summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--lib/libcrypto/bn/asm/modexp512-x86_64.pl112
-rwxr-xr-xlib/libcrypto/bn/asm/x86_64-mont.pl1428
-rw-r--r--lib/libcrypto/camellia/asm/cmll-x86_64.pl223
-rwxr-xr-xlib/libcrypto/md5/asm/md5-x86_64.pl53
-rw-r--r--lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl110
-rwxr-xr-xlib/libcrypto/rc4/asm/rc4-x86_64.pl134
-rwxr-xr-xlib/libcrypto/sha/asm/sha512-x86_64.pl118
-rw-r--r--lib/libcrypto/whrlpool/asm/wp-x86_64.pl112
8 files changed, 1464 insertions, 826 deletions
diff --git a/lib/libcrypto/bn/asm/modexp512-x86_64.pl b/lib/libcrypto/bn/asm/modexp512-x86_64.pl
index 54aeb01921e..43172828353 100644
--- a/lib/libcrypto/bn/asm/modexp512-x86_64.pl
+++ b/lib/libcrypto/bn/asm/modexp512-x86_64.pl
@@ -61,14 +61,13 @@ $flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $flavour $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
use strict;
my $code=".text\n\n";
@@ -1373,113 +1372,6 @@ end_main_loop_a3b:
.size mod_exp_512, . - mod_exp_512
___
-if ($win64) {
-# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
-# CONTEXT *context,DISPATCHER_CONTEXT *disp)
-my $rec="%rcx";
-my $frame="%rdx";
-my $context="%r8";
-my $disp="%r9";
-
-$code.=<<___;
-.extern __imp_RtlVirtualUnwind
-.type mod_exp_512_se_handler,\@abi-omnipotent
-.align 16
-mod_exp_512_se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- sub \$64,%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- lea .Lbody(%rip),%r10
- cmp %r10,%rbx # context->Rip<prologue label
- jb .Lin_prologue
-
- mov 152($context),%rax # pull context->Rsp
-
- lea .Lepilogue(%rip),%r10
- cmp %r10,%rbx # context->Rip>=epilogue label
- jae .Lin_prologue
-
- mov $rsp_offset(%rax),%rax # pull saved Rsp
-
- mov 32(%rax),%rbx
- mov 40(%rax),%rbp
- mov 24(%rax),%r12
- mov 16(%rax),%r13
- mov 8(%rax),%r14
- mov 0(%rax),%r15
- lea 48(%rax),%rax
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R13
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
-
-.Lin_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
- mov 40($disp),%rdi # disp->ContextRecord
- mov $context,%rsi # context
- mov \$154,%ecx # sizeof(CONTEXT)
- .long 0xa548f3fc # cld; rep movsq
-
- mov $disp,%rsi
- xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
- mov 8(%rsi),%rdx # arg2, disp->ImageBase
- mov 0(%rsi),%r8 # arg3, disp->ControlPc
- mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
- mov 40(%rsi),%r10 # disp->ContextRecord
- lea 56(%rsi),%r11 # &disp->HandlerData
- lea 24(%rsi),%r12 # &disp->EstablisherFrame
- mov %r10,32(%rsp) # arg5
- mov %r11,40(%rsp) # arg6
- mov %r12,48(%rsp) # arg7
- mov %rcx,56(%rsp) # arg8, (NULL)
- call *__imp_RtlVirtualUnwind(%rip)
-
- mov \$1,%eax # ExceptionContinueSearch
- add \$64,%rsp
- popfq
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
- pop %rdi
- pop %rsi
- ret
-.size mod_exp_512_se_handler,.-mod_exp_512_se_handler
-
-.section .pdata
-.align 4
- .rva .LSEH_begin_mod_exp_512
- .rva .LSEH_end_mod_exp_512
- .rva .LSEH_info_mod_exp_512
-
-.section .xdata
-.align 8
-.LSEH_info_mod_exp_512:
- .byte 9,0,0,0
- .rva mod_exp_512_se_handler
-___
-}
-
sub reg_part {
my ($reg,$conv)=@_;
if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
diff --git a/lib/libcrypto/bn/asm/x86_64-mont.pl b/lib/libcrypto/bn/asm/x86_64-mont.pl
index c43b69592a5..90c717f1152 100755
--- a/lib/libcrypto/bn/asm/x86_64-mont.pl
+++ b/lib/libcrypto/bn/asm/x86_64-mont.pl
@@ -1,7 +1,7 @@
#!/usr/bin/env perl
# ====================================================================
-# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
@@ -15,14 +15,31 @@
# respectful 50%. It remains to be seen if loop unrolling and
# dedicated squaring routine can provide further improvement...
-$output=shift;
+# July 2011.
+#
+# Add dedicated squaring procedure. Performance improvement varies
+# from platform to platform, but in average it's ~5%/15%/25%/33%
+# for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
+
+# August 2011.
+#
+# Unroll and modulo-schedule inner loops in such manner that they
+# are "fallen through" for input lengths of 8, which is critical for
+# 1024-bit RSA *sign*. Average performance improvement in comparison
+# to *initial* version of this module from 2005 is ~0%/30%/40%/45%
+# for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
+
+$flavour = shift;
+$output = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
# int bn_mul_mont(
$rp="%rdi"; # BN_ULONG *rp,
@@ -33,7 +50,6 @@ $n0="%r8"; # const BN_ULONG *n0,
$num="%r9"; # int num);
$lo0="%r10";
$hi0="%r11";
-$bp="%r12"; # reassign $bp
$hi1="%r13";
$i="%r14";
$j="%r15";
@@ -47,6 +63,16 @@ $code=<<___;
.type bn_mul_mont,\@function,6
.align 16
bn_mul_mont:
+ test \$3,${num}d
+ jnz .Lmul_enter
+ cmp \$8,${num}d
+ jb .Lmul_enter
+ cmp $ap,$bp
+ jne .Lmul4x_enter
+ jmp .Lsqr4x_enter
+
+.align 16
+.Lmul_enter:
push %rbx
push %rbp
push %r12
@@ -55,54 +81,73 @@ bn_mul_mont:
push %r15
mov ${num}d,${num}d
- lea 2($num),%rax
- mov %rsp,%rbp
- neg %rax
- lea (%rsp,%rax,8),%rsp # tp=alloca(8*(num+2))
+ lea 2($num),%r10
+ mov %rsp,%r11
+ neg %r10
+ lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
and \$-1024,%rsp # minimize TLB usage
- mov %rbp,8(%rsp,$num,8) # tp[num+1]=%rsp
- mov %rdx,$bp # $bp reassigned, remember?
-
+ mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
+.Lmul_body:
+ mov $bp,%r12 # reassign $bp
+___
+ $bp="%r12";
+$code.=<<___;
mov ($n0),$n0 # pull n0[0] value
+ mov ($bp),$m0 # m0=bp[0]
+ mov ($ap),%rax
xor $i,$i # i=0
xor $j,$j # j=0
- mov ($bp),$m0 # m0=bp[0]
- mov ($ap),%rax
+ mov $n0,$m1
mulq $m0 # ap[0]*bp[0]
mov %rax,$lo0
- mov %rdx,$hi0
+ mov ($np),%rax
- imulq $n0,%rax # "tp[0]"*n0
- mov %rax,$m1
+ imulq $lo0,$m1 # "tp[0]"*n0
+ mov %rdx,$hi0
- mulq ($np) # np[0]*m1
- add $lo0,%rax # discarded
+ mulq $m1 # np[0]*m1
+ add %rax,$lo0 # discarded
+ mov 8($ap),%rax
adc \$0,%rdx
mov %rdx,$hi1
lea 1($j),$j # j++
+ jmp .L1st_enter
+
+.align 16
.L1st:
+ add %rax,$hi1
mov ($ap,$j,8),%rax
- mulq $m0 # ap[j]*bp[0]
- add $hi0,%rax
adc \$0,%rdx
- mov %rax,$lo0
+ add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
+ mov $lo0,$hi0
+ adc \$0,%rdx
+ mov $hi1,-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$hi1
+
+.L1st_enter:
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$hi0
mov ($np,$j,8),%rax
- mov %rdx,$hi0
+ adc \$0,%rdx
+ lea 1($j),$j # j++
+ mov %rdx,$lo0
mulq $m1 # np[j]*m1
- add $hi1,%rax
- lea 1($j),$j # j++
+ cmp $num,$j
+ jne .L1st
+
+ add %rax,$hi1
+ mov ($ap),%rax # ap[0]
adc \$0,%rdx
- add $lo0,%rax # np[j]*m1+ap[j]*bp[0]
+ add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
adc \$0,%rdx
- mov %rax,-16(%rsp,$j,8) # tp[j-1]
- cmp $num,$j
+ mov $hi1,-16(%rsp,$j,8) # tp[j-1]
mov %rdx,$hi1
- jl .L1st
+ mov $lo0,$hi0
xor %rdx,%rdx
add $hi0,$hi1
@@ -111,50 +156,64 @@ bn_mul_mont:
mov %rdx,(%rsp,$num,8) # store upmost overflow bit
lea 1($i),$i # i++
-.align 4
+ jmp .Louter
+.align 16
.Louter:
- xor $j,$j # j=0
-
mov ($bp,$i,8),$m0 # m0=bp[i]
- mov ($ap),%rax # ap[0]
+ xor $j,$j # j=0
+ mov $n0,$m1
+ mov (%rsp),$lo0
mulq $m0 # ap[0]*bp[i]
- add (%rsp),%rax # ap[0]*bp[i]+tp[0]
+ add %rax,$lo0 # ap[0]*bp[i]+tp[0]
+ mov ($np),%rax
adc \$0,%rdx
- mov %rax,$lo0
- mov %rdx,$hi0
- imulq $n0,%rax # tp[0]*n0
- mov %rax,$m1
+ imulq $lo0,$m1 # tp[0]*n0
+ mov %rdx,$hi0
- mulq ($np,$j,8) # np[0]*m1
- add $lo0,%rax # discarded
- mov 8(%rsp),$lo0 # tp[1]
+ mulq $m1 # np[0]*m1
+ add %rax,$lo0 # discarded
+ mov 8($ap),%rax
adc \$0,%rdx
+ mov 8(%rsp),$lo0 # tp[1]
mov %rdx,$hi1
lea 1($j),$j # j++
-.align 4
+ jmp .Linner_enter
+
+.align 16
.Linner:
+ add %rax,$hi1
mov ($ap,$j,8),%rax
- mulq $m0 # ap[j]*bp[i]
- add $hi0,%rax
adc \$0,%rdx
- add %rax,$lo0 # ap[j]*bp[i]+tp[j]
+ add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
+ mov (%rsp,$j,8),$lo0
+ adc \$0,%rdx
+ mov $hi1,-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$hi1
+
+.Linner_enter:
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$hi0
mov ($np,$j,8),%rax
adc \$0,%rdx
+ add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
mov %rdx,$hi0
+ adc \$0,$hi0
+ lea 1($j),$j # j++
mulq $m1 # np[j]*m1
- add $hi1,%rax
- lea 1($j),$j # j++
- adc \$0,%rdx
- add $lo0,%rax # np[j]*m1+ap[j]*bp[i]+tp[j]
+ cmp $num,$j
+ jne .Linner
+
+ add %rax,$hi1
+ mov ($ap),%rax # ap[0]
adc \$0,%rdx
+ add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
mov (%rsp,$j,8),$lo0
- cmp $num,$j
- mov %rax,-16(%rsp,$j,8) # tp[j-1]
+ adc \$0,%rdx
+ mov $hi1,-16(%rsp,$j,8) # tp[j-1]
mov %rdx,$hi1
- jl .Linner
xor %rdx,%rdx
add $hi0,$hi1
@@ -168,46 +227,1277 @@ bn_mul_mont:
cmp $num,$i
jl .Louter
- lea (%rsp),$ap # borrow ap for tp
- lea -1($num),$j # j=num-1
-
- mov ($ap),%rax # tp[0]
xor $i,$i # i=0 and clear CF!
