my $bits1="THIRTY_TWO_BIT ";
my $bits2="SIXTY_FOUR_BIT ";
-my $x86_elf_asm="x86cpuid-elf.o:bn86-elf.o co86-elf.o:dx86-elf.o yx86-elf.o:ax86-elf.o:bx86-elf.o:mx86-elf.o:sx86-elf.o s512sse2-elf.o:cx86-elf.o:rx86-elf.o:rm86-elf.o:r586-elf.o";
-my $x86_coff_asm="x86cpuid-cof.o:bn86-cof.o co86-cof.o:dx86-cof.o yx86-cof.o:ax86-cof.o:bx86-cof.o:mx86-cof.o:sx86-cof.o s512sse2-cof.o:cx86-cof.o:rx86-cof.o:rm86-cof.o:r586-cof.o";
-my $x86_out_asm="x86cpuid-out.o:bn86-out.o co86-out.o:dx86-out.o yx86-out.o:ax86-out.o:bx86-out.o:mx86-out.o:sx86-out.o s512sse2-out.o:cx86-out.o:rx86-out.o:rm86-out.o:r586-out.o";
+my $x86_elf_asm="x86cpuid-elf.o:bn86-elf.o co86-elf.o mo86-elf.o:dx86-elf.o yx86-elf.o:ax86-elf.o:bx86-elf.o:mx86-elf.o:sx86-elf.o s512sse2-elf.o:cx86-elf.o:rx86-elf.o:rm86-elf.o:r586-elf.o";
+my $x86_coff_asm="x86cpuid-cof.o:bn86-cof.o co86-cof.o mo86-cof.o:dx86-cof.o yx86-cof.o:ax86-cof.o:bx86-cof.o:mx86-cof.o:sx86-cof.o s512sse2-cof.o:cx86-cof.o:rx86-cof.o:rm86-cof.o:r586-cof.o";
+my $x86_out_asm="x86cpuid-out.o:bn86-out.o co86-out.o mo86-out.o:dx86-out.o yx86-out.o:ax86-out.o:bx86-out.o:mx86-out.o:sx86-out.o s512sse2-out.o:cx86-out.o:rx86-out.o:rm86-out.o:r586-out.o";
my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o::";
my $ia64_asm="ia64cpuid.o:bn-ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o::";
$cflags.=" -DOPENSSL_BN_ASM_PART_WORDS" if ($bn_obj =~ /bn86/);
$cflags.=" -DOPENSSL_IA32_SSE2" if (!$no_sse2 && $bn_obj =~ /bn86/);
-$cflags.=" -DOPENSSL_BN_ASM_MONT" if ($bn_obj =~ /\-mont/);
+$cflags.=" -DOPENSSL_BN_ASM_MONT" if ($bn_obj =~ /\-mont|mo86\-/);
$des_obj=$des_enc unless ($des_obj =~ /\.o$/);
$bf_obj=$bf_enc unless ($bf_obj =~ /\.o$/);
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-out.o
-$bn_obj = bn86-out.o co86-out.o
+$bn_obj = bn86-out.o co86-out.o mo86-out.o
$des_obj = dx86-out.o yx86-out.o
$aes_obj = ax86-out.o
$bf_obj = bx86-out.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-cof.o
-$bn_obj = bn86-cof.o co86-cof.o
+$bn_obj = bn86-cof.o co86-cof.o mo86-cof.o
$des_obj = dx86-cof.o yx86-cof.o
$aes_obj = ax86-cof.o
$bf_obj = bx86-cof.o
$lflags = -L/dev/env/WATT_ROOT/lib -lwatt
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-out.o
-$bn_obj = bn86-out.o co86-out.o
+$bn_obj = bn86-out.o co86-out.o mo86-out.o
$des_obj = dx86-out.o yx86-out.o
$aes_obj = ax86-out.o
$bf_obj = bx86-out.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lefence -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -rdynamic -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lnsl -lsocket
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags =
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-out.o
-$bn_obj = bn86-out.o co86-out.o
+$bn_obj = bn86-out.o co86-out.o mo86-out.o
$des_obj = dx86-out.o yx86-out.o
$aes_obj = ax86-out.o
$bf_obj = bx86-out.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lwsock32 -lgdi32
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT EXPORT_VAR_AS_FN
$cpuid_obj = x86cpuid-cof.o
-$bn_obj = bn86-cof.o co86-cof.o
+$bn_obj = bn86-cof.o co86-cof.o mo86-cof.o
$des_obj = dx86-cof.o yx86-cof.o
$aes_obj = ax86-cof.o
$bf_obj = bx86-cof.o
$lflags = -lsocket -lnsl
$bn_ops = DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lsocket -lnsl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lsocket -lnsl -ldl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lsocket -lnsl
