/B   1.36x
Intel Core2 T8100   20.84 c/B   15.07 c/B   1.38x

Signed-off-by: Jussi Kivilinna <jussi.kivilinna at iki.fi>
---
 cipher/Makefile.am          |    2 
 cipher/sha256-ssse3-amd64.S |  526 +++++++++++++++++++++++++++++++++++++++++++
 cipher/sha256.c             |   45 ++++
 configure.ac                |   45 ++++
 src/g10lib.h                |    1 
 src/global.c                |    1 
 src/hwf-x86.c               |    3 
 7 files changed, 621 insertions(+), 2 deletions(-)
 create mode 100644 cipher/sha256-ssse3-amd64.S

diff --git a/cipher/Makefile.am b/cipher/Makefile.am
index ff9deca..34f74e2 100644
--- a/cipher/Makefile.am
+++ b/cipher/Makefile.am
@@ -78,7 +78,7 @@ scrypt.c \
 seed.c \
 serpent.c serpent-sse2-amd64.S serpent-avx2-amd64.S \
 sha1.c \
-sha256.c \
+sha256.c sha256-ssse3-amd64.S \
 sha512.c sha512-armv7-neon.S \
 stribog.c \
 tiger.c \
diff --git a/cipher/sha256-ssse3-amd64.S b/cipher/sha256-ssse3-amd64.S
new file mode 100644
index 0000000..d7d5980
--- /dev/null
+++ b/cipher/sha256-ssse3-amd64.S
@@ -0,0 +1,526 @@
+/*
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; Copyright (c) 2012, Intel Corporation
+;
+; All rights reserved.
+;
+; 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 the 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 INTEL CORPORATION "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 INTEL CORPORATION 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.
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;
+; 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.
+; The paper is expected to be released roughly at the end of April, 2012
+;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; This code schedules 1 blocks at a time, with 4 lanes per block
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+*/
+/*
+ * Conversion to GAS assembly and integration to libgcrypt
+ *  by Jussi Kivilinna <jussi.kivilinna at iki.fi>
+ *
+ * Note: original implementation was named as SHA256-SSE4. However, only SSSE3
+ *       is required.
+ */
+
+#ifdef __x86_64
+#include <config.h>
+#if defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
+    defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
+    defined(HAVE_GCC_INLINE_ASM_SSSE3) && defined(USE_SHA256)
+
+#ifdef __PIC__
+#  define ADD_RIP +rip
+#else
+#  define ADD_RIP
+#endif
+
+.intel_syntax noprefix
+
+#define	MOVDQ movdqu /* assume buffers not aligned */
+
+/*;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; Define Macros*/
+
+/* addm [mem], reg
+ * Add reg to mem using reg-mem add and store */
+.macro addm p1 p2
+	add	\p2, \p1
+	mov	\p1, \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
+	MOVDQ \p1, \p2
+	pshufb \p1, \p3
+.endm
+
+/*;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;*/
+
+X0 = xmm4
+X1 = xmm5
+X2 = xmm6
+X3 = xmm7
+