+ mov (%rsp),%rax # tp[0]
+ lea (%rsp),$ap # borrow ap for tp
+ mov $num,$j # j=num
jmp .Lsub
.align 16
.Lsub: sbb ($np,$i,8),%rax
mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
- dec $j # doesn't affect CF!
mov 8($ap,$i,8),%rax # tp[i+1]
lea 1($i),$i # i++
- jge .Lsub
+ dec $j # doesnn't affect CF!
+ jnz .Lsub
sbb \$0,%rax # handle upmost overflow bit
+ xor $i,$i
and %rax,$ap
not %rax
mov $rp,$np
and %rax,$np
- lea -1($num),$j
+ mov $num,$j # j=num
or $np,$ap # ap=borrow?tp:rp
.align 16
.Lcopy: # copy or in-place refresh
+ mov ($ap,$i,8),%rax
+ mov $i,(%rsp,$i,8) # zap temporary vector
+ mov %rax,($rp,$i,8) # rp[i]=tp[i]
+ lea 1($i),$i
+ sub \$1,$j
+ jnz .Lcopy
+
+ mov 8(%rsp,$num,8),%rsi # restore %rsp
+ mov \$1,%rax
+ mov (%rsi),%r15
+ mov 8(%rsi),%r14
+ mov 16(%rsi),%r13
+ mov 24(%rsi),%r12
+ mov 32(%rsi),%rbp
+ mov 40(%rsi),%rbx
+ lea 48(%rsi),%rsp
+.Lmul_epilogue:
+ ret
+.size bn_mul_mont,.-bn_mul_mont
+___
+{{{
+my @A=("%r10","%r11");
+my @N=("%r13","%rdi");
+$code.=<<___;
+.type bn_mul4x_mont,\@function,6
+.align 16
+bn_mul4x_mont:
+.Lmul4x_enter:
+ push %rbx
+ push %rbp
+ push %r12
+ push %r13
+ push %r14
+ push %r15
+
+ mov ${num}d,${num}d
+ lea 4($num),%r10
+ mov %rsp,%r11
+ neg %r10
+ lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
+ and \$-1024,%rsp # minimize TLB usage
+
+ mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
+.Lmul4x_body:
+ mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
+ mov %rdx,%r12 # reassign $bp
+___
+ $bp="%r12";
+$code.=<<___;
+ mov ($n0),$n0 # pull n0[0] value
+ mov ($bp),$m0 # m0=bp[0]
+ mov ($ap),%rax
+
+ xor $i,$i # i=0
+ xor $j,$j # j=0
+
+ mov $n0,$m1
+ mulq $m0 # ap[0]*bp[0]
+ mov %rax,$A[0]
+ mov ($np),%rax
+
+ imulq $A[0],$m1 # "tp[0]"*n0
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[0]*m1
+ add %rax,$A[0] # discarded
+ mov 8($ap),%rax
+ adc \$0,%rdx
+ mov %rdx,$N[1]
+
+ mulq $m0
+ add %rax,$A[1]
+ mov 8($np),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[0]
+
+ mulq $m1
+ add %rax,$N[1]
+ mov 16($ap),%rax
+ adc \$0,%rdx
+ add $A[1],$N[1]
+ lea 4($j),$j # j++
+ adc \$0,%rdx
+ mov $N[1],(%rsp)
+ mov %rdx,$N[0]
+ jmp .L1st4x
+.align 16
+.L1st4x:
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[0]
+ mov -16($np,$j,8),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov -8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[0],-24(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[1]
+ mov -8($np,$j,8),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov ($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[1],-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[0]
+ mov ($np,$j,8),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov 8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[0],-8(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[1]
+ mov 8($np,$j,8),%rax
+ adc \$0,%rdx
+ lea 4($j),$j # j++
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov -16($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[1],-32(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+ cmp $num,$j
+ jl .L1st4x
+
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[0]
+ mov -16($np,$j,8),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov -8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[0],-24(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[0]
+ add %rax,$A[1]
+ mov -8($np,$j,8),%rax
+ adc \$0,%rdx
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov ($ap),%rax # ap[0]
+ adc \$0,%rdx
+ add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
+ adc \$0,%rdx
+ mov $N[1],-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+
+ xor $N[1],$N[1]
+ add $A[0],$N[0]
+ adc \$0,$N[1]
+ mov $N[0],-8(%rsp,$j,8)
+ mov $N[1],(%rsp,$j,8) # store upmost overflow bit
+
+ lea 1($i),$i # i++
+.align 4
+.Louter4x:
+ mov ($bp,$i,8),$m0 # m0=bp[i]
+ xor $j,$j # j=0
+ mov (%rsp),$A[0]
+ mov $n0,$m1
+ mulq $m0 # ap[0]*bp[i]
+ add %rax,$A[0] # ap[0]*bp[i]+tp[0]
+ mov ($np),%rax
+ adc \$0,%rdx
+
+ imulq $A[0],$m1 # tp[0]*n0
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[0]*m1
+ add %rax,$A[0] # "$N[0]", discarded
+ mov 8($ap),%rax
+ adc \$0,%rdx
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[1]
+ mov 8($np),%rax
+ adc \$0,%rdx
+ add 8(%rsp),$A[1] # +tp[1]
+ adc \$0,%rdx
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov 16($ap),%rax
+ adc \$0,%rdx
+ add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
+ lea 4($j),$j # j+=2
+ adc \$0,%rdx
+ mov $N[1],(%rsp) # tp[j-1]
+ mov %rdx,$N[0]
+ jmp .Linner4x
+.align 16
+.Linner4x:
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[0]
+ mov -16($np,$j,8),%rax
+ adc \$0,%rdx
+ add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov -8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0]
+ adc \$0,%rdx
+ mov $N[0],-24(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[1]
+ mov -8($np,$j,8),%rax
+ adc \$0,%rdx
+ add -8(%rsp,$j,8),$A[1]
+ adc \$0,%rdx
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
mov ($ap,$j,8),%rax
- mov %rax,($rp,$j,8) # rp[i]=tp[i]
- mov $i,(%rsp,$j,8) # zap temporary vector
+ adc \$0,%rdx
+ add $A[1],$N[1]
+ adc \$0,%rdx
+ mov $N[1],-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[0]
+ mov ($np,$j,8),%rax
+ adc \$0,%rdx
+ add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov 8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0]
+ adc \$0,%rdx
+ mov $N[0],-8(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[1]
+ mov 8($np,$j,8),%rax
+ adc \$0,%rdx
+ add 8(%rsp,$j,8),$A[1]
+ adc \$0,%rdx
+ lea 4($j),$j # j++
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov -16($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[1],$N[1]
+ adc \$0,%rdx
+ mov $N[1],-32(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+ cmp $num,$j
+ jl .Linner4x
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[0]
+ mov -16($np,$j,8),%rax
+ adc \$0,%rdx
+ add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
+ adc \$0,%rdx
+ mov %rdx,$A[1]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[0]
+ mov -8($ap,$j,8),%rax
+ adc \$0,%rdx
+ add $A[0],$N[0]
+ adc \$0,%rdx
+ mov $N[0],-24(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[1]
+
+ mulq $m0 # ap[j]*bp[i]
+ add %rax,$A[1]
+ mov -8($np,$j,8),%rax
+ adc \$0,%rdx
+ add -8(%rsp,$j,8),$A[1]
+ adc \$0,%rdx
+ lea 1($i),$i # i++
+ mov %rdx,$A[0]
+
+ mulq $m1 # np[j]*m1
+ add %rax,$N[1]
+ mov ($ap),%rax # ap[0]
+ adc \$0,%rdx
+ add $A[1],$N[1]
+ adc \$0,%rdx
+ mov $N[1],-16(%rsp,$j,8) # tp[j-1]
+ mov %rdx,$N[0]
+
+ xor $N[1],$N[1]
+ add $A[0],$N[0]
+ adc \$0,$N[1]
+ add (%rsp,$num,8),$N[0] # pull upmost overflow bit
+ adc \$0,$N[1]
+ mov $N[0],-8(%rsp,$j,8)
+ mov $N[1],(%rsp,$j,8) # store upmost overflow bit
+
+ cmp $num,$i
+ jl .Louter4x
+___
+{
+my @ri=("%rax","%rdx",$m0,$m1);
+$code.=<<___;
+ mov 16(%rsp,$num,8),$rp # restore $rp
+ mov 0(%rsp),@ri[0] # tp[0]
+ pxor %xmm0,%xmm0
+ mov 8(%rsp),@ri[1] # tp[1]
+ shr \$2,$num # num/=4
+ lea (%rsp),$ap # borrow ap for tp
+ xor $i,$i # i=0 and clear CF!
