# on P4, where it kills performance, nor Sandy Bridge, where folded
# loop is approximately as fast...
#
+# June 2012.
+#
+# Add AMD XOP-specific code path, >30% improvement on Bulldozer over
+# May version, >60% over original. Add AVX+shrd code path, >25%
+# improvement on Sandy Bridge over May version, 60% over original.
+#
# Performance in clock cycles per processed byte (less is better):
#
-# PIII P4 AMD K8 Core2 SB(**) Atom Bldzr
+# PIII P4 AMD K8 Core2 SB Atom Bldzr
# gcc 36 41 27 26 25 50 36
# icc 33 38 25 23 - - -
-# x86 asm(*) 27/24 28 19/15.5 18/15.6 16(**) 30/25 27/22
-# x86_64 asm(***) 17.5 15 15.5 17.5 23 21
+# x86 asm(*) 27/24 28 19/15.5 18/15.6 12.5 30/25 16.6
+# x86_64 asm(**) 17.5 15 15.5 17.5 23 21
#
# (*) numbers after slash are for unrolled loop, where available;
-# (**) for Sandy Bridge executing code path with ror replaced with
-# equivalent shrd;
-# (***) x86_64 assembly performance is presented for reference
+# (**) x86_64 assembly performance is presented for reference
# purposes.
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
&asm_init($ARGV[0],"sha512-586.pl",$ARGV[$#ARGV] eq "386");
+$xmm=$ymm=0;
+for (@ARGV) { $xmm=1 if (/-DOPENSSL_IA32_SSE2/); }
+
+$ymm=1 if ($xmm &&
+ `$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+ =~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
+ $1>=2.19); # first version supporting AVX
+
+$ymm=1 if ($xmm && !$ymm && $ARGV[0] eq "win32n" &&
+ `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
+ $1>=2.03); # first version supporting AVX
+
+$ymm=1 if ($xmm && !$ymm && $ARGV[0] eq "win32" &&
+ `ml 2>&1` =~ /Version ([0-9]+)\./ &&
+ $1>=10); # first version supporting AVX
+
$unroll_after = 64*4; # If pre-evicted from L1P cache first spin of
# fully unrolled loop was measured to run about
# 3-4x slower. If slowdown coefficient is N and
&mov ("edx",&DWP(4,"edx"));
&test ("ecx",1<<20); # check for P4
&jnz (&label("loop"));
+ &test ("edx",1<<11); # check for XOP
+ &jnz (&label("XOP"));
&and ("ecx",1<<30); # mask "Intel CPU" bit
&and ("edx",1<<28); # mask AVX bit
&or ("ecx","edx");
&cmp ("ecx",1<<28|1<<30);
- &je (&label("loop_shrd"));
+ &je (&label("AVX")) if ($ymm);
+ &je (&label("loop_shrd")) if (!$ymm);
if ($unroll_after) {
&sub ("eax","edi");
&cmp ("eax",$unroll_after);
&COMPACT_LOOP();
&mov ("esp",&DWP(12,"esp")); # restore sp
&function_end_A();
- if (!$i386) {
+ if (!$i386 && !$ymm) {
# ~20% improvement on Sandy Bridge
local *ror = sub { &shrd(@_[0],@_) };
&COMPACT_LOOP("_shrd");
&mov ("esp",&DWP(96+12,"esp")); # restore sp
&function_end_A();
+
+ if ($ymm) {{{
+my @X = map("xmm$_",(0..3));
+my ($t0,$t1,$t2,$t3) = map("xmm$_",(4..7));
+my @AH = ($A,$T);
+
+&set_label("XOP",16);
+ &lea ("esp",&DWP(-96,"esp"));
+ &vzeroall ();
+ # copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
+ &mov ($AH[0],&DWP(0,"esi"));
+ &mov ($AH[1],&DWP(4,"esi"));
+ &mov ("ecx",&DWP(8,"esi"));
