* [PATCH v2 00/10] Optimize SHA256 and SHA512 for Intel x86_64 with SSSE3, AVX or AVX2 instructions
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:58 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Herbert,
The following patch series provides optimized SHA256 and SHA512 routines
using the SSSE3, AVX or AVX2 instructions on x86_64 for Intel cpus.
Depending on cpu capabilities, speedup between 40% to 70% or more can be achieved
over the generic SHA256 and SHA512 routines.
Tim
Version 2:
1. Check AVX2 feature directly in glue code
2. Add CONFGI_AS_AVX2 check for AVX2 code
3. Use ENTRY/ENDPROC macros for assembly routines
4. Fix SSSE3 feature check in glue code
Thanks to Peter Anvin, Jussi Kivilinna and Jim Kukunas for their reviews and comments.
Tim Chen (10):
Expose SHA256 generic routine to be callable externally.
Optimized sha256 x86_64 assembly routine using Supplemental SSE3
instructions.
Optimized sha256 x86_64 assembly routine with AVX instructions.
Optimized sha256 x86_64 routine using AVX2's RORX instructions
Create module providing optimized SHA256 routines using SSSE3, AVX or
AVX2 instructions.
Expose generic sha512 routine to be callable from other modules
Optimized SHA512 x86_64 assembly routine using Supplemental SSE3
instructions.
Optimized SHA512 x86_64 assembly routine using AVX instructions.
Optimized SHA512 x86_64 assembly routine using AVX2 RORX instruction.
Create module providing optimized SHA512 routines using SSSE3, AVX or
AVX2 instructions.
arch/x86/crypto/Makefile | 4 +
arch/x86/crypto/sha256-avx-asm.S | 496 +++++++++++++++++++++++
arch/x86/crypto/sha256-avx2-asm.S | 772 ++++++++++++++++++++++++++++++++++++
arch/x86/crypto/sha256-ssse3-asm.S | 506 +++++++++++++++++++++++
arch/x86/crypto/sha256_ssse3_glue.c | 275 +++++++++++++
arch/x86/crypto/sha512-avx-asm.S | 423 ++++++++++++++++++++
arch/x86/crypto/sha512-avx2-asm.S | 743 ++++++++++++++++++++++++++++++++++
arch/x86/crypto/sha512-ssse3-asm.S | 421 ++++++++++++++++++++
arch/x86/crypto/sha512_ssse3_glue.c | 282 +++++++++++++
crypto/Kconfig | 22 +
crypto/sha256_generic.c | 11 +-
crypto/sha512_generic.c | 13 +-
include/crypto/sha.h | 5 +
13 files changed, 3962 insertions(+), 11 deletions(-)
create mode 100644 arch/x86/crypto/sha256-avx-asm.S
create mode 100644 arch/x86/crypto/sha256-avx2-asm.S
create mode 100644 arch/x86/crypto/sha256-ssse3-asm.S
create mode 100644 arch/x86/crypto/sha256_ssse3_glue.c
create mode 100644 arch/x86/crypto/sha512-avx-asm.S
create mode 100644 arch/x86/crypto/sha512-avx2-asm.S
create mode 100644 arch/x86/crypto/sha512-ssse3-asm.S
create mode 100644 arch/x86/crypto/sha512_ssse3_glue.c
--
1.7.11.7
^ permalink raw reply [flat|nested] 32+ messages in thread
* [PATCH v2 00/10] Optimize SHA256 and SHA512 for Intel x86_64 with SSSE3, AVX or AVX2 instructions
@ 2013-03-26 20:58 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Herbert,
The following patch series provides optimized SHA256 and SHA512 routines
using the SSSE3, AVX or AVX2 instructions on x86_64 for Intel cpus.
Depending on cpu capabilities, speedup between 40% to 70% or more can be achieved
over the generic SHA256 and SHA512 routines.
Tim
Version 2:
1. Check AVX2 feature directly in glue code
2. Add CONFGI_AS_AVX2 check for AVX2 code
3. Use ENTRY/ENDPROC macros for assembly routines
4. Fix SSSE3 feature check in glue code
Thanks to Peter Anvin, Jussi Kivilinna and Jim Kukunas for their reviews and comments.
Tim Chen (10):
Expose SHA256 generic routine to be callable externally.
Optimized sha256 x86_64 assembly routine using Supplemental SSE3
instructions.
Optimized sha256 x86_64 assembly routine with AVX instructions.
Optimized sha256 x86_64 routine using AVX2's RORX instructions
Create module providing optimized SHA256 routines using SSSE3, AVX or
AVX2 instructions.
Expose generic sha512 routine to be callable from other modules
Optimized SHA512 x86_64 assembly routine using Supplemental SSE3
instructions.
Optimized SHA512 x86_64 assembly routine using AVX instructions.
Optimized SHA512 x86_64 assembly routine using AVX2 RORX instruction.
Create module providing optimized SHA512 routines using SSSE3, AVX or
AVX2 instructions.
arch/x86/crypto/Makefile | 4 +
arch/x86/crypto/sha256-avx-asm.S | 496 +++++++++++++++++++++++
arch/x86/crypto/sha256-avx2-asm.S | 772 ++++++++++++++++++++++++++++++++++++
arch/x86/crypto/sha256-ssse3-asm.S | 506 +++++++++++++++++++++++
arch/x86/crypto/sha256_ssse3_glue.c | 275 +++++++++++++
arch/x86/crypto/sha512-avx-asm.S | 423 ++++++++++++++++++++
arch/x86/crypto/sha512-avx2-asm.S | 743 ++++++++++++++++++++++++++++++++++
arch/x86/crypto/sha512-ssse3-asm.S | 421 ++++++++++++++++++++
arch/x86/crypto/sha512_ssse3_glue.c | 282 +++++++++++++
crypto/Kconfig | 22 +
crypto/sha256_generic.c | 11 +-
crypto/sha512_generic.c | 13 +-
include/crypto/sha.h | 5 +
13 files changed, 3962 insertions(+), 11 deletions(-)
create mode 100644 arch/x86/crypto/sha256-avx-asm.S
create mode 100644 arch/x86/crypto/sha256-avx2-asm.S
create mode 100644 arch/x86/crypto/sha256-ssse3-asm.S
create mode 100644 arch/x86/crypto/sha256_ssse3_glue.c
create mode 100644 arch/x86/crypto/sha512-avx-asm.S
create mode 100644 arch/x86/crypto/sha512-avx2-asm.S
create mode 100644 arch/x86/crypto/sha512-ssse3-asm.S
create mode 100644 arch/x86/crypto/sha512_ssse3_glue.c
--
1.7.11.7
^ permalink raw reply [flat|nested] 32+ messages in thread
* [PATCH v2 01/10] Expose SHA256 generic routine to be callable externally.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:58 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Other SHA256 routine may need to use the generic routine when
FPU is not available.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
crypto/sha256_generic.c | 11 ++++++-----
include/crypto/sha.h | 2 ++
2 files changed, 8 insertions(+), 5 deletions(-)
diff --git a/crypto/sha256_generic.c b/crypto/sha256_generic.c
index c3ed4ec..5433667 100644
--- a/crypto/sha256_generic.c
+++ b/crypto/sha256_generic.c
@@ -246,7 +246,7 @@ static int sha256_init(struct shash_desc *desc)
return 0;
}
-static int sha256_update(struct shash_desc *desc, const u8 *data,
+int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
unsigned int len)
{
struct sha256_state *sctx = shash_desc_ctx(desc);
@@ -277,6 +277,7 @@ static int sha256_update(struct shash_desc *desc, const u8 *data,
return 0;
}
+EXPORT_SYMBOL(crypto_sha256_update);
static int sha256_final(struct shash_desc *desc, u8 *out)
{
@@ -293,10 +294,10 @@ static int sha256_final(struct shash_desc *desc, u8 *out)
/* Pad out to 56 mod 64. */
index = sctx->count & 0x3f;
pad_len = (index < 56) ? (56 - index) : ((64+56) - index);
- sha256_update(desc, padding, pad_len);
+ crypto_sha256_update(desc, padding, pad_len);
/* Append length (before padding) */
- sha256_update(desc, (const u8 *)&bits, sizeof(bits));
+ crypto_sha256_update(desc, (const u8 *)&bits, sizeof(bits));
/* Store state in digest */
for (i = 0; i < 8; i++)
@@ -339,7 +340,7 @@ static int sha256_import(struct shash_desc *desc, const void *in)
static struct shash_alg sha256_algs[2] = { {
.digestsize = SHA256_DIGEST_SIZE,
.init = sha256_init,
- .update = sha256_update,
+ .update = crypto_sha256_update,
.final = sha256_final,
.export = sha256_export,
.import = sha256_import,
@@ -355,7 +356,7 @@ static struct shash_alg sha256_algs[2] = { {
}, {
.digestsize = SHA224_DIGEST_SIZE,
.init = sha224_init,
- .update = sha256_update,
+ .update = crypto_sha256_update,
.final = sha224_final,
.descsize = sizeof(struct sha256_state),
.base = {
diff --git a/include/crypto/sha.h b/include/crypto/sha.h
index c6c9c1f..f46ff61 100644
--- a/include/crypto/sha.h
+++ b/include/crypto/sha.h
@@ -87,4 +87,6 @@ struct shash_desc;
extern int crypto_sha1_update(struct shash_desc *desc, const u8 *data,
unsigned int len);
+extern int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len);
#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 01/10] Expose SHA256 generic routine to be callable externally.
@ 2013-03-26 20:58 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Other SHA256 routine may need to use the generic routine when
FPU is not available.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
crypto/sha256_generic.c | 11 ++++++-----
include/crypto/sha.h | 2 ++
2 files changed, 8 insertions(+), 5 deletions(-)
diff --git a/crypto/sha256_generic.c b/crypto/sha256_generic.c
index c3ed4ec..5433667 100644
--- a/crypto/sha256_generic.c
+++ b/crypto/sha256_generic.c
@@ -246,7 +246,7 @@ static int sha256_init(struct shash_desc *desc)
return 0;
}
-static int sha256_update(struct shash_desc *desc, const u8 *data,
+int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
unsigned int len)
{
struct sha256_state *sctx = shash_desc_ctx(desc);
@@ -277,6 +277,7 @@ static int sha256_update(struct shash_desc *desc, const u8 *data,
return 0;
}
+EXPORT_SYMBOL(crypto_sha256_update);
static int sha256_final(struct shash_desc *desc, u8 *out)
{
@@ -293,10 +294,10 @@ static int sha256_final(struct shash_desc *desc, u8 *out)
/* Pad out to 56 mod 64. */
index = sctx->count & 0x3f;
pad_len = (index < 56) ? (56 - index) : ((64+56) - index);
- sha256_update(desc, padding, pad_len);
+ crypto_sha256_update(desc, padding, pad_len);
/* Append length (before padding) */
- sha256_update(desc, (const u8 *)&bits, sizeof(bits));
+ crypto_sha256_update(desc, (const u8 *)&bits, sizeof(bits));
/* Store state in digest */
for (i = 0; i < 8; i++)
@@ -339,7 +340,7 @@ static int sha256_import(struct shash_desc *desc, const void *in)
static struct shash_alg sha256_algs[2] = { {
.digestsize = SHA256_DIGEST_SIZE,
.init = sha256_init,
- .update = sha256_update,
+ .update = crypto_sha256_update,
.final = sha256_final,
.export = sha256_export,
.import = sha256_import,
@@ -355,7 +356,7 @@ static struct shash_alg sha256_algs[2] = { {
}, {
.digestsize = SHA224_DIGEST_SIZE,
.init = sha224_init,
- .update = sha256_update,
+ .update = crypto_sha256_update,
.final = sha224_final,
.descsize = sizeof(struct sha256_state),
.base = {
diff --git a/include/crypto/sha.h b/include/crypto/sha.h
index c6c9c1f..f46ff61 100644
--- a/include/crypto/sha.h
+++ b/include/crypto/sha.h
@@ -87,4 +87,6 @@ struct shash_desc;
extern int crypto_sha1_update(struct shash_desc *desc, const u8 *data,
unsigned int len);
+extern int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len);
#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 02/10] Optimized sha256 x86_64 assembly routine using Supplemental SSE3 instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:58 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA256 x86_64 assembly routine optimized with SSSE3 instructions.
Speedup of 40% or more has been measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256-ssse3-asm.S | 506 +++++++++++++++++++++++++++++++++++++
1 file changed, 506 insertions(+)
create mode 100644 arch/x86/crypto/sha256-ssse3-asm.S
diff --git a/arch/x86/crypto/sha256-ssse3-asm.S b/arch/x86/crypto/sha256-ssse3-asm.S
new file mode 100644
index 0000000..98d3c39
--- /dev/null
+++ b/arch/x86/crypto/sha256-ssse3-asm.S
@@ -0,0 +1,506 @@
+########################################################################
+# Implement fast SHA-256 with SSSE3 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.
+#
+########################################################################
+
+#include <linux/linkage.h>
+
+## assume buffers not aligned
+#define MOVDQ movdqu
+
+################################ 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
+
+################################
+
+# 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 \p2, \p1
+ pshufb \p3, \p1
+.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
+
+
+
+_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
+ movdqa X3, XTMP0
+ mov e, y0 # y0 = e
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ palignr $4, X2, XTMP0 # XTMP0 = W[-7]
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ movdqa X1, XTMP1
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ xor g, y2 # y2 = f^g
+ paddd X0, XTMP0 # XTMP0 = W[-7] + W[-16]
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+ and e, y2 # y2 = (f^g)&e
+ ror $(13-2), y1 # y1 = (a >> (13-2)) ^ (a >> (22-2))
+ ## compute s0
+ palignr $4, X0, XTMP1 # XTMP1 = W[-15]
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ movdqa XTMP1, XTMP2 # XTMP2 = W[-15]
+ 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
+ movdqa XTMP1, XTMP3 # XTMP3 = W[-15]
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pslld $(32-7), XTMP1 #
+ or c, y0 # y0 = a|c
+ add h, d # d = d + h + S1 + CH + k + w
+ and c, y2 # y2 = a&c
+ psrld $7, XTMP2 #
+ and b, y0 # y0 = (a|c)&b
+ add y1, h # h = h + S1 + CH + k + w + S0
+ por XTMP2, XTMP1 # XTMP1 = W[-15] 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 #
+ movdqa XTMP3, XTMP2 # XTMP2 = W[-15]
+ mov e, y0 # y0 = e
+ mov a, y1 # y1 = a
+ movdqa XTMP3, XTMP4 # XTMP4 = W[-15]
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ pslld $(32-18), XTMP3 #
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ xor g, y2 # y2 = f^g
+ psrld $18, XTMP2 #
+ 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
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ pxor XTMP3, XTMP1
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ psrld $3, XTMP4 # XTMP4 = W[-15] >> 3
+ add y0, y2 # y2 = S1 + CH
+ add (1*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+ pxor XTMP2, XTMP1 # XTMP1 = W[-15] ror 7 ^ W[-15] ror 18
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pxor XTMP4, 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
+ pshufd $0b11111010, X3, XTMP2 # XTMP2 = W[-2] {BBAA}
+ and b, y0 # y0 = (a|c)&b
+ add y1, h # h = h + S1 + CH + k + w + S0
+ paddd XTMP1, 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
+ movdqa XTMP2, XTMP3 # XTMP3 = W[-2] {BBAA}
+ mov e, y0 # y0 = e
+ mov a, y1 # y1 = a
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ movdqa XTMP2, XTMP4 # XTMP4 = W[-2] {BBAA}
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ mov f, y2 # y2 = f
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ psrlq $17, XTMP2 # XTMP2 = W[-2] ror 17 {xBxA}
+ xor g, y2 # y2 = f^g
+ psrlq $19, XTMP3 # XTMP3 = W[-2] ror 19 {xBxA}
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+ and e, y2 # y2 = (f^g)&e
+ psrld $10, XTMP4 # XTMP4 = W[-2] >> 10 {BBAA}
+ 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
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ pxor XTMP3, XTMP2
+ add y0, y2 # y2 = S1 + CH
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+ add (2*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ pxor XTMP2, XTMP4 # XTMP4 = s1 {xBxA}
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pshufb SHUF_00BA, 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
+ paddd XTMP4, 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
+ pshufd $0b01010000, XTMP0, XTMP2 # XTMP2 = W[-2] {BBAA}
+ or y2, y0 # y0 = MAJ = (a|c)&b)|(a&c)
+ add y0, h # h = h + S1 + CH + k + w + S0 + MAJ
+ #
+ ROTATE_ARGS #
+ movdqa XTMP2, XTMP3 # XTMP3 = W[-2] {DDCC}
+ mov e, y0 # y0 = e
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ movdqa XTMP2, X0 # X0 = W[-2] {DDCC}
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ psrlq $17, XTMP2 # XTMP2 = W[-2] ror 17 {xDxC}
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ xor g, y2 # y2 = f^g
+ psrlq $19, XTMP3 # XTMP3 = W[-2] ror 19 {xDxC}
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25
+ and e, y2 # y2 = (f^g)&e
+ ror $(13-2), y1 # y1 = (a >> (13-2)) ^ (a >> (22-2))
+ psrld $10, X0 # X0 = W[-2] >> 10 {DDCC}
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>2
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ pxor XTMP3, XTMP2 #
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>2
+ add y0, y2 # y2 = S1 + CH
+ add (3*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ pxor XTMP2, X0 # X0 = s1 {xDxC}
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pshufb SHUF_DC00, X0 # X0 = s1 {DC00}
+ or c, y0 # y0 = a|c
+ add h, d # d = d + h + S1 + CH + k + w
+ and c, y2 # y2 = a&c
+ paddd XTMP0, 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
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ mov f, y2 # y2 = f
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ 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))
+ 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))
+ 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
+ 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_ssse3(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_ssse3)
+.align 32
+ pushq %rbx
+ pushq %rbp
+ pushq %r13
+ pushq %r14
+ pushq %r15
+ pushq %r12
+
+ mov %rsp, %r12
+ subq $STACK_SIZE, %rsp
+ and $~15, %rsp
+
+ shl $6, NUM_BLKS # convert to bytes
+ jz done_hash
+ add INP, NUM_BLKS
+ mov NUM_BLKS, _INP_END(%rsp) # pointer to end of data
+
+ ## 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
+
+ movdqa PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
+ movdqa _SHUF_00BA(%rip), SHUF_00BA
+ movdqa _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:
+ movdqa (TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 1*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 2*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 3*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ add $4*16, TBL
+ FOUR_ROUNDS_AND_SCHED
+
+ sub $1, SRND
+ jne loop1
+
+ mov $2, SRND
+loop2:
+ paddd (TBL), X0
+ movdqa X0, _XFER(%rsp)
+ DO_ROUND 0
+ DO_ROUND 1
+ DO_ROUND 2
+ DO_ROUND 3
+ paddd 1*16(TBL), X1
+ movdqa X1, _XFER(%rsp)
+ add $2*16, TBL
+ DO_ROUND 0
+ DO_ROUND 1
+ DO_ROUND 2
+ DO_ROUND 3
+
+ movdqa X2, X0
+ movdqa 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_ssse3)
+
+.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
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 02/10] Optimized sha256 x86_64 assembly routine using Supplemental SSE3 instructions.
@ 2013-03-26 20:58 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:58 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA256 x86_64 assembly routine optimized with SSSE3 instructions.
Speedup of 40% or more has been measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256-ssse3-asm.S | 506 +++++++++++++++++++++++++++++++++++++
1 file changed, 506 insertions(+)
create mode 100644 arch/x86/crypto/sha256-ssse3-asm.S
diff --git a/arch/x86/crypto/sha256-ssse3-asm.S b/arch/x86/crypto/sha256-ssse3-asm.S
new file mode 100644
index 0000000..98d3c39
--- /dev/null
+++ b/arch/x86/crypto/sha256-ssse3-asm.S
@@ -0,0 +1,506 @@
+########################################################################
+# Implement fast SHA-256 with SSSE3 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.
+#
+########################################################################
+
+#include <linux/linkage.h>
+
+## assume buffers not aligned
+#define MOVDQ movdqu
+
+################################ 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
+
+################################
+
+# 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 \p2, \p1
+ pshufb \p3, \p1
+.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
+
+
+
+_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
+ movdqa X3, XTMP0
+ mov e, y0 # y0 = e
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ palignr $4, X2, XTMP0 # XTMP0 = W[-7]
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ movdqa X1, XTMP1
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ xor g, y2 # y2 = f^g
+ paddd X0, XTMP0 # XTMP0 = W[-7] + W[-16]
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+ and e, y2 # y2 = (f^g)&e
+ ror $(13-2), y1 # y1 = (a >> (13-2)) ^ (a >> (22-2))
+ ## compute s0
+ palignr $4, X0, XTMP1 # XTMP1 = W[-15]
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ movdqa XTMP1, XTMP2 # XTMP2 = W[-15]
+ 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
+ movdqa XTMP1, XTMP3 # XTMP3 = W[-15]
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pslld $(32-7), XTMP1 #
+ or c, y0 # y0 = a|c
+ add h, d # d = d + h + S1 + CH + k + w
+ and c, y2 # y2 = a&c
+ psrld $7, XTMP2 #
+ and b, y0 # y0 = (a|c)&b
+ add y1, h # h = h + S1 + CH + k + w + S0
+ por XTMP2, XTMP1 # XTMP1 = W[-15] 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 #
+ movdqa XTMP3, XTMP2 # XTMP2 = W[-15]
+ mov e, y0 # y0 = e
+ mov a, y1 # y1 = a
+ movdqa XTMP3, XTMP4 # XTMP4 = W[-15]
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ pslld $(32-18), XTMP3 #
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ xor g, y2 # y2 = f^g
+ psrld $18, XTMP2 #
+ 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
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ pxor XTMP3, XTMP1
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22-2))
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ psrld $3, XTMP4 # XTMP4 = W[-15] >> 3
+ add y0, y2 # y2 = S1 + CH
+ add (1*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+ pxor XTMP2, XTMP1 # XTMP1 = W[-15] ror 7 ^ W[-15] ror 18
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pxor XTMP4, 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
+ pshufd $0b11111010, X3, XTMP2 # XTMP2 = W[-2] {BBAA}
+ and b, y0 # y0 = (a|c)&b
+ add y1, h # h = h + S1 + CH + k + w + S0
+ paddd XTMP1, 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
+ movdqa XTMP2, XTMP3 # XTMP3 = W[-2] {BBAA}
+ mov e, y0 # y0 = e
+ mov a, y1 # y1 = a
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ movdqa XTMP2, XTMP4 # XTMP4 = W[-2] {BBAA}
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ mov f, y2 # y2 = f
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ psrlq $17, XTMP2 # XTMP2 = W[-2] ror 17 {xBxA}
+ xor g, y2 # y2 = f^g
+ psrlq $19, XTMP3 # XTMP3 = W[-2] ror 19 {xBxA}
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25-6))
+ and e, y2 # y2 = (f^g)&e
+ psrld $10, XTMP4 # XTMP4 = W[-2] >> 10 {BBAA}
+ 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
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>25)
+ pxor XTMP3, XTMP2
+ add y0, y2 # y2 = S1 + CH
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>22)
+ add (2*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ pxor XTMP2, XTMP4 # XTMP4 = s1 {xBxA}
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pshufb SHUF_00BA, 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
+ paddd XTMP4, 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
+ pshufd $0b01010000, XTMP0, XTMP2 # XTMP2 = W[-2] {BBAA}
+ or y2, y0 # y0 = MAJ = (a|c)&b)|(a&c)
+ add y0, h # h = h + S1 + CH + k + w + S0 + MAJ
+ #
+ ROTATE_ARGS #
+ movdqa XTMP2, XTMP3 # XTMP3 = W[-2] {DDCC}
+ mov e, y0 # y0 = e
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ movdqa XTMP2, X0 # X0 = W[-2] {DDCC}
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ mov f, y2 # y2 = f
+ ror $(11-6), y0 # y0 = (e >> (11-6)) ^ (e >> (25-6))
+ psrlq $17, XTMP2 # XTMP2 = W[-2] ror 17 {xDxC}
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ xor g, y2 # y2 = f^g
+ psrlq $19, XTMP3 # XTMP3 = W[-2] ror 19 {xDxC}
+ xor e, y0 # y0 = e ^ (e >> (11-6)) ^ (e >> (25
+ and e, y2 # y2 = (f^g)&e
+ ror $(13-2), y1 # y1 = (a >> (13-2)) ^ (a >> (22-2))
+ psrld $10, X0 # X0 = W[-2] >> 10 {DDCC}
+ xor a, y1 # y1 = a ^ (a >> (13-2)) ^ (a >> (22
+ ror $6, y0 # y0 = S1 = (e>>6) & (e>>11) ^ (e>>2
+ xor g, y2 # y2 = CH = ((f^g)&e)^g
+ pxor XTMP3, XTMP2 #
+ ror $2, y1 # y1 = S0 = (a>>2) ^ (a>>13) ^ (a>>2
+ add y0, y2 # y2 = S1 + CH
+ add (3*4 + _XFER)(%rsp), y2 # y2 = k + w + S1 + CH
+ pxor XTMP2, X0 # X0 = s1 {xDxC}
+ mov a, y0 # y0 = a
+ add y2, h # h = h + S1 + CH + k + w
+ mov a, y2 # y2 = a
+ pshufb SHUF_DC00, X0 # X0 = s1 {DC00}
+ or c, y0 # y0 = a|c
+ add h, d # d = d + h + S1 + CH + k + w
+ and c, y2 # y2 = a&c
+ paddd XTMP0, 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
+ ror $(25-11), y0 # y0 = e >> (25-11)
+ mov a, y1 # y1 = a
+ xor e, y0 # y0 = e ^ (e >> (25-11))
+ ror $(22-13), y1 # y1 = a >> (22-13)
+ mov f, y2 # y2 = f
+ xor a, y1 # y1 = a ^ (a >> (22-13)
+ 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))
+ 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))
+ 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
+ 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_ssse3(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_ssse3)
+.align 32
+ pushq %rbx
+ pushq %rbp
+ pushq %r13
+ pushq %r14
+ pushq %r15
+ pushq %r12
+
+ mov %rsp, %r12
+ subq $STACK_SIZE, %rsp
+ and $~15, %rsp
+
+ shl $6, NUM_BLKS # convert to bytes
+ jz done_hash
+ add INP, NUM_BLKS
+ mov NUM_BLKS, _INP_END(%rsp) # pointer to end of data
+
+ ## 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
+
+ movdqa PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
+ movdqa _SHUF_00BA(%rip), SHUF_00BA
+ movdqa _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:
+ movdqa (TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 1*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 2*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ movdqa 3*16(TBL), XFER
+ paddd X0, XFER
+ movdqa XFER, _XFER(%rsp)
+ add $4*16, TBL
+ FOUR_ROUNDS_AND_SCHED
+
+ sub $1, SRND
+ jne loop1
+
+ mov $2, SRND
+loop2:
+ paddd (TBL), X0
+ movdqa X0, _XFER(%rsp)
+ DO_ROUND 0
+ DO_ROUND 1
+ DO_ROUND 2
+ DO_ROUND 3
+ paddd 1*16(TBL), X1
+ movdqa X1, _XFER(%rsp)
+ add $2*16, TBL
+ DO_ROUND 0
+ DO_ROUND 1
+ DO_ROUND 2
+ DO_ROUND 3
+
+ movdqa X2, X0
+ movdqa 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_ssse3)
+
+.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
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 03/10] Optimized sha256 x86_64 assembly routine with AVX instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
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>
---
arch/x86/crypto/sha256-avx-asm.S | 496 +++++++++++++++++++++++++++++++++++++++
1 file changed, 496 insertions(+)
create mode 100644 arch/x86/crypto/sha256-avx-asm.S
diff --git a/arch/x86/crypto/sha256-avx-asm.S b/arch/x86/crypto/sha256-avx-asm.S
new file mode 100644
index 0000000..56610c4
--- /dev/null
+++ b/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
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 03/10] Optimized sha256 x86_64 assembly routine with AVX instructions.
