^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1) #! /usr/bin/env perl
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) # SPDX-License-Identifier: GPL-2.0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4) # This code is taken from the OpenSSL project but the author (Andy Polyakov)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) # has relicensed it under the GPLv2. Therefore this program is free software;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) # you can redistribute it and/or modify it under the terms of the GNU General
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) # Public License version 2 as published by the Free Software Foundation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) # The original headers, including the original license headers, are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) # included below for completeness.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) # Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) # Licensed under the OpenSSL license (the "License"). You may not use
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) # this file except in compliance with the License. You can obtain a copy
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) # in the file LICENSE in the source distribution or at
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17) # https://www.openssl.org/source/license.html
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20) # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) # project. The module is, however, dual licensed under OpenSSL and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) # CRYPTOGAMS licenses depending on where you obtain it. For further
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23) # details see http://www.openssl.org/~appro/cryptogams/.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26) # SHA256/512 for ARMv8.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) # Performance in cycles per processed byte and improvement coefficient
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29) # over code generated with "default" compiler:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31) # SHA256-hw SHA256(*) SHA512
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) # Apple A7 1.97 10.5 (+33%) 6.73 (-1%(**))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) # Cortex-A53 2.38 15.5 (+115%) 10.0 (+150%(***))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34) # Cortex-A57 2.31 11.6 (+86%) 7.51 (+260%(***))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) # Denver 2.01 10.5 (+26%) 6.70 (+8%)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36) # X-Gene 20.0 (+100%) 12.8 (+300%(***))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) # Mongoose 2.36 13.0 (+50%) 8.36 (+33%)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39) # (*) Software SHA256 results are of lesser relevance, presented
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40) # mostly for informational purposes.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) # (**) The result is a trade-off: it's possible to improve it by
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) # 10% (or by 1 cycle per round), but at the cost of 20% loss
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43) # on Cortex-A53 (or by 4 cycles per round).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) # (***) Super-impressive coefficients over gcc-generated code are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45) # indication of some compiler "pathology", most notably code
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46) # generated with -mgeneral-regs-only is significanty faster
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) # and the gap is only 40-90%.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49) # October 2016.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) # Originally it was reckoned that it makes no sense to implement NEON
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52) # version of SHA256 for 64-bit processors. This is because performance
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53) # improvement on most wide-spread Cortex-A5x processors was observed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) # to be marginal, same on Cortex-A53 and ~10% on A57. But then it was
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) # observed that 32-bit NEON SHA256 performs significantly better than
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) # 64-bit scalar version on *some* of the more recent processors. As
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57) # result 64-bit NEON version of SHA256 was added to provide best
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) # all-round performance. For example it executes ~30% faster on X-Gene
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59) # and Mongoose. [For reference, NEON version of SHA512 is bound to
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) # deliver much less improvement, likely *negative* on Cortex-A5x.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61) # Which is why NEON support is limited to SHA256.]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63) $output=pop;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) $flavour=pop;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66) if ($flavour && $flavour ne "void") {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70) die "can't locate arm-xlate.pl";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72) open OUT,"| \"$^X\" $xlate $flavour $output";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73) *STDOUT=*OUT;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) open STDOUT,">$output";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) if ($output =~ /512/) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) $BITS=512;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80) $SZ=8;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81) @Sigma0=(28,34,39);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82) @Sigma1=(14,18,41);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) @sigma0=(1, 8, 7);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84) @sigma1=(19,61, 6);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) $rounds=80;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) $reg_t="x";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) $BITS=256;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) $SZ=4;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90) @Sigma0=( 2,13,22);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) @Sigma1=( 6,11,25);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) @sigma0=( 7,18, 3);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93) @sigma1=(17,19,10);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) $rounds=64;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) $reg_t="w";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) $func="sha${BITS}_block_data_order";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) ($ctx,$inp,$num,$Ktbl)=map("x$_",(0..2,30));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) @X=map("$reg_t$_",(3..15,0..2));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("$reg_t$_",(20..27));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104) ($t0,$t1,$t2,$t3)=map("$reg_t$_",(16,17,19,28));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) sub BODY_00_xx {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) my $j=($i+1)&15;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) my ($T0,$T1,$T2)=(@X[($i-8)&15],@X[($i-9)&15],@X[($i-10)&15]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) $T0=@X[$i+3] if ($i<11);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) $code.=<<___ if ($i<16);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) #ifndef __AARCH64EB__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) rev @X[$i],@X[$i] // $i
