# PIII 63 /77 16 24
# P4 96 /122 30 84(***)
# Opteron 50 /71 21 30
-# Core2 54 /68 13 18
+# Core2 54 /68 12.5 18
#
# (*) gcc 3.4.x was observed to generate few percent slower code,
# which is one of reasons why 2.95.3 results were chosen,
# position-independent;
# (***) see comment in non-MMX routine for further details;
#
-# To summarize, it's 2-3 times faster than gcc-generated code. To
+# To summarize, it's >2-3 times faster than gcc-generated code. To
# anchor it to something else SHA1 assembler processes one byte in
# 11-13 cycles on contemporary x86 cores.
+# May 2010
+#
+# Add PCLMULQDQ version performing at 2.13 cycles per processed byte.
+# The question is how close is it to theoretical limit? The pclmulqdq
+# instruction latency appears to be 14 cycles and there can't be more
+# than 2 of them executing at any given time. This means that single
+# Karatsuba multiplication would take 28 cycles *plus* few cycles for
+# pre- and post-processing. Then multiplication has to be followed by
+# modulo-reduction. Given that aggregated reduction method [see
+# "Carry-less Multiplication and Its Usage for Computing the GCM Mode"
+# white paper by Intel] allows you to perform reduction only once in
+# a while we can assume that asymptotic performance can be estimated
+# as (28+Tmod/Naggr)/16, where Tmod is time to perform reduction
+# and Naggr is the aggregation factor.
+#
+# Before we proceed to this implementation let's have closer look at
+# the best-performing code suggested by Intel in their white paper.
+# By tracing inter-register dependencies Tmod is estimated as ~19
+# cycles and Naggr is 4, resulting in 2.05 cycles per processed byte.
+# As implied, this is quite optimistic estimate, because it does not
+# account for Karatsuba pre- and post-processing, which for a single
+# multiplication is ~5 cycles. Unfortunately Intel does not provide
+# performance data for GHASH alone, only for fused GCM mode. But
+# we can estimate it by subtracting CTR performance result provided
+# in "AES Instruction Set" white paper: 3.54-1.38=2.16 cycles per
+# processed byte or 5% off the estimate. It should be noted though
+# that 3.54 is GCM result for 16KB block size, while 1.38 is CTR for
+# 1KB block size, meaning that real number is likely to be a bit
+# further from estimate.
+#
+# Moving on to the implementation in question. Tmod is estimated as
+# ~13 cycles and Naggr is 2, giving asymptotic performance of ...
+# 2.16. How is it possible that measured performance is better than
+# optimistic theoretical estimate? There is one thing Intel failed
+# to recognize. By fusing GHASH with CTR former's performance is
+# really limited to above (Tmul + Tmod/Naggr) equation. But if GHASH
+# procedure is detached, the modulo-reduction can be interleaved with
+# Naggr-1 multiplications and under ideal conditions even disappear
+# from the equation. So that optimistic theoretical estimate for this
+# implementation is ... 28/16=1.75, and not 2.16. Well, it's probably
+# way too optimistic, at least for such small Naggr. I'd argue that
+# (28+Tproc/Naggr), where Tproc is time required for Karatsuba pre-
+# and post-processing, is more realistic estimate. In this case it
+# gives ... 1.91 cycles per processed byte. Or in other words,
+# depending on how well we can interleave reduction and one of the
+# two multiplications the performance should be betwen 1.91 and 2.16.
+# As already mentioned, this implementation processes one byte [out
+# of 1KB buffer] in 2.13 cycles, while x86_64 counterpart - in 2.07.
+# x86_64 performance is better, because larger register bank allows
+# to interleave reduction and multiplication better.
+#
+# Does it make sense to increase Naggr? To start with it's virtually
+# impossible in 32-bit mode, because of limited register bank
+# capacity. Otherwise improvement has to be weighed agiainst slower
+# setup, as well as code size and complexity increase. As even
+# optimistic estimate doesn't promise 30% performance improvement,
+# there are currently no plans to increase Naggr.
+
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
-&asm_init($ARGV[0],"gcm-x86.pl",$x86only = $ARGV[$#ARGV] eq "386");
+&asm_init($ARGV[0],"ghash-x86.pl",$x86only = $ARGV[$#ARGV] eq "386");
-&static_label("rem_4bit") if (!$x86only);
+$sse2=0;
+for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
-$Zhh = "ebp";
-$Zhl = "edx";
-$Zlh = "ecx";
-$Zll = "ebx";
+($Zhh,$Zhl,$Zlh,$Zll) = ("ebp","edx","ecx","ebx");
$inp = "edi";
$Htbl = "esi";
-
+\f
$unroll = 0; # Affects x86 loop. Folded loop performs ~7% worse
# than unrolled, which has to be weighted against
- # 1.7x code size reduction. Well, *overall* 1.7x,
- # x86-specific code itself shrinks by 2.5x...
-
-sub mmx_loop() {
-# MMX version performs 2.8 times better on P4 (see comment in non-MMX
-# routine for further details), 40% better on Opteron, 50% better
-# on PIII and Core2... In other words effort is considered to be well
-# spent...