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
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>
---
arch/x86/crypto/sha256-avx-asm.S | 496 +++++++++++++++++++++++++++++++++++++++
1 file changed, 496 insertions(+)
create mode 100644 arch/x86/crypto/sha256-avx-asm.S
diff --git a/arch/x86/crypto/sha256-avx-asm.S b/arch/x86/crypto/sha256-avx-asm.S
new file mode 100644
index 0000000..56610c4
--- /dev/null
+++ b/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
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 04/10] Optimized sha256 x86_64 routine using AVX2's RORX instructions
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA256 x86_64 assembly routine optimized with SSE, AVX and
AVX2's RORX instructions. Speedup of 70% or more has been
measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256-avx2-asm.S | 772 ++++++++++++++++++++++++++++++++++++++
1 file changed, 772 insertions(+)
create mode 100644 arch/x86/crypto/sha256-avx2-asm.S
diff --git a/arch/x86/crypto/sha256-avx2-asm.S b/arch/x86/crypto/sha256-avx2-asm.S
new file mode 100644
index 0000000..9e86944
--- /dev/null
+++ b/arch/x86/crypto/sha256-avx2-asm.S
@@ -0,0 +1,772 @@
+########################################################################
+# Implement fast SHA-256 with AVX2 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 2 blocks at a time, with 4 lanes per block
+########################################################################
+
+#ifdef CONFIG_AS_AVX2
+#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
+
+################################
+
+X0 = %ymm4
+X1 = %ymm5
+X2 = %ymm6
+X3 = %ymm7
+
+# XMM versions of above
+XWORD0 = %xmm4
+XWORD1 = %xmm5
+XWORD2 = %xmm6
+XWORD3 = %xmm7
+
+XTMP0 = %ymm0
+XTMP1 = %ymm1
+XTMP2 = %ymm2
+XTMP3 = %ymm3
+XTMP4 = %ymm8
+XFER = %ymm9
+XTMP5 = %ymm11
+
+SHUF_00BA = %ymm10 # shuffle xBxA -> 00BA
+SHUF_DC00 = %ymm12 # shuffle xDxC -> DC00
+BYTE_FLIP_MASK = %ymm13
+
+X_BYTE_FLIP_MASK = %xmm13 # XMM version of BYTE_FLIP_MASK
+
+NUM_BLKS = %rdx # 3rd arg
+CTX = %rsi # 2nd arg
+INP = %rdi # 1st arg
+c = %ecx
+d = %r8d
+e = %edx # clobbers NUM_BLKS
+y3 = %edi # clobbers INP
+
+
+TBL = %rbp
+SRND = CTX # SRND is same register as CTX
+
+a = %eax
+b = %ebx
+f = %r9d
+g = %r10d
+h = %r11d
+old_h = %r11d
+
+T1 = %r12d
+y0 = %r13d
+y1 = %r14d
+y2 = %r15d
+
+
+_XFER_SIZE = 2*64*4 # 2 blocks, 64 rounds, 4 bytes/round
+_XMM_SAVE_SIZE = 0
+_INP_END_SIZE = 8
+_INP_SIZE = 8
+_CTX_SIZE = 8
+_RSP_SIZE = 8
+
+_XFER = 0
+_XMM_SAVE = _XFER + _XFER_SIZE
+_INP_END = _XMM_SAVE + _XMM_SAVE_SIZE
+_INP = _INP_END + _INP_END_SIZE
+_CTX = _INP + _INP_SIZE
+_RSP = _CTX + _CTX_SIZE
+STACK_SIZE = _RSP + _RSP_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
+ old_h = h
+ TMP_ = h
+ h = g
+ g = f
+ f = e
+ e = d
+ d = c
+ c = b
+ b = a
+ a = TMP_
+.endm
+
+.macro FOUR_ROUNDS_AND_SCHED disp
+################################### RND N + 0 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+
+ addl \disp(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+ vpalignr $4, X2, X3, XTMP0 # XTMP0 = W[-7]
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+ vpaddd X0, XTMP0, XTMP0 # XTMP0 = W[-7] + W[-16]# y1 = (e >> 6)# S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+
+ and e, y2 # y2 = (f^g)&e # CH
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ vpalignr $4, X0, X1, XTMP1 # XTMP1 = W[-15]
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ vpsrld $7, XTMP1, XTMP2
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+
+ add y0, y2 # y2 = S1 + CH # --
+ vpslld $(32-7), XTMP1, XTMP3
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ vpor XTMP2, XTMP3, XTMP3 # XTMP3 = W[-15] ror 7
+
+ vpsrld $18, XTMP1, XTMP2
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+
+ ROTATE_ARGS
+
+################################### RND N + 1 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ offset = \disp + 1*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ vpsrld $3, XTMP1, XTMP4 # XTMP4 = W[-15] >> 3
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+
+ vpslld $(32-18), XTMP1, XTMP1
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+
+ vpxor XTMP1, XTMP3, XTMP3
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpxor XTMP2, XTMP3, XTMP3 # XTMP3 = W[-15] ror 7 ^ W[-15] ror 18
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ vpxor XTMP4, XTMP3, XTMP1 # XTMP1 = s0
+ vpshufd $0b11111010, X3, XTMP2 # XTMP2 = W[-2] {BBAA}
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ vpaddd XTMP1, XTMP0, XTMP0 # XTMP0 = W[-16] + W[-7] + s0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ vpsrld $10, XTMP2, XTMP4 # XTMP4 = W[-2] >> 10 {BBAA}
+
+
+ ROTATE_ARGS
+
+################################### RND N + 2 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ offset = \disp + 2*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+
+ vpsrlq $19, XTMP2, XTMP3 # XTMP3 = W[-2] ror 19 {xBxA}
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ xor g, y2 # y2 = f^g # CH
+
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ vpsrlq $17, XTMP2, XTMP2 # XTMP2 = W[-2] ror 17 {xBxA}
+ and e, y2 # y2 = (f^g)&e # CH
+
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ vpxor XTMP3, XTMP2, XTMP2
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ vpxor XTMP2, XTMP4, XTMP4 # XTMP4 = s1 {xBxA}
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpshufb SHUF_00BA, XTMP4, XTMP4 # XTMP4 = s1 {00BA}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a ,T1 # T1 = (a >> 2) # S0
+ vpaddd XTMP4, XTMP0, XTMP0 # XTMP0 = {..., ..., W[1], W[0]}
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+ vpshufd $0b01010000, XTMP0, XTMP2 # XTMP2 = W[-2] {DDCC}
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1,h # h = k + w + h + S0 # --
+ add y2,d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2,h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3,h # h = t1 + S0 + MAJ # --
+
+
+ ROTATE_ARGS
+
+################################### RND N + 3 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ offset = \disp + 3*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ vpsrld $10, XTMP2, XTMP5 # XTMP5 = W[-2] >> 10 {DDCC}
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ vpsrlq $19, XTMP2, XTMP3 # XTMP3 = W[-2] ror 19 {xDxC}
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ vpsrlq $17, XTMP2, XTMP2 # XTMP2 = W[-2] ror 17 {xDxC}
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpxor XTMP3, XTMP2, XTMP2
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ add y0, y2 # y2 = S1 + CH # --
+
+ vpxor XTMP2, XTMP5, XTMP5 # XTMP5 = s1 {xDxC}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ vpshufb SHUF_DC00, XTMP5, XTMP5 # XTMP5 = s1 {DC00}
+
+ vpaddd XTMP0, XTMP5, X0 # X0 = {W[3], W[2], W[1], W[0]}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+
+ add y1, h # h = k + w + h + S0 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ ROTATE_ARGS
+ rotate_Xs
+.endm
+
+.macro DO_4ROUNDS disp
+################################### RND N + 0 ###########################
+
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ addl \disp(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 1 ###########################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*1 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 2 ##############################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*2 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 3 ###########################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*3 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ ROTATE_ARGS
+
+.endm
+
+########################################################################
+## void sha256_transform_rorx(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_rorx)
+.align 32
+ pushq %rbx
+ pushq %rbp
+ pushq %r12
+ pushq %r13
+ pushq %r14
+ pushq %r15
+
+ mov %rsp, %rax
+ subq $STACK_SIZE, %rsp
+ and $-32, %rsp # align rsp to 32 byte boundary
+ mov %rax, _RSP(%rsp)
+
+
+ shl $6, NUM_BLKS # convert to bytes
+ jz done_hash
+ lea -64(INP, NUM_BLKS), NUM_BLKS # pointer to last block
+ mov NUM_BLKS, _INP_END(%rsp)
+
+ cmp NUM_BLKS, INP
+ je only_one_block
+
+ ## load initial digest
+ mov (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
+
+ mov CTX, _CTX(%rsp)
+
+loop0:
+ lea K256(%rip), TBL
+
+ ## Load first 16 dwords from two blocks
+ VMOVDQ 0*32(INP),XTMP0
+ VMOVDQ 1*32(INP),XTMP1
+ VMOVDQ 2*32(INP),XTMP2
+ VMOVDQ 3*32(INP),XTMP3
+
+ ## byte swap data
+ vpshufb BYTE_FLIP_MASK, XTMP0, XTMP0
+ vpshufb BYTE_FLIP_MASK, XTMP1, XTMP1
+ vpshufb BYTE_FLIP_MASK, XTMP2, XTMP2
+ vpshufb BYTE_FLIP_MASK, XTMP3, XTMP3
+
+ ## transpose data into high/low halves
+ vperm2i128 $0x20, XTMP2, XTMP0, X0
+ vperm2i128 $0x31, XTMP2, XTMP0, X1
+ vperm2i128 $0x20, XTMP3, XTMP1, X2
+ vperm2i128 $0x31, XTMP3, XTMP1, X3
+
+last_block_enter:
+ add $64, INP
+ mov INP, _INP(%rsp)
+
+ ## schedule 48 input dwords, by doing 3 rounds of 12 each
+ xor SRND, SRND
+
+.align 16
+loop1:
+ vpaddd 0*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 0*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 0*32
+
+ vpaddd 1*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 1*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 1*32
+
+ vpaddd 2*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 2*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 2*32
+
+ vpaddd 3*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 3*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 3*32
+
+ add $4*32, SRND
+ cmp $3*4*32, SRND
+ jb loop1
+
+loop2:
+ ## Do last 16 rounds with no scheduling
+ vpaddd 0*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 0*32+_XFER(%rsp, SRND)
+ DO_4ROUNDS _XFER + 0*32
+ vpaddd 1*32(TBL, SRND), X1, XFER
+ vmovdqa XFER, 1*32+_XFER(%rsp, SRND)
+ DO_4ROUNDS _XFER + 1*32
+ add $2*32, SRND
+
+ vmovdqa X2, X0
+ vmovdqa X3, X1
+
+ cmp $4*4*32, SRND
+ jb loop2
+
+ mov _CTX(%rsp), CTX
+ mov _INP(%rsp), INP
+
+ 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
+
+ cmp _INP_END(%rsp), INP
+ ja done_hash
+
+ #### Do second block using previously scheduled results
+ xor SRND, SRND
+.align 16
+loop3:
+ DO_4ROUNDS _XFER + 0*32 + 16
+ DO_4ROUNDS _XFER + 1*32 + 16
+ add $2*32, SRND
+ cmp $4*4*32, SRND
+ jb loop3
+
+ mov _CTX(%rsp), CTX
+ mov _INP(%rsp), INP
+ add $64, INP
+
+ 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
+
+ cmp _INP_END(%rsp), INP
+ jb loop0
+ ja done_hash
+
+do_last_block:
+ #### do last block
+ lea K256(%rip), TBL
+
+ VMOVDQ 0*16(INP),XWORD0
+ VMOVDQ 1*16(INP),XWORD1
+ VMOVDQ 2*16(INP),XWORD2
+ VMOVDQ 3*16(INP),XWORD3
+
+ vpshufb X_BYTE_FLIP_MASK, XWORD0, XWORD0
+ vpshufb X_BYTE_FLIP_MASK, XWORD1, XWORD1
+ vpshufb X_BYTE_FLIP_MASK, XWORD2, XWORD2
+ vpshufb X_BYTE_FLIP_MASK, XWORD3, XWORD3
+
+ jmp last_block_enter
+
+only_one_block:
+
+ ## 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
+
+ mov CTX, _CTX(%rsp)
+ jmp do_last_block
+
+done_hash:
+
+ mov _RSP(%rsp), %rsp
+
+ popq %r15
+ popq %r14
+ popq %r13
+ popq %r12
+ popq %rbp
+ popq %rbx
+ ret
+ENDPROC(sha256_transform_rorx)
+
+.data
+.align 64
+K256:
+ .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+ .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+ .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+ .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+ .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+ .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+ .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+ .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+ .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+ .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+ .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+ .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+ .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+ .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+ .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+ .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+ .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+ .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+ .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+ .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+ .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+ .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+ .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+ .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+ .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+ .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+ .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+ .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+ .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+ .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+ .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+ .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+
+PSHUFFLE_BYTE_FLIP_MASK:
+ .octa 0x0c0d0e0f08090a0b0405060700010203,0x0c0d0e0f08090a0b0405060700010203
+
+# shuffle xBxA -> 00BA
+_SHUF_00BA:
+ .octa 0xFFFFFFFFFFFFFFFF0b0a090803020100,0xFFFFFFFFFFFFFFFF0b0a090803020100
+
+# shuffle xDxC -> DC00
+_SHUF_DC00:
+ .octa 0x0b0a090803020100FFFFFFFFFFFFFFFF,0x0b0a090803020100FFFFFFFFFFFFFFFF
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 04/10] Optimized sha256 x86_64 routine using AVX2's RORX instructions
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA256 x86_64 assembly routine optimized with SSE, AVX and
AVX2's RORX instructions. Speedup of 70% or more has been
measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256-avx2-asm.S | 772 ++++++++++++++++++++++++++++++++++++++
1 file changed, 772 insertions(+)
create mode 100644 arch/x86/crypto/sha256-avx2-asm.S
diff --git a/arch/x86/crypto/sha256-avx2-asm.S b/arch/x86/crypto/sha256-avx2-asm.S
new file mode 100644
index 0000000..9e86944
--- /dev/null
+++ b/arch/x86/crypto/sha256-avx2-asm.S
@@ -0,0 +1,772 @@
+########################################################################
+# Implement fast SHA-256 with AVX2 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 2 blocks at a time, with 4 lanes per block
+########################################################################
+
+#ifdef CONFIG_AS_AVX2
+#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
+
+################################
+
+X0 = %ymm4
+X1 = %ymm5
+X2 = %ymm6
+X3 = %ymm7
+
+# XMM versions of above
+XWORD0 = %xmm4
+XWORD1 = %xmm5
+XWORD2 = %xmm6
+XWORD3 = %xmm7
+
+XTMP0 = %ymm0
+XTMP1 = %ymm1
+XTMP2 = %ymm2
+XTMP3 = %ymm3
+XTMP4 = %ymm8
+XFER = %ymm9
+XTMP5 = %ymm11
+
+SHUF_00BA = %ymm10 # shuffle xBxA -> 00BA
+SHUF_DC00 = %ymm12 # shuffle xDxC -> DC00
+BYTE_FLIP_MASK = %ymm13
+
+X_BYTE_FLIP_MASK = %xmm13 # XMM version of BYTE_FLIP_MASK
+
+NUM_BLKS = %rdx # 3rd arg
+CTX = %rsi # 2nd arg
+INP = %rdi # 1st arg
+c = %ecx
+d = %r8d
+e = %edx # clobbers NUM_BLKS
+y3 = %edi # clobbers INP
+
+
+TBL = %rbp
+SRND = CTX # SRND is same register as CTX
+
+a = %eax
+b = %ebx
+f = %r9d
+g = %r10d
+h = %r11d
+old_h = %r11d
+
+T1 = %r12d
+y0 = %r13d
+y1 = %r14d
+y2 = %r15d
+
+
+_XFER_SIZE = 2*64*4 # 2 blocks, 64 rounds, 4 bytes/round
+_XMM_SAVE_SIZE = 0
+_INP_END_SIZE = 8
+_INP_SIZE = 8
+_CTX_SIZE = 8
+_RSP_SIZE = 8
+
+_XFER = 0
+_XMM_SAVE = _XFER + _XFER_SIZE
+_INP_END = _XMM_SAVE + _XMM_SAVE_SIZE
+_INP = _INP_END + _INP_END_SIZE
+_CTX = _INP + _INP_SIZE
+_RSP = _CTX + _CTX_SIZE
+STACK_SIZE = _RSP + _RSP_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
+ old_h = h
+ TMP_ = h
+ h = g
+ g = f
+ f = e
+ e = d
+ d = c
+ c = b
+ b = a
+ a = TMP_
+.endm
+
+.macro FOUR_ROUNDS_AND_SCHED disp
+################################### RND N + 0 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+
+ addl \disp(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+ vpalignr $4, X2, X3, XTMP0 # XTMP0 = W[-7]
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+ vpaddd X0, XTMP0, XTMP0 # XTMP0 = W[-7] + W[-16]# y1 = (e >> 6)# S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+
+ and e, y2 # y2 = (f^g)&e # CH
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ vpalignr $4, X0, X1, XTMP1 # XTMP1 = W[-15]
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ vpsrld $7, XTMP1, XTMP2
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+
+ add y0, y2 # y2 = S1 + CH # --
+ vpslld $(32-7), XTMP1, XTMP3
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ vpor XTMP2, XTMP3, XTMP3 # XTMP3 = W[-15] ror 7
+
+ vpsrld $18, XTMP1, XTMP2
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+
+ ROTATE_ARGS
+
+################################### RND N + 1 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ offset = \disp + 1*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ vpsrld $3, XTMP1, XTMP4 # XTMP4 = W[-15] >> 3
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+
+ vpslld $(32-18), XTMP1, XTMP1
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+
+ vpxor XTMP1, XTMP3, XTMP3
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpxor XTMP2, XTMP3, XTMP3 # XTMP3 = W[-15] ror 7 ^ W[-15] ror 18
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ vpxor XTMP4, XTMP3, XTMP1 # XTMP1 = s0
+ vpshufd $0b11111010, X3, XTMP2 # XTMP2 = W[-2] {BBAA}
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ vpaddd XTMP1, XTMP0, XTMP0 # XTMP0 = W[-16] + W[-7] + s0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ vpsrld $10, XTMP2, XTMP4 # XTMP4 = W[-2] >> 10 {BBAA}
+
+
+ ROTATE_ARGS
+
+################################### RND N + 2 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ offset = \disp + 2*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+
+ vpsrlq $19, XTMP2, XTMP3 # XTMP3 = W[-2] ror 19 {xBxA}
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ xor g, y2 # y2 = f^g # CH
+
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ vpsrlq $17, XTMP2, XTMP2 # XTMP2 = W[-2] ror 17 {xBxA}
+ and e, y2 # y2 = (f^g)&e # CH
+
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ vpxor XTMP3, XTMP2, XTMP2
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ vpxor XTMP2, XTMP4, XTMP4 # XTMP4 = s1 {xBxA}
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpshufb SHUF_00BA, XTMP4, XTMP4 # XTMP4 = s1 {00BA}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a ,T1 # T1 = (a >> 2) # S0
+ vpaddd XTMP4, XTMP0, XTMP0 # XTMP0 = {..., ..., W[1], W[0]}
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+ vpshufd $0b01010000, XTMP0, XTMP2 # XTMP2 = W[-2] {DDCC}
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1,h # h = k + w + h + S0 # --
+ add y2,d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2,h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3,h # h = t1 + S0 + MAJ # --
+
+
+ ROTATE_ARGS
+
+################################### RND N + 3 ############################
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ offset = \disp + 3*4
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ vpsrld $10, XTMP2, XTMP5 # XTMP5 = W[-2] >> 10 {DDCC}
+ mov f, y2 # y2 = f # CH
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ vpsrlq $19, XTMP2, XTMP3 # XTMP3 = W[-2] ror 19 {xDxC}
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ vpsrlq $17, XTMP2, XTMP2 # XTMP2 = W[-2] ror 17 {xDxC}
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ vpxor XTMP3, XTMP2, XTMP2
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ add y0, y2 # y2 = S1 + CH # --
+
+ vpxor XTMP2, XTMP5, XTMP5 # XTMP5 = s1 {xDxC}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ vpshufb SHUF_DC00, XTMP5, XTMP5 # XTMP5 = s1 {DC00}
+
+ vpaddd XTMP0, XTMP5, X0 # X0 = {W[3], W[2], W[1], W[0]}
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+
+ add y1, h # h = k + w + h + S0 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ ROTATE_ARGS
+ rotate_Xs
+.endm
+
+.macro DO_4ROUNDS disp
+################################### RND N + 0 ###########################
+
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ addl \disp(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 1 ###########################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*1 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 2 ##############################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*2 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ ROTATE_ARGS
+
+################################### RND N + 3 ###########################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $25, e, y0 # y0 = e >> 25 # S1A
+ rorx $11, e, y1 # y1 = e >> 11 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) # S1
+ rorx $6, e, y1 # y1 = (e >> 6) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>25) ^ (e>>11) ^ (e>>6) # S1
+ rorx $13, a, T1 # T1 = a >> 13 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $22, a, y1 # y1 = a >> 22 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) # S0
+ rorx $2, a, T1 # T1 = (a >> 2) # S0
+ offset = 4*3 + \disp
+ addl offset(%rsp, SRND), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>22) ^ (a>>13) ^ (a>>2) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ ROTATE_ARGS
+
+.endm
+
+########################################################################
+## void sha256_transform_rorx(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_rorx)
+.align 32
+ pushq %rbx
+ pushq %rbp
+ pushq %r12
+ pushq %r13
+ pushq %r14
+ pushq %r15
+
+ mov %rsp, %rax
+ subq $STACK_SIZE, %rsp
+ and $-32, %rsp # align rsp to 32 byte boundary
+ mov %rax, _RSP(%rsp)
+
+
+ shl $6, NUM_BLKS # convert to bytes
+ jz done_hash
+ lea -64(INP, NUM_BLKS), NUM_BLKS # pointer to last block
+ mov NUM_BLKS, _INP_END(%rsp)
+
+ cmp NUM_BLKS, INP
+ je only_one_block
+
+ ## load initial digest
+ mov (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
+
+ mov CTX, _CTX(%rsp)
+
+loop0:
+ lea K256(%rip), TBL
+
+ ## Load first 16 dwords from two blocks
+ VMOVDQ 0*32(INP),XTMP0
+ VMOVDQ 1*32(INP),XTMP1
+ VMOVDQ 2*32(INP),XTMP2
+ VMOVDQ 3*32(INP),XTMP3
+
+ ## byte swap data
+ vpshufb BYTE_FLIP_MASK, XTMP0, XTMP0
+ vpshufb BYTE_FLIP_MASK, XTMP1, XTMP1
+ vpshufb BYTE_FLIP_MASK, XTMP2, XTMP2
+ vpshufb BYTE_FLIP_MASK, XTMP3, XTMP3
+
+ ## transpose data into high/low halves
+ vperm2i128 $0x20, XTMP2, XTMP0, X0
+ vperm2i128 $0x31, XTMP2, XTMP0, X1
+ vperm2i128 $0x20, XTMP3, XTMP1, X2
+ vperm2i128 $0x31, XTMP3, XTMP1, X3
+
+last_block_enter:
+ add $64, INP
+ mov INP, _INP(%rsp)
+
+ ## schedule 48 input dwords, by doing 3 rounds of 12 each
+ xor SRND, SRND
+
+.align 16
+loop1:
+ vpaddd 0*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 0*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 0*32
+
+ vpaddd 1*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 1*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 1*32
+
+ vpaddd 2*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 2*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 2*32
+
+ vpaddd 3*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 3*32+_XFER(%rsp, SRND)
+ FOUR_ROUNDS_AND_SCHED _XFER + 3*32
+
+ add $4*32, SRND
+ cmp $3*4*32, SRND
+ jb loop1
+
+loop2:
+ ## Do last 16 rounds with no scheduling
+ vpaddd 0*32(TBL, SRND), X0, XFER
+ vmovdqa XFER, 0*32+_XFER(%rsp, SRND)
+ DO_4ROUNDS _XFER + 0*32
+ vpaddd 1*32(TBL, SRND), X1, XFER
+ vmovdqa XFER, 1*32+_XFER(%rsp, SRND)
+ DO_4ROUNDS _XFER + 1*32
+ add $2*32, SRND
+
+ vmovdqa X2, X0
+ vmovdqa X3, X1
+
+ cmp $4*4*32, SRND
+ jb loop2
+
+ mov _CTX(%rsp), CTX
+ mov _INP(%rsp), INP
+
+ 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
+
+ cmp _INP_END(%rsp), INP
+ ja done_hash
+
+ #### Do second block using previously scheduled results
+ xor SRND, SRND
+.align 16
+loop3:
+ DO_4ROUNDS _XFER + 0*32 + 16
+ DO_4ROUNDS _XFER + 1*32 + 16
+ add $2*32, SRND
+ cmp $4*4*32, SRND
+ jb loop3
+
+ mov _CTX(%rsp), CTX
+ mov _INP(%rsp), INP
+ add $64, INP
+
+ 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
+
+ cmp _INP_END(%rsp), INP
+ jb loop0
+ ja done_hash
+
+do_last_block:
+ #### do last block
+ lea K256(%rip), TBL
+
+ VMOVDQ 0*16(INP),XWORD0
+ VMOVDQ 1*16(INP),XWORD1
+ VMOVDQ 2*16(INP),XWORD2
+ VMOVDQ 3*16(INP),XWORD3
+
+ vpshufb X_BYTE_FLIP_MASK, XWORD0, XWORD0
+ vpshufb X_BYTE_FLIP_MASK, XWORD1, XWORD1
+ vpshufb X_BYTE_FLIP_MASK, XWORD2, XWORD2
+ vpshufb X_BYTE_FLIP_MASK, XWORD3, XWORD3
+
+ jmp last_block_enter
+
+only_one_block:
+
+ ## 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
+
+ mov CTX, _CTX(%rsp)
+ jmp do_last_block
+
+done_hash:
+
+ mov _RSP(%rsp), %rsp
+
+ popq %r15
+ popq %r14
+ popq %r13
+ popq %r12
+ popq %rbp
+ popq %rbx
+ ret
+ENDPROC(sha256_transform_rorx)
+
+.data
+.align 64
+K256:
+ .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+ .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
+ .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+ .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
+ .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+ .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
+ .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+ .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
+ .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+ .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
+ .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+ .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
+ .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+ .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
+ .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+ .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
+ .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+ .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
+ .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+ .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
+ .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+ .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
+ .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+ .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
+ .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+ .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
+ .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+ .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
+ .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+ .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
+ .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+ .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
+
+PSHUFFLE_BYTE_FLIP_MASK:
+ .octa 0x0c0d0e0f08090a0b0405060700010203,0x0c0d0e0f08090a0b0405060700010203
+
+# shuffle xBxA -> 00BA
+_SHUF_00BA:
+ .octa 0xFFFFFFFFFFFFFFFF0b0a090803020100,0xFFFFFFFFFFFFFFFF0b0a090803020100
+
+# shuffle xDxC -> DC00
+_SHUF_DC00:
+ .octa 0x0b0a090803020100FFFFFFFFFFFFFFFF,0x0b0a090803020100FFFFFFFFFFFFFFFF
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 05/10] Create module providing optimized SHA256 routines using SSSE3, AVX or AVX2 instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
We added glue code and config options to create crypto
module that uses SSE/AVX/AVX2 optimized SHA256 x86_64 assembly routines.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha256_ssse3_glue.c | 275 ++++++++++++++++++++++++++++++++++++
crypto/Kconfig | 11 ++
3 files changed, 288 insertions(+)
create mode 100644 arch/x86/crypto/sha256_ssse3_glue.c
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index ca96761..b45f599 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -28,6 +28,7 @@ obj-$(CONFIG_CRYPTO_GHASH_CLMUL_NI_INTEL) += ghash-clmulni-intel.o
obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
obj-$(CONFIG_CRYPTO_CRC32_PCLMUL) += crc32-pclmul.o
+obj-$(CONFIG_CRYPTO_SHA256_SSSE3) += sha256-ssse3.o
aes-i586-y := aes-i586-asm_32.o aes_glue.o
twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
@@ -54,3 +55,4 @@ sha1-ssse3-y := sha1_ssse3_asm.o sha1_ssse3_glue.o
crc32c-intel-y := crc32c-intel_glue.o
crc32c-intel-$(CONFIG_64BIT) += crc32c-pcl-intel-asm_64.o
crc32-pclmul-y := crc32-pclmul_asm.o crc32-pclmul_glue.o
+sha256-ssse3-y := sha256-ssse3-asm.o sha256-avx-asm.o sha256-avx2-asm.o sha256_ssse3_glue.o
diff --git a/arch/x86/crypto/sha256_ssse3_glue.c b/arch/x86/crypto/sha256_ssse3_glue.c
new file mode 100644
index 0000000..fa65453
--- /dev/null
+++ b/arch/x86/crypto/sha256_ssse3_glue.c
@@ -0,0 +1,275 @@
+/*
+ * Cryptographic API.