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117) $code.=<<___ if ($i<13 && ($i&1));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) ldp @X[$i+1],@X[$i+2],[$inp],#2*$SZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) $code.=<<___ if ($i==13);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) ldp @X[14],@X[15],[$inp]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) $code.=<<___ if ($i>=14);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) ldr @X[($i-11)&15],[sp,#`$SZ*(($i-11)%4)`]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) $code.=<<___ if ($i>0 && $i<16);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) add $a,$a,$t1 // h+=Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) $code.=<<___ if ($i>=11);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) str @X[($i-8)&15],[sp,#`$SZ*(($i-8)%4)`]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) # While ARMv8 specifies merged rotate-n-logical operation such as
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) # 'eor x,y,z,ror#n', it was found to negatively affect performance
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) # on Apple A7. The reason seems to be that it requires even 'y' to
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) # be available earlier. This means that such merged instruction is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) # not necessarily best choice on critical path... On the other hand
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) # Cortex-A5x handles merged instructions much better than disjoint
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) # rotate and logical... See (**) footnote above.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) $code.=<<___ if ($i<15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) ror $t0,$e,#$Sigma1[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) add $h,$h,$t2 // h+=K[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) eor $T0,$e,$e,ror#`$Sigma1[2]-$Sigma1[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) and $t1,$f,$e
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) bic $t2,$g,$e
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) add $h,$h,@X[$i&15] // h+=X[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) orr $t1,$t1,$t2 // Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) eor $t2,$a,$b // a^b, b^c in next round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148) eor $t0,$t0,$T0,ror#$Sigma1[1] // Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) ror $T0,$a,#$Sigma0[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) add $h,$h,$t1 // h+=Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) eor $t1,$a,$a,ror#`$Sigma0[2]-$Sigma0[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) add $h,$h,$t0 // h+=Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) and $t3,$t3,$t2 // (b^c)&=(a^b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) add $d,$d,$h // d+=h
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155) eor $t3,$t3,$b // Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) eor $t1,$T0,$t1,ror#$Sigma0[1] // Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) add $h,$h,$t3 // h+=Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) ldr $t3,[$Ktbl],#$SZ // *K++, $t2 in next round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) //add $h,$h,$t1 // h+=Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) $code.=<<___ if ($i>=15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) ror $t0,$e,#$Sigma1[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) add $h,$h,$t2 // h+=K[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) ror $T1,@X[($j+1)&15],#$sigma0[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) and $t1,$f,$e
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) ror $T2,@X[($j+14)&15],#$sigma1[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) bic $t2,$g,$e
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) ror $T0,$a,#$Sigma0[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) add $h,$h,@X[$i&15] // h+=X[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170) eor $t0,$t0,$e,ror#$Sigma1[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) eor $T1,$T1,@X[($j+1)&15],ror#$sigma0[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) orr $t1,$t1,$t2 // Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) eor $t2,$a,$b // a^b, b^c in next round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) eor $t0,$t0,$e,ror#$Sigma1[2] // Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) eor $T0,$T0,$a,ror#$Sigma0[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) add $h,$h,$t1 // h+=Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) and $t3,$t3,$t2 // (b^c)&=(a^b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) eor $T2,$T2,@X[($j+14)&15],ror#$sigma1[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) eor $T1,$T1,@X[($j+1)&15],lsr#$sigma0[2] // sigma0(X[i+1])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) add $h,$h,$t0 // h+=Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) eor $t3,$t3,$b // Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) eor $t1,$T0,$a,ror#$Sigma0[2] // Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) eor $T2,$T2,@X[($j+14)&15],lsr#$sigma1[2] // sigma1(X[i+14])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) add @X[$j],@X[$j],@X[($j+9)&15]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) add $d,$d,$h // d+=h
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) add $h,$h,$t3 // h+=Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) ldr $t3,[$Ktbl],#$SZ // *K++, $t2 in next round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) add @X[$j],@X[$j],$T1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) add $h,$h,$t1 // h+=Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) add @X[$j],@X[$j],$T2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) ($t2,$t3)=($t3,$t2);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) # include "arm_arch.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) .text
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) .extern OPENSSL_armcap_P
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) .globl $func
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204) .type $func,%function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) .align 6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206) $func:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) $code.=<<___ if ($SZ==4);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) # ifdef __ILP32__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) ldrsw x16,.LOPENSSL_armcap_P
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) # else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) ldr x16,.LOPENSSL_armcap_P
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) # endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) adr x17,.LOPENSSL_armcap_P
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) add x16,x16,x17
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) ldr w16,[x16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218) tst w16,#ARMV8_SHA256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) b.ne .Lv8_entry
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220) tst w16,#ARMV7_NEON
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) b.ne .Lneon_entry
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) stp x29,x30,[sp,#-128]!
