Update ghashv8-armx.pl from upstream.
This syncs this file up to e7ff223a20697e5a401d2d9bb7a75e699ed46633 from
upstream's OpenSSL_1_1_1-stable branch. The main change of note is the
4x loop from upstream's 7ff2fa4b9281232f0ca1db03d42a954c462ef77d,
9ee020f8dc7813db82a119058d8f57e70e7e8904,
aa7bf316980259a11dcbaf6128ed86d33dc24b97, and
603ebe03529101424670051aa0c616dc6e037b28.
Benchmarks on a Pixel 4a.
Before:
Did 14069000 AES-128-GCM (16 bytes) seal operations in 2000042us (112.5 MB/sec)
Did 6768000 AES-128-GCM (256 bytes) seal operations in 2000182us (866.2 MB/sec)
Did 1902000 AES-128-GCM (1350 bytes) seal operations in 2000479us (1283.5 MB/sec)
Did 359000 AES-128-GCM (8192 bytes) seal operations in 2003942us (1467.6 MB/sec)
Did 182000 AES-128-GCM (16384 bytes) seal operations in 2002245us (1489.3 MB/sec)
Did 13388000 AES-256-GCM (16 bytes) seal operations in 2000144us (107.1 MB/sec)
Did 6069000 AES-256-GCM (256 bytes) seal operations in 2000276us (776.7 MB/sec)
Did 1638000 AES-256-GCM (1350 bytes) seal operations in 2001076us (1105.1 MB/sec)
Did 305000 AES-256-GCM (8192 bytes) seal operations in 2000040us (1249.3 MB/sec)
Did 155000 AES-256-GCM (16384 bytes) seal operations in 2009398us (1263.8 MB/sec)
After:
Did 13837000 AES-128-GCM (16 bytes) seal operations in 2000131us (110.7 MB/sec) [-1.7%]
Did 7506000 AES-128-GCM (256 bytes) seal operations in 2000197us (960.7 MB/sec) [+10.9%]
Did 2289000 AES-128-GCM (1350 bytes) seal operations in 2000734us (1544.5 MB/sec) [+20.3%]
Did 443000 AES-128-GCM (8192 bytes) seal operations in 2000321us (1814.2 MB/sec) [+23.6%]
Did 225000 AES-128-GCM (16384 bytes) seal operations in 2002308us (1841.1 MB/sec) [+23.6%]
Did 13280000 AES-256-GCM (16 bytes) seal operations in 2000011us (106.2 MB/sec) [-0.8%]
Did 6630000 AES-256-GCM (256 bytes) seal operations in 2000229us (848.5 MB/sec) [+9.2%]
Did 1916000 AES-256-GCM (1350 bytes) seal operations in 2000373us (1293.1 MB/sec) [+17.0%]
Did 365000 AES-256-GCM (8192 bytes) seal operations in 2001519us (1493.9 MB/sec) [+19.6%]
Did 185000 AES-256-GCM (16384 bytes) seal operations in 2006588us (1510.5 MB/sec) [+19.5%]
(See cl/387919990 for some notes I made in reviewing, though likely
future me will find them incomprehensible anyway.)
Change-Id: Id386e80143611487e07b2fbfda15d0abc54ea145
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/48726
Reviewed-by: Adam Langley <agl@google.com>
diff --git a/crypto/fipsmodule/modes/asm/ghashv8-armx.pl b/crypto/fipsmodule/modes/asm/ghashv8-armx.pl
index 18b87a2..24eb773 100644
--- a/crypto/fipsmodule/modes/asm/ghashv8-armx.pl
+++ b/crypto/fipsmodule/modes/asm/ghashv8-armx.pl
@@ -1,5 +1,5 @@
#! /usr/bin/env perl
-# Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
+# Copyright 2014-2020 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
@@ -17,23 +17,31 @@
# GHASH for ARMv8 Crypto Extension, 64-bit polynomial multiplication.
#
# June 2014
-# Initial version was developed in tight cooperation with Ard Biesheuvel
-# of Linaro from bits-n-pieces from other assembly modules. Just like
-# aesv8-armx.pl this module supports both AArch32 and AArch64 execution modes.
+#
+# Initial version was developed in tight cooperation with Ard
+# Biesheuvel of Linaro from bits-n-pieces from other assembly modules.
+# Just like aesv8-armx.pl this module supports both AArch32 and
+# AArch64 execution modes.
