openssl/crypto/modes/asm/ghash-armv4.pl
Andy Polyakov 2d22e08083 ARM assembler pack: reschedule instructions for dual-issue pipeline.
Modest improvement coefficients mean that code already had some
parallelism and there was not very much room for improvement. Special
thanks to Ted Krovetz for benchmarking the code with such patience.
2010-07-13 14:03:31 +00:00

283 lines
7.5 KiB
Prolog

#!/usr/bin/env perl
#
# ====================================================================
# 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/.
# ====================================================================
#
# April 2010
#
# The module implements "4-bit" GCM GHASH function and underlying
# single multiplication operation in GF(2^128). "4-bit" means that it
# uses 256 bytes per-key table [+32 bytes shared table]. There is no
# experimental performance data available yet. The only approximation
# that can be made at this point is based on code size. Inner loop is
# 32 instructions long and on single-issue core should execute in <40
# cycles. Having verified that gcc 3.4 didn't unroll corresponding
# loop, this assembler loop body was found to be ~3x smaller than
# compiler-generated one...
#
# July 2010
#
# Rescheduling for dual-issue pipeline resulted in 8.5% improvement on
# Cortex A8 core and ~25 cycles per processed byte (which was observed
# to be ~3 times faster than gcc-generated code:-)
#
# Note about "528B" variant. In ARM case it makes lesser sense to
# implement it for following reasons:
#
# - performance improvement won't be anywhere near 50%, because 128-
# bit shift operation is neatly fused with 128-bit xor here, and
# "538B" variant would eliminate only 4-5 instructions out of 32
# in the inner loop (meaning that estimated improvement is ~15%);
# - ARM-based systems are often embedded ones and extra memory
# consumption might be unappreciated (for so little improvement);
#
# Byte order [in]dependence. =========================================
#
# Caller is expected to maintain specific *dword* order in Htable,
# namely with *least* significant dword of 128-bit value at *lower*
# address. This differs completely from C code and has everything to
# do with ldm instruction and order in which dwords are "consumed" by
# algorithm. *Byte* order within these dwords in turn is whatever
# *native* byte order on current platform. See gcm128.c for working
# example...
while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
$Xi="r0"; # argument block
$Htbl="r1";
$inp="r2";
$len="r3";
$Zll="r4"; # variables
$Zlh="r5";
$Zhl="r6";
$Zhh="r7";
$Tll="r8";
$Tlh="r9";
$Thl="r10";
$Thh="r11";
$nlo="r12";
################# r13 is stack pointer
$nhi="r14";
################# r15 is program counter
$rem_4bit=$inp; # used in gcm_gmult_4bit
$cnt=$len;
sub Zsmash() {
my $i=12;
my @args=@_;
for ($Zll,$Zlh,$Zhl,$Zhh) {
# can be reduced to single "str $_,[$Xi,$i]" on big-endian platforms
$code.=<<___;
mov $Tlh,$_,lsr#8
strb $_,[$Xi,#$i+3]
mov $Thl,$_,lsr#16
strb $Tlh,[$Xi,#$i+2]
mov $Thh,$_,lsr#24
strb $Thl,[$Xi,#$i+1]
strb $Thh,[$Xi,#$i]
___
$code.="\t".shift(@args)."\n";
$i-=4;
}
}
$code=<<___;
.text
.code 32
.type rem_4bit,%object
.align 5
rem_4bit:
.short 0x0000,0x1C20,0x3840,0x2460
.short 0x7080,0x6CA0,0x48C0,0x54E0
.short 0xE100,0xFD20,0xD940,0xC560
.short 0x9180,0x8DA0,0xA9C0,0xB5E0
.size rem_4bit,.-rem_4bit
.type rem_4bit_get,%function
rem_4bit_get:
sub $rem_4bit,pc,#8
sub $rem_4bit,$rem_4bit,#32 @ &rem_4bit
b .Lrem_4bit_got
nop
.size rem_4bit_get,.-rem_4bit_get
.global gcm_ghash_4bit
.type gcm_ghash_4bit,%function
gcm_ghash_4bit:
sub r12,pc,#8
add $len,$inp,$len @ $len to point at the end
stmdb sp!,{r3-r11,lr} @ save $len/end too
sub r12,r12,#48 @ &rem_4bit
ldmia r12,{r4-r11} @ copy rem_4bit ...