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) add x29,sp,#0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) stp x19,x20,[sp,#16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) stp x21,x22,[sp,#32]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) stp x23,x24,[sp,#48]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) stp x25,x26,[sp,#64]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) stp x27,x28,[sp,#80]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) sub sp,sp,#4*$SZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) ldp $A,$B,[$ctx] // load context
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236) ldp $C,$D,[$ctx,#2*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) ldp $E,$F,[$ctx,#4*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) add $num,$inp,$num,lsl#`log(16*$SZ)/log(2)` // end of input
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) ldp $G,$H,[$ctx,#6*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) adr $Ktbl,.LK$BITS
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) stp $ctx,$num,[x29,#96]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) .Loop:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) ldp @X[0],@X[1],[$inp],#2*$SZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) ldr $t2,[$Ktbl],#$SZ // *K++
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246) eor $t3,$B,$C // magic seed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) str $inp,[x29,#112]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) for ($i=0;$i<16;$i++) { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) $code.=".Loop_16_xx:\n";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) for (;$i<32;$i++) { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) cbnz $t2,.Loop_16_xx
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) ldp $ctx,$num,[x29,#96]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) ldr $inp,[x29,#112]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) sub $Ktbl,$Ktbl,#`$SZ*($rounds+1)` // rewind
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259) ldp @X[0],@X[1],[$ctx]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) ldp @X[2],@X[3],[$ctx,#2*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) add $inp,$inp,#14*$SZ // advance input pointer
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) ldp @X[4],@X[5],[$ctx,#4*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263) add $A,$A,@X[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) ldp @X[6],@X[7],[$ctx,#6*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) add $B,$B,@X[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) add $C,$C,@X[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) add $D,$D,@X[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) stp $A,$B,[$ctx]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) add $E,$E,@X[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) add $F,$F,@X[5]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) stp $C,$D,[$ctx,#2*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) add $G,$G,@X[6]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273) add $H,$H,@X[7]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) cmp $inp,$num
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) stp $E,$F,[$ctx,#4*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) stp $G,$H,[$ctx,#6*$SZ]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277) b.ne .Loop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) ldp x19,x20,[x29,#16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280) add sp,sp,#4*$SZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) ldp x21,x22,[x29,#32]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282) ldp x23,x24,[x29,#48]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) ldp x25,x26,[x29,#64]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) ldp x27,x28,[x29,#80]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) ldp x29,x30,[sp],#128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) .size $func,.-$func
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) .align 6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) .type .LK$BITS,%object
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291) .LK$BITS:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) $code.=<<___ if ($SZ==8);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) .quad 0x428a2f98d728ae22,0x7137449123ef65cd
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) .quad 0x3956c25bf348b538,0x59f111f1b605d019
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298) .quad 0xd807aa98a3030242,0x12835b0145706fbe
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) .quad 0x9bdc06a725c71235,0xc19bf174cf692694
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305) .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) .quad 0x983e5152ee66dfab,0xa831c66d2db43210
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307) .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) .quad 0x06ca6351e003826f,0x142929670a0e6e70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310) .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311) .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) .quad 0x650a73548baf63de,0x766a0abb3c77b2a8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) .quad 0x81c2c92e47edaee6,0x92722c851482353b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) .quad 0xc24b8b70d0f89791,0xc76c51a30654be30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316) .quad 0xd192e819d6ef5218,0xd69906245565a910
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) .quad 0xf40e35855771202a,0x106aa07032bbd1b8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319) .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322) .quad 0x748f82ee5defb2fc,0x78a5636f43172f60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324) .quad 0x90befffa23631e28,0xa4506cebde82bde9
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) .quad 0xca273eceea26619c,0xd186b8c721c0c207
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) .quad 0x113f9804bef90dae,0x1b710b35131c471b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330) .quad 0x28db77f523047d84,0x32caab7b40c72493
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331) .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334) .quad 0 // terminator
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336) $code.=<<___ if ($SZ==4);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338) .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340) .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346) .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347) .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348) .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351) .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352) .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) .long 0 //terminator