+ *
+ * Glue code for the SHA256 Secure Hash Algorithm assembler
+ * implementation using supplemental SSE3 / AVX / AVX2 instructions.
+ *
+ * This file is based on sha256_generic.c
+ *
+ * Copyright (C) 2013 Intel Corporation.
+ *
+ * Author:
+ * Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * 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.
+ */
+
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <crypto/sha.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+#include <linux/string.h>
+
+asmlinkage void sha256_transform_ssse3(const char *data, u32 *digest,
+ u64 rounds);
+#ifdef CONFIG_AS_AVX
+asmlinkage void sha256_transform_avx(const char *data, u32 *digest,
+ u64 rounds);
+#endif
+#ifdef CONFIG_AS_AVX2
+asmlinkage void sha256_transform_rorx(const char *data, u32 *digest,
+ u64 rounds);
+#endif
+
+asmlinkage void (*sha256_transform_asm)(const char *, u32 *, u64);
+
+
+static int sha256_ssse3_init(struct shash_desc *desc)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ sctx->state[0] = SHA256_H0;
+ sctx->state[1] = SHA256_H1;
+ sctx->state[2] = SHA256_H2;
+ sctx->state[3] = SHA256_H3;
+ sctx->state[4] = SHA256_H4;
+ sctx->state[5] = SHA256_H5;
+ sctx->state[6] = SHA256_H6;
+ sctx->state[7] = SHA256_H7;
+ sctx->count = 0;
+
+ return 0;
+}
+
+static int __sha256_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len, unsigned int partial)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int done = 0;
+
+ sctx->count += len;
+
+ if (partial) {
+ done = SHA256_BLOCK_SIZE - partial;
+ memcpy(sctx->buf + partial, data, done);
+ sha256_transform_asm(sctx->buf, sctx->state, 1);
+ }
+
+ if (len - done >= SHA256_BLOCK_SIZE) {
+ const unsigned int rounds = (len - done) / SHA256_BLOCK_SIZE;
+
+ sha256_transform_asm(data + done, sctx->state, (u64) rounds);
+
+ done += rounds * SHA256_BLOCK_SIZE;
+ }
+
+ memcpy(sctx->buf, data + done, len - done);
+
+ return 0;
+}
+
+static int sha256_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int partial = sctx->count % SHA256_BLOCK_SIZE;
+ int res;
+
+ /* Handle the fast case right here */
+ if (partial + len < SHA256_BLOCK_SIZE) {
+ sctx->count += len;
+ memcpy(sctx->buf + partial, data, len);
+
+ return 0;
+ }
+
+ if (!irq_fpu_usable()) {
+ res = crypto_sha256_update(desc, data, len);
+ } else {
+ kernel_fpu_begin();
+ res = __sha256_ssse3_update(desc, data, len, partial);
+ kernel_fpu_end();
+ }
+
+ return res;
+}
+
+
+/* Add padding and return the message digest. */
+static int sha256_ssse3_final(struct shash_desc *desc, u8 *out)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int i, index, padlen;
+ __be32 *dst = (__be32 *)out;
+ __be64 bits;
+ static const u8 padding[SHA256_BLOCK_SIZE] = { 0x80, };
+
+ bits = cpu_to_be64(sctx->count << 3);
+
+ /* Pad out to 56 mod 64 and append length */
+ index = sctx->count % SHA256_BLOCK_SIZE;
+ padlen = (index < 56) ? (56 - index) : ((SHA256_BLOCK_SIZE+56)-index);
+
+ if (!irq_fpu_usable()) {
+ crypto_sha256_update(desc, padding, padlen);
+ crypto_sha256_update(desc, (const u8 *)&bits, sizeof(bits));
+ } else {
+ kernel_fpu_begin();
+ /* We need to fill a whole block for __sha256_ssse3_update() */
+ if (padlen <= 56) {
+ sctx->count += padlen;
+ memcpy(sctx->buf + index, padding, padlen);
+ } else {
+ __sha256_ssse3_update(desc, padding, padlen, index);
+ }
+ __sha256_ssse3_update(desc, (const u8 *)&bits,
+ sizeof(bits), 56);
+ kernel_fpu_end();
+ }
+
+ /* Store state in digest */
+ for (i = 0; i < 8; i++)
+ dst[i] = cpu_to_be32(sctx->state[i]);
+
+ /* Wipe context */
+ memset(sctx, 0, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha256_ssse3_export(struct shash_desc *desc, void *out)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(out, sctx, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha256_ssse3_import(struct shash_desc *desc, const void *in)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(sctx, in, sizeof(*sctx));
+
+ return 0;
+}
+
+static struct shash_alg alg = {
+ .digestsize = SHA256_DIGEST_SIZE,
+ .init = sha256_ssse3_init,
+ .update = sha256_ssse3_update,
+ .final = sha256_ssse3_final,
+ .export = sha256_ssse3_export,
+ .import = sha256_ssse3_import,
+ .descsize = sizeof(struct sha256_state),
+ .statesize = sizeof(struct sha256_state),
+ .base = {
+ .cra_name = "sha256",
+ .cra_driver_name = "sha256-ssse3",
+ .cra_priority = 150,
+ .cra_flags = CRYPTO_ALG_TYPE_SHASH,
+ .cra_blocksize = SHA256_BLOCK_SIZE,
+ .cra_module = THIS_MODULE,
+ }
+};
+
+#ifdef CONFIG_AS_AVX
+static bool __init avx_usable(void)
+{
+ u64 xcr0;
+
+ if (!cpu_has_avx || !cpu_has_osxsave)
+ return false;
+
+ xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
+ if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
+ pr_info("AVX detected but unusable.\n");
+
+ return false;
+ }
+
+ return true;
+}
+#endif
+
+static int __init sha256_ssse3_mod_init(void)
+{
+ /* test for SSE3 first */
+ if (cpu_has_ssse3)
+ sha256_transform_asm = sha256_transform_ssse3;
+
+#ifdef CONFIG_AS_AVX
+ /* allow AVX to override SSSE3, it's a little faster */
+ if (avx_usable()) {
+#ifdef CONFIG_AS_AVX2
+ if (boot_cpu_has(X86_FEATURE_AVX2))
+ sha256_transform_asm = sha256_transform_rorx;
+ else
+#endif
+ sha256_transform_asm = sha256_transform_avx;
+ }
+#endif
+
+ if (sha256_transform_asm) {
+#ifdef CONFIG_AS_AVX
+ if (sha256_transform_asm == sha256_transform_avx)
+ pr_info("Using AVX optimized SHA-256 implementation\n");
+#ifdef CONFIG_AS_AVX2
+ else if (sha256_transform_asm == sha256_transform_rorx)
+ pr_info("Using AVX2 optimized SHA-256 implementation\n");
+#endif
+ else
+#endif
+ pr_info("Using SSSE3 optimized SHA-256 implementation\n");
+ return crypto_register_shash(&alg);
+ }
+ pr_info("Neither AVX nor SSSE3 is available/usable.\n");
+
+ return -ENODEV;
+}
+
+static void __exit sha256_ssse3_mod_fini(void)
+{
+ crypto_unregister_shash(&alg);
+}
+
+module_init(sha256_ssse3_mod_init);
+module_exit(sha256_ssse3_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA256 Secure Hash Algorithm, Supplemental SSE3 accelerated");
+
+MODULE_ALIAS("sha256");
diff --git a/crypto/Kconfig b/crypto/Kconfig
index aed52b2..8064ef1 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -470,6 +470,17 @@ config CRYPTO_SHA1_SSSE3
using Supplemental SSE3 (SSSE3) instructions or Advanced Vector
Extensions (AVX), when available.
+config CRYPTO_SHA256_SSSE3
+ tristate "SHA256 digest algorithm (SSSE3/AVX/AVX2)"
+ depends on X86 && 64BIT
+ select CRYPTO_SHA256
+ select CRYPTO_HASH
+ help
+ SHA-256 secure hash standard (DFIPS 180-2) implemented
+ using Supplemental SSE3 (SSSE3) instructions, or Advanced Vector
+ Extensions version 1 (AVX1), or Advanced Vector Extensions
+ version 2 (AVX2) instructions, when available.
+
config CRYPTO_SHA1_SPARC64
tristate "SHA1 digest algorithm (SPARC64)"
depends on SPARC64
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 05/10] Create module providing optimized SHA256 routines using SSSE3, AVX or AVX2 instructions.
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
We added glue code and config options to create crypto
module that uses SSE/AVX/AVX2 optimized SHA256 x86_64 assembly routines.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha256_ssse3_glue.c | 275 ++++++++++++++++++++++++++++++++++++
crypto/Kconfig | 11 ++
3 files changed, 288 insertions(+)
create mode 100644 arch/x86/crypto/sha256_ssse3_glue.c
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index ca96761..b45f599 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -28,6 +28,7 @@ obj-$(CONFIG_CRYPTO_GHASH_CLMUL_NI_INTEL) += ghash-clmulni-intel.o
obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
obj-$(CONFIG_CRYPTO_CRC32_PCLMUL) += crc32-pclmul.o
+obj-$(CONFIG_CRYPTO_SHA256_SSSE3) += sha256-ssse3.o
aes-i586-y := aes-i586-asm_32.o aes_glue.o
twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
@@ -54,3 +55,4 @@ sha1-ssse3-y := sha1_ssse3_asm.o sha1_ssse3_glue.o
crc32c-intel-y := crc32c-intel_glue.o
crc32c-intel-$(CONFIG_64BIT) += crc32c-pcl-intel-asm_64.o
crc32-pclmul-y := crc32-pclmul_asm.o crc32-pclmul_glue.o
+sha256-ssse3-y := sha256-ssse3-asm.o sha256-avx-asm.o sha256-avx2-asm.o sha256_ssse3_glue.o
diff --git a/arch/x86/crypto/sha256_ssse3_glue.c b/arch/x86/crypto/sha256_ssse3_glue.c
new file mode 100644
index 0000000..fa65453
--- /dev/null
+++ b/arch/x86/crypto/sha256_ssse3_glue.c
@@ -0,0 +1,275 @@
+/*
+ * Cryptographic API.
+ *
+ * Glue code for the SHA256 Secure Hash Algorithm assembler
+ * implementation using supplemental SSE3 / AVX / AVX2 instructions.
+ *
+ * This file is based on sha256_generic.c
+ *
+ * Copyright (C) 2013 Intel Corporation.
+ *
+ * Author:
+ * Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * 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.
+ */
+
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <crypto/sha.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+#include <linux/string.h>
+
+asmlinkage void sha256_transform_ssse3(const char *data, u32 *digest,
+ u64 rounds);
+#ifdef CONFIG_AS_AVX
+asmlinkage void sha256_transform_avx(const char *data, u32 *digest,
+ u64 rounds);
+#endif
+#ifdef CONFIG_AS_AVX2
+asmlinkage void sha256_transform_rorx(const char *data, u32 *digest,
+ u64 rounds);
+#endif
+
+asmlinkage void (*sha256_transform_asm)(const char *, u32 *, u64);
+
+
+static int sha256_ssse3_init(struct shash_desc *desc)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ sctx->state[0] = SHA256_H0;
+ sctx->state[1] = SHA256_H1;
+ sctx->state[2] = SHA256_H2;
+ sctx->state[3] = SHA256_H3;
+ sctx->state[4] = SHA256_H4;
+ sctx->state[5] = SHA256_H5;
+ sctx->state[6] = SHA256_H6;
+ sctx->state[7] = SHA256_H7;
+ sctx->count = 0;
+
+ return 0;
+}
+
+static int __sha256_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len, unsigned int partial)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int done = 0;
+
+ sctx->count += len;
+
+ if (partial) {
+ done = SHA256_BLOCK_SIZE - partial;
+ memcpy(sctx->buf + partial, data, done);
+ sha256_transform_asm(sctx->buf, sctx->state, 1);
+ }
+
+ if (len - done >= SHA256_BLOCK_SIZE) {
+ const unsigned int rounds = (len - done) / SHA256_BLOCK_SIZE;
+
+ sha256_transform_asm(data + done, sctx->state, (u64) rounds);
+
+ done += rounds * SHA256_BLOCK_SIZE;
+ }
+
+ memcpy(sctx->buf, data + done, len - done);
+
+ return 0;
+}
+
+static int sha256_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int partial = sctx->count % SHA256_BLOCK_SIZE;
+ int res;
+
+ /* Handle the fast case right here */
+ if (partial + len < SHA256_BLOCK_SIZE) {
+ sctx->count += len;
+ memcpy(sctx->buf + partial, data, len);
+
+ return 0;
+ }
+
+ if (!irq_fpu_usable()) {
+ res = crypto_sha256_update(desc, data, len);
+ } else {
+ kernel_fpu_begin();
+ res = __sha256_ssse3_update(desc, data, len, partial);
+ kernel_fpu_end();
+ }
+
+ return res;
+}
+
+
+/* Add padding and return the message digest. */
+static int sha256_ssse3_final(struct shash_desc *desc, u8 *out)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+ unsigned int i, index, padlen;
+ __be32 *dst = (__be32 *)out;
+ __be64 bits;
+ static const u8 padding[SHA256_BLOCK_SIZE] = { 0x80, };
+
+ bits = cpu_to_be64(sctx->count << 3);
+
+ /* Pad out to 56 mod 64 and append length */
+ index = sctx->count % SHA256_BLOCK_SIZE;
+ padlen = (index < 56) ? (56 - index) : ((SHA256_BLOCK_SIZE+56)-index);
+
+ if (!irq_fpu_usable()) {
+ crypto_sha256_update(desc, padding, padlen);
+ crypto_sha256_update(desc, (const u8 *)&bits, sizeof(bits));
+ } else {
+ kernel_fpu_begin();
+ /* We need to fill a whole block for __sha256_ssse3_update() */
+ if (padlen <= 56) {
+ sctx->count += padlen;
+ memcpy(sctx->buf + index, padding, padlen);
+ } else {
+ __sha256_ssse3_update(desc, padding, padlen, index);
+ }
+ __sha256_ssse3_update(desc, (const u8 *)&bits,
+ sizeof(bits), 56);
+ kernel_fpu_end();
+ }
+
+ /* Store state in digest */
+ for (i = 0; i < 8; i++)
+ dst[i] = cpu_to_be32(sctx->state[i]);
+
+ /* Wipe context */
+ memset(sctx, 0, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha256_ssse3_export(struct shash_desc *desc, void *out)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(out, sctx, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha256_ssse3_import(struct shash_desc *desc, const void *in)
+{
+ struct sha256_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(sctx, in, sizeof(*sctx));
+
+ return 0;
+}
+
+static struct shash_alg alg = {
+ .digestsize = SHA256_DIGEST_SIZE,
+ .init = sha256_ssse3_init,
+ .update = sha256_ssse3_update,
+ .final = sha256_ssse3_final,
+ .export = sha256_ssse3_export,
+ .import = sha256_ssse3_import,
+ .descsize = sizeof(struct sha256_state),
+ .statesize = sizeof(struct sha256_state),
+ .base = {
+ .cra_name = "sha256",
+ .cra_driver_name = "sha256-ssse3",
+ .cra_priority = 150,
+ .cra_flags = CRYPTO_ALG_TYPE_SHASH,
+ .cra_blocksize = SHA256_BLOCK_SIZE,
+ .cra_module = THIS_MODULE,
+ }
+};
+
+#ifdef CONFIG_AS_AVX
+static bool __init avx_usable(void)
+{
+ u64 xcr0;
+
+ if (!cpu_has_avx || !cpu_has_osxsave)
+ return false;
+
+ xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
+ if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
+ pr_info("AVX detected but unusable.\n");
+
+ return false;
+ }
+
+ return true;
+}
+#endif
+
+static int __init sha256_ssse3_mod_init(void)
+{
+ /* test for SSE3 first */
+ if (cpu_has_ssse3)
+ sha256_transform_asm = sha256_transform_ssse3;
+
+#ifdef CONFIG_AS_AVX
+ /* allow AVX to override SSSE3, it's a little faster */
+ if (avx_usable()) {
+#ifdef CONFIG_AS_AVX2
+ if (boot_cpu_has(X86_FEATURE_AVX2))
+ sha256_transform_asm = sha256_transform_rorx;
+ else
+#endif
+ sha256_transform_asm = sha256_transform_avx;
+ }
+#endif
+
+ if (sha256_transform_asm) {
+#ifdef CONFIG_AS_AVX
+ if (sha256_transform_asm == sha256_transform_avx)
+ pr_info("Using AVX optimized SHA-256 implementation\n");
+#ifdef CONFIG_AS_AVX2
+ else if (sha256_transform_asm == sha256_transform_rorx)
+ pr_info("Using AVX2 optimized SHA-256 implementation\n");
+#endif
+ else
+#endif
+ pr_info("Using SSSE3 optimized SHA-256 implementation\n");
+ return crypto_register_shash(&alg);
+ }
+ pr_info("Neither AVX nor SSSE3 is available/usable.\n");
+
+ return -ENODEV;
+}
+
+static void __exit sha256_ssse3_mod_fini(void)
+{
+ crypto_unregister_shash(&alg);
+}
+
+module_init(sha256_ssse3_mod_init);
+module_exit(sha256_ssse3_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA256 Secure Hash Algorithm, Supplemental SSE3 accelerated");
+
+MODULE_ALIAS("sha256");
diff --git a/crypto/Kconfig b/crypto/Kconfig
index aed52b2..8064ef1 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -470,6 +470,17 @@ config CRYPTO_SHA1_SSSE3
using Supplemental SSE3 (SSSE3) instructions or Advanced Vector
Extensions (AVX), when available.
+config CRYPTO_SHA256_SSSE3
+ tristate "SHA256 digest algorithm (SSSE3/AVX/AVX2)"
+ depends on X86 && 64BIT
+ select CRYPTO_SHA256
+ select CRYPTO_HASH
+ help
+ SHA-256 secure hash standard (DFIPS 180-2) implemented
+ using Supplemental SSE3 (SSSE3) instructions, or Advanced Vector
+ Extensions version 1 (AVX1), or Advanced Vector Extensions
+ version 2 (AVX2) instructions, when available.