+XTMP0 = xmm0
+XTMP1 = xmm1
+XTMP2 = xmm2
+XTMP3 = xmm3
+XTMP4 = xmm8
+XFER  = xmm9
+
+SHUF_00BA = xmm10 /* shuffle xBxA -> 00BA */
+SHUF_DC00 = xmm11 /* shuffle xDxC -> DC00 */
+BYTE_FLIP_MASK = xmm12
+
+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
+
+
+
+#define _INP_END_SIZE	8
+#define _INP_SIZE	8
+#define _XFER_SIZE	8
+#define _XMM_SAVE_SIZE	0
+/* STACK_SIZE plus pushes must be an odd multiple of 8 */
+#define _ALIGN_SIZE	8
+
+#define _INP_END	0
+#define _INP		(_INP_END  + _INP_END_SIZE)
+#define _XFER		(_INP      + _INP_SIZE)
+#define _XMM_SAVE	(_XFER     + _XFER_SIZE + _ALIGN_SIZE)
+#define 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 */
+		movdqa	XTMP0, X3
+	mov	y0, e		/* y0 = e */
+	ror	y0, (25-11)	/* y0 = e >> (25-11) */
+	mov	y1, a		/* y1 = a */
+		palignr	XTMP0, X2, 4	/* XTMP0 = W[-7] */
+	ror	y1, (22-13)	/* y1 = a >> (22-13) */
+	xor	y0, e		/* y0 = e ^ (e >> (25-11)) */
+	mov	y2, f		/* y2 = f */
+	ror	y0, (11-6)	/* y0 = (e >> (11-6)) ^ (e >> (25-6)) */
+		movdqa	XTMP1, X1
+	xor	y1, a		/* y1 = a ^ (a >> (22-13) */
+	xor	y2, g		/* y2 = f^g */
+		paddd	XTMP0, X0	/* XTMP0 = W[-7] + W[-16] */
+	xor	y0, e		/* y0 = e ^ (e >> (11-6)) ^ (e >> (25-6)) */
+	and	y2, e		/* y2 = (f^g)&e */
+	ror	y1, (13-2)	/* y1 = (a >> (13-2)) ^ (a >> (22-2)) */
+		/* compute s0 */
+		palignr	XTMP1, X0, 4	/* XTMP1 = W[-15] */
+	xor	y1, a		/* y1 = a ^ (a >> (13-2)) ^ (a >> (22-2)) */
+	ror	y0, 6		/* y0 = S1 = (e>>6) & (e>>11) ^ (e>>25) */
+	xor	y2, g		/* y2 = CH = ((f^g)&e)^g */
+		movdqa	XTMP2, XTMP1	/* XTMP2 = W[-15] */
+	ror	y1, 2		/* y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22) */
+	add	y2, y0		/* y2 = S1 + CH */
+	add	y2, [rsp + _XFER + 0*4]	/* y2 = k + w + S1 + CH */
+		movdqa	XTMP3, XTMP1	/* XTMP3 = W[-15] */
+	mov	y0, a		/* y0 = a */
+	add	h, y2		/* h = h + S1 + CH + k + w */
+	mov	y2, a		/* y2 = a */
+		pslld	XTMP1, (32-7)
+	or	y0, c		/* y0 = a|c */
+	add	d, h		/* d = d + h + S1 + CH + k + w */
+	and	y2, c		/* y2 = a&c */
+		psrld	XTMP2, 7
+	and	y0, b		/* y0 = (a|c)&b */
+	add	h, y1		/* h = h + S1 + CH + k + w + S0 */
+		por	XTMP1, XTMP2	/* XTMP1 = W[-15] ror 7 */
+	or	y0, y2		/* y0 = MAJ = (a|c)&b)|(a&c) */
+	add	h, y0		/* h = h + S1 + CH + k + w + S0 + MAJ */
+
+ROTATE_ARGS
+		movdqa	XTMP2, XTMP3	/* XTMP2 = W[-15] */
+	mov	y0, e		/* y0 = e */
+	mov	y1, a		/* y1 = a */
+		movdqa	XTMP4, XTMP3	/* XTMP4 = W[-15] */
+	ror	y0, (25-11)	/* y0 = e >> (25-11) */
+	xor	y0, e		/* y0 = e ^ (e >> (25-11)) */
+	mov	y2, f		/* y2 = f */
+	ror	y1, (22-13)	/* y1 = a >> (22-13) */
+		pslld	XTMP3, (32-18)
+	xor	y1, a		/* y1 = a ^ (a >> (22-13) */
+	ror	y0, (11-6)	/* y0 = (e >> (11-6)) ^ (e >> (25-6)) */
+	xor	y2, g		/* y2 = f^g */
+		psrld	XTMP2, 18
+	ror	y1, (13-2)	/* y1 = (a >> (13-2)) ^ (a >> (22-2)) */
+	xor	y0, e		/* y0 = e ^ (e >> (11-6)) ^ (e >> (25-6)) */
+	and	y2, e		/* y2 = (f^g)&e */