+ &mov ("edi",&DWP(12,"esi"));
+ #&mov (&DWP(0,"esp"),$AH[0]);
+ &mov (&DWP(4,"esp"),$AH[1]);
+ &xor ($AH[1],"ecx"); # magic
+ &mov (&DWP(8,"esp"),"ecx");
+ &mov (&DWP(12,"esp"),"edi");
+ &mov ($E,&DWP(16,"esi"));
+ &mov ("edi",&DWP(20,"esi"));
+ &mov ("ecx",&DWP(24,"esi"));
+ &mov ("esi",&DWP(28,"esi"));
+ #&mov (&DWP(16,"esp"),$E);
+ &mov (&DWP(20,"esp"),"edi");
+ &mov ("edi",&DWP(96+4,"esp")); # inp
+ &mov (&DWP(24,"esp"),"ecx");
+ &mov (&DWP(28,"esp"),"esi");
+ &vmovdqa ($t3,&DWP(256,$K256));
+ &jmp (&label("grand_xop"));
+
+&set_label("grand_xop",16);
+ # load input, reverse byte order, add K256[0..15], save to stack
+ &vmovdqu (@X[0],&QWP(0,"edi"));
+ &vmovdqu (@X[1],&QWP(16,"edi"));
+ &vmovdqu (@X[2],&QWP(32,"edi"));
+ &vmovdqu (@X[3],&QWP(48,"edi"));
+ &add ("edi",64);
+ &vpshufb (@X[0],@X[0],$t3);
+ &mov (&DWP(96+4,"esp"),"edi");
+ &vpshufb (@X[1],@X[1],$t3);
+ &vpshufb (@X[2],@X[2],$t3);
+ &vpaddd ($t0,@X[0],&QWP(0,$K256));
+ &vpshufb (@X[3],@X[3],$t3);
+ &vpaddd ($t1,@X[1],&QWP(16,$K256));
+ &vpaddd ($t2,@X[2],&QWP(32,$K256));
+ &vpaddd ($t3,@X[3],&QWP(48,$K256));
+ &vmovdqa (&QWP(32+0,"esp"),$t0);
+ &vmovdqa (&QWP(32+16,"esp"),$t1);
+ &vmovdqa (&QWP(32+32,"esp"),$t2);
+ &vmovdqa (&QWP(32+48,"esp"),$t3);
+ &jmp (&label("xop_00_47"));
+
+&set_label("xop_00_47",16);
+ &add ($K256,64);
+
+sub XOP_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body); # 120 instructions
+
+ &vpalignr ($t0,@X[1],@X[0],4); # X[1..4]
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpalignr ($t3,@X[3],@X[2],4); # X[9..12]
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t1,$t0,14);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpsrld ($t0,$t0,3);
+ &vpaddd (@X[0],@X[0],$t3); # X[0..3] += X[9..12]
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t2,$t1,25-14);
+ &vpxor ($t0,$t0,$t1);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t3,@X[3],13);
+ &vpxor ($t0,$t0,$t2); # sigma0(X[1..4])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpsrld ($t2,@X[3],10);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpaddd (@X[0],@X[0],$t0); # X[0..3] += sigma0(X[1..4])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t1,$t3,15-13);
+ &vpxor ($t3,$t3,$t2);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpxor ($t3,$t3,$t1); # sigma1(X[14..15])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpsrldq ($t3,$t3,8);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpaddd (@X[0],@X[0],$t3); # X[0..1] += sigma1(X[14..15])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t3,@X[0],13);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpsrld ($t2,@X[0],10);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vprotd ($t1,$t3,15-13);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpxor ($t3,$t3,$t2);
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpxor ($t3,$t3,$t1); # sigma1(X[16..17])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpslldq ($t3,$t3,8); # 22 instructions