+
config CRYPTO_SHA1_SPARC64
tristate "SHA1 digest algorithm (SPARC64)"
depends on SPARC64
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 06/10] Expose generic sha512 routine to be callable from other modules
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Other SHA512 routines may need to use the generic routine when
FPU is not available.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
crypto/sha512_generic.c | 13 +++++++------
include/crypto/sha.h | 3 +++
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/crypto/sha512_generic.c b/crypto/sha512_generic.c
index 71fcf36..4c58620 100644
--- a/crypto/sha512_generic.c
+++ b/crypto/sha512_generic.c
@@ -163,8 +163,8 @@ sha384_init(struct shash_desc *desc)
return 0;
}
-static int
-sha512_update(struct shash_desc *desc, const u8 *data, unsigned int len)
+int crypto_sha512_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
{
struct sha512_state *sctx = shash_desc_ctx(desc);
@@ -197,6 +197,7 @@ sha512_update(struct shash_desc *desc, const u8 *data, unsigned int len)
return 0;
}
+EXPORT_SYMBOL(crypto_sha512_update);
static int
sha512_final(struct shash_desc *desc, u8 *hash)
@@ -215,10 +216,10 @@ sha512_final(struct shash_desc *desc, u8 *hash)
/* Pad out to 112 mod 128. */
index = sctx->count[0] & 0x7f;
pad_len = (index < 112) ? (112 - index) : ((128+112) - index);
- sha512_update(desc, padding, pad_len);
+ crypto_sha512_update(desc, padding, pad_len);
/* Append length (before padding) */
- sha512_update(desc, (const u8 *)bits, sizeof(bits));
+ crypto_sha512_update(desc, (const u8 *)bits, sizeof(bits));
/* Store state in digest */
for (i = 0; i < 8; i++)
@@ -245,7 +246,7 @@ static int sha384_final(struct shash_desc *desc, u8 *hash)
static struct shash_alg sha512_algs[2] = { {
.digestsize = SHA512_DIGEST_SIZE,
.init = sha512_init,
- .update = sha512_update,
+ .update = crypto_sha512_update,
.final = sha512_final,
.descsize = sizeof(struct sha512_state),
.base = {
@@ -257,7 +258,7 @@ static struct shash_alg sha512_algs[2] = { {
}, {
.digestsize = SHA384_DIGEST_SIZE,
.init = sha384_init,
- .update = sha512_update,
+ .update = crypto_sha512_update,
.final = sha384_final,
.descsize = sizeof(struct sha512_state),
.base = {
diff --git a/include/crypto/sha.h b/include/crypto/sha.h
index f46ff61..190f8a0 100644
--- a/include/crypto/sha.h
+++ b/include/crypto/sha.h
@@ -89,4 +89,7 @@ extern int crypto_sha1_update(struct shash_desc *desc, const u8 *data,
extern int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
unsigned int len);
+
+extern int crypto_sha512_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len);
#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 06/10] Expose generic sha512 routine to be callable from other modules
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Other SHA512 routines may need to use the generic routine when
FPU is not available.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
crypto/sha512_generic.c | 13 +++++++------
include/crypto/sha.h | 3 +++
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/crypto/sha512_generic.c b/crypto/sha512_generic.c
index 71fcf36..4c58620 100644
--- a/crypto/sha512_generic.c
+++ b/crypto/sha512_generic.c
@@ -163,8 +163,8 @@ sha384_init(struct shash_desc *desc)
return 0;
}
-static int
-sha512_update(struct shash_desc *desc, const u8 *data, unsigned int len)
+int crypto_sha512_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
{
struct sha512_state *sctx = shash_desc_ctx(desc);
@@ -197,6 +197,7 @@ sha512_update(struct shash_desc *desc, const u8 *data, unsigned int len)
return 0;
}
+EXPORT_SYMBOL(crypto_sha512_update);
static int
sha512_final(struct shash_desc *desc, u8 *hash)
@@ -215,10 +216,10 @@ sha512_final(struct shash_desc *desc, u8 *hash)
/* Pad out to 112 mod 128. */
index = sctx->count[0] & 0x7f;
pad_len = (index < 112) ? (112 - index) : ((128+112) - index);
- sha512_update(desc, padding, pad_len);
+ crypto_sha512_update(desc, padding, pad_len);
/* Append length (before padding) */
- sha512_update(desc, (const u8 *)bits, sizeof(bits));
+ crypto_sha512_update(desc, (const u8 *)bits, sizeof(bits));
/* Store state in digest */
for (i = 0; i < 8; i++)
@@ -245,7 +246,7 @@ static int sha384_final(struct shash_desc *desc, u8 *hash)
static struct shash_alg sha512_algs[2] = { {
.digestsize = SHA512_DIGEST_SIZE,
.init = sha512_init,
- .update = sha512_update,
+ .update = crypto_sha512_update,
.final = sha512_final,
.descsize = sizeof(struct sha512_state),
.base = {
@@ -257,7 +258,7 @@ static struct shash_alg sha512_algs[2] = { {
}, {
.digestsize = SHA384_DIGEST_SIZE,
.init = sha384_init,
- .update = sha512_update,
+ .update = crypto_sha512_update,
.final = sha384_final,
.descsize = sizeof(struct sha512_state),
.base = {
diff --git a/include/crypto/sha.h b/include/crypto/sha.h
index f46ff61..190f8a0 100644
--- a/include/crypto/sha.h
+++ b/include/crypto/sha.h
@@ -89,4 +89,7 @@ extern int crypto_sha1_update(struct shash_desc *desc, const u8 *data,
extern int crypto_sha256_update(struct shash_desc *desc, const u8 *data,
unsigned int len);
+
+extern int crypto_sha512_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len);
#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 07/10] Optimized SHA512 x86_64 assembly routine using Supplemental SSE3 instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 x86_64 assembly routine optimized with SSSE3 instructions.
Speedup of 40% or more has been measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha512-ssse3-asm.S | 421 +++++++++++++++++++++++++++++++++++++
1 file changed, 421 insertions(+)
create mode 100644 arch/x86/crypto/sha512-ssse3-asm.S
diff --git a/arch/x86/crypto/sha512-ssse3-asm.S b/arch/x86/crypto/sha512-ssse3-asm.S
new file mode 100644
index 0000000..fb56855
--- /dev/null
+++ b/arch/x86/crypto/sha512-ssse3-asm.S
@@ -0,0 +1,421 @@
+########################################################################
+# Implement fast SHA-512 with SSSE3 instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 Implementations on Intel Architecture Processors"
+#
+# To find it, surf to http://www.intel.com/p/en_US/embedded
+# and search for that title.
+#
+########################################################################
+
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+# ARG1
+msg = %rdi
+# ARG2
+digest = %rsi
+# ARG3
+msglen = %rdx
+T1 = %rcx
+T2 = %r8
+a_64 = %r9
+b_64 = %r10
+c_64 = %r11
+d_64 = %r12
+e_64 = %r13
+f_64 = %r14
+g_64 = %r15
+h_64 = %rbx
+tmp0 = %rax
+
+# Local variables (stack frame)
+
+W_SIZE = 80*8
+WK_SIZE = 2*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 5*8
+
+frame_W = 0
+frame_WK = frame_W + W_SIZE
+frame_RSPSAVE = frame_WK + WK_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+# Useful QWORD "arrays" for simpler memory references
+# MSG, DIGEST, K_t, W_t are arrays
+# WK_2(t) points to 1 of 2 qwords at frame.WK depdending on t being odd/even
+
+# Input message (arg1)
+#define MSG(i) 8*i(msg)
+
+# Output Digest (arg2)
+#define DIGEST(i) 8*i(digest)
+
+# SHA Constants (static mem)
+#define K_t(i) 8*i+K512(%rip)
+
+# Message Schedule (stack frame)
+#define W_t(i) 8*i+frame_W(%rsp)
+
+# W[t]+K[t] (stack frame)
+#define WK_2(i) 8*((i%2))+frame_WK(%rsp)
+
+.macro RotateState
+ # Rotate symbols a..h right
+ TMP = h_64
+ h_64 = g_64
+ g_64 = f_64
+ f_64 = e_64
+ e_64 = d_64
+ d_64 = c_64
+ c_64 = b_64
+ b_64 = a_64
+ a_64 = TMP
+.endm
+
+.macro SHA512_Round rnd
+
+ # Compute Round %%t
+ mov f_64, T1 # T1 = f
+ mov e_64, tmp0 # tmp = e
+ xor g_64, T1 # T1 = f ^ g
+ ror $23, tmp0 # 41 # tmp = e ror 23
+ and e_64, T1 # T1 = (f ^ g) & e
+ xor e_64, tmp0 # tmp = (e ror 23) ^ e
+ xor g_64, T1 # T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
+ idx = \rnd
+ add WK_2(idx), T1 # W[t] + K[t] from message scheduler
+ ror $4, tmp0 # 18 # tmp = ((e ror 23) ^ e) ror 4
+ xor e_64, tmp0 # tmp = (((e ror 23) ^ e) ror 4) ^ e
+ mov a_64, T2 # T2 = a
+ add h_64, T1 # T1 = CH(e,f,g) + W[t] + K[t] + h
+ ror $14, tmp0 # 14 # tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
+ add tmp0, T1 # T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
+ mov a_64, tmp0 # tmp = a
+ xor c_64, T2 # T2 = a ^ c
+ and c_64, tmp0 # tmp = a & c
+ and b_64, T2 # T2 = (a ^ c) & b
+ xor tmp0, T2 # T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
+ mov a_64, tmp0 # tmp = a
+ ror $5, tmp0 # 39 # tmp = a ror 5
+ xor a_64, tmp0 # tmp = (a ror 5) ^ a
+ add T1, d_64 # e(next_state) = d + T1
+ ror $6, tmp0 # 34 # tmp = ((a ror 5) ^ a) ror 6
+ xor a_64, tmp0 # tmp = (((a ror 5) ^ a) ror 6) ^ a
+ lea (T1, T2), h_64 # a(next_state) = T1 + Maj(a,b,c)
+ ror $28, tmp0 # 28 # tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
+ add tmp0, h_64 # a(next_state) = T1 + Maj(a,b,c) S0(a)
+ RotateState
+.endm
+
+.macro SHA512_2Sched_2Round_sse rnd
+
+ # Compute rounds t-2 and t-1
+ # Compute message schedule QWORDS t and t+1
+
+ # Two rounds are computed based on the values for K[t-2]+W[t-2] and
+ # K[t-1]+W[t-1] which were previously stored at WK_2 by the message
+ # scheduler.
+ # The two new schedule QWORDS are stored at [W_t(%%t)] and [W_t(%%t+1)].
+ # They are then added to their respective SHA512 constants at
+ # [K_t(%%t)] and [K_t(%%t+1)] and stored at dqword [WK_2(%%t)]
+ # For brievity, the comments following vectored instructions only refer to
+ # the first of a pair of QWORDS.
+ # Eg. XMM2=W[t-2] really means XMM2={W[t-2]|W[t-1]}
+ # The computation of the message schedule and the rounds are tightly
+ # stitched to take advantage of instruction-level parallelism.
+ # For clarity, integer instructions (for the rounds calculation) are indented
+ # by one tab. Vectored instructions (for the message scheduler) are indented
+ # by two tabs.
+
+ mov f_64, T1
+ idx = \rnd -2
+ movdqa W_t(idx), %xmm2 # XMM2 = W[t-2]
+ xor g_64, T1
+ and e_64, T1
+ movdqa %xmm2, %xmm0 # XMM0 = W[t-2]
+ xor g_64, T1
+ idx = \rnd
+ add WK_2(idx), T1
+ idx = \rnd - 15
+ movdqu W_t(idx), %xmm5 # XMM5 = W[t-15]
+ mov e_64, tmp0
+ ror $23, tmp0 # 41
+ movdqa %xmm5, %xmm3 # XMM3 = W[t-15]
+ xor e_64, tmp0
+ ror $4, tmp0 # 18
+ psrlq $61-19, %xmm0 # XMM0 = W[t-2] >> 42
+ xor e_64, tmp0
+ ror $14, tmp0 # 14
+ psrlq $(8-7), %xmm3 # XMM3 = W[t-15] >> 1
+ add tmp0, T1
+ add h_64, T1
+ pxor %xmm2, %xmm0 # XMM0 = (W[t-2] >> 42) ^ W[t-2]
+ mov a_64, T2
+ xor c_64, T2
+ pxor %xmm5, %xmm3 # XMM3 = (W[t-15] >> 1) ^ W[t-15]
+ and b_64, T2
+ mov a_64, tmp0
+ psrlq $(19-6), %xmm0 # XMM0 = ((W[t-2]>>42)^W[t-2])>>13
+ and c_64, tmp0
+ xor tmp0, T2
+ psrlq $(7-1), %xmm3 # XMM3 = ((W[t-15]>>1)^W[t-15])>>6
+ mov a_64, tmp0
+ ror $5, tmp0 # 39
+ pxor %xmm2, %xmm0 # XMM0 = (((W[t-2]>>42)^W[t-2])>>13)^W[t-2]
+ xor a_64, tmp0
+ ror $6, tmp0 # 34
+ pxor %xmm5, %xmm3 # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]
+ xor a_64, tmp0
+ ror $28, tmp0 # 28
+ psrlq $6, %xmm0 # XMM0 = ((((W[t-2]>>42)^W[t-2])>>13)^W[t-2])>>6
+ add tmp0, T2
+ add T1, d_64
+ psrlq $1, %xmm3 # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]>>1
+ lea (T1, T2), h_64
+ RotateState
+ movdqa %xmm2, %xmm1 # XMM1 = W[t-2]
+ mov f_64, T1
+ xor g_64, T1
+ movdqa %xmm5, %xmm4 # XMM4 = W[t-15]
+ and e_64, T1
+ xor g_64, T1
+ psllq $(64-19)-(64-61) , %xmm1 # XMM1 = W[t-2] << 42
+ idx = \rnd + 1
+ add WK_2(idx), T1
+ mov e_64, tmp0
+ psllq $(64-1)-(64-8), %xmm4 # XMM4 = W[t-15] << 7
+ ror $23, tmp0 # 41
+ xor e_64, tmp0
+ pxor %xmm2, %xmm1 # XMM1 = (W[t-2] << 42)^W[t-2]
+ ror $4, tmp0 # 18
+ xor e_64, tmp0
+ pxor %xmm5, %xmm4 # XMM4 = (W[t-15]<<7)^W[t-15]
+ ror $14, tmp0 # 14
+ add tmp0, T1
+ psllq $(64-61), %xmm1 # XMM1 = ((W[t-2] << 42)^W[t-2])<<3
+ add h_64, T1
+ mov a_64, T2
+ psllq $(64-8), %xmm4 # XMM4 = ((W[t-15]<<7)^W[t-15])<<56
+ xor c_64, T2
+ and b_64, T2
+ pxor %xmm1, %xmm0 # XMM0 = s1(W[t-2])
+ mov a_64, tmp0
+ and c_64, tmp0
+ idx = \rnd - 7
+ movdqu W_t(idx), %xmm1 # XMM1 = W[t-7]
+ xor tmp0, T2
+ pxor %xmm4, %xmm3 # XMM3 = s0(W[t-15])
+ mov a_64, tmp0
+ paddq %xmm3, %xmm0 # XMM0 = s1(W[t-2]) + s0(W[t-15])
+ ror $5, tmp0 # 39
+ idx =\rnd-16
+ paddq W_t(idx), %xmm0 # XMM0 = s1(W[t-2]) + s0(W[t-15]) + W[t-16]
+ xor a_64, tmp0
+ paddq %xmm1, %xmm0 # XMM0 = s1(W[t-2]) + W[t-7] + s0(W[t-15]) + W[t-16]
+ ror $6, tmp0 # 34
+ movdqa %xmm0, W_t(\rnd) # Store scheduled qwords
+ xor a_64, tmp0
+ paddq K_t(\rnd), %xmm0 # Compute W[t]+K[t]
+ ror $28, tmp0 # 28
+ idx = \rnd
+ movdqa %xmm0, WK_2(idx) # Store W[t]+K[t] for next rounds
+ add tmp0, T2
+ add T1, d_64
+ lea (T1, T2), h_64
+ RotateState
+.endm
+
+########################################################################
+# void sha512_transform_ssse3(const void* M, void* D, u64 L)#
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks.
+########################################################################
+ENTRY(sha512_transform_ssse3)
+
+ cmp $0, msglen
+ je nowork
+
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbx, frame_GPRSAVE(%rsp)
+ mov %r12, frame_GPRSAVE +8*1(%rsp)
+ mov %r13, frame_GPRSAVE +8*2(%rsp)
+ mov %r14, frame_GPRSAVE +8*3(%rsp)
+ mov %r15, frame_GPRSAVE +8*4(%rsp)
+
+updateblock:
+
+# Load state variables
+ mov DIGEST(0), a_64
+ mov DIGEST(1), b_64
+ mov DIGEST(2), c_64
+ mov DIGEST(3), d_64
+ mov DIGEST(4), e_64
+ mov DIGEST(5), f_64
+ mov DIGEST(6), g_64
+ mov DIGEST(7), h_64
+
+ t = 0
+ .rept 80/2 + 1
+ # (80 rounds) / (2 rounds/iteration) + (1 iteration)
+ # +1 iteration because the scheduler leads hashing by 1 iteration
+ .if t < 2
+ # BSWAP 2 QWORDS
+ movdqa XMM_QWORD_BSWAP(%rip), %xmm1
+ movdqu MSG(t), %xmm0
+ pshufb %xmm1, %xmm0 # BSWAP
+ movdqa %xmm0, W_t(t) # Store Scheduled Pair
+ paddq K_t(t), %xmm0 # Compute W[t]+K[t]
+ movdqa %xmm0, WK_2(t) # Store into WK for rounds
+ .elseif t < 16
+ # BSWAP 2 QWORDS# Compute 2 Rounds
+ movdqu MSG(t), %xmm0
+ pshufb %xmm1, %xmm0 # BSWAP
+ SHA512_Round t-2 # Round t-2
+ movdqa %xmm0, W_t(t) # Store Scheduled Pair
+ paddq K_t(t), %xmm0 # Compute W[t]+K[t]
+ SHA512_Round t-1 # Round t-1
+ movdqa %xmm0, WK_2(t) # Store W[t]+K[t] into WK
+ .elseif t < 79
+ # Schedule 2 QWORDS# Compute 2 Rounds
+ SHA512_2Sched_2Round_sse t
+ .else
+ # Compute 2 Rounds
+ SHA512_Round t-2
+ SHA512_Round t-1
+ .endif
+ t = t+2
+ .endr
+
+ # Update digest
+ add a_64, DIGEST(0)
+ add b_64, DIGEST(1)
+ add c_64, DIGEST(2)
+ add d_64, DIGEST(3)
+ add e_64, DIGEST(4)
+ add f_64, DIGEST(5)
+ add g_64, DIGEST(6)
+ add h_64, DIGEST(7)
+
+ # Advance to next message block
+ add $16*8, msg
+ dec msglen
+ jnz updateblock
+
+ # Restore GPRs
+ mov frame_GPRSAVE(%rsp), %rbx
+ mov frame_GPRSAVE +8*1(%rsp), %r12
+ mov frame_GPRSAVE +8*2(%rsp), %r13
+ mov frame_GPRSAVE +8*3(%rsp), %r14
+ mov frame_GPRSAVE +8*4(%rsp), %r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+
+nowork:
+ ret
+ENDPROC(sha512_transform_ssse3)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 16
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+XMM_QWORD_BSWAP:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 07/10] Optimized SHA512 x86_64 assembly routine using Supplemental SSE3 instructions.
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 x86_64 assembly routine optimized with SSSE3 instructions.
Speedup of 40% or more has been measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha512-ssse3-asm.S | 421 +++++++++++++++++++++++++++++++++++++
1 file changed, 421 insertions(+)
create mode 100644 arch/x86/crypto/sha512-ssse3-asm.S
diff --git a/arch/x86/crypto/sha512-ssse3-asm.S b/arch/x86/crypto/sha512-ssse3-asm.S
new file mode 100644
index 0000000..fb56855
--- /dev/null
+++ b/arch/x86/crypto/sha512-ssse3-asm.S
@@ -0,0 +1,421 @@
+########################################################################
+# Implement fast SHA-512 with SSSE3 instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 Implementations on Intel Architecture Processors"
+#
+# To find it, surf to http://www.intel.com/p/en_US/embedded
+# and search for that title.
+#
+########################################################################
+
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+# ARG1
+msg = %rdi
+# ARG2
+digest = %rsi
+# ARG3
+msglen = %rdx
+T1 = %rcx
+T2 = %r8
+a_64 = %r9
+b_64 = %r10
+c_64 = %r11
+d_64 = %r12
+e_64 = %r13
+f_64 = %r14
+g_64 = %r15
+h_64 = %rbx
+tmp0 = %rax
+
+# Local variables (stack frame)
+
+W_SIZE = 80*8
+WK_SIZE = 2*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 5*8
+
+frame_W = 0
+frame_WK = frame_W + W_SIZE
+frame_RSPSAVE = frame_WK + WK_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+# Useful QWORD "arrays" for simpler memory references
+# MSG, DIGEST, K_t, W_t are arrays
+# WK_2(t) points to 1 of 2 qwords at frame.WK depdending on t being odd/even
+
+# Input message (arg1)
+#define MSG(i) 8*i(msg)
+
+# Output Digest (arg2)
+#define DIGEST(i) 8*i(digest)
+
+# SHA Constants (static mem)
+#define K_t(i) 8*i+K512(%rip)
+
+# Message Schedule (stack frame)
+#define W_t(i) 8*i+frame_W(%rsp)
+
+# W[t]+K[t] (stack frame)
+#define WK_2(i) 8*((i%2))+frame_WK(%rsp)
+
+.macro RotateState
+ # Rotate symbols a..h right
+ TMP = h_64
+ h_64 = g_64
+ g_64 = f_64
+ f_64 = e_64
+ e_64 = d_64
+ d_64 = c_64
+ c_64 = b_64
+ b_64 = a_64
+ a_64 = TMP
+.endm
+
+.macro SHA512_Round rnd
+
+ # Compute Round %%t
+ mov f_64, T1 # T1 = f
+ mov e_64, tmp0 # tmp = e
+ xor g_64, T1 # T1 = f ^ g
+ ror $23, tmp0 # 41 # tmp = e ror 23
+ and e_64, T1 # T1 = (f ^ g) & e
+ xor e_64, tmp0 # tmp = (e ror 23) ^ e
+ xor g_64, T1 # T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
+ idx = \rnd
+ add WK_2(idx), T1 # W[t] + K[t] from message scheduler
+ ror $4, tmp0 # 18 # tmp = ((e ror 23) ^ e) ror 4
+ xor e_64, tmp0 # tmp = (((e ror 23) ^ e) ror 4) ^ e
+ mov a_64, T2 # T2 = a
+ add h_64, T1 # T1 = CH(e,f,g) + W[t] + K[t] + h
+ ror $14, tmp0 # 14 # tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
+ add tmp0, T1 # T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
+ mov a_64, tmp0 # tmp = a
+ xor c_64, T2 # T2 = a ^ c
+ and c_64, tmp0 # tmp = a & c
+ and b_64, T2 # T2 = (a ^ c) & b
+ xor tmp0, T2 # T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
+ mov a_64, tmp0 # tmp = a
+ ror $5, tmp0 # 39 # tmp = a ror 5
+ xor a_64, tmp0 # tmp = (a ror 5) ^ a
+ add T1, d_64 # e(next_state) = d + T1
+ ror $6, tmp0 # 34 # tmp = ((a ror 5) ^ a) ror 6
+ xor a_64, tmp0 # tmp = (((a ror 5) ^ a) ror 6) ^ a
+ lea (T1, T2), h_64 # a(next_state) = T1 + Maj(a,b,c)
+ ror $28, tmp0 # 28 # tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
+ add tmp0, h_64 # a(next_state) = T1 + Maj(a,b,c) S0(a)
+ RotateState
+.endm
+
+.macro SHA512_2Sched_2Round_sse rnd
+
+ # Compute rounds t-2 and t-1
+ # Compute message schedule QWORDS t and t+1
+
+ # Two rounds are computed based on the values for K[t-2]+W[t-2] and
+ # K[t-1]+W[t-1] which were previously stored at WK_2 by the message
+ # scheduler.
+ # The two new schedule QWORDS are stored at [W_t(%%t)] and [W_t(%%t+1)].
+ # They are then added to their respective SHA512 constants at
+ # [K_t(%%t)] and [K_t(%%t+1)] and stored at dqword [WK_2(%%t)]
+ # For brievity, the comments following vectored instructions only refer to
+ # the first of a pair of QWORDS.
+ # Eg. XMM2=W[t-2] really means XMM2={W[t-2]|W[t-1]}
+ # The computation of the message schedule and the rounds are tightly
+ # stitched to take advantage of instruction-level parallelism.
+ # For clarity, integer instructions (for the rounds calculation) are indented
+ # by one tab. Vectored instructions (for the message scheduler) are indented
+ # by two tabs.