+
+ sub 0($np),@ri[0]
+ mov 16($ap),@ri[2] # tp[2]
+ mov 24($ap),@ri[3] # tp[3]
+ sbb 8($np),@ri[1]
+ lea -1($num),$j # j=num/4-1
+ jmp .Lsub4x
+.align 16
+.Lsub4x:
+ mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
+ mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 16($np,$i,8),@ri[2]
+ mov 32($ap,$i,8),@ri[0] # tp[i+1]
+ mov 40($ap,$i,8),@ri[1]
+ sbb 24($np,$i,8),@ri[3]
+ mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
+ mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 32($np,$i,8),@ri[0]
+ mov 48($ap,$i,8),@ri[2]
+ mov 56($ap,$i,8),@ri[3]
+ sbb 40($np,$i,8),@ri[1]
+ lea 4($i),$i # i++
+ dec $j # doesnn't affect CF!
+ jnz .Lsub4x
+
+ mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
+ mov 32($ap,$i,8),@ri[0] # load overflow bit
+ sbb 16($np,$i,8),@ri[2]
+ mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 24($np,$i,8),@ri[3]
+ mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
+
+ sbb \$0,@ri[0] # handle upmost overflow bit
+ mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
+ xor $i,$i # i=0
+ and @ri[0],$ap
+ not @ri[0]
+ mov $rp,$np
+ and @ri[0],$np
+ lea -1($num),$j
+ or $np,$ap # ap=borrow?tp:rp
+
+ movdqu ($ap),%xmm1
+ movdqa %xmm0,(%rsp)
+ movdqu %xmm1,($rp)
+ jmp .Lcopy4x
+.align 16
+.Lcopy4x: # copy or in-place refresh
+ movdqu 16($ap,$i),%xmm2
+ movdqu 32($ap,$i),%xmm1
+ movdqa %xmm0,16(%rsp,$i)
+ movdqu %xmm2,16($rp,$i)
+ movdqa %xmm0,32(%rsp,$i)
+ movdqu %xmm1,32($rp,$i)
+ lea 32($i),$i
dec $j
- jge .Lcopy
+ jnz .Lcopy4x
- mov 8(%rsp,$num,8),%rsp # restore %rsp
+ shl \$2,$num
+ movdqu 16($ap,$i),%xmm2
+ movdqa %xmm0,16(%rsp,$i)
+ movdqu %xmm2,16($rp,$i)
+___
+}
+$code.=<<___;
+ mov 8(%rsp,$num,8),%rsi # restore %rsp
mov \$1,%rax
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
+ mov (%rsi),%r15
+ mov 8(%rsi),%r14
+ mov 16(%rsi),%r13
+ mov 24(%rsi),%r12
+ mov 32(%rsi),%rbp
+ mov 40(%rsi),%rbx
+ lea 48(%rsi),%rsp
+.Lmul4x_epilogue:
ret
-.size bn_mul_mont,.-bn_mul_mont
+.size bn_mul4x_mont,.-bn_mul4x_mont
+___
+}}}
+ {{{
+######################################################################
+# void bn_sqr4x_mont(
+my $rptr="%rdi"; # const BN_ULONG *rptr,
+my $aptr="%rsi"; # const BN_ULONG *aptr,
+my $bptr="%rdx"; # not used
+my $nptr="%rcx"; # const BN_ULONG *nptr,
+my $n0 ="%r8"; # const BN_ULONG *n0);
+my $num ="%r9"; # int num, has to be divisible by 4 and
+ # not less than 8
+
+my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
+my @A0=("%r10","%r11");
+my @A1=("%r12","%r13");
+my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
+
+$code.=<<___;
+.type bn_sqr4x_mont,\@function,6
+.align 16
+bn_sqr4x_mont:
+.Lsqr4x_enter:
+ push %rbx
+ push %rbp
+ push %r12
+ push %r13
+ push %r14
+ push %r15
+
+ shl \$3,${num}d # convert $num to bytes
+ xor %r10,%r10
+ mov %rsp,%r11 # put aside %rsp
+ sub $num,%r10 # -$num
+ mov ($n0),$n0 # *n0
+ lea -72(%rsp,%r10,2),%rsp # alloca(frame+2*$num)
+ and \$-1024,%rsp # minimize TLB usage
+ ##############################################################
+ # Stack layout
+ #
+ # +0 saved $num, used in reduction section
+ # +8 &t[2*$num], used in reduction section
+ # +32 saved $rptr
+ # +40 saved $nptr
+ # +48 saved *n0
+ # +56 saved %rsp
+ # +64 t[2*$num]
+ #
+ mov $rptr,32(%rsp) # save $rptr
+ mov $nptr,40(%rsp)
+ mov $n0, 48(%rsp)
+ mov %r11, 56(%rsp) # save original %rsp
+.Lsqr4x_body:
+ ##############################################################
+ # Squaring part:
+ #
+ # a) multiply-n-add everything but a[i]*a[i];
+ # b) shift result of a) by 1 to the left and accumulate
+ # a[i]*a[i] products;
+ #
+ lea 32(%r10),$i # $i=-($num-32)
+ lea ($aptr,$num),$aptr # end of a[] buffer, ($aptr,$i)=&ap[2]
+
+ mov $num,$j # $j=$num
+
+ # comments apply to $num==8 case
+ mov -32($aptr,$i),$a0 # a[0]
+ lea 64(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
+ mov -24($aptr,$i),%rax # a[1]
+ lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
+ mov -16($aptr,$i),$ai # a[2]
+ mov %rax,$a1
+
+ mul $a0 # a[1]*a[0]
+ mov %rax,$A0[0] # a[1]*a[0]
+ mov $ai,%rax # a[2]
+ mov %rdx,$A0[1]
+ mov $A0[0],-24($tptr,$i) # t[1]
+
+ xor $A0[0],$A0[0]
+ mul $a0 # a[2]*a[0]
+ add %rax,$A0[1]
+ mov $ai,%rax
+ adc %rdx,$A0[0]
+ mov $A0[1],-16($tptr,$i) # t[2]
+
+ lea -16($i),$j # j=-16
+
+
+ mov 8($aptr,$j),$ai # a[3]
+ mul $a1 # a[2]*a[1]
+ mov %rax,$A1[0] # a[2]*a[1]+t[3]
+ mov $ai,%rax
+ mov %rdx,$A1[1]
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ lea 16($j),$j
+ adc \$0,$A0[1]
+ mul $a0 # a[3]*a[0]
+ add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],-8($tptr,$j) # t[3]
+ jmp .Lsqr4x_1st
+
+.align 16
+.Lsqr4x_1st:
+ mov ($aptr,$j),$ai # a[4]
+ xor $A1[0],$A1[0]
+ mul $a1 # a[3]*a[1]
+ add %rax,$A1[1] # a[3]*a[1]+t[4]
+ mov $ai,%rax
+ adc %rdx,$A1[0]
+
+ xor $A0[0],$A0[0]
+ add $A1[1],$A0[1]
+ adc \$0,$A0[0]
+ mul $a0 # a[4]*a[0]
+ add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
+ mov $ai,%rax # a[3]
+ adc %rdx,$A0[0]
+ mov $A0[1],($tptr,$j) # t[4]
+
+
+ mov 8($aptr,$j),$ai # a[5]
+ xor $A1[1],$A1[1]
+ mul $a1 # a[4]*a[3]
+ add %rax,$A1[0] # a[4]*a[3]+t[5]
+ mov $ai,%rax
+ adc %rdx,$A1[1]
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ adc \$0,$A0[1]
+ mul $a0 # a[5]*a[2]
+ add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],8($tptr,$j) # t[5]
+
+ mov 16($aptr,$j),$ai # a[6]
+ xor $A1[0],$A1[0]
+ mul $a1 # a[5]*a[3]
+ add %rax,$A1[1] # a[5]*a[3]+t[6]
+ mov $ai,%rax
+ adc %rdx,$A1[0]
+
+ xor $A0[0],$A0[0]
+ add $A1[1],$A0[1]
+ adc \$0,$A0[0]
+ mul $a0 # a[6]*a[2]
+ add %rax,$A0[1] # a[6]*a[2]+a[5]*a[3]+t[6]
+ mov $ai,%rax # a[3]
+ adc %rdx,$A0[0]
+ mov $A0[1],16($tptr,$j) # t[6]
+
+
+ mov 24($aptr,$j),$ai # a[7]
+ xor $A1[1],$A1[1]
+ mul $a1 # a[6]*a[5]
+ add %rax,$A1[0] # a[6]*a[5]+t[7]
+ mov $ai,%rax
+ adc %rdx,$A1[1]
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ lea 32($j),$j
+ adc \$0,$A0[1]
+ mul $a0 # a[7]*a[4]
+ add %rax,$A0[0] # a[7]*a[4]+a[6]*a[5]+t[6]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],-8($tptr,$j) # t[7]
+
+ cmp \$0,$j
+ jne .Lsqr4x_1st
+
+ xor $A1[0],$A1[0]
+ add $A0[1],$A1[1]
+ adc \$0,$A1[0]
+ mul $a1 # a[7]*a[5]
+ add %rax,$A1[1]
+ adc %rdx,$A1[0]
+
+ mov $A1[1],($tptr) # t[8]
+ lea 16($i),$i
+ mov $A1[0],8($tptr) # t[9]
+ jmp .Lsqr4x_outer
+
+.align 16
+.Lsqr4x_outer: # comments apply to $num==6 case
+ mov -32($aptr,$i),$a0 # a[0]
+ lea 64(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
+ mov -24($aptr,$i),%rax # a[1]
+ lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
+ mov -16($aptr,$i),$ai # a[2]
+ mov %rax,$a1
+
+ mov -24($tptr,$i),$A0[0] # t[1]
+ xor $A0[1],$A0[1]
+ mul $a0 # a[1]*a[0]
+ add %rax,$A0[0] # a[1]*a[0]+t[1]
+ mov $ai,%rax # a[2]