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpaddd (@X[0],@X[0],$t3); # X[2..3] += sigma1(X[16..17])
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ &vpaddd ($t2,@X[0],&QWP(16*$j,$K256));
+
+ foreach (@insns) { eval; } # remaining instructions
+
+ &vmovdqa (&QWP(32+16*$j,"esp"),$t2);
+}
+
+sub body_00_15 () {
+ (
+ '&mov ("ecx",$E);',
+ '&mov ("esi",&off($f));',
+ '&ror ($E,25-11);',
+ '&mov ("edi",&off($g));',
+ '&xor ($E,"ecx");',
+ '&xor ("esi","edi");',
+ '&ror ($E,11-6);',
+ '&and ("esi","ecx");',
+ '&mov (&off($e),"ecx");', # save $E, modulo-scheduled
+ '&xor ($E,"ecx");',
+ '&xor ("edi","esi");', # Ch(e,f,g)
+ '&ror ($E,6);', # T = Sigma1(e)
+ '&mov ("ecx",$AH[0]);',
+ '&mov ("esi",$AH[0]);',
+ '&add ($E,&off($h));', # T += h
+
+ '&ror ("ecx",22-13);',
+ '&add ($E,"edi");', # T += Ch(e,f,g)
+ '&mov ("edi",&off($b));',
+ '&xor ("ecx",$AH[0]);',
+ '&mov (&off($a),$AH[0]);', # save $A, modulo-scheduled
+ '&xor ($AH[0],"edi");', # a ^= b, (b^c) in next round
+ '&ror ("ecx",13-2);',
+ '&and ($AH[1],$AH[0]);', # (b^c) &= (a^b)
+ '&add ($E,&DWP(32+4*($i&15),"esp"));', # T += K[i]+X[i]
+ '&xor ("ecx","esi");',
+ '&xor ($AH[1],"edi");', # h = Maj(a,b,c) = Ch(a^b,c,b)
+ '&ror ("ecx",2);', # Sigma0(a)
+
+ '&add ($AH[1],$E);', # h += T
+ '&add ($E,&off($d));', # d += T
+ '&add ($AH[1],"ecx");'. # h += Sigma0(a)
+
+ '@AH = reverse(@AH); $i++;' # rotate(a,h)
+ );
+}
+
+ for ($i=0,$j=0; $j<4; $j++) {
+ &XOP_00_47($j,\&body_00_15,@X);
+ push(@X,shift(@X)); # rotate(@X)
+ }
+ &cmp (&DWP(16*$j,$K256),0x00010203);
+ &jne (&label("xop_00_47"));
+
+ for ($i=0; $i<16; ) {
+ foreach(body_00_15()) { eval; }
+ }
+
+ &mov ("esi",&DWP(96,"esp")); #ctx
+ #&mov ($AH[0],&DWP(0,"esp"));
+ &xor ($AH[1],"edi"); #&mov ($AH[1],&DWP(4,"esp"));
+ #&mov ("edi", &DWP(8,"esp"));
+ &mov ("ecx",&DWP(12,"esp"));
+ &add ($AH[0],&DWP(0,"esi"));
+ &add ($AH[1],&DWP(4,"esi"));
+ &add ("edi",&DWP(8,"esi"));
+ &add ("ecx",&DWP(12,"esi"));
+ &mov (&DWP(0,"esi"),$AH[0]);
+ &mov (&DWP(4,"esi"),$AH[1]);
+ &mov (&DWP(8,"esi"),"edi");
+ &mov (&DWP(12,"esi"),"ecx");
+ #&mov (&DWP(0,"esp"),$AH[0]);
+ &mov (&DWP(4,"esp"),$AH[1]);
+ &xor ($AH[1],"edi"); # magic
+ &mov (&DWP(8,"esp"),"edi");
+ &mov (&DWP(12,"esp"),"ecx");
+ #&mov ($E,&DWP(16,"esp"));
+ &mov ("edi",&DWP(20,"esp"));
+ &mov ("ecx",&DWP(24,"esp"));
+ &add ($E,&DWP(16,"esi"));
+ &add ("edi",&DWP(20,"esi"));
+ &add ("ecx",&DWP(24,"esi"));
+ &mov (&DWP(16,"esi"),$E);
+ &mov (&DWP(20,"esi"),"edi");
+ &mov (&DWP(20,"esp"),"edi");
+ &mov ("edi",&DWP(28,"esp"));
+ &mov (&DWP(24,"esi"),"ecx");
+ #&mov (&DWP(16,"esp"),$E);
+ &add ("edi",&DWP(28,"esi"));
+ &mov (&DWP(24,"esp"),"ecx");
+ &mov (&DWP(28,"esi"),"edi");
+ &mov (&DWP(28,"esp"),"edi");
+ &mov ("edi",&DWP(96+4,"esp")); # inp
+
+ &vmovdqa ($t3,&QWP(64,$K256));
+ &sub ($K256,3*64); # rewind K
+ &cmp ("edi",&DWP(96+8,"esp")); # are we done yet?