+
+ mov f_64, T1
+ idx = \rnd -2
+ movdqa W_t(idx), %xmm2 # XMM2 = W[t-2]
+ xor g_64, T1
+ and e_64, T1
+ movdqa %xmm2, %xmm0 # XMM0 = W[t-2]
+ xor g_64, T1
+ idx = \rnd
+ add WK_2(idx), T1
+ idx = \rnd - 15
+ movdqu W_t(idx), %xmm5 # XMM5 = W[t-15]
+ mov e_64, tmp0
+ ror $23, tmp0 # 41
+ movdqa %xmm5, %xmm3 # XMM3 = W[t-15]
+ xor e_64, tmp0
+ ror $4, tmp0 # 18
+ psrlq $61-19, %xmm0 # XMM0 = W[t-2] >> 42
+ xor e_64, tmp0
+ ror $14, tmp0 # 14
+ psrlq $(8-7), %xmm3 # XMM3 = W[t-15] >> 1
+ add tmp0, T1
+ add h_64, T1
+ pxor %xmm2, %xmm0 # XMM0 = (W[t-2] >> 42) ^ W[t-2]
+ mov a_64, T2
+ xor c_64, T2
+ pxor %xmm5, %xmm3 # XMM3 = (W[t-15] >> 1) ^ W[t-15]
+ and b_64, T2
+ mov a_64, tmp0
+ psrlq $(19-6), %xmm0 # XMM0 = ((W[t-2]>>42)^W[t-2])>>13
+ and c_64, tmp0
+ xor tmp0, T2
+ psrlq $(7-1), %xmm3 # XMM3 = ((W[t-15]>>1)^W[t-15])>>6
+ mov a_64, tmp0
+ ror $5, tmp0 # 39
+ pxor %xmm2, %xmm0 # XMM0 = (((W[t-2]>>42)^W[t-2])>>13)^W[t-2]
+ xor a_64, tmp0
+ ror $6, tmp0 # 34
+ pxor %xmm5, %xmm3 # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]
+ xor a_64, tmp0
+ ror $28, tmp0 # 28
+ psrlq $6, %xmm0 # XMM0 = ((((W[t-2]>>42)^W[t-2])>>13)^W[t-2])>>6
+ add tmp0, T2
+ add T1, d_64
+ psrlq $1, %xmm3 # XMM3 = (((W[t-15]>>1)^W[t-15])>>6)^W[t-15]>>1
+ lea (T1, T2), h_64
+ RotateState
+ movdqa %xmm2, %xmm1 # XMM1 = W[t-2]
+ mov f_64, T1
+ xor g_64, T1
+ movdqa %xmm5, %xmm4 # XMM4 = W[t-15]
+ and e_64, T1
+ xor g_64, T1
+ psllq $(64-19)-(64-61) , %xmm1 # XMM1 = W[t-2] << 42
+ idx = \rnd + 1
+ add WK_2(idx), T1
+ mov e_64, tmp0
+ psllq $(64-1)-(64-8), %xmm4 # XMM4 = W[t-15] << 7
+ ror $23, tmp0 # 41
+ xor e_64, tmp0
+ pxor %xmm2, %xmm1 # XMM1 = (W[t-2] << 42)^W[t-2]
+ ror $4, tmp0 # 18
+ xor e_64, tmp0
+ pxor %xmm5, %xmm4 # XMM4 = (W[t-15]<<7)^W[t-15]
+ ror $14, tmp0 # 14
+ add tmp0, T1
+ psllq $(64-61), %xmm1 # XMM1 = ((W[t-2] << 42)^W[t-2])<<3
+ add h_64, T1
+ mov a_64, T2
+ psllq $(64-8), %xmm4 # XMM4 = ((W[t-15]<<7)^W[t-15])<<56
+ xor c_64, T2
+ and b_64, T2
+ pxor %xmm1, %xmm0 # XMM0 = s1(W[t-2])
+ mov a_64, tmp0
+ and c_64, tmp0
+ idx = \rnd - 7
+ movdqu W_t(idx), %xmm1 # XMM1 = W[t-7]
+ xor tmp0, T2
+ pxor %xmm4, %xmm3 # XMM3 = s0(W[t-15])
+ mov a_64, tmp0
+ paddq %xmm3, %xmm0 # XMM0 = s1(W[t-2]) + s0(W[t-15])
+ ror $5, tmp0 # 39
+ idx =\rnd-16
+ paddq W_t(idx), %xmm0 # XMM0 = s1(W[t-2]) + s0(W[t-15]) + W[t-16]
+ xor a_64, tmp0
+ paddq %xmm1, %xmm0 # XMM0 = s1(W[t-2]) + W[t-7] + s0(W[t-15]) + W[t-16]
+ ror $6, tmp0 # 34
+ movdqa %xmm0, W_t(\rnd) # Store scheduled qwords
+ xor a_64, tmp0
+ paddq K_t(\rnd), %xmm0 # Compute W[t]+K[t]
+ ror $28, tmp0 # 28
+ idx = \rnd
+ movdqa %xmm0, WK_2(idx) # Store W[t]+K[t] for next rounds
+ add tmp0, T2
+ add T1, d_64
+ lea (T1, T2), h_64
+ RotateState
+.endm
+
+########################################################################
+# void sha512_transform_ssse3(const void* M, void* D, u64 L)#
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks.
+########################################################################
+ENTRY(sha512_transform_ssse3)
+
+ cmp $0, msglen
+ je nowork
+
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbx, frame_GPRSAVE(%rsp)
+ mov %r12, frame_GPRSAVE +8*1(%rsp)
+ mov %r13, frame_GPRSAVE +8*2(%rsp)
+ mov %r14, frame_GPRSAVE +8*3(%rsp)
+ mov %r15, frame_GPRSAVE +8*4(%rsp)
+
+updateblock:
+
+# Load state variables
+ mov DIGEST(0), a_64
+ mov DIGEST(1), b_64
+ mov DIGEST(2), c_64
+ mov DIGEST(3), d_64
+ mov DIGEST(4), e_64
+ mov DIGEST(5), f_64
+ mov DIGEST(6), g_64
+ mov DIGEST(7), h_64
+
+ t = 0
+ .rept 80/2 + 1
+ # (80 rounds) / (2 rounds/iteration) + (1 iteration)
+ # +1 iteration because the scheduler leads hashing by 1 iteration
+ .if t < 2
+ # BSWAP 2 QWORDS
+ movdqa XMM_QWORD_BSWAP(%rip), %xmm1
+ movdqu MSG(t), %xmm0
+ pshufb %xmm1, %xmm0 # BSWAP
+ movdqa %xmm0, W_t(t) # Store Scheduled Pair
+ paddq K_t(t), %xmm0 # Compute W[t]+K[t]
+ movdqa %xmm0, WK_2(t) # Store into WK for rounds
+ .elseif t < 16
+ # BSWAP 2 QWORDS# Compute 2 Rounds
+ movdqu MSG(t), %xmm0
+ pshufb %xmm1, %xmm0 # BSWAP
+ SHA512_Round t-2 # Round t-2
+ movdqa %xmm0, W_t(t) # Store Scheduled Pair
+ paddq K_t(t), %xmm0 # Compute W[t]+K[t]
+ SHA512_Round t-1 # Round t-1
+ movdqa %xmm0, WK_2(t) # Store W[t]+K[t] into WK
+ .elseif t < 79
+ # Schedule 2 QWORDS# Compute 2 Rounds
+ SHA512_2Sched_2Round_sse t
+ .else
+ # Compute 2 Rounds
+ SHA512_Round t-2
+ SHA512_Round t-1
+ .endif
+ t = t+2
+ .endr
+
+ # Update digest
+ add a_64, DIGEST(0)
+ add b_64, DIGEST(1)
+ add c_64, DIGEST(2)
+ add d_64, DIGEST(3)
+ add e_64, DIGEST(4)
+ add f_64, DIGEST(5)
+ add g_64, DIGEST(6)
+ add h_64, DIGEST(7)
+
+ # Advance to next message block
+ add $16*8, msg
+ dec msglen
+ jnz updateblock
+
+ # Restore GPRs
+ mov frame_GPRSAVE(%rsp), %rbx
+ mov frame_GPRSAVE +8*1(%rsp), %r12
+ mov frame_GPRSAVE +8*2(%rsp), %r13
+ mov frame_GPRSAVE +8*3(%rsp), %r14
+ mov frame_GPRSAVE +8*4(%rsp), %r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+
+nowork:
+ ret
+ENDPROC(sha512_transform_ssse3)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 16
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+XMM_QWORD_BSWAP:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 08/10] Optimized SHA512 x86_64 assembly routine using AVX instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 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>
---
arch/x86/crypto/sha512-avx-asm.S | 423 +++++++++++++++++++++++++++++++++++++++
1 file changed, 423 insertions(+)
create mode 100644 arch/x86/crypto/sha512-avx-asm.S
diff --git a/arch/x86/crypto/sha512-avx-asm.S b/arch/x86/crypto/sha512-avx-asm.S
new file mode 100644
index 0000000..974dde9
--- /dev/null
+++ b/arch/x86/crypto/sha512-avx-asm.S
@@ -0,0 +1,423 @@
+########################################################################
+# Implement fast SHA-512 with AVX instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 Implementations on Intel Architecture Processors"
+#
+# To find it, surf to http://www.intel.com/p/en_US/embedded
+# and search for that title.
+#
+########################################################################
+
+#ifdef CONFIG_AS_AVX
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+# ARG1
+msg = %rdi
+# ARG2
+digest = %rsi
+# ARG3
+msglen = %rdx
+T1 = %rcx
+T2 = %r8
+a_64 = %r9
+b_64 = %r10
+c_64 = %r11
+d_64 = %r12
+e_64 = %r13
+f_64 = %r14
+g_64 = %r15
+h_64 = %rbx
+tmp0 = %rax
+
+# Local variables (stack frame)
+
+# Message Schedule
+W_SIZE = 80*8
+# W[t] + K[t] | W[t+1] + K[t+1]
+WK_SIZE = 2*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 5*8
+
+frame_W = 0
+frame_WK = frame_W + W_SIZE
+frame_RSPSAVE = frame_WK + WK_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+# Useful QWORD "arrays" for simpler memory references
+# MSG, DIGEST, K_t, W_t are arrays
+# WK_2(t) points to 1 of 2 qwords at frame.WK depdending on t being odd/even
+
+# Input message (arg1)
+#define MSG(i) 8*i(msg)
+
+# Output Digest (arg2)
+#define DIGEST(i) 8*i(digest)
+
+# SHA Constants (static mem)
+#define K_t(i) 8*i+K512(%rip)
+
+# Message Schedule (stack frame)
+#define W_t(i) 8*i+frame_W(%rsp)
+
+# W[t]+K[t] (stack frame)
+#define WK_2(i) 8*((i%2))+frame_WK(%rsp)
+
+.macro RotateState
+ # Rotate symbols a..h right
+ TMP = h_64
+ h_64 = g_64
+ g_64 = f_64
+ f_64 = e_64
+ e_64 = d_64
+ d_64 = c_64
+ c_64 = b_64
+ b_64 = a_64
+ a_64 = TMP
+.endm
+
+.macro RORQ p1 p2
+ # shld is faster than ror on Sandybridge
+ shld $(64-\p2), \p1, \p1
+.endm
+
+.macro SHA512_Round rnd
+ # Compute Round %%t
+ mov f_64, T1 # T1 = f
+ mov e_64, tmp0 # tmp = e
+ xor g_64, T1 # T1 = f ^ g
+ RORQ tmp0, 23 # 41 # tmp = e ror 23
+ and e_64, T1 # T1 = (f ^ g) & e
+ xor e_64, tmp0 # tmp = (e ror 23) ^ e
+ xor g_64, T1 # T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
+ idx = \rnd
+ add WK_2(idx), T1 # W[t] + K[t] from message scheduler
+ RORQ tmp0, 4 # 18 # tmp = ((e ror 23) ^ e) ror 4
+ xor e_64, tmp0 # tmp = (((e ror 23) ^ e) ror 4) ^ e
+ mov a_64, T2 # T2 = a
+ add h_64, T1 # T1 = CH(e,f,g) + W[t] + K[t] + h
+ RORQ tmp0, 14 # 14 # tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
+ add tmp0, T1 # T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
+ mov a_64, tmp0 # tmp = a
+ xor c_64, T2 # T2 = a ^ c
+ and c_64, tmp0 # tmp = a & c
+ and b_64, T2 # T2 = (a ^ c) & b
+ xor tmp0, T2 # T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
+ mov a_64, tmp0 # tmp = a
+ RORQ tmp0, 5 # 39 # tmp = a ror 5
+ xor a_64, tmp0 # tmp = (a ror 5) ^ a
+ add T1, d_64 # e(next_state) = d + T1
+ RORQ tmp0, 6 # 34 # tmp = ((a ror 5) ^ a) ror 6
+ xor a_64, tmp0 # tmp = (((a ror 5) ^ a) ror 6) ^ a
+ lea (T1, T2), h_64 # a(next_state) = T1 + Maj(a,b,c)
+ RORQ tmp0, 28 # 28 # tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
+ add tmp0, h_64 # a(next_state) = T1 + Maj(a,b,c) S0(a)
+ RotateState
+.endm
+
+.macro SHA512_2Sched_2Round_avx rnd
+ # Compute rounds t-2 and t-1
+ # Compute message schedule QWORDS t and t+1
+
+ # Two rounds are computed based on the values for K[t-2]+W[t-2] and
+ # K[t-1]+W[t-1] which were previously stored at WK_2 by the message
+ # scheduler.
+ # The two new schedule QWORDS are stored at [W_t(t)] and [W_t(t+1)].
+ # They are then added to their respective SHA512 constants at
+ # [K_t(t)] and [K_t(t+1)] and stored at dqword [WK_2(t)]
+ # For brievity, the comments following vectored instructions only refer to
+ # the first of a pair of QWORDS.
+ # Eg. XMM4=W[t-2] really means XMM4={W[t-2]|W[t-1]}
+ # The computation of the message schedule and the rounds are tightly
+ # stitched to take advantage of instruction-level parallelism.
+
+ idx = \rnd - 2
+ vmovdqa W_t(idx), %xmm4 # XMM4 = W[t-2]
+ idx = \rnd - 15
+ vmovdqu W_t(idx), %xmm5 # XMM5 = W[t-15]
+ mov f_64, T1
+ vpsrlq $61, %xmm4, %xmm0 # XMM0 = W[t-2]>>61
+ mov e_64, tmp0
+ vpsrlq $1, %xmm5, %xmm6 # XMM6 = W[t-15]>>1
+ xor g_64, T1
+ RORQ tmp0, 23 # 41
+ vpsrlq $19, %xmm4, %xmm1 # XMM1 = W[t-2]>>19
+ and e_64, T1
+ xor e_64, tmp0
+ vpxor %xmm1, %xmm0, %xmm0 # XMM0 = W[t-2]>>61 ^ W[t-2]>>19
+ xor g_64, T1
+ idx = \rnd
+ add WK_2(idx), T1#
+ vpsrlq $8, %xmm5, %xmm7 # XMM7 = W[t-15]>>8
+ RORQ tmp0, 4 # 18
+ vpsrlq $6, %xmm4, %xmm2 # XMM2 = W[t-2]>>6
+ xor e_64, tmp0
+ mov a_64, T2
+ add h_64, T1
+ vpxor %xmm7, %xmm6, %xmm6 # XMM6 = W[t-15]>>1 ^ W[t-15]>>8
+ RORQ tmp0, 14 # 14
+ add tmp0, T1
+ vpsrlq $7, %xmm5, %xmm8 # XMM8 = W[t-15]>>7
+ mov a_64, tmp0
+ xor c_64, T2
+ vpsllq $(64-61), %xmm4, %xmm3 # XMM3 = W[t-2]<<3
+ and c_64, tmp0
+ and b_64, T2
+ vpxor %xmm3, %xmm2, %xmm2 # XMM2 = W[t-2]>>6 ^ W[t-2]<<3
+ xor tmp0, T2
+ mov a_64, tmp0
+ vpsllq $(64-1), %xmm5, %xmm9 # XMM9 = W[t-15]<<63
+ RORQ tmp0, 5 # 39
+ vpxor %xmm9, %xmm8, %xmm8 # XMM8 = W[t-15]>>7 ^ W[t-15]<<63
+ xor a_64, tmp0
+ add T1, d_64
+ RORQ tmp0, 6 # 34
+ xor a_64, tmp0
+ vpxor %xmm8, %xmm6, %xmm6 # XMM6 = W[t-15]>>1 ^ W[t-15]>>8 ^
+ # W[t-15]>>7 ^ W[t-15]<<63
+ lea (T1, T2), h_64
+ RORQ tmp0, 28 # 28
+ vpsllq $(64-19), %xmm4, %xmm4 # XMM4 = W[t-2]<<25
+ add tmp0, h_64
+ RotateState
+ vpxor %xmm4, %xmm0, %xmm0 # XMM0 = W[t-2]>>61 ^ W[t-2]>>19 ^
+ # W[t-2]<<25
+ mov f_64, T1
+ vpxor %xmm2, %xmm0, %xmm0 # XMM0 = s1(W[t-2])
+ mov e_64, tmp0
+ xor g_64, T1
+ idx = \rnd - 16
+ vpaddq W_t(idx), %xmm0, %xmm0 # XMM0 = s1(W[t-2]) + W[t-16]
+ idx = \rnd - 7
+ vmovdqu W_t(idx), %xmm1 # XMM1 = W[t-7]
+ RORQ tmp0, 23 # 41
+ and e_64, T1
+ xor e_64, tmp0
+ xor g_64, T1
+ vpsllq $(64-8), %xmm5, %xmm5 # XMM5 = W[t-15]<<56
+ idx = \rnd + 1
+ add WK_2(idx), T1
+ vpxor %xmm5, %xmm6, %xmm6 # XMM6 = s0(W[t-15])
+ RORQ tmp0, 4 # 18
+ vpaddq %xmm6, %xmm0, %xmm0 # XMM0 = s1(W[t-2]) + W[t-16] + s0(W[t-15])
+ xor e_64, tmp0
+ vpaddq %xmm1, %xmm0, %xmm0 # XMM0 = W[t] = s1(W[t-2]) + W[t-7] +
+ # s0(W[t-15]) + W[t-16]
+ mov a_64, T2
+ add h_64, T1
+ RORQ tmp0, 14 # 14
+ add tmp0, T1
+ idx = \rnd
+ vmovdqa %xmm0, W_t(idx) # Store W[t]
+ vpaddq K_t(idx), %xmm0, %xmm0 # Compute W[t]+K[t]
+ vmovdqa %xmm0, WK_2(idx) # Store W[t]+K[t] for next rounds
+ mov a_64, tmp0
+ xor c_64, T2
+ and c_64, tmp0
+ and b_64, T2
+ xor tmp0, T2
+ mov a_64, tmp0
+ RORQ tmp0, 5 # 39
+ xor a_64, tmp0
+ add T1, d_64
+ RORQ tmp0, 6 # 34
+ xor a_64, tmp0
+ lea (T1, T2), h_64
+ RORQ tmp0, 28 # 28
+ add tmp0, h_64
+ RotateState
+.endm
+
+########################################################################
+# void sha512_transform_avx(const void* M, void* D, u64 L)
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks
+########################################################################
+ENTRY(sha512_transform_avx)
+ cmp $0, msglen
+ je nowork
+
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbx, frame_GPRSAVE(%rsp)
+ mov %r12, frame_GPRSAVE +8*1(%rsp)
+ mov %r13, frame_GPRSAVE +8*2(%rsp)
+ mov %r14, frame_GPRSAVE +8*3(%rsp)
+ mov %r15, frame_GPRSAVE +8*4(%rsp)
+
+updateblock:
+
+ # Load state variables
+ mov DIGEST(0), a_64
+ mov DIGEST(1), b_64
+ mov DIGEST(2), c_64
+ mov DIGEST(3), d_64
+ mov DIGEST(4), e_64
+ mov DIGEST(5), f_64
+ mov DIGEST(6), g_64
+ mov DIGEST(7), h_64
+
+ t = 0
+ .rept 80/2 + 1
+ # (80 rounds) / (2 rounds/iteration) + (1 iteration)
+ # +1 iteration because the scheduler leads hashing by 1 iteration
+ .if t < 2
+ # BSWAP 2 QWORDS
+ vmovdqa XMM_QWORD_BSWAP(%rip), %xmm1
+ vmovdqu MSG(t), %xmm0
+ vpshufb %xmm1, %xmm0, %xmm0 # BSWAP
+ vmovdqa %xmm0, W_t(t) # Store Scheduled Pair
+ vpaddq K_t(t), %xmm0, %xmm0 # Compute W[t]+K[t]
+ vmovdqa %xmm0, WK_2(t) # Store into WK for rounds
+ .elseif t < 16
+ # BSWAP 2 QWORDS# Compute 2 Rounds
+ vmovdqu MSG(t), %xmm0
+ vpshufb %xmm1, %xmm0, %xmm0 # BSWAP
+ SHA512_Round t-2 # Round t-2
+ vmovdqa %xmm0, W_t(t) # Store Scheduled Pair
+ vpaddq K_t(t), %xmm0, %xmm0 # Compute W[t]+K[t]
+ SHA512_Round t-1 # Round t-1
+ vmovdqa %xmm0, WK_2(t)# Store W[t]+K[t] into WK
+ .elseif t < 79
+ # Schedule 2 QWORDS# Compute 2 Rounds
+ SHA512_2Sched_2Round_avx t
+ .else
+ # Compute 2 Rounds
+ SHA512_Round t-2
+ SHA512_Round t-1
+ .endif
+ t = t+2
+ .endr
+
+ # Update digest
+ add a_64, DIGEST(0)
+ add b_64, DIGEST(1)
+ add c_64, DIGEST(2)
+ add d_64, DIGEST(3)
+ add e_64, DIGEST(4)
+ add f_64, DIGEST(5)
+ add g_64, DIGEST(6)
+ add h_64, DIGEST(7)
+
+ # Advance to next message block
+ add $16*8, msg
+ dec msglen
+ jnz updateblock
+
+ # Restore GPRs
+ mov frame_GPRSAVE(%rsp), %rbx
+ mov frame_GPRSAVE +8*1(%rsp), %r12
+ mov frame_GPRSAVE +8*2(%rsp), %r13
+ mov frame_GPRSAVE +8*3(%rsp), %r14
+ mov frame_GPRSAVE +8*4(%rsp), %r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+
+nowork:
+ ret
+ENDPROC(sha512_transform_avx)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 16
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+XMM_QWORD_BSWAP:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 08/10] Optimized SHA512 x86_64 assembly routine using AVX instructions.
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 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>
---
arch/x86/crypto/sha512-avx-asm.S | 423 +++++++++++++++++++++++++++++++++++++++
1 file changed, 423 insertions(+)
create mode 100644 arch/x86/crypto/sha512-avx-asm.S
diff --git a/arch/x86/crypto/sha512-avx-asm.S b/arch/x86/crypto/sha512-avx-asm.S
new file mode 100644
index 0000000..974dde9
--- /dev/null
+++ b/arch/x86/crypto/sha512-avx-asm.S
@@ -0,0 +1,423 @@
+########################################################################
+# Implement fast SHA-512 with AVX instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 Implementations on Intel Architecture Processors"
+#
+# To find it, surf to http://www.intel.com/p/en_US/embedded
+# and search for that title.
+#
+########################################################################
+
+#ifdef CONFIG_AS_AVX
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+# ARG1
+msg = %rdi
+# ARG2
+digest = %rsi
+# ARG3
+msglen = %rdx
+T1 = %rcx
+T2 = %r8
+a_64 = %r9
+b_64 = %r10
+c_64 = %r11
+d_64 = %r12
+e_64 = %r13
+f_64 = %r14
+g_64 = %r15
+h_64 = %rbx
+tmp0 = %rax
+
+# Local variables (stack frame)
+
+# Message Schedule
+W_SIZE = 80*8
+# W[t] + K[t] | W[t+1] + K[t+1]
+WK_SIZE = 2*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 5*8
+
+frame_W = 0
+frame_WK = frame_W + W_SIZE
+frame_RSPSAVE = frame_WK + WK_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+# Useful QWORD "arrays" for simpler memory references
+# MSG, DIGEST, K_t, W_t are arrays
+# WK_2(t) points to 1 of 2 qwords at frame.WK depdending on t being odd/even
+
+# Input message (arg1)
+#define MSG(i) 8*i(msg)
+
+# Output Digest (arg2)
+#define DIGEST(i) 8*i(digest)
+
+# SHA Constants (static mem)
+#define K_t(i) 8*i+K512(%rip)
+
+# Message Schedule (stack frame)
+#define W_t(i) 8*i+frame_W(%rsp)
+
+# W[t]+K[t] (stack frame)
+#define WK_2(i) 8*((i%2))+frame_WK(%rsp)
+
+.macro RotateState
+ # Rotate symbols a..h right
+ TMP = h_64
+ h_64 = g_64
+ g_64 = f_64
+ f_64 = e_64
+ e_64 = d_64
+ d_64 = c_64
+ c_64 = b_64
+ b_64 = a_64
+ a_64 = TMP
+.endm
+
+.macro RORQ p1 p2
+ # shld is faster than ror on Sandybridge
+ shld $(64-\p2), \p1, \p1
+.endm
+
+.macro SHA512_Round rnd
+ # Compute Round %%t
+ mov f_64, T1 # T1 = f
+ mov e_64, tmp0 # tmp = e
+ xor g_64, T1 # T1 = f ^ g
+ RORQ tmp0, 23 # 41 # tmp = e ror 23
+ and e_64, T1 # T1 = (f ^ g) & e
+ xor e_64, tmp0 # tmp = (e ror 23) ^ e
+ xor g_64, T1 # T1 = ((f ^ g) & e) ^ g = CH(e,f,g)
+ idx = \rnd
+ add WK_2(idx), T1 # W[t] + K[t] from message scheduler
+ RORQ tmp0, 4 # 18 # tmp = ((e ror 23) ^ e) ror 4
+ xor e_64, tmp0 # tmp = (((e ror 23) ^ e) ror 4) ^ e
+ mov a_64, T2 # T2 = a
+ add h_64, T1 # T1 = CH(e,f,g) + W[t] + K[t] + h
+ RORQ tmp0, 14 # 14 # tmp = ((((e ror23)^e)ror4)^e)ror14 = S1(e)
+ add tmp0, T1 # T1 = CH(e,f,g) + W[t] + K[t] + S1(e)
+ mov a_64, tmp0 # tmp = a
+ xor c_64, T2 # T2 = a ^ c
+ and c_64, tmp0 # tmp = a & c
+ and b_64, T2 # T2 = (a ^ c) & b
+ xor tmp0, T2 # T2 = ((a ^ c) & b) ^ (a & c) = Maj(a,b,c)
+ mov a_64, tmp0 # tmp = a
+ RORQ tmp0, 5 # 39 # tmp = a ror 5
+ xor a_64, tmp0 # tmp = (a ror 5) ^ a
+ add T1, d_64 # e(next_state) = d + T1
+ RORQ tmp0, 6 # 34 # tmp = ((a ror 5) ^ a) ror 6
+ xor a_64, tmp0 # tmp = (((a ror 5) ^ a) ror 6) ^ a
+ lea (T1, T2), h_64 # a(next_state) = T1 + Maj(a,b,c)
+ RORQ tmp0, 28 # 28 # tmp = ((((a ror5)^a)ror6)^a)ror28 = S0(a)
+ add tmp0, h_64 # a(next_state) = T1 + Maj(a,b,c) S0(a)
+ RotateState
+.endm
+
+.macro SHA512_2Sched_2Round_avx rnd
+ # Compute rounds t-2 and t-1
+ # Compute message schedule QWORDS t and t+1
+
+ # Two rounds are computed based on the values for K[t-2]+W[t-2] and
+ # K[t-1]+W[t-1] which were previously stored at WK_2 by the message
+ # scheduler.
