kris/boringssl
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
ef793f4b6f
boringssl/crypto/bn/asm/armv4-mont.pl

695 lines
18 KiB
Perl
Raw Normal View History

Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
#!/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/.
# ====================================================================
# January 2007.
# Montgomery multiplication for ARMv4.
#
# Performance improvement naturally varies among CPU implementations
# and compilers. The code was observed to provide +65-35% improvement
# [depending on key length, less for longer keys] on ARM920T, and
# +115-80% on Intel IXP425. This is compared to pre-bn_mul_mont code
# base and compiler generated code with in-lined umull and even umlal
# instructions. The latter means that this code didn't really have an
# "advantage" of utilizing some "secret" instruction.
#
# The code is interoperable with Thumb ISA and is rather compact, less
# than 1/2KB. Windows CE port would be trivial, as it's exclusively
# about decorations, ABI and instruction syntax are identical.
# November 2013
#
# Add NEON code path, which handles lengths divisible by 8. RSA/DSA
# performance improvement on Cortex-A8 is ~45-100% depending on key
# length, more for longer keys. On Cortex-A15 the span is ~10-105%.
# On Snapdragon S4 improvement was measured to vary from ~70% to
# incredible ~380%, yes, 4.8x faster, for RSA4096 sign. But this is
# rather because original integer-only code seems to perform
# suboptimally on S4. Situation on Cortex-A9 is unfortunately
# different. It's being looked into, but the trouble is that
# performance for vectors longer than 256 bits is actually couple
# of percent worse than for integer-only code. The code is chosen
# for execution on all NEON-capable processors, because gain on
# others outweighs the marginal loss on Cortex-A9.
Add assembly support for 32-bit iOS. (Imported from upstream's 313e6ec11fb8a7bda1676ce5804bee8755664141) BUG=338886 Change-Id: Id635e78b9afaad5ca311e3aeed888c9aedeb9637 Reviewed-on: https://boringssl-review.googlesource.com/4490 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:07:38 +01:00
$flavour = shift;
if ($flavour=~/^\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
else { while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {} }
if ($flavour && $flavour ne "void") {
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";
open STDOUT,"| \"$^X\" $xlate $flavour $output";
} else {
open STDOUT,">$output";
}
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
$num="r0"; # starts as num argument, but holds &tp[num-1]
$ap="r1";
$bp="r2"; $bi="r2"; $rp="r2";
$np="r3";
$tp="r4";
$aj="r5";
$nj="r6";
$tj="r7";
$n0="r8";
########### # r9 is reserved by ELF as platform specific, e.g. TLS pointer
$alo="r10"; # sl, gcc uses it to keep @GOT
$ahi="r11"; # fp
$nlo="r12"; # ip
########### # r13 is stack pointer
$nhi="r14"; # lr
########### # r15 is program counter
#### argument block layout relative to &tp[num-1], a.k.a. $num
$_rp="$num,#12*4";
# ap permanently resides in r1
$_bp="$num,#13*4";
# np permanently resides in r3
$_n0="$num,#14*4";
$_num="$num,#15*4"; $_bpend=$_num;
$code=<<___;
Move arm_arch.h and fix up lots of include paths. arm_arch.h is included from ARM asm files, but lives in crypto/, not openssl/include/. Since the asm files are often built from a different location than their position in the source tree, relative include paths are unlikely to work so, rather than having crypto/ be a de-facto, second global include path, this change moves arm_arch.h to include/openssl/. It also removes entries from many include paths because they should be needed as relative includes are always based on the locations of the source file. Change-Id: I638ff43d641ca043a4fc06c0d901b11c6ff73542 Reviewed-on: https://boringssl-review.googlesource.com/5746 Reviewed-by: Adam Langley <agl@google.com>
2015-08-25 02:03:17 +01:00
#include <openssl/arm_arch.h>
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.text
.code 32
Remove inconsistency in ARM support. This facilitates "universal" builds, ones that target multiple architectures, e.g. ARMv5 through ARMv7. (Imported from upstream's c1669e1c205dc8e695fb0c10a655f434e758b9f7) This is a change from a while ago which was a source of divergence between our perlasm and upstream's. This change in upstream came with the following comment in Configure: Note that -march is not among compiler options in below linux-armv4 target line. Not specifying one is intentional to give you choice to: a) rely on your compiler default by not specifying one; b) specify your target platform explicitly for optimal performance, e.g. -march=armv6 or -march=armv7-a; c) build "universal" binary that targets *range* of platforms by specifying minimum and maximum supported architecture; As for c) option. It actually makes no sense to specify maximum to be less than ARMv7, because it's the least requirement for run-time switch between platform-specific code paths. And without run-time switch performance would be equivalent to one for minimum. Secondly, there are some natural limitations that you'd have to accept and respect. Most notably you can *not* build "universal" binary for big-endian platform. This is because ARMv7 processor always picks instructions in little-endian order. Another similar limitation is that -mthumb can't "cross" -march=armv6t2 boundary, because that's where it became Thumb-2. Well, this limitation is a bit artificial, because it's not really impossible, but it's deemed too tricky to support. And of course you have to be sure that your binutils are actually up to the task of handling maximum target platform. Change-Id: Ie5f674d603393f0a1354a0d0973987484a4a650c Reviewed-on: https://boringssl-review.googlesource.com/4488 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:27:38 +01:00