#
# July 2014
+#
# Implement 2x aggregated reduction [see ghash-x86.pl for background
# information].
#
+# November 2017
+#
+# AArch64 register bank to "accommodate" 4x aggregated reduction and
+# improve performance by 20-70% depending on processor.
+#
# Current performance in cycles per processed byte:
#
-# PMULL[2] 32-bit NEON(*)
-# Apple A7 0.92 5.62
-# Cortex-A53 1.01 8.39
-# Cortex-A57 1.17 7.61
-# Denver 0.71 6.02
-# Mongoose 1.10 8.06
-# Kryo 1.16 8.00
+# 64-bit PMULL 32-bit PMULL 32-bit NEON(*)
+# Apple A7 0.58 0.92 5.62
+# Cortex-A53 0.85 1.01 8.39
+# Cortex-A57 0.73 1.17 7.61
+# Denver 0.51 0.65 6.02
+# Mongoose 0.65 1.10 8.06
+# Kryo 0.76 1.16 8.00
#
# (*) presented for reference/comparison purposes;
@@ -62,6 +70,7 @@
$code=<<___;
#include <openssl/arm_arch.h>
+#if __ARM_MAX_ARCH__>=7
.text
___
$code.=".arch armv8-a+crypto\n" if ($flavour =~ /64/);
@@ -129,8 +138,56 @@
vext.8 $t1,$H2,$H2,#8 @ Karatsuba pre-processing
veor $t1,$t1,$H2
vext.8 $Hhl,$t0,$t1,#8 @ pack Karatsuba pre-processed
- vst1.64 {$Hhl-$H2},[x0] @ store Htable[1..2]
+ vst1.64 {$Hhl-$H2},[x0],#32 @ store Htable[1..2]
+___
+if ($flavour =~ /64/) {
+my ($t3,$Yl,$Ym,$Yh) = map("q$_",(4..7));
+$code.=<<___;
+ @ calculate H^3 and H^4
+ vpmull.p64 $Xl,$H, $H2
+ vpmull.p64 $Yl,$H2,$H2
+ vpmull2.p64 $Xh,$H, $H2
+ vpmull2.p64 $Yh,$H2,$H2
+ vpmull.p64 $Xm,$t0,$t1
+ vpmull.p64 $Ym,$t1,$t1
+
+ vext.8 $t0,$Xl,$Xh,#8 @ Karatsuba post-processing
+ vext.8 $t1,$Yl,$Yh,#8
+ veor $t2,$Xl,$Xh
+ veor $Xm,$Xm,$t0
+ veor $t3,$Yl,$Yh
+ veor $Ym,$Ym,$t1
+ veor $Xm,$Xm,$t2
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase
+ veor $Ym,$Ym,$t3
+ vpmull.p64 $t3,$Yl,$xC2
+
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Yh#lo,$Ym#hi
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ vmov $Ym#hi,$Yl#lo
+ veor $Xl,$Xm,$t2
+ veor $Yl,$Ym,$t3
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase
+ vext.8 $t3,$Yl,$Yl,#8
+ vpmull.p64 $Xl,$Xl,$xC2
+ vpmull.p64 $Yl,$Yl,$xC2
+ veor $t2,$t2,$Xh
+ veor $t3,$t3,$Yh
+ veor $H, $Xl,$t2 @ H^3
+ veor $H2,$Yl,$t3 @ H^4
+
+ vext.8 $t0,$H, $H,#8 @ Karatsuba pre-processing
+ vext.8 $t1,$H2,$H2,#8
+ veor $t0,$t0,$H
+ veor $t1,$t1,$H2
+ vext.8 $Hhl,$t0,$t1,#8 @ pack Karatsuba pre-processed
+ vst1.64 {$H-$H2},[x0] @ store Htable[3..5]
+___
+}
+$code.=<<___;
ret
.size gcm_init_v8,.-gcm_init_v8
___
@@ -201,6 +258,10 @@
gcm_ghash_v8:
AARCH64_VALID_CALL_TARGET