+ # The two new schedule QWORDS are stored at [W_t(t)] and [W_t(t+1)].
+ # They are then added to their respective SHA512 constants at
+ # [K_t(t)] and [K_t(t+1)] and stored at dqword [WK_2(t)]
+ # For brievity, the comments following vectored instructions only refer to
+ # the first of a pair of QWORDS.
+ # Eg. XMM4=W[t-2] really means XMM4={W[t-2]|W[t-1]}
+ # The computation of the message schedule and the rounds are tightly
+ # stitched to take advantage of instruction-level parallelism.
+
+ idx = \rnd - 2
+ vmovdqa W_t(idx), %xmm4 # XMM4 = W[t-2]
+ idx = \rnd - 15
+ vmovdqu W_t(idx), %xmm5 # XMM5 = W[t-15]
+ mov f_64, T1
+ vpsrlq $61, %xmm4, %xmm0 # XMM0 = W[t-2]>>61
+ mov e_64, tmp0
+ vpsrlq $1, %xmm5, %xmm6 # XMM6 = W[t-15]>>1
+ xor g_64, T1
+ RORQ tmp0, 23 # 41
+ vpsrlq $19, %xmm4, %xmm1 # XMM1 = W[t-2]>>19
+ and e_64, T1
+ xor e_64, tmp0
+ vpxor %xmm1, %xmm0, %xmm0 # XMM0 = W[t-2]>>61 ^ W[t-2]>>19
+ xor g_64, T1
+ idx = \rnd
+ add WK_2(idx), T1#
+ vpsrlq $8, %xmm5, %xmm7 # XMM7 = W[t-15]>>8
+ RORQ tmp0, 4 # 18
+ vpsrlq $6, %xmm4, %xmm2 # XMM2 = W[t-2]>>6
+ xor e_64, tmp0
+ mov a_64, T2
+ add h_64, T1
+ vpxor %xmm7, %xmm6, %xmm6 # XMM6 = W[t-15]>>1 ^ W[t-15]>>8
+ RORQ tmp0, 14 # 14
+ add tmp0, T1
+ vpsrlq $7, %xmm5, %xmm8 # XMM8 = W[t-15]>>7
+ mov a_64, tmp0
+ xor c_64, T2
+ vpsllq $(64-61), %xmm4, %xmm3 # XMM3 = W[t-2]<<3
+ and c_64, tmp0
+ and b_64, T2
+ vpxor %xmm3, %xmm2, %xmm2 # XMM2 = W[t-2]>>6 ^ W[t-2]<<3
+ xor tmp0, T2
+ mov a_64, tmp0
+ vpsllq $(64-1), %xmm5, %xmm9 # XMM9 = W[t-15]<<63
+ RORQ tmp0, 5 # 39
+ vpxor %xmm9, %xmm8, %xmm8 # XMM8 = W[t-15]>>7 ^ W[t-15]<<63
+ xor a_64, tmp0
+ add T1, d_64
+ RORQ tmp0, 6 # 34
+ xor a_64, tmp0
+ vpxor %xmm8, %xmm6, %xmm6 # XMM6 = W[t-15]>>1 ^ W[t-15]>>8 ^
+ # W[t-15]>>7 ^ W[t-15]<<63
+ lea (T1, T2), h_64
+ RORQ tmp0, 28 # 28
+ vpsllq $(64-19), %xmm4, %xmm4 # XMM4 = W[t-2]<<25
+ add tmp0, h_64
+ RotateState
+ vpxor %xmm4, %xmm0, %xmm0 # XMM0 = W[t-2]>>61 ^ W[t-2]>>19 ^
+ # W[t-2]<<25
+ mov f_64, T1
+ vpxor %xmm2, %xmm0, %xmm0 # XMM0 = s1(W[t-2])
+ mov e_64, tmp0
+ xor g_64, T1
+ idx = \rnd - 16
+ vpaddq W_t(idx), %xmm0, %xmm0 # XMM0 = s1(W[t-2]) + W[t-16]
+ idx = \rnd - 7
+ vmovdqu W_t(idx), %xmm1 # XMM1 = W[t-7]
+ RORQ tmp0, 23 # 41
+ and e_64, T1
+ xor e_64, tmp0
+ xor g_64, T1
+ vpsllq $(64-8), %xmm5, %xmm5 # XMM5 = W[t-15]<<56
+ idx = \rnd + 1
+ add WK_2(idx), T1
+ vpxor %xmm5, %xmm6, %xmm6 # XMM6 = s0(W[t-15])
+ RORQ tmp0, 4 # 18
+ vpaddq %xmm6, %xmm0, %xmm0 # XMM0 = s1(W[t-2]) + W[t-16] + s0(W[t-15])
+ xor e_64, tmp0
+ vpaddq %xmm1, %xmm0, %xmm0 # XMM0 = W[t] = s1(W[t-2]) + W[t-7] +
+ # s0(W[t-15]) + W[t-16]
+ mov a_64, T2
+ add h_64, T1
+ RORQ tmp0, 14 # 14
+ add tmp0, T1
+ idx = \rnd
+ vmovdqa %xmm0, W_t(idx) # Store W[t]
+ vpaddq K_t(idx), %xmm0, %xmm0 # Compute W[t]+K[t]
+ vmovdqa %xmm0, WK_2(idx) # Store W[t]+K[t] for next rounds
+ mov a_64, tmp0
+ xor c_64, T2
+ and c_64, tmp0
+ and b_64, T2
+ xor tmp0, T2
+ mov a_64, tmp0
+ RORQ tmp0, 5 # 39
+ xor a_64, tmp0
+ add T1, d_64
+ RORQ tmp0, 6 # 34
+ xor a_64, tmp0
+ lea (T1, T2), h_64
+ RORQ tmp0, 28 # 28
+ add tmp0, h_64
+ RotateState
+.endm
+
+########################################################################
+# void sha512_transform_avx(const void* M, void* D, u64 L)
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks
+########################################################################
+ENTRY(sha512_transform_avx)
+ cmp $0, msglen
+ je nowork
+
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbx, frame_GPRSAVE(%rsp)
+ mov %r12, frame_GPRSAVE +8*1(%rsp)
+ mov %r13, frame_GPRSAVE +8*2(%rsp)
+ mov %r14, frame_GPRSAVE +8*3(%rsp)
+ mov %r15, frame_GPRSAVE +8*4(%rsp)
+
+updateblock:
+
+ # Load state variables
+ mov DIGEST(0), a_64
+ mov DIGEST(1), b_64
+ mov DIGEST(2), c_64
+ mov DIGEST(3), d_64
+ mov DIGEST(4), e_64
+ mov DIGEST(5), f_64
+ mov DIGEST(6), g_64
+ mov DIGEST(7), h_64
+
+ t = 0
+ .rept 80/2 + 1
+ # (80 rounds) / (2 rounds/iteration) + (1 iteration)
+ # +1 iteration because the scheduler leads hashing by 1 iteration
+ .if t < 2
+ # BSWAP 2 QWORDS
+ vmovdqa XMM_QWORD_BSWAP(%rip), %xmm1
+ vmovdqu MSG(t), %xmm0
+ vpshufb %xmm1, %xmm0, %xmm0 # BSWAP
+ vmovdqa %xmm0, W_t(t) # Store Scheduled Pair
+ vpaddq K_t(t), %xmm0, %xmm0 # Compute W[t]+K[t]
+ vmovdqa %xmm0, WK_2(t) # Store into WK for rounds
+ .elseif t < 16
+ # BSWAP 2 QWORDS# Compute 2 Rounds
+ vmovdqu MSG(t), %xmm0
+ vpshufb %xmm1, %xmm0, %xmm0 # BSWAP
+ SHA512_Round t-2 # Round t-2
+ vmovdqa %xmm0, W_t(t) # Store Scheduled Pair
+ vpaddq K_t(t), %xmm0, %xmm0 # Compute W[t]+K[t]
+ SHA512_Round t-1 # Round t-1
+ vmovdqa %xmm0, WK_2(t)# Store W[t]+K[t] into WK
+ .elseif t < 79
+ # Schedule 2 QWORDS# Compute 2 Rounds
+ SHA512_2Sched_2Round_avx t
+ .else
+ # Compute 2 Rounds
+ SHA512_Round t-2
+ SHA512_Round t-1
+ .endif
+ t = t+2
+ .endr
+
+ # Update digest
+ add a_64, DIGEST(0)
+ add b_64, DIGEST(1)
+ add c_64, DIGEST(2)
+ add d_64, DIGEST(3)
+ add e_64, DIGEST(4)
+ add f_64, DIGEST(5)
+ add g_64, DIGEST(6)
+ add h_64, DIGEST(7)
+
+ # Advance to next message block
+ add $16*8, msg
+ dec msglen
+ jnz updateblock
+
+ # Restore GPRs
+ mov frame_GPRSAVE(%rsp), %rbx
+ mov frame_GPRSAVE +8*1(%rsp), %r12
+ mov frame_GPRSAVE +8*2(%rsp), %r13
+ mov frame_GPRSAVE +8*3(%rsp), %r14
+ mov frame_GPRSAVE +8*4(%rsp), %r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+
+nowork:
+ ret
+ENDPROC(sha512_transform_avx)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 16
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+XMM_QWORD_BSWAP:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 09/10] Optimized SHA512 x86_64 assembly routine using AVX2 RORX instruction.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 20:59 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 x86_64 assembly routine optimized with SSE, AVX and
AVX2's RORX instructions. Speedup of 70% or more has been
measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha512-avx2-asm.S | 743 ++++++++++++++++++++++++++++++++++++++
1 file changed, 743 insertions(+)
create mode 100644 arch/x86/crypto/sha512-avx2-asm.S
diff --git a/arch/x86/crypto/sha512-avx2-asm.S b/arch/x86/crypto/sha512-avx2-asm.S
new file mode 100644
index 0000000..568b961
--- /dev/null
+++ b/arch/x86/crypto/sha512-avx2-asm.S
@@ -0,0 +1,743 @@
+########################################################################
+# Implement fast SHA-512 with AVX2 instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 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 blocks at a time, with 4 lanes per block
+########################################################################
+
+#ifdef CONFIG_AS_AVX2
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+Y_0 = %ymm4
+Y_1 = %ymm5
+Y_2 = %ymm6
+Y_3 = %ymm7
+
+YTMP0 = %ymm0
+YTMP1 = %ymm1
+YTMP2 = %ymm2
+YTMP3 = %ymm3
+YTMP4 = %ymm8
+XFER = YTMP0
+
+BYTE_FLIP_MASK = %ymm9
+
+# 1st arg
+INP = %rdi
+# 2nd arg
+CTX = %rsi
+# 3rd arg
+NUM_BLKS = %rdx
+
+c = %rcx
+d = %r8
+e = %rdx
+y3 = %rdi
+
+TBL = %rbp
+
+a = %rax
+b = %rbx
+
+f = %r9
+g = %r10
+h = %r11
+old_h = %r11
+
+T1 = %r12
+y0 = %r13
+y1 = %r14
+y2 = %r15
+
+y4 = %r12
+
+# Local variables (stack frame)
+XFER_SIZE = 4*8
+SRND_SIZE = 1*8
+INP_SIZE = 1*8
+INPEND_SIZE = 1*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 6*8
+
+frame_XFER = 0
+frame_SRND = frame_XFER + XFER_SIZE
+frame_INP = frame_SRND + SRND_SIZE
+frame_INPEND = frame_INP + INP_SIZE
+frame_RSPSAVE = frame_INPEND + INPEND_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+## assume buffers not aligned
+#define VMOVDQ vmovdqu
+
+# addm [mem], reg
+# Add reg to mem using reg-mem add and store
+.macro addm p1 p2
+ add \p1, \p2
+ mov \p2, \p1
+.endm
+
+
+# COPY_YMM_AND_BSWAP ymm, [mem], byte_flip_mask
+# Load ymm with mem and byte swap each dword
+.macro COPY_YMM_AND_BSWAP p1 p2 p3
+ VMOVDQ \p2, \p1
+ vpshufb \p3, \p1, \p1
+.endm
+# rotate_Ys
+# Rotate values of symbols Y0...Y3
+.macro rotate_Ys
+ Y_ = Y_0
+ Y_0 = Y_1
+ Y_1 = Y_2
+ Y_2 = Y_3
+ Y_3 = Y_
+.endm
+
+# RotateState
+.macro RotateState
+ # Rotate symbols a..h right
+ old_h = h
+ TMP_ = h
+ h = g
+ g = f
+ f = e
+ e = d
+ d = c
+ c = b
+ b = a
+ a = TMP_
+.endm
+
+# macro MY_VPALIGNR YDST, YSRC1, YSRC2, RVAL
+# YDST = {YSRC1, YSRC2} >> RVAL*8
+.macro MY_VPALIGNR YDST YSRC1 YSRC2 RVAL
+ vperm2f128 $0x3, \YSRC2, \YSRC1, \YDST # YDST = {YS1_LO, YS2_HI}
+ vpalignr $\RVAL, \YSRC2, \YDST, \YDST # YDST = {YDS1, YS2} >> RVAL*8
+.endm
+
+.macro FOUR_ROUNDS_AND_SCHED
+################################### RND N + 0 #########################################
+
+ # Extract w[t-7]
+ MY_VPALIGNR YTMP0, Y_3, Y_2, 8 # YTMP0 = W[-7]
+ # Calculate w[t-16] + w[t-7]
+ vpaddq Y_0, YTMP0, YTMP0 # YTMP0 = W[-7] + W[-16]
+ # Extract w[t-15]
+ MY_VPALIGNR YTMP1, Y_1, Y_0, 8 # YTMP1 = W[-15]
+
+ # Calculate sigma0
+
+ # Calculate w[t-15] ror 1
+ vpsrlq $1, YTMP1, YTMP2
+ vpsllq $(64-1), YTMP1, YTMP3
+ vpor YTMP2, YTMP3, YTMP3 # YTMP3 = W[-15] ror 1
+ # Calculate w[t-15] shr 7
+ vpsrlq $7, YTMP1, YTMP4 # YTMP4 = W[-15] >> 7
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add frame_XFER(%rsp),h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+
+ and e, y2 # y2 = (f^g)&e # CH
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+
+ add y0, y2 # y2 = S1 + CH # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+################################### RND N + 1 #########################################
+
+ # Calculate w[t-15] ror 8
+ vpsrlq $8, YTMP1, YTMP2
+ vpsllq $(64-8), YTMP1, YTMP1
+ vpor YTMP2, YTMP1, YTMP1 # YTMP1 = W[-15] ror 8
+ # XOR the three components
+ vpxor YTMP4, YTMP3, YTMP3 # YTMP3 = W[-15] ror 1 ^ W[-15] >> 7
+ vpxor YTMP1, YTMP3, YTMP1 # YTMP1 = s0
+
+
+ # Add three components, w[t-16], w[t-7] and sigma0
+ vpaddq YTMP1, YTMP0, YTMP0 # YTMP0 = W[-16] + W[-7] + s0
+ # Move to appropriate lanes for calculating w[16] and w[17]
+ vperm2f128 $0x0, YTMP0, YTMP0, Y_0 # Y_0 = W[-16] + W[-7] + s0 {BABA}
+ # Move to appropriate lanes for calculating w[18] and w[19]
+ vpand MASK_YMM_LO(%rip), YTMP0, YTMP0 # YTMP0 = W[-16] + W[-7] + s0 {DC00}
+
+ # Calculate w[16] and w[17] in both 128 bit lanes
+
+ # Calculate sigma1 for w[16] and w[17] on both 128 bit lanes
+ vperm2f128 $0x11, Y_3, Y_3, YTMP2 # YTMP2 = W[-2] {BABA}
+ vpsrlq $6, YTMP2, YTMP4 # YTMP4 = W[-2] >> 6 {BABA}
+
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add 1*8+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+
+################################### RND N + 2 #########################################
+
+ vpsrlq $19, YTMP2, YTMP3 # YTMP3 = W[-2] >> 19 {BABA}
+ vpsllq $(64-19), YTMP2, YTMP1 # YTMP1 = W[-2] << 19 {BABA}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 19 {BABA}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {BABA}
+ vpsrlq $61, YTMP2, YTMP3 # YTMP3 = W[-2] >> 61 {BABA}
+ vpsllq $(64-61), YTMP2, YTMP1 # YTMP1 = W[-2] << 61 {BABA}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 61 {BABA}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = s1 = (W[-2] ror 19) ^
+ # (W[-2] ror 61) ^ (W[-2] >> 6) {BABA}
+
+ # Add sigma1 to the other compunents to get w[16] and w[17]
+ vpaddq YTMP4, Y_0, Y_0 # Y_0 = {W[1], W[0], W[1], W[0]}
+
+ # Calculate sigma1 for w[18] and w[19] for upper 128 bit lane
+ vpsrlq $6, Y_0, YTMP4 # YTMP4 = W[-2] >> 6 {DC--}
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ add 2*8+frame_XFER(%rsp), h # h = k + w + h # --
+
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ xor g, y2 # y2 = f^g # CH
+
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+################################### RND N + 3 #########################################
+
+ vpsrlq $19, Y_0, YTMP3 # YTMP3 = W[-2] >> 19 {DC--}
+ vpsllq $(64-19), Y_0, YTMP1 # YTMP1 = W[-2] << 19 {DC--}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 19 {DC--}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {DC--}
+ vpsrlq $61, Y_0, YTMP3 # YTMP3 = W[-2] >> 61 {DC--}
+ vpsllq $(64-61), Y_0, YTMP1 # YTMP1 = W[-2] << 61 {DC--}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 61 {DC--}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = s1 = (W[-2] ror 19) ^
+ # (W[-2] ror 61) ^ (W[-2] >> 6) {DC--}
+
+ # Add the sigma0 + w[t-7] + w[t-16] for w[18] and w[19]
+ # to newly calculated sigma1 to get w[18] and w[19]
+ vpaddq YTMP4, YTMP0, YTMP2 # YTMP2 = {W[3], W[2], --, --}
+
+ # Form w[19, w[18], w17], w[16]
+ vpblendd $0xF0, YTMP2, Y_0, Y_0 # Y_0 = {W[3], W[2], W[1], W[0]}
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add 3*8+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ add y0, y2 # y2 = S1 + CH # --
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+
+ add y1, h # h = k + w + h + S0 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+ rotate_Ys
+.endm
+
+.macro DO_4ROUNDS
+
+################################### RND N + 0 #########################################
+
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 1 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*1+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 2 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*2+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 3 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*3+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+.endm
+
+########################################################################
+# void sha512_transform_rorx(const void* M, void* D, uint64_t L)#
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks
+########################################################################
+ENTRY(sha512_transform_rorx)
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbp, frame_GPRSAVE(%rsp)
+ mov %rbx, 8*1+frame_GPRSAVE(%rsp)
+ mov %r12, 8*2+frame_GPRSAVE(%rsp)
+ mov %r13, 8*3+frame_GPRSAVE(%rsp)
+ mov %r14, 8*4+frame_GPRSAVE(%rsp)
+ mov %r15, 8*5+frame_GPRSAVE(%rsp)
+
+ shl $7, NUM_BLKS # convert to bytes
+ jz done_hash
+ add INP, NUM_BLKS # pointer to end of data
+ mov NUM_BLKS, frame_INPEND(%rsp)
+
+ ## load initial digest
+ mov 8*0(CTX),a
+ mov 8*1(CTX),b
+ mov 8*2(CTX),c
+ mov 8*3(CTX),d
+ mov 8*4(CTX),e
+ mov 8*5(CTX),f
+ mov 8*6(CTX),g
+ mov 8*7(CTX),h
+
+ vmovdqa PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
+
+loop0:
+ lea K512(%rip), TBL
+
+ ## byte swap first 16 dwords
+ COPY_YMM_AND_BSWAP Y_0, (INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_1, 1*32(INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_2, 2*32(INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_3, 3*32(INP), BYTE_FLIP_MASK
+
+ mov INP, frame_INP(%rsp)
+
+ ## schedule 64 input dwords, by doing 12 rounds of 4 each
+ movq $4, frame_SRND(%rsp)
+
+.align 16
+loop1:
+ vpaddq (TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 1*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 2*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 3*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ add $(4*32), TBL
+ FOUR_ROUNDS_AND_SCHED
+
+ subq $1, frame_SRND(%rsp)
+ jne loop1
+
+ movq $2, frame_SRND(%rsp)
+loop2:
+ vpaddq (TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ DO_4ROUNDS
+ vpaddq 1*32(TBL), Y_1, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ add $(2*32), TBL
+ DO_4ROUNDS
+
+ vmovdqa Y_2, Y_0
+ vmovdqa Y_3, Y_1
+
+ subq $1, frame_SRND(%rsp)
+ jne loop2
+
+ addm 8*0(CTX),a
+ addm 8*1(CTX),b
+ addm 8*2(CTX),c
+ addm 8*3(CTX),d
+ addm 8*4(CTX),e
+ addm 8*5(CTX),f
+ addm 8*6(CTX),g
+ addm 8*7(CTX),h
+
+ mov frame_INP(%rsp), INP
+ add $128, INP
+ cmp frame_INPEND(%rsp), INP
+ jne loop0
+
+done_hash:
+
+# Restore GPRs
+ mov frame_GPRSAVE(%rsp) ,%rbp
+ mov 8*1+frame_GPRSAVE(%rsp) ,%rbx
+ mov 8*2+frame_GPRSAVE(%rsp) ,%r12
+ mov 8*3+frame_GPRSAVE(%rsp) ,%r13
+ mov 8*4+frame_GPRSAVE(%rsp) ,%r14
+ mov 8*5+frame_GPRSAVE(%rsp) ,%r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+ ret
+ENDPROC(sha512_transform_rorx)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 64
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
+
+.align 32
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+PSHUFFLE_BYTE_FLIP_MASK:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+ .octa 0x18191a1b1c1d1e1f1011121314151617
+
+MASK_YMM_LO:
+ .octa 0x00000000000000000000000000000000
+ .octa 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 09/10] Optimized SHA512 x86_64 assembly routine using AVX2 RORX instruction.
@ 2013-03-26 20:59 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 20:59 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
Provides SHA512 x86_64 assembly routine optimized with SSE, AVX and
AVX2's RORX instructions. Speedup of 70% or more has been
measured over the generic implementation.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha512-avx2-asm.S | 743 ++++++++++++++++++++++++++++++++++++++
1 file changed, 743 insertions(+)
create mode 100644 arch/x86/crypto/sha512-avx2-asm.S
diff --git a/arch/x86/crypto/sha512-avx2-asm.S b/arch/x86/crypto/sha512-avx2-asm.S
new file mode 100644
index 0000000..568b961
--- /dev/null
+++ b/arch/x86/crypto/sha512-avx2-asm.S
@@ -0,0 +1,743 @@
+########################################################################
+# Implement fast SHA-512 with AVX2 instructions. (x86_64)
+#
+# Copyright (C) 2013 Intel Corporation.