$bn_ops = BN_LLONG MD2_CHAR RC4_INDEX DES_PTR DES_RISC1 DES_UNROLL
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
$lflags = -lsocket -lnsl
$bn_ops = BN_LLONG DES_PTR DES_RISC1 DES_UNROLL RC4_INDEX MD2_INT
$cpuid_obj = x86cpuid-elf.o
-$bn_obj = bn86-elf.o co86-elf.o
+$bn_obj = bn86-elf.o co86-elf.o mo86-elf.o
$des_obj = dx86-elf.o yx86-elf.o
$aes_obj = ax86-elf.o
$bf_obj = bx86-elf.o
(cd asm; $(PERL) bn-586.pl elf $(CFLAGS) > ../$@)
co86-elf.s: asm/co-586.pl ../perlasm/x86asm.pl
(cd asm; $(PERL) co-586.pl elf $(CFLAGS) > ../$@)
+mo86-elf.s: asm/x86-mont.pl ../perlasm/x86asm.pl
+ (cd asm; $(PERL) x86-mont.pl elf $(CFLAGS) > ../$@)
# COFF
bn86-cof.s: asm/bn-586.pl ../perlasm/x86asm.pl
(cd asm; $(PERL) bn-586.pl coff $(CFLAGS) > ../$@)
co86-cof.s: asm/co-586.pl ../perlasm/x86asm.pl
(cd asm; $(PERL) co-586.pl coff $(CFLAGS) > ../$@)
+mo86-cof.s: asm/x86-mont.pl ../perlasm/x86asm.pl
+ (cd asm; $(PERL) x86-mont.pl coff $(CFLAGS) > ../$@)
# a.out
bn86-out.s: asm/bn-586.pl ../perlasm/x86asm.pl
(cd asm; $(PERL) bn-586.pl a.out $(CFLAGS) > ../$@)
co86-out.s: asm/co-586.pl ../perlasm/x86asm.pl
(cd asm; $(PERL) co-586.pl a.out $(CFLAGS) > ../$@)
+mo86-out.s: asm/x86-mont.pl ../perlasm/x86asm.pl
+ (cd asm; $(PERL) x86-mont.pl a.out $(CFLAGS) > ../$@)
sparcv8.o: asm/sparcv8.S
$(CC) $(CFLAGS) -c asm/sparcv8.S
--- /dev/null
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. Rights for redistribution and usage in source and binary
+# forms are granted according to the OpenSSL license.
+# ====================================================================
+
+# This is a "teaser" code, as it can be improved in several ways...
+# First of all non-SSE2 path should be implemented (yes, for now it
+# performs Montgomery multiplication/convolution only on SSE2-capable
+# CPUs such as P4, others fall down to original code). Then inner loop
+# can be unrolled and modulo-scheduled to improve ILP and possibly
+# moved to 128-bit XMM register bank (though it would require input
+# rearrangement and/or increase bus bandwidth utilization). Dedicated
+# squaring procedure should give further performance improvement...
+# Yet, for being draft, the code improves rsa512 *sign* benchmark by
+# 110%(!), rsa1024 one - by 70% and rsa4096 - by 20%:-)
+
+push(@INC,"perlasm","../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],$0);
+
+$sse2=0;
+for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
+
+&external_label("OPENSSL_ia32cap_P") if ($sse2);
+
+&function_begin("bn_mul_mont",$sse2?"EXTRN\t_OPENSSL_ia32cap_P:DWORD":"");
+
+$i="ebx";
+$j="ecx";
+$ap="esi";
+$rp="edi"; $bp="edi"; # overlapping variables!!!
+$np="edx";
+$num="ebp";
+$tp="esp";
+
+$bias=2; # amount of extra words in tp
+ # (rounded up to even value)
+$_rp=&DWP(4*($bias+0),"esp",$num,4); # stack frame layout below tp
+$_ap=&DWP(4*($bias+1),"esp",$num,4);
+$_bp=&DWP(4*($bias+2),"esp",$num,4);
+$_np=&DWP(4*($bias+3),"esp",$num,4);
+$_n0=&DWP(4*($bias+4),"esp",$num,4);
+$_sp=&DWP(4*($bias+5),"esp",$num,4);
+
+$acc0="mm0"; # mmx register bank layout
+$acc1="mm1";
+$car0="mm2";
+$car1="mm3";
+$mul0="mm4";
+$mul1="mm5";
+$temp="mm6";
+$mask="mm7";
+
+if($sse2) {
+ &picmeup("eax","OPENSSL_ia32cap_P");
+ &bt (&DWP(0,"eax"),26);
+ &mov ("eax",0); # zero signals "we did nothing"
+ &jnc (&label("non_sse2"));
+
+ ################################# load argument block...
