mirror/openssl
2
0
Fork 0
mirror of https://github.com/openssl/openssl.git synced 2024-11-27 05:21:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

rc4-586.pl: optimize unused code path.

Browse Source
This commit is contained in:
Andy Polyakov 2011-05-25 09:36:13 +00:00
parent 67d8487bb8
commit 6a99984b57
1 changed files with 37 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
crypto/rc4/asm/rc4-586.pl
View File

@ -30,8 +30,8 @@
# May 2011 # May 2011
# #
# Optimize for Core2 and Westmere [and Opteron]. Current performance # Optimize for Core2 and Westmere [and incidentally Opteron]. Current
# in cycles per processed byte (less is better) is: # performance in cycles per processed byte (less is better) is:
# #
# Pentium 10.2 # original numbers # Pentium 10.2 # original numbers
# Pentium III 7.8(*) # Pentium III 7.8(*)
@ -43,11 +43,12 @@
# Sandy Bridge 5.4/0% # Sandy Bridge 5.4/0%
# #
# (*) PIII can actually deliver 6.6 cycles per byte with MMX code, # (*) PIII can actually deliver 6.6 cycles per byte with MMX code,
# but this specific code performs poorly on Core2. While below # but this specific code performs poorly on Core2. And vice
# MMX code delivering 5.8 on Core2 performs at 8.0 on PIII:-( # versa, below MMX/SSE code delivering 5.8/7.1 on Core2 performs
# As PIII is not a "hot" CPU [anymore], I chose not to introduce # poorly on PIII, at 8.0/14.5:-( As PIII is not a "hot" CPU
# PIII-specific code path, which is why MMX code path is quarded # [anymore], I chose to discard PIII-specific code path and opt
# by SSE2 bit (see below), not MMX. # for original IALU-only code, which is why MMX/SSE code path
# is guarded by SSE2 bit (see below), not MMX/SSE.
# (**) Performance vs. block size on Core2 and Westmere had a maximum # (**) Performance vs. block size on Core2 and Westmere had a maximum
# at ... 64 bytes block size. And it was quite a maximum, 40-60% # at ... 64 bytes block size. And it was quite a maximum, 40-60%
# in comparison to largest 8KB block size. Above improvement # in comparison to largest 8KB block size. Above improvement
@ -88,31 +89,40 @@ sub RC4_loop {
} }
if ($alt=0) { if ($alt=0) {
# works ~5% faster on Atom and ~20% slower on Core2 # >20% faster on Atom and Sandy Bridge[!], 8% faster on Opteron,
# but ~40% slower on Core2 and Westmere... Attempt to add movz
# brings down Opteron by 25%, Atom and Sandy Bridge by 15%, yet
# on Core2 with movz it's almost 20% slower than below alternative
# code... Yes, it's a total mess...
my @XX=($xx,$out); my @XX=($xx,$out);
$RC4_loop_mmx = sub { $RC4_loop_mmx = sub { # SSE actually...
my $i=shift; my $i=shift;
my $j=$i<=0?0:$i>>1;
my $mm=$i<=0?"mm0":"mm".($i&1);
&add ($yy,$tx); &add (&LB($yy),&LB($tx));
&movz ($yy,&LB($yy));
&lea (@XX[1],&DWP(1,@XX[0])); &lea (@XX[1],&DWP(1,@XX[0]));
&psllq ("mm1",8*(($i-1)&7)) if (abs($i)!=1); &pxor ("mm2","mm0") if ($i==0);
&psllq ("mm1",8) if ($i==0);
&and (@XX[1],0xff); &and (@XX[1],0xff);
&pxor ("mm0","mm0") if ($i<=0);
&mov ($ty,&DWP(0,$dat,$yy,4)); &mov ($ty,&DWP(0,$dat,$yy,4));
&mov (&DWP(0,$dat,$yy,4),$tx); &mov (&DWP(0,$dat,$yy,4),$tx);
&pxor ("mm1","mm2") if ($i==0);
&mov (&DWP(0,$dat,$XX[0],4),$ty); &mov (&DWP(0,$dat,$XX[0],4),$ty);
&add ($ty,$tx); &add (&LB($ty),&LB($tx));
&movd (@XX[0],"mm7") if ($i==0); &movd (@XX[0],"mm7") if ($i==0);
&movz ($ty,&LB($ty));
&mov ($tx,&DWP(0,$dat,@XX[1],4)); &mov ($tx,&DWP(0,$dat,@XX[1],4));
&pxor ("mm2",$i==1?"mm0":"mm1") if ($i>=0); &pxor ("mm1","mm1") if ($i==1);
&movq ("mm0",&QWP(0,$inp)) if ($i<=0); &movq ("mm2",&QWP(0,$inp)) if ($i==1);
&movq (&QWP(-8,(@XX[0],$inp)),"mm2") if ($i==0); &movq (&QWP(-8,(@XX[0],$inp)),"mm1") if ($i==0);
&movd ($i>0?"mm1":"mm2",&DWP(0,$dat,$ty,4)); &pinsrw ($mm,&DWP(0,$dat,$ty,4),$j);
push (@XX,shift(@XX)) if ($i>=0); push (@XX,shift(@XX)) if ($i>=0);
} }
} else { } else {
# Using pinsrw here improves performane on Intel CPUs by 2-3%, but
# brings down AMD by 7%...
$RC4_loop_mmx = sub { $RC4_loop_mmx = sub {
my $i=shift; my $i=shift;
@ -185,8 +195,6 @@ if ($alt=0) {
&movd ("mm7",&wparam(3)) if ($alt); &movd ("mm7",&wparam(3)) if ($alt);
&and ($ty,-8); &and ($ty,-8);
&lea ($ty,&DWP(-8,$inp,$ty)); &lea ($ty,&DWP(-8,$inp,$ty));
&mov (&wparam(2),$ty);
&mov (&DWP(-4,$dat),$ty); # save input+(len/8)*8-8 &mov (&DWP(-4,$dat),$ty); # save input+(len/8)*8-8
&$RC4_loop_mmx(-1); &$RC4_loop_mmx(-1);
@ -200,10 +208,17 @@ if ($alt=0) {
&lea ($inp,&DWP(8,$inp)); &lea ($inp,&DWP(8,$inp));
&jb (&label("loop_mmx")); &jb (&label("loop_mmx"));
&movd ($out,"mm7") if ($alt); if ($alt) {
&movd ($out,"mm7");
&pxor ("mm2","mm0");
&psllq ("mm1",8);
&pxor ("mm1","mm2");
&movq (&QWP(-8,$out,$inp),"mm1");
} else {
&psllq ("mm1",56); &psllq ("mm1",56);
&pxor ("mm2","mm1"); &pxor ("mm2","mm1");
&movq (&QWP(-8,$out,$inp),"mm2"); &movq (&QWP(-8,$out,$inp),"mm2");
}
&emms (); &emms ();
&cmp ($inp,&wparam(1)); # compare to input+len &cmp ($inp,&wparam(1)); # compare to input+len
@ -267,7 +282,7 @@ if ($alt=0) {
&set_label("done"); &set_label("done");
&dec (&LB($xx)); &dec (&LB($xx));
&mov (&BP(-4,$dat),&LB($yy)); # save key->y &mov (&DWP(-4,$dat),$yy); # save key->y
&mov (&BP(-8,$dat),&LB($xx)); # save key->x &mov (&BP(-8,$dat),&LB($xx)); # save key->x
&set_label("abort"); &set_label("abort");
&function_end("RC4"); &function_end("RC4");