+#
+# Authors:
+# James Guilford <james.guilford@intel.com>
+# Kirk Yap <kirk.s.yap@intel.com>
+# David Cote <david.m.cote@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-512 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 blocks at a time, with 4 lanes per block
+########################################################################
+
+#ifdef CONFIG_AS_AVX2
+#include <linux/linkage.h>
+
+.text
+
+# Virtual Registers
+Y_0 = %ymm4
+Y_1 = %ymm5
+Y_2 = %ymm6
+Y_3 = %ymm7
+
+YTMP0 = %ymm0
+YTMP1 = %ymm1
+YTMP2 = %ymm2
+YTMP3 = %ymm3
+YTMP4 = %ymm8
+XFER = YTMP0
+
+BYTE_FLIP_MASK = %ymm9
+
+# 1st arg
+INP = %rdi
+# 2nd arg
+CTX = %rsi
+# 3rd arg
+NUM_BLKS = %rdx
+
+c = %rcx
+d = %r8
+e = %rdx
+y3 = %rdi
+
+TBL = %rbp
+
+a = %rax
+b = %rbx
+
+f = %r9
+g = %r10
+h = %r11
+old_h = %r11
+
+T1 = %r12
+y0 = %r13
+y1 = %r14
+y2 = %r15
+
+y4 = %r12
+
+# Local variables (stack frame)
+XFER_SIZE = 4*8
+SRND_SIZE = 1*8
+INP_SIZE = 1*8
+INPEND_SIZE = 1*8
+RSPSAVE_SIZE = 1*8
+GPRSAVE_SIZE = 6*8
+
+frame_XFER = 0
+frame_SRND = frame_XFER + XFER_SIZE
+frame_INP = frame_SRND + SRND_SIZE
+frame_INPEND = frame_INP + INP_SIZE
+frame_RSPSAVE = frame_INPEND + INPEND_SIZE
+frame_GPRSAVE = frame_RSPSAVE + RSPSAVE_SIZE
+frame_size = frame_GPRSAVE + GPRSAVE_SIZE
+
+## assume buffers not aligned
+#define VMOVDQ vmovdqu
+
+# addm [mem], reg
+# Add reg to mem using reg-mem add and store
+.macro addm p1 p2
+ add \p1, \p2
+ mov \p2, \p1
+.endm
+
+
+# COPY_YMM_AND_BSWAP ymm, [mem], byte_flip_mask
+# Load ymm with mem and byte swap each dword
+.macro COPY_YMM_AND_BSWAP p1 p2 p3
+ VMOVDQ \p2, \p1
+ vpshufb \p3, \p1, \p1
+.endm
+# rotate_Ys
+# Rotate values of symbols Y0...Y3
+.macro rotate_Ys
+ Y_ = Y_0
+ Y_0 = Y_1
+ Y_1 = Y_2
+ Y_2 = Y_3
+ Y_3 = Y_
+.endm
+
+# RotateState
+.macro RotateState
+ # Rotate symbols a..h right
+ old_h = h
+ TMP_ = h
+ h = g
+ g = f
+ f = e
+ e = d
+ d = c
+ c = b
+ b = a
+ a = TMP_
+.endm
+
+# macro MY_VPALIGNR YDST, YSRC1, YSRC2, RVAL
+# YDST = {YSRC1, YSRC2} >> RVAL*8
+.macro MY_VPALIGNR YDST YSRC1 YSRC2 RVAL
+ vperm2f128 $0x3, \YSRC2, \YSRC1, \YDST # YDST = {YS1_LO, YS2_HI}
+ vpalignr $\RVAL, \YSRC2, \YDST, \YDST # YDST = {YDS1, YS2} >> RVAL*8
+.endm
+
+.macro FOUR_ROUNDS_AND_SCHED
+################################### RND N + 0 #########################################
+
+ # Extract w[t-7]
+ MY_VPALIGNR YTMP0, Y_3, Y_2, 8 # YTMP0 = W[-7]
+ # Calculate w[t-16] + w[t-7]
+ vpaddq Y_0, YTMP0, YTMP0 # YTMP0 = W[-7] + W[-16]
+ # Extract w[t-15]
+ MY_VPALIGNR YTMP1, Y_1, Y_0, 8 # YTMP1 = W[-15]
+
+ # Calculate sigma0
+
+ # Calculate w[t-15] ror 1
+ vpsrlq $1, YTMP1, YTMP2
+ vpsllq $(64-1), YTMP1, YTMP3
+ vpor YTMP2, YTMP3, YTMP3 # YTMP3 = W[-15] ror 1
+ # Calculate w[t-15] shr 7
+ vpsrlq $7, YTMP1, YTMP4 # YTMP4 = W[-15] >> 7
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add frame_XFER(%rsp),h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+
+ and e, y2 # y2 = (f^g)&e # CH
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+
+ add y0, y2 # y2 = S1 + CH # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+################################### RND N + 1 #########################################
+
+ # Calculate w[t-15] ror 8
+ vpsrlq $8, YTMP1, YTMP2
+ vpsllq $(64-8), YTMP1, YTMP1
+ vpor YTMP2, YTMP1, YTMP1 # YTMP1 = W[-15] ror 8
+ # XOR the three components
+ vpxor YTMP4, YTMP3, YTMP3 # YTMP3 = W[-15] ror 1 ^ W[-15] >> 7
+ vpxor YTMP1, YTMP3, YTMP1 # YTMP1 = s0
+
+
+ # Add three components, w[t-16], w[t-7] and sigma0
+ vpaddq YTMP1, YTMP0, YTMP0 # YTMP0 = W[-16] + W[-7] + s0
+ # Move to appropriate lanes for calculating w[16] and w[17]
+ vperm2f128 $0x0, YTMP0, YTMP0, Y_0 # Y_0 = W[-16] + W[-7] + s0 {BABA}
+ # Move to appropriate lanes for calculating w[18] and w[19]
+ vpand MASK_YMM_LO(%rip), YTMP0, YTMP0 # YTMP0 = W[-16] + W[-7] + s0 {DC00}
+
+ # Calculate w[16] and w[17] in both 128 bit lanes
+
+ # Calculate sigma1 for w[16] and w[17] on both 128 bit lanes
+ vperm2f128 $0x11, Y_3, Y_3, YTMP2 # YTMP2 = W[-2] {BABA}
+ vpsrlq $6, YTMP2, YTMP4 # YTMP4 = W[-2] >> 6 {BABA}
+
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add 1*8+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+
+ and b, y3 # y3 = (a|c)&b # MAJA
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+
+################################### RND N + 2 #########################################
+
+ vpsrlq $19, YTMP2, YTMP3 # YTMP3 = W[-2] >> 19 {BABA}
+ vpsllq $(64-19), YTMP2, YTMP1 # YTMP1 = W[-2] << 19 {BABA}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 19 {BABA}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {BABA}
+ vpsrlq $61, YTMP2, YTMP3 # YTMP3 = W[-2] >> 61 {BABA}
+ vpsllq $(64-61), YTMP2, YTMP1 # YTMP1 = W[-2] << 61 {BABA}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 61 {BABA}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = s1 = (W[-2] ror 19) ^
+ # (W[-2] ror 61) ^ (W[-2] >> 6) {BABA}
+
+ # Add sigma1 to the other compunents to get w[16] and w[17]
+ vpaddq YTMP4, Y_0, Y_0 # Y_0 = {W[1], W[0], W[1], W[0]}
+
+ # Calculate sigma1 for w[18] and w[19] for upper 128 bit lane
+ vpsrlq $6, Y_0, YTMP4 # YTMP4 = W[-2] >> 6 {DC--}
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ add 2*8+frame_XFER(%rsp), h # h = k + w + h # --
+
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ or c, y3 # y3 = a|c # MAJA
+ mov f, y2 # y2 = f # CH
+ xor g, y2 # y2 = f^g # CH
+
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+################################### RND N + 3 #########################################
+
+ vpsrlq $19, Y_0, YTMP3 # YTMP3 = W[-2] >> 19 {DC--}
+ vpsllq $(64-19), Y_0, YTMP1 # YTMP1 = W[-2] << 19 {DC--}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 19 {DC--}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = W[-2] ror 19 ^ W[-2] >> 6 {DC--}
+ vpsrlq $61, Y_0, YTMP3 # YTMP3 = W[-2] >> 61 {DC--}
+ vpsllq $(64-61), Y_0, YTMP1 # YTMP1 = W[-2] << 61 {DC--}
+ vpor YTMP1, YTMP3, YTMP3 # YTMP3 = W[-2] ror 61 {DC--}
+ vpxor YTMP3, YTMP4, YTMP4 # YTMP4 = s1 = (W[-2] ror 19) ^
+ # (W[-2] ror 61) ^ (W[-2] >> 6) {DC--}
+
+ # Add the sigma0 + w[t-7] + w[t-16] for w[18] and w[19]
+ # to newly calculated sigma1 to get w[18] and w[19]
+ vpaddq YTMP4, YTMP0, YTMP2 # YTMP2 = {W[3], W[2], --, --}
+
+ # Form w[19, w[18], w17], w[16]
+ vpblendd $0xF0, YTMP2, Y_0, Y_0 # Y_0 = {W[3], W[2], W[1], W[0]}
+
+ mov a, y3 # y3 = a # MAJA
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ add 3*8+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+
+ mov f, y2 # y2 = f # CH
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ xor g, y2 # y2 = f^g # CH
+
+
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add h, d # d = k + w + h + d # --
+ and b, y3 # y3 = (a|c)&b # MAJA
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ add y0, y2 # y2 = S1 + CH # --
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and c, T1 # T1 = a&c # MAJB
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+
+ add y1, h # h = k + w + h + S0 # --
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+ rotate_Ys
+.endm
+
+.macro DO_4ROUNDS
+
+################################### RND N + 0 #########################################
+
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 1 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*1+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 2 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*2+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ RotateState
+
+################################### RND N + 3 #########################################
+
+ add y2, old_h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+ mov f, y2 # y2 = f # CH
+ rorx $41, e, y0 # y0 = e >> 41 # S1A
+ rorx $18, e, y1 # y1 = e >> 18 # S1B
+ xor g, y2 # y2 = f^g # CH
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) # S1
+ rorx $14, e, y1 # y1 = (e >> 14) # S1
+ and e, y2 # y2 = (f^g)&e # CH
+ add y3, old_h # h = t1 + S0 + MAJ # --
+
+ xor y1, y0 # y0 = (e>>41) ^ (e>>18) ^ (e>>14) # S1
+ rorx $34, a, T1 # T1 = a >> 34 # S0B
+ xor g, y2 # y2 = CH = ((f^g)&e)^g # CH
+ rorx $39, a, y1 # y1 = a >> 39 # S0A
+ mov a, y3 # y3 = a # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) # S0
+ rorx $28, a, T1 # T1 = (a >> 28) # S0
+ add 8*3+frame_XFER(%rsp), h # h = k + w + h # --
+ or c, y3 # y3 = a|c # MAJA
+
+ xor T1, y1 # y1 = (a>>39) ^ (a>>34) ^ (a>>28) # S0
+ mov a, T1 # T1 = a # MAJB
+ and b, y3 # y3 = (a|c)&b # MAJA
+ and c, T1 # T1 = a&c # MAJB
+ add y0, y2 # y2 = S1 + CH # --
+
+
+ add h, d # d = k + w + h + d # --
+ or T1, y3 # y3 = MAJ = (a|c)&b)|(a&c) # MAJ
+ add y1, h # h = k + w + h + S0 # --
+
+ add y2, d # d = k + w + h + d + S1 + CH = d + t1 # --
+
+ add y2, h # h = k + w + h + S0 + S1 + CH = t1 + S0# --
+
+ add y3, h # h = t1 + S0 + MAJ # --
+
+ RotateState
+
+.endm
+
+########################################################################
+# void sha512_transform_rorx(const void* M, void* D, uint64_t L)#
+# Purpose: Updates the SHA512 digest stored at D with the message stored in M.
+# The size of the message pointed to by M must be an integer multiple of SHA512
+# message blocks.
+# L is the message length in SHA512 blocks
+########################################################################
+ENTRY(sha512_transform_rorx)
+ # Allocate Stack Space
+ mov %rsp, %rax
+ sub $frame_size, %rsp
+ and $~(0x20 - 1), %rsp
+ mov %rax, frame_RSPSAVE(%rsp)
+
+ # Save GPRs
+ mov %rbp, frame_GPRSAVE(%rsp)
+ mov %rbx, 8*1+frame_GPRSAVE(%rsp)
+ mov %r12, 8*2+frame_GPRSAVE(%rsp)
+ mov %r13, 8*3+frame_GPRSAVE(%rsp)
+ mov %r14, 8*4+frame_GPRSAVE(%rsp)
+ mov %r15, 8*5+frame_GPRSAVE(%rsp)
+
+ shl $7, NUM_BLKS # convert to bytes
+ jz done_hash
+ add INP, NUM_BLKS # pointer to end of data
+ mov NUM_BLKS, frame_INPEND(%rsp)
+
+ ## load initial digest
+ mov 8*0(CTX),a
+ mov 8*1(CTX),b
+ mov 8*2(CTX),c
+ mov 8*3(CTX),d
+ mov 8*4(CTX),e
+ mov 8*5(CTX),f
+ mov 8*6(CTX),g
+ mov 8*7(CTX),h
+
+ vmovdqa PSHUFFLE_BYTE_FLIP_MASK(%rip), BYTE_FLIP_MASK
+
+loop0:
+ lea K512(%rip), TBL
+
+ ## byte swap first 16 dwords
+ COPY_YMM_AND_BSWAP Y_0, (INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_1, 1*32(INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_2, 2*32(INP), BYTE_FLIP_MASK
+ COPY_YMM_AND_BSWAP Y_3, 3*32(INP), BYTE_FLIP_MASK
+
+ mov INP, frame_INP(%rsp)
+
+ ## schedule 64 input dwords, by doing 12 rounds of 4 each
+ movq $4, frame_SRND(%rsp)
+
+.align 16
+loop1:
+ vpaddq (TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 1*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 2*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ FOUR_ROUNDS_AND_SCHED
+
+ vpaddq 3*32(TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ add $(4*32), TBL
+ FOUR_ROUNDS_AND_SCHED
+
+ subq $1, frame_SRND(%rsp)
+ jne loop1
+
+ movq $2, frame_SRND(%rsp)
+loop2:
+ vpaddq (TBL), Y_0, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ DO_4ROUNDS
+ vpaddq 1*32(TBL), Y_1, XFER
+ vmovdqa XFER, frame_XFER(%rsp)
+ add $(2*32), TBL
+ DO_4ROUNDS
+
+ vmovdqa Y_2, Y_0
+ vmovdqa Y_3, Y_1
+
+ subq $1, frame_SRND(%rsp)
+ jne loop2
+
+ addm 8*0(CTX),a
+ addm 8*1(CTX),b
+ addm 8*2(CTX),c
+ addm 8*3(CTX),d
+ addm 8*4(CTX),e
+ addm 8*5(CTX),f
+ addm 8*6(CTX),g
+ addm 8*7(CTX),h
+
+ mov frame_INP(%rsp), INP
+ add $128, INP
+ cmp frame_INPEND(%rsp), INP
+ jne loop0
+
+done_hash:
+
+# Restore GPRs
+ mov frame_GPRSAVE(%rsp) ,%rbp
+ mov 8*1+frame_GPRSAVE(%rsp) ,%rbx
+ mov 8*2+frame_GPRSAVE(%rsp) ,%r12
+ mov 8*3+frame_GPRSAVE(%rsp) ,%r13
+ mov 8*4+frame_GPRSAVE(%rsp) ,%r14
+ mov 8*5+frame_GPRSAVE(%rsp) ,%r15
+
+ # Restore Stack Pointer
+ mov frame_RSPSAVE(%rsp), %rsp
+ ret
+ENDPROC(sha512_transform_rorx)
+
+########################################################################
+### Binary Data
+
+.data
+
+.align 64
+# K[t] used in SHA512 hashing
+K512:
+ .quad 0x428a2f98d728ae22,0x7137449123ef65cd
+ .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
+ .quad 0x3956c25bf348b538,0x59f111f1b605d019
+ .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
+ .quad 0xd807aa98a3030242,0x12835b0145706fbe
+ .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
+ .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
+ .quad 0x9bdc06a725c71235,0xc19bf174cf692694
+ .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
+ .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
+ .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
+ .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
+ .quad 0x983e5152ee66dfab,0xa831c66d2db43210
+ .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
+ .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
+ .quad 0x06ca6351e003826f,0x142929670a0e6e70
+ .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
+ .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
+ .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
+ .quad 0x81c2c92e47edaee6,0x92722c851482353b
+ .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
+ .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
+ .quad 0xd192e819d6ef5218,0xd69906245565a910
+ .quad 0xf40e35855771202a,0x106aa07032bbd1b8
+ .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
+ .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
+ .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
+ .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
+ .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
+ .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
+ .quad 0x90befffa23631e28,0xa4506cebde82bde9
+ .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
+ .quad 0xca273eceea26619c,0xd186b8c721c0c207
+ .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
+ .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
+ .quad 0x113f9804bef90dae,0x1b710b35131c471b
+ .quad 0x28db77f523047d84,0x32caab7b40c72493
+ .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
+ .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
+ .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
+
+.align 32
+
+# Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
+PSHUFFLE_BYTE_FLIP_MASK:
+ .octa 0x08090a0b0c0d0e0f0001020304050607
+ .octa 0x18191a1b1c1d1e1f1011121314151617
+
+MASK_YMM_LO:
+ .octa 0x00000000000000000000000000000000
+ .octa 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
+#endif
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 10/10] Create module providing optimized SHA512 routines using SSSE3, AVX or AVX2 instructions.
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
@ 2013-03-26 21:00 ` Tim Chen
2013-03-26 20:58 ` Tim Chen
` (9 subsequent siblings)
10 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 21:00 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
We added glue code and config options to create crypto
module that uses SSE/AVX/AVX2 optimized SHA512 x86_64 assembly routines.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha512_ssse3_glue.c | 282 ++++++++++++++++++++++++++++++++++++
crypto/Kconfig | 11 ++
3 files changed, 295 insertions(+)
create mode 100644 arch/x86/crypto/sha512_ssse3_glue.c
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index b45f599..3b3d1aa 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -29,6 +29,7 @@ obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
obj-$(CONFIG_CRYPTO_CRC32_PCLMUL) += crc32-pclmul.o
obj-$(CONFIG_CRYPTO_SHA256_SSSE3) += sha256-ssse3.o
+obj-$(CONFIG_CRYPTO_SHA512_SSSE3) += sha512-ssse3.o
aes-i586-y := aes-i586-asm_32.o aes_glue.o
twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
@@ -56,3 +57,4 @@ crc32c-intel-y := crc32c-intel_glue.o
crc32c-intel-$(CONFIG_64BIT) += crc32c-pcl-intel-asm_64.o
crc32-pclmul-y := crc32-pclmul_asm.o crc32-pclmul_glue.o
sha256-ssse3-y := sha256-ssse3-asm.o sha256-avx-asm.o sha256-avx2-asm.o sha256_ssse3_glue.o
+sha512-ssse3-y := sha512-ssse3-asm.o sha512-avx-asm.o sha512-avx2-asm.o sha512_ssse3_glue.o
diff --git a/arch/x86/crypto/sha512_ssse3_glue.c b/arch/x86/crypto/sha512_ssse3_glue.c
new file mode 100644
index 0000000..295f790
--- /dev/null
+++ b/arch/x86/crypto/sha512_ssse3_glue.c
@@ -0,0 +1,282 @@
+/*
+ * Cryptographic API.
+ *
+ * Glue code for the SHA512 Secure Hash Algorithm assembler
+ * implementation using supplemental SSE3 / AVX / AVX2 instructions.
+ *
+ * This file is based on sha512_generic.c
+ *
+ * Copyright (C) 2013 Intel Corporation
+ * Author: Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * 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.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <crypto/sha.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+
+#include <linux/string.h>
+
+asmlinkage void sha512_transform_ssse3(const char *data, u64 *digest,
+ u64 rounds);
+#ifdef CONFIG_AS_AVX
+asmlinkage void sha512_transform_avx(const char *data, u64 *digest,
+ u64 rounds);
+#endif
+#ifdef CONFIG_AS_AVX2
+asmlinkage void sha512_transform_rorx(const char *data, u64 *digest,
+ u64 rounds);
+#endif
+
+asmlinkage void (*sha512_transform_asm)(const char *, u64 *, u64);
+
+
+static int sha512_ssse3_init(struct shash_desc *desc)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ sctx->state[0] = SHA512_H0;
+ sctx->state[1] = SHA512_H1;
+ sctx->state[2] = SHA512_H2;
+ sctx->state[3] = SHA512_H3;
+ sctx->state[4] = SHA512_H4;
+ sctx->state[5] = SHA512_H5;
+ sctx->state[6] = SHA512_H6;
+ sctx->state[7] = SHA512_H7;
+ sctx->count[0] = sctx->count[1] = 0;
+
+ return 0;
+}
+
+static int __sha512_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len, unsigned int partial)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int done = 0;
+
+ sctx->count[0] += len;
+ if (sctx->count[0] < len)
+ sctx->count[1]++;
+
+ if (partial) {
+ done = SHA512_BLOCK_SIZE - partial;
+ memcpy(sctx->buf + partial, data, done);
+ sha512_transform_asm(sctx->buf, sctx->state, 1);
+ }
+
+ if (len - done >= SHA512_BLOCK_SIZE) {
+ const unsigned int rounds = (len - done) / SHA512_BLOCK_SIZE;
+
+ sha512_transform_asm(data + done, sctx->state, (u64) rounds);
+
+ done += rounds * SHA512_BLOCK_SIZE;
+ }
+
+ memcpy(sctx->buf, data + done, len - done);
+
+ return 0;
+}
+
+static int sha512_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int partial = sctx->count[0] % SHA512_BLOCK_SIZE;
+ int res;
+
+ /* Handle the fast case right here */
+ if (partial + len < SHA512_BLOCK_SIZE) {
+ sctx->count[0] += len;
+ if (sctx->count[0] < len)
+ sctx->count[1]++;
+ memcpy(sctx->buf + partial, data, len);
+
+ return 0;
+ }
+
+ if (!irq_fpu_usable()) {
+ res = crypto_sha512_update(desc, data, len);
+ } else {
+ kernel_fpu_begin();
+ res = __sha512_ssse3_update(desc, data, len, partial);
+ kernel_fpu_end();
+ }
+
+ return res;
+}
+
+
+/* Add padding and return the message digest. */
+static int sha512_ssse3_final(struct shash_desc *desc, u8 *out)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int i, index, padlen;
+ __be64 *dst = (__be64 *)out;
+ __be64 bits[2];
+ static const u8 padding[SHA512_BLOCK_SIZE] = { 0x80, };
+
+ /* save number of bits */
+ bits[1] = cpu_to_be64(sctx->count[0] << 3);
+ bits[0] = cpu_to_be64(sctx->count[1] << 3) | sctx->count[0] >> 61;
+
+ /* Pad out to 112 mod 128 and append length */
+ index = sctx->count[0] & 0x7f;
+ padlen = (index < 112) ? (112 - index) : ((128+112) - index);
+
+ if (!irq_fpu_usable()) {
+ crypto_sha512_update(desc, padding, padlen);
+ crypto_sha512_update(desc, (const u8 *)&bits, sizeof(bits));
+ } else {
+ kernel_fpu_begin();
+ /* We need to fill a whole block for __sha512_ssse3_update() */
+ if (padlen <= 112) {
+ sctx->count[0] += padlen;
+ if (sctx->count[0] < padlen)
+ sctx->count[1]++;
+ memcpy(sctx->buf + index, padding, padlen);
+ } else {
+ __sha512_ssse3_update(desc, padding, padlen, index);
+ }
+ __sha512_ssse3_update(desc, (const u8 *)&bits,
+ sizeof(bits), 112);
+ kernel_fpu_end();
+ }
+
+ /* Store state in digest */
+ for (i = 0; i < 8; i++)
+ dst[i] = cpu_to_be64(sctx->state[i]);
+
+ /* Wipe context */
+ memset(sctx, 0, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha512_ssse3_export(struct shash_desc *desc, void *out)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(out, sctx, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha512_ssse3_import(struct shash_desc *desc, const void *in)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(sctx, in, sizeof(*sctx));
+
+ return 0;
+}
+
+static struct shash_alg alg = {
+ .digestsize = SHA512_DIGEST_SIZE,
+ .init = sha512_ssse3_init,
+ .update = sha512_ssse3_update,
+ .final = sha512_ssse3_final,
+ .export = sha512_ssse3_export,
+ .import = sha512_ssse3_import,
+ .descsize = sizeof(struct sha512_state),
+ .statesize = sizeof(struct sha512_state),
+ .base = {
+ .cra_name = "sha512",
+ .cra_driver_name = "sha512-ssse3",
+ .cra_priority = 150,
+ .cra_flags = CRYPTO_ALG_TYPE_SHASH,
+ .cra_blocksize = SHA512_BLOCK_SIZE,
+ .cra_module = THIS_MODULE,
+ }
+};
+
+#ifdef CONFIG_AS_AVX
+static bool __init avx_usable(void)
+{
+ u64 xcr0;
+
+ if (!cpu_has_avx || !cpu_has_osxsave)
+ return false;
+
+ xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
+ if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
+ pr_info("AVX detected but unusable.\n");
+
+ return false;
+ }
+
+ return true;
+}
+#endif
+
+static int __init sha512_ssse3_mod_init(void)
+{
+ /* test for SSE3 first */
+ if (cpu_has_ssse3)
+ sha512_transform_asm = sha512_transform_ssse3;
+
+#ifdef CONFIG_AS_AVX
+ /* allow AVX to override SSSE3, it's a little faster */
+ if (avx_usable()) {
+#ifdef CONFIG_AS_AVX2
+ if (boot_cpu_has(X86_FEATURE_AVX2))
+ sha512_transform_asm = sha512_transform_rorx;
+ else
+#endif
+ sha512_transform_asm = sha512_transform_avx;
+ }
+#endif
+
+ if (sha512_transform_asm) {
+#ifdef CONFIG_AS_AVX
+ if (sha512_transform_asm == sha512_transform_avx)
+ pr_info("Using AVX optimized SHA-512 implementation\n");
+#ifdef CONFIG_AS_AVX2
+ else if (sha512_transform_asm == sha512_transform_rorx)
+ pr_info("Using AVX2 optimized SHA-512 implementation\n");
+#endif
+ else
+#endif
+ pr_info("Using SSSE3 optimized SHA-512 implementation\n");
+ return crypto_register_shash(&alg);
+ }
+ pr_info("Neither AVX nor SSSE3 is available/usable.\n");
+
+ return -ENODEV;
+}
+
+static void __exit sha512_ssse3_mod_fini(void)
+{
+ crypto_unregister_shash(&alg);
+}
+
+module_init(sha512_ssse3_mod_init);
+module_exit(sha512_ssse3_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA512 Secure Hash Algorithm, Supplemental SSE3 accelerated");
+
+MODULE_ALIAS("sha512");
diff --git a/crypto/Kconfig b/crypto/Kconfig
index 8064ef1..a654b13 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -481,6 +481,17 @@ config CRYPTO_SHA256_SSSE3
Extensions version 1 (AVX1), or Advanced Vector Extensions
version 2 (AVX2) instructions, when available.
+config CRYPTO_SHA512_SSSE3
+ tristate "SHA512 digest algorithm (SSSE3/AVX/AVX2)"
+ depends on X86 && 64BIT
+ select CRYPTO_SHA512
+ select CRYPTO_HASH
+ help
+ SHA-512 secure hash standard (DFIPS 180-2) implemented
+ using Supplemental SSE3 (SSSE3) instructions, or Advanced Vector
+ Extensions version 1 (AVX1), or Advanced Vector Extensions
+ version 2 (AVX2) instructions, when available.
+
config CRYPTO_SHA1_SPARC64
tristate "SHA1 digest algorithm (SPARC64)"
depends on SPARC64
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH v2 10/10] Create module providing optimized SHA512 routines using SSSE3, AVX or AVX2 instructions.
