#
# SHA256 block transform for x86. September 2007.
#
+# Performance improvement over compiler generated code varies from
+# 10% to 40% [see below]. Not very impressive on some µ-archs, but
+# it's 5 times smaller and optimizies amount of writes.
+#
+# May 2012.
+#
+# Optimization including one of Pavel Semjanov's ideas resulted in
+# ~5% improvement on P4, AMD and Sandy Bridge, and ~13% on Atom.
+# Pavel also suggested full unroll. While his code runs ~20%/13%/6%
+# faster on K8/Core2/Sandy Bridge, it's 9.6x larger and ~6%/18%/24%
+# slower on P4/Atom/Pentium...
+#
# Performance in clock cycles per processed byte (less is better):
#
-# Pentium PIII P4 AMD K8 Core2
+# Pentium PIII P4 AMD K8 Core2 SB(**) Atom
# 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 asm 39 29 31 19 18 19(**) 31
+# x86_64 asm(*) - - 21 16 16 18 25
#
-# (*) 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.
+# (*) x86_64 assembler performance is presented for reference
+# purposes.
+# (**) Sandy Bridge results can be improved by ~20% by replacing
+# ror with equivalent shrd.
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
$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");
+$Aoff=&DWP(4,"esp");
+$Boff=&DWP(8,"esp");
+$Coff=&DWP(12,"esp");
+$Doff=&DWP(16,"esp");
+$Eoff=&DWP(20,"esp");
+$Foff=&DWP(24,"esp");
+$Goff=&DWP(28,"esp");
+$Hoff=&DWP(32,"esp");
+$Xoff=&DWP(36,"esp");
$K256="ebp";
+# *ror = sub { &shrd(@_[0],@_) };
+
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",$Foff);
+ &ror ("ecx",25-11);
+ &add ($T,"edi") if ($in_16_63); # T += sigma1(X[-2])
&xor ("ecx",$E);
+ &mov ("edi",$Goff);
&ror ("ecx",11-6);
- &mov (&DWP(4*(8+15),"esp"),$T) if ($in_16_63); # save X[0]
+ &xor ("esi","edi");
&xor ("ecx",$E);
+ &mov (&DWP(4*(9+15),"esp"),$T) if ($in_16_63); # save X[0]
&ror ("ecx",6); # Sigma1(e)
- &mov ("edi",$Goff);
+ &and ("esi",$E);
&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 ($T,"esi"); # T += Ch(e,f,g)
+ &add ($T,$Hoff); # T += h
+ &mov ("ecx",$A);
+ &mov ($E,$Doff); # e becomes d, which is e in next iteration
&ror ("ecx",22-13);
- &add ($T,$Hoff); # T += h
+ &add ($T,"esi"); # T += Ch(e,f,g)
&xor ("ecx",$A);
+ &mov ("esi",&DWP(0,$K256));
&ror ("ecx",13-2);
- &mov ("esi",$Boff);
+ &mov ("edi",$Boff);
&xor ("ecx",$A);
+ &mov ($Aoff,$A); # modulo-scheduled
&ror ("ecx",2); # Sigma0(a)
- &add ($E,$T); # d += T
- &mov ("edi",$Coff);
+ &xor ($A,"edi"); # a ^= b
+ &add ($T,"esi");
+ &push ($A); # (b^c) in next round
+ &add ($E,$T); # d += T
+ &and ($A,&DWP(4,"esp")); # a &= (b^c)
&add ($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 ($A,"ecx"); # h=Maj(a,b,c)
+ &xor ($A,"edi"); # h = Maj(a,b,c) = Ch(a^b,c,b)