+	ror	y0, 6		/* y0 = S1 = (e>>6) & (e>>11) ^ (e>>25) */
+		pxor	XTMP1, XTMP3
+	xor	y1, a		/* y1 = a ^ (a >> (13-2)) ^ (a >> (22-2)) */
+	xor	y2, g		/* y2 = CH = ((f^g)&e)^g */
+		psrld	XTMP4, 3	/* XTMP4 = W[-15] >> 3 */
+	add	y2, y0		/* y2 = S1 + CH */
+	add	y2, [rsp + _XFER + 1*4]	/* y2 = k + w + S1 + CH */
+	ror	y1, 2		/* y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22) */
+		pxor	XTMP1, XTMP2	/* XTMP1 = W[-15] ror 7 ^ W[-15] ror 18 */
+	mov	y0, a		/* y0 = a */
+	add	h, y2		/* h = h + S1 + CH + k + w */
+	mov	y2, a		/* y2 = a */
+		pxor	XTMP1, XTMP4	/* XTMP1 = s0 */
+	or	y0, c		/* y0 = a|c */
+	add	d, h		/* d = d + h + S1 + CH + k + w */
+	and	y2, c		/* y2 = a&c */
+		/* compute low s1 */
+		pshufd	XTMP2, X3, 0b11111010	/* XTMP2 = W[-2] {BBAA} */
+	and	y0, b		/* y0 = (a|c)&b */
+	add	h, y1		/* h = h + S1 + CH + k + w + S0 */
+		paddd	XTMP0, XTMP1	/* XTMP0 = W[-16] + W[-7] + s0 */
+	or	y0, y2		/* y0 = MAJ = (a|c)&b)|(a&c) */
+	add	h, y0		/* h = h + S1 + CH + k + w + S0 + MAJ */
+
+ROTATE_ARGS
+		movdqa	XTMP3, XTMP2	/* XTMP3 = W[-2] {BBAA} */
+	mov	y0, e		/* y0 = e */
+	mov	y1, a		/* y1 = a */
+	ror	y0, (25-11)	/* y0 = e >> (25-11) */
+		movdqa	XTMP4, XTMP2	/* XTMP4 = W[-2] {BBAA} */
+	xor	y0, e		/* y0 = e ^ (e >> (25-11)) */
+	ror	y1, (22-13)	/* y1 = a >> (22-13) */
+	mov	y2, f		/* y2 = f */
+	xor	y1, a		/* y1 = a ^ (a >> (22-13) */
+	ror	y0, (11-6)	/* y0 = (e >> (11-6)) ^ (e >> (25-6)) */
+		psrlq	XTMP2, 17	/* XTMP2 = W[-2] ror 17 {xBxA} */
+	xor	y2, g		/* y2 = f^g */
+		psrlq	XTMP3, 19	/* XTMP3 = W[-2] ror 19 {xBxA} */
+	xor	y0, e		/* y0 = e ^ (e >> (11-6)) ^ (e >> (25-6)) */
+	and	y2, e		/* y2 = (f^g)&e */
+		psrld	XTMP4, 10	/* XTMP4 = W[-2] >> 10 {BBAA} */
+	ror	y1, (13-2)	/* y1 = (a >> (13-2)) ^ (a >> (22-2)) */
+	xor	y1, a		/* y1 = a ^ (a >> (13-2)) ^ (a >> (22-2)) */
+	xor	y2, g		/* y2 = CH = ((f^g)&e)^g */
+	ror	y0, 6		/* y0 = S1 = (e>>6) & (e>>11) ^ (e>>25) */
+		pxor	XTMP2, XTMP3
+	add	y2, y0		/* y2 = S1 + CH */
+	ror	y1, 2		/* y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22) */
+	add	y2, [rsp + _XFER + 2*4]	/* y2 = k + w + S1 + CH */
+		pxor	XTMP4, XTMP2	/* XTMP4 = s1 {xBxA} */
+	mov	y0, a		/* y0 = a */
+	add	h, y2		/* h = h + S1 + CH + k + w */
+	mov	y2, a		/* y2 = a */
+		pshufb	XTMP4, SHUF_00BA	/* XTMP4 = s1 {00BA} */
+	or	y0, c		/* y0 = a|c */
+	add	d, h		/* d = d + h + S1 + CH + k + w */
+	and	y2, c		/* y2 = a&c */
+		paddd	XTMP0, XTMP4	/* XTMP0 = {..., ..., W[1], W[0]} */
+	and	y0, b		/* y0 = (a|c)&b */
+	add	h, y1		/* h = h + S1 + CH + k + w + S0 */
+		/* compute high s1 */
+		pshufd	XTMP2, XTMP0, 0b01010000 /* XTMP2 = W[-2] {DDCC} */
+	or	y0, y2		/* y0 = MAJ = (a|c)&b)|(a&c) */
+	add	h, y0		/* h = h + S1 + CH + k + w + S0 + MAJ */
+
+ROTATE_ARGS
+		movdqa	XTMP3, XTMP2	/* XTMP3 = W[-2] {DDCC} */
+	mov	y0, e		/* y0 = e */
