7237N/A#!/usr/bin/env perl
7237N/A
7237N/A# ====================================================================
7237N/A# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
7237N/A# project. The module is, however, dual licensed under OpenSSL and
7237N/A# CRYPTOGAMS licenses depending on where you obtain it. For further
7237N/A# details see http://www.openssl.org/~appro/cryptogams/.
7237N/A#
7237N/A# Hardware SPARC T4 support by David S. Miller <davem@davemloft.net>.
7237N/A# ====================================================================
7237N/A
7237N/A# SHA256 performance improvement over compiler generated code varies
7237N/A# from 40% for Sun C [32-bit build] to 70% for gcc [3.3, 64-bit
7237N/A# build]. Just like in SHA1 module I aim to ensure scalability on
7237N/A# UltraSPARC T1 by packing X[16] to 8 64-bit registers.
7237N/A
7237N/A# SHA512 on pre-T1 UltraSPARC.
7237N/A#
7237N/A# Performance is >75% better than 64-bit code generated by Sun C and
7237N/A# over 2x than 32-bit code. X[16] resides on stack, but access to it
7237N/A# is scheduled for L2 latency and staged through 32 least significant
7237N/A# bits of %l0-%l7. The latter is done to achieve 32-/64-bit ABI
7237N/A# duality. Nevetheless it's ~40% faster than SHA256, which is pretty
7237N/A# good [optimal coefficient is 50%].
7237N/A#
7237N/A# SHA512 on UltraSPARC T1.
7237N/A#
7237N/A# It's not any faster than 64-bit code generated by Sun C 5.8. This is
7237N/A# because 64-bit code generator has the advantage of using 64-bit
7237N/A# loads(*) to access X[16], which I consciously traded for 32-/64-bit
7237N/A# ABI duality [as per above]. But it surpasses 32-bit Sun C generated
7237N/A# code by 60%, not to mention that it doesn't suffer from severe decay
7237N/A# when running 4 times physical cores threads and that it leaves gcc
7237N/A# [3.4] behind by over 4x factor! If compared to SHA256, single thread
7237N/A# performance is only 10% better, but overall throughput for maximum
7237N/A# amount of threads for given CPU exceeds corresponding one of SHA256
7237N/A# by 30% [again, optimal coefficient is 50%].
7237N/A#
7237N/A# (*) Unlike pre-T1 UltraSPARC loads on T1 are executed strictly
7237N/A# in-order, i.e. load instruction has to complete prior next
7237N/A# instruction in given thread is executed, even if the latter is
7237N/A# not dependent on load result! This means that on T1 two 32-bit
7237N/A# loads are always slower than one 64-bit load. Once again this
7237N/A# is unlike pre-T1 UltraSPARC, where, if scheduled appropriately,
7237N/A# 2x32-bit loads can be as fast as 1x64-bit ones.
7237N/A#
7237N/A# SPARC T4 SHA256/512 hardware achieves 3.17/2.01 cycles per byte,
7237N/A# which is 9.3x/11.1x faster than software. Multi-process benchmark
7237N/A# saturates at 11.5x single-process result on 8-core processor, or
7237N/A# ~11/16GBps per 2.85GHz socket.
7237N/A
7237N/A$output=shift;
7237N/Aopen STDOUT,">$output";
7237N/A
7237N/Aif ($output =~ /512/) {
7237N/A $label="512";
7237N/A $SZ=8;
7237N/A $LD="ldx"; # load from memory
7237N/A $ST="stx"; # store to memory
7237N/A $SLL="sllx"; # shift left logical
7237N/A $SRL="srlx"; # shift right logical
7237N/A @Sigma0=(28,34,39);
7237N/A @Sigma1=(14,18,41);
7237N/A @sigma0=( 7, 1, 8); # right shift first
7237N/A @sigma1=( 6,19,61); # right shift first
7237N/A $lastK=0x817;
7237N/A $rounds=80;
7237N/A $align=4;
7237N/A
7237N/A $locals=16*$SZ; # X[16]
7237N/A
7237N/A $A="%o0";
7237N/A $B="%o1";
7237N/A $C="%o2";
7237N/A $D="%o3";
7237N/A $E="%o4";
7237N/A $F="%o5";
7237N/A $G="%g1";
7237N/A $H="%o7";
7237N/A @V=($A,$B,$C,$D,$E,$F,$G,$H);
7237N/A} else {
7237N/A $label="256";
7237N/A $SZ=4;
7237N/A $LD="ld"; # load from memory
7237N/A $ST="st"; # store to memory
7237N/A $SLL="sll"; # shift left logical
7237N/A $SRL="srl"; # shift right logical
7237N/A @Sigma0=( 2,13,22);
7237N/A @Sigma1=( 6,11,25);
7237N/A @sigma0=( 3, 7,18); # right shift first