+ &jb (&label("grand_xop"));
+
+ &mov ("esp",&DWP(96+12,"esp")); # restore sp
+ &vzeroall ();
+&function_end_A();
+
+&set_label("AVX",16);
+ &lea ("esp",&DWP(-96,"esp"));
+ &vzeroall ();
+ # copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
+ &mov ($AH[0],&DWP(0,"esi"));
+ &mov ($AH[1],&DWP(4,"esi"));
+ &mov ("ecx",&DWP(8,"esi"));
+ &mov ("edi",&DWP(12,"esi"));
+ #&mov (&DWP(0,"esp"),$AH[0]);
+ &mov (&DWP(4,"esp"),$AH[1]);
+ &xor ($AH[1],"ecx"); # magic
+ &mov (&DWP(8,"esp"),"ecx");
+ &mov (&DWP(12,"esp"),"edi");
+ &mov ($E,&DWP(16,"esi"));
+ &mov ("edi",&DWP(20,"esi"));
+ &mov ("ecx",&DWP(24,"esi"));
+ &mov ("esi",&DWP(28,"esi"));
+ #&mov (&DWP(16,"esp"),$E);
+ &mov (&DWP(20,"esp"),"edi");
+ &mov ("edi",&DWP(96+4,"esp")); # inp
+ &mov (&DWP(24,"esp"),"ecx");
+ &mov (&DWP(28,"esp"),"esi");
+ &vmovdqa ($t3,&DWP(256,$K256));
+ &jmp (&label("grand_avx"));
+
+&set_label("grand_avx",16);
+ # load input, reverse byte order, add K256[0..15], save to stack
+ &vmovdqu (@X[0],&QWP(0,"edi"));
+ &vmovdqu (@X[1],&QWP(16,"edi"));
+ &vmovdqu (@X[2],&QWP(32,"edi"));
+ &vmovdqu (@X[3],&QWP(48,"edi"));
+ &add ("edi",64);
+ &vpshufb (@X[0],@X[0],$t3);
+ &mov (&DWP(96+4,"esp"),"edi");
+ &vpshufb (@X[1],@X[1],$t3);
+ &vpshufb (@X[2],@X[2],$t3);
+ &vpaddd ($t0,@X[0],&QWP(0,$K256));
+ &vpshufb (@X[3],@X[3],$t3);
+ &vpaddd ($t1,@X[1],&QWP(16,$K256));
+ &vpaddd ($t2,@X[2],&QWP(32,$K256));
+ &vpaddd ($t3,@X[3],&QWP(48,$K256));
+ &vmovdqa (&QWP(32+0,"esp"),$t0);
+ &vmovdqa (&QWP(32+16,"esp"),$t1);
+ &vmovdqa (&QWP(32+32,"esp"),$t2);
+ &vmovdqa (&QWP(32+48,"esp"),$t3);
+ &jmp (&label("avx_00_47"));
+
+&set_label("avx_00_47",16);
+ &add ($K256,64);
+
+sub Xupdate_AVX () {
+ (
+ '&vpalignr ($t0,@X[1],@X[0],4);', # X[1..4]
+ '&vpalignr ($t3,@X[3],@X[2],4);', # X[9..12]
+ '&vpsrld ($t2,$t0,7);',
+ '&vpaddd (@X[0],@X[0],$t3);', # X[0..3] += X[9..16]
+ '&vpsrld ($t3,$t0,3);',
+ '&vpslld ($t1,$t0,14);',
+ '&vpxor ($t0,$t3,$t2);',
+ '&vpsrld ($t2,$t2,18-7);',
+ '&vpxor ($t0,$t0,$t1);',
+ '&vpslld ($t1,$t1,25-14);',
+ '&vpxor ($t0,$t0,$t2);',
+ '&vpsrld ($t3,@X[3],10);',
+ '&vpxor ($t0,$t0,$t1);', # sigma0(X[1..4])
+ '&vpslld ($t2,@X[3],13);',
+ '&vpaddd (@X[0],@X[0],$t0);', # X[0..3] += sigma0(X[1..4])
+ '&vpsrld ($t1,@X[3],17);',
+ '&vpxor ($t3,$t3,$t2);',
+ '&vpslld ($t2,$t2,15-13);',
+ '&vpxor ($t3,$t3,$t1);',
+ '&vpsrld ($t1,$t1,19-17);',
+ '&vpxor ($t3,$t3,$t2);',
+ '&vpxor ($t3,$t3,$t1);', # sigma1(X[14..15])
+ '&vpsrldq ($t3,$t3,8);',
+ '&vpaddd (@X[0],@X[0],$t3);', # X[0..1] += sigma1(X[14..15])