- my $inp = shift;
- my $rem_4bit = shift;
- my $cnt = $Zhh;
- my $nhi = $Zhl;
- my $nlo = $Zlh;
- my $rem = $Zll;
-
- my $Zlo = "mm0";
- my $Zhi = "mm1";
- my $tmp = "mm2";
-
- &xor ($nlo,$nlo); # avoid partial register stalls on PIII
- &mov ($nhi,$Zll);
- &mov (&LB($nlo),&LB($nhi));
- &mov ($cnt,14);
- &shl (&LB($nlo),4);
- &and ($nhi,0xf0);
- &movq ($Zlo,&QWP(8,$Htbl,$nlo));
- &movq ($Zhi,&QWP(0,$Htbl,$nlo));
- &movd ($rem,$Zlo);
- &jmp (&label("mmx_loop"));
-
- &set_label("mmx_loop",16);
- &psrlq ($Zlo,4);
- &and ($rem,0xf);
- &pxor ($Zlo,&QWP(8,$Htbl,$nhi));
- &movq ($tmp,$Zhi);
- &psrlq ($Zhi,4);
- &mov (&LB($nlo),&BP(0,$inp,$cnt));
- &dec ($cnt);
- &psllq ($tmp,60);
- &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
- &movd ($rem,$Zlo);
- &pxor ($Zhi,&QWP(0,$Htbl,$nhi));
- &mov ($nhi,$nlo);
- &pxor ($Zlo,$tmp);
- &js (&label("mmx_break"));
-
- &shl (&LB($nlo),4);
- &and ($rem,0xf);
- &psrlq ($Zlo,4);
- &and ($nhi,0xf0);
- &movq ($tmp,$Zhi);
- &psrlq ($Zhi,4);
- &pxor ($Zlo,&QWP(8,$Htbl,$nlo));
- &psllq ($tmp,60);
- &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
- &movd ($rem,$Zlo);
- &pxor ($Zhi,&QWP(0,$Htbl,$nlo));
- &pxor ($Zlo,$tmp);
- &jmp (&label("mmx_loop"));
-
- &set_label("mmx_break",16);
- &shl (&LB($nlo),4);
- &and ($rem,0xf);
- &psrlq ($Zlo,4);
- &and ($nhi,0xf0);
- &movq ($tmp,$Zhi);
- &psrlq ($Zhi,4);
- &pxor ($Zlo,&QWP(8,$Htbl,$nlo));
- &psllq ($tmp,60);
- &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
- &movd ($rem,$Zlo);
- &pxor ($Zhi,&QWP(0,$Htbl,$nlo));
- &pxor ($Zlo,$tmp);
-
- &psrlq ($Zlo,4);
- &and ($rem,0xf);
- &pxor ($Zlo,&QWP(8,$Htbl,$nhi));
- &movq ($tmp,$Zhi);
- &psrlq ($Zhi,4);
- &psllq ($tmp,60);
- &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
- &movd ($rem,$Zlo);
- &pxor ($Zhi,&QWP(0,$Htbl,$nhi));
- &mov ($nhi,$nlo);
- &pxor ($Zlo,$tmp);
-
- &psrlq ($Zlo,32); # lower part of Zlo is already there
- &movd ($Zhl,$Zhi);
- &psrlq ($Zhi,32);
- &movd ($Zlh,$Zlo);
- &movd ($Zhh,$Zhi);
-
- &bswap ($Zll);
- &bswap ($Zhl);
- &bswap ($Zlh);
- &bswap ($Zhh);
-}
+ # 2.5x x86-specific code size reduction.
sub x86_loop {
my $off = shift;
&function_end_B("_x86_gmult_4bit_inner");
}
-&function_begin("gcm_gmult_4bit");
- if (!$x86only) {
- &call (&label("pic_point"));
- &set_label("pic_point");
- &blindpop("eax");
- &picmeup("ebp","OPENSSL_ia32cap_P","eax",&label("pic_point"));
- &bt (&DWP(0,"ebp"),23); # check for MMX bit
- &jnc (&label("x86"));
-
- &lea ("eax",&DWP(&label("rem_4bit")."-".&label("pic_point"),"eax"));
-
- &mov ($inp,&wparam(0)); # load Xi
- &mov ($Htbl,&wparam(1)); # load Htable
-
- &movz ($Zll,&BP(15,$inp));
-
- &mmx_loop($inp,"eax");
+sub deposit_rem_4bit {
+ my $bias = shift;
- &emms ();