7237N/A @sigma1=(10,17,19); # right shift first
7237N/A $lastK=0x8f2;
7237N/A $rounds=64;
7237N/A $align=8;
7237N/A
7237N/A $locals=0; # X[16] is register resident
7237N/A @X=("%o0","%o1","%o2","%o3","%o4","%o5","%g1","%o7");
7237N/A
7237N/A $A="%l0";
7237N/A $B="%l1";
7237N/A $C="%l2";
7237N/A $D="%l3";
7237N/A $E="%l4";
7237N/A $F="%l5";
7237N/A $G="%l6";
7237N/A $H="%l7";
7237N/A @V=($A,$B,$C,$D,$E,$F,$G,$H);
7237N/A}
7237N/A$T1="%g2";
7237N/A$tmp0="%g3";
7237N/A$tmp1="%g4";
7237N/A$tmp2="%g5";
7237N/A
7237N/A$ctx="%i0";
7237N/A$inp="%i1";
7237N/A$len="%i2";
7237N/A$Ktbl="%i3";
7237N/A$tmp31="%i4";
7237N/A$tmp32="%i5";
7237N/A
7237N/A########### SHA256
7237N/A$Xload = sub {
7237N/Amy ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
7237N/A
7237N/A if ($i==0) {
7237N/A$code.=<<___;
7237N/A ldx [$inp+0],@X[0]
7237N/A ldx [$inp+16],@X[2]
7237N/A ldx [$inp+32],@X[4]
7237N/A ldx [$inp+48],@X[6]
7237N/A ldx [$inp+8],@X[1]
7237N/A ldx [$inp+24],@X[3]
7237N/A subcc %g0,$tmp31,$tmp32 ! should be 64-$tmp31, but -$tmp31 works too
7237N/A ldx [$inp+40],@X[5]
7237N/A bz,pt %icc,.Laligned
7237N/A ldx [$inp+56],@X[7]
7237N/A
7237N/A sllx @X[0],$tmp31,@X[0]
7237N/A ldx [$inp+64],$T1
7237N/A___
7237N/Afor($j=0;$j<7;$j++)
7237N/A{ $code.=<<___;
7237N/A srlx @X[$j+1],$tmp32,$tmp1
7237N/A sllx @X[$j+1],$tmp31,@X[$j+1]
7237N/A or $tmp1,@X[$j],@X[$j]
7237N/A___
7237N/A}
7237N/A$code.=<<___;
7237N/A srlx $T1,$tmp32,$T1
7237N/A or $T1,@X[7],@X[7]
7237N/A.Laligned:
7237N/A___
7237N/A }
7237N/A
7237N/A if ($i&1) {
7237N/A $code.="\tadd @X[$i/2],$h,$T1\n";
7237N/A } else {
7237N/A $code.="\tsrlx @X[$i/2],32,$T1\n\tadd $h,$T1,$T1\n";
7237N/A }
7237N/A} if ($SZ==4);
7237N/A
7237N/A########### SHA512
7237N/A$Xload = sub {
7237N/Amy ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
7237N/Amy @pair=("%l".eval(($i*2)%8),"%l".eval(($i*2)%8+1),"%l".eval((($i+1)*2)%8));
7237N/A
7237N/A$code.=<<___ if ($i==0);
7237N/A ld [$inp+0],%l0
7237N/A ld [$inp+4],%l1
7237N/A ld [$inp+8],%l2
7237N/A ld [$inp+12],%l3
7237N/A ld [$inp+16],%l4
7237N/A ld [$inp+20],%l5
7237N/A ld [$inp+24],%l6
7237N/A cmp $tmp31,0
7237N/A ld [$inp+28],%l7
7237N/A___
7237N/A$code.=<<___ if ($i<15);
7237N/A sllx @pair[1],$tmp31,$tmp2 ! Xload($i)
7237N/A add $tmp31,32,$tmp0
7237N/A sllx @pair[0],$tmp0,$tmp1
7237N/A `"ld [$inp+".eval(32+0+$i*8)."],@pair[0]" if ($i<12)`
7237N/A srlx @pair[2],$tmp32,@pair[1]
7237N/A or $tmp1,$tmp2,$tmp2
7237N/A or @pair[1],$tmp2,$tmp2
7237N/A `"ld [$inp+".eval(32+4+$i*8)."],@pair[1]" if ($i<12)`
7237N/A add $h,$tmp2,$T1
7237N/A $ST $tmp2,[%sp+STACK_BIAS+STACK_FRAME+`$i*$SZ`]
7237N/A___
7237N/A$code.=<<___ if ($i==12);
7237N/A bnz,a,pn %icc,.+8
7237N/A ld [$inp+128],%l0
7237N/A___
7237N/A$code.=<<___ if ($i==15);
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+1)%16)*$SZ+0`],%l2
7237N/A sllx @pair[1],$tmp31,$tmp2 ! Xload($i)
7237N/A add $tmp31,32,$tmp0
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+1)%16)*$SZ+4`],%l3
7237N/A sllx @pair[0],$tmp0,$tmp1
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+9)%16)*$SZ+0`],%l4
7237N/A srlx @pair[2],$tmp32,@pair[1]
7237N/A or $tmp1,$tmp2,$tmp2
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+9)%16)*$SZ+4`],%l5
7237N/A or @pair[1],$tmp2,$tmp2
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+14)%16)*$SZ+0`],%l6
7237N/A add $h,$tmp2,$T1
7237N/A $ST $tmp2,[%sp+STACK_BIAS+STACK_FRAME+`$i*$SZ`]
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+14)%16)*$SZ+4`],%l7
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+0)%16)*$SZ+0`],%l0
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+0)%16)*$SZ+4`],%l1
7237N/A___
7237N/A} if ($SZ==8);
7237N/A
7237N/A########### common
7237N/Asub BODY_00_15 {
7237N/Amy ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
7237N/A
7237N/A if ($i<16) {
7237N/A &$Xload(@_);
7237N/A } else {
7237N/A $code.="\tadd $h,$T1,$T1\n";
7237N/A }
7237N/A
7237N/A$code.=<<___;
7237N/A $SRL $e,@Sigma1[0],$h !! $i
7237N/A xor $f,$g,$tmp2
