# non-MMX implementation would actually pay off, but till
# opposite is proved "unlikely" is assumed.
-push(@INC,"../../perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
&asm_init($ARGV[0],"wp-mmx.pl");
&mov (&DWP(12,"ebx"),"esi"); # zero round counter
&set_label("round",16);
- &movq (@mm[0],&DWP(2048*$SCALE,$tbl,"esi",8)); # rc[r]
+ &movq (@mm[0],&QWP(2048*$SCALE,$tbl,"esi",8)); # rc[r]
&mov ("eax",&DWP(0,"esp"));
&mov ("ebx",&DWP(4,"esp"));
+ &movz ("ecx",&LB("eax"));
+ &movz ("edx",&HB("eax"));
for($i=0;$i<8;$i++) {
- my $func = ($i==0)? movq : pxor;
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
+ my $func = ($i==0)? \&movq : \&pxor;
+ &shr ("eax",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
- &shr ("eax",16);
+ &movz ("edx",&HB("eax"));
&pxor (@mm[0],&QWP(&row(0),$tbl,"esi",8));
&$func (@mm[1],&QWP(&row(1),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
&mov ("eax",&DWP(($i+1)*8,"esp"));
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
+ &movz ("edx",&HB("ebx"));
&$func (@mm[2],&QWP(&row(2),$tbl,"esi",8));
&$func (@mm[3],&QWP(&row(3),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
+ &shr ("ebx",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
- &shr ("ebx",16);
+ &movz ("edx",&HB("ebx"));
&$func (@mm[4],&QWP(&row(4),$tbl,"esi",8));
&$func (@mm[5],&QWP(&row(5),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
&mov ("ebx",&DWP(($i+1)*8+4,"esp"));
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
+ &movz ("edx",&HB("eax"));
&$func (@mm[6],&QWP(&row(6),$tbl,"esi",8));
&$func (@mm[7],&QWP(&row(7),$tbl,"edi",8));
push(@mm,shift(@mm));
for($i=0;$i<8;$i++) { &movq(&QWP($i*8,"esp"),@mm[$i]); } # K=L
for($i=0;$i<8;$i++) {
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
+ &shr ("eax",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
- &shr ("eax",16);
+ &movz ("edx",&HB("eax"));
&pxor (@mm[0],&QWP(&row(0),$tbl,"esi",8));
&pxor (@mm[1],&QWP(&row(1),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
&mov ("eax",&DWP(64+($i+1)*8,"esp")) if ($i<7);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
+ &movz ("edx",&HB("ebx"));
&pxor (@mm[2],&QWP(&row(2),$tbl,"esi",8));
&pxor (@mm[3],&QWP(&row(3),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
+ &shr ("ebx",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
- &shr ("ebx",16);
+ &movz ("edx",&HB("ebx"));
&pxor (@mm[4],&QWP(&row(4),$tbl,"esi",8));
&pxor (@mm[5],&QWP(&row(5),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
&mov ("ebx",&DWP(64+($i+1)*8+4,"esp")) if ($i<7);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
+ &movz ("edx",&HB("eax"));
&pxor (@mm[6],&QWP(&row(6),$tbl,"esi",8));
&pxor (@mm[7],&QWP(&row(7),$tbl,"edi",8));
push(@mm,shift(@mm));
&L(0xfb,0xee,0x7c,0x66,0xdd,0x17,0x47,0x9e);
&L(0xca,0x2d,0xbf,0x07,0xad,0x5a,0x83,0x33);
-&function_end_B("whrilpool_block_mmx");
+&function_end_B("whirlpool_block_mmx");
&asm_finish();