#if __ARM_MAX_ARCH__>=7
Move .align directives next to their labels for ARM. 2ab24a2d40fd193a25d7c94072596a1180cbb460 added sections to ARM assembly files. However, in cases where .align directives were not next to the labels that they were intended to apply to, the section directives would cause them to be ignored. Change-Id: I32117f6747ff8545b80c70dd3b8effdc6e6f67e0 Reviewed-on: https://boringssl-review.googlesource.com/6050 Reviewed-by: David Benjamin <davidben@chromium.org> Reviewed-by: Adam Langley <agl@google.com>
2015-09-30 18:38:38 +01:00
.align 5
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.LOPENSSL_armcap:
Add assembly support for 32-bit iOS. (Imported from upstream's 313e6ec11fb8a7bda1676ce5804bee8755664141) BUG=338886 Change-Id: Id635e78b9afaad5ca311e3aeed888c9aedeb9637 Reviewed-on: https://boringssl-review.googlesource.com/4490 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:07:38 +01:00
.word OPENSSL_armcap_P-.Lbn_mul_mont
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
#endif
.global bn_mul_mont
Add visibility rules. This change marks public symbols as dynamically exported. This means that it becomes viable to build a shared library of libcrypto and libssl with -fvisibility=hidden. On Windows, one not only needs to mark functions for export in a component, but also for import when using them from a different component. Because of this we have to build with |BORINGSSL_IMPLEMENTATION| defined when building the code. Other components, when including our headers, won't have that defined and then the |OPENSSL_EXPORT| tag becomes an import tag instead. See the #defines in base.h In the asm code, symbols are now hidden by default and those that need to be exported are wrapped by a C function. In order to support Chromium, a couple of libssl functions were moved to ssl.h from ssl_locl.h: ssl_get_new_session and ssl_update_cache. Change-Id: Ib4b76e2f1983ee066e7806c24721e8626d08a261 Reviewed-on: https://boringssl-review.googlesource.com/1350 Reviewed-by: Adam Langley <agl@google.com>
2014-07-31 00:02:14 +01:00
.hidden bn_mul_mont
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.type bn_mul_mont,%function
.align 5
bn_mul_mont:
Add assembly support for 32-bit iOS. (Imported from upstream's 313e6ec11fb8a7bda1676ce5804bee8755664141) BUG=338886 Change-Id: Id635e78b9afaad5ca311e3aeed888c9aedeb9637 Reviewed-on: https://boringssl-review.googlesource.com/4490 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:07:38 +01:00
.Lbn_mul_mont:
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
ldr ip,[sp,#4] @ load num
stmdb sp!,{r0,r2} @ sp points at argument block
Remove inconsistency in ARM support. This facilitates "universal" builds, ones that target multiple architectures, e.g. ARMv5 through ARMv7. (Imported from upstream's c1669e1c205dc8e695fb0c10a655f434e758b9f7) This is a change from a while ago which was a source of divergence between our perlasm and upstream's. This change in upstream came with the following comment in Configure: Note that -march is not among compiler options in below linux-armv4 target line. Not specifying one is intentional to give you choice to: a) rely on your compiler default by not specifying one; b) specify your target platform explicitly for optimal performance, e.g. -march=armv6 or -march=armv7-a; c) build "universal" binary that targets *range* of platforms by specifying minimum and maximum supported architecture; As for c) option. It actually makes no sense to specify maximum to be less than ARMv7, because it's the least requirement for run-time switch between platform-specific code paths. And without run-time switch performance would be equivalent to one for minimum. Secondly, there are some natural limitations that you'd have to accept and respect. Most notably you can *not* build "universal" binary for big-endian platform. This is because ARMv7 processor always picks instructions in little-endian order. Another similar limitation is that -mthumb can't "cross" -march=armv6t2 boundary, because that's where it became Thumb-2. Well, this limitation is a bit artificial, because it's not really impossible, but it's deemed too tricky to support. And of course you have to be sure that your binutils are actually up to the task of handling maximum target platform. Change-Id: Ie5f674d603393f0a1354a0d0973987484a4a650c Reviewed-on: https://boringssl-review.googlesource.com/4488 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:27:38 +01:00
#if __ARM_MAX_ARCH__>=7
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
tst ip,#7
bne .Lialu
adr r0,bn_mul_mont
ldr r2,.LOPENSSL_armcap
ldr r0,[r0,r2]
Add assembly support for 32-bit iOS. (Imported from upstream's 313e6ec11fb8a7bda1676ce5804bee8755664141) BUG=338886 Change-Id: Id635e78b9afaad5ca311e3aeed888c9aedeb9637 Reviewed-on: https://boringssl-review.googlesource.com/4490 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:07:38 +01:00
#ifdef __APPLE__
ldr r0,[r0]
#endif
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
tst r0,#1 @ NEON available?