7237N/A $SLL $e,`$SZ*8-@Sigma1[2]`,$tmp1
7237N/A and $e,$tmp2,$tmp2
7237N/A $SRL $e,@Sigma1[1],$tmp0
7237N/A xor $tmp1,$h,$h
7237N/A $SLL $e,`$SZ*8-@Sigma1[1]`,$tmp1
7237N/A xor $tmp0,$h,$h
7237N/A $SRL $e,@Sigma1[2],$tmp0
7237N/A xor $tmp1,$h,$h
7237N/A $SLL $e,`$SZ*8-@Sigma1[0]`,$tmp1
7237N/A xor $tmp0,$h,$h
7237N/A xor $g,$tmp2,$tmp2 ! Ch(e,f,g)
7237N/A xor $tmp1,$h,$tmp0 ! Sigma1(e)
7237N/A
7237N/A $SRL $a,@Sigma0[0],$h
7237N/A add $tmp2,$T1,$T1
7237N/A $LD [$Ktbl+`$i*$SZ`],$tmp2 ! K[$i]
7237N/A $SLL $a,`$SZ*8-@Sigma0[2]`,$tmp1
7237N/A add $tmp0,$T1,$T1
7237N/A $SRL $a,@Sigma0[1],$tmp0
7237N/A xor $tmp1,$h,$h
7237N/A $SLL $a,`$SZ*8-@Sigma0[1]`,$tmp1
7237N/A xor $tmp0,$h,$h
7237N/A $SRL $a,@Sigma0[2],$tmp0
7237N/A xor $tmp1,$h,$h
7237N/A $SLL $a,`$SZ*8-@Sigma0[0]`,$tmp1
7237N/A xor $tmp0,$h,$h
7237N/A xor $tmp1,$h,$h ! Sigma0(a)
7237N/A
7237N/A or $a,$b,$tmp0
7237N/A and $a,$b,$tmp1
7237N/A and $c,$tmp0,$tmp0
7237N/A or $tmp0,$tmp1,$tmp1 ! Maj(a,b,c)
7237N/A add $tmp2,$T1,$T1 ! +=K[$i]
7237N/A add $tmp1,$h,$h
7237N/A
7237N/A add $T1,$d,$d
7237N/A add $T1,$h,$h
7237N/A___
7237N/A}
7237N/A
7237N/A########### SHA256
7237N/A$BODY_16_XX = sub {
7237N/Amy $i=@_[0];
7237N/Amy $xi;
7237N/A
7237N/A if ($i&1) {
7237N/A $xi=$tmp32;
7237N/A $code.="\tsrlx @X[(($i+1)/2)%8],32,$xi\n";
7237N/A } else {
7237N/A $xi=@X[(($i+1)/2)%8];
7237N/A }
7237N/A$code.=<<___;
7237N/A srl $xi,@sigma0[0],$T1 !! Xupdate($i)
7237N/A sll $xi,`32-@sigma0[2]`,$tmp1
7237N/A srl $xi,@sigma0[1],$tmp0
7237N/A xor $tmp1,$T1,$T1
7237N/A sll $tmp1,`@sigma0[2]-@sigma0[1]`,$tmp1
7237N/A xor $tmp0,$T1,$T1
7237N/A srl $xi,@sigma0[2],$tmp0
7237N/A xor $tmp1,$T1,$T1
7237N/A___
7237N/A if ($i&1) {
7237N/A $xi=@X[(($i+14)/2)%8];
7237N/A } else {
7237N/A $xi=$tmp32;
7237N/A $code.="\tsrlx @X[(($i+14)/2)%8],32,$xi\n";
7237N/A }
7237N/A$code.=<<___;
7237N/A srl $xi,@sigma1[0],$tmp2
7237N/A xor $tmp0,$T1,$T1 ! T1=sigma0(X[i+1])
7237N/A sll $xi,`32-@sigma1[2]`,$tmp1
7237N/A srl $xi,@sigma1[1],$tmp0
7237N/A xor $tmp1,$tmp2,$tmp2
7237N/A sll $tmp1,`@sigma1[2]-@sigma1[1]`,$tmp1
7237N/A xor $tmp0,$tmp2,$tmp2
7237N/A srl $xi,@sigma1[2],$tmp0
7237N/A xor $tmp1,$tmp2,$tmp2
7237N/A___
7237N/A if ($i&1) {
7237N/A $xi=@X[($i/2)%8];
7237N/A$code.=<<___;
7237N/A srlx @X[(($i+9)/2)%8],32,$tmp1 ! X[i+9]
7237N/A xor $tmp0,$tmp2,$tmp2 ! sigma1(X[i+14])
7237N/A srl @X[($i/2)%8],0,$tmp0
7237N/A add $tmp2,$tmp1,$tmp1
7237N/A add $xi,$T1,$T1 ! +=X[i]
7237N/A xor $tmp0,@X[($i/2)%8],@X[($i/2)%8]
7237N/A add $tmp1,$T1,$T1
7237N/A
7237N/A srl $T1,0,$T1
7237N/A or $T1,@X[($i/2)%8],@X[($i/2)%8]
7237N/A___
7237N/A } else {
7237N/A $xi=@X[(($i+9)/2)%8];
7237N/A$code.=<<___;
7237N/A srlx @X[($i/2)%8],32,$tmp1 ! X[i]
7237N/A xor $tmp0,$tmp2,$tmp2 ! sigma1(X[i+14])
7237N/A add $xi,$T1,$T1 ! +=X[i+9]
7237N/A add $tmp2,$tmp1,$tmp1
7237N/A srl @X[($i/2)%8],0,@X[($i/2)%8]
7237N/A add $tmp1,$T1,$T1
7237N/A
7237N/A sllx $T1,32,$tmp0
7237N/A or $tmp0,@X[($i/2)%8],@X[($i/2)%8]
7237N/A___
7237N/A }
7237N/A &BODY_00_15(@_);
7237N/A} if ($SZ==4);
7237N/A
7237N/A########### SHA512
7237N/A$BODY_16_XX = sub {
7237N/Amy $i=@_[0];
7237N/Amy @pair=("%l".eval(($i*2)%8),"%l".eval(($i*2)%8+1));
7237N/A
7237N/A$code.=<<___;
7237N/A sllx %l2,32,$tmp0 !! Xupdate($i)
7237N/A or %l3,$tmp0,$tmp0
7237N/A
7237N/A srlx $tmp0,@sigma0[0],$T1
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+1)%16)*$SZ+0`],%l2
7237N/A sllx $tmp0,`64-@sigma0[2]`,$tmp1
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+1)%16)*$SZ+4`],%l3
7237N/A srlx $tmp0,@sigma0[1],$tmp0
7237N/A xor $tmp1,$T1,$T1
7237N/A sllx $tmp1,`@sigma0[2]-@sigma0[1]`,$tmp1
7237N/A xor $tmp0,$T1,$T1
7237N/A srlx $tmp0,`@sigma0[2]-@sigma0[1]`,$tmp0
7237N/A xor $tmp1,$T1,$T1
7237N/A sllx %l6,32,$tmp2
7237N/A xor $tmp0,$T1,$T1 ! sigma0(X[$i+1])
7237N/A or %l7,$tmp2,$tmp2
7237N/A
7237N/A srlx $tmp2,@sigma1[0],$tmp1
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+14)%16)*$SZ+0`],%l6
7237N/A sllx $tmp2,`64-@sigma1[2]`,$tmp0