&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 ($A,"esi"); # h += K256[i]
+ &add ($A,$T); # h += T
+ &mov ($T,&DWP(4*(9+15+16-1),"esp")) if ($in_16_63); # preload T
}
&function_begin("sha256_block_data_order");
&mov ("eax",&DWP($i*16+0,"edi"));
&mov ("ebx",&DWP($i*16+4,"edi"));
&mov ("ecx",&DWP($i*16+8,"edi"));
- &mov ("edx",&DWP($i*16+12,"edi"));
&bswap ("eax");
+ &mov ("edx",&DWP($i*16+12,"edi"));
&bswap ("ebx");
- &bswap ("ecx");
- &bswap ("edx");
&push ("eax");
+ &bswap ("ecx");
&push ("ebx");
+ &bswap ("edx");
&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");
+ &lea ("esp",&DWP(-4*9,"esp"));# place for A,B,C,D,E,F,G,H
+ &mov (&DWP(4*(9+16)+4,"esp"),"edi");
# copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
&mov ($A,&DWP(0,"esi"));
&mov ("edi",&DWP(12,"esi"));
# &mov ($Aoff,$A);
&mov ($Boff,"ebx");
+ &xor ("ebx","ecx");
&mov ($Coff,"ecx");
&mov ($Doff,"edi");
+ &mov (&DWP(0,"esp"),"ebx"); # magic
&mov ($E,&DWP(16,"esi"));
&mov ("ebx",&DWP(20,"esi"));
&mov ("ecx",&DWP(24,"esi"));
&mov ($Hoff,"edi");
&set_label("00_15",16);
- &mov ($T,&DWP(4*(8+15),"esp"));
+ &mov ($T,&DWP(4*(9+15),"esp"));
&BODY_00_15();
&cmp ("esi",0xc19bf174);
&jne (&label("00_15"));
- &mov ($T,&DWP(4*(8+15+16-1),"esp")); # preloaded in BODY_00_15(1)
+ &mov ($T,&DWP(4*(9+15+16-1),"esp")); # preloaded in BODY_00_15(1)
&set_label("16_63",16);
&mov ("esi",$T);
- &mov ("ecx",&DWP(4*(8+15+16-14),"esp"));
+ &mov ("ecx",&DWP(4*(9+15+16-14),"esp"));
&ror ("esi",18-7);
&mov ("edi","ecx");
&xor ("esi",$T);
+ &shr ($T,3);
&ror ("esi",7);
- &shr ($T,3);
&ror ("edi",19-17);
&xor ($T,"esi"); # T = sigma0(X[-15])
&xor ("edi","ecx");
+ &shr ("ecx",10);
&ror ("edi",17);
- &shr ("ecx",10);
- &add ($T,&DWP(4*(8+15+16),"esp")); # T += X[-16]
+ &add ($T,&DWP(4*(9+15+16),"esp")); # T += X[-16]
&xor ("edi","ecx"); # sigma1(X[-2])
- &add ($T,&DWP(4*(8+15+16-9),"esp")); # T += X[-7]
+ &add ($T,&DWP(4*(9+15+16-9),"esp")); # T += X[-7]
# &add ($T,"edi"); # T += sigma1(X[-2])
- # &mov (&DWP(4*(8+15),"esp"),$T); # save X[0]
+ # &mov (&DWP(4*(9+15),"esp"),$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 ("esi",&DWP(4*(9+16+64)+0,"esp"));#ctx
# &mov ($A,$Aoff);
&mov ("ebx",$Boff);
&mov ("ecx",$Coff);
&mov ("eax",$Foff);
&mov ("ebx",$Goff);
&mov ("ecx",$Hoff);
- &mov ("edi",&DWP(4*(8+16+64)+4,"esp"));#inp
+ &mov ("edi",&DWP(4*(9+16+64)+4,"esp"));#inp
&add ($E,&DWP(16,"esi"));
&add ("eax",&DWP(20,"esi"));
&add ("ebx",&DWP(24,"esi"));
&mov (&DWP(24,"esi"),"ebx");
&mov (&DWP(28,"esi"),"ecx");
- &add ("esp",4*(8+16+64)); # destroy frame
+ &lea ("esp",&DWP(4*(9+16+64),"esp"));# destroy frame
&sub ($K256,4*64); # rewind K
&cmp ("edi",&DWP(8,"esp")); # are we done yet?
projects / openssl.git / commitdiff