- &mov (&DWP(12,$inp),$Zll);
- &mov (&DWP(4,$inp),$Zhl);
- &mov (&DWP(8,$inp),$Zlh);
- &mov (&DWP(0,$inp),$Zhh);
+ &mov (&DWP($bias+0, "esp"),0x0000<<16);
+ &mov (&DWP($bias+4, "esp"),0x1C20<<16);
+ &mov (&DWP($bias+8, "esp"),0x3840<<16);
+ &mov (&DWP($bias+12,"esp"),0x2460<<16);
+ &mov (&DWP($bias+16,"esp"),0x7080<<16);
+ &mov (&DWP($bias+20,"esp"),0x6CA0<<16);
+ &mov (&DWP($bias+24,"esp"),0x48C0<<16);
+ &mov (&DWP($bias+28,"esp"),0x54E0<<16);
+ &mov (&DWP($bias+32,"esp"),0xE100<<16);
+ &mov (&DWP($bias+36,"esp"),0xFD20<<16);
+ &mov (&DWP($bias+40,"esp"),0xD940<<16);
+ &mov (&DWP($bias+44,"esp"),0xC560<<16);
+ &mov (&DWP($bias+48,"esp"),0x9180<<16);
+ &mov (&DWP($bias+52,"esp"),0x8DA0<<16);
+ &mov (&DWP($bias+56,"esp"),0xA9C0<<16);
+ &mov (&DWP($bias+60,"esp"),0xB5E0<<16);
+}
+\f
+$suffix = $x86only ? "" : "_x86";
- &function_end_A();
- &set_label("x86",16);
- }
+&function_begin("gcm_gmult_4bit".$suffix);
&stack_push(16+4+1); # +1 for stack alignment
&mov ($inp,&wparam(0)); # load Xi
&mov ($Htbl,&wparam(1)); # load Htable
&mov (&DWP(4,$inp),$Zhl);
&mov (&DWP(0,$inp),$Zhh);
&stack_pop(16+4+1);
-&function_end("gcm_gmult_4bit");
+&function_end("gcm_gmult_4bit".$suffix);
+
+&function_begin("gcm_ghash_4bit".$suffix);
+ &stack_push(16+4+1); # +1 for 64-bit alignment
+ &mov ($Zll,&wparam(0)); # load Xi
+ &mov ($Htbl,&wparam(1)); # load Htable
+ &mov ($inp,&wparam(2)); # load in
+ &mov ("ecx",&wparam(3)); # load len
+ &add ("ecx",$inp);
+ &mov (&wparam(3),"ecx");
+
+ &mov ($Zhh,&DWP(0,$Zll)); # load Xi[16]
+ &mov ($Zhl,&DWP(4,$Zll));
+ &mov ($Zlh,&DWP(8,$Zll));
+ &mov ($Zll,&DWP(12,$Zll));
+
+ &deposit_rem_4bit(16);
+
+ &set_label("x86_outer_loop",16);
+ &xor ($Zll,&DWP(12,$inp)); # xor with input
+ &xor ($Zlh,&DWP(8,$inp));
+ &xor ($Zhl,&DWP(4,$inp));
+ &xor ($Zhh,&DWP(0,$inp));
+ &mov (&DWP(12,"esp"),$Zll); # dump it on stack
+ &mov (&DWP(8,"esp"),$Zlh);
+ &mov (&DWP(4,"esp"),$Zhl);
+ &mov (&DWP(0,"esp"),$Zhh);
+
+ &shr ($Zll,20);
+ &and ($Zll,0xf0);
+
+ if ($unroll) {
+ &call ("_x86_gmult_4bit_inner");
+ } else {
+ &x86_loop(0);
+ &mov ($inp,&wparam(2));
+ }
+ &lea ($inp,&DWP(16,$inp));
+ &cmp ($inp,&wparam(3));
+ &mov (&wparam(2),$inp) if (!$unroll);
+ &jb (&label("x86_outer_loop"));
+
+ &mov ($inp,&wparam(0)); # load Xi
+ &mov (&DWP(12,$inp),$Zll);
+ &mov (&DWP(8,$inp),$Zlh);
+ &mov (&DWP(4,$inp),$Zhl);
+ &mov (&DWP(0,$inp),$Zhh);
+ &stack_pop(16+4+1);
+&function_end("gcm_ghash_4bit".$suffix);
+\f
+if (!$x86only) {{{
+
+&static_label("rem_4bit");
+
+sub mmx_loop() {
+# MMX version performs 2.8 times better on P4 (see comment in non-MMX
+# routine for further details), 40% better on Opteron and Core2, 50%
+# better on PIII... In other words effort is considered to be well
+# spent...