ldmia sp, {r0,r2}
beq .Lialu
add sp,sp,#8
b bn_mul8x_mont_neon
.align 4
.Lialu:
#endif
cmp ip,#2
mov $num,ip @ load num
movlt r0,#0
addlt sp,sp,#2*4
blt .Labrt
stmdb sp!,{r4-r12,lr} @ save 10 registers
mov $num,$num,lsl#2 @ rescale $num for byte count
sub sp,sp,$num @ alloca(4*num)
sub sp,sp,#4 @ +extra dword
sub $num,$num,#4 @ "num=num-1"
add $tp,$bp,$num @ &bp[num-1]
add $num,sp,$num @ $num to point at &tp[num-1]
ldr $n0,[$_n0] @ &n0
ldr $bi,[$bp] @ bp[0]
ldr $aj,[$ap],#4 @ ap[0],ap++
ldr $nj,[$np],#4 @ np[0],np++
ldr $n0,[$n0] @ *n0
str $tp,[$_bpend] @ save &bp[num]
umull $alo,$ahi,$aj,$bi @ ap[0]*bp[0]
str $n0,[$_n0] @ save n0 value
mul $n0,$alo,$n0 @ "tp[0]"*n0
mov $nlo,#0
umlal $alo,$nlo,$nj,$n0 @ np[0]*n0+"t[0]"
mov $tp,sp
.L1st:
ldr $aj,[$ap],#4 @ ap[j],ap++
mov $alo,$ahi
ldr $nj,[$np],#4 @ np[j],np++
mov $ahi,#0
umlal $alo,$ahi,$aj,$bi @ ap[j]*bp[0]
mov $nhi,#0
umlal $nlo,$nhi,$nj,$n0 @ np[j]*n0
adds $nlo,$nlo,$alo
str $nlo,[$tp],#4 @ tp[j-1]=,tp++
adc $nlo,$nhi,#0
cmp $tp,$num
bne .L1st
adds $nlo,$nlo,$ahi
ldr $tp,[$_bp] @ restore bp
mov $nhi,#0
ldr $n0,[$_n0] @ restore n0
adc $nhi,$nhi,#0
str $nlo,[$num] @ tp[num-1]=
str $nhi,[$num,#4] @ tp[num]=
.Louter:
sub $tj,$num,sp @ "original" $num-1 value
sub $ap,$ap,$tj @ "rewind" ap to &ap[1]
ldr $bi,[$tp,#4]! @ *(++bp)
sub $np,$np,$tj @ "rewind" np to &np[1]
ldr $aj,[$ap,#-4] @ ap[0]
ldr $alo,[sp] @ tp[0]
ldr $nj,[$np,#-4] @ np[0]
ldr $tj,[sp,#4] @ tp[1]
mov $ahi,#0
umlal $alo,$ahi,$aj,$bi @ ap[0]*bp[i]+tp[0]
str $tp,[$_bp] @ save bp
mul $n0,$alo,$n0
mov $nlo,#0
umlal $alo,$nlo,$nj,$n0 @ np[0]*n0+"tp[0]"
mov $tp,sp
.Linner:
ldr $aj,[$ap],#4 @ ap[j],ap++
adds $alo,$ahi,$tj @ +=tp[j]
ldr $nj,[$np],#4 @ np[j],np++
mov $ahi,#0
umlal $alo,$ahi,$aj,$bi @ ap[j]*bp[i]
mov $nhi,#0
umlal $nlo,$nhi,$nj,$n0 @ np[j]*n0
adc $ahi,$ahi,#0
ldr $tj,[$tp,#8] @ tp[j+1]
adds $nlo,$nlo,$alo
str $nlo,[$tp],#4 @ tp[j-1]=,tp++
adc $nlo,$nhi,#0
cmp $tp,$num
bne .Linner
adds $nlo,$nlo,$ahi
mov $nhi,#0
ldr $tp,[$_bp] @ restore bp
adc $nhi,$nhi,#0
ldr $n0,[$_n0] @ restore n0
adds $nlo,$nlo,$tj
ldr $tj,[$_bpend] @ restore &bp[num]
adc $nhi,$nhi,#0
str $nlo,[$num] @ tp[num-1]=
str $nhi,[$num,#4] @ tp[num]=
cmp $tp,$tj
bne .Louter
ldr $rp,[$_rp] @ pull rp
add $num,$num,#4 @ $num to point at &tp[num]
sub $aj,$num,sp @ "original" num value
mov $tp,sp @ "rewind" $tp
mov $ap,$tp @ "borrow" $ap
sub $np,$np,$aj @ "rewind" $np to &np[0]
subs $tj,$tj,$tj @ "clear" carry flag
.Lsub: ldr $tj,[$tp],#4
ldr $nj,[$np],#4
sbcs $tj,$tj,$nj @ tp[j]-np[j]
str $tj,[$rp],#4 @ rp[j]=
teq $tp,$num @ preserve carry
bne .Lsub
sbcs $nhi,$nhi,#0 @ upmost carry
mov $tp,sp @ "rewind" $tp
sub $rp,$rp,$aj @ "rewind" $rp
and $ap,$tp,$nhi
bic $np,$rp,$nhi
orr $ap,$ap,$np @ ap=borrow?tp:rp
.Lcopy: ldr $tj,[$ap],#4 @ copy or in-place refresh
str sp,[$tp],#4 @ zap tp
str $tj,[$rp],#4
cmp $tp,$num
bne .Lcopy
add sp,$num,#4 @ skip over tp[num+1]
ldmia sp!,{r4-r12,lr} @ restore registers
add sp,sp,#2*4 @ skip over {r0,r2}
mov r0,#1
ARM assembly pack: get ARMv7 instruction endianness right. Pointer out and suggested by: Ard Biesheuvel. (Imported from upstream's 5dcf70a1c57c2019bfad640fe14fd4a73212860a) This is from a while ago, but it's one source of divergence between our copy of these files and master's. Change-Id: I6525a27f25eb86a92420c32996af47ecc42ee020 Reviewed-on: https://boringssl-review.googlesource.com/4487 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:21:51 +01:00
.Labrt:
#if __ARM_ARCH__>=5
ret @ bx lr
#else
tst lr,#1
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
moveq pc,lr @ be binary compatible with V4, yet
bx lr @ interoperable with Thumb ISA:-)
