[thirdparty/openssl.git] / crypto / sha / asm / sha256-586.pl

#!/usr/bin/env perl
#
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> 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/.
# ====================================================================
#
# SHA256 block transform for x86. September 2007.
#
# Performance in clock cycles per processed byte (less is better):
#
#		Pentium	PIII	P4	AMD K8	Core2
# gcc		46	36	41	27	26
# icc		57	33	38	25	23	
# x86 asm	40	30	33	20	18
# x86_64 asm(*)	-	-	21	16	16
#
# (*) x86_64 assembler performance is presented for reference
#     purposes.
#
# Performance improvement over compiler generated code varies from
# 10% to 40% [see above]. Not very impressive on some µ-archs, but
# it's 5 times smaller and optimizies amount of writes.

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";

&asm_init($ARGV[0],"sha512-586.pl",$ARGV[$#ARGV] eq "386");

$A="eax";
$E="edx";
$T="ebx";
$Aoff=&DWP(0,"esp");
$Boff=&DWP(4,"esp");
$Coff=&DWP(8,"esp");
$Doff=&DWP(12,"esp");
$Eoff=&DWP(16,"esp");
$Foff=&DWP(20,"esp");
$Goff=&DWP(24,"esp");
$Hoff=&DWP(28,"esp");
$Xoff=&DWP(32,"esp");
$K256="ebp";

sub BODY_00_15() {
    my $in_16_63=shift;

	&mov	("ecx",$E);
	 &add	($T,"edi")			if ($in_16_63);	# T += sigma1(X[-2])
	&ror	("ecx",25-11);
	 &mov	("esi",$Foff);
	&xor	("ecx",$E);
	&ror	("ecx",11-6);
	 &mov	(&DWP(4*(8+15),"esp"),$T)	if ($in_16_63);	# save X[0]
	&xor	("ecx",$E);
	&ror	("ecx",6);	# Sigma1(e)
	 &mov	("edi",$Goff);
	&add	($T,"ecx");	# T += Sigma1(e)

	&xor	("esi","edi");
	 &mov	($Eoff,$E);	# modulo-scheduled
	 &mov	("ecx",$A);
	&and	("esi",$E);
	 &mov	($E,$Doff);	# e becomes d, which is e in next iteration
	&xor	("esi","edi");	# Ch(e,f,g)
	 &mov	("edi",$A);
	&add	($T,"esi");	# T += Ch(e,f,g)

	&ror	("ecx",22-13);
	 &add	($T,$Hoff);	# T += h
	&xor	("ecx",$A);
	&ror	("ecx",13-2);
	 &mov	("esi",$Boff);
	&xor	("ecx",$A);
	&ror	("ecx",2);	# Sigma0(a)
	 &add	($E,$T);	# d += T
	 &mov	("edi",$Coff);

	&add	($T,"ecx");	# T += Sigma0(a)
	 &mov	($Aoff,$A);	# modulo-scheduled

	&mov	("ecx",$A);
	 &sub	("esp",4);
	&or	($A,"esi");	# a becomes h, which is a in next iteration
	&and	("ecx","esi");
	&and	($A,"edi");
	 &mov	("esi",&DWP(0,$K256));
	&or	($A,"ecx");	# h=Maj(a,b,c)

	&add	($K256,4);
	&add	($A,$T);	# h += T
	 &mov	($T,&DWP(4*(8+15+16-1),"esp"))	if ($in_16_63);	# preload T
	&add	($E,"esi");	# d += K256[i]
	&add	($A,"esi");	# h += K256[i]
}

&function_begin("sha256_block_data_order");
	&mov	("esi",wparam(0));	# ctx
	&mov	("edi",wparam(1));	# inp
	&mov	("eax",wparam(2));	# num
	&mov	("ebx","esp");		# saved sp

	&call	(&label("pic_point"));	# make it PIC!
&set_label("pic_point");
	&blindpop($K256);
	&lea	($K256,&DWP(&label("K256")."-".&label("pic_point"),$K256));

	&sub	("esp",16);
	&and	("esp",-64);

	&shl	("eax",6);
	&add	("eax","edi");
	&mov	(&DWP(0,"esp"),"esi");	# ctx
	&mov	(&DWP(4,"esp"),"edi");	# inp
	&mov	(&DWP(8,"esp"),"eax");	# inp+num*128
	&mov	(&DWP(12,"esp"),"ebx");	# saved sp