+ &mov ("eax",&wparam(0)); # BN_ULONG *rp
+ &mov ("ebx",&wparam(1)); # const BN_ULONG *ap
+ &mov ("ecx",&wparam(2)); # const BN_ULONG *bp
+ &mov ("edx",&wparam(3)); # const BN_ULONG *np
+ &mov ("esi",&wparam(4)); # BN_ULONG n0
+ &mov ($num,&wparam(5)); # int num
+
+ &mov ("edi","esp"); # saved stack pointer!
+ &add ($num,$bias+6);
+ &neg ($num);
+ &lea ("esp",&DWP(0,"esp",$num,4)); # alloca(4*(num+$bias+6))
+ &neg ($num);
+ &and ("esp",-1024); # minimize TLB utilization
+ &sub ($num,$bias+6); # num is restored to its original value
+ # and will remain constant from now...
+
+ &mov ($_rp,"eax"); # ... save a copy of argument block
+ &mov ($_ap,"ebx");
+ &mov ($_bp,"ecx");
+ &mov ($_np,"edx");
+ &mov ($_n0,"esi");
+ &mov ($_sp,"edi"); # saved stack pointer!
+
+ &mov ("eax",-1);
+ &movd ($mask,"eax"); # mask 32 lower bits
+
+ &mov ($ap,$_ap); # load input pointers
+ &mov ($bp,$_bp);
+ &mov ($np,$_np);
+
+ &xor ($i,$i); # i=0
+ &xor ($j,$j); # j=0
+
+ &movd ($mul0,&DWP(0,$bp)); # bp[0]
+ &movd ($mul1,&DWP(0,$ap)); # ap[0]
+ &movd ($car1,&DWP(0,$np)); # np[0]
+
+ &pmuludq($mul1,$mul0); # ap[0]*bp[0]
+ &movq ($car0,$mul1);
+ &movq ($acc0,$mul1); # I wish movd worked for
+ &pand ($acc0,$mask); # inter-register transfers
+
+ &pmuludq($mul1,$_n0); # *=n0
+
+ &pmuludq($car1,$mul1); # "t[0]"*np[0]*n0
+ &paddq ($car1,$acc0);
+
+ &psrlq ($car0,32);
+ &psrlq ($car1,32);
+
+ &inc ($j); # j++
+&set_label("1st");
+ &movd ($acc0,&DWP(0,$ap,$j,4)); # ap[j]
+ &movd ($acc1,&DWP(0,$np,$j,4)); # np[j]
+ &pmuludq($acc0,$mul0); # ap[j]*bp[0]
+ &pmuludq($acc1,$mul1); # np[j]*m1
+
+ &paddq ($car0,$acc0); # +=c0
+ &movq ($acc0,$car0);
+ &pand ($acc0,$mask);
+
+ &paddq ($car1,$acc1); # +=c1
+ &paddq ($car1,$acc0); # +=ap[j]*bp[0];
+ &movd (&DWP(-4,"esp",$j,4),$car1); # tp[j-1]=
+
+ &psrlq ($car0,32);
+ &psrlq ($car1,32);
+
+ &lea ($j,&DWP(1,$j));
+ &cmp ($j,$num);
+ &jl (&label("1st"));
+
+ &paddq ($car1,$car0);
+ &movq (&DWP(-4,"esp",$num,4),$car1);
+
+ &inc ($i); # i++
+&set_label("outer");
+ &xor ($j,$j); # j=0
+
+ &movd ($mul0,&DWP(0,$bp,$i,4)); # bp[i]
+ &movd ($mul1,&DWP(0,$ap)); # ap[0]
+ &movd ($temp,&DWP(0,"esp")); # tp[0]
+ &movd ($car1,&DWP(0,$np,$j,4)); # np[0]
+ &pmuludq($mul1,$mul0); # ap[0]*bp[i]
+
+ &paddq ($mul1,$temp); # +=tp[0]
+ &movq ($acc0,$mul1);
+ &movq ($car0,$mul1);
+ &pand ($acc0,$mask);
+
+ &pmuludq($mul1,$_n0); # *=n0
+
+ &pmuludq($car1,$mul1);
+ &paddq ($car1,$acc0);
+
+ &psrlq ($car0,32);
+ &psrlq ($car1,32);
+
+ &inc ($j); # j++
+&set_label("inner");
+ &movd ($acc0,&DWP(0,$ap,$j,4)); # ap[j]
+ &movd ($acc1,&DWP(0,$np,$j,4)); # np[j]
+ &movd ($temp,&DWP(0,"esp",$j,4)); # tp[j]
