From: Andy Polyakov Date: Sun, 24 Jul 2005 12:28:04 +0000 (+0000) Subject: SHA-256/-512 x86_64 assembler module. X-Git-Tag: OpenSSL_0_9_8k^2~1909 X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=commitdiff_plain;h=2337eb58233a68b3f97cfa18f3d30f7385478f0c SHA-256/-512 x86_64 assembler module. --- diff --git a/crypto/sha/asm/sha512-x86_64.pl b/crypto/sha/asm/sha512-x86_64.pl new file mode 100755 index 0000000000..28937de0fa --- /dev/null +++ b/crypto/sha/asm/sha512-x86_64.pl @@ -0,0 +1,358 @@ +#!/usr/bin/env perl +# +# ==================================================================== +# Written by Andy Polyakov for the OpenSSL +# project. Rights for redistribution and usage in source and binary +# forms are granted according to the OpenSSL license. +# ==================================================================== +# +# sha256/512_block procedure for x86_64. +# +# 40% improvement over compiler-generated code on Opteron. No magical +# tricks, just straight implementation... I really wonder why gcc +# [being armed with inline assembler] fails to generate as fast code. +# The only thing which is cool about this module is that it's very +# same instruction sequence used for both SHA-256 and SHA-512. In +# former case the instructions operate on 32-bit operands, while in +# latter - on 64-bit ones. All I had to do is to get one flavor right, +# the other one passed the test right away:-) +# +# sha256_block runs in ~1005 cycles on Opteron, which gives you +# asymptotic performance of 64*1000/1005=63.7MBps times CPU clock +# frequency in GHz. sha512_block runs in ~1275 cycles, which results +# in 128*1000/1275=100MBps per GHz. Is there room for improvement? +# Well, if you compare it to IA-64 implementation, which maintains +# X[16] in register bank[!], tends to 4 instructions per CPU clock +# cycle and runs in 1003 cycles, 1275 is very good result for 3-way +# issue Opteron pipeline and X[16] maintained in memory. So that *if* +# there is a way to improve it, *then* the only way would be to try to +# offload X[16] updates to SSE unit, but that would require "deeper" +# loop unroll, which in turn would naturally cause size blow-up, not +# to mention increased complexity! And once again, only *if* it's +# actually possible to noticeably improve overall ILP, instruction +# level parallelism, on a given CPU implementation in this case. +# +# Special note on Intel EM64T. While Opteron CPU exhibits perfect +# perfromance ratio of 1.5 between 64- and 32-bit flavors [see above], +# [currently available] EM64T CPUs apparently are far from it. 64-bit +# version, sha512_block, is hardly faster than 32-bit one. This is +# presumably because 64-bit shifts/rotates apparently are not atomic +# instructions, but implemented in microcode. + +$output=shift; +open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output"; + +if ($output =~ /512/) { + $func="sha512_block"; + $TABLE="K512"; + $SZ=8; + @ROT=($A,$B,$C,$D,$E,$F,$G,$H)=("%rax","%rbx","%rcx","%rdx", + "%r8", "%r9", "%r10","%r11"); + ($T1,$a0,$a1,$a2)=("%r12","%r13","%r14","%r15"); + @Sigma0=(28,34,39); + @Sigma1=(14,18,41); + @sigma0=(1, 8, 7); + @sigma1=(19,61, 6); + $rounds=80; +} else { + $func="sha256_block"; + $TABLE="K256"; + $SZ=4; + @ROT=($A,$B,$C,$D,$E,$F,$G,$H)=("%eax","%ebx","%ecx","%edx", + "%r8d","%r9d","%r10d","%r11d"); + ($T1,$a0,$a1,$a2)=("%r12d","%r13d","%r14d","%r15d"); + @Sigma0=( 2,13,22); + @Sigma1=( 6,11,25); + @sigma0=( 7,18, 3); + @sigma1=(17,19,10); + $rounds=64; +} + +$ctx="%rdi"; # 1st arg +$round="%rdi"; # zaps $ctx +$inp="%rsi"; # 2nd arg +$Tbl="%rbp"; + +$_ctx="16*$SZ+0*8(%rsp)"; +$_inp="16*$SZ+1*8(%rsp)"; +$_end="16*$SZ+2*8(%rsp)"; +$_ord="16*$SZ+3*8(%rsp)"; +$_rsp="16*$SZ+4*8(%rsp)"; +$framesz="16*$SZ+5*8"; + + +sub ROUND_00_15() +{ my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_; + +$code.