+ my $inp = shift;
+ my $rem_4bit = shift;
+ my $cnt = $Zhh;
+ my $nhi = $Zhl;
+ my $nlo = $Zlh;
+ my $rem = $Zll;
+
+ my ($Zlo,$Zhi) = ("mm0","mm1");
+ my $tmp = "mm2";
+
+ &xor ($nlo,$nlo); # avoid partial register stalls on PIII
+ &mov ($nhi,$Zll);
+ &mov (&LB($nlo),&LB($nhi));
+ &mov ($cnt,14);
+ &shl (&LB($nlo),4);
+ &and ($nhi,0xf0);
+ &movq ($Zlo,&QWP(8,$Htbl,$nlo));
+ &movq ($Zhi,&QWP(0,$Htbl,$nlo));
+ &movd ($rem,$Zlo);
+ &jmp (&label("mmx_loop"));
+
+ &set_label("mmx_loop",16);
+ &psrlq ($Zlo,4);
+ &and ($rem,0xf);
+ &movq ($tmp,$Zhi);
+ &psrlq ($Zhi,4);
+ &pxor ($Zlo,&QWP(8,$Htbl,$nhi));
+ &mov (&LB($nlo),&BP(0,$inp,$cnt));
+ &psllq ($tmp,60);
+ &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
+ &dec ($cnt);
+ &movd ($rem,$Zlo);
+ &pxor ($Zhi,&QWP(0,$Htbl,$nhi));
+ &mov ($nhi,$nlo);
+ &pxor ($Zlo,$tmp);
+ &js (&label("mmx_break"));
+
+ &shl (&LB($nlo),4);
+ &and ($rem,0xf);
+ &psrlq ($Zlo,4);
+ &and ($nhi,0xf0);
+ &movq ($tmp,$Zhi);
+ &psrlq ($Zhi,4);
+ &pxor ($Zlo,&QWP(8,$Htbl,$nlo));
+ &psllq ($tmp,60);
+ &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
+ &movd ($rem,$Zlo);
+ &pxor ($Zhi,&QWP(0,$Htbl,$nlo));
+ &pxor ($Zlo,$tmp);
+ &jmp (&label("mmx_loop"));
+
+ &set_label("mmx_break",16);
+ &shl (&LB($nlo),4);
+ &and ($rem,0xf);
+ &psrlq ($Zlo,4);
+ &and ($nhi,0xf0);
+ &movq ($tmp,$Zhi);
+ &psrlq ($Zhi,4);
+ &pxor ($Zlo,&QWP(8,$Htbl,$nlo));
+ &psllq ($tmp,60);
+ &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
+ &movd ($rem,$Zlo);
+ &pxor ($Zhi,&QWP(0,$Htbl,$nlo));
+ &pxor ($Zlo,$tmp);
+
+ &psrlq ($Zlo,4);
+ &and ($rem,0xf);
+ &movq ($tmp,$Zhi);
+ &psrlq ($Zhi,4);
+ &pxor ($Zlo,&QWP(8,$Htbl,$nhi));
+ &psllq ($tmp,60);
+ &pxor ($Zhi,&QWP(0,$rem_4bit,$rem,8));
+ &movd ($rem,$Zlo);
+ &pxor ($Zhi,&QWP(0,$Htbl,$nhi));
+ &pxor ($Zlo,$tmp);
+
+ &psrlq ($Zlo,32); # lower part of Zlo is already there
+ &movd ($Zhl,$Zhi);
+ &psrlq ($Zhi,32);
+ &movd ($Zlh,$Zlo);
+ &movd ($Zhh,$Zhi);
+
+ &bswap ($Zll);
+ &bswap ($Zhl);
+ &bswap ($Zlh);
+ &bswap ($Zhh);
+}
+
+&function_begin("gcm_gmult_4bit_mmx");
+ &mov ($inp,&wparam(0)); # load Xi
+ &mov ($Htbl,&wparam(1)); # load Htable
-# Streamed version performs 20% better on P4, 7% on Opteron,
-# 10% on Core2 and PIII...
-&function_begin("gcm_ghash_4bit");
- if (!$x86only) {
&call (&label("pic_point"));
&set_label("pic_point");
&blindpop("eax");
- &picmeup("ebp","OPENSSL_ia32cap_P","eax",&label("pic_point"));
- &bt (&DWP(0,"ebp"),23); # check for MMX bit
- &jnc (&label("x86"));
-
&lea ("eax",&DWP(&label("rem_4bit")."-".&label("pic_point"),"eax"));
+ &movz ($Zll,&BP(15,$inp));
+
+ &mmx_loop($inp,"eax");
+
+ &emms ();
+ &mov (&DWP(12,$inp),$Zll);
+ &mov (&DWP(4,$inp),$Zhl);
+ &mov (&DWP(8,$inp),$Zlh);
+ &mov (&DWP(0,$inp),$Zhh);
+&function_end("gcm_gmult_4bit_mmx");
+\f
+# Streamed version performs 20% better on P4, 7% on Opteron,
+# 10% on Core2 and PIII...
+&function_begin("gcm_ghash_4bit_mmx");
&mov ($Zhh,&wparam(0)); # load Xi
&mov ($Htbl,&wparam(1)); # load Htable
&mov ($inp,&wparam(2)); # load in
&mov ($Zlh,&wparam(3)); # load len
+
+ &call (&label("pic_point"));
+ &set_label("pic_point");
+ &blindpop("eax");
+ &lea ("eax",&DWP(&label("rem_4bit")."-".&label("pic_point"),"eax"));
+
&add ($Zlh,$inp);
&mov (&wparam(3),$Zlh); # len to point at the end of input
&stack_push(4+1); # +1 for stack alignment
+
&mov ($Zll,&DWP(12,$Zhh)); # load Xi[16]
&mov ($Zhl,&DWP(4,$Zhh));
&mov ($Zlh,&DWP(8,$Zhh));
&mov ($Zhh,&DWP(0,$Zhh));
+ &jmp (&label("mmx_outer_loop"));
&set_label("mmx_outer_loop",16);
&xor ($Zll,&DWP(12,$inp));
&mov (&DWP(0,$inp),$Zhh);
&stack_pop(4+1);
- &function_end_A();
- &set_label("x86",16);
- }
- &stack_push(16+4+1); # +1 for 64-bit alignment
- &mov ($Zll,&wparam(0)); # load Xi
- &mov ($Htbl,&wparam(1)); # load Htable
- &mov ($inp,&wparam(2)); # load in
- &mov ("ecx",&wparam(3)); # load len
- &add ("ecx",$inp);
- &mov (&wparam(3),"ecx");
+&function_end("gcm_ghash_4bit_mmx");
+\f
+if ($sse2) {{
+######################################################################
+# PCLMULQDQ version.