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+14)%16)*$SZ+4`],%l7
7237N/A srlx $tmp2,@sigma1[1],$tmp2
7237N/A xor $tmp0,$tmp1,$tmp1
7237N/A sllx $tmp0,`@sigma1[2]-@sigma1[1]`,$tmp0
7237N/A xor $tmp2,$tmp1,$tmp1
7237N/A srlx $tmp2,`@sigma1[2]-@sigma1[1]`,$tmp2
7237N/A xor $tmp0,$tmp1,$tmp1
7237N/A sllx %l4,32,$tmp0
7237N/A xor $tmp2,$tmp1,$tmp1 ! sigma1(X[$i+14])
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+9)%16)*$SZ+0`],%l4
7237N/A or %l5,$tmp0,$tmp0
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+9)%16)*$SZ+4`],%l5
7237N/A
7237N/A sllx %l0,32,$tmp2
7237N/A add $tmp1,$T1,$T1
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+0)%16)*$SZ+0`],%l0
7237N/A or %l1,$tmp2,$tmp2
7237N/A add $tmp0,$T1,$T1 ! +=X[$i+9]
7237N/A ld [%sp+STACK_BIAS+STACK_FRAME+`(($i+1+0)%16)*$SZ+4`],%l1
7237N/A add $tmp2,$T1,$T1 ! +=X[$i]
7237N/A $ST $T1,[%sp+STACK_BIAS+STACK_FRAME+`($i%16)*$SZ`]
7237N/A___
7237N/A &BODY_00_15(@_);
7237N/A} if ($SZ==8);
7237N/A
7237N/A$code.=<<___;
7237N/A#include "sparc_arch.h"
7237N/A#include <openssl/fipssyms.h>
7237N/A
7237N/A#ifdef __arch64__
7237N/A.register %g2,#scratch
7237N/A.register %g3,#scratch
7237N/A#endif
7237N/A
7237N/A.section ".text",#alloc,#execinstr
7237N/A
7237N/A.align 64
7237N/AK${label}:
7237N/A.type K${label},#object
7237N/A___
7237N/Aif ($SZ==4) {
7237N/A$code.=<<___;
7237N/A .long 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5
7237N/A .long 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5
7237N/A .long 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3
7237N/A .long 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174
7237N/A .long 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc
7237N/A .long 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da
7237N/A .long 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7
7237N/A .long 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967
7237N/A .long 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13
7237N/A .long 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85
7237N/A .long 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3
7237N/A .long 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070
7237N/A .long 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5
7237N/A .long 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3
7237N/A .long 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208
7237N/A .long 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
7237N/A___
7237N/A} else {
7237N/A$code.=<<___;
7237N/A .long 0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd
7237N/A .long 0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc
7237N/A .long 0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019
7237N/A .long 0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118
7237N/A .long 0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe
7237N/A .long 0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2
7237N/A .long 0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1
7237N/A .long 0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694
7237N/A .long 0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3
7237N/A .long 0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65
7237N/A .long 0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483
7237N/A .long 0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5
7237N/A .long 0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210
7237N/A .long 0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4
7237N/A .long 0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725
7237N/A .long 0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70
7237N/A .long 0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926
7237N/A .long 0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df
7237N/A .long 0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8
7237N/A .long 0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b
