[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openssl-cvs
Subject:    [CVS] OpenSSL: openssl/crypto/des/asm/ des-586.pl
From:       "Andy Polyakov" <appro () openssl ! org>
Date:       2005-10-25 15:47:10
Message-ID: 20051025154710.7933D1AC6572 () master ! openssl ! org
[Download RAW message or body]

  OpenSSL CVS Repository
  http://cvs.openssl.org/
  ____________________________________________________________________________

  Server: cvs.openssl.org                  Name:   Andy Polyakov
  Root:   /v/openssl/cvs                   Email:  appro@openssl.org
  Module: openssl                          Date:   25-Oct-2005 17:47:10
  Branch: HEAD                             Handle: 2005102516470900

  Modified files:
    openssl/crypto/des/asm  des-586.pl

  Log:
    Eliminate ~3.5KB of duplicate code in des-586.pl and reserve for folded
    loop option, which can give further 3KB code reduction.

  Summary:
    Revision    Changes     Path
    1.9         +89 -32     openssl/crypto/des/asm/des-586.pl
  ____________________________________________________________________________

  patch -p0 <<'@@ .'
  Index: openssl/crypto/des/asm/des-586.pl
  ============================================================================
  $ cvs diff -u -r1.8 -r1.9 des-586.pl
  --- openssl/crypto/des/asm/des-586.pl	25 Oct 2005 15:43:18 -0000	1.8
  +++ openssl/crypto/des/asm/des-586.pl	25 Oct 2005 15:47:09 -0000	1.9
  @@ -18,8 +18,14 @@
   
   $L="edi";
   $R="esi";
  +$trans="ebp";
  +$small_footprint=1 if (grep(/\-DOPENSSL_SMALL_FOOTPRINT/,@ARGV));
  +# one can discuss setting this variable to 1 unconditionally, as
  +# the folded loop is only 3% slower than unrolled, but >7 times smaller
   
   &external_label("DES_SPtrans");
  +&DES_encrypt_internal();
  +&DES_decrypt_internal();
   &DES_encrypt("DES_encrypt1",1);
   &DES_encrypt("DES_encrypt2",0);
   &DES_encrypt3("DES_encrypt3",1);
  @@ -30,6 +36,80 @@
   
   &asm_finish();
   
  +sub DES_encrypt_internal()
  +	{
  +	&function_begin_B("_x86_DES_encrypt");
  +
  +	if ($small_footprint)
  +	    {
  +	    &lea("edx",&DWP(128,"ecx"));
  +	    &push("edx");
  +	    &push("ecx");
  +	    &set_label("eloop");
  +		&D_ENCRYPT(0,$L,$R,0,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("");
  +		&D_ENCRYPT(1,$R,$L,2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("");
  +		&add("ecx",16);
  +		&cmp("ecx",&swtmp(1));
  +		&mov(&swtmp(0),"ecx");
  +		&jb(&label("eloop"));
  +	    &add("esp",8);
  +	    }
  +	else
  +	    {
  +	    &push("ecx");
  +	    for ($i=0; $i<16; $i+=2)
  +		{
  +		&comment("Round $i");
  +		&D_ENCRYPT($i,$L,$R,$i*2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("Round ".sprintf("%d",$i+1));
  +		&D_ENCRYPT($i+1,$R,$L,($i+1)*2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		}
  +	    &add("esp",4);
  +	}
  +	&ret();
  +
  +	&function_end_B("_x86_DES_encrypt");
  +	}
  +	
  +sub DES_decrypt_internal()
  +	{
  +	&function_begin_B("_x86_DES_decrypt");
  +
  +	if ($small_footprint)
  +	    {
  +	    &push("ecx");
  +	    &lea("ecx",&DWP(128,"ecx"));
  +	    &push("ecx");
  +	    &set_label("dloop");
  +		&D_ENCRYPT(0,$L,$R,-2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("");
  +		&D_ENCRYPT(1,$R,$L,-4,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("");
  +		&sub("ecx",16);
  +		&cmp("ecx",&swtmp(1));
  +		&mov(&swtmp(0),"ecx");
  +		&ja(&label("dloop"));
  +	    &add("esp",8);
  +	    }
  +	else
  +	    {
  +	    &push("ecx");
  +	    for ($i=15; $i>0; $i-=2)
  +		{
  +		&comment("Round $i");
  +		&D_ENCRYPT(15-$i,$L,$R,$i*2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		&comment("Round ".sprintf("%d",$i-1));
  +		&D_ENCRYPT(15-$i+1,$R,$L,($i-1)*2,$trans,"eax","ebx","ecx","edx",&swtmp(0));
  +		}
  +	    &add("esp",4);
  +	    }
  +	&ret();
  +
  +	&function_end_B("_x86_DES_decrypt");
  +	}
  +	
   sub DES_encrypt
   	{
   	local($name,$do_ip)=@_;
  @@ -41,7 +121,6 @@
   