___
+$code.=<<___ if ($flavour =~ /64/);
+ cmp $len,#64
+ b.hs .Lgcm_ghash_v8_4x
+___
$code.=<<___ if ($flavour !~ /64/);
vstmdb sp!,{d8-d15} @ 32-bit ABI says so
___
@@ -348,10 +409,301 @@
ret
.size gcm_ghash_v8,.-gcm_ghash_v8
___
+
+if ($flavour =~ /64/) { # 4x subroutine
+my ($I0,$j1,$j2,$j3,
+ $I1,$I2,$I3,$H3,$H34,$H4,$Yl,$Ym,$Yh) = map("q$_",(4..7,15..23));
+
+$code.=<<___;
+.type gcm_ghash_v8_4x,%function
+.align 4
+gcm_ghash_v8_4x:
+.Lgcm_ghash_v8_4x:
+ vld1.64 {$Xl},[$Xi] @ load [rotated] Xi
+ vld1.64 {$H-$H2},[$Htbl],#48 @ load twisted H, ..., H^2
+ vmov.i8 $xC2,#0xe1
+ vld1.64 {$H3-$H4},[$Htbl] @ load twisted H^3, ..., H^4
+ vshl.u64 $xC2,$xC2,#57 @ compose 0xc2.0 constant
+
+ vld1.64 {$I0-$j3},[$inp],#64
+#ifndef __ARMEB__
+ vrev64.8 $Xl,$Xl
+ vrev64.8 $j1,$j1
+ vrev64.8 $j2,$j2
+ vrev64.8 $j3,$j3
+ vrev64.8 $I0,$I0
+#endif
+ vext.8 $I3,$j3,$j3,#8
+ vext.8 $I2,$j2,$j2,#8
+ vext.8 $I1,$j1,$j1,#8
+
+ vpmull.p64 $Yl,$H,$I3 @ H·Ii+3
+ veor $j3,$j3,$I3
+ vpmull2.p64 $Yh,$H,$I3
+ vpmull.p64 $Ym,$Hhl,$j3
+
+ vpmull.p64 $t0,$H2,$I2 @ H^2·Ii+2
+ veor $j2,$j2,$I2
+ vpmull2.p64 $I2,$H2,$I2
+ vpmull2.p64 $j2,$Hhl,$j2
+
+ veor $Yl,$Yl,$t0
+ veor $Yh,$Yh,$I2
+ veor $Ym,$Ym,$j2
+
+ vpmull.p64 $j3,$H3,$I1 @ H^3·Ii+1
+ veor $j1,$j1,$I1
+ vpmull2.p64 $I1,$H3,$I1
+ vpmull.p64 $j1,$H34,$j1
+
+ veor $Yl,$Yl,$j3
+ veor $Yh,$Yh,$I1
+ veor $Ym,$Ym,$j1
+
+ subs $len,$len,#128
+ b.lo .Ltail4x
+
+ b .Loop4x
+
+.align 4
+.Loop4x:
+ veor $t0,$I0,$Xl
+ vld1.64 {$I0-$j3},[$inp],#64
+ vext.8 $IN,$t0,$t0,#8
+#ifndef __ARMEB__
+ vrev64.8 $j1,$j1
+ vrev64.8 $j2,$j2
+ vrev64.8 $j3,$j3
+ vrev64.8 $I0,$I0
+#endif
+
+ vpmull.p64 $Xl,$H4,$IN @ H^4·(Xi+Ii)
+ veor $t0,$t0,$IN
+ vpmull2.p64 $Xh,$H4,$IN
+ vext.8 $I3,$j3,$j3,#8
+ vpmull2.p64 $Xm,$H34,$t0
+
+ veor $Xl,$Xl,$Yl
+ veor $Xh,$Xh,$Yh
+ vext.8 $I2,$j2,$j2,#8
+ veor $Xm,$Xm,$Ym
+ vext.8 $I1,$j1,$j1,#8
+
+ vext.8 $t1,$Xl,$Xh,#8 @ Karatsuba post-processing
+ veor $t2,$Xl,$Xh
+ vpmull.p64 $Yl,$H,$I3 @ H·Ii+3
+ veor $j3,$j3,$I3
+ veor $Xm,$Xm,$t1
+ vpmull2.p64 $Yh,$H,$I3
+ veor $Xm,$Xm,$t2
+ vpmull.p64 $Ym,$Hhl,$j3
+
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase of reduction
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ vpmull.p64 $t0,$H2,$I2 @ H^2·Ii+2
+ veor $j2,$j2,$I2
+ vpmull2.p64 $I2,$H2,$I2
+ veor $Xl,$Xm,$t2
+ vpmull2.p64 $j2,$Hhl,$j2
+
+ veor $Yl,$Yl,$t0
+ veor $Yh,$Yh,$I2
+ veor $Ym,$Ym,$j2
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase of reduction