+
+$Xip="eax";
+$Htbl="edx";
+$const="ecx";
+$inp="esi";
+$len="ebx";
+
+($Xi,$Xhi)=("xmm0","xmm1"); $Hkey="xmm2";
+($T1,$T2,$T3)=("xmm3","xmm4","xmm5");
+($Xn,$Xhn)=("xmm6","xmm7");
+
+&static_label("bswap");
+
+sub clmul64x64_T2 { # minimal "register" pressure
+my ($Xhi,$Xi,$Hkey)=@_;
+
+ &movdqa ($Xhi,$Xi); #
+ &pshufd ($T1,$Xi,0b01001110);
+ &pshufd ($T2,$Hkey,0b01001110);
+ &pxor ($T1,$Xi); #
+ &pxor ($T2,$Hkey);
+
+ &pclmulqdq ($Xi,$Hkey,0x00); #######
+ &pclmulqdq ($Xhi,$Hkey,0x11); #######
+ &pclmulqdq ($T1,$T2,0x00); #######
+ &pxor ($T1,$Xi); #
+ &pxor ($T1,$Xhi); #
+
+ &movdqa ($T2,$T1); #
+ &psrldq ($T1,8);
+ &pslldq ($T2,8); #
+ &pxor ($Xhi,$T1);
+ &pxor ($Xi,$T2); #
+}
- &mov ($Zhh,&DWP(0,$Zll)); # load Xi[16]
- &mov ($Zhl,&DWP(4,$Zll));
- &mov ($Zlh,&DWP(8,$Zll));
- &mov ($Zll,&DWP(12,$Zll));
+sub clmul64x64_T3 {
+# Even though this subroutine offers visually better ILP, it
+# was empirically found to be a tad slower than above version.
+# At least in gcm_ghash_clmul context. But it's just as well,
+# because loop modulo-scheduling is possible only thanks to
+# minimized "register" pressure...
+my ($Xhi,$Xi,$Hkey)=@_;
+
+ &movdqa ($T1,$Xi); #
+ &movdqa ($Xhi,$Xi);
+ &pclmulqdq ($Xi,$Hkey,0x00); #######
+ &pclmulqdq ($Xhi,$Hkey,0x11); #######
+ &pshufd ($T2,$T1,0b01001110); #
+ &pshufd ($T3,$Hkey,0b01001110);
+ &pxor ($T2,$T1); #
+ &pxor ($T3,$Hkey);
+ &pclmulqdq ($T2,$T3,0x00); #######
+ &pxor ($T2,$Xi); #
+ &pxor ($T2,$Xhi); #
+
+ &movdqa ($T3,$T2); #
+ &psrldq ($T2,8);
+ &pslldq ($T3,8); #
+ &pxor ($Xhi,$T2);
+ &pxor ($Xi,$T3); #
+}
+\f
+if (1) { # Algorithm 9 with <<1 twist.
+ # Reduction is shorter and uses only two
+ # temporary registers, which makes it better
+ # candidate for interleaving with 64x64
+ # multiplication. Pre-modulo-scheduled loop
+ # was found to be ~20% faster than Algorithm 5
+ # below. Algorithm 9 was then chosen and
+ # optimized further...
+
+sub reduction_alg9 { # 17/13 times faster than Intel version
+my ($Xhi,$Xi) = @_;
+
+ # 1st phase
+ &movdqa ($T1,$Xi) #
+ &psllq ($Xi,1);
+ &pxor ($Xi,$T1); #
+ &psllq ($Xi,5); #
+ &pxor ($Xi,$T1); #
+ &psllq ($Xi,57); #
+ &movdqa ($T2,$Xi); #
+ &pslldq ($Xi,8);
+ &psrldq ($T2,8); #
+ &pxor ($Xi,$T1);
+ &pxor ($Xhi,$T2); #
+
+ # 2nd phase
+ &movdqa ($T2,$Xi);
+ &psrlq ($Xi,5);
+ &pxor ($Xi,$T2); #
+ &psrlq ($Xi,1); #
+ &pxor ($Xi,$T2); #
+ &pxor ($T2,$Xhi);
+ &psrlq ($Xi,1); #
+ &pxor ($Xi,$T2); #
+}
- &deposit_rem_4bit(16);
+&function_begin_B("gcm_init_clmul");
+ &mov ($Htbl,&wparam(0));
+ &mov ($Xip,&wparam(1));
- &set_label("x86_outer_loop",16);
- &xor ($Zll,&DWP(12,$inp)); # xor with input
- &xor ($Zlh,&DWP(8,$inp));
- &xor ($Zhl,&DWP(4,$inp));
- &xor ($Zhh,&DWP(0,$inp));
- &mov (&DWP(12,"esp"),$Zll); # dump it on stack
- &mov (&DWP(8,"esp"),$Zlh);
- &mov (&DWP(4,"esp"),$Zhl);
- &mov (&DWP(0,"esp"),$Zhh);
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
- &shr ($Zll,20);
- &and ($Zll,0xf0);
+ &movdqu ($Hkey,&QWP(0,$Xip));
+ &pshufd ($Hkey,$Hkey,0b01001110);# dword swap