&set_label("loop",16);
    # copy input block to stack reversing byte and dword order
    for($i=0;$i<4;$i++) {
	&mov	("eax",&DWP($i*16+0,"edi"));
	&mov	("ebx",&DWP($i*16+4,"edi"));
	&mov	("ecx",&DWP($i*16+8,"edi"));
	&mov	("edx",&DWP($i*16+12,"edi"));
	&bswap	("eax");
	&bswap	("ebx");
	&bswap	("ecx");
	&bswap	("edx");
	&push	("eax");
	&push	("ebx");
	&push	("ecx");
	&push	("edx");
    }
	&add	("edi",64);
	&sub	("esp",4*8);		# place for A,B,C,D,E,F,G,H
	&mov	(&DWP(4*(8+16)+4,"esp"),"edi");

	# copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
	&mov	($A,&DWP(0,"esi"));
	&mov	("ebx",&DWP(4,"esi"));
	&mov	("ecx",&DWP(8,"esi"));
	&mov	("edi",&DWP(12,"esi"));
	# &mov	($Aoff,$A);
	&mov	($Boff,"ebx");
	&mov	($Coff,"ecx");
	&mov	($Doff,"edi");
	&mov	($E,&DWP(16,"esi"));	
	&mov	("ebx",&DWP(20,"esi"));
	&mov	("ecx",&DWP(24,"esi"));
	&mov	("edi",&DWP(28,"esi"));
	# &mov	($Eoff,$E);
	&mov	($Foff,"ebx");
	&mov	($Goff,"ecx");
	&mov	($Hoff,"edi");

&set_label("00_15",16);
	&mov	($T,&DWP(4*(8+15),"esp"));

	&BODY_00_15();

	&cmp	("esi",0xc19bf174);
	&jne	(&label("00_15"));

	&mov	($T,&DWP(4*(8+15+16-1),"esp"));	# preloaded in BODY_00_15(1)
&set_label("16_63",16);
	&mov	("esi",$T);
	 &mov	("ecx",&DWP(4*(8+15+16-14),"esp"));
	&ror	("esi",18-7);
	 &mov	("edi","ecx");
	&xor	("esi",$T);
	&ror	("esi",7);
	&shr	($T,3);

	&ror	("edi",19-17);
	 &xor	($T,"esi");			# T = sigma0(X[-15])
	&xor	("edi","ecx");
	&ror	("edi",17);
	&shr	("ecx",10);
	 &add	($T,&DWP(4*(8+15+16),"esp"));	# T += X[-16]
	&xor	("edi","ecx");			# sigma1(X[-2])

	 &add	($T,&DWP(4*(8+15+16-9),"esp"));	# T += X[-7]
	# &add	($T,"edi");			# T += sigma1(X[-2])
	# &mov	(&DWP(4*(8+15),"esp"),$T);	# save X[0]

	&BODY_00_15(1);

	&cmp	("esi",0xc67178f2);
	&jne	(&label("16_63"));

	&mov	("esi",&DWP(4*(8+16+64)+0,"esp"));#ctx
	# &mov	($A,$Aoff);
	&mov	("ebx",$Boff);
	&mov	("ecx",$Coff);
	&mov	("edi",$Doff);
	&add	($A,&DWP(0,"esi"));
	&add	("ebx",&DWP(4,"esi"));
	&add	("ecx",&DWP(8,"esi"));
	&add	("edi",&DWP(12,"esi"));
	&mov	(&DWP(0,"esi"),$A);
	&mov	(&DWP(4,"esi"),"ebx");
	&mov	(&DWP(8,"esi"),"ecx");
	&mov	(&DWP(12,"esi"),"edi");
	# &mov	($E,$Eoff);
	&mov	("eax",$Foff);
	&mov	("ebx",$Goff);
	&mov	("ecx",$Hoff);
	&mov	("edi",&DWP(4*(8+16+64)+4,"esp"));#inp
	&add	($E,&DWP(16,"esi"));
	&add	("eax",&DWP(20,"esi"));
	&add	("ebx",&DWP(24,"esi"));
	&add	("ecx",&DWP(28,"esi"));
	&mov	(&DWP(16,"esi"),$E);
	&mov	(&DWP(20,"esi"),"eax");
	&mov	(&DWP(24,"esi"),"ebx");
	&mov	(&DWP(28,"esi"),"ecx");