@ 2013-03-26 21:00 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-03-26 21:00 UTC (permalink / raw)
To: Herbert Xu, H. Peter Anvin, David S.Miller
Cc: Jussi Kivilinna, Jim Kukunas, Kirk Yap, David Cote,
James Guilford, Wajdi Feghali, linux-kernel, linux-crypto
We added glue code and config options to create crypto
module that uses SSE/AVX/AVX2 optimized SHA512 x86_64 assembly routines.
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/Makefile | 2 +
arch/x86/crypto/sha512_ssse3_glue.c | 282 ++++++++++++++++++++++++++++++++++++
crypto/Kconfig | 11 ++
3 files changed, 295 insertions(+)
create mode 100644 arch/x86/crypto/sha512_ssse3_glue.c
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index b45f599..3b3d1aa 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -29,6 +29,7 @@ obj-$(CONFIG_CRYPTO_CRC32C_INTEL) += crc32c-intel.o
obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
obj-$(CONFIG_CRYPTO_CRC32_PCLMUL) += crc32-pclmul.o
obj-$(CONFIG_CRYPTO_SHA256_SSSE3) += sha256-ssse3.o
+obj-$(CONFIG_CRYPTO_SHA512_SSSE3) += sha512-ssse3.o
aes-i586-y := aes-i586-asm_32.o aes_glue.o
twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
@@ -56,3 +57,4 @@ crc32c-intel-y := crc32c-intel_glue.o
crc32c-intel-$(CONFIG_64BIT) += crc32c-pcl-intel-asm_64.o
crc32-pclmul-y := crc32-pclmul_asm.o crc32-pclmul_glue.o
sha256-ssse3-y := sha256-ssse3-asm.o sha256-avx-asm.o sha256-avx2-asm.o sha256_ssse3_glue.o
+sha512-ssse3-y := sha512-ssse3-asm.o sha512-avx-asm.o sha512-avx2-asm.o sha512_ssse3_glue.o
diff --git a/arch/x86/crypto/sha512_ssse3_glue.c b/arch/x86/crypto/sha512_ssse3_glue.c
new file mode 100644
index 0000000..295f790
--- /dev/null
+++ b/arch/x86/crypto/sha512_ssse3_glue.c
@@ -0,0 +1,282 @@
+/*
+ * Cryptographic API.
+ *
+ * Glue code for the SHA512 Secure Hash Algorithm assembler
+ * implementation using supplemental SSE3 / AVX / AVX2 instructions.
+ *
+ * This file is based on sha512_generic.c
+ *
+ * Copyright (C) 2013 Intel Corporation
+ * Author: Tim Chen <tim.c.chen@linux.intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * 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.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mm.h>
+#include <linux/cryptohash.h>
+#include <linux/types.h>
+#include <crypto/sha.h>
+#include <asm/byteorder.h>
+#include <asm/i387.h>
+#include <asm/xcr.h>
+#include <asm/xsave.h>
+
+#include <linux/string.h>
+
+asmlinkage void sha512_transform_ssse3(const char *data, u64 *digest,
+ u64 rounds);
+#ifdef CONFIG_AS_AVX
+asmlinkage void sha512_transform_avx(const char *data, u64 *digest,
+ u64 rounds);
+#endif
+#ifdef CONFIG_AS_AVX2
+asmlinkage void sha512_transform_rorx(const char *data, u64 *digest,
+ u64 rounds);
+#endif
+
+asmlinkage void (*sha512_transform_asm)(const char *, u64 *, u64);
+
+
+static int sha512_ssse3_init(struct shash_desc *desc)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ sctx->state[0] = SHA512_H0;
+ sctx->state[1] = SHA512_H1;
+ sctx->state[2] = SHA512_H2;
+ sctx->state[3] = SHA512_H3;
+ sctx->state[4] = SHA512_H4;
+ sctx->state[5] = SHA512_H5;
+ sctx->state[6] = SHA512_H6;
+ sctx->state[7] = SHA512_H7;
+ sctx->count[0] = sctx->count[1] = 0;
+
+ return 0;
+}
+
+static int __sha512_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len, unsigned int partial)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int done = 0;
+
+ sctx->count[0] += len;
+ if (sctx->count[0] < len)
+ sctx->count[1]++;
+
+ if (partial) {
+ done = SHA512_BLOCK_SIZE - partial;
+ memcpy(sctx->buf + partial, data, done);
+ sha512_transform_asm(sctx->buf, sctx->state, 1);
+ }
+
+ if (len - done >= SHA512_BLOCK_SIZE) {
+ const unsigned int rounds = (len - done) / SHA512_BLOCK_SIZE;
+
+ sha512_transform_asm(data + done, sctx->state, (u64) rounds);
+
+ done += rounds * SHA512_BLOCK_SIZE;
+ }
+
+ memcpy(sctx->buf, data + done, len - done);
+
+ return 0;
+}
+
+static int sha512_ssse3_update(struct shash_desc *desc, const u8 *data,
+ unsigned int len)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int partial = sctx->count[0] % SHA512_BLOCK_SIZE;
+ int res;
+
+ /* Handle the fast case right here */
+ if (partial + len < SHA512_BLOCK_SIZE) {
+ sctx->count[0] += len;
+ if (sctx->count[0] < len)
+ sctx->count[1]++;
+ memcpy(sctx->buf + partial, data, len);
+
+ return 0;
+ }
+
+ if (!irq_fpu_usable()) {
+ res = crypto_sha512_update(desc, data, len);
+ } else {
+ kernel_fpu_begin();
+ res = __sha512_ssse3_update(desc, data, len, partial);
+ kernel_fpu_end();
+ }
+
+ return res;
+}
+
+
+/* Add padding and return the message digest. */
+static int sha512_ssse3_final(struct shash_desc *desc, u8 *out)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+ unsigned int i, index, padlen;
+ __be64 *dst = (__be64 *)out;
+ __be64 bits[2];
+ static const u8 padding[SHA512_BLOCK_SIZE] = { 0x80, };
+
+ /* save number of bits */
+ bits[1] = cpu_to_be64(sctx->count[0] << 3);
+ bits[0] = cpu_to_be64(sctx->count[1] << 3) | sctx->count[0] >> 61;
+
+ /* Pad out to 112 mod 128 and append length */
+ index = sctx->count[0] & 0x7f;
+ padlen = (index < 112) ? (112 - index) : ((128+112) - index);
+
+ if (!irq_fpu_usable()) {
+ crypto_sha512_update(desc, padding, padlen);
+ crypto_sha512_update(desc, (const u8 *)&bits, sizeof(bits));
+ } else {
+ kernel_fpu_begin();
+ /* We need to fill a whole block for __sha512_ssse3_update() */
+ if (padlen <= 112) {
+ sctx->count[0] += padlen;
+ if (sctx->count[0] < padlen)
+ sctx->count[1]++;
+ memcpy(sctx->buf + index, padding, padlen);
+ } else {
+ __sha512_ssse3_update(desc, padding, padlen, index);
+ }
+ __sha512_ssse3_update(desc, (const u8 *)&bits,
+ sizeof(bits), 112);
+ kernel_fpu_end();
+ }
+
+ /* Store state in digest */
+ for (i = 0; i < 8; i++)
+ dst[i] = cpu_to_be64(sctx->state[i]);
+
+ /* Wipe context */
+ memset(sctx, 0, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha512_ssse3_export(struct shash_desc *desc, void *out)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(out, sctx, sizeof(*sctx));
+
+ return 0;
+}
+
+static int sha512_ssse3_import(struct shash_desc *desc, const void *in)
+{
+ struct sha512_state *sctx = shash_desc_ctx(desc);
+
+ memcpy(sctx, in, sizeof(*sctx));
+
+ return 0;
+}
+
+static struct shash_alg alg = {
+ .digestsize = SHA512_DIGEST_SIZE,
+ .init = sha512_ssse3_init,
+ .update = sha512_ssse3_update,
+ .final = sha512_ssse3_final,
+ .export = sha512_ssse3_export,
+ .import = sha512_ssse3_import,
+ .descsize = sizeof(struct sha512_state),
+ .statesize = sizeof(struct sha512_state),
+ .base = {
+ .cra_name = "sha512",
+ .cra_driver_name = "sha512-ssse3",
+ .cra_priority = 150,
+ .cra_flags = CRYPTO_ALG_TYPE_SHASH,
+ .cra_blocksize = SHA512_BLOCK_SIZE,
+ .cra_module = THIS_MODULE,
+ }
+};
+
+#ifdef CONFIG_AS_AVX
+static bool __init avx_usable(void)
+{
+ u64 xcr0;
+
+ if (!cpu_has_avx || !cpu_has_osxsave)
+ return false;
+
+ xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
+ if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
+ pr_info("AVX detected but unusable.\n");
+
+ return false;
+ }
+
+ return true;
+}
+#endif
+
+static int __init sha512_ssse3_mod_init(void)
+{
+ /* test for SSE3 first */
+ if (cpu_has_ssse3)
+ sha512_transform_asm = sha512_transform_ssse3;
+
+#ifdef CONFIG_AS_AVX
+ /* allow AVX to override SSSE3, it's a little faster */
+ if (avx_usable()) {
+#ifdef CONFIG_AS_AVX2
+ if (boot_cpu_has(X86_FEATURE_AVX2))
+ sha512_transform_asm = sha512_transform_rorx;
+ else
+#endif
+ sha512_transform_asm = sha512_transform_avx;
+ }
+#endif
+
+ if (sha512_transform_asm) {
+#ifdef CONFIG_AS_AVX
+ if (sha512_transform_asm == sha512_transform_avx)
+ pr_info("Using AVX optimized SHA-512 implementation\n");
+#ifdef CONFIG_AS_AVX2
+ else if (sha512_transform_asm == sha512_transform_rorx)
+ pr_info("Using AVX2 optimized SHA-512 implementation\n");
+#endif
+ else
+#endif
+ pr_info("Using SSSE3 optimized SHA-512 implementation\n");
+ return crypto_register_shash(&alg);
+ }
+ pr_info("Neither AVX nor SSSE3 is available/usable.\n");
+
+ return -ENODEV;
+}
+
+static void __exit sha512_ssse3_mod_fini(void)
+{
+ crypto_unregister_shash(&alg);
+}
+
+module_init(sha512_ssse3_mod_init);
+module_exit(sha512_ssse3_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("SHA512 Secure Hash Algorithm, Supplemental SSE3 accelerated");
+
+MODULE_ALIAS("sha512");
diff --git a/crypto/Kconfig b/crypto/Kconfig
index 8064ef1..a654b13 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -481,6 +481,17 @@ config CRYPTO_SHA256_SSSE3
Extensions version 1 (AVX1), or Advanced Vector Extensions
version 2 (AVX2) instructions, when available.
+config CRYPTO_SHA512_SSSE3
+ tristate "SHA512 digest algorithm (SSSE3/AVX/AVX2)"
+ depends on X86 && 64BIT
+ select CRYPTO_SHA512
+ select CRYPTO_HASH
+ help
+ SHA-512 secure hash standard (DFIPS 180-2) implemented
+ using Supplemental SSE3 (SSSE3) instructions, or Advanced Vector
+ Extensions version 1 (AVX1), or Advanced Vector Extensions
+ version 2 (AVX2) instructions, when available.
+
config CRYPTO_SHA1_SPARC64
tristate "SHA1 digest algorithm (SPARC64)"
depends on SPARC64
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* Re: [PATCH v2 00/10] Optimize SHA256 and SHA512 for Intel x86_64 with SSSE3, AVX or AVX2 instructions
2013-03-26 20:58 ` Tim Chen
@ 2013-04-03 1:53 ` Herbert Xu
-1 siblings, 0 replies; 32+ messages in thread
From: Herbert Xu @ 2013-04-03 1:53 UTC (permalink / raw)
To: Tim Chen
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Tue, Mar 26, 2013 at 01:58:39PM -0700, Tim Chen wrote:
> Herbert,
>
> The following patch series provides optimized SHA256 and SHA512 routines
> using the SSSE3, AVX or AVX2 instructions on x86_64 for Intel cpus.
> Depending on cpu capabilities, speedup between 40% to 70% or more can be achieved
> over the generic SHA256 and SHA512 routines.
All applied. Thanks!
--
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH v2 00/10] Optimize SHA256 and SHA512 for Intel x86_64 with SSSE3, AVX or AVX2 instructions
@ 2013-04-03 1:53 ` Herbert Xu
0 siblings, 0 replies; 32+ messages in thread
From: Herbert Xu @ 2013-04-03 1:53 UTC (permalink / raw)
To: Tim Chen
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Tue, Mar 26, 2013 at 01:58:39PM -0700, Tim Chen wrote:
> Herbert,
>
> The following patch series provides optimized SHA256 and SHA512 routines
> using the SSSE3, AVX or AVX2 instructions on x86_64 for Intel cpus.
> Depending on cpu capabilities, speedup between 40% to 70% or more can be achieved
> over the generic SHA256 and SHA512 routines.
All applied. Thanks!
--
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
^ permalink raw reply [flat|nested] 32+ messages in thread
* [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
2013-03-26 20:58 ` Tim Chen
@ 2013-04-19 20:25 ` Tim Chen
-1 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-04-19 20:25 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
Herbert,
This is a follow on patch to the optimized sha256 and sha512 patch series that's just
merged into the crypto-dev. Let me know if you prefer me to respin the
patch series.
This patch corrects the prototype of sha256_transform_asm and
sha512_transform_asm function pointer declaration to static. It also
fixes a typo in sha512_ssse3_final function that affects the computation
of upper 64 bits of the buffer size.
Thanks.
Tim
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256_ssse3_glue.c | 2 +-
arch/x86/crypto/sha512_ssse3_glue.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/x86/crypto/sha256_ssse3_glue.c b/arch/x86/crypto/sha256_ssse3_glue.c
index fa65453..8a0b711 100644
--- a/arch/x86/crypto/sha256_ssse3_glue.c
+++ b/arch/x86/crypto/sha256_ssse3_glue.c
@@ -53,7 +53,7 @@ asmlinkage void sha256_transform_rorx(const char *data, u32 *digest,
u64 rounds);
#endif
-asmlinkage void (*sha256_transform_asm)(const char *, u32 *, u64);
+static void (*sha256_transform_asm)(const char *, u32 *, u64);
static int sha256_ssse3_init(struct shash_desc *desc)
diff --git a/arch/x86/crypto/sha512_ssse3_glue.c b/arch/x86/crypto/sha512_ssse3_glue.c
index 295f790..3c844f2 100644
--- a/arch/x86/crypto/sha512_ssse3_glue.c
+++ b/arch/x86/crypto/sha512_ssse3_glue.c
@@ -52,7 +52,7 @@ asmlinkage void sha512_transform_rorx(const char *data, u64 *digest,
u64 rounds);
#endif
-asmlinkage void (*sha512_transform_asm)(const char *, u64 *, u64);
+static void (*sha512_transform_asm)(const char *, u64 *, u64);
static int sha512_ssse3_init(struct shash_desc *desc)
@@ -141,7 +141,7 @@ static int sha512_ssse3_final(struct shash_desc *desc, u8 *out)
/* save number of bits */
bits[1] = cpu_to_be64(sctx->count[0] << 3);
- bits[0] = cpu_to_be64(sctx->count[1] << 3) | sctx->count[0] >> 61;
+ bits[0] = cpu_to_be64(sctx->count[1] << 3 | sctx->count[0] >> 61);
/* Pad out to 112 mod 128 and append length */
index = sctx->count[0] & 0x7f;
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
@ 2013-04-19 20:25 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-04-19 20:25 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
Herbert,
This is a follow on patch to the optimized sha256 and sha512 patch series that's just
merged into the crypto-dev. Let me know if you prefer me to respin the
patch series.
This patch corrects the prototype of sha256_transform_asm and
sha512_transform_asm function pointer declaration to static. It also
fixes a typo in sha512_ssse3_final function that affects the computation
of upper 64 bits of the buffer size.
Thanks.
Tim
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
arch/x86/crypto/sha256_ssse3_glue.c | 2 +-
arch/x86/crypto/sha512_ssse3_glue.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/x86/crypto/sha256_ssse3_glue.c b/arch/x86/crypto/sha256_ssse3_glue.c
index fa65453..8a0b711 100644
--- a/arch/x86/crypto/sha256_ssse3_glue.c
+++ b/arch/x86/crypto/sha256_ssse3_glue.c
@@ -53,7 +53,7 @@ asmlinkage void sha256_transform_rorx(const char *data, u32 *digest,
u64 rounds);
#endif
-asmlinkage void (*sha256_transform_asm)(const char *, u32 *, u64);
+static void (*sha256_transform_asm)(const char *, u32 *, u64);
static int sha256_ssse3_init(struct shash_desc *desc)
diff --git a/arch/x86/crypto/sha512_ssse3_glue.c b/arch/x86/crypto/sha512_ssse3_glue.c
index 295f790..3c844f2 100644
--- a/arch/x86/crypto/sha512_ssse3_glue.c
+++ b/arch/x86/crypto/sha512_ssse3_glue.c
@@ -52,7 +52,7 @@ asmlinkage void sha512_transform_rorx(const char *data, u64 *digest,
u64 rounds);
#endif
-asmlinkage void (*sha512_transform_asm)(const char *, u64 *, u64);
+static void (*sha512_transform_asm)(const char *, u64 *, u64);
static int sha512_ssse3_init(struct shash_desc *desc)
@@ -141,7 +141,7 @@ static int sha512_ssse3_final(struct shash_desc *desc, u8 *out)
/* save number of bits */
bits[1] = cpu_to_be64(sctx->count[0] << 3);
- bits[0] = cpu_to_be64(sctx->count[1] << 3) | sctx->count[0] >> 61;
+ bits[0] = cpu_to_be64(sctx->count[1] << 3 | sctx->count[0] >> 61);
/* Pad out to 112 mod 128 and append length */
index = sctx->count[0] & 0x7f;
--
1.7.11.7
^ permalink raw reply related [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
2013-04-19 20:25 ` Tim Chen
@ 2013-04-25 13:04 ` Herbert Xu
-1 siblings, 0 replies; 32+ messages in thread
From: Herbert Xu @ 2013-04-25 13:04 UTC (permalink / raw)
To: Tim Chen
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> Herbert,
>
> This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> merged into the crypto-dev. Let me know if you prefer me to respin the
> patch series.
>
> This patch corrects the prototype of sha256_transform_asm and
> sha512_transform_asm function pointer declaration to static. It also
> fixes a typo in sha512_ssse3_final function that affects the computation
> of upper 64 bits of the buffer size.
Thanks, I've folded this into the original patches in order to
maintain bisectability.
--
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
@ 2013-04-25 13:04 ` Herbert Xu
0 siblings, 0 replies; 32+ messages in thread
From: Herbert Xu @ 2013-04-25 13:04 UTC (permalink / raw)
To: Tim Chen
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> Herbert,
>
> This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> merged into the crypto-dev. Let me know if you prefer me to respin the
> patch series.
>
> This patch corrects the prototype of sha256_transform_asm and
> sha512_transform_asm function pointer declaration to static. It also
> fixes a typo in sha512_ssse3_final function that affects the computation
> of upper 64 bits of the buffer size.
Thanks, I've folded this into the original patches in order to
maintain bisectability.
--
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
2013-04-25 13:04 ` Herbert Xu
@ 2013-06-07 16:47 ` Tim Chen
-1 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-06-07 16:47 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Thu, 2013-04-25 at 21:04 +0800, Herbert Xu wrote:
> On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> > Herbert,
> >
> > This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> > merged into the crypto-dev. Let me know if you prefer me to respin the
> > patch series.
> >
> > This patch corrects the prototype of sha256_transform_asm and
> > sha512_transform_asm function pointer declaration to static. It also
> > fixes a typo in sha512_ssse3_final function that affects the computation
> > of upper 64 bits of the buffer size.
>
> Thanks, I've folded this into the original patches in order to
> maintain bisectability.
Hi Herbert,
Wonder if you had a chance to merge this patch? I think it is not in
the latest crypto-dev.
Thanks.
Tim
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
@ 2013-06-07 16:47 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-06-07 16:47 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Thu, 2013-04-25 at 21:04 +0800, Herbert Xu wrote:
> On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> > Herbert,
> >
> > This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> > merged into the crypto-dev. Let me know if you prefer me to respin the
> > patch series.
> >
> > This patch corrects the prototype of sha256_transform_asm and
> > sha512_transform_asm function pointer declaration to static. It also
> > fixes a typo in sha512_ssse3_final function that affects the computation
> > of upper 64 bits of the buffer size.
>
> Thanks, I've folded this into the original patches in order to
> maintain bisectability.
Hi Herbert,
Wonder if you had a chance to merge this patch? I think it is not in
the latest crypto-dev.
Thanks.
Tim
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
2013-06-07 16:47 ` Tim Chen
@ 2013-06-07 16:51 ` Tim Chen
-1 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-06-07 16:51 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Fri, 2013-06-07 at 09:47 -0700, Tim Chen wrote:
> On Thu, 2013-04-25 at 21:04 +0800, Herbert Xu wrote:
> > On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> > > Herbert,
> > >
> > > This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> > > merged into the crypto-dev. Let me know if you prefer me to respin the
> > > patch series.
> > >
> > > This patch corrects the prototype of sha256_transform_asm and
> > > sha512_transform_asm function pointer declaration to static. It also
> > > fixes a typo in sha512_ssse3_final function that affects the computation
> > > of upper 64 bits of the buffer size.
> >
> > Thanks, I've folded this into the original patches in order to
> > maintain bisectability.
>
> Hi Herbert,
>
> Wonder if you had a chance to merge this patch? I think it is not in
> the latest crypto-dev.
>
Actually it has been merged. The kbuild test robot was picking up an
old tree.
Sorry for the noise. Thanks.
Tim
^ permalink raw reply [flat|nested] 32+ messages in thread
* Re: [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm
@ 2013-06-07 16:51 ` Tim Chen
0 siblings, 0 replies; 32+ messages in thread
From: Tim Chen @ 2013-06-07 16:51 UTC (permalink / raw)
To: Herbert Xu
Cc: H. Peter Anvin, David S.Miller, Jussi Kivilinna, Jim Kukunas,
Kirk Yap, David Cote, James Guilford, Wajdi Feghali,
linux-kernel, linux-crypto
On Fri, 2013-06-07 at 09:47 -0700, Tim Chen wrote:
> On Thu, 2013-04-25 at 21:04 +0800, Herbert Xu wrote:
> > On Fri, Apr 19, 2013 at 01:25:12PM -0700, Tim Chen wrote:
> > > Herbert,
> > >
> > > This is a follow on patch to the optimized sha256 and sha512 patch series that's just
> > > merged into the crypto-dev. Let me know if you prefer me to respin the
> > > patch series.
> > >
> > > This patch corrects the prototype of sha256_transform_asm and
> > > sha512_transform_asm function pointer declaration to static. It also
> > > fixes a typo in sha512_ssse3_final function that affects the computation
> > > of upper 64 bits of the buffer size.
> >
> > Thanks, I've folded this into the original patches in order to
> > maintain bisectability.
>
> Hi Herbert,
>
> Wonder if you had a chance to merge this patch? I think it is not in
> the latest crypto-dev.
>
Actually it has been merged. The kbuild test robot was picking up an
old tree.
Sorry for the noise. Thanks.
Tim
^ permalink raw reply [flat|nested] 32+ messages in thread
end of thread, other threads:[~2013-06-07 16:52 UTC | newest]
Thread overview: 32+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
[not found] <cover.1364303479.git.tim.c.chen@linux.intel.com>
2013-03-26 20:58 ` [PATCH v2 00/10] Optimize SHA256 and SHA512 for Intel x86_64 with SSSE3, AVX or AVX2 instructions Tim Chen
2013-03-26 20:58 ` Tim Chen
2013-04-03 1:53 ` Herbert Xu
2013-04-03 1:53 ` Herbert Xu
2013-04-19 20:25 ` [PATCH] Fix prototype definitions of sha256_transform_asm, sha512_transform_asm Tim Chen
2013-04-19 20:25 ` Tim Chen
2013-04-25 13:04 ` Herbert Xu
2013-04-25 13:04 ` Herbert Xu
2013-06-07 16:47 ` Tim Chen
2013-06-07 16:47 ` Tim Chen
2013-06-07 16:51 ` Tim Chen
2013-06-07 16:51 ` Tim Chen
2013-03-26 20:58 ` [PATCH v2 01/10] Expose SHA256 generic routine to be callable externally Tim Chen
2013-03-26 20:58 ` Tim Chen
2013-03-26 20:58 ` [PATCH v2 02/10] Optimized sha256 x86_64 assembly routine using Supplemental SSE3 instructions Tim Chen
2013-03-26 20:58 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 03/10] Optimized sha256 x86_64 assembly routine with AVX instructions Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 04/10] Optimized sha256 x86_64 routine using AVX2's RORX instructions Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 05/10] Create module providing optimized SHA256 routines using SSSE3, AVX or AVX2 instructions Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 06/10] Expose generic sha512 routine to be callable from other modules Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 07/10] Optimized SHA512 x86_64 assembly routine using Supplemental SSE3 instructions Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 08/10] Optimized SHA512 x86_64 assembly routine using AVX instructions Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 20:59 ` [PATCH v2 09/10] Optimized SHA512 x86_64 assembly routine using AVX2 RORX instruction Tim Chen
2013-03-26 20:59 ` Tim Chen
2013-03-26 21:00 ` [PATCH v2 10/10] Create module providing optimized SHA512 routines using SSSE3, AVX or AVX2 instructions Tim Chen
2013-03-26 21:00 ` Tim Chen
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.