-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2012-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License"). You may not use
+# this file except in compliance with the License. You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
# ====================================================================
-# Written by Andy Polyakov <appro@opensl.org> for the OpenSSL
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
+#
+# Hardware SPARC T4 support by David S. Miller <davem@davemloft.net>.
# ====================================================================
-# MD5 for SPARCv9, 7.5 cycles per byte on UltraSPARC, >40% faster than
+# MD5 for SPARCv9, 6.9 cycles per byte on UltraSPARC, >40% faster than
# code generated by Sun C 5.2.
-$bits=32;
-for (@ARGV) { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
-if ($bits==64) { $bias=2047; $frame=192; }
-else { $bias=0; $frame=112; }
+# SPARC T4 MD5 hardware achieves 3.20 cycles per byte, which is 2.1x
+# faster than software. Multi-process benchmark saturates at 12x
+# single-process result on 8-core processor, or ~11GBps per 2.85GHz
+# socket.
-$output=shift;
+$output=pop;
open STDOUT,">$output";
use integer;
___
}
-$code.=<<___ if ($bits==64);
+$code.=<<___;
+#include "sparc_arch.h"
+
+#ifdef __arch64__
.register %g2,#scratch
.register %g3,#scratch
-___
-$code.=<<___;
+#endif
+
.section ".text",#alloc,#execinstr
+#ifdef __PIC__
+SPARC_PIC_THUNK(%g1)
+#endif
+
.globl md5_block_asm_data_order
.align 32
md5_block_asm_data_order:
- save %sp,-$frame,%sp
+ SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
+ ld [%g1+4],%g1 ! OPENSSL_sparcv9cap_P[1]
+
+ andcc %g1, CFR_MD5, %g0
+ be .Lsoftware
+ nop
+
+ mov 4, %g1
+ andcc %o1, 0x7, %g0
+ lda [%o0 + %g0]0x88, %f0 ! load context
+ lda [%o0 + %g1]0x88, %f1
+ add %o0, 8, %o0
+ lda [%o0 + %g0]0x88, %f2
+ lda [%o0 + %g1]0x88, %f3
+ bne,pn %icc, .Lhwunaligned
+ sub %o0, 8, %o0
+
+.Lhw_loop:
+ ldd [%o1 + 0x00], %f8
+ ldd [%o1 + 0x08], %f10
+ ldd [%o1 + 0x10], %f12
+ ldd [%o1 + 0x18], %f14
+ ldd [%o1 + 0x20], %f16
+ ldd [%o1 + 0x28], %f18
+ ldd [%o1 + 0x30], %f20
+ subcc %o2, 1, %o2 ! done yet?
+ ldd [%o1 + 0x38], %f22
+ add %o1, 0x40, %o1
+ prefetch [%o1 + 63], 20
+
+ .word 0x81b02800 ! MD5
+
+ bne,pt SIZE_T_CC, .Lhw_loop
+ nop
+
+.Lhwfinish:
+ sta %f0, [%o0 + %g0]0x88 ! store context
+ sta %f1, [%o0 + %g1]0x88
+ add %o0, 8, %o0
+ sta %f2, [%o0 + %g0]0x88
+ sta %f3, [%o0 + %g1]0x88
+ retl
+ nop
+
+.align 8
+.Lhwunaligned:
+ alignaddr %o1, %g0, %o1
+
+ ldd [%o1 + 0x00], %f10
+.Lhwunaligned_loop:
+ ldd [%o1 + 0x08], %f12
+ ldd [%o1 + 0x10], %f14
+ ldd [%o1 + 0x18], %f16
+ ldd [%o1 + 0x20], %f18
+ ldd [%o1 + 0x28], %f20
+ ldd [%o1 + 0x30], %f22
+ ldd [%o1 + 0x38], %f24
+ subcc %o2, 1, %o2 ! done yet?
+ ldd [%o1 + 0x40], %f26
+ add %o1, 0x40, %o1
+ prefetch [%o1 + 63], 20
+
+ faligndata %f10, %f12, %f8
+ faligndata %f12, %f14, %f10
+ faligndata %f14, %f16, %f12
+ faligndata %f16, %f18, %f14
+ faligndata %f18, %f20, %f16
+ faligndata %f20, %f22, %f18
+ faligndata %f22, %f24, %f20
+ faligndata %f24, %f26, %f22
+
+ .word 0x81b02800 ! MD5
+
+ bne,pt SIZE_T_CC, .Lhwunaligned_loop
+ for %f26, %f26, %f10 ! %f10=%f26
+
+ ba .Lhwfinish
+ nop
+
+.align 16
+.Lsoftware:
+ save %sp,-STACK_FRAME,%sp
rd %asi,$saved_asi
wr %g0,0x88,%asi ! ASI_PRIMARY_LITTLE
add $t2,$C,$C
add $CD,$D,$D
srl $B,0,$B ! clruw $B
- bne `$bits==64?"%xcc":"%icc"`,.Loop
+ bne SIZE_T_CC,.Loop
srl $D,0,$D ! clruw $D
st $A,[$ctx+0] ! write out ctx
.align 4
___
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
-print $code;
+# Purpose of these subroutines is to explicitly encode VIS instructions,
+# so that one can compile the module without having to specify VIS
+# extensions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
+# Idea is to reserve for option to produce "universal" binary and let
+# programmer detect if current CPU is VIS capable at run-time.
+sub unvis {
+my ($mnemonic,$rs1,$rs2,$rd)=@_;
+my $ref,$opf;
+my %visopf = ( "faligndata" => 0x048,
+ "for" => 0x07c );
+
+ $ref = "$mnemonic\t$rs1,$rs2,$rd";
+
+ if ($opf=$visopf{$mnemonic}) {
+ foreach ($rs1,$rs2,$rd) {
+ return $ref if (!/%f([0-9]{1,2})/);
+ $_=$1;
+ if ($1>=32) {
+ return $ref if ($1&1);
+ # re-encode for upper double register addressing
+ $_=($1|$1>>5)&31;
+ }
+ }
+
+ return sprintf ".word\t0x%08x !%s",
+ 0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
+ $ref;
+ } else {
+ return $ref;
+ }
+}
+sub unalignaddr {
+my ($mnemonic,$rs1,$rs2,$rd)=@_;
+my %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
+my $ref="$mnemonic\t$rs1,$rs2,$rd";
+
+ foreach ($rs1,$rs2,$rd) {
+ if (/%([goli])([0-7])/) { $_=$bias{$1}+$2; }
+ else { return $ref; }
+ }
+ return sprintf ".word\t0x%08x !%s",
+ 0x81b00300|$rd<<25|$rs1<<14|$rs2,
+ $ref;
+}
+
+foreach (split("\n",$code)) {
+ s/\`([^\`]*)\`/eval $1/ge;
+
+ s/\b(f[^\s]*)\s+(%f[0-9]{1,2}),\s*(%f[0-9]{1,2}),\s*(%f[0-9]{1,2})/
+ &unvis($1,$2,$3,$4)
+ /ge;
+ s/\b(alignaddr)\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/
+ &unalignaddr($1,$2,$3,$4)
+ /ge;
+
+ print $_,"\n";
+}
+
close STDOUT;
projects / openssl.git / blobdiff