+ adc %rdx,$A0[1]
+ mov $A0[0],-24($tptr,$i) # t[1]
+
+ xor $A0[0],$A0[0]
+ add -16($tptr,$i),$A0[1] # a[2]*a[0]+t[2]
+ adc \$0,$A0[0]
+ mul $a0 # a[2]*a[0]
+ add %rax,$A0[1]
+ mov $ai,%rax
+ adc %rdx,$A0[0]
+ mov $A0[1],-16($tptr,$i) # t[2]
+
+ lea -16($i),$j # j=-16
+ xor $A1[0],$A1[0]
+
+
+ mov 8($aptr,$j),$ai # a[3]
+ xor $A1[1],$A1[1]
+ add 8($tptr,$j),$A1[0]
+ adc \$0,$A1[1]
+ mul $a1 # a[2]*a[1]
+ add %rax,$A1[0] # a[2]*a[1]+t[3]
+ mov $ai,%rax
+ adc %rdx,$A1[1]
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ adc \$0,$A0[1]
+ mul $a0 # a[3]*a[0]
+ add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],8($tptr,$j) # t[3]
+
+ lea 16($j),$j
+ jmp .Lsqr4x_inner
+
+.align 16
+.Lsqr4x_inner:
+ mov ($aptr,$j),$ai # a[4]
+ xor $A1[0],$A1[0]
+ add ($tptr,$j),$A1[1]
+ adc \$0,$A1[0]
+ mul $a1 # a[3]*a[1]
+ add %rax,$A1[1] # a[3]*a[1]+t[4]
+ mov $ai,%rax
+ adc %rdx,$A1[0]
+
+ xor $A0[0],$A0[0]
+ add $A1[1],$A0[1]
+ adc \$0,$A0[0]
+ mul $a0 # a[4]*a[0]
+ add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
+ mov $ai,%rax # a[3]
+ adc %rdx,$A0[0]
+ mov $A0[1],($tptr,$j) # t[4]
+
+ mov 8($aptr,$j),$ai # a[5]
+ xor $A1[1],$A1[1]
+ add 8($tptr,$j),$A1[0]
+ adc \$0,$A1[1]
+ mul $a1 # a[4]*a[3]
+ add %rax,$A1[0] # a[4]*a[3]+t[5]
+ mov $ai,%rax
+ adc %rdx,$A1[1]
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ lea 16($j),$j # j++
+ adc \$0,$A0[1]
+ mul $a0 # a[5]*a[2]
+ add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],-8($tptr,$j) # t[5], "preloaded t[1]" below
+
+ cmp \$0,$j
+ jne .Lsqr4x_inner
+
+ xor $A1[0],$A1[0]
+ add $A0[1],$A1[1]
+ adc \$0,$A1[0]
+ mul $a1 # a[5]*a[3]
+ add %rax,$A1[1]
+ adc %rdx,$A1[0]
+
+ mov $A1[1],($tptr) # t[6], "preloaded t[2]" below
+ mov $A1[0],8($tptr) # t[7], "preloaded t[3]" below
+
+ add \$16,$i
+ jnz .Lsqr4x_outer
+
+ # comments apply to $num==4 case
+ mov -32($aptr),$a0 # a[0]
+ lea 64(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
+ mov -24($aptr),%rax # a[1]
+ lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
+ mov -16($aptr),$ai # a[2]
+ mov %rax,$a1
+
+ xor $A0[1],$A0[1]
+ mul $a0 # a[1]*a[0]
+ add %rax,$A0[0] # a[1]*a[0]+t[1], preloaded t[1]
+ mov $ai,%rax # a[2]
+ adc %rdx,$A0[1]
+ mov $A0[0],-24($tptr) # t[1]
+
+ xor $A0[0],$A0[0]
+ add $A1[1],$A0[1] # a[2]*a[0]+t[2], preloaded t[2]
+ adc \$0,$A0[0]
+ mul $a0 # a[2]*a[0]
+ add %rax,$A0[1]
+ mov $ai,%rax
+ adc %rdx,$A0[0]
+ mov $A0[1],-16($tptr) # t[2]
+
+ mov -8($aptr),$ai # a[3]
+ mul $a1 # a[2]*a[1]
+ add %rax,$A1[0] # a[2]*a[1]+t[3], preloaded t[3]
+ mov $ai,%rax
+ adc \$0,%rdx
+
+ xor $A0[1],$A0[1]
+ add $A1[0],$A0[0]
+ mov %rdx,$A1[1]
+ adc \$0,$A0[1]
+ mul $a0 # a[3]*a[0]
+ add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
+ mov $ai,%rax
+ adc %rdx,$A0[1]
+ mov $A0[0],-8($tptr) # t[3]
+
+ xor $A1[0],$A1[0]
+ add $A0[1],$A1[1]
+ adc \$0,$A1[0]
+ mul $a1 # a[3]*a[1]
+ add %rax,$A1[1]
+ mov -16($aptr),%rax # a[2]
+ adc %rdx,$A1[0]
+
+ mov $A1[1],($tptr) # t[4]
+ mov $A1[0],8($tptr) # t[5]
+
+ mul $ai # a[2]*a[3]
+___
+{
+my ($shift,$carry)=($a0,$a1);
+my @S=(@A1,$ai,$n0);
+$code.=<<___;
+ add \$16,$i
+ xor $shift,$shift
+ sub $num,$i # $i=16-$num
+ xor $carry,$carry
+
+ add $A1[0],%rax # t[5]
+ adc \$0,%rdx
+ mov %rax,8($tptr) # t[5]
+ mov %rdx,16($tptr) # t[6]
+ mov $carry,24($tptr) # t[7]
+
+ mov -16($aptr,$i),%rax # a[0]
+ lea 64(%rsp,$num,2),$tptr
+ xor $A0[0],$A0[0] # t[0]
+ mov -24($tptr,$i,2),$A0[1] # t[1]
+
+ lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[1] # | t[2*i]>>63
+ mov -16($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov -8($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[0]
+ mov -8($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[0],-32($tptr,$i,2)
+ adc %rdx,$S[1]
+
+ lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
+ mov $S[1],-24($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[3] # | t[2*i]>>63
+ mov 0($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov 8($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[2]
+ mov 0($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[2],-16($tptr,$i,2)
+ adc %rdx,$S[3]
+ lea 16($i),$i
+ mov $S[3],-40($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ jmp .Lsqr4x_shift_n_add
+
+.align 16
+.Lsqr4x_shift_n_add:
+ lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[1] # | t[2*i]>>63
+ mov -16($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov -8($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[0]
+ mov -8($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[0],-32($tptr,$i,2)
+ adc %rdx,$S[1]
+
+ lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
+ mov $S[1],-24($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[3] # | t[2*i]>>63
+ mov 0($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov 8($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[2]
+ mov 0($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[2],-16($tptr,$i,2)
+ adc %rdx,$S[3]
+
+ lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
+ mov $S[3],-8($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[1] # | t[2*i]>>63
+ mov 16($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov 24($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[0]
+ mov 8($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[0],0($tptr,$i,2)
+ adc %rdx,$S[1]
+
+ lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
+ mov $S[1],8($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[3] # | t[2*i]>>63
+ mov 32($tptr,$i,2),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov 40($tptr,$i,2),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[2]
+ mov 16($aptr,$i),%rax # a[i+1] # prefetch
+ mov $S[2],16($tptr,$i,2)
+ adc %rdx,$S[3]
+ mov $S[3],24($tptr,$i,2)
+ sbb $carry,$carry # mov cf,$carry
+ add \$32,$i
+ jnz .Lsqr4x_shift_n_add
+
+ lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[1] # | t[2*i]>>63
+ mov -16($tptr),$A0[0] # t[2*i+2] # prefetch
+ mov $A0[1],$shift # shift=t[2*i+1]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ mov -8($tptr),$A0[1] # t[2*i+2+1] # prefetch
+ adc %rax,$S[0]
+ mov -8($aptr),%rax # a[i+1] # prefetch
+ mov $S[0],-32($tptr)
+ adc %rdx,$S[1]
+
+ lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1|shift
+ mov $S[1],-24($tptr)
+ sbb $carry,$carry # mov cf,$carry
+ shr \$63,$A0[0]
+ lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
+ shr \$63,$A0[1]
+ or $A0[0],$S[3] # | t[2*i]>>63
+ mul %rax # a[i]*a[i]
+ neg $carry # mov $carry,cf
+ adc %rax,$S[2]
+ adc %rdx,$S[3]
+ mov $S[2],-16($tptr)
+ mov $S[3],-8($tptr)
+___
+}
+##############################################################
+# Montgomery reduction part, "word-by-word" algorithm.