ARM assembly pack: get ARMv7 instruction endianness right. Pointer out and suggested by: Ard Biesheuvel. (Imported from upstream's 5dcf70a1c57c2019bfad640fe14fd4a73212860a) This is from a while ago, but it's one source of divergence between our copy of these files and master's. Change-Id: I6525a27f25eb86a92420c32996af47ecc42ee020 Reviewed-on: https://boringssl-review.googlesource.com/4487 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:21:51 +01:00
#endif
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.size bn_mul_mont,.-bn_mul_mont
___
{
sub Dlo() { shift=~m|q([1]?[0-9])|?"d".($1*2):""; }
sub Dhi() { shift=~m|q([1]?[0-9])|?"d".($1*2+1):""; }
my ($A0,$A1,$A2,$A3)=map("d$_",(0..3));
my ($N0,$N1,$N2,$N3)=map("d$_",(4..7));
my ($Z,$Temp)=("q4","q5");
my ($A0xB,$A1xB,$A2xB,$A3xB,$A4xB,$A5xB,$A6xB,$A7xB)=map("q$_",(6..13));
my ($Bi,$Ni,$M0)=map("d$_",(28..31));
my $zero=&Dlo($Z);
my $temp=&Dlo($Temp);
my ($rptr,$aptr,$bptr,$nptr,$n0,$num)=map("r$_",(0..5));
my ($tinptr,$toutptr,$inner,$outer)=map("r$_",(6..9));
$code.=<<___;
Remove inconsistency in ARM support. This facilitates "universal" builds, ones that target multiple architectures, e.g. ARMv5 through ARMv7. (Imported from upstream's c1669e1c205dc8e695fb0c10a655f434e758b9f7) This is a change from a while ago which was a source of divergence between our perlasm and upstream's. This change in upstream came with the following comment in Configure: Note that -march is not among compiler options in below linux-armv4 target line. Not specifying one is intentional to give you choice to: a) rely on your compiler default by not specifying one; b) specify your target platform explicitly for optimal performance, e.g. -march=armv6 or -march=armv7-a; c) build "universal" binary that targets *range* of platforms by specifying minimum and maximum supported architecture; As for c) option. It actually makes no sense to specify maximum to be less than ARMv7, because it's the least requirement for run-time switch between platform-specific code paths. And without run-time switch performance would be equivalent to one for minimum. Secondly, there are some natural limitations that you'd have to accept and respect. Most notably you can *not* build "universal" binary for big-endian platform. This is because ARMv7 processor always picks instructions in little-endian order. Another similar limitation is that -mthumb can't "cross" -march=armv6t2 boundary, because that's where it became Thumb-2. Well, this limitation is a bit artificial, because it's not really impossible, but it's deemed too tricky to support. And of course you have to be sure that your binutils are actually up to the task of handling maximum target platform. Change-Id: Ie5f674d603393f0a1354a0d0973987484a4a650c Reviewed-on: https://boringssl-review.googlesource.com/4488 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:27:38 +01:00
#if __ARM_MAX_ARCH__>=7
.arch armv7-a
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.fpu neon
.type bn_mul8x_mont_neon,%function
.align 5
bn_mul8x_mont_neon:
mov ip,sp
stmdb sp!,{r4-r11}
vstmdb sp!,{d8-d15} @ ABI specification says so
ldmia ip,{r4-r5} @ load rest of parameter block
sub $toutptr,sp,#16
vld1.32 {${Bi}[0]}, [$bptr,:32]!
sub $toutptr,$toutptr,$num,lsl#4
vld1.32 {$A0-$A3}, [$aptr]! @ can't specify :32 :-(
and $toutptr,$toutptr,#-64
vld1.32 {${M0}[0]}, [$n0,:32]
mov sp,$toutptr @ alloca
veor $zero,$zero,$zero
subs $inner,$num,#8
vzip.16 $Bi,$zero
vmull.u32 $A0xB,$Bi,${A0}[0]
vmull.u32 $A1xB,$Bi,${A0}[1]
vmull.u32 $A2xB,$Bi,${A1}[0]
vshl.i64 $temp,`&Dhi("$A0xB")`,#16
vmull.u32 $A3xB,$Bi,${A1}[1]
vadd.u64 $temp,$temp,`&Dlo("$A0xB")`
veor $zero,$zero,$zero
vmul.u32 $Ni,$temp,$M0
vmull.u32 $A4xB,$Bi,${A2}[0]
vld1.32 {$N0-$N3}, [$nptr]!
vmull.u32 $A5xB,$Bi,${A2}[1]
vmull.u32 $A6xB,$Bi,${A3}[0]
vzip.16 $Ni,$zero
vmull.u32 $A7xB,$Bi,${A3}[1]
bne .LNEON_1st
@ special case for num=8, everything is in register bank...