7237N/A .long 0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001
7237N/A .long 0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30
7237N/A .long 0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910
7237N/A .long 0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8
7237N/A .long 0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53
7237N/A .long 0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8
7237N/A .long 0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb
7237N/A .long 0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3
7237N/A .long 0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60
7237N/A .long 0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec
7237N/A .long 0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9
7237N/A .long 0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b
7237N/A .long 0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207
7237N/A .long 0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178
7237N/A .long 0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6
7237N/A .long 0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b
7237N/A .long 0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493
7237N/A .long 0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c
7237N/A .long 0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a
7237N/A .long 0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817
7237N/A___
7237N/A}
7237N/A$code.=<<___;
7237N/A.size K${label},.-K${label}
7237N/A
7237N/A#ifdef __PIC__
7237N/ASPARC_PIC_THUNK(%g1)
7237N/A#endif
7237N/A
7237N/A.globl sha${label}_block_data_order
7237N/A.align 32
7237N/Asha${label}_block_data_order:
7237N/A SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
7237N/A ld [%g1+4],%g1 ! OPENSSL_sparcv9cap_P[1]
7237N/A
7237N/A andcc %g1, CFR_SHA${label}, %g0
7237N/A be .Lsoftware
7237N/A nop
7237N/A___
7237N/A$code.=<<___ if ($SZ==8); # SHA512
7237N/A ldd [%o0 + 0x00], %f0 ! load context
7237N/A ldd [%o0 + 0x08], %f2
7237N/A ldd [%o0 + 0x10], %f4
7237N/A ldd [%o0 + 0x18], %f6
7237N/A ldd [%o0 + 0x20], %f8
7237N/A ldd [%o0 + 0x28], %f10
7237N/A andcc %o1, 0x7, %g0
7237N/A ldd [%o0 + 0x30], %f12
7237N/A bne,pn %icc, .Lhwunaligned
7237N/A ldd [%o0 + 0x38], %f14
7237N/A
7237N/A.Lhwaligned_loop:
7237N/A ldd [%o1 + 0x00], %f16
7237N/A ldd [%o1 + 0x08], %f18
7237N/A ldd [%o1 + 0x10], %f20
7237N/A ldd [%o1 + 0x18], %f22
7237N/A ldd [%o1 + 0x20], %f24
7237N/A ldd [%o1 + 0x28], %f26
7237N/A ldd [%o1 + 0x30], %f28
7237N/A ldd [%o1 + 0x38], %f30
7237N/A ldd [%o1 + 0x40], %f32
7237N/A ldd [%o1 + 0x48], %f34
7237N/A ldd [%o1 + 0x50], %f36
7237N/A ldd [%o1 + 0x58], %f38
7237N/A ldd [%o1 + 0x60], %f40
7237N/A ldd [%o1 + 0x68], %f42
7237N/A ldd [%o1 + 0x70], %f44
7237N/A subcc %o2, 1, %o2 ! done yet?
7237N/A ldd [%o1 + 0x78], %f46
7237N/A add %o1, 0x80, %o1
7237N/A prefetch [%o1 + 63], 20
7237N/A prefetch [%o1 + 64+63], 20
7237N/A
7237N/A .word 0x81b02860 ! SHA512
7237N/A
7237N/A bne,pt SIZE_T_CC, .Lhwaligned_loop
7237N/A nop
7237N/A
7237N/A.Lhwfinish:
7237N/A std %f0, [%o0 + 0x00] ! store context
7237N/A std %f2, [%o0 + 0x08]
7237N/A std %f4, [%o0 + 0x10]
7237N/A std %f6, [%o0 + 0x18]
7237N/A std %f8, [%o0 + 0x20]
7237N/A std %f10, [%o0 + 0x28]
7237N/A std %f12, [%o0 + 0x30]
7237N/A retl
7237N/A std %f14, [%o0 + 0x38]
7237N/A
7237N/A.align 16
7237N/A.Lhwunaligned:
7237N/A alignaddr %o1, %g0, %o1
7237N/A
7237N/A ldd [%o1 + 0x00], %f18
7237N/A.Lhwunaligned_loop:
7237N/A ldd [%o1 + 0x08], %f20
7237N/A ldd [%o1 + 0x10], %f22
7237N/A ldd [%o1 + 0x18], %f24
7237N/A ldd [%o1 + 0x20], %f26
7237N/A ldd [%o1 + 0x28], %f28
7237N/A ldd [%o1 + 0x30], %f30
7237N/A ldd [%o1 + 0x38], %f32
7237N/A ldd [%o1 + 0x40], %f34
7237N/A ldd [%o1 + 0x48], %f36
7237N/A ldd [%o1 + 0x50], %f38
7237N/A ldd [%o1 + 0x58], %f40
7237N/A ldd [%o1 + 0x60], %f42
7237N/A ldd [%o1 + 0x68], %f44
7237N/A ldd [%o1 + 0x70], %f46
7237N/A ldd [%o1 + 0x78], %f48
7237N/A subcc %o2, 1, %o2 ! done yet?