	&add	("esp",4*(8+16+64));		# destroy frame
	&sub	($K256,4*64);			# rewind K

	&cmp	("edi",&DWP(8,"esp"));		# are we done yet?
	&jb	(&label("loop"));

	&mov	("esp",&DWP(12,"esp"));		# restore sp
&function_end_A();

&set_label("K256",64);	# Yes! I keep it in the code segment!
	&data_word(0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5);
	&data_word(0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5);
	&data_word(0xd807aa98,0x12835b01,0x243185be,0x550c7dc3);
	&data_word(0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174);
	&data_word(0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc);
	&data_word(0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da);
	&data_word(0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7);
	&data_word(0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967);
	&data_word(0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13);
	&data_word(0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85);
	&data_word(0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3);
	&data_word(0xd192e819,0xd6990624,0xf40e3585,0x106aa070);
	&data_word(0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5);
	&data_word(0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3);
	&data_word(0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208);
	&data_word(0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2);
&function_end_B("sha256_block_data_order");
&asciz("SHA256 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>");

&asm_finish();
Commit	Line	Data
ee0449b1 AP	1	#!/usr/bin/env perl
	2	#
	3	# ====================================================================
	4	# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
	5	# project. The module is, however, dual licensed under OpenSSL and
	6	# CRYPTOGAMS licenses depending on where you obtain it. For further
	7	# details see http://www.openssl.org/~appro/cryptogams/.
	8	# ====================================================================
	9	#
	10	# SHA256 block transform for x86. September 2007.
	11	#
	12	# Performance in clock cycles per processed byte (less is better):
	13	#
	14	# Pentium PIII P4 AMD K8 Core2
	15	# gcc 46 36 41 27 26
	16	# icc 57 33 38 25 23
2357ae17 AP	17	# x86 asm 40 30 33 20 18
2357ae17 AP	18	# x86_64 asm(*) - - 21 16 16
ee0449b1 AP	19	#
	20	# (*) x86_64 assembler performance is presented for reference
	21	# purposes.
	22	#
	23	# Performance improvement over compiler generated code varies from
a027bba2	24	# 10% to 40% [see above]. Not very impressive on some µ-archs, but
ee0449b1 AP	25	# it's 5 times smaller and optimizies amount of writes.
	26
	27	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
	28	push(@INC,"${dir}","${dir}../../perlasm");
	29	require "x86asm.pl";
	30
	31	&asm_init($ARGV[0],"sha512-586.pl",$ARGV[$#ARGV] eq "386");
	32
	33	$A="eax";
	34	$E="edx";
	35	$T="ebx";
	36	$Aoff=&DWP(0,"esp");
	37	$Boff=&DWP(4,"esp");
	38	$Coff=&DWP(8,"esp");
	39	$Doff=&DWP(12,"esp");
	40	$Eoff=&DWP(16,"esp");
	41	$Foff=&DWP(20,"esp");
	42	$Goff=&DWP(24,"esp");
	43	$Hoff=&DWP(28,"esp");
	44	$Xoff=&DWP(32,"esp");
	45	$K256="ebp";
	46
	47	sub BODY_00_15() {
b5e5760d	48	my $in_16_63=shift;
8dc899de	49
ee0449b1	50	&mov ("ecx",$E);
2357ae17 AP	51	&add ($T,"edi") if ($in_16_63); # T += sigma1(X[-2])
2357ae17 AP	52	&ror ("ecx",25-11);
ee0449b1	53	&mov ("esi",$Foff);
2357ae17 AP	54	&xor ("ecx",$E);