vmlal.u32 $A0xB,$Ni,${N0}[0]
sub $outer,$num,#1
vmlal.u32 $A1xB,$Ni,${N0}[1]
vmlal.u32 $A2xB,$Ni,${N1}[0]
vmlal.u32 $A3xB,$Ni,${N1}[1]
vmlal.u32 $A4xB,$Ni,${N2}[0]
vmov $Temp,$A0xB
vmlal.u32 $A5xB,$Ni,${N2}[1]
vmov $A0xB,$A1xB
vmlal.u32 $A6xB,$Ni,${N3}[0]
vmov $A1xB,$A2xB
vmlal.u32 $A7xB,$Ni,${N3}[1]
vmov $A2xB,$A3xB
vmov $A3xB,$A4xB
vshr.u64 $temp,$temp,#16
vmov $A4xB,$A5xB
vmov $A5xB,$A6xB
vadd.u64 $temp,$temp,`&Dhi("$Temp")`
vmov $A6xB,$A7xB
veor $A7xB,$A7xB
vshr.u64 $temp,$temp,#16
b .LNEON_outer8
.align 4
.LNEON_outer8:
vld1.32 {${Bi}[0]}, [$bptr,:32]!
veor $zero,$zero,$zero
vzip.16 $Bi,$zero
vadd.u64 `&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
vmlal.u32 $A0xB,$Bi,${A0}[0]
vmlal.u32 $A1xB,$Bi,${A0}[1]
vmlal.u32 $A2xB,$Bi,${A1}[0]
vshl.i64 $temp,`&Dhi("$A0xB")`,#16
vmlal.u32 $A3xB,$Bi,${A1}[1]
vadd.u64 $temp,$temp,`&Dlo("$A0xB")`
veor $zero,$zero,$zero
subs $outer,$outer,#1
vmul.u32 $Ni,$temp,$M0
vmlal.u32 $A4xB,$Bi,${A2}[0]
vmlal.u32 $A5xB,$Bi,${A2}[1]
vmlal.u32 $A6xB,$Bi,${A3}[0]
vzip.16 $Ni,$zero
vmlal.u32 $A7xB,$Bi,${A3}[1]
vmlal.u32 $A0xB,$Ni,${N0}[0]
vmlal.u32 $A1xB,$Ni,${N0}[1]
vmlal.u32 $A2xB,$Ni,${N1}[0]
vmlal.u32 $A3xB,$Ni,${N1}[1]
vmlal.u32 $A4xB,$Ni,${N2}[0]
vmov $Temp,$A0xB
vmlal.u32 $A5xB,$Ni,${N2}[1]
vmov $A0xB,$A1xB
vmlal.u32 $A6xB,$Ni,${N3}[0]
vmov $A1xB,$A2xB
vmlal.u32 $A7xB,$Ni,${N3}[1]
vmov $A2xB,$A3xB
vmov $A3xB,$A4xB
vshr.u64 $temp,$temp,#16
vmov $A4xB,$A5xB
vmov $A5xB,$A6xB
vadd.u64 $temp,$temp,`&Dhi("$Temp")`
vmov $A6xB,$A7xB
veor $A7xB,$A7xB
vshr.u64 $temp,$temp,#16
bne .LNEON_outer8
vadd.u64 `&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
mov $toutptr,sp
vshr.u64 $temp,`&Dlo("$A0xB")`,#16
mov $inner,$num
vadd.u64 `&Dhi("$A0xB")`,`&Dhi("$A0xB")`,$temp
add $tinptr,sp,#16
vshr.u64 $temp,`&Dhi("$A0xB")`,#16
vzip.16 `&Dlo("$A0xB")`,`&Dhi("$A0xB")`
b .LNEON_tail2
.align 4
.LNEON_1st:
vmlal.u32 $A0xB,$Ni,${N0}[0]
vld1.32 {$A0-$A3}, [$aptr]!
vmlal.u32 $A1xB,$Ni,${N0}[1]
subs $inner,$inner,#8
vmlal.u32 $A2xB,$Ni,${N1}[0]
vmlal.u32 $A3xB,$Ni,${N1}[1]
vmlal.u32 $A4xB,$Ni,${N2}[0]
vld1.32 {$N0-$N1}, [$nptr]!
vmlal.u32 $A5xB,$Ni,${N2}[1]
vst1.64 {$A0xB-$A1xB}, [$toutptr,:256]!
vmlal.u32 $A6xB,$Ni,${N3}[0]
vmlal.u32 $A7xB,$Ni,${N3}[1]
vst1.64 {$A2xB-$A3xB}, [$toutptr,:256]!
vmull.u32 $A0xB,$Bi,${A0}[0]
vld1.32 {$N2-$N3}, [$nptr]!
vmull.u32 $A1xB,$Bi,${A0}[1]
vst1.64 {$A4xB-$A5xB}, [$toutptr,:256]!
vmull.u32 $A2xB,$Bi,${A1}[0]
vmull.u32 $A3xB,$Bi,${A1}[1]
vst1.64 {$A6xB-$A7xB}, [$toutptr,:256]!
vmull.u32 $A4xB,$Bi,${A2}[0]
vmull.u32 $A5xB,$Bi,${A2}[1]
vmull.u32 $A6xB,$Bi,${A3}[0]
vmull.u32 $A7xB,$Bi,${A3}[1]
bne .LNEON_1st
vmlal.u32 $A0xB,$Ni,${N0}[0]
add $tinptr,sp,#16
vmlal.u32 $A1xB,$Ni,${N0}[1]
sub $aptr,$aptr,$num,lsl#2 @ rewind $aptr
vmlal.u32 $A2xB,$Ni,${N1}[0]
vld1.64 {$Temp}, [sp,:128]
vmlal.u32 $A3xB,$Ni,${N1}[1]
sub $outer,$num,#1
vmlal.u32 $A4xB,$Ni,${N2}[0]
vst1.64 {$A0xB-$A1xB}, [$toutptr,:256]!