   	&comment("");
   	&comment("Load the 2 words");
  -	$trans="ebp";
   
   	if ($do_ip)
   		{
  @@ -75,38 +154,16 @@
   
   	# PIC-ification:-)
   	&picmeup($trans,"DES_SPtrans");
  -	#if ($cpp)	{ &picmeup($trans,"DES_SPtrans");   }
  -	#else		{ &lea($trans,&DWP("DES_SPtrans")); }
   
   	&mov(	"ecx",	&wparam(1)	);
  -	&cmp("ebx","0");
  -	&je(&label("start_decrypt"));
  -
  -	for ($i=0; $i<16; $i+=2)
  -		{
  -		&comment("");
  -		&comment("Round $i");
  -		&D_ENCRYPT($i,$L,$R,$i*2,$trans,"eax","ebx","ecx","edx");
  -
  -		&comment("");
  -		&comment("Round ".sprintf("%d",$i+1));
  -		&D_ENCRYPT($i+1,$R,$L,($i+1)*2,$trans,"eax","ebx","ecx","edx");
  -		}
  -	&jmp(&label("end"));
  -
  -	&set_label("start_decrypt");
   
  -	for ($i=15; $i>0; $i-=2)
  -		{
  -		&comment("");
  -		&comment("Round $i");
  -		&D_ENCRYPT(15-$i,$L,$R,$i*2,$trans,"eax","ebx","ecx","edx");
  -		&comment("");
  -		&comment("Round ".sprintf("%d",$i-1));
  -		&D_ENCRYPT(15-$i+1,$R,$L,($i-1)*2,$trans,"eax","ebx","ecx","edx");
  -		}
  -
  -	&set_label("end");
  +	&cmp("ebx","0");
  +	&je(&label("decrypt"));
  +	&call("_x86_DES_encrypt");
  +	&jmp(&label("done"));
  +	&set_label("decrypt");
  +	&call("_x86_DES_decrypt");
  +	&set_label("done");
   
   	if ($do_ip)
   		{
  @@ -140,7 +197,7 @@
   
   sub D_ENCRYPT
   	{
  -	local($r,$L,$R,$S,$trans,$u,$tmp1,$tmp2,$t)=@_;
  +	local($r,$L,$R,$S,$trans,$u,$tmp1,$tmp2,$t,$wp1)=@_;
   
   	 &mov(	$u,		&DWP(&n2a($S*4),$tmp2,"",0));
   	&xor(	$tmp1,		$tmp1);
  @@ -167,7 +224,7 @@
   	&and(	$t,		"0xff"	);
   	 &xor(	$L,		&DWP("0x600",$trans,$tmp1,0));
   	 &xor(	$L,		&DWP("0x700",$trans,$tmp2,0));
  -	&mov(	$tmp2,		&wparam(1)	);
  +	&mov(	$tmp2,		$wp1	);
   	 &xor(	$L,		&DWP("0x400",$trans,$u,0));
   	 &xor(	$L,		&DWP("0x500",$trans,$t,0));
   	}
  @@ .
______________________________________________________________________
OpenSSL Project                                 http://www.openssl.org
CVS Repository Commit List                     openssl-cvs@openssl.org
Automated List Manager                           majordomo@openssl.org

[prev in list] [next in list] [prev in thread] [next in thread]