rpm/openssl-1_1
SHA256
1
0
Fork 0
forked from pool/openssl-1_1
Code Pull Requests Activity
5c42b46766
openssl-1_1/openssl-ec-powerpc64le-Add-asm-implementation-of-felem_-squa.patch

411 lines
10 KiB
Diff
Raw Normal View History

Accepting request 1119558 from home:ohollmann:branches:security:tls - Performance enhancements for cryptography from OpenSSL 3.x [jsc#PED-5086, jsc#PED-3514] * Add patches: - openssl-ec-Use-static-linkage-on-nistp521-felem_-square-mul-.patch - openssl-ec-56-bit-Limb-Solinas-Strategy-for-secp384r1.patch - openssl-ec-powerpc64le-Add-asm-implementation-of-felem_-squa.patch - openssl-ecc-Remove-extraneous-parentheses-in-secp384r1.patch - openssl-powerpc-ecc-Fix-stack-allocation-secp384r1-asm.patch - openssl-Improve-performance-for-6x-unrolling-with-vpermxor-i.patch OBS-URL: https://build.opensuse.org/request/show/1119558 OBS-URL: https://build.opensuse.org/package/show/security:tls/openssl-1_1?expand=0&rev=148
2023-10-25 09:52:22 +02:00
From 966047ee13188e8634af25af348940acceb9316d Mon Sep 17 00:00:00 2001
From: Rohan McLure <rohanmclure@linux.ibm.com>
Date: Wed, 31 May 2023 14:32:26 +1000
Subject: [PATCH] ec: powerpc64le: Add asm implementation of felem_{square,mul}
Add an assembly implementation of felem_{square,mul}, which will be
implemented whenever Altivec support is present and the core implements
ISA 3.0 (Power 9) or greater.
Signed-off-by: Rohan McLure <rohanmclure@linux.ibm.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Todd Short <todd.short@me.com>
(Merged from https://github.com/openssl/openssl/pull/21471)
---
crypto/ec/asm/ecp_nistp384-ppc64.pl | 355 ++++++++++++++++++++++++++++++++++++
crypto/ec/build.info | 2
crypto/ec/ecp_nistp384.c | 9
3 files changed, 366 insertions(+)
create mode 100755 crypto/ec/asm/ecp_nistp384-ppc64.pl
--- /dev/null
+++ b/crypto/ec/asm/ecp_nistp384-ppc64.pl
@@ -0,0 +1,355 @@
+#! /usr/bin/env perl
+# Copyright 2023 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the Apache License 2.0 (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 Rohan McLure <rmclure@linux.ibm.com> for the OpenSSL
+# project.
+# ====================================================================
+#
+# p384 lower-level primitives for PPC64 using vector instructions.
+#
+
+use strict;
+use warnings;
+
+my $flavour = shift;
+my $output = "";
+while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
+if (!$output) {
+ $output = "-";
+}
+
+my ($xlate, $dir);
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+my $code = "";
+
+my ($sp, $outp, $savelr, $savesp) = ("r1", "r3", "r10", "r12");
+
+my $vzero = "v32";
+
+sub startproc($)
+{
+ my ($name) = @_;
+
+ $code.=<<___;
+ .globl ${name}
+ .align 5
+${name}:
+
+___
+}
+
+sub endproc($)
+{
+ my ($name) = @_;
+
+ $code.=<<___;
+ blr
+ .size ${name},.-${name}
+
+___
+}
+
+
+sub push_vrs($$)
+{
+ my ($min, $max) = @_;
+
+ my $count = $max - $min + 1;
+
+ $code.=<<___;
+ mr $savesp,$sp
+ stdu $sp,-16*`$count+1`($sp)
+
+___
+ for (my $i = $min; $i <= $max; $i++) {
+ my $mult = $max - $i + 1;
+ $code.=<<___;
+ stxv $i,-16*$mult($savesp)
+___
+
+ }
+
+ $code.=<<___;
+
+___
+}
+
+sub pop_vrs($$)