+#
+{
+my ($topbit,$nptr)=("%rbp",$aptr);
+my ($m0,$m1)=($a0,$a1);
+my @Ni=("%rbx","%r9");
+$code.=<<___;
+ mov 40(%rsp),$nptr # restore $nptr
+ mov 48(%rsp),$n0 # restore *n0
+ xor $j,$j
+ mov $num,0(%rsp) # save $num
+ sub $num,$j # $j=-$num
+ mov 64(%rsp),$A0[0] # t[0] # modsched #
+ mov $n0,$m0 # # modsched #
+ lea 64(%rsp,$num,2),%rax # end of t[] buffer
+ lea 64(%rsp,$num),$tptr # end of t[] window
+ mov %rax,8(%rsp) # save end of t[] buffer
+ lea ($nptr,$num),$nptr # end of n[] buffer
+ xor $topbit,$topbit # $topbit=0
+
+ mov 0($nptr,$j),%rax # n[0] # modsched #
+ mov 8($nptr,$j),$Ni[1] # n[1] # modsched #
+ imulq $A0[0],$m0 # m0=t[0]*n0 # modsched #
+ mov %rax,$Ni[0] # # modsched #
+ jmp .Lsqr4x_mont_outer
+
+.align 16
+.Lsqr4x_mont_outer:
+ xor $A0[1],$A0[1]
+ mul $m0 # n[0]*m0
+ add %rax,$A0[0] # n[0]*m0+t[0]
+ mov $Ni[1],%rax
+ adc %rdx,$A0[1]
+ mov $n0,$m1
+
+ xor $A0[0],$A0[0]
+ add 8($tptr,$j),$A0[1]
+ adc \$0,$A0[0]
+ mul $m0 # n[1]*m0
+ add %rax,$A0[1] # n[1]*m0+t[1]
+ mov $Ni[0],%rax
+ adc %rdx,$A0[0]
+
+ imulq $A0[1],$m1
+
+ mov 16($nptr,$j),$Ni[0] # n[2]
+ xor $A1[1],$A1[1]
+ add $A0[1],$A1[0]
+ adc \$0,$A1[1]
+ mul $m1 # n[0]*m1
+ add %rax,$A1[0] # n[0]*m1+"t[1]"
+ mov $Ni[0],%rax
+ adc %rdx,$A1[1]
+ mov $A1[0],8($tptr,$j) # "t[1]"
+
+ xor $A0[1],$A0[1]
+ add 16($tptr,$j),$A0[0]
+ adc \$0,$A0[1]
+ mul $m0 # n[2]*m0
+ add %rax,$A0[0] # n[2]*m0+t[2]
+ mov $Ni[1],%rax
+ adc %rdx,$A0[1]
+
+ mov 24($nptr,$j),$Ni[1] # n[3]
+ xor $A1[0],$A1[0]
+ add $A0[0],$A1[1]
+ adc \$0,$A1[0]
+ mul $m1 # n[1]*m1
+ add %rax,$A1[1] # n[1]*m1+"t[2]"
+ mov $Ni[1],%rax
+ adc %rdx,$A1[0]
+ mov $A1[1],16($tptr,$j) # "t[2]"
+
+ xor $A0[0],$A0[0]
+ add 24($tptr,$j),$A0[1]
+ lea 32($j),$j
+ adc \$0,$A0[0]
+ mul $m0 # n[3]*m0
+ add %rax,$A0[1] # n[3]*m0+t[3]
+ mov $Ni[0],%rax
+ adc %rdx,$A0[0]
+ jmp .Lsqr4x_mont_inner
+
+.align 16
+.Lsqr4x_mont_inner:
+ mov ($nptr,$j),$Ni[0] # n[4]
+ xor $A1[1],$A1[1]
+ add $A0[1],$A1[0]
+ adc \$0,$A1[1]
+ mul $m1 # n[2]*m1
+ add %rax,$A1[0] # n[2]*m1+"t[3]"
+ mov $Ni[0],%rax
+ adc %rdx,$A1[1]
+ mov $A1[0],-8($tptr,$j) # "t[3]"
+
+ xor $A0[1],$A0[1]
+ add ($tptr,$j),$A0[0]
+ adc \$0,$A0[1]
+ mul $m0 # n[4]*m0
+ add %rax,$A0[0] # n[4]*m0+t[4]
+ mov $Ni[1],%rax
+ adc %rdx,$A0[1]
+
+ mov 8($nptr,$j),$Ni[1] # n[5]
+ xor $A1[0],$A1[0]
+ add $A0[0],$A1[1]
+ adc \$0,$A1[0]
+ mul $m1 # n[3]*m1
+ add %rax,$A1[1] # n[3]*m1+"t[4]"
+ mov $Ni[1],%rax
+ adc %rdx,$A1[0]
+ mov $A1[1],($tptr,$j) # "t[4]"
+
+ xor $A0[0],$A0[0]
+ add 8($tptr,$j),$A0[1]
+ adc \$0,$A0[0]
+ mul $m0 # n[5]*m0
+ add %rax,$A0[1] # n[5]*m0+t[5]
+ mov $Ni[0],%rax
+ adc %rdx,$A0[0]
+
+
+ mov 16($nptr,$j),$Ni[0] # n[6]
+ xor $A1[1],$A1[1]
+ add $A0[1],$A1[0]
+ adc \$0,$A1[1]
+ mul $m1 # n[4]*m1
+ add %rax,$A1[0] # n[4]*m1+"t[5]"
+ mov $Ni[0],%rax
+ adc %rdx,$A1[1]
+ mov $A1[0],8($tptr,$j) # "t[5]"
+
+ xor $A0[1],$A0[1]
+ add 16($tptr,$j),$A0[0]
+ adc \$0,$A0[1]
+ mul $m0 # n[6]*m0
+ add %rax,$A0[0] # n[6]*m0+t[6]
+ mov $Ni[1],%rax
+ adc %rdx,$A0[1]
+
+ mov 24($nptr,$j),$Ni[1] # n[7]
+ xor $A1[0],$A1[0]
+ add $A0[0],$A1[1]
+ adc \$0,$A1[0]
+ mul $m1 # n[5]*m1
+ add %rax,$A1[1] # n[5]*m1+"t[6]"
+ mov $Ni[1],%rax
+ adc %rdx,$A1[0]
+ mov $A1[1],16($tptr,$j) # "t[6]"
+
+ xor $A0[0],$A0[0]
+ add 24($tptr,$j),$A0[1]
+ lea 32($j),$j
+ adc \$0,$A0[0]
+ mul $m0 # n[7]*m0
+ add %rax,$A0[1] # n[7]*m0+t[7]
+ mov $Ni[0],%rax
+ adc %rdx,$A0[0]
+ cmp \$0,$j
+ jne .Lsqr4x_mont_inner
+
+ sub 0(%rsp),$j # $j=-$num # modsched #
+ mov $n0,$m0 # # modsched #
+
+ xor $A1[1],$A1[1]
+ add $A0[1],$A1[0]
+ adc \$0,$A1[1]
+ mul $m1 # n[6]*m1
+ add %rax,$A1[0] # n[6]*m1+"t[7]"
+ mov $Ni[1],%rax
+ adc %rdx,$A1[1]
+ mov $A1[0],-8($tptr) # "t[7]"
+
+ xor $A0[1],$A0[1]
+ add ($tptr),$A0[0] # +t[8]
+ adc \$0,$A0[1]
+ mov 0($nptr,$j),$Ni[0] # n[0] # modsched #
+ add $topbit,$A0[0]
+ adc \$0,$A0[1]
+
+ imulq 16($tptr,$j),$m0 # m0=t[0]*n0 # modsched #
+ xor $A1[0],$A1[0]
+ mov 8($nptr,$j),$Ni[1] # n[1] # modsched #
+ add $A0[0],$A1[1]
+ mov 16($tptr,$j),$A0[0] # t[0] # modsched #
+ adc \$0,$A1[0]
+ mul $m1 # n[7]*m1
+ add %rax,$A1[1] # n[7]*m1+"t[8]"
+ mov $Ni[0],%rax # # modsched #
+ adc %rdx,$A1[0]
+ mov $A1[1],($tptr) # "t[8]"
+
+ xor $topbit,$topbit
+ add 8($tptr),$A1[0] # +t[9]
+ adc $topbit,$topbit
+ add $A0[1],$A1[0]
+ lea 16($tptr),$tptr # "t[$num]>>128"
+ adc \$0,$topbit
+ mov $A1[0],-8($tptr) # "t[9]"
+ cmp 8(%rsp),$tptr # are we done?
+ jb .Lsqr4x_mont_outer
+
+ mov 0(%rsp),$num # restore $num
+ mov $topbit,($tptr) # save $topbit
+___
+}
+##############################################################
+# Post-condition, 4x unrolled copy from bn_mul_mont
+#
+{
+my ($tptr,$nptr)=("%rbx",$aptr);
+my @ri=("%rax","%rdx","%r10","%r11");
+$code.=<<___;
+ mov 64(%rsp,$num),@ri[0] # tp[0]
+ lea 64(%rsp,$num),$tptr # upper half of t[2*$num] holds result
+ mov 40(%rsp),$nptr # restore $nptr
+ shr \$5,$num # num/4
+ mov 8($tptr),@ri[1] # t[1]
+ xor $i,$i # i=0 and clear CF!
+
+ mov 32(%rsp),$rptr # restore $rptr
+ sub 0($nptr),@ri[0]
+ mov 16($tptr),@ri[2] # t[2]
+ mov 24($tptr),@ri[3] # t[3]
+ sbb 8($nptr),@ri[1]
+ lea -1($num),$j # j=num/4-1
+ jmp .Lsqr4x_sub
+.align 16
+.Lsqr4x_sub:
+ mov @ri[0],0($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ mov @ri[1],8($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 16($nptr,$i,8),@ri[2]
+ mov 32($tptr,$i,8),@ri[0] # tp[i+1]
+ mov 40($tptr,$i,8),@ri[1]
+ sbb 24($nptr,$i,8),@ri[3]
+ mov @ri[2],16($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ mov @ri[3],24($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 32($nptr,$i,8),@ri[0]
+ mov 48($tptr,$i,8),@ri[2]
+ mov 56($tptr,$i,8),@ri[3]
+ sbb 40($nptr,$i,8),@ri[1]
+ lea 4($i),$i # i++
+ dec $j # doesn't affect CF!