+ &pmuludq($acc0,$mul0); # ap[j]*bp[i]
+ &pmuludq($acc1,$mul1); # np[j]*m1
+ &paddq ($car0,$temp); # +=tp[j]
+ &paddq ($car0,$acc0); # +=c0
+ &movq ($acc0,$car0);
+ &pand ($acc0,$mask);
+
+ &paddq ($car1,$acc1); # +=c1
+ &paddq ($car1,$acc0); # +=ap[j]*bp[i]+tp[j]
+ &movd (&DWP(-4,"esp",$j,4),$car1); # tp[j-1]
+
+ &psrlq ($car0,32);
+ &psrlq ($car1,32);
+
+ &lea ($j,&DWP(1,$j)); # j++
+ &cmp ($j,$num);
+ &jl (&label("inner"));
+
+ &movd ($temp,&DWP(0,"esp",$num,4));
+ &paddq ($car1,$car0);
+ &paddq ($car1,$temp);
+ &movq (&DWP(-4,"esp",$num,4),$car1);
+
+ &lea ($i,&DWP(1,$i)); # i++
+ &cmp ($i,$num);
+ &jl (&label("outer"));
+
+ &emms (); # done with mmx bank
+
+ &mov ("esi",&DWP(0,"esp",$num,4)); # load upmost overflow bit
+ &mov ($rp,$_rp); # load result pointer
+ # [$ap and $bp are zapped]
+ &xor ($i,$i); # i=0
+ &lea ($j,&DWP(-1,$num)); # j=num-1
+ &cmp ("esi",0); # clears CF unconditionally
+ &jnz (&label("sub"));
+ &mov ("eax",&DWP(0,"esp",$j,4));
+ &cmp ("eax",&DWP(0,$np,$j,4)); # tp[num-1]-np[num-1]?
+ &jae (&label("sub")); # if taken CF is cleared
+&set_label("copy");
+ &mov ("eax",&DWP(0,"esp",$j,4));
+ &mov (&DWP(0,$rp,$j,4),"eax"); # rp[i]=tp[i]
+ &mov (&DWP(0,"esp",$j,4),$j); # zap temporary vector
+ &dec ($j);
+ &jge (&label("copy"));
+ &jmp (&label("exit_sse2"));
+
+&set_label("sub",4);
+ &mov ("eax",&DWP(0,"esp",$i,4));
+ &sbb ("eax",&DWP(0,$np,$i,4));
+ &mov (&DWP(0,$rp,$i,4),"eax"); # rp[i]=tp[i]-np[i]
+ &lea ($i,&DWP(1,$i)); # i++
+ &dec ($j); # doesn't affect CF!
+ &jge (&label("sub"));
+ &lea ($j,&DWP(-1,$num)); # j=num-1
+ &sbb ("esi",0); # esi holds upmost overflow bit
+ &jc (&label("copy"));
+&set_label("zap");
+ &mov (&DWP(0,"esp",$j,4),$i); # zap temporary vector
+ &dec ($j);
+ &jge (&label("zap"));
+
+&set_label("exit_sse2");
+ &mov ("esp",$_sp); # pull saved stack pointer
+ &mov ("eax",1);
+&set_label("non_sse2");
+}
+
+&function_end("bn_mul_mont");
+
+&asm_finish();
# SSE2
sub main'emms { &out0("emms"); }
sub main'movd { &out2("movd",@_); }
-sub main'movq { &out2("movq",@_); }
sub main'movdqu { &out2("movdqu",@_); }
sub main'movdqa { &out2("movdqa",@_); }
sub main'movdq2q{ &out2("movdq2q",@_); }
sub main'pxor { &out2("pxor",@_); }
sub main'por { &out2("por",@_); }
sub main'pand { &out2("pand",@_); }
+sub main'movq {
+ local($p1,$p2)=@_;
+ if ($p1=~/^mm[0-7]$/ && $p2=~/^mm[0-7]$/)
+ # movq between mmx registers sinks Intel CPUs
+ { push(@out,"\tpshufw\t\$0xe4,%$p2,%$p1\n"); }
+ else { &out2("movq",@_); }
+ }
# The bswapl instruction is new for the 486. Emulate if i386.
sub main'bswap
projects / openssl.git / commitdiff