+	ror	y0, (25-11)	/* y0 = e >> (25-11) */
+	mov	y1, a		/* y1 = a */
+		movdqa	X0,    XTMP2	/* X0    = W[-2] {DDCC} */
+	ror	y1, (22-13)	/* y1 = a >> (22-13) */
+	xor	y0, e		/* y0 = e ^ (e >> (25-11)) */
+	mov	y2, f		/* y2 = f */
+	ror	y0, (11-6)	/* y0 = (e >> (11-6)) ^ (e >> (25-6)) */
+		psrlq	XTMP2, 17	/* XTMP2 = W[-2] ror 17 {xDxC} */
+	xor	y1, a		/* y1 = a ^ (a >> (22-13) */
+	xor	y2, g		/* y2 = f^g */
+		psrlq	XTMP3, 19	/* XTMP3 = W[-2] ror 19 {xDxC} */
+	xor	y0, e		/* y0 = e ^ (e >> (11-6)) ^ (e >> (25-6)) */
+	and	y2, e		/* y2 = (f^g)&e */
+	ror	y1, (13-2)	/* y1 = (a >> (13-2)) ^ (a >> (22-2)) */
+		psrld	X0,    10	/* X0 = W[-2] >> 10 {DDCC} */
+	xor	y1, a		/* y1 = a ^ (a >> (13-2)) ^ (a >> (22-2)) */
+	ror	y0, 6		/* y0 = S1 = (e>>6) & (e>>11) ^ (e>>25) */
+	xor	y2, g		/* y2 = CH = ((f^g)&e)^g */
+		pxor	XTMP2, XTMP3
+	ror	y1, 2		/* y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22) */
+	add	y2, y0		/* y2 = S1 + CH */
+	add	y2, [rsp + _XFER + 3*4]	/* y2 = k + w + S1 + CH */
+		pxor	X0, XTMP2	/* X0 = s1 {xDxC} */
+	mov	y0, a		/* y0 = a */
+	add	h, y2		/* h = h + S1 + CH + k + w */
+	mov	y2, a		/* y2 = a */
+		pshufb	X0, SHUF_DC00	/* X0 = s1 {DC00} */
+	or	y0, c		/* y0 = a|c */
+	add	d, h		/* d = d + h + S1 + CH + k + w */
+	and	y2, c		/* y2 = a&c */
+		paddd	X0, XTMP0	/* X0 = {W[3], W[2], W[1], W[0]} */
+	and	y0, b		/* y0 = (a|c)&b */
+	add	h, y1		/* h = h + S1 + CH + k + w + S0 */
+	or	y0, y2		/* y0 = MAJ = (a|c)&b)|(a&c) */
+	add	h, y0		/* h = h + S1 + CH + k + w + S0 + MAJ */
+
+ROTATE_ARGS
+rotate_Xs
+.endm
+
+/* input is [rsp + _XFER + %1 * 4] */
+.macro DO_ROUND i1
+	mov	y0, e		/* y0 = e */
+	ror	y0, (25-11)	/* y0 = e >> (25-11) */
+	mov	y1, a		/* y1 = a */
+	xor	y0, e		/* y0 = e ^ (e >> (25-11)) */
+	ror	y1, (22-13)	/* y1 = a >> (22-13) */
+	mov	y2, f		/* y2 = f */
+	xor	y1, a		/* y1 = a ^ (a >> (22-13) */
+	ror	y0, (11-6)	/* y0 = (e >> (11-6)) ^ (e >> (25-6)) */
+	xor	y2, g		/* y2 = f^g */
+	xor	y0, e		/* y0 = e ^ (e >> (11-6)) ^ (e >> (25-6)) */
+	ror	y1, (13-2)	/* y1 = (a >> (13-2)) ^ (a >> (22-2)) */
+	and	y2, e		/* y2 = (f^g)&e */
+	xor	y1, a		/* y1 = a ^ (a >> (13-2)) ^ (a >> (22-2)) */
+	ror	y0, 6		/* y0 = S1 = (e>>6) & (e>>11) ^ (e>>25) */
+	xor	y2, g		/* y2 = CH = ((f^g)&e)^g */
+	add	y2, y0		/* y2 = S1 + CH */
+	ror	y1, 2		/* y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22) */
+	add	y2, [rsp + _XFER + \i1 * 4]	/* y2 = k + w + S1 + CH */
+	mov	y0, a		/* y0 = a */
+	add	h, y2		/* h = h + S1 + CH + k + w */
+	mov	y2, a		/* y2 = a */
+	or	y0, c		/* y0 = a|c */
+	add	d, h		/* d = d + h + S1 + CH + k + w */
+	and	y2, c		/* y2 = a&c */
+	and	y0, b		/* y0 = (a|c)&b */
+	add	h, y1		/* h = h + S1 + CH + k + w + S0 */
+	or	y0, y2		/* y0 = MAJ = (a|c)&b)|(a&c) */
+	add	h, y0		/* h = h + S1 + CH + k + w + S0 + MAJ */
+	ROTATE_ARGS
+.endm
+
+/*