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) .size .LK$BITS,.-.LK$BITS
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) .align 3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) .LOPENSSL_armcap_P:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) # ifdef __ILP32__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361) .long OPENSSL_armcap_P-.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) # else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) .quad OPENSSL_armcap_P-.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364) # endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) .asciz "SHA$BITS block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) .align 2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) if ($SZ==4) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371) my $Ktbl="x3";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) my ($ABCD,$EFGH,$abcd)=map("v$_.16b",(0..2));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374) my @MSG=map("v$_.16b",(4..7));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375) my ($W0,$W1)=("v16.4s","v17.4s");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376) my ($ABCD_SAVE,$EFGH_SAVE)=("v18.16b","v19.16b");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380) .type sha256_block_armv8,%function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381) .align 6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) sha256_block_armv8:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) .Lv8_entry:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) stp x29,x30,[sp,#-16]!
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385) add x29,sp,#0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) ld1.32 {$ABCD,$EFGH},[$ctx]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388) adr $Ktbl,.LK256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390) .Loop_hw:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) ld1 {@MSG[0]-@MSG[3]},[$inp],#64
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392) sub $num,$num,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393) ld1.32 {$W0},[$Ktbl],#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) rev32 @MSG[0],@MSG[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) rev32 @MSG[1],@MSG[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) rev32 @MSG[2],@MSG[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) rev32 @MSG[3],@MSG[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398) orr $ABCD_SAVE,$ABCD,$ABCD // offload
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399) orr $EFGH_SAVE,$EFGH,$EFGH
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401) for($i=0;$i<12;$i++) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) ld1.32 {$W1},[$Ktbl],#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404) add.i32 $W0,$W0,@MSG[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405) sha256su0 @MSG[0],@MSG[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406) orr $abcd,$ABCD,$ABCD
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) sha256h $ABCD,$EFGH,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408) sha256h2 $EFGH,$abcd,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409) sha256su1 @MSG[0],@MSG[2],@MSG[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411) ($W0,$W1)=($W1,$W0); push(@MSG,shift(@MSG));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) ld1.32 {$W1},[$Ktbl],#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415) add.i32 $W0,$W0,@MSG[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416) orr $abcd,$ABCD,$ABCD
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) sha256h $ABCD,$EFGH,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418) sha256h2 $EFGH,$abcd,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420) ld1.32 {$W0},[$Ktbl],#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) add.i32 $W1,$W1,@MSG[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422) orr $abcd,$ABCD,$ABCD
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) sha256h $ABCD,$EFGH,$W1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424) sha256h2 $EFGH,$abcd,$W1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426) ld1.32 {$W1},[$Ktbl]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) add.i32 $W0,$W0,@MSG[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) sub $Ktbl,$Ktbl,#$rounds*$SZ-16 // rewind
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429) orr $abcd,$ABCD,$ABCD
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) sha256h $ABCD,$EFGH,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) sha256h2 $EFGH,$abcd,$W0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433) add.i32 $W1,$W1,@MSG[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434) orr $abcd,$ABCD,$ABCD
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) sha256h $ABCD,$EFGH,$W1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436) sha256h2 $EFGH,$abcd,$W1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438) add.i32 $ABCD,$ABCD,$ABCD_SAVE
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) add.i32 $EFGH,$EFGH,$EFGH_SAVE
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441) cbnz $num,.Loop_hw
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443) st1.32 {$ABCD,$EFGH},[$ctx]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445) ldr x29,[sp],#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) .size sha256_block_armv8,.-sha256_block_armv8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) if ($SZ==4) { ######################################### NEON stuff #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) # You'll surely note a lot of similarities with sha256-armv4 module,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454) # and of course it's not a coincidence. sha256-armv4 was used as