- if ($unroll) {
- &call ("_x86_gmult_4bit_inner");
- } else {
- &x86_loop(0);
- &mov ($inp,&wparam(2));
- }
- &lea ($inp,&DWP(16,$inp));
- &cmp ($inp,&wparam(3));
- &mov (&wparam(2),$inp) if (!$unroll);
- &jb (&label("x86_outer_loop"));
+ # <<1 twist
+ &pshufd ($T2,$Hkey,0b11111111); # broadcast uppermost dword
+ &movdqa ($T1,$Hkey);
+ &psllq ($Hkey,1);
+ &pxor ($T3,$T3); #
+ &psrlq ($T1,63);
+ &pcmpgtd ($T3,$T2); # broadcast carry bit
+ &pslldq ($T1,8);
+ &por ($Hkey,$T1); # H<<=1
- &mov ($inp,&wparam(0)); # load Xi
- &mov (&DWP(12,$inp),$Zll);
- &mov (&DWP(8,$inp),$Zlh);
- &mov (&DWP(4,$inp),$Zhl);
- &mov (&DWP(0,$inp),$Zhh);
- &stack_pop(16+4+1);
-&function_end("gcm_ghash_4bit");
+ # magic reduction
+ &pand ($T3,&QWP(16,$const)); # 0x1c2_polynomial
+ &pxor ($Hkey,$T3); # if(carry) H^=0x1c2_polynomial
-sub deposit_rem_4bit {
- my $bias = shift;
+ # calculate H^2
+ &movdqa ($Xi,$Hkey);
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey);
+ &reduction_alg9 ($Xhi,$Xi);
- &mov (&DWP($bias+0, "esp"),0x0000<<16);
- &mov (&DWP($bias+4, "esp"),0x1C20<<16);
- &mov (&DWP($bias+8, "esp"),0x3840<<16);
- &mov (&DWP($bias+12,"esp"),0x2460<<16);
- &mov (&DWP($bias+16,"esp"),0x7080<<16);
- &mov (&DWP($bias+20,"esp"),0x6CA0<<16);
- &mov (&DWP($bias+24,"esp"),0x48C0<<16);
- &mov (&DWP($bias+28,"esp"),0x54E0<<16);
- &mov (&DWP($bias+32,"esp"),0xE100<<16);
- &mov (&DWP($bias+36,"esp"),0xFD20<<16);
- &mov (&DWP($bias+40,"esp"),0xD940<<16);
- &mov (&DWP($bias+44,"esp"),0xC560<<16);
- &mov (&DWP($bias+48,"esp"),0x9180<<16);
- &mov (&DWP($bias+52,"esp"),0x8DA0<<16);
- &mov (&DWP($bias+56,"esp"),0xA9C0<<16);
- &mov (&DWP($bias+60,"esp"),0xB5E0<<16);
+ &movdqu (&QWP(0,$Htbl),$Hkey); # save H
+ &movdqu (&QWP(16,$Htbl),$Xi); # save H^2
+
+ &ret ();
+&function_end_B("gcm_init_clmul");
+
+&function_begin_B("gcm_gmult_clmul");
+ &mov ($Xip,&wparam(0));
+ &mov ($Htbl,&wparam(1));
+
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
+
+ &movdqu ($Xi,&QWP(0,$Xip));
+ &movdqa ($T3,&QWP(0,$const));
+ &movdqu ($Hkey,&QWP(0,$Htbl));
+ &pshufb ($Xi,$T3);
+
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey);
+ &reduction_alg9 ($Xhi,$Xi);
+
+ &pshufb ($Xi,$T3);
+ &movdqu (&QWP(0,$Xip),$Xi);
+
+ &ret ();
+&function_end_B("gcm_gmult_clmul");
+
+&function_begin("gcm_ghash_clmul");
+ &mov ($Xip,&wparam(0));
+ &mov ($Htbl,&wparam(1));
+ &mov ($inp,&wparam(2));
+ &mov ($len,&wparam(3));
+
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
+
+ &movdqu ($Xi,&QWP(0,$Xip));
+ &movdqa ($T3,&QWP(0,$const));
+ &movdqu ($Hkey,&QWP(0,$Htbl));
+ &pshufb ($Xi,$T3);
+
+ &sub ($len,0x10);
+ &jz (&label("odd_tail"));
+
+ #######
+ # Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
+ # [(H*Ii+1) + (H*Xi+1)] mod P =
+ # [(H*Ii+1) + H^2*(Ii+Xi)] mod P
+ #
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &movdqu ($Xn,&QWP(16,$inp)); # Ii+1
+ &pshufb ($T1,$T3);
+ &pshufb ($Xn,$T3);
+ &pxor ($Xi,$T1); # Ii+Xi
+
+ &clmul64x64_T2 ($Xhn,$Xn,$Hkey); # H*Ii+1
+ &movdqu ($Hkey,&QWP(16,$Htbl)); # load H^2