2357ae17 AP	55	&ror ("ecx",11-6);
b5e5760d	56	&mov (&DWP(4*(8+15),"esp"),$T) if ($in_16_63); # save X[0]
2357ae17 AP	57	&xor ("ecx",$E);
2357ae17 AP	58	&ror ("ecx",6); # Sigma1(e)
ee0449b1 AP	59	&mov ("edi",$Goff);
ee0449b1 AP	60	&add ($T,"ecx"); # T += Sigma1(e)
ee0449b1 AP	61
ee0449b1 AP	62	&xor ("esi","edi");
2357ae17	63	&mov ($Eoff,$E); # modulo-scheduled
ee0449b1 AP	64	&mov ("ecx",$A);
	65	&and ("esi",$E);
	66	&mov ($E,$Doff); # e becomes d, which is e in next iteration
	67	&xor ("esi","edi"); # Ch(e,f,g)
	68	&mov ("edi",$A);
	69	&add ($T,"esi"); # T += Ch(e,f,g)
	70
2357ae17	71	&ror ("ecx",22-13);
ee0449b1	72	&add ($T,$Hoff); # T += h
2357ae17 AP	73	&xor ("ecx",$A);
2357ae17 AP	74	&ror ("ecx",13-2);
ee0449b1	75	&mov ("esi",$Boff);
2357ae17 AP	76	&xor ("ecx",$A);
2357ae17 AP	77	&ror ("ecx",2); # Sigma0(a)
ee0449b1	78	&add ($E,$T); # d += T
ee0449b1 AP	79	&mov ("edi",$Coff);
	80
	81	&add ($T,"ecx"); # T += Sigma0(a)
	82	&mov ($Aoff,$A); # modulo-scheduled
	83
	84	&mov ("ecx",$A);
	85	&sub ("esp",4);
	86	&or ($A,"esi"); # a becomes h, which is a in next iteration
	87	&and ("ecx","esi");
	88	&and ($A,"edi");
	89	&mov ("esi",&DWP(0,$K256));
	90	&or ($A,"ecx"); # h=Maj(a,b,c)
	91
	92	&add ($K256,4);
	93	&add ($A,$T); # h += T
b5e5760d	94	&mov ($T,&DWP(4*(8+15+16-1),"esp")) if ($in_16_63); # preload T
ee0449b1 AP	95	&add ($E,"esi"); # d += K256[i]
	96	&add ($A,"esi"); # h += K256[i]
	97	}
	98
87facba3	99	&function_begin("sha256_block_data_order");
ee0449b1 AP	100	&mov ("esi",wparam(0)); # ctx
	101	&mov ("edi",wparam(1)); # inp
	102	&mov ("eax",wparam(2)); # num
	103	&mov ("ebx","esp"); # saved sp
	104
	105	&call (&label("pic_point")); # make it PIC!
	106	&set_label("pic_point");
	107	&blindpop($K256);
	108	&lea ($K256,&DWP(&label("K256")."-".&label("pic_point"),$K256));
	109
	110	&sub ("esp",16);
	111	&and ("esp",-64);
	112
	113	&shl ("eax",6);
	114	&add ("eax","edi");
	115	&mov (&DWP(0,"esp"),"esi"); # ctx
	116	&mov (&DWP(4,"esp"),"edi"); # inp
	117	&mov (&DWP(8,"esp"),"eax"); # inp+num*128
	118	&mov (&DWP(12,"esp"),"ebx"); # saved sp
	119
	120	&set_label("loop",16);
	121	# copy input block to stack reversing byte and dword order
	122	for($i=0;$i<4;$i++) {
	123	&mov ("eax",&DWP($i*16+0,"edi"));
	124	&mov ("ebx",&DWP($i*16+4,"edi"));
	125	&mov ("ecx",&DWP($i*16+8,"edi"));
	126	&mov ("edx",&DWP($i*16+12,"edi"));
	127	&bswap ("eax");
	128	&bswap ("ebx");
	129	&bswap ("ecx");
	130	&bswap ("edx");
	131	&push ("eax");
	132	&push ("ebx");
	133	&push ("ecx");
	134	&push ("edx");
	135	}
	136	&add ("edi",64);
	137	&sub ("esp",4*8); # place for A,B,C,D,E,F,G,H
	138	&mov (&DWP(4*(8+16)+4,"esp"),"edi");
	139
	140	# copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
	141	&mov ($A,&DWP(0,"esi"));
	142	&mov ("ebx",&DWP(4,"esi"));
	143	&mov ("ecx",&DWP(8,"esi"));
	144	&mov ("edi",&DWP(12,"esi"));
	145	# &mov ($Aoff,$A);
	146	&mov ($Boff,"ebx");
	147	&mov ($Coff,"ecx");
	148	&mov ($Doff,"edi");
	149	&mov ($E,&DWP(16,"esi"));
	150	&mov ("ebx",&DWP(20,"esi"));
	151	&mov ("ecx",&DWP(24,"esi"));
	152	&mov ("edi",&DWP(28,"esi"));
	153	# &mov ($Eoff,$E);
	154	&mov ($Foff,"ebx");
	155	&mov ($Goff,"ecx");
	156	&mov ($Hoff,"edi");
	157
	158	&set_label("00_15",16);
	159	&mov ($T,&DWP(4*(8+15),"esp"));
	160
	161	&BODY_00_15();
	162
	163	&cmp ("esi",0xc19bf174);
164	&jne (&label("00_15"));
165