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) # initial template, but was adapted for ARMv8 instruction set and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456) # extensively re-tuned for all-round performance.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) my @V = ($A,$B,$C,$D,$E,$F,$G,$H) = map("w$_",(3..10));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) my ($t0,$t1,$t2,$t3,$t4) = map("w$_",(11..15));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460) my $Ktbl="x16";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) my $Xfer="x17";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462) my @X = map("q$_",(0..3));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) my ($T0,$T1,$T2,$T3,$T4,$T5,$T6,$T7) = map("q$_",(4..7,16..19));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464) my $j=0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) sub AUTOLOAD() # thunk [simplified] x86-style perlasm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 468) my $arg = pop;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 469) $arg = "#$arg" if ($arg*1 eq $arg);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 470) $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 471) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 472)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 473) sub Dscalar { shift =~ m|[qv]([0-9]+)|?"d$1":""; }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 474) sub Dlo { shift =~ m|[qv]([0-9]+)|?"v$1.d[0]":""; }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 475) sub Dhi { shift =~ m|[qv]([0-9]+)|?"v$1.d[1]":""; }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 476)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 477) sub Xupdate()
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 478) { use integer;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 479) my $body = shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 480) my @insns = (&$body,&$body,&$body,&$body);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 481) my ($a,$b,$c,$d,$e,$f,$g,$h);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 482)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 483) &ext_8 ($T0,@X[0],@X[1],4); # X[1..4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 484) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 485) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 486) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 487) &ext_8 ($T3,@X[2],@X[3],4); # X[9..12]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 488) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 489) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 490) &mov (&Dscalar($T7),&Dhi(@X[3])); # X[14..15]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 491) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 492) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 493) &ushr_32 ($T2,$T0,$sigma0[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 494) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 495) &ushr_32 ($T1,$T0,$sigma0[2]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 496) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 497) &add_32 (@X[0],@X[0],$T3); # X[0..3] += X[9..12]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 498) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 499) &sli_32 ($T2,$T0,32-$sigma0[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 500) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 501) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 502) &ushr_32 ($T3,$T0,$sigma0[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 503) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 504) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 505) &eor_8 ($T1,$T1,$T2);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 506) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 507) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 508) &sli_32 ($T3,$T0,32-$sigma0[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 509) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 510) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 511) &ushr_32 ($T4,$T7,$sigma1[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 512) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 513) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 514) &eor_8 ($T1,$T1,$T3); # sigma0(X[1..4])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 515) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 516) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 517) &sli_32 ($T4,$T7,32-$sigma1[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 518) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 519) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 520) &ushr_32 ($T5,$T7,$sigma1[2]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 521) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 522) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 523) &ushr_32 ($T3,$T7,$sigma1[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 524) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 525) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 526) &add_32 (@X[0],@X[0],$T1); # X[0..3] += sigma0(X[1..4])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 527) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 528) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 529) &sli_u32 ($T3,$T7,32-$sigma1[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 530) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 531) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 532) &eor_8 ($T5,$T5,$T4);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 533) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 534) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 535) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 536) &eor_8 ($T5,$T5,$T3); # sigma1(X[14..15])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 537) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 538) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 539) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 540) &add_32 (@X[0],@X[0],$T5); # X[0..1] += sigma1(X[14..15])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 541) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 542) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 543) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 544) &ushr_32 ($T6,@X[0],$sigma1[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 545) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 546) &ushr_32 ($T7,@X[0],$sigma1[2]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 547) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 548) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 549) &sli_32 ($T6,@X[0],32-$sigma1[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 550) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 551) &ushr_32 ($T5,@X[0],$sigma1[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 552) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 553) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 554) &eor_8 ($T7,$T7,$T6);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 555) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 556) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 557) &sli_32 ($T5,@X[0],32-$sigma1[1]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 558) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 559) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 560) &ld1_32 ("{$T0}","[$Ktbl], #16");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 561) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 562) &eor_8 ($T7,$T7,$T5); # sigma1(X[16..17])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 563) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 564) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 565) &eor_8 ($T5,$T5,$T5);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 566) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 567) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 568) &mov (&Dhi($T5), &Dlo($T7));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 569) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 570) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 571) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 572) &add_32 (@X[0],@X[0],$T5); # X[2..3] += sigma1(X[16..17])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 573) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 574) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 575) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 576) &add_32 ($T0,$T0,@X[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 577) while($#insns>=1) { eval(shift(@insns)); }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 578) &st1_32 ("{$T0}","[$Xfer], #16");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 579) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 580)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 581) push(@X,shift(@X)); # "rotate" X[]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 582) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 583)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 584) sub Xpreload()
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 585) { use integer;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 586) my $body = shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 587) my @insns = (&$body,&$body,&$body,&$body);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 588) my ($a,$b,$c,$d,$e,$f,$g,$h);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 589)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 590) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 591) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 592) &ld1_8 ("{@X[0]}","[$inp],#16");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 593) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 594) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 595) &ld1_32 ("{$T0}","[$Ktbl],#16");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 596) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 597) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 598) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 599) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 600) &rev32 (@X[0],@X[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 601) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 602) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 603) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 604) eval(shift(@insns));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 605) &add_32 ($T0,$T0,@X[0]);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 606) foreach (@insns) { eval; } # remaining instructions
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 607) &st1_32 ("{$T0}","[$Xfer], #16");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 608)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 609) push(@X,shift(@X)); # "rotate" X[]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 610) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 611)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 612) sub body_00_15 () {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 613) (
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 614) '($a,$b,$c,$d,$e,$f,$g,$h)=@V;'.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 615) '&add ($h,$h,$t1)', # h+=X[i]+K[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 616) '&add ($a,$a,$t4);'. # h+=Sigma0(a) from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 617) '&and ($t1,$f,$e)',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 618) '&bic ($t4,$g,$e)',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 619) '&eor ($t0,$e,$e,"ror#".($Sigma1[1]-$Sigma1[0]))',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 620) '&add ($a,$a,$t2)', # h+=Maj(a,b,c) from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 621) '&orr ($t1,$t1,$t4)', # Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 622) '&eor ($t0,$t0,$e,"ror#".($Sigma1[2]-$Sigma1[0]))', # Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 623) '&eor ($t4,$a,$a,"ror#".($Sigma0[1]-$Sigma0[0]))',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 624) '&add ($h,$h,$t1)', # h+=Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 625) '&ror ($t0,$t0,"#$Sigma1[0]")',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 626) '&eor ($t2,$a,$b)', # a^b, b^c in next round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 627) '&eor ($t4,$t4,$a,"ror#".($Sigma0[2]-$Sigma0[0]))', # Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 628) '&add ($h,$h,$t0)', # h+=Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 629) '&ldr ($t1,sprintf "[sp,#%d]",4*(($j+1)&15)) if (($j&15)!=15);'.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 630) '&ldr ($t1,"[$Ktbl]") if ($j==15);'.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 631) '&and ($t3,$t3,$t2)', # (b^c)&=(a^b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 632) '&ror ($t4,$t4,"#$Sigma0[0]")',
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 633) '&add ($d,$d,$h)', # d+=h
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 634) '&eor ($t3,$t3,$b)', # Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 635) '$j++; unshift(@V,pop(@V)); ($t2,$t3)=($t3,$t2);'
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 636) )
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 637) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 638)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 639) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 640) #ifdef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 641) .globl sha256_block_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 642) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 643) .type sha256_block_neon,%function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 644) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 645) sha256_block_neon:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 646) .Lneon_entry:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 647) stp x29, x30, [sp, #-16]!