+
+ &lea ($inp,&DWP(32,$inp)); # i+=2
+ &sub ($len,0x20);
+ &jbe (&label("even_tail"));
+
+&set_label("mod_loop");
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey); # H^2*(Ii+Xi)
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &movdqu ($Hkey,&QWP(0,$Htbl)); # load H
+
+ &pxor ($Xi,$Xn); # (H*Ii+1) + H^2*(Ii+Xi)
+ &pxor ($Xhi,$Xhn);
+
+ &movdqu ($Xn,&QWP(16,$inp)); # Ii+1
+ &pshufb ($T1,$T3);
+ &pshufb ($Xn,$T3);
+
+ &movdqa ($T3,$Xn); #&clmul64x64_TX ($Xhn,$Xn,$Hkey); H*Ii+1
+ &movdqa ($Xhn,$Xn);
+ &pxor ($Xhi,$T1); # "Ii+Xi", consume early
+
+ &movdqa ($T1,$Xi) #&reduction_alg9($Xhi,$Xi); 1st phase
+ &psllq ($Xi,1);
+ &pxor ($Xi,$T1); #
+ &psllq ($Xi,5); #
+ &pxor ($Xi,$T1); #
+ &pclmulqdq ($Xn,$Hkey,0x00); #######
+ &psllq ($Xi,57); #
+ &movdqa ($T2,$Xi); #
+ &pslldq ($Xi,8);
+ &psrldq ($T2,8); #
+ &pxor ($Xi,$T1);
+ &pshufd ($T1,$T3,0b01001110);
+ &pxor ($Xhi,$T2); #
+ &pxor ($T1,$T3);
+ &pshufd ($T3,$Hkey,0b01001110);
+ &pxor ($T3,$Hkey); #
+
+ &pclmulqdq ($Xhn,$Hkey,0x11); #######
+ &movdqa ($T2,$Xi); # 2nd phase
+ &psrlq ($Xi,5);
+ &pxor ($Xi,$T2); #
+ &psrlq ($Xi,1); #
+ &pxor ($Xi,$T2); #
+ &pxor ($T2,$Xhi);
+ &psrlq ($Xi,1); #
+ &pxor ($Xi,$T2); #
+
+ &pclmulqdq ($T1,$T3,0x00); #######
+ &movdqu ($Hkey,&QWP(16,$Htbl)); # load H^2
+ &pxor ($T1,$Xn); #
+ &pxor ($T1,$Xhn); #
+
+ &movdqa ($T3,$T1); #
+ &psrldq ($T1,8);
+ &pslldq ($T3,8); #
+ &pxor ($Xhn,$T1);
+ &pxor ($Xn,$T3); #
+ &movdqa ($T3,&QWP(0,$const));
+
+ &lea ($inp,&DWP(32,$inp));
+ &sub ($len,0x20);
+ &ja (&label("mod_loop"));
+
+&set_label("even_tail");
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey); # H^2*(Ii+Xi)
+
+ &pxor ($Xi,$Xn); # (H*Ii+1) + H^2*(Ii+Xi)
+ &pxor ($Xhi,$Xhn);
+
+ &reduction_alg9 ($Xhi,$Xi);
+
+ &test ($len,$len);
+ &jnz (&label("done"));
+
+ &movdqu ($Hkey,&QWP(0,$Htbl)); # load H
+&set_label("odd_tail");
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &pshufb ($T1,$T3);
+ &pxor ($Xi,$T1); # Ii+Xi
+
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey); # H*(Ii+Xi)
+ &reduction_alg9 ($Xhi,$Xi);
+
+&set_label("done");
+ &pshufb ($Xi,$T3);
+ &movdqu (&QWP(0,$Xip),$Xi);
+&function_end("gcm_ghash_clmul");
+\f
+} else { # Algorith 5. Kept for reference purposes.
+
+sub reduction_alg5 { # 19/16 times faster than Intel version
+my ($Xhi,$Xi)=@_;
+
+ # <<1
+ &movdqa ($T1,$Xi); #
+ &movdqa ($T2,$Xhi);
+ &pslld ($Xi,1);
+ &pslld ($Xhi,1); #
+ &psrld ($T1,31);
+ &psrld ($T2,31); #
+ &movdqa ($T3,$T1);
+ &pslldq ($T1,4);
+ &psrldq ($T3,12); #
+ &pslldq ($T2,4);
+ &por ($Xhi,$T3); #
+ &por ($Xi,$T1);
+ &por ($Xhi,$T2); #
+
+ # 1st phase
+ &movdqa ($T1,$Xi);
+ &movdqa ($T2,$Xi);
+ &movdqa ($T3,$Xi); #
+ &pslld ($T1,31);
+ &pslld ($T2,30);
+ &pslld ($Xi,25); #
+ &pxor ($T1,$T2);
+ &pxor ($T1,$Xi); #
+ &movdqa ($T2,$T1); #
+ &pslldq ($T1,12);
+ &psrldq ($T2,4); #
+ &pxor ($T3,$T1);
+
+ # 2nd phase
+ &pxor ($Xhi,$T3); #
+ &movdqa ($Xi,$T3);
+ &movdqa ($T1,$T3);
+ &psrld ($Xi,1); #
+ &psrld ($T1,2);
+ &psrld ($T3,7); #
+ &pxor ($Xi,$T1);