+ jnz .Lsqr4x_sub
+
+ mov @ri[0],0($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ mov 32($tptr,$i,8),@ri[0] # load overflow bit
+ sbb 16($nptr,$i,8),@ri[2]
+ mov @ri[1],8($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ sbb 24($nptr,$i,8),@ri[3]
+ mov @ri[2],16($rptr,$i,8) # rp[i]=tp[i]-np[i]
+
+ sbb \$0,@ri[0] # handle upmost overflow bit
+ mov @ri[3],24($rptr,$i,8) # rp[i]=tp[i]-np[i]
+ xor $i,$i # i=0
+ and @ri[0],$tptr
+ not @ri[0]
+ mov $rptr,$nptr
+ and @ri[0],$nptr
+ lea -1($num),$j
+ or $nptr,$tptr # tp=borrow?tp:rp
+
+ pxor %xmm0,%xmm0
+ lea 64(%rsp,$num,8),$nptr
+ movdqu ($tptr),%xmm1
+ lea ($nptr,$num,8),$nptr
+ movdqa %xmm0,64(%rsp) # zap lower half of temporary vector
+ movdqa %xmm0,($nptr) # zap upper half of temporary vector
+ movdqu %xmm1,($rptr)
+ jmp .Lsqr4x_copy
+.align 16
+.Lsqr4x_copy: # copy or in-place refresh
+ movdqu 16($tptr,$i),%xmm2
+ movdqu 32($tptr,$i),%xmm1
+ movdqa %xmm0,80(%rsp,$i) # zap lower half of temporary vector
+ movdqa %xmm0,96(%rsp,$i) # zap lower half of temporary vector
+ movdqa %xmm0,16($nptr,$i) # zap upper half of temporary vector
+ movdqa %xmm0,32($nptr,$i) # zap upper half of temporary vector
+ movdqu %xmm2,16($rptr,$i)
+ movdqu %xmm1,32($rptr,$i)
+ lea 32($i),$i
+ dec $j
+ jnz .Lsqr4x_copy
+
+ movdqu 16($tptr,$i),%xmm2
+ movdqa %xmm0,80(%rsp,$i) # zap lower half of temporary vector
+ movdqa %xmm0,16($nptr,$i) # zap upper half of temporary vector
+ movdqu %xmm2,16($rptr,$i)
+___
+}
+$code.=<<___;
+ mov 56(%rsp),%rsi # restore %rsp
+ mov \$1,%rax
+ mov 0(%rsi),%r15
+ mov 8(%rsi),%r14
+ mov 16(%rsi),%r13
+ mov 24(%rsi),%r12
+ mov 32(%rsi),%rbp
+ mov 40(%rsi),%rbx
+ lea 48(%rsi),%rsp
+.Lsqr4x_epilogue:
+ ret
+.size bn_sqr4x_mont,.-bn_sqr4x_mont
+___
+}}}
+$code.=<<___;
.asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align 16
___
print $code;
diff --git a/lib/libcrypto/camellia/asm/cmll-x86_64.pl b/lib/libcrypto/camellia/asm/cmll-x86_64.pl
index c683646ca7c..a171c654b2d 100644
--- a/lib/libcrypto/camellia/asm/cmll-x86_64.pl
+++ b/lib/libcrypto/camellia/asm/cmll-x86_64.pl
@@ -33,14 +33,13 @@ $flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $flavour $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
sub hi() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1h/; $r; }
sub lo() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/;
@@ -56,7 +55,7 @@ $inp="%r12";
$out="%r13";
$key="%r14";
$keyend="%r15";
-$arg0d=$win64?"%ecx":"%edi";
+$arg0d="%edi";
# const unsigned int Camellia_SBOX[4][256];
# Well, sort of... Camellia_SBOX[0][] is interleaved with [1][],
@@ -656,7 +655,7 @@ Camellia_cbc_encrypt:
mov %rsi,$out # out argument
mov %r8,%rbx # ivp argument
mov %rcx,$key # key argument
- mov 272(%rcx),$keyend # grandRounds
+ mov 272(%rcx),${keyend}d # grandRounds
mov %r8,$_ivp
mov %rbp,$_rsp
@@ -859,219 +858,7 @@ Camellia_cbc_encrypt:
ret
.size Camellia_cbc_encrypt,.-Camellia_cbc_encrypt
-.asciz "Camellia for x86_64 by <appro@openssl.org>"
-___
-}
-
-# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
-# CONTEXT *context,DISPATCHER_CONTEXT *disp)
-if ($win64) {
-$rec="%rcx";
-$frame="%rdx";
-$context="%r8";
-$disp="%r9";
-
-$code.=<<___;
-.extern __imp_RtlVirtualUnwind
-.type common_se_handler,\@abi-omnipotent
-.align 16
-common_se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- lea -64(%rsp),%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- mov 8($disp),%rsi # disp->ImageBase
- mov 56($disp),%r11 # disp->HandlerData
-
- mov 0(%r11),%r10d # HandlerData[0]
- lea (%rsi,%r10),%r10 # prologue label
- cmp %r10,%rbx # context->Rip<prologue label
- jb .Lin_prologue
-
- mov 152($context),%rax # pull context->Rsp
-
- mov 4(%r11),%r10d # HandlerData[1]
- lea (%rsi,%r10),%r10 # epilogue label
- cmp %r10,%rbx # context->Rip>=epilogue label
- jae .Lin_prologue
-
- lea 40(%rax),%rax
- mov -8(%rax),%rbx
- mov -16(%rax),%rbp
- mov -24(%rax),%r13
- mov -32(%rax),%r14
- mov -40(%rax),%r15
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r13,224($context) # restore context->R13
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
-
-.Lin_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
- jmp .Lcommon_seh_exit
-.size common_se_handler,.-common_se_handler
-
-.type cbc_se_handler,\@abi-omnipotent
-.align 16
-cbc_se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- lea -64(%rsp),%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- lea .Lcbc_prologue(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lcbc_prologue
- jb .Lin_cbc_prologue
-
- lea .Lcbc_body(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lcbc_body
- jb .Lin_cbc_frame_setup
-
- mov 152($context),%rax # pull context->Rsp
-
- lea .Lcbc_abort(%rip),%r10
- cmp %r10,%rbx # context->Rip>=.Lcbc_abort
- jae .Lin_cbc_prologue
-
- # handle pushf/popf in Camellia_cbc_encrypt
- lea .Lcbc_enc_pushf(%rip),%r10
- cmp %r10,%rbx # context->Rip<=.Lcbc_enc_pushf
- jbe .Lin_cbc_no_flag
- lea 8(%rax),%rax
- lea .Lcbc_enc_popf(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lcbc_enc_popf
- jb .Lin_cbc_no_flag
- lea -8(%rax),%rax
- lea .Lcbc_dec_pushf(%rip),%r10
- cmp %r10,%rbx # context->Rip<=.Lcbc_dec_pushf
- jbe .Lin_cbc_no_flag
- lea 8(%rax),%rax
- lea .Lcbc_dec_popf(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lcbc_dec_popf
- jb .Lin_cbc_no_flag
- lea -8(%rax),%rax
-
-.Lin_cbc_no_flag:
- mov 48(%rax),%rax # $_rsp
- lea 48(%rax),%rax
-
-.Lin_cbc_frame_setup:
- mov -8(%rax),%rbx
- mov -16(%rax),%rbp
- mov -24(%rax),%r12
- mov -32(%rax),%r13
- mov -40(%rax),%r14
- mov -48(%rax),%r15
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R13
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
-
-.Lin_cbc_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
-.align 4
-.Lcommon_seh_exit:
-
- mov 40($disp),%rdi # disp->ContextRecord
- mov $context,%rsi # context
- mov \$`1232/8`,%ecx # sizeof(CONTEXT)
- .long 0xa548f3fc # cld; rep movsq
-
- mov $disp,%rsi
- xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
- mov 8(%rsi),%rdx # arg2, disp->ImageBase
- mov 0(%rsi),%r8 # arg3, disp->ControlPc
- mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
- mov 40(%rsi),%r10 # disp->ContextRecord
- lea 56(%rsi),%r11 # &disp->HandlerData
- lea 24(%rsi),%r12 # &disp->EstablisherFrame
- mov %r10,32(%rsp) # arg5
- mov %r11,40(%rsp) # arg6
- mov %r12,48(%rsp) # arg7
- mov %rcx,56(%rsp) # arg8, (NULL)
- call *__imp_RtlVirtualUnwind(%rip)
-
- mov \$1,%eax # ExceptionContinueSearch
- lea 64(%rsp),%rsp
- popfq
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
- pop %rdi
- pop %rsi
- ret
-.size cbc_se_handler,.-cbc_se_handler
-
-.section .pdata
-.align 4
- .rva .LSEH_begin_Camellia_EncryptBlock_Rounds
- .rva .LSEH_end_Camellia_EncryptBlock_Rounds
- .rva .LSEH_info_Camellia_EncryptBlock_Rounds
-
- .rva .LSEH_begin_Camellia_DecryptBlock_Rounds
- .rva .LSEH_end_Camellia_DecryptBlock_Rounds
- .rva .LSEH_info_Camellia_DecryptBlock_Rounds
-
- .rva .LSEH_begin_Camellia_Ekeygen
- .rva .LSEH_end_Camellia_Ekeygen
- .rva .LSEH_info_Camellia_Ekeygen
-
- .rva .LSEH_begin_Camellia_cbc_encrypt
- .rva .LSEH_end_Camellia_cbc_encrypt
- .rva .LSEH_info_Camellia_cbc_encrypt
-
-.section .xdata
-.align 8
-.LSEH_info_Camellia_EncryptBlock_Rounds:
- .byte 9,0,0,0
- .rva common_se_handler
- .rva .Lenc_prologue,.Lenc_epilogue # HandlerData[]
-.LSEH_info_Camellia_DecryptBlock_Rounds:
- .byte 9,0,0,0
- .rva common_se_handler
- .rva .Ldec_prologue,.Ldec_epilogue # HandlerData[]
-.LSEH_info_Camellia_Ekeygen:
- .byte 9,0,0,0
- .rva common_se_handler
- .rva .Lkey_prologue,.Lkey_epilogue # HandlerData[]
-.LSEH_info_Camellia_cbc_encrypt:
- .byte 9,0,0,0
- .rva cbc_se_handler
+.asciz "Camellia for x86_64 by <appro\@openssl.org>"
___
}
diff --git a/lib/libcrypto/md5/asm/md5-x86_64.pl b/lib/libcrypto/md5/asm/md5-x86_64.pl
index 9a6fa67224e..c902a1b532f 100755
--- a/lib/libcrypto/md5/asm/md5-x86_64.pl
+++ b/lib/libcrypto/md5/asm/md5-x86_64.pl
@@ -15,7 +15,7 @@ my $code;
# dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
# %r10d = X[k_next]
# %r11d = z' (copy of z for the next step)
-# Each round1_step() takes about 5.71 clocks (9 instructions, 1.58 IPC)
+# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
sub round1_step
{
my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -37,22 +37,26 @@ EOF