vmlal.u32 $A5xB,$Ni,${N2}[1]
vshr.u64 $temp,$temp,#16
vld1.64 {$A0xB}, [$tinptr, :128]!
vmlal.u32 $A6xB,$Ni,${N3}[0]
vst1.64 {$A2xB-$A3xB}, [$toutptr,:256]!
vmlal.u32 $A7xB,$Ni,${N3}[1]
vst1.64 {$A4xB-$A5xB}, [$toutptr,:256]!
vadd.u64 $temp,$temp,`&Dhi("$Temp")`
veor $Z,$Z,$Z
vst1.64 {$A6xB-$A7xB}, [$toutptr,:256]!
vld1.64 {$A1xB-$A2xB}, [$tinptr, :256]!
vst1.64 {$Z}, [$toutptr,:128]
vshr.u64 $temp,$temp,#16
b .LNEON_outer
.align 4
.LNEON_outer:
vld1.32 {${Bi}[0]}, [$bptr,:32]!
sub $nptr,$nptr,$num,lsl#2 @ rewind $nptr
vld1.32 {$A0-$A3}, [$aptr]!
veor $zero,$zero,$zero
mov $toutptr,sp
vzip.16 $Bi,$zero
sub $inner,$num,#8
vadd.u64 `&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
vmlal.u32 $A0xB,$Bi,${A0}[0]
vld1.64 {$A3xB-$A4xB},[$tinptr,:256]!
vmlal.u32 $A1xB,$Bi,${A0}[1]
vmlal.u32 $A2xB,$Bi,${A1}[0]
vld1.64 {$A5xB-$A6xB},[$tinptr,:256]!
vmlal.u32 $A3xB,$Bi,${A1}[1]
vshl.i64 $temp,`&Dhi("$A0xB")`,#16
veor $zero,$zero,$zero
vadd.u64 $temp,$temp,`&Dlo("$A0xB")`
vld1.64 {$A7xB},[$tinptr,:128]!
vmul.u32 $Ni,$temp,$M0
vmlal.u32 $A4xB,$Bi,${A2}[0]
vld1.32 {$N0-$N3}, [$nptr]!
vmlal.u32 $A5xB,$Bi,${A2}[1]
vmlal.u32 $A6xB,$Bi,${A3}[0]
vzip.16 $Ni,$zero
vmlal.u32 $A7xB,$Bi,${A3}[1]
.LNEON_inner:
vmlal.u32 $A0xB,$Ni,${N0}[0]
vld1.32 {$A0-$A3}, [$aptr]!
vmlal.u32 $A1xB,$Ni,${N0}[1]
subs $inner,$inner,#8
vmlal.u32 $A2xB,$Ni,${N1}[0]
vmlal.u32 $A3xB,$Ni,${N1}[1]
vst1.64 {$A0xB-$A1xB}, [$toutptr,:256]!
vmlal.u32 $A4xB,$Ni,${N2}[0]
vld1.64 {$A0xB}, [$tinptr, :128]!
vmlal.u32 $A5xB,$Ni,${N2}[1]
vst1.64 {$A2xB-$A3xB}, [$toutptr,:256]!
vmlal.u32 $A6xB,$Ni,${N3}[0]
vld1.64 {$A1xB-$A2xB}, [$tinptr, :256]!
vmlal.u32 $A7xB,$Ni,${N3}[1]
vst1.64 {$A4xB-$A5xB}, [$toutptr,:256]!
vmlal.u32 $A0xB,$Bi,${A0}[0]
vld1.64 {$A3xB-$A4xB}, [$tinptr, :256]!
vmlal.u32 $A1xB,$Bi,${A0}[1]
vst1.64 {$A6xB-$A7xB}, [$toutptr,:256]!
vmlal.u32 $A2xB,$Bi,${A1}[0]
vld1.64 {$A5xB-$A6xB}, [$tinptr, :256]!
vmlal.u32 $A3xB,$Bi,${A1}[1]
vld1.32 {$N0-$N3}, [$nptr]!
vmlal.u32 $A4xB,$Bi,${A2}[0]
vld1.64 {$A7xB}, [$tinptr, :128]!
vmlal.u32 $A5xB,$Bi,${A2}[1]
vmlal.u32 $A6xB,$Bi,${A3}[0]
vmlal.u32 $A7xB,$Bi,${A3}[1]
bne .LNEON_inner
vmlal.u32 $A0xB,$Ni,${N0}[0]
add $tinptr,sp,#16
vmlal.u32 $A1xB,$Ni,${N0}[1]
sub $aptr,$aptr,$num,lsl#2 @ rewind $aptr
vmlal.u32 $A2xB,$Ni,${N1}[0]
vld1.64 {$Temp}, [sp,:128]
vmlal.u32 $A3xB,$Ni,${N1}[1]
subs $outer,$outer,#1
vmlal.u32 $A4xB,$Ni,${N2}[0]
vst1.64 {$A0xB-$A1xB}, [$toutptr,:256]!
vmlal.u32 $A5xB,$Ni,${N2}[1]
vld1.64 {$A0xB}, [$tinptr, :128]!
vshr.u64 $temp,$temp,#16
vst1.64 {$A2xB-$A3xB}, [$toutptr,:256]!
