sha1-armv4-large.pl performance improvement. On PXA255 it gives +10% on
authorAndy Polyakov <appro@openssl.org>
Wed, 6 Aug 2008 08:47:07 +0000 (08:47 +0000)
committerAndy Polyakov <appro@openssl.org>
Wed, 6 Aug 2008 08:47:07 +0000 (08:47 +0000)
8KB block, +60% on 1KB, +160% on 256B...

crypto/sha/asm/sha1-armv4-large.pl

index d468be3..102cf03 100644 (file)
@@ -18,8 +18,8 @@
 # thumb                304             3212            4420
 # armv4-small  392/+29%        1958/+64%       2250/+96%
 # armv4-compact        740/+89%        1552/+26%       1840/+22%
-# armv4-large  1420/+92%       1307/+19%       1500/+23%
-# full unroll  ~5100/+260%     ~1260/+4%       ~1500/+0%
+# armv4-large  1420/+92%       1307/+19%       1370/+34%[***]
+# full unroll  ~5100/+260%     ~1260/+4%       ~1300/+5%
 # ====================================================================
 # thumb                = same as 'small' but in Thumb instructions[**] and
 #                with recurring code in two private functions;
@@ -37,6 +37,7 @@
 #      modes are limited. As result it takes more instructions to do
 #      the same job in Thumb, therefore the code is never twice as
 #      small and always slower.
+# [***]        which is also ~35% better than compiler generated code.
 
 $output=shift;
 open STDOUT,">$output";
@@ -50,9 +51,10 @@ $c="r5";
 $d="r6";
 $e="r7";
 $K="r8";
-$t0="r10";
-$t1="r11";
-$t2="r12";
+$t0="r9";
+$t1="r10";
+$t2="r11";
+$t3="r12";
 $Xi="r14";
 @V=($a,$b,$c,$d,$e);
 
@@ -64,14 +66,14 @@ $code.=<<___;
        ldrb    $t0,[$inp],#4
        ldrb    $t1,[$inp,#-3]
        ldrb    $t2,[$inp,#-2]
+       ldrb    $t3,[$inp,#-1]
        add     $e,$K,$e,ror#2                  @ E+=K_00_19
        orr     $t0,$t1,$t0,lsl#8
-       ldrb    $t1,[$inp,#-1]
-       orr     $t0,$t2,$t0,lsl#8
        add     $e,$e,$a,ror#27                 @ E+=ROR(A,27)
-       orr     $t0,$t1,$t0,lsl#8
-       add     $e,$e,$t0                       @ E+=X[i]
+       orr     $t0,$t2,$t0,lsl#8
        eor     $t1,$c,$d                       @ F_xx_xx
+       orr     $t0,$t3,$t0,lsl#8
+       add     $e,$e,$t0                       @ E+=X[i]
        str     $t0,[$Xi,#-4]!
 ___
 }
@@ -81,12 +83,12 @@ $code.=<<___;
        ldr     $t0,[$Xi,#15*4]
        ldr     $t1,[$Xi,#13*4]
        ldr     $t2,[$Xi,#7*4]
+       ldr     $t3,[$Xi,#2*4]
        add     $e,$K,$e,ror#2                  @ E+=K_xx_xx
        eor     $t0,$t0,$t1
-       ldr     $t1,[$Xi,#2*4]
-       add     $e,$e,$a,ror#27                 @ E+=ROR(A,27)
        eor     $t0,$t0,$t2
-       eor     $t0,$t0,$t1
+       eor     $t0,$t0,$t3
+       add     $e,$e,$a,ror#27                 @ E+=ROR(A,27)
 ___
 $code.=<<___ if (!defined($flag));
        eor     $t1,$c,$d                       @ F_xx_xx, but not in 40_59