+{
+ my ($min, $max) = @_;
+
+ $code.=<<___;
+ ld $savesp,0($sp)
+___
+ for (my $i = $min; $i <= $max; $i++) {
+ my $mult = $max - $i + 1;
+ $code.=<<___;
+ lxv $i,-16*$mult($savesp)
+___
+ }
+
+ $code.=<<___;
+ mr $sp,$savesp
+
+___
+}
+
+sub load_vrs($$)
+{
+ my ($pointer, $reg_list) = @_;
+
+ for (my $i = 0; $i <= 6; $i++) {
+ my $offset = $i * 8;
+ $code.=<<___;
+ lxsd $reg_list->[$i],$offset($pointer)
+___
+ }
+
+ $code.=<<___;
+
+___
+}
+
+sub store_vrs($$)
+{
+ my ($pointer, $reg_list) = @_;
+
+ for (my $i = 0; $i <= 12; $i++) {
+ my $offset = $i * 16;
+ $code.=<<___;
+ stxv $reg_list->[$i],$offset($pointer)
+___
+ }
+
+ $code.=<<___;
+
+___
+}
+
+$code.=<<___;
+.machine "any"
+.text
+
+___
+
+{
+ # mul/square common
+ my ($t1, $t2, $t3, $t4) = ("v33", "v34", "v42", "v43");
+ my ($zero, $one) = ("r8", "r9");
+ my $out = "v51";
+
+ {
+ #
+ # p384_felem_mul
+ #
+
+ my ($in1p, $in2p) = ("r4", "r5");
+ my @in1 = map("v$_",(44..50));
+ my @in2 = map("v$_",(35..41));
+
+ startproc("p384_felem_mul");
+
+ push_vrs(52, 63);
+
+ $code.=<<___;
+ vspltisw $vzero,0
+
+___
+
+ load_vrs($in1p, \@in1);
+ load_vrs($in2p, \@in2);
+
+ $code.=<<___;
+ vmsumudm $out,$in1[0],$in2[0],$vzero
+ stxv $out,0($outp)
+
+ xxpermdi $t1,$in1[0],$in1[1],0b00
+ xxpermdi $t2,$in2[1],$in2[0],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ stxv $out,16($outp)
+
+ xxpermdi $t2,$in2[2],$in2[1],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$in1[2],$in2[0],$out
+ stxv $out,32($outp)
+
+ xxpermdi $t2,$in2[1],$in2[0],0b00
+ xxpermdi $t3,$in1[2],$in1[3],0b00
+ xxpermdi $t4,$in2[3],$in2[2],0b00
+ vmsumudm $out,$t1,$t4,$vzero
+ vmsumudm $out,$t3,$t2,$out
+ stxv $out,48($outp)
+
+ xxpermdi $t2,$in2[4],$in2[3],0b00
+ xxpermdi $t4,$in2[2],$in2[1],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$t3,$t4,$out
+ vmsumudm $out,$in1[4],$in2[0],$out
+ stxv $out,64($outp)
+
+ xxpermdi $t2,$in2[5],$in2[4],0b00
+ xxpermdi $t4,$in2[3],$in2[2],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$t3,$t4,$out
+ xxpermdi $t4,$in2[1],$in2[0],0b00
+ xxpermdi $t1,$in1[4],$in1[5],0b00
+ vmsumudm $out,$t1,$t4,$out
+ stxv $out,80($outp)
+
+ xxpermdi $t1,$in1[0],$in1[1],0b00
+ xxpermdi $t2,$in2[6],$in2[5],0b00
+ xxpermdi $t4,$in2[4],$in2[3],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$t3,$t4,$out
+ xxpermdi $t2,$in2[2],$in2[1],0b00
+ xxpermdi $t1,$in1[4],$in1[5],0b00
+ vmsumudm $out,$t1,$t2,$out
+ vmsumudm $out,$in1[6],$in2[0],$out
+ stxv $out,96($outp)
+
+ xxpermdi $t1,$in1[1],$in1[2],0b00
+ xxpermdi $t2,$in2[6],$in2[5],0b00
+ xxpermdi $t3,$in1[3],$in1[4],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$t3,$t4,$out
+ xxpermdi $t3,$in2[2],$in2[1],0b00
+ xxpermdi $t1,$in1[5],$in1[6],0b00
+ vmsumudm $out,$t1,$t3,$out
+ stxv $out,112($outp)
+
+ xxpermdi $t1,$in1[2],$in1[3],0b00
+ xxpermdi $t3,$in1[4],$in1[5],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$t3,$t4,$out
+ vmsumudm $out,$in1[6],$in2[2],$out
+ stxv $out,128($outp)
+
+ xxpermdi $t1,$in1[3],$in1[4],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ xxpermdi $t1,$in1[5],$in1[6],0b00
+ vmsumudm $out,$t1,$t4,$out
+ stxv $out,144($outp)
+