vmlal.u32 $A6xB,$Ni,${N3}[0]
vld1.64 {$A1xB-$A2xB}, [$tinptr, :256]!
vmlal.u32 $A7xB,$Ni,${N3}[1]
vst1.64 {$A4xB-$A5xB}, [$toutptr,:256]!
vadd.u64 $temp,$temp,`&Dhi("$Temp")`
vst1.64 {$A6xB-$A7xB}, [$toutptr,:256]!
vshr.u64 $temp,$temp,#16
bne .LNEON_outer
mov $toutptr,sp
mov $inner,$num
.LNEON_tail:
vadd.u64 `&Dlo("$A0xB")`,`&Dlo("$A0xB")`,$temp
vld1.64 {$A3xB-$A4xB}, [$tinptr, :256]!
vshr.u64 $temp,`&Dlo("$A0xB")`,#16
vadd.u64 `&Dhi("$A0xB")`,`&Dhi("$A0xB")`,$temp
vld1.64 {$A5xB-$A6xB}, [$tinptr, :256]!
vshr.u64 $temp,`&Dhi("$A0xB")`,#16
vld1.64 {$A7xB}, [$tinptr, :128]!
vzip.16 `&Dlo("$A0xB")`,`&Dhi("$A0xB")`
.LNEON_tail2:
vadd.u64 `&Dlo("$A1xB")`,`&Dlo("$A1xB")`,$temp
vst1.32 {`&Dlo("$A0xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A1xB")`,#16
vadd.u64 `&Dhi("$A1xB")`,`&Dhi("$A1xB")`,$temp
vshr.u64 $temp,`&Dhi("$A1xB")`,#16
vzip.16 `&Dlo("$A1xB")`,`&Dhi("$A1xB")`
vadd.u64 `&Dlo("$A2xB")`,`&Dlo("$A2xB")`,$temp
vst1.32 {`&Dlo("$A1xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A2xB")`,#16
vadd.u64 `&Dhi("$A2xB")`,`&Dhi("$A2xB")`,$temp
vshr.u64 $temp,`&Dhi("$A2xB")`,#16
vzip.16 `&Dlo("$A2xB")`,`&Dhi("$A2xB")`
vadd.u64 `&Dlo("$A3xB")`,`&Dlo("$A3xB")`,$temp
vst1.32 {`&Dlo("$A2xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A3xB")`,#16
vadd.u64 `&Dhi("$A3xB")`,`&Dhi("$A3xB")`,$temp
vshr.u64 $temp,`&Dhi("$A3xB")`,#16
vzip.16 `&Dlo("$A3xB")`,`&Dhi("$A3xB")`
vadd.u64 `&Dlo("$A4xB")`,`&Dlo("$A4xB")`,$temp
vst1.32 {`&Dlo("$A3xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A4xB")`,#16
vadd.u64 `&Dhi("$A4xB")`,`&Dhi("$A4xB")`,$temp
vshr.u64 $temp,`&Dhi("$A4xB")`,#16
vzip.16 `&Dlo("$A4xB")`,`&Dhi("$A4xB")`
vadd.u64 `&Dlo("$A5xB")`,`&Dlo("$A5xB")`,$temp
vst1.32 {`&Dlo("$A4xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A5xB")`,#16
vadd.u64 `&Dhi("$A5xB")`,`&Dhi("$A5xB")`,$temp
vshr.u64 $temp,`&Dhi("$A5xB")`,#16
vzip.16 `&Dlo("$A5xB")`,`&Dhi("$A5xB")`
vadd.u64 `&Dlo("$A6xB")`,`&Dlo("$A6xB")`,$temp
vst1.32 {`&Dlo("$A5xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A6xB")`,#16
vadd.u64 `&Dhi("$A6xB")`,`&Dhi("$A6xB")`,$temp
vld1.64 {$A0xB}, [$tinptr, :128]!
vshr.u64 $temp,`&Dhi("$A6xB")`,#16
vzip.16 `&Dlo("$A6xB")`,`&Dhi("$A6xB")`
vadd.u64 `&Dlo("$A7xB")`,`&Dlo("$A7xB")`,$temp
vst1.32 {`&Dlo("$A6xB")`[0]}, [$toutptr, :32]!
vshr.u64 $temp,`&Dlo("$A7xB")`,#16
vadd.u64 `&Dhi("$A7xB")`,`&Dhi("$A7xB")`,$temp
vld1.64 {$A1xB-$A2xB}, [$tinptr, :256]!
vshr.u64 $temp,`&Dhi("$A7xB")`,#16
vzip.16 `&Dlo("$A7xB")`,`&Dhi("$A7xB")`
subs $inner,$inner,#8
vst1.32 {`&Dlo("$A7xB")`[0]}, [$toutptr, :32]!