# round2_step() does:
# dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
# %r10d = X[k_next]
-# %r11d = y' (copy of y for the next step)
-# Each round2_step() takes about 6.22 clocks (9 instructions, 1.45 IPC)
+# %r11d = z' (copy of z for the next step)
+# %r12d = z' (copy of z for the next step)
+# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
sub round2_step
{
my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
$code .= " mov 1*4(%rsi), %r10d /* (NEXT STEP) X[1] */\n" if ($pos == -1);
- $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
+ $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
+ $code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
$code .= <<EOF;
- xor $x, %r11d /* x ^ ... */
+ not %r11d /* not z */
lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
- and $z, %r11d /* z & ... */
- xor $y, %r11d /* y ^ ... */
+ and $x, %r12d /* x & z */
+ and $y, %r11d /* y & (not z) */
mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
- add %r11d, $dst /* dst += ... */
+ or %r11d, %r12d /* (y & (not z)) | (x & z) */
+ mov $y, %r11d /* (NEXT STEP) z' = $y */
+ add %r12d, $dst /* dst += ... */
+ mov $y, %r12d /* (NEXT STEP) z' = $y */
rol \$$s, $dst /* dst <<< s */
- mov $x, %r11d /* (NEXT STEP) y' = $x */
add $x, $dst /* dst += x */
EOF
}
@@ -61,7 +65,7 @@ EOF
# dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
# %r10d = X[k_next]
# %r11d = y' (copy of y for the next step)
-# Each round3_step() takes about 4.26 clocks (8 instructions, 1.88 IPC)
+# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
sub round3_step
{
my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -83,7 +87,7 @@ EOF
# dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
# %r10d = X[k_next]
# %r11d = not z' (copy of not z for the next step)
-# Each round4_step() takes about 5.27 clocks (9 instructions, 1.71 IPC)
+# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
sub round4_step
{
my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -104,8 +108,18 @@ sub round4_step
EOF
}
-my $output = shift;
-open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
+my $flavour = shift;
+my $output = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+no warnings qw(uninitialized);
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
$code .= <<EOF;
.text
@@ -116,8 +130,10 @@ $code .= <<EOF;
md5_block_asm_data_order:
push %rbp
push %rbx
+ push %r12
push %r14
push %r15
+.Lprologue:
# rdi = arg #1 (ctx, MD5_CTX pointer)
# rsi = arg #2 (ptr, data pointer)
@@ -232,10 +248,13 @@ $code .= <<EOF;
mov %ecx, 2*4(%rbp) # ctx->C = C
mov %edx, 3*4(%rbp) # ctx->D = D
- pop %r15
- pop %r14
- pop %rbx
- pop %rbp
+ mov (%rsp),%r15
+ mov 8(%rsp),%r14
+ mov 16(%rsp),%r12
+ mov 24(%rsp),%rbx
+ mov 32(%rsp),%rbp
+ add \$40,%rsp
+.Lepilogue:
ret
.size md5_block_asm_data_order,.-md5_block_asm_data_order
EOF
diff --git a/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl b/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
index 7f684092d40..501d9e936bb 100644
--- a/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
+++ b/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
@@ -44,14 +44,13 @@ my $flavour = shift;
my $output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $flavour $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
my ($dat,$in0,$out,$ctx,$inp,$len, $func,$nargs);
@@ -504,111 +503,6 @@ RC4_options:
.size RC4_options,.-RC4_options
___
}
-# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
-# CONTEXT *context,DISPATCHER_CONTEXT *disp)
-if ($win64) {
-my $rec="%rcx";
-my $frame="%rdx";
-my $context="%r8";
-my $disp="%r9";
-
-$code.=<<___;
-.extern __imp_RtlVirtualUnwind
-.type se_handler,\@abi-omnipotent
-.align 16
-se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- sub \$64,%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- lea .Lbody(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lbody
- jb .Lin_prologue
-
- mov 152($context),%rax # pull context->Rsp
-
- lea .Lepilogue(%rip),%r10
- cmp %r10,%rbx # context->Rip>=.Lepilogue
- jae .Lin_prologue
-
- mov 40(%rax),%r15
- mov 48(%rax),%r14
- mov 56(%rax),%r13
- mov 64(%rax),%r12
- mov 72(%rax),%rbp
- mov 80(%rax),%rbx
- lea 88(%rax),%rax
-
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R12
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
-
-.Lin_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
- mov 40($disp),%rdi # disp->ContextRecord
- mov $context,%rsi # context
- mov \$154,%ecx # sizeof(CONTEXT)
- .long 0xa548f3fc # cld; rep movsq
-
- mov $disp,%rsi
- xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
- mov 8(%rsi),%rdx # arg2, disp->ImageBase
- mov 0(%rsi),%r8 # arg3, disp->ControlPc
- mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
- mov 40(%rsi),%r10 # disp->ContextRecord
- lea 56(%rsi),%r11 # &disp->HandlerData
- lea 24(%rsi),%r12 # &disp->EstablisherFrame
- mov %r10,32(%rsp) # arg5
- mov %r11,40(%rsp) # arg6
- mov %r12,48(%rsp) # arg7
- mov %rcx,56(%rsp) # arg8, (NULL)
- call *__imp_RtlVirtualUnwind(%rip)
-
- mov \$1,%eax # ExceptionContinueSearch
- add \$64,%rsp
- popfq
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
- pop %rdi
- pop %rsi
- ret
-.size se_handler,.-se_handler
-
-.section .pdata
-.align 4
- .rva .LSEH_begin_$func
- .rva .LSEH_end_$func
- .rva .LSEH_info_$func
-
-.section .xdata
-.align 8
-.LSEH_info_$func:
- .byte 9,0,0,0
- .rva se_handler
-___
-}
sub reg_part {
my ($reg,$conv)=@_;
diff --git a/lib/libcrypto/rc4/asm/rc4-x86_64.pl b/lib/libcrypto/rc4/asm/rc4-x86_64.pl
index 75750dbf334..2bed1e279fc 100755
--- a/lib/libcrypto/rc4/asm/rc4-x86_64.pl
+++ b/lib/libcrypto/rc4/asm/rc4-x86_64.pl
@@ -105,8 +105,6 @@ $flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
@@ -528,138 +526,6 @@ RC4_options:
.size RC4_options,.-RC4_options
___
-# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
-# CONTEXT *context,DISPATCHER_CONTEXT *disp)
-if ($win64) {
-$rec="%rcx";
-$frame="%rdx";
-$context="%r8";
-$disp="%r9";
-
-$code.=<<___;
-.extern __imp_RtlVirtualUnwind
-.type stream_se_handler,\@abi-omnipotent
-.align 16
-stream_se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- sub \$64,%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- lea .Lprologue(%rip),%r10
- cmp %r10,%rbx # context->Rip<prologue label
- jb .Lin_prologue
-
- mov 152($context),%rax # pull context->Rsp
-
- lea .Lepilogue(%rip),%r10
- cmp %r10,%rbx # context->Rip>=epilogue label
- jae .Lin_prologue
-
- lea 24(%rax),%rax
-
- mov -8(%rax),%rbx
- mov -16(%rax),%r12
- mov -24(%rax),%r13
- mov %rbx,144($context) # restore context->Rbx
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R13
-
-.Lin_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
- jmp .Lcommon_seh_exit
-.size stream_se_handler,.-stream_se_handler
-
-.type key_se_handler,\@abi-omnipotent
-.align 16
-key_se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- sub \$64,%rsp
-
- mov 152($context),%rax # pull context->Rsp
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
-.Lcommon_seh_exit:
-
- mov 40($disp),%rdi # disp->ContextRecord
- mov $context,%rsi # context
- mov \$154,%ecx # sizeof(CONTEXT)
- .long 0xa548f3fc # cld; rep movsq
-
- mov $disp,%rsi
- xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
- mov 8(%rsi),%rdx # arg2, disp->ImageBase
- mov 0(%rsi),%r8 # arg3, disp->ControlPc
- mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
- mov 40(%rsi),%r10 # disp->ContextRecord
- lea 56(%rsi),%r11 # &disp->HandlerData
- lea 24(%rsi),%r12 # &disp->EstablisherFrame
- mov %r10,32(%rsp) # arg5
- mov %r11,40(%rsp) # arg6
- mov %r12,48(%rsp) # arg7
- mov %rcx,56(%rsp) # arg8, (NULL)
- call *__imp_RtlVirtualUnwind(%rip)
-
- mov \$1,%eax # ExceptionContinueSearch
- add \$64,%rsp
- popfq
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
- pop %rdi
- pop %rsi
- ret
-.size key_se_handler,.-key_se_handler
-
-.section .pdata
-.align 4
- .rva .LSEH_begin_RC4
- .rva .LSEH_end_RC4
- .rva .LSEH_info_RC4
-
- .rva .LSEH_begin_private_RC4_set_key
- .rva .LSEH_end_private_RC4_set_key
- .rva .LSEH_info_private_RC4_set_key
-
-.section .xdata
-.align 8
-.LSEH_info_RC4:
- .byte 9,0,0,0
- .rva stream_se_handler
-.LSEH_info_private_RC4_set_key:
- .byte 9,0,0,0
- .rva key_se_handler
-___
-}
-
sub reg_part {
my ($reg,$conv)=@_;
if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
diff --git a/lib/libcrypto/sha/asm/sha512-x86_64.pl b/lib/libcrypto/sha/asm/sha512-x86_64.pl