7237N/A ldd [%o1 + 0x80], %f50
7237N/A add %o1, 0x80, %o1
7237N/A prefetch [%o1 + 63], 20
7237N/A prefetch [%o1 + 64+63], 20
7237N/A
7237N/A faligndata %f18, %f20, %f16
7237N/A faligndata %f20, %f22, %f18
7237N/A faligndata %f22, %f24, %f20
7237N/A faligndata %f24, %f26, %f22
7237N/A faligndata %f26, %f28, %f24
7237N/A faligndata %f28, %f30, %f26
7237N/A faligndata %f30, %f32, %f28
7237N/A faligndata %f32, %f34, %f30
7237N/A faligndata %f34, %f36, %f32
7237N/A faligndata %f36, %f38, %f34
7237N/A faligndata %f38, %f40, %f36
7237N/A faligndata %f40, %f42, %f38
7237N/A faligndata %f42, %f44, %f40
7237N/A faligndata %f44, %f46, %f42
7237N/A faligndata %f46, %f48, %f44
7237N/A faligndata %f48, %f50, %f46
7237N/A
7237N/A .word 0x81b02860 ! SHA512
7237N/A
7237N/A bne,pt SIZE_T_CC, .Lhwunaligned_loop
7237N/A for %f50, %f50, %f18 ! %f18=%f50
7237N/A
7237N/A ba .Lhwfinish
7237N/A nop
7237N/A___
7237N/A$code.=<<___ if ($SZ==4); # SHA256
7237N/A ld [%o0 + 0x00], %f0
7237N/A ld [%o0 + 0x04], %f1
7237N/A ld [%o0 + 0x08], %f2
7237N/A ld [%o0 + 0x0c], %f3
7237N/A ld [%o0 + 0x10], %f4
7237N/A ld [%o0 + 0x14], %f5
7237N/A andcc %o1, 0x7, %g0
7237N/A ld [%o0 + 0x18], %f6
7237N/A bne,pn %icc, .Lhwunaligned
7237N/A ld [%o0 + 0x1c], %f7
7237N/A
7237N/A.Lhwloop:
7237N/A ldd [%o1 + 0x00], %f8
7237N/A ldd [%o1 + 0x08], %f10
7237N/A ldd [%o1 + 0x10], %f12
7237N/A ldd [%o1 + 0x18], %f14
7237N/A ldd [%o1 + 0x20], %f16
7237N/A ldd [%o1 + 0x28], %f18
7237N/A ldd [%o1 + 0x30], %f20
7237N/A subcc %o2, 1, %o2 ! done yet?
7237N/A ldd [%o1 + 0x38], %f22
7237N/A add %o1, 0x40, %o1
7237N/A prefetch [%o1 + 63], 20
7237N/A
7237N/A .word 0x81b02840 ! SHA256
7237N/A
7237N/A bne,pt SIZE_T_CC, .Lhwloop
7237N/A nop
7237N/A
7237N/A.Lhwfinish:
7237N/A st %f0, [%o0 + 0x00] ! store context
7237N/A st %f1, [%o0 + 0x04]
7237N/A st %f2, [%o0 + 0x08]
7237N/A st %f3, [%o0 + 0x0c]
7237N/A st %f4, [%o0 + 0x10]
7237N/A st %f5, [%o0 + 0x14]
7237N/A st %f6, [%o0 + 0x18]
7237N/A retl
7237N/A st %f7, [%o0 + 0x1c]
7237N/A
7237N/A.align 8
7237N/A.Lhwunaligned:
7237N/A alignaddr %o1, %g0, %o1
7237N/A
7237N/A ldd [%o1 + 0x00], %f10
7237N/A.Lhwunaligned_loop:
7237N/A ldd [%o1 + 0x08], %f12
7237N/A ldd [%o1 + 0x10], %f14
7237N/A ldd [%o1 + 0x18], %f16
7237N/A ldd [%o1 + 0x20], %f18
7237N/A ldd [%o1 + 0x28], %f20
7237N/A ldd [%o1 + 0x30], %f22
7237N/A ldd [%o1 + 0x38], %f24
7237N/A subcc %o2, 1, %o2 ! done yet?