+ vmsumudm $out,$t3,$t2,$vzero
+ vmsumudm $out,$in1[6],$in2[4],$out
+ stxv $out,160($outp)
+
+ vmsumudm $out,$t1,$t2,$vzero
+ stxv $out,176($outp)
+
+ vmsumudm $out,$in1[6],$in2[6],$vzero
+ stxv $out,192($outp)
+___
+
+ endproc("p384_felem_mul");
+ }
+
+ {
+ #
+ # p384_felem_square
+ #
+
+ my ($inp) = ("r4");
+ my @in = map("v$_",(44..50));
+ my @inx2 = map("v$_",(35..41));
+
+ startproc("p384_felem_square");
+
+ push_vrs(52, 63);
+
+ $code.=<<___;
+ vspltisw $vzero,0
+
+___
+
+ load_vrs($inp, \@in);
+
+ $code.=<<___;
+ li $zero,0
+ li $one,1
+ mtvsrdd $t1,$one,$zero
+___
+
+ for (my $i = 0; $i <= 6; $i++) {
+ $code.=<<___;
+ vsld $inx2[$i],$in[$i],$t1
+___
+ }
+
+ $code.=<<___;
+ vmsumudm $out,$in[0],$in[0],$vzero
+ stxv $out,0($outp)
+
+ vmsumudm $out,$in[0],$inx2[1],$vzero
+ stxv $out,16($outp)
+
+ vmsumudm $out,$in[0],$inx2[2],$vzero
+ vmsumudm $out,$in[1],$in[1],$out
+ stxv $out,32($outp)
+
+ xxpermdi $t1,$in[0],$in[1],0b00
+ xxpermdi $t2,$inx2[3],$inx2[2],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ stxv $out,48($outp)
+
+ xxpermdi $t4,$inx2[4],$inx2[3],0b00
+ vmsumudm $out,$t1,$t4,$vzero
+ vmsumudm $out,$in[2],$in[2],$out
+ stxv $out,64($outp)
+
+ xxpermdi $t2,$inx2[5],$inx2[4],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$in[2],$inx2[3],$out
+ stxv $out,80($outp)
+
+ xxpermdi $t2,$inx2[6],$inx2[5],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$in[2],$inx2[4],$out
+ vmsumudm $out,$in[3],$in[3],$out
+ stxv $out,96($outp)
+
+ xxpermdi $t3,$in[1],$in[2],0b00
+ vmsumudm $out,$t3,$t2,$vzero
+ vmsumudm $out,$in[3],$inx2[4],$out
+ stxv $out,112($outp)
+
+ xxpermdi $t1,$in[2],$in[3],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ vmsumudm $out,$in[4],$in[4],$out
+ stxv $out,128($outp)
+
+ xxpermdi $t1,$in[3],$in[4],0b00
+ vmsumudm $out,$t1,$t2,$vzero
+ stxv $out,144($outp)
+
+ vmsumudm $out,$in[4],$inx2[6],$vzero
+ vmsumudm $out,$in[5],$in[5],$out
+ stxv $out,160($outp)
+
+ vmsumudm $out,$in[5],$inx2[6],$vzero
+ stxv $out,176($outp)
+
+ vmsumudm $out,$in[6],$in[6],$vzero
+ stxv $out,192($outp)
+___
+
+ endproc("p384_felem_square");
+ }
+}
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+print $code;
+close STDOUT or die "error closing STDOUT: $!";
--- a/crypto/ec/build.info
+++ b/crypto/ec/build.info
@@ -31,6 +31,8 @@ GENERATE[ecp_nistz256-armv8.S]=asm/ecp_n
INCLUDE[ecp_nistz256-armv8.o]=..
GENERATE[ecp_nistz256-ppc64.s]=asm/ecp_nistz256-ppc64.pl $(PERLASM_SCHEME)
+GENERATE[ecp_nistp384-ppc64.s]=asm/ecp_nistp384-ppc64.pl $(PERLASM_SCHEME)
+INCLUDE[ecp_nistp384.o]=..
GENERATE[ecp_nistp521-ppc64.s]=asm/ecp_nistp521-ppc64.pl $(PERLASM_SCHEME)
GENERATE[x25519-x86_64.s]=asm/x25519-x86_64.pl $(PERLASM_SCHEME)
--- a/crypto/ec/ecp_nistp384.c
+++ b/crypto/ec/ecp_nistp384.c
@@ -691,6 +691,15 @@ void p384_felem_mul(widefelem out, const
static void felem_select(void)
{
+# if defined(_ARCH_PPC64)
+ if ((OPENSSL_ppccap_P & PPC_MADD300) && (OPENSSL_ppccap_P & PPC_ALTIVEC)) {
+ felem_square_p = p384_felem_square;
+ felem_mul_p = p384_felem_mul;
+
+ return;
+ }
+# endif
+
/* Default */
felem_square_p = felem_square_ref;
felem_mul_p = felem_mul_ref;
Copy Permalink