8dc899de	166	&mov ($T,&DWP(4*(8+15+16-1),"esp")); # preloaded in BODY_00_15(1)
ee0449b1	167	&set_label("16_63",16);
ee0449b1	168	&mov ("esi",$T);
8dc899de	169	&mov ("ecx",&DWP(4*(8+15+16-14),"esp"));
ee0449b1 AP	170	&ror ("esi",18-7);
ee0449b1 AP	171	&mov ("edi","ecx");
2357ae17 AP	172	&xor ("esi",$T);
	173	&ror ("esi",7);
	174	&shr ($T,3);
ee0449b1	175
ee0449b1	176	&ror ("edi",19-17);
2357ae17 AP	177	&xor ($T,"esi"); # T = sigma0(X[-15])
	178	&xor ("edi","ecx");
	179	&ror ("edi",17);
	180	&shr ("ecx",10);
	181	&add ($T,&DWP(4*(8+15+16),"esp")); # T += X[-16]
	182	&xor ("edi","ecx"); # sigma1(X[-2])
ee0449b1	183
2357ae17 AP	184	&add ($T,&DWP(4*(8+15+16-9),"esp")); # T += X[-7]
2357ae17 AP	185	# &add ($T,"edi"); # T += sigma1(X[-2])
8dc899de	186	# &mov (&DWP(4*(8+15),"esp"),$T); # save X[0]
ee0449b1	187
8dc899de	188	&BODY_00_15(1);
ee0449b1 AP	189
	190	&cmp ("esi",0xc67178f2);
	191	&jne (&label("16_63"));
	192
	193	&mov ("esi",&DWP(4*(8+16+64)+0,"esp"));#ctx
	194	# &mov ($A,$Aoff);
	195	&mov ("ebx",$Boff);
	196	&mov ("ecx",$Coff);
	197	&mov ("edi",$Doff);
	198	&add ($A,&DWP(0,"esi"));
	199	&add ("ebx",&DWP(4,"esi"));
	200	&add ("ecx",&DWP(8,"esi"));
	201	&add ("edi",&DWP(12,"esi"));
	202	&mov (&DWP(0,"esi"),$A);
	203	&mov (&DWP(4,"esi"),"ebx");
	204	&mov (&DWP(8,"esi"),"ecx");
	205	&mov (&DWP(12,"esi"),"edi");
	206	# &mov ($E,$Eoff);
	207	&mov ("eax",$Foff);
	208	&mov ("ebx",$Goff);
	209	&mov ("ecx",$Hoff);
	210	&mov ("edi",&DWP(4*(8+16+64)+4,"esp"));#inp
	211	&add ($E,&DWP(16,"esi"));
	212	&add ("eax",&DWP(20,"esi"));
	213	&add ("ebx",&DWP(24,"esi"));
	214	&add ("ecx",&DWP(28,"esi"));
	215	&mov (&DWP(16,"esi"),$E);
	216	&mov (&DWP(20,"esi"),"eax");
	217	&mov (&DWP(24,"esi"),"ebx");
	218	&mov (&DWP(28,"esi"),"ecx");
	219
	220	&add ("esp",4*(8+16+64)); # destroy frame
	221	&sub ($K256,4*64); # rewind K
	222
	223	&cmp ("edi",&DWP(8,"esp")); # are we done yet?
	224	&jb (&label("loop"));
	225
	226	&mov ("esp",&DWP(12,"esp")); # restore sp
	227	&function_end_A();
	228
	229	&set_label("K256",64); # Yes! I keep it in the code segment!
	230	&data_word(0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5);
	231	&data_word(0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5);
	232	&data_word(0xd807aa98,0x12835b01,0x243185be,0x550c7dc3);
	233	&data_word(0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174);
	234	&data_word(0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc);
	235	&data_word(0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da);
	236	&data_word(0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7);
	237	&data_word(0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967);
	238	&data_word(0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13);
	239	&data_word(0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85);
	240	&data_word(0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3);
	241	&data_word(0xd192e819,0xd6990624,0xf40e3585,0x106aa070);
	242	&data_word(0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5);
	243	&data_word(0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3);
	244	&data_word(0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208);
	245	&data_word(0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2);
	246	&function_end_B("sha256_block_data_order");
	247	&asciz("SHA256 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>");
	248
	249	&asm_finish();