7237N/A ldd [%o1 + 0x40], %f26
7237N/A add %o1, 0x40, %o1
7237N/A prefetch [%o1 + 63], 20
7237N/A
7237N/A faligndata %f10, %f12, %f8
7237N/A faligndata %f12, %f14, %f10
7237N/A faligndata %f14, %f16, %f12
7237N/A faligndata %f16, %f18, %f14
7237N/A faligndata %f18, %f20, %f16
7237N/A faligndata %f20, %f22, %f18
7237N/A faligndata %f22, %f24, %f20
7237N/A faligndata %f24, %f26, %f22
7237N/A
7237N/A .word 0x81b02840 ! SHA256
7237N/A
7237N/A bne,pt SIZE_T_CC, .Lhwunaligned_loop
7237N/A for %f26, %f26, %f10 ! %f10=%f26
7237N/A
7237N/A ba .Lhwfinish
7237N/A nop
7237N/A___
7237N/A$code.=<<___;
7237N/A.align 16
7237N/A.Lsoftware:
7237N/A save %sp,-STACK_FRAME-$locals,%sp
7237N/A and $inp,`$align-1`,$tmp31
7237N/A sllx $len,`log(16*$SZ)/log(2)`,$len
7237N/A andn $inp,`$align-1`,$inp
7237N/A sll $tmp31,3,$tmp31
7237N/A add $inp,$len,$len
7237N/A___
7237N/A$code.=<<___ if ($SZ==8); # SHA512
7237N/A mov 32,$tmp32
7237N/A sub $tmp32,$tmp31,$tmp32
7237N/A___
7237N/A$code.=<<___;
7237N/A.Lpic: call .+8
7237N/A add %o7,K${label}-.Lpic,$Ktbl
7237N/A
7237N/A $LD [$ctx+`0*$SZ`],$A
7237N/A $LD [$ctx+`1*$SZ`],$B
7237N/A $LD [$ctx+`2*$SZ`],$C
7237N/A $LD [$ctx+`3*$SZ`],$D
7237N/A $LD [$ctx+`4*$SZ`],$E
7237N/A $LD [$ctx+`5*$SZ`],$F
7237N/A $LD [$ctx+`6*$SZ`],$G
7237N/A $LD [$ctx+`7*$SZ`],$H
7237N/A
7237N/A.Lloop:
7237N/A___
7237N/Afor ($i=0;$i<16;$i++) { &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
7237N/A$code.=".L16_xx:\n";
7237N/Afor (;$i<32;$i++) { &$BODY_16_XX($i,@V); unshift(@V,pop(@V)); }
7237N/A$code.=<<___;
7237N/A and $tmp2,0xfff,$tmp2
7237N/A cmp $tmp2,$lastK
7237N/A bne .L16_xx
7237N/A add $Ktbl,`16*$SZ`,$Ktbl ! Ktbl+=16
7237N/A
7237N/A___
7237N/A$code.=<<___ if ($SZ==4); # SHA256
7237N/A $LD [$ctx+`0*$SZ`],@X[0]
7237N/A $LD [$ctx+`1*$SZ`],@X[1]
7237N/A $LD [$ctx+`2*$SZ`],@X[2]
7237N/A $LD [$ctx+`3*$SZ`],@X[3]
7237N/A $LD [$ctx+`4*$SZ`],@X[4]
7237N/A $LD [$ctx+`5*$SZ`],@X[5]
7237N/A $LD [$ctx+`6*$SZ`],@X[6]
7237N/A $LD [$ctx+`7*$SZ`],@X[7]
7237N/A
7237N/A add $A,@X[0],$A
7237N/A $ST $A,[$ctx+`0*$SZ`]
7237N/A add $B,@X[1],$B
7237N/A $ST $B,[$ctx+`1*$SZ`]
7237N/A add $C,@X[2],$C
7237N/A $ST $C,[$ctx+`2*$SZ`]
7237N/A add $D,@X[3],$D
7237N/A $ST $D,[$ctx+`3*$SZ`]
7237N/A add $E,@X[4],$E
7237N/A $ST $E,[$ctx+`4*$SZ`]
7237N/A add $F,@X[5],$F
7237N/A $ST $F,[$ctx+`5*$SZ`]
7237N/A add $G,@X[6],$G
7237N/A $ST $G,[$ctx+`6*$SZ`]
7237N/A add $H,@X[7],$H
7237N/A $ST $H,[$ctx+`7*$SZ`]
7237N/A___
7237N/A$code.=<<___ if ($SZ==8); # SHA512
7237N/A ld [$ctx+`0*$SZ+0`],%l0
7237N/A ld [$ctx+`0*$SZ+4`],%l1
7237N/A ld [$ctx+`1*$SZ+0`],%l2
7237N/A ld [$ctx+`1*$SZ+4`],%l3
7237N/A ld [$ctx+`2*$SZ+0`],%l4
7237N/A ld [$ctx+`2*$SZ+4`],%l5
7237N/A ld [$ctx+`3*$SZ+0`],%l6
7237N/A
7237N/A sllx %l0,32,$tmp0
7237N/A ld [$ctx+`3*$SZ+4`],%l7
7237N/A sllx %l2,32,$tmp1
7237N/A or %l1,$tmp0,$tmp0
7237N/A or %l3,$tmp1,$tmp1
7237N/A add $tmp0,$A,$A
7237N/A add $tmp1,$B,$B
7237N/A $ST $A,[$ctx+`0*$SZ`]
7237N/A sllx %l4,32,$tmp2
7237N/A $ST $B,[$ctx+`1*$SZ`]
7237N/A sllx %l6,32,$T1
7237N/A or %l5,$tmp2,$tmp2
7237N/A or %l7,$T1,$T1
7237N/A add $tmp2,$C,$C
7237N/A $ST $C,[$ctx+`2*$SZ`]
