[openssl] master update
tomas at openssl.org
tomas at openssl.org
Thu Nov 11 09:59:06 UTC 2021
The branch master has been updated
via 8f9842fd03945d9484dcc9e5fab37dce7caa5f50 (commit)
from 90c311315c15a4fea895fd317d9c8fe801ba04a0 (commit)
- Log -----------------------------------------------------------------
commit 8f9842fd03945d9484dcc9e5fab37dce7caa5f50
Author: Anton Blanchard <anton at linux.ibm.com>
Date: Tue Mar 12 16:03:56 2019 +1100
sha/asm/keccak1600-ppc64.pl: Load data in 8 byte chunks on little endian
We currently load data byte by byte in order to byteswap it on big
endian. On little endian we can just do 8 byte loads.
A SHAKE128 benchmark runs 10% faster on POWER9 with this patch applied.
Reviewed-by: Paul Dale <pauli at openssl.org>
Reviewed-by: Tomas Mraz <tomas at openssl.org>
(Merged from https://github.com/openssl/openssl/pull/8455)
-----------------------------------------------------------------------
Summary of changes:
crypto/sha/asm/keccak1600-ppc64.pl | 69 +++++++++++++++++++++++---------------
1 file changed, 42 insertions(+), 27 deletions(-)
diff --git a/crypto/sha/asm/keccak1600-ppc64.pl b/crypto/sha/asm/keccak1600-ppc64.pl
index 83f8d8ef33..bff0d78585 100755
--- a/crypto/sha/asm/keccak1600-ppc64.pl
+++ b/crypto/sha/asm/keccak1600-ppc64.pl
@@ -51,6 +51,16 @@ if ($flavour =~ /64/) {
$PUSH ="std";
} else { die "nonsense $flavour"; }
+$LITTLE_ENDIAN = ($flavour=~/le$/) ? 1 : 0;
+
+if ($LITTLE_ENDIAN) {
+ $DWORD_LE_LOAD = "ldu r0,8(r3)";
+ $LE_LOAD_SIZE = "8";
+} else {
+ $DWORD_LE_LOAD = "bl dword_le_load";
+ $LE_LOAD_SIZE = "1";
+}
+
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@@ -384,7 +394,9 @@ KeccakF1600:
.byte 0,12,4,1,0x80,18,1,0
.long 0
.size KeccakF1600,.-KeccakF1600
-
+___
+if (!$LITTLE_ENDIAN) {
+$code.=<<___;
.type dword_le_load,\@function
.align 5
dword_le_load:
@@ -408,7 +420,10 @@ dword_le_load:
.byte 0,12,0x14,0,0,0,1,0
.long 0
.size dword_le_load,.-dword_le_load
+___
+}
+$code.=<<___;
.globl SHA3_absorb
.type SHA3_absorb,\@function
.align 5
@@ -436,7 +451,7 @@ SHA3_absorb:
$PUSH r0,`$FRAME+$LRSAVE`($sp)
bl PICmeup
- subi r4,r4,1 ; prepare for lbzu
+ subi r4,r4,$LE_LOAD_SIZE ; prepare for ldu or lbzu
subi r12,r12,8 ; prepare for ldu
$PUSH r3,`$LOCALS+0*$SIZE_T`($sp) ; save A[][]
@@ -487,79 +502,79 @@ SHA3_absorb:
srwi r5,r5,3
$PUSH r4,`$LOCALS+2*$SIZE_T`($sp) ; save len
mtctr r5
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[0][0],$A[0][0],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[0][1],$A[0][1],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[0][2],$A[0][2],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[0][3],$A[0][3],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[0][4],$A[0][4],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[1][0],$A[1][0],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[1][1],$A[1][1],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[1][2],$A[1][2],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[1][3],$A[1][3],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[1][4],$A[1][4],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[2][0],$A[2][0],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[2][1],$A[2][1],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[2][2],$A[2][2],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[2][3],$A[2][3],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[2][4],$A[2][4],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[3][0],$A[3][0],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[3][1],$A[3][1],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[3][2],$A[3][2],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[3][3],$A[3][3],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[3][4],$A[3][4],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[4][0],$A[4][0],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[4][1],$A[4][1],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[4][2],$A[4][2],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[4][3],$A[4][3],r0
bdz .Lprocess_block
- bl dword_le_load ; *inp++
+ $DWORD_LE_LOAD ; *inp++
xor $A[4][4],$A[4][4],r0
.Lprocess_block:
More information about the openssl-commits
mailing list