+ '&vpsrld ($t3,@X[0],10);',
+ '&vpslld ($t2,@X[0],13);',
+ '&vpsrld ($t1,@X[0],17);',
+ '&vpxor ($t3,$t3,$t2);',
+ '&vpslld ($t2,$t2,15-13);',
+ '&vpxor ($t3,$t3,$t1);',
+ '&vpsrld ($t1,$t1,19-17);',
+ '&vpxor ($t3,$t3,$t2);',
+ '&vpxor ($t3,$t3,$t1);', # sigma1(X[16..17])
+ '&vpslldq ($t3,$t3,8);',
+ '&vpaddd (@X[0],@X[0],$t3);' # X[2..3] += sigma1(X[16..17])
+ );
+}
+
+local *ror = sub { &shrd(@_[0],@_) };
+sub AVX_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body); # 120 instructions
+
+ foreach (Xupdate_AVX()) { # 35 instructions
+ eval;
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns));
+ }
+ &vpaddd ($t2,@X[0],&QWP(16*$j,$K256));
+ foreach (@insns) { eval; } # remaining instructions
+ &vmovdqa (&QWP(32+16*$j,"esp"),$t2);
+}
+
+ for ($i=0,$j=0; $j<4; $j++) {
+ &AVX_00_47($j,\&body_00_15,@X);
+ push(@X,shift(@X)); # rotate(@X)
+ }
+ &cmp (&DWP(16*$j,$K256),0x00010203);
+ &jne (&label("avx_00_47"));
+
+ for ($i=0; $i<16; ) {
+ foreach(body_00_15()) { eval; }
+ }
+
+ &mov ("esi",&DWP(96,"esp")); #ctx
+ #&mov ($AH[0],&DWP(0,"esp"));
+ &xor ($AH[1],"edi"); #&mov ($AH[1],&DWP(4,"esp"));
+ #&mov ("edi", &DWP(8,"esp"));
+ &mov ("ecx",&DWP(12,"esp"));
+ &add ($AH[0],&DWP(0,"esi"));
+ &add ($AH[1],&DWP(4,"esi"));
+ &add ("edi",&DWP(8,"esi"));
+ &add ("ecx",&DWP(12,"esi"));
+ &mov (&DWP(0,"esi"),$AH[0]);
+ &mov (&DWP(4,"esi"),$AH[1]);
+ &mov (&DWP(8,"esi"),"edi");
+ &mov (&DWP(12,"esi"),"ecx");
+ #&mov (&DWP(0,"esp"),$AH[0]);
+ &mov (&DWP(4,"esp"),$AH[1]);
+ &xor ($AH[1],"edi"); # magic
+ &mov (&DWP(8,"esp"),"edi");
+ &mov (&DWP(12,"esp"),"ecx");
+ #&mov ($E,&DWP(16,"esp"));
+ &mov ("edi",&DWP(20,"esp"));
+ &mov ("ecx",&DWP(24,"esp"));
+ &add ($E,&DWP(16,"esi"));
+ &add ("edi",&DWP(20,"esi"));
+ &add ("ecx",&DWP(24,"esi"));
+ &mov (&DWP(16,"esi"),$E);
+ &mov (&DWP(20,"esi"),"edi");
+ &mov (&DWP(20,"esp"),"edi");
+ &mov ("edi",&DWP(28,"esp"));
+ &mov (&DWP(24,"esi"),"ecx");
+ #&mov (&DWP(16,"esp"),$E);
+ &add ("edi",&DWP(28,"esi"));
+ &mov (&DWP(24,"esp"),"ecx");
+ &mov (&DWP(28,"esi"),"edi");
+ &mov (&DWP(28,"esp"),"edi");
+ &mov ("edi",&DWP(96+4,"esp")); # inp
+
+ &vmovdqa ($t3,&QWP(64,$K256));
+ &sub ($K256,3*64); # rewind K
+ &cmp ("edi",&DWP(96+8,"esp")); # are we done yet?
+ &jb (&label("grand_avx"));
+
+ &mov ("esp",&DWP(96+12,"esp")); # restore sp
+ &vzeroall ();
+&function_end_A();
+ }}}
}
&function_end_B("sha256_block_data_order");
&asciz("SHA256 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>");
projects / openssl / commitdiff