7237N/A add $T1,$D,$D
7237N/A $ST $D,[$ctx+`3*$SZ`]
7237N/A
7237N/A ld [$ctx+`4*$SZ+0`],%l0
7237N/A ld [$ctx+`4*$SZ+4`],%l1
7237N/A ld [$ctx+`5*$SZ+0`],%l2
7237N/A ld [$ctx+`5*$SZ+4`],%l3
7237N/A ld [$ctx+`6*$SZ+0`],%l4
7237N/A ld [$ctx+`6*$SZ+4`],%l5
7237N/A ld [$ctx+`7*$SZ+0`],%l6
7237N/A
7237N/A sllx %l0,32,$tmp0
7237N/A ld [$ctx+`7*$SZ+4`],%l7
7237N/A sllx %l2,32,$tmp1
7237N/A or %l1,$tmp0,$tmp0
7237N/A or %l3,$tmp1,$tmp1
7237N/A add $tmp0,$E,$E
7237N/A add $tmp1,$F,$F
7237N/A $ST $E,[$ctx+`4*$SZ`]
7237N/A sllx %l4,32,$tmp2
7237N/A $ST $F,[$ctx+`5*$SZ`]
7237N/A sllx %l6,32,$T1
7237N/A or %l5,$tmp2,$tmp2
7237N/A or %l7,$T1,$T1
7237N/A add $tmp2,$G,$G
7237N/A $ST $G,[$ctx+`6*$SZ`]
7237N/A add $T1,$H,$H
7237N/A $ST $H,[$ctx+`7*$SZ`]
7237N/A___
7237N/A$code.=<<___;
7237N/A add $inp,`16*$SZ`,$inp ! advance inp
7237N/A cmp $inp,$len
7237N/A bne SIZE_T_CC,.Lloop
7237N/A sub $Ktbl,`($rounds-16)*$SZ`,$Ktbl ! rewind Ktbl
7237N/A
7237N/A ret
7237N/A restore
7237N/A.type sha${label}_block_data_order,#function
7237N/A.size sha${label}_block_data_order,(.-sha${label}_block_data_order)
7237N/A.asciz "SHA${label} block transform for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
7237N/A.align 4
7237N/A___
7237N/A
7237N/A# Purpose of these subroutines is to explicitly encode VIS instructions,
7237N/A# so that one can compile the module without having to specify VIS
7237N/A# extentions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
7237N/A# Idea is to reserve for option to produce "universal" binary and let
7237N/A# programmer detect if current CPU is VIS capable at run-time.
7237N/Asub unvis {
7237N/Amy ($mnemonic,$rs1,$rs2,$rd)=@_;
7237N/Amy $ref,$opf;
7237N/Amy %visopf = ( "faligndata" => 0x048,
7237N/A "for" => 0x07c );
7237N/A
7237N/A $ref = "$mnemonic\t$rs1,$rs2,$rd";
7237N/A
7237N/A if ($opf=$visopf{$mnemonic}) {
7237N/A foreach ($rs1,$rs2,$rd) {
7237N/A return $ref if (!/%f([0-9]{1,2})/);
7237N/A $_=$1;
7237N/A if ($1>=32) {
7237N/A return $ref if ($1&1);
7237N/A # re-encode for upper double register addressing
7237N/A $_=($1|$1>>5)&31;
7237N/A }
7237N/A }
7237N/A
7237N/A return sprintf ".word\t0x%08x !%s",
7237N/A 0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
7237N/A $ref;
7237N/A } else {
7237N/A return $ref;
7237N/A }
7237N/A}
7237N/Asub unalignaddr {
7237N/Amy ($mnemonic,$rs1,$rs2,$rd)=@_;
7237N/Amy %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
7237N/Amy $ref="$mnemonic\t$rs1,$rs2,$rd";
7237N/A
7237N/A foreach ($rs1,$rs2,$rd) {
7237N/A if (/%([goli])([0-7])/) { $_=$bias{$1}+$2; }
7237N/A else { return $ref; }
7237N/A }
7237N/A return sprintf ".word\t0x%08x !%s",
7237N/A 0x81b00300|$rd<<25|$rs1<<14|$rs2,
7237N/A $ref;
7237N/A}
7237N/A
7237N/Aforeach (split("\n",$code)) {
7237N/A s/\`([^\`]*)\`/eval $1/ge;
7237N/A
7237N/A s/\b(f[^\s]*)\s+(%f[0-9]{1,2}),\s*(%f[0-9]{1,2}),\s*(%f[0-9]{1,2})/
7237N/A &unvis($1,$2,$3,$4)
7237N/A /ge;
7237N/A s/\b(alignaddr)\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/
7237N/A &unalignaddr($1,$2,$3,$4)
7237N/A /ge;
7237N/A
7237N/A print $_,"\n";
7237N/A}
7237N/A
7237N/Aclose STDOUT;