bne .LNEON_tail
vst1.32 {${temp}[0]}, [$toutptr, :32] @ top-most bit
sub $nptr,$nptr,$num,lsl#2 @ rewind $nptr
subs $aptr,sp,#0 @ clear carry flag
add $bptr,sp,$num,lsl#2
.LNEON_sub:
ldmia $aptr!, {r4-r7}
ldmia $nptr!, {r8-r11}
sbcs r8, r4,r8
sbcs r9, r5,r9
sbcs r10,r6,r10
sbcs r11,r7,r11
teq $aptr,$bptr @ preserves carry
stmia $rptr!, {r8-r11}
bne .LNEON_sub
ldr r10, [$aptr] @ load top-most bit
veor q0,q0,q0
sub r11,$bptr,sp @ this is num*4
veor q1,q1,q1
mov $aptr,sp
sub $rptr,$rptr,r11 @ rewind $rptr
mov $nptr,$bptr @ second 3/4th of frame
sbcs r10,r10,#0 @ result is carry flag
.LNEON_copy_n_zap:
ldmia $aptr!, {r4-r7}
ldmia $rptr, {r8-r11}
movcc r8, r4
vst1.64 {q0-q1}, [$nptr,:256]! @ wipe
movcc r9, r5
movcc r10,r6
vst1.64 {q0-q1}, [$nptr,:256]! @ wipe
movcc r11,r7
ldmia $aptr, {r4-r7}
stmia $rptr!, {r8-r11}
sub $aptr,$aptr,#16
ldmia $rptr, {r8-r11}
movcc r8, r4
vst1.64 {q0-q1}, [$aptr,:256]! @ wipe
movcc r9, r5
movcc r10,r6
vst1.64 {q0-q1}, [$nptr,:256]! @ wipe
movcc r11,r7
teq $aptr,$bptr @ preserves carry
stmia $rptr!, {r8-r11}
bne .LNEON_copy_n_zap
sub sp,ip,#96
vldmia sp!,{d8-d15}
ldmia sp!,{r4-r11}
ARM assembly pack: get ARMv7 instruction endianness right. Pointer out and suggested by: Ard Biesheuvel. (Imported from upstream's 5dcf70a1c57c2019bfad640fe14fd4a73212860a) This is from a while ago, but it's one source of divergence between our copy of these files and master's. Change-Id: I6525a27f25eb86a92420c32996af47ecc42ee020 Reviewed-on: https://boringssl-review.googlesource.com/4487 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:21:51 +01:00
ret @ bx lr
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.size bn_mul8x_mont_neon,.-bn_mul8x_mont_neon
#endif
___
}
$code.=<<___;
.asciz "Montgomery multiplication for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
.align 2
Remove inconsistency in ARM support. This facilitates "universal" builds, ones that target multiple architectures, e.g. ARMv5 through ARMv7. (Imported from upstream's c1669e1c205dc8e695fb0c10a655f434e758b9f7) This is a change from a while ago which was a source of divergence between our perlasm and upstream's. This change in upstream came with the following comment in Configure: Note that -march is not among compiler options in below linux-armv4 target line. Not specifying one is intentional to give you choice to: a) rely on your compiler default by not specifying one; b) specify your target platform explicitly for optimal performance, e.g. -march=armv6 or -march=armv7-a; c) build "universal" binary that targets *range* of platforms by specifying minimum and maximum supported architecture; As for c) option. It actually makes no sense to specify maximum to be less than ARMv7, because it's the least requirement for run-time switch between platform-specific code paths. And without run-time switch performance would be equivalent to one for minimum. Secondly, there are some natural limitations that you'd have to accept and respect. Most notably you can *not* build "universal" binary for big-endian platform. This is because ARMv7 processor always picks instructions in little-endian order. Another similar limitation is that -mthumb can't "cross" -march=armv6t2 boundary, because that's where it became Thumb-2. Well, this limitation is a bit artificial, because it's not really impossible, but it's deemed too tricky to support. And of course you have to be sure that your binutils are actually up to the task of handling maximum target platform. Change-Id: Ie5f674d603393f0a1354a0d0973987484a4a650c Reviewed-on: https://boringssl-review.googlesource.com/4488 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:27:38 +01:00
#if __ARM_MAX_ARCH__>=7
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
.comm OPENSSL_armcap_P,4,4
Mark OPENSSL_armcap_P as hidden in ARM asm. This is an import from ARM. Without this, one of the Android builds of BoringSSL was failing with: (sha512-armv4.o): requires unsupported dynamic reloc R_ARM_REL32; recompile with -fPIC This is (I believe) a very misleading error message. The R_ARM_REL32 relocation type is the correct type for position independent code. But unless the target symbol is hidden then the linker doesn't know that it's not going to be overridden by a different ELF module. Chromium probably gets away with this because of different default compiler flags than Android. Change-Id: I967eabc4d6b33d1e6635caaf6e7a306e4e77c101 Reviewed-on: https://boringssl-review.googlesource.com/3471 Reviewed-by: Adam Langley <agl@google.com>
2015-02-13 22:57:49 +00:00
.hidden OPENSSL_armcap_P
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
#endif
___
$code =~ s/\`([^\`]*)\`/eval $1/gem;
$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm; # make it possible to compile with -march=armv4
ARM assembly pack: get ARMv7 instruction endianness right. Pointer out and suggested by: Ard Biesheuvel. (Imported from upstream's 5dcf70a1c57c2019bfad640fe14fd4a73212860a) This is from a while ago, but it's one source of divergence between our copy of these files and master's. Change-Id: I6525a27f25eb86a92420c32996af47ecc42ee020 Reviewed-on: https://boringssl-review.googlesource.com/4487 Reviewed-by: Adam Langley <agl@google.com>
2015-04-21 02:21:51 +01:00
$code =~ s/\bret\b/bx lr/gm;
Inital import. Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta). (This change contains substantial changes from the original and effectively starts a new history.)
2014-06-20 20:00:00 +01:00
print $code;
close STDOUT;
Reference in New Issue Copy Permalink