+ vpmull.p64 $Xl,$Xl,$xC2
+ vpmull.p64 $j3,$H3,$I1 @ H^3·Ii+1
+ veor $j1,$j1,$I1
+ veor $t2,$t2,$Xh
+ vpmull2.p64 $I1,$H3,$I1
+ vpmull.p64 $j1,$H34,$j1
+
+ veor $Xl,$Xl,$t2
+ veor $Yl,$Yl,$j3
+ veor $Yh,$Yh,$I1
+ vext.8 $Xl,$Xl,$Xl,#8
+ veor $Ym,$Ym,$j1
+
+ subs $len,$len,#64
+ b.hs .Loop4x
+
+.Ltail4x:
+ veor $t0,$I0,$Xl
+ vext.8 $IN,$t0,$t0,#8
+
+ vpmull.p64 $Xl,$H4,$IN @ H^4·(Xi+Ii)
+ veor $t0,$t0,$IN
+ vpmull2.p64 $Xh,$H4,$IN
+ vpmull2.p64 $Xm,$H34,$t0
+
+ veor $Xl,$Xl,$Yl
+ veor $Xh,$Xh,$Yh
+ veor $Xm,$Xm,$Ym
+
+ adds $len,$len,#64
+ b.eq .Ldone4x
+
+ cmp $len,#32
+ b.lo .Lone
+ b.eq .Ltwo
+.Lthree:
+ vext.8 $t1,$Xl,$Xh,#8 @ Karatsuba post-processing
+ veor $t2,$Xl,$Xh
+ veor $Xm,$Xm,$t1
+ vld1.64 {$I0-$j2},[$inp]
+ veor $Xm,$Xm,$t2
+#ifndef __ARMEB__
+ vrev64.8 $j1,$j1
+ vrev64.8 $j2,$j2
+ vrev64.8 $I0,$I0
+#endif
+
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase of reduction
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ vext.8 $I2,$j2,$j2,#8
+ vext.8 $I1,$j1,$j1,#8
+ veor $Xl,$Xm,$t2
+
+ vpmull.p64 $Yl,$H,$I2 @ H·Ii+2
+ veor $j2,$j2,$I2
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase of reduction
+ vpmull.p64 $Xl,$Xl,$xC2
+ veor $t2,$t2,$Xh
+ vpmull2.p64 $Yh,$H,$I2
+ vpmull.p64 $Ym,$Hhl,$j2
+ veor $Xl,$Xl,$t2
+ vpmull.p64 $j3,$H2,$I1 @ H^2·Ii+1
+ veor $j1,$j1,$I1
+ vext.8 $Xl,$Xl,$Xl,#8
+
+ vpmull2.p64 $I1,$H2,$I1
+ veor $t0,$I0,$Xl
+ vpmull2.p64 $j1,$Hhl,$j1
+ vext.8 $IN,$t0,$t0,#8
+
+ veor $Yl,$Yl,$j3
+ veor $Yh,$Yh,$I1
+ veor $Ym,$Ym,$j1
+
+ vpmull.p64 $Xl,$H3,$IN @ H^3·(Xi+Ii)
+ veor $t0,$t0,$IN
+ vpmull2.p64 $Xh,$H3,$IN
+ vpmull.p64 $Xm,$H34,$t0
+
+ veor $Xl,$Xl,$Yl
+ veor $Xh,$Xh,$Yh
+ veor $Xm,$Xm,$Ym
+ b .Ldone4x
+
+.align 4
+.Ltwo:
+ vext.8 $t1,$Xl,$Xh,#8 @ Karatsuba post-processing
+ veor $t2,$Xl,$Xh
+ veor $Xm,$Xm,$t1
+ vld1.64 {$I0-$j1},[$inp]
+ veor $Xm,$Xm,$t2
+#ifndef __ARMEB__
+ vrev64.8 $j1,$j1
+ vrev64.8 $I0,$I0
+#endif
+
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase of reduction
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ vext.8 $I1,$j1,$j1,#8
+ veor $Xl,$Xm,$t2
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase of reduction
+ vpmull.p64 $Xl,$Xl,$xC2
+ veor $t2,$t2,$Xh
+ veor $Xl,$Xl,$t2
+ vext.8 $Xl,$Xl,$Xl,#8
+
+ vpmull.p64 $Yl,$H,$I1 @ H·Ii+1
+ veor $j1,$j1,$I1
+
+ veor $t0,$I0,$Xl
+ vext.8 $IN,$t0,$t0,#8
+
+ vpmull2.p64 $Yh,$H,$I1
+ vpmull.p64 $Ym,$Hhl,$j1