stmdb sp!,{r4-r11} @ ... to stack
ldrb $nlo,[$inp,#15]
ldrb $nhi,[$Xi,#15]
.Louter:
eor $nlo,$nlo,$nhi
and $nhi,$nlo,#0xf0
and $nlo,$nlo,#0x0f
mov $cnt,#14
add $Zhh,$Htbl,$nlo,lsl#4
ldmia $Zhh,{$Zll-$Zhh} @ load Htbl[nlo]
add $Thh,$Htbl,$nhi
ldrb $nlo,[$inp,#14]
and $nhi,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nhi]
add $nhi,$nhi,$nhi
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[sp,$nhi] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
ldrb $nhi,[$Xi,#14]
eor $Zlh,$Tlh,$Zlh,lsr#4
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eor $Zhh,$Thh,$Zhh,lsr#4
eor $nlo,$nlo,$nhi
and $nhi,$nlo,#0xf0
and $nlo,$nlo,#0x0f
eor $Zhh,$Zhh,$Tll,lsl#16
.Loop:
add $Thh,$Htbl,$nlo,lsl#4
subs $cnt,$cnt,#1
and $nlo,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nlo]
add $nlo,$nlo,$nlo
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[sp,$nlo] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
eor $Zlh,$Tlh,$Zlh,lsr#4
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eor $Zhh,$Thh,$Zhh,lsr#4
ldrplb $nlo,[$inp,$cnt]
add $Thh,$Htbl,$nhi
eor $Zhh,$Zhh,$Tll,lsl#16 @ ^= rem_4bit[rem]
and $nhi,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nhi]
add $nhi,$nhi,$nhi
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[sp,$nhi] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
eor $Zlh,$Tlh,$Zlh,lsr#4
ldrplb $nhi,[$Xi,$cnt]
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eorpl $nlo,$nlo,$nhi
eor $Zhh,$Thh,$Zhh,lsr#4
andpl $nhi,$nlo,#0xf0
andpl $nlo,$nlo,#0x0f
eor $Zhh,$Zhh,$Tll,lsl#16 @ ^= rem_4bit[rem]
bpl .Loop
ldr $len,[sp,#32] @ re-load $len/end
add $inp,$inp,#16
mov $nhi,$Zll
___
&Zsmash("cmp\t$inp,$len","ldrneb\t$nlo,[$inp,#15]");
$code.=<<___;
bne .Louter
add sp,sp,#36
ldmia sp!,{r4-r11,lr}
tst lr,#1
moveq pc,lr @ be binary compatible with V4, yet
bx lr @ interoperable with Thumb ISA:-)
.size gcm_ghash_4bit,.-gcm_ghash_4bit
.global gcm_gmult_4bit
.type gcm_gmult_4bit,%function
gcm_gmult_4bit:
stmdb sp!,{r4-r11,lr}
ldrb $nlo,[$Xi,#15]
b rem_4bit_get
.Lrem_4bit_got:
and $nhi,$nlo,#0xf0
and $nlo,$nlo,#0x0f
mov $cnt,#14
add $Zhh,$Htbl,$nlo,lsl#4
ldmia $Zhh,{$Zll-$Zhh} @ load Htbl[nlo]
ldrb $nlo,[$Xi,#14]
add $Thh,$Htbl,$nhi
and $nhi,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nhi]
add $nhi,$nhi,$nhi
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[$rem_4bit,$nhi] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
eor $Zlh,$Tlh,$Zlh,lsr#4
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eor $Zhh,$Thh,$Zhh,lsr#4
and $nhi,$nlo,#0xf0
eor $Zhh,$Zhh,$Tll,lsl#16
and $nlo,$nlo,#0x0f
.Loop2:
add $Thh,$Htbl,$nlo,lsl#4
subs $cnt,$cnt,#1
and $nlo,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nlo]
add $nlo,$nlo,$nlo
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[$rem_4bit,$nlo] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
eor $Zlh,$Tlh,$Zlh,lsr#4
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eor $Zhh,$Thh,$Zhh,lsr#4
ldrplb $nlo,[$Xi,$cnt]
add $Thh,$Htbl,$nhi
eor $Zhh,$Zhh,$Tll,lsl#16 @ ^= rem_4bit[rem]
and $nhi,$Zll,#0xf @ rem
ldmia $Thh,{$Tll-$Thh} @ load Htbl[nhi]
add $nhi,$nhi,$nhi
eor $Zll,$Tll,$Zll,lsr#4
ldrh $Tll,[$rem_4bit,$nhi] @ rem_4bit[rem]
eor $Zll,$Zll,$Zlh,lsl#28
eor $Zlh,$Tlh,$Zlh,lsr#4
eor $Zlh,$Zlh,$Zhl,lsl#28
eor $Zhl,$Thl,$Zhl,lsr#4
eor $Zhl,$Zhl,$Zhh,lsl#28
eor $Zhh,$Thh,$Zhh,lsr#4
andpl $nhi,$nlo,#0xf0
andpl $nlo,$nlo,#0x0f
eor $Zhh,$Zhh,$Tll,lsl#16 @ ^= rem_4bit[rem]
bpl .Loop2
___
&Zsmash();
$code.=<<___;
ldmia sp!,{r4-r11,lr}
tst lr,#1
moveq pc,lr @ be binary compatible with V4, yet
bx lr @ interoperable with Thumb ISA:-)
.size gcm_gmult_4bit,.-gcm_gmult_4bit
.asciz "GHASH for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
.align 2
___
$code =~ s/\`([^\`]*)\`/eval $1/gem;
$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm; # make it possible to compile with -march=armv4
print $code;
close STDOUT; # enforce flush