index b6252d31eca..feb0f9e7767 100755
--- a/lib/libcrypto/sha/asm/sha512-x86_64.pl
+++ b/lib/libcrypto/sha/asm/sha512-x86_64.pl
@@ -40,14 +40,17 @@
# sha256_block:-( This is presumably because 64-bit shifts/rotates
# apparently are not atomic instructions, but implemented in microcode.
-$output=shift;
+$flavour = shift;
+$output = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
if ($output =~ /512/) {
$func="sha512_block_data_order";
@@ -91,50 +94,44 @@ sub ROUND_00_15()
{ my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
$code.=<<___;
- mov $e,$a0
- mov $e,$a1
+ ror \$`$Sigma1[2]-$Sigma1[1]`,$a0
mov $f,$a2
+ mov $T1,`$SZ*($i&0xf)`(%rsp)
- ror \$$Sigma1[0],$a0
- ror \$$Sigma1[1],$a1
+ ror \$`$Sigma0[2]-$Sigma0[1]`,$a1
+ xor $e,$a0
xor $g,$a2 # f^g
- xor $a1,$a0
- ror \$`$Sigma1[2]-$Sigma1[1]`,$a1
+ ror \$`$Sigma1[1]-$Sigma1[0]`,$a0
+ add $h,$T1 # T1+=h
+ xor $a,$a1
+
+ add ($Tbl,$round,$SZ),$T1 # T1+=K[round]
and $e,$a2 # (f^g)&e
- mov $T1,`$SZ*($i&0xf)`(%rsp)
+ mov $b,$h
- xor $a1,$a0 # Sigma1(e)
+ ror \$`$Sigma0[1]-$Sigma0[0]`,$a1
+ xor $e,$a0
xor $g,$a2 # Ch(e,f,g)=((f^g)&e)^g
- add $h,$T1 # T1+=h
-
- mov $a,$h
- add $a0,$T1 # T1+=Sigma1(e)
+ xor $c,$h # b^c
+ xor $a,$a1
add $a2,$T1 # T1+=Ch(e,f,g)
- mov $a,$a0
- mov $a,$a1
+ mov $b,$a2
- ror \$$Sigma0[0],$h
- ror \$$Sigma0[1],$a0
- mov $a,$a2
- add ($Tbl,$round,$SZ),$T1 # T1+=K[round]
+ ror \$$Sigma1[0],$a0 # Sigma1(e)
+ and $a,$h # h=(b^c)&a
+ and $c,$a2 # b&c
- xor $a0,$h
- ror \$`$Sigma0[2]-$Sigma0[1]`,$a0
- or $c,$a1 # a|c
+ ror \$$Sigma0[0],$a1 # Sigma0(a)
+ add $a0,$T1 # T1+=Sigma1(e)
+ add $a2,$h # h+=b&c (completes +=Maj(a,b,c)
- xor $a0,$h # h=Sigma0(a)
- and $c,$a2 # a&c
add $T1,$d # d+=T1
-
- and $b,$a1 # (a|c)&b
add $T1,$h # h+=T1
-
- or $a2,$a1 # Maj(a,b,c)=((a|c)&b)|(a&c)
lea 1($round),$round # round++
+ add $a1,$h # h+=Sigma0(a)
- add $a1,$h # h+=Maj(a,b,c)
___
}
@@ -143,32 +140,30 @@ sub ROUND_16_XX()
$code.=<<___;
mov `$SZ*(($i+1)&0xf)`(%rsp),$a0
- mov `$SZ*(($i+14)&0xf)`(%rsp),$T1
-
- mov $a0,$a2
+ mov `$SZ*(($i+14)&0xf)`(%rsp),$a1
+ mov $a0,$T1
+ mov $a1,$a2
+ ror \$`$sigma0[1]-$sigma0[0]`,$T1
+ xor $a0,$T1
shr \$$sigma0[2],$a0
- ror \$$sigma0[0],$a2
-
- xor $a2,$a0
- ror \$`$sigma0[1]-$sigma0[0]`,$a2
-
- xor $a2,$a0 # sigma0(X[(i+1)&0xf])
- mov $T1,$a1
- shr \$$sigma1[2],$T1
- ror \$$sigma1[0],$a1
+ ror \$$sigma0[0],$T1
+ xor $T1,$a0 # sigma0(X[(i+1)&0xf])
+ mov `$SZ*(($i+9)&0xf)`(%rsp),$T1
- xor $a1,$T1
- ror \$`$sigma1[1]-$sigma1[0]`,$a1
-
- xor $a1,$T1 # sigma1(X[(i+14)&0xf])
+ ror \$`$sigma1[1]-$sigma1[0]`,$a2
+ xor $a1,$a2
+ shr \$$sigma1[2],$a1
+ ror \$$sigma1[0],$a2
add $a0,$T1
-
- add `$SZ*(($i+9)&0xf)`(%rsp),$T1
+ xor $a2,$a1 # sigma1(X[(i+14)&0xf])
add `$SZ*($i&0xf)`(%rsp),$T1
+ mov $e,$a0
+ add $a1,$T1
+ mov $a,$a1
___
&ROUND_00_15(@_);
}
@@ -186,7 +181,7 @@ $func:
push %r13
push %r14
push %r15
- mov %rsp,%rbp # copy %rsp
+ mov %rsp,%r11 # copy %rsp
shl \$4,%rdx # num*16
sub \$$framesz,%rsp
lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ
@@ -194,10 +189,10 @@ $func:
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %rbp,$_rsp # save copy of %rsp
+ mov %r11,$_rsp # save copy of %rsp
+.Lprologue:
- .picmeup $Tbl
- lea $TABLE-.($Tbl),$Tbl
+ lea $TABLE(%rip),$Tbl
mov $SZ*0($ctx),$A
mov $SZ*1($ctx),$B
@@ -215,6 +210,8 @@ $func:
___
for($i=0;$i<16;$i++) {
$code.=" mov $SZ*$i($inp),$T1\n";
+ $code.=" mov @ROT[4],$a0\n";
+ $code.=" mov @ROT[0],$a1\n";
$code.=" bswap $T1\n";
&ROUND_00_15($i,@ROT);
unshift(@ROT,pop(@ROT));
@@ -257,14 +254,15 @@ $code.=<<___;
mov $H,$SZ*7($ctx)
jb .Lloop
- mov $_rsp,%rsp
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
-
+ mov $_rsp,%rsi
+ mov (%rsi),%r15
+ mov 8(%rsi),%r14
+ mov 16(%rsi),%r13
+ mov 24(%rsi),%r12
+ mov 32(%rsi),%rbp
+ mov 40(%rsi),%rbx
+ lea 48(%rsi),%rsp
+.Lepilogue:
ret
.size $func,.-$func
___
diff --git a/lib/libcrypto/whrlpool/asm/wp-x86_64.pl b/lib/libcrypto/whrlpool/asm/wp-x86_64.pl
index 87c0843dc1d..afadd5d2f15 100644
--- a/lib/libcrypto/whrlpool/asm/wp-x86_64.pl
+++ b/lib/libcrypto/whrlpool/asm/wp-x86_64.pl
@@ -34,14 +34,13 @@ $flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
-$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
-
$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $flavour $output";
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
sub L() { $code.=".byte ".join(',',@_)."\n"; }
sub LL(){ $code.=".byte ".join(',',@_).",".join(',',@_)."\n"; }
@@ -477,113 +476,6 @@ ___
&L(0xfb,0xee,0x7c,0x66,0xdd,0x17,0x47,0x9e);
&L(0xca,0x2d,0xbf,0x07,0xad,0x5a,0x83,0x33);
-# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
-# CONTEXT *context,DISPATCHER_CONTEXT *disp)
-if ($win64) {
-$rec="%rcx";
-$frame="%rdx";
-$context="%r8";
-$disp="%r9";
-
-$code.=<<___;
-.extern __imp_RtlVirtualUnwind
-.type se_handler,\@abi-omnipotent
-.align 16
-se_handler:
- push %rsi
- push %rdi
- push %rbx
- push %rbp
- push %r12
- push %r13
- push %r14
- push %r15
- pushfq
- sub \$64,%rsp
-
- mov 120($context),%rax # pull context->Rax
- mov 248($context),%rbx # pull context->Rip
-
- lea .Lprologue(%rip),%r10
- cmp %r10,%rbx # context->Rip<.Lprologue
- jb .Lin_prologue
-
- mov 152($context),%rax # pull context->Rsp
-
- lea .Lepilogue(%rip),%r10
- cmp %r10,%rbx # context->Rip>=.Lepilogue
- jae .Lin_prologue
-
- mov 128+32(%rax),%rax # pull saved stack pointer
- lea 48(%rax),%rax
-
- mov -8(%rax),%rbx
- mov -16(%rax),%rbp
- mov -24(%rax),%r12
- mov -32(%rax),%r13
- mov -40(%rax),%r14
- mov -48(%rax),%r15
- mov %rbx,144($context) # restore context->Rbx
- mov %rbp,160($context) # restore context->Rbp
- mov %r12,216($context) # restore context->R12
- mov %r13,224($context) # restore context->R13
- mov %r14,232($context) # restore context->R14
- mov %r15,240($context) # restore context->R15
-
-.Lin_prologue:
- mov 8(%rax),%rdi
- mov 16(%rax),%rsi
- mov %rax,152($context) # restore context->Rsp
- mov %rsi,168($context) # restore context->Rsi
- mov %rdi,176($context) # restore context->Rdi
-
- mov 40($disp),%rdi # disp->ContextRecord
- mov $context,%rsi # context
- mov \$154,%ecx # sizeof(CONTEXT)
- .long 0xa548f3fc # cld; rep movsq
-
- mov $disp,%rsi
- xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
- mov 8(%rsi),%rdx # arg2, disp->ImageBase
- mov 0(%rsi),%r8 # arg3, disp->ControlPc
- mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
- mov 40(%rsi),%r10 # disp->ContextRecord
- lea 56(%rsi),%r11 # &disp->HandlerData
- lea 24(%rsi),%r12 # &disp->EstablisherFrame
- mov %r10,32(%rsp) # arg5
- mov %r11,40(%rsp) # arg6
- mov %r12,48(%rsp) # arg7
- mov %rcx,56(%rsp) # arg8, (NULL)
- call *__imp_RtlVirtualUnwind(%rip)
-
- mov \$1,%eax # ExceptionContinueSearch
- add \$64,%rsp
- popfq
- pop %r15
- pop %r14
- pop %r13
- pop %r12
- pop %rbp
- pop %rbx
- pop %rdi
- pop %rsi
- ret
-.size se_handler,.-se_handler
-
-.section .pdata
-.align 4
- .rva .LSEH_begin_$func
- .rva .LSEH_end_$func
- .rva .LSEH_info_$func
-
-.section .xdata
-.align 8
-.LSEH_info_$func:
- .byte 9,0,0,0
- .rva se_handler
-___
-}
-
$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-22 20:13:54 +0000