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; void sha256_sse4(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
+.globl _gcry_sha256_transform_amd64_ssse3
+.type  _gcry_sha256_transform_amd64_ssse3, at function;
+.align 32
+_gcry_sha256_transform_amd64_ssse3:
+	push	rbx
+	push	rbp
+	push	r13
+	push	r14
+	push	r15
+
+	sub	rsp, STACK_SIZE
+
+	shl	NUM_BLKS, 6	/* convert to bytes */
+	jz	.Ldone_hash
+	add	NUM_BLKS, INP	/* pointer to end of data */
+	mov	[rsp + _INP_END], NUM_BLKS
+
+	/* load initial digest */
+	mov	a,[4*0 + CTX]
+	mov	b,[4*1 + CTX]
+	mov	c,[4*2 + CTX]
+	mov	d,[4*3 + CTX]
+	mov	e,[4*4 + CTX]
+	mov	f,[4*5 + CTX]
+	mov	g,[4*6 + CTX]
+	mov	h,[4*7 + CTX]
+
+	movdqa	BYTE_FLIP_MASK, [.LPSHUFFLE_BYTE_FLIP_MASK ADD_RIP]
+	movdqa	SHUF_00BA, [.L_SHUF_00BA ADD_RIP]
+	movdqa	SHUF_DC00, [.L_SHUF_DC00 ADD_RIP]
+
+.Loop0:
+	lea	TBL, [.LK256 ADD_RIP]
+
+	/* byte swap first 16 dwords */
+	COPY_XMM_AND_BSWAP	X0, [INP + 0*16], BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP	X1, [INP + 1*16], BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP	X2, [INP + 2*16], BYTE_FLIP_MASK
+	COPY_XMM_AND_BSWAP	X3, [INP + 3*16], BYTE_FLIP_MASK
+
+	mov	[rsp + _INP], INP
+
+	/* schedule 48 input dwords, by doing 3 rounds of 16 each */
+	mov	SRND, 3
+.align 16
+.Loop1:
+	movdqa	XFER, [TBL + 0*16]
+	paddd	XFER, X0
+	movdqa	[rsp + _XFER], XFER
+	FOUR_ROUNDS_AND_SCHED
+
+	movdqa	XFER, [TBL + 1*16]
+	paddd	XFER, X0
+	movdqa	[rsp + _XFER], XFER
+	FOUR_ROUNDS_AND_SCHED
+
+	movdqa	XFER, [TBL + 2*16]
+	paddd	XFER, X0
+	movdqa	[rsp + _XFER], XFER
+	FOUR_ROUNDS_AND_SCHED
+
+	movdqa	XFER, [TBL + 3*16]
+	paddd	XFER, X0
+	movdqa	[rsp + _XFER], XFER
+	add	TBL, 4*16
+	FOUR_ROUNDS_AND_SCHED
+
+	sub	SRND, 1
+	jne	.Loop1
+
+	mov	SRND, 2
+.Loop2:
+	paddd	X0, [TBL + 0*16]
+	movdqa	[rsp + _XFER], X0
+	DO_ROUND	0
+	DO_ROUND	1
+	DO_ROUND	2
+	DO_ROUND	3
+	paddd	X1, [TBL + 1*16]
+	movdqa	[rsp + _XFER], X1
+	add	TBL, 2*16
+	DO_ROUND	0
+	DO_ROUND	1
+	DO_ROUND	2
+	DO_ROUND	3
+
+	movdqa	X0, X2
+	movdqa	X1, X3
+
+	sub	SRND, 1
+	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	INP, 64
+	cmp	INP, [rsp + _INP_END]
+	jne	.Loop0
+
+.Ldone_hash:
+	add	rsp, STACK_SIZE
+
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	rbp
+	pop	rbx
+
+	mov     rax, STACK_SIZE
+	ret
+
+
+.data
+.align 64
+.LK256:
+	.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
+
+.LPSHUFFLE_BYTE_FLIP_MASK: .octa 0x0c0d0e0f08090a0b0405060700010203
+
+/* shuffle xBxA -> 00BA */
+.L_SHUF_00BA:              .octa 0xFFFFFFFFFFFFFFFF0b0a090803020100
+
+/* shuffle xDxC -> DC00 */
+.L_SHUF_DC00:              .octa 0x0b0a090803020100FFFFFFFFFFFFFFFF
+
+#endif
+#endif
diff --git a/cipher/sha256.c b/cipher/sha256.c
index bd5a412..f3c1d62 100644
--- a/cipher/sha256.c
+++ b/cipher/sha256.c
@@ -46,11 +46,25 @@
 #include "cipher.h"
 #include "hash-common.h"
 