=<<___; + mov $e,$a0 + mov $e,$a1 + mov $f,$a2 + + ror \$$Sigma1[0],$a0 + ror \$$Sigma1[1],$a1 + xor $g,$a2 # f^g + + xor $a1,$a0 + ror \$`$Sigma1[2]-$Sigma1[1]`,$a1 + and $e,$a2 # (f^g)&e + mov $T1,`$SZ*($i&0xf)`(%rsp) + + xor $a1,$a0 # Sigma1(e) + xor $g,$a2 # Ch(e,f,g)=((f^g)&e)^g + add $h,$T1 # T1+=h + + mov $a,$h + add $a0,$T1 # T1+=Sigma1(e) + + add $a2,$T1 # T1+=Ch(e,f,g) + mov $a,$a0 + mov $a,$a1 + + ror \$$Sigma0[0],$h + ror \$$Sigma0[1],$a0 + mov $a,$a2 + add ($Tbl,$round,$SZ),$T1 # T1+=K[round] + + xor $a0,$h + ror \$`$Sigma0[2]-$Sigma0[1]`,$a0 + or $c,$a1 # a|c + + xor $a0,$h # h=Sigma0(a) + and $c,$a2 # a&c + add $T1,$d # d+=T1 + + and $b,$a1 # (a|c)&b + add $T1,$h # h+=T1 + + or $a2,$a1 # Maj(a,b,c)=((a|c)&b)|(a&c) + lea 1($round),$round # round++ + + add $a1,$h # h+=Maj(a,b,c) +___ +} + +sub ROUND_16_XX() +{ my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_; + +$code.=<<___; + mov `$SZ*(($i+1)&0xf)`(%rsp),$a0 + mov `$SZ*(($i+14)&0xf)`(%rsp),$T1 + + mov $a0,$a2 + + shr \$$sigma0[2],$a0 + ror \$$sigma0[0],$a2 + + xor $a2,$a0 + ror \$`$sigma0[1]-$sigma0[0]`,$a2 + + xor $a2,$a0 # sigma0(X[(i+1)&0xf]) + mov $T1,$a1 + + shr \$$sigma1[2],$T1 + ror \$$sigma1[0],$a1 + + xor $a1,$T1 + ror \$`$sigma1[1]-$sigma1[0]`,$a1 + + xor $a1,$T1 # sigma1(X[(i+14)&0xf]) + + add $a0,$T1 + + add `$SZ*(($i+9)&0xf)`(%rsp),$T1 + + add `$SZ*($i&0xf)`(%rsp),$T1 +___ + &ROUND_00_15(@_); +} + +$code=<<___; +.text + +.globl $func +.type $func,\@function,4 +.align 16 +$func: + push %rbx + push %rbp + push %r12 + push %r13 + push %r14 + push %r15 + mov %rsp,%rbp # copy %rsp + shl \$4,%rdx # num*16 + sub \$$framesz,%rsp + lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ + and \$-64,%rsp # align stack frame + mov $ctx,$_ctx # save ctx, 1st arg + mov $inp,$_inp # save inp, 2nd arh + mov %rdx,$_end # save end pointer, "3rd" arg + mov %ecx,$_ord # save host, 4th arg + mov %rbp,$_rsp # save copy of %rsp + + .picmeup $Tbl + lea $TABLE-.($Tbl),$Tbl + + mov $SZ*0($ctx),$A + mov $SZ*1($ctx),$B + mov $SZ*2($ctx),$C + mov $SZ*3($ctx),$D + mov $SZ*4($ctx),$E + mov $SZ*5($ctx),$F + mov $SZ*6($ctx),$G + mov $SZ*7($ctx),$H + jmp .Lloop + +.align 16 +.Lloop: + xor $round,$round +___ +if ($SZ==4) { +$code.=<<___; + cmpl \$0,$_ord + je .Ldata_order +.align 16 +.Lhost_order: +___ + + for($i=0;$i<16;$i++) { + $code.=" mov $SZ*$i($inp),$T1\n"; + &ROUND_00_15($i,@ROT); + unshift(@ROT,pop(@ROT)); + } +$code.=<<___; + jmp .Lrounds_16_xx +.align 16 +.Ldata_order: +___ +} # 256 + for($i=0;$i<16;$i++) { + $code.=" mov $SZ*$i($inp),$T1\n"; + $code.=" bswap $T1\n"; + &ROUND_00_15($i,@ROT); + unshift(@ROT,pop(@ROT)); + } +$code.=<<___; + jmp .Lrounds_16_xx +.align 16 +.Lrounds_16_xx: +___ + for(;$i<32;$i++) { + &ROUND_16_XX($i,@ROT); + unshift(@ROT,pop(@ROT)); + } + +$code.=<<___; + cmp \$$rounds,$round + jb .Lrounds_16_xx + + mov $_ctx,$ctx + lea 16*$SZ($inp),$inp + + add $SZ*0($ctx),$A + add $SZ*1($ctx),$B + add $SZ*2($ctx),$C + add $SZ*3($ctx),$D + add $SZ*4($ctx),$E + add $SZ*5($ctx),$F + add $SZ*6($ctx),$G + add $SZ*7($ctx),$H + + cmp $_end,$inp + + mov $A,$SZ*0($ctx) + mov $B,$SZ*1($ctx) + mov $C,$SZ*2($ctx) + mov $D,$SZ*3($ctx) + mov $E,$SZ*4($ctx) + mov $F,$SZ*5($ctx) + mov $G,$SZ*6($ctx) + mov $H,$SZ*7($ctx) + jb .Lloop + + mov $_rsp,%rsp + pop %r15 + pop %r14 + pop %r13 + pop %r12 + pop %rbp + pop %rbx + + ret +.size $func,.-$func +___ + +if ($SZ==4) { +$code.=<<___; +.align 64 +.type $TABLE,\@object +$TABLE: + .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 +___ +} else { +$code.=<<___; +.align 64 +.type $TABLE,\@object +$TABLE: + .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 +___ +} + +$code =~ s/\`([^\`]*)\`/eval $1/gem; +print $code; +close STDOUT;