+
+ vpmull.p64 $Xl,$H2,$IN @ H^2·(Xi+Ii)
+ veor $t0,$t0,$IN
+ vpmull2.p64 $Xh,$H2,$IN
+ vpmull2.p64 $Xm,$Hhl,$t0
+
+ veor $Xl,$Xl,$Yl
+ veor $Xh,$Xh,$Yh
+ veor $Xm,$Xm,$Ym
+ b .Ldone4x
+
+.align 4
+.Lone:
+ vext.8 $t1,$Xl,$Xh,#8 @ Karatsuba post-processing
+ veor $t2,$Xl,$Xh
+ veor $Xm,$Xm,$t1
+ vld1.64 {$I0},[$inp]
+ veor $Xm,$Xm,$t2
+#ifndef __ARMEB__
+ vrev64.8 $I0,$I0
+#endif
+
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase of reduction
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ veor $Xl,$Xm,$t2
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase of reduction
+ vpmull.p64 $Xl,$Xl,$xC2
+ veor $t2,$t2,$Xh
+ veor $Xl,$Xl,$t2
+ vext.8 $Xl,$Xl,$Xl,#8
+
+ veor $t0,$I0,$Xl
+ vext.8 $IN,$t0,$t0,#8
+
+ vpmull.p64 $Xl,$H,$IN
+ veor $t0,$t0,$IN
+ vpmull2.p64 $Xh,$H,$IN
+ vpmull.p64 $Xm,$Hhl,$t0
+
+.Ldone4x:
+ vext.8 $t1,$Xl,$Xh,#8 @ Karatsuba post-processing
+ veor $t2,$Xl,$Xh
+ veor $Xm,$Xm,$t1
+ veor $Xm,$Xm,$t2
+
+ vpmull.p64 $t2,$Xl,$xC2 @ 1st phase of reduction
+ vmov $Xh#lo,$Xm#hi @ Xh|Xm - 256-bit result
+ vmov $Xm#hi,$Xl#lo @ Xm is rotated Xl
+ veor $Xl,$Xm,$t2
+
+ vext.8 $t2,$Xl,$Xl,#8 @ 2nd phase of reduction
+ vpmull.p64 $Xl,$Xl,$xC2
+ veor $t2,$t2,$Xh
+ veor $Xl,$Xl,$t2
+ vext.8 $Xl,$Xl,$Xl,#8
+
+#ifndef __ARMEB__
+ vrev64.8 $Xl,$Xl
+#endif
+ vst1.64 {$Xl},[$Xi] @ write out Xi
+
+ ret
+.size gcm_ghash_v8_4x,.-gcm_ghash_v8_4x
+___
+
}
+}
+
$code.=<<___;
.asciz "GHASH for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
.align 2
+#endif
___
if ($flavour =~ /64/) { ######## 64-bit code
@@ -359,7 +711,8 @@
my $arg=shift;
$arg =~ m/q([0-9]+)#(lo|hi),\s*q([0-9]+)#(lo|hi)/o &&
- sprintf "ins v%d.d[%d],v%d.d[%d]",$1,($2 eq "lo")?0:1,$3,($4 eq "lo")?0:1;
+ sprintf "ins v%d.d[%d],v%d.d[%d]",$1<8?$1:$1+8,($2 eq "lo")?0:1,
+ $3<8?$3:$3+8,($4 eq "lo")?0:1;
}
foreach(split("\n",$code)) {
s/cclr\s+([wx])([^,]+),\s*([a-z]+)/csel $1$2,$1zr,$1$2,$3/o or
diff --git a/crypto/fipsmodule/modes/gcm_test.cc b/crypto/fipsmodule/modes/gcm_test.cc
index 031b06c..02ba2d1 100644
--- a/crypto/fipsmodule/modes/gcm_test.cc
+++ b/crypto/fipsmodule/modes/gcm_test.cc
@@ -125,7 +125,7 @@
UINT64_C(0x66e94bd4ef8a2c3b),
UINT64_C(0x884cfa59ca342b2e),
};
- static const size_t kBlockCounts[] = {1, 2, 3, 4, 7, 8, 15, 16, 31, 32};
+ static const size_t kBlockCounts[] = {1, 2, 3, 4, 5, 6, 7, 8, 15, 16, 31, 32};
uint8_t buf[16 * 32];
OPENSSL_memset(buf, 42, sizeof(buf));