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 648) mov x29, sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 649) sub sp,sp,#16*4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 650)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 651) adr $Ktbl,.LK256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 652) add $num,$inp,$num,lsl#6 // len to point at the end of inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 653)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 654) ld1.8 {@X[0]},[$inp], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 655) ld1.8 {@X[1]},[$inp], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 656) ld1.8 {@X[2]},[$inp], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 657) ld1.8 {@X[3]},[$inp], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 658) ld1.32 {$T0},[$Ktbl], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 659) ld1.32 {$T1},[$Ktbl], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 660) ld1.32 {$T2},[$Ktbl], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 661) ld1.32 {$T3},[$Ktbl], #16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 662) rev32 @X[0],@X[0] // yes, even on
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 663) rev32 @X[1],@X[1] // big-endian
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 664) rev32 @X[2],@X[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 665) rev32 @X[3],@X[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 666) mov $Xfer,sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 667) add.32 $T0,$T0,@X[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 668) add.32 $T1,$T1,@X[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 669) add.32 $T2,$T2,@X[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 670) st1.32 {$T0-$T1},[$Xfer], #32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 671) add.32 $T3,$T3,@X[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 672) st1.32 {$T2-$T3},[$Xfer]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 673) sub $Xfer,$Xfer,#32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 674)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 675) ldp $A,$B,[$ctx]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 676) ldp $C,$D,[$ctx,#8]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 677) ldp $E,$F,[$ctx,#16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 678) ldp $G,$H,[$ctx,#24]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 679) ldr $t1,[sp,#0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 680) mov $t2,wzr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 681) eor $t3,$B,$C
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 682) mov $t4,wzr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 683) b .L_00_48
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 684)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 685) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 686) .L_00_48:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 687) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 688) &Xupdate(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 689) &Xupdate(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 690) &Xupdate(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 691) &Xupdate(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 692) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 693) cmp $t1,#0 // check for K256 terminator
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 694) ldr $t1,[sp,#0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 695) sub $Xfer,$Xfer,#64
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 696) bne .L_00_48
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 697)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 698) sub $Ktbl,$Ktbl,#256 // rewind $Ktbl
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 699) cmp $inp,$num
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 700) mov $Xfer, #64
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 701) csel $Xfer, $Xfer, xzr, eq
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 702) sub $inp,$inp,$Xfer // avoid SEGV
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 703) mov $Xfer,sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 704) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 705) &Xpreload(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 706) &Xpreload(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 707) &Xpreload(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 708) &Xpreload(\&body_00_15);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 709) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 710) add $A,$A,$t4 // h+=Sigma0(a) from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 711) ldp $t0,$t1,[$ctx,#0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 712) add $A,$A,$t2 // h+=Maj(a,b,c) from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 713) ldp $t2,$t3,[$ctx,#8]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 714) add $A,$A,$t0 // accumulate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 715) add $B,$B,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 716) ldp $t0,$t1,[$ctx,#16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 717) add $C,$C,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 718) add $D,$D,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 719) ldp $t2,$t3,[$ctx,#24]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 720) add $E,$E,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 721) add $F,$F,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 722) ldr $t1,[sp,#0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 723) stp $A,$B,[$ctx,#0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 724) add $G,$G,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 725) mov $t2,wzr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 726) stp $C,$D,[$ctx,#8]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 727) add $H,$H,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 728) stp $E,$F,[$ctx,#16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 729) eor $t3,$B,$C
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 730) stp $G,$H,[$ctx,#24]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 731) mov $t4,wzr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 732) mov $Xfer,sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 733) b.ne .L_00_48
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 734)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 735) ldr x29,[x29]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 736) add sp,sp,#16*4+16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 737) ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 738) .size sha256_block_neon,.-sha256_block_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 739) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 740) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 741)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 742) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 743) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 744) .comm OPENSSL_armcap_P,4,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 745) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 746) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 747)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 748) { my %opcode = (
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 749) "sha256h" => 0x5e004000, "sha256h2" => 0x5e005000,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 750) "sha256su0" => 0x5e282800, "sha256su1" => 0x5e006000 );
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 751)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 752) sub unsha256 {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 753) my ($mnemonic,$arg)=@_;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 754)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 755) $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 756) &&
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 757) sprintf ".inst\t0x%08x\t//%s %s",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 758) $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 759) $mnemonic,$arg;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 760) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 761) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 762)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 763) open SELF,$0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 764) while(<SELF>) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 765) next if (/^#!/);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 766) last if (!s/^#/\/\// and !/^$/);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 767) print;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 768) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 769) close SELF;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 770)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 771) foreach(split("\n",$code)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 772)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 773) s/\`([^\`]*)\`/eval($1)/ge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 774)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 775) s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/ge;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 776)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 777) s/\bq([0-9]+)\b/v$1.16b/g; # old->new registers
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 778)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 779) s/\.[ui]?8(\s)/$1/;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 780) s/\.\w?32\b// and s/\.16b/\.4s/g;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 781) m/(ld|st)1[^\[]+\[0\]/ and s/\.4s/\.s/g;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 782)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 783) print $_,"\n";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 784) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 785)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 786) close STDOUT;