+
+/* USE_SSSE3 indicates whether to compile with Intel SSSE3 code. */
+#undef USE_SSSE3
+#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
+    defined(HAVE_GCC_INLINE_ASM_SSSE3) && \
+    defined(HAVE_INTEL_SYNTAX_PLATFORM_AS)
+# define USE_SSSE3 1
+#endif
+
+
 typedef struct {
   gcry_md_block_ctx_t bctx;
   u32  h0,h1,h2,h3,h4,h5,h6,h7;
+#ifdef USE_SSSE3
+  unsigned int use_ssse3:1;
+#endif
 } SHA256_CONTEXT;
 
+
 static unsigned int
 transform (void *c, const unsigned char *data);
 
@@ -74,6 +88,10 @@ sha256_init (void *context)
   hd->bctx.count = 0;
   hd->bctx.blocksize = 64;
   hd->bctx.bwrite = transform;
+
+#ifdef USE_SSSE3
+  hd->use_ssse3 = (_gcry_get_hw_features () & HWF_INTEL_SSSE3) != 0;
+#endif
 }
 
 
@@ -96,6 +114,10 @@ sha224_init (void *context)
   hd->bctx.count = 0;
   hd->bctx.blocksize = 64;
   hd->bctx.bwrite = transform;
+
+#ifdef USE_SSSE3
+  hd->use_ssse3 = (_gcry_get_hw_features () & HWF_INTEL_SSSE3) != 0;
+#endif
 }
 
 
@@ -148,7 +170,7 @@ Sum1 (u32 x)
 
 
 static unsigned int
-transform (void *ctx, const unsigned char *data)
+_transform (void *ctx, const unsigned char *data)
 {
   SHA256_CONTEXT *hd = ctx;
   static const u32 K[64] = {
@@ -254,6 +276,27 @@ transform (void *ctx, const unsigned char *data)
 #undef R
 
 
+#ifdef USE_SSSE3
+unsigned int _gcry_sha256_transform_amd64_ssse3(const void *input_data,
+					        u32 state[8], size_t num_blks);
+#endif
+
+
+static unsigned int
+transform (void *ctx, const unsigned char *data)
+{
+  SHA256_CONTEXT *hd = ctx;
+
+#ifdef USE_SSSE3
+  if (hd->use_ssse3)
+    return _gcry_sha256_transform_amd64_ssse3 (data, &hd->h0, 1)
+           + 4 * sizeof(void*);
+#endif
+
+  return _transform (hd, data);
+}
+
+
 /*
    The routine finally terminates the computation and returns the
    digest.  The handle is prepared for a new cycle, but adding bytes
diff --git a/configure.ac b/configure.ac
index 6d40343..eb0dd29 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1077,6 +1077,44 @@ fi
 