+ &pxor ($Xhi,$T2);
+ &pxor ($Xi,$T3); #
+ &pxor ($Xi,$Xhi); #
}
-if (!$x86only) {
+&function_begin_B("gcm_init_clmul");
+ &mov ($Htbl,&wparam(0));
+ &mov ($Xip,&wparam(1));
+
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
+
+ &movdqu ($Hkey,&QWP(0,$Xip));
+ &pshufd ($Hkey,$Hkey,0b01001110);# dword swap
+
+ # calculate H^2
+ &movdqa ($Xi,$Hkey);
+ &clmul64x64_T3 ($Xhi,$Xi,$Hkey);
+ &reduction_alg5 ($Xhi,$Xi);
+
+ &movdqu (&QWP(0,$Htbl),$Hkey); # save H
+ &movdqu (&QWP(16,$Htbl),$Xi); # save H^2
+
+ &ret ();
+&function_end_B("gcm_init_clmul");
+
+&function_begin_B("gcm_gmult_clmul");
+ &mov ($Xip,&wparam(0));
+ &mov ($Htbl,&wparam(1));
+
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
+
+ &movdqu ($Xi,&QWP(0,$Xip));
+ &movdqa ($Xn,&QWP(0,$const));
+ &movdqu ($Hkey,&QWP(0,$Htbl));
+ &pshufb ($Xi,$Xn);
+
+ &clmul64x64_T3 ($Xhi,$Xi,$Hkey);
+ &reduction_alg5 ($Xhi,$Xi);
+
+ &pshufb ($Xi,$Xn);
+ &movdqu (&QWP(0,$Xip),$Xi);
+
+ &ret ();
+&function_end_B("gcm_gmult_clmul");
+
+&function_begin("gcm_ghash_clmul");
+ &mov ($Xip,&wparam(0));
+ &mov ($Htbl,&wparam(1));
+ &mov ($inp,&wparam(2));
+ &mov ($len,&wparam(3));
+
+ &call (&label("pic"));
+&set_label("pic");
+ &blindpop ($const);
+ &lea ($const,&DWP(&label("bswap")."-".&label("pic"),$const));
+
+ &movdqu ($Xi,&QWP(0,$Xip));
+ &movdqa ($T3,&QWP(0,$const));
+ &movdqu ($Hkey,&QWP(0,$Htbl));
+ &pshufb ($Xi,$T3);
+
+ &sub ($len,0x10);
+ &jz (&label("odd_tail"));
+
+ #######
+ # Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
+ # [(H*Ii+1) + (H*Xi+1)] mod P =
+ # [(H*Ii+1) + H^2*(Ii+Xi)] mod P
+ #
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &movdqu ($Xn,&QWP(16,$inp)); # Ii+1
+ &pshufb ($T1,$T3);
+ &pshufb ($Xn,$T3);
+ &pxor ($Xi,$T1); # Ii+Xi
+
+ &clmul64x64_T3 ($Xhn,$Xn,$Hkey); # H*Ii+1
+ &movdqu ($Hkey,&QWP(16,$Htbl)); # load H^2
+
+ &sub ($len,0x20);
+ &lea ($inp,&DWP(32,$inp)); # i+=2
+ &jbe (&label("even_tail"));
+
+&set_label("mod_loop");
+ &clmul64x64_T3 ($Xhi,$Xi,$Hkey); # H^2*(Ii+Xi)
+ &movdqu ($Hkey,&QWP(0,$Htbl)); # load H
+
+ &pxor ($Xi,$Xn); # (H*Ii+1) + H^2*(Ii+Xi)
+ &pxor ($Xhi,$Xhn);
+
+ &reduction_alg5 ($Xhi,$Xi);
+
+ #######
+ &movdqa ($T3,&QWP(0,$const));
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &movdqu ($Xn,&QWP(16,$inp)); # Ii+1
+ &pshufb ($T1,$T3);
+ &pshufb ($Xn,$T3);
+ &pxor ($Xi,$T1); # Ii+Xi
+
+ &clmul64x64_T3 ($Xhn,$Xn,$Hkey); # H*Ii+1
+ &movdqu ($Hkey,&QWP(16,$Htbl)); # load H^2
+
+ &sub ($len,0x20);
+ &lea ($inp,&DWP(32,$inp));
+ &ja (&label("mod_loop"));
+
+&set_label("even_tail");
+ &clmul64x64_T3 ($Xhi,$Xi,$Hkey); # H^2*(Ii+Xi)
+
+ &pxor ($Xi,$Xn); # (H*Ii+1) + H^2*(Ii+Xi)
+ &pxor ($Xhi,$Xhn);
+
+ &reduction_alg5 ($Xhi,$Xi);
+
+ &movdqa ($T3,&QWP(0,$const));
+ &test ($len,$len);
+ &jnz (&label("done"));
+
+ &movdqu ($Hkey,&QWP(0,$Htbl)); # load H
+&set_label("odd_tail");
+ &movdqu ($T1,&QWP(0,$inp)); # Ii
+ &pshufb ($T1,$T3);
+ &pxor ($Xi,$T1); # Ii+Xi
+
+ &clmul64x64_T3 ($Xhi,$Xi,$Hkey); # H*(Ii+Xi)
+ &reduction_alg5 ($Xhi,$Xi);
+
+ &movdqa ