]> WPIA git - cassiopeia.git/blobdiff - lib/openssl/crypto/sha/asm/sha512-armv8.pl
upd: openssl to 1.1.0
[cassiopeia.git] / lib / openssl / crypto / sha / asm / sha512-armv8.pl
diff --git a/lib/openssl/crypto/sha/asm/sha512-armv8.pl b/lib/openssl/crypto/sha/asm/sha512-armv8.pl
new file mode 100644 (file)
index 0000000..c1aaf77
--- /dev/null
@@ -0,0 +1,446 @@
+#! /usr/bin/env perl
+# Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# SHA256/512 for ARMv8.
+#
+# Performance in cycles per processed byte and improvement coefficient
+# over code generated with "default" compiler:
+#
+#              SHA256-hw       SHA256(*)       SHA512
+# Apple A7     1.97            10.5 (+33%)     6.73 (-1%(**))
+# Cortex-A53   2.38            15.5 (+115%)    10.0 (+150%(***))
+# Cortex-A57   2.31            11.6 (+86%)     7.51 (+260%(***))
+# Denver       2.01            10.5 (+26%)     6.70 (+8%)
+# X-Gene                       20.0 (+100%)    12.8 (+300%(***))
+# Mongoose     2.36            13.0 (+50%)     8.36 (+33%)
+# 
+# (*)  Software SHA256 results are of lesser relevance, presented
+#      mostly for informational purposes.
+# (**) The result is a trade-off: it's possible to improve it by
+#      10% (or by 1 cycle per round), but at the cost of 20% loss
+#      on Cortex-A53 (or by 4 cycles per round).
+# (***)        Super-impressive coefficients over gcc-generated code are
+#      indication of some compiler "pathology", most notably code
+#      generated with -mgeneral-regs-only is significanty faster
+#      and the gap is only 40-90%.
+
+$flavour=shift;
+$output=shift;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+die "can't locate arm-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+if ($output =~ /512/) {
+       $BITS=512;
+       $SZ=8;
+       @Sigma0=(28,34,39);
+       @Sigma1=(14,18,41);
+       @sigma0=(1,  8, 7);
+       @sigma1=(19,61, 6);
+       $rounds=80;
+       $reg_t="x";
+} else {
+       $BITS=256;
+       $SZ=4;
+       @Sigma0=( 2,13,22);
+       @Sigma1=( 6,11,25);
+       @sigma0=( 7,18, 3);
+       @sigma1=(17,19,10);
+       $rounds=64;
+       $reg_t="w";
+}
+
+$func="sha${BITS}_block_data_order";
+
+($ctx,$inp,$num,$Ktbl)=map("x$_",(0..2,30));
+
+@X=map("$reg_t$_",(3..15,0..2));
+@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("$reg_t$_",(20..27));
+($t0,$t1,$t2,$t3)=map("$reg_t$_",(16,17,19,28));
+
+sub BODY_00_xx {
+my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
+my $j=($i+1)&15;
+my ($T0,$T1,$T2)=(@X[($i-8)&15],@X[($i-9)&15],@X[($i-10)&15]);
+   $T0=@X[$i+3] if ($i<11);
+
+$code.=<<___   if ($i<16);
+#ifndef        __ARMEB__
+       rev     @X[$i],@X[$i]                   // $i
+#endif
+___
+$code.=<<___   if ($i<13 && ($i&1));
+       ldp     @X[$i+1],@X[$i+2],[$inp],#2*$SZ
+___
+$code.=<<___   if ($i==13);
+       ldp     @X[14],@X[15],[$inp]
+___
+$code.=<<___   if ($i>=14);
+       ldr     @X[($i-11)&15],[sp,#`$SZ*(($i-11)%4)`]
+___
+$code.=<<___   if ($i>0 && $i<16);
+       add     $a,$a,$t1                       // h+=Sigma0(a)
+___
+$code.=<<___   if ($i>=11);
+       str     @X[($i-8)&15],[sp,#`$SZ*(($i-8)%4)`]
+___
+# While ARMv8 specifies merged rotate-n-logical operation such as
+# 'eor x,y,z,ror#n', it was found to negatively affect performance
+# on Apple A7. The reason seems to be that it requires even 'y' to
+# be available earlier. This means that such merged instruction is
+# not necessarily best choice on critical path... On the other hand
+# Cortex-A5x handles merged instructions much better than disjoint
+# rotate and logical... See (**) footnote above.
+$code.=<<___   if ($i<15);
+       ror     $t0,$e,#$Sigma1[0]
+       add     $h,$h,$t2                       // h+=K[i]
+       eor     $T0,$e,$e,ror#`$Sigma1[2]-$Sigma1[1]`
+       and     $t1,$f,$e
+       bic     $t2,$g,$e
+       add     $h,$h,@X[$i&15]                 // h+=X[i]
+       orr     $t1,$t1,$t2                     // Ch(e,f,g)
+       eor     $t2,$a,$b                       // a^b, b^c in next round
+       eor     $t0,$t0,$T0,ror#$Sigma1[1]      // Sigma1(e)
+       ror     $T0,$a,#$Sigma0[0]
+       add     $h,$h,$t1                       // h+=Ch(e,f,g)
+       eor     $t1,$a,$a,ror#`$Sigma0[2]-$Sigma0[1]`
+       add     $h,$h,$t0                       // h+=Sigma1(e)
+       and     $t3,$t3,$t2                     // (b^c)&=(a^b)
+       add     $d,$d,$h                        // d+=h
+       eor     $t3,$t3,$b                      // Maj(a,b,c)
+       eor     $t1,$T0,$t1,ror#$Sigma0[1]      // Sigma0(a)
+       add     $h,$h,$t3                       // h+=Maj(a,b,c)
+       ldr     $t3,[$Ktbl],#$SZ                // *K++, $t2 in next round
+       //add   $h,$h,$t1                       // h+=Sigma0(a)
+___
+$code.=<<___   if ($i>=15);
+       ror     $t0,$e,#$Sigma1[0]
+       add     $h,$h,$t2                       // h+=K[i]
+       ror     $T1,@X[($j+1)&15],#$sigma0[0]
+       and     $t1,$f,$e
+       ror     $T2,@X[($j+14)&15],#$sigma1[0]
+       bic     $t2,$g,$e
+       ror     $T0,$a,#$Sigma0[0]
+       add     $h,$h,@X[$i&15]                 // h+=X[i]
+       eor     $t0,$t0,$e,ror#$Sigma1[1]
+       eor     $T1,$T1,@X[($j+1)&15],ror#$sigma0[1]
+       orr     $t1,$t1,$t2                     // Ch(e,f,g)
+       eor     $t2,$a,$b                       // a^b, b^c in next round
+       eor     $t0,$t0,$e,ror#$Sigma1[2]       // Sigma1(e)
+       eor     $T0,$T0,$a,ror#$Sigma0[1]
+       add     $h,$h,$t1                       // h+=Ch(e,f,g)
+       and     $t3,$t3,$t2                     // (b^c)&=(a^b)
+       eor     $T2,$T2,@X[($j+14)&15],ror#$sigma1[1]
+       eor     $T1,$T1,@X[($j+1)&15],lsr#$sigma0[2]    // sigma0(X[i+1])
+       add     $h,$h,$t0                       // h+=Sigma1(e)
+       eor     $t3,$t3,$b                      // Maj(a,b,c)
+       eor     $t1,$T0,$a,ror#$Sigma0[2]       // Sigma0(a)
+       eor     $T2,$T2,@X[($j+14)&15],lsr#$sigma1[2]   // sigma1(X[i+14])
+       add     @X[$j],@X[$j],@X[($j+9)&15]
+       add     $d,$d,$h                        // d+=h
+       add     $h,$h,$t3                       // h+=Maj(a,b,c)
+       ldr     $t3,[$Ktbl],#$SZ                // *K++, $t2 in next round
+       add     @X[$j],@X[$j],$T1
+       add     $h,$h,$t1                       // h+=Sigma0(a)
+       add     @X[$j],@X[$j],$T2
+___
+       ($t2,$t3)=($t3,$t2);
+}
+
+$code.=<<___;
+#include "arm_arch.h"
+
+.text
+
+.extern        OPENSSL_armcap_P
+.globl $func
+.type  $func,%function
+.align 6
+$func:
+___
+$code.=<<___   if ($SZ==4);
+#ifdef __ILP32__
+       ldrsw   x16,.LOPENSSL_armcap_P
+#else
+       ldr     x16,.LOPENSSL_armcap_P
+#endif
+       adr     x17,.LOPENSSL_armcap_P
+       add     x16,x16,x17
+       ldr     w16,[x16]
+       tst     w16,#ARMV8_SHA256
+       b.ne    .Lv8_entry
+___
+$code.=<<___;
+       stp     x29,x30,[sp,#-128]!
+       add     x29,sp,#0
+
+       stp     x19,x20,[sp,#16]
+       stp     x21,x22,[sp,#32]
+       stp     x23,x24,[sp,#48]
+       stp     x25,x26,[sp,#64]
+       stp     x27,x28,[sp,#80]
+       sub     sp,sp,#4*$SZ
+
+       ldp     $A,$B,[$ctx]                            // load context
+       ldp     $C,$D,[$ctx,#2*$SZ]
+       ldp     $E,$F,[$ctx,#4*$SZ]
+       add     $num,$inp,$num,lsl#`log(16*$SZ)/log(2)` // end of input
+       ldp     $G,$H,[$ctx,#6*$SZ]
+       adr     $Ktbl,.LK$BITS
+       stp     $ctx,$num,[x29,#96]
+
+.Loop:
+       ldp     @X[0],@X[1],[$inp],#2*$SZ
+       ldr     $t2,[$Ktbl],#$SZ                        // *K++
+       eor     $t3,$B,$C                               // magic seed
+       str     $inp,[x29,#112]
+___
+for ($i=0;$i<16;$i++)  { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
+$code.=".Loop_16_xx:\n";
+for (;$i<32;$i++)      { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+       cbnz    $t2,.Loop_16_xx
+
+       ldp     $ctx,$num,[x29,#96]
+       ldr     $inp,[x29,#112]
+       sub     $Ktbl,$Ktbl,#`$SZ*($rounds+1)`          // rewind
+
+       ldp     @X[0],@X[1],[$ctx]
+       ldp     @X[2],@X[3],[$ctx,#2*$SZ]
+       add     $inp,$inp,#14*$SZ                       // advance input pointer
+       ldp     @X[4],@X[5],[$ctx,#4*$SZ]
+       add     $A,$A,@X[0]
+       ldp     @X[6],@X[7],[$ctx,#6*$SZ]
+       add     $B,$B,@X[1]
+       add     $C,$C,@X[2]
+       add     $D,$D,@X[3]
+       stp     $A,$B,[$ctx]
+       add     $E,$E,@X[4]
+       add     $F,$F,@X[5]
+       stp     $C,$D,[$ctx,#2*$SZ]
+       add     $G,$G,@X[6]
+       add     $H,$H,@X[7]
+       cmp     $inp,$num
+       stp     $E,$F,[$ctx,#4*$SZ]
+       stp     $G,$H,[$ctx,#6*$SZ]
+       b.ne    .Loop
+
+       ldp     x19,x20,[x29,#16]
+       add     sp,sp,#4*$SZ
+       ldp     x21,x22,[x29,#32]
+       ldp     x23,x24,[x29,#48]
+       ldp     x25,x26,[x29,#64]
+       ldp     x27,x28,[x29,#80]
+       ldp     x29,x30,[sp],#128
+       ret
+.size  $func,.-$func
+
+.align 6
+.type  .LK$BITS,%object
+.LK$BITS:
+___
+$code.=<<___ if ($SZ==8);
+       .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
+       .quad   0       // terminator
+___
+$code.=<<___ if ($SZ==4);
+       .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
+       .long   0       //terminator
+___
+$code.=<<___;
+.size  .LK$BITS,.-.LK$BITS
+.align 3
+.LOPENSSL_armcap_P:
+#ifdef __ILP32__
+       .long   OPENSSL_armcap_P-.
+#else
+       .quad   OPENSSL_armcap_P-.
+#endif
+.asciz "SHA$BITS block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
+.align 2
+___
+
+if ($SZ==4) {
+my $Ktbl="x3";
+
+my ($ABCD,$EFGH,$abcd)=map("v$_.16b",(0..2));
+my @MSG=map("v$_.16b",(4..7));
+my ($W0,$W1)=("v16.4s","v17.4s");
+my ($ABCD_SAVE,$EFGH_SAVE)=("v18.16b","v19.16b");
+
+$code.=<<___;
+.type  sha256_block_armv8,%function
+.align 6
+sha256_block_armv8:
+.Lv8_entry:
+       stp             x29,x30,[sp,#-16]!
+       add             x29,sp,#0
+
+       ld1.32          {$ABCD,$EFGH},[$ctx]
+       adr             $Ktbl,.LK256
+
+.Loop_hw:
+       ld1             {@MSG[0]-@MSG[3]},[$inp],#64
+       sub             $num,$num,#1
+       ld1.32          {$W0},[$Ktbl],#16
+       rev32           @MSG[0],@MSG[0]
+       rev32           @MSG[1],@MSG[1]
+       rev32           @MSG[2],@MSG[2]
+       rev32           @MSG[3],@MSG[3]
+       orr             $ABCD_SAVE,$ABCD,$ABCD          // offload
+       orr             $EFGH_SAVE,$EFGH,$EFGH
+___
+for($i=0;$i<12;$i++) {
+$code.=<<___;
+       ld1.32          {$W1},[$Ktbl],#16
+       add.i32         $W0,$W0,@MSG[0]
+       sha256su0       @MSG[0],@MSG[1]
+       orr             $abcd,$ABCD,$ABCD
+       sha256h         $ABCD,$EFGH,$W0
+       sha256h2        $EFGH,$abcd,$W0
+       sha256su1       @MSG[0],@MSG[2],@MSG[3]
+___
+       ($W0,$W1)=($W1,$W0);    push(@MSG,shift(@MSG));
+}
+$code.=<<___;
+       ld1.32          {$W1},[$Ktbl],#16
+       add.i32         $W0,$W0,@MSG[0]
+       orr             $abcd,$ABCD,$ABCD
+       sha256h         $ABCD,$EFGH,$W0
+       sha256h2        $EFGH,$abcd,$W0
+
+       ld1.32          {$W0},[$Ktbl],#16
+       add.i32         $W1,$W1,@MSG[1]
+       orr             $abcd,$ABCD,$ABCD
+       sha256h         $ABCD,$EFGH,$W1
+       sha256h2        $EFGH,$abcd,$W1
+
+       ld1.32          {$W1},[$Ktbl]
+       add.i32         $W0,$W0,@MSG[2]
+       sub             $Ktbl,$Ktbl,#$rounds*$SZ-16     // rewind
+       orr             $abcd,$ABCD,$ABCD
+       sha256h         $ABCD,$EFGH,$W0
+       sha256h2        $EFGH,$abcd,$W0
+
+       add.i32         $W1,$W1,@MSG[3]
+       orr             $abcd,$ABCD,$ABCD
+       sha256h         $ABCD,$EFGH,$W1
+       sha256h2        $EFGH,$abcd,$W1
+
+       add.i32         $ABCD,$ABCD,$ABCD_SAVE
+       add.i32         $EFGH,$EFGH,$EFGH_SAVE
+
+       cbnz            $num,.Loop_hw
+
+       st1.32          {$ABCD,$EFGH},[$ctx]
+
+       ldr             x29,[sp],#16
+       ret
+.size  sha256_block_armv8,.-sha256_block_armv8
+___
+}
+
+$code.=<<___;
+.comm  OPENSSL_armcap_P,4,4
+___
+
+{   my  %opcode = (
+       "sha256h"       => 0x5e004000,  "sha256h2"      => 0x5e005000,
+       "sha256su0"     => 0x5e282800,  "sha256su1"     => 0x5e006000   );
+
+    sub unsha256 {
+       my ($mnemonic,$arg)=@_;
+
+       $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
+       &&
+       sprintf ".inst\t0x%08x\t//%s %s",
+                       $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
+                       $mnemonic,$arg;
+    }
+}
+
+foreach(split("\n",$code)) {
+
+       s/\`([^\`]*)\`/eval($1)/geo;
+
+       s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/geo;
+
+       s/\.\w?32\b//o          and s/\.16b/\.4s/go;
+       m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go;
+
+       print $_,"\n";
+}
+
+close STDOUT;