 
 #
+# Check whether GCC assembler supports features needed for our amd64 assembly
+# implementations that use Intel syntax
+#
+if test $amd64_as_feature_detection = yes; then
+  if test $gcry_cv_gcc_amd64_platform_as_ok = yes; then
+    AC_CACHE_CHECK([whether GCC assembler is compatible for Intel syntax assembly implementations],
+       [gcry_cv_gcc_amd64_platform_as_ok_for_intel_syntax],
+       [gcry_cv_gcc_amd64_platform_as_ok_for_intel_syntax=no
+        AC_COMPILE_IFELSE([AC_LANG_SOURCE(
+          [[__asm__(
+                ".intel_syntax noprefix\n\t"
+                "pxor xmm1, xmm7;\n\t"
+                /* Intel syntax implementation also use GAS macros, so check
+                 * for them here. */
+                "VAL_A = xmm8\n\t"
+                "VAL_B = xmm9\n\t"
+                ".macro SET_VAL_A p1\n\t"
+                "  VAL_A = \\\\p1 \n\t"
+                ".endm\n\t"
+                ".macro SET_VAL_B p1\n\t"
+                "  VAL_B = \\\\p1 \n\t"
+                ".endm\n\t"
+                "vmovdqa VAL_A, VAL_B;\n\t"
+                "SET_VAL_A eax\n\t"
+                "SET_VAL_B r8d\n\t"
+                "add VAL_A, VAL_B;\n\t"
+                "add VAL_B, 0b10101;\n\t"
+            );]])],
+          [gcry_cv_gcc_amd64_platform_as_ok_for_intel_syntax=yes])])
+    if test "$gcry_cv_gcc_amd64_platform_as_ok_for_intel_syntax" = "yes" ; then
+      AC_DEFINE(HAVE_INTEL_SYNTAX_PLATFORM_AS,1,
+                [Defined if underlying assembler is compatible with Intel syntax assembly implementations])
+    fi
+  fi
+fi
+
+
+#
 # Check whether compiler is configured for ARMv6 or newer architecture
 #
 AC_CACHE_CHECK([whether compiler is configured for ARMv6 or newer architecture],
@@ -1676,6 +1714,13 @@ LIST_MEMBER(sha256, $enabled_digests)
 if test "$found" = "1" ; then
    GCRYPT_DIGESTS="$GCRYPT_DIGESTS sha256.lo"
    AC_DEFINE(USE_SHA256, 1, [Defined if this module should be included])
+
+   case "${host}" in
+      x86_64-*-*)
+         # Build with the assembly implementation
+         GCRYPT_DIGESTS="$GCRYPT_DIGESTS sha256-ssse3-amd64.lo"
+      ;;
+   esac
 fi
 
 LIST_MEMBER(sha512, $enabled_digests)
diff --git a/src/g10lib.h b/src/g10lib.h
index a326ad5..1dcadfa 100644
--- a/src/g10lib.h
+++ b/src/g10lib.h
@@ -167,6 +167,7 @@ int _gcry_log_verbosity( int level );
 #define HWF_PADLOCK_SHA  4
 #define HWF_PADLOCK_MMUL 8
 
+#define HWF_INTEL_SSSE3  64
 #define HWF_INTEL_PCLMUL 128
 #define HWF_INTEL_AESNI  256
 #define HWF_INTEL_RDRAND 512
diff --git a/src/global.c b/src/global.c
index 8a5d310..de99286 100644
--- a/src/global.c
+++ b/src/global.c
@@ -66,6 +66,7 @@ static struct
     { HWF_PADLOCK_AES, "padlock-aes" },
     { HWF_PADLOCK_SHA, "padlock-sha" },
     { HWF_PADLOCK_MMUL,"padlock-mmul"},
+    { HWF_INTEL_SSSE3, "intel-ssse3" },
     { HWF_INTEL_PCLMUL,"intel-pclmul" },
     { HWF_INTEL_AESNI, "intel-aesni" },
     { HWF_INTEL_RDRAND,"intel-rdrand" },
diff --git a/src/hwf-x86.c b/src/hwf-x86.c
index 784fe2a..ab6dacd 100644
--- a/src/hwf-x86.c
+++ b/src/hwf-x86.c
@@ -206,6 +206,9 @@ detect_x86_gnuc (void)
   if (features & 0x00000002)
      result |= HWF_INTEL_PCLMUL;
 #endif
+  /* Test bit 9 for SSSE3.  */
+  if (features & 0x00000200)
+     result |= HWF_INTEL_SSSE3;
 #ifdef ENABLE_AESNI_SUPPORT
   /* Test bit 25 for AES-NI.  */
   if (features & 0x02000000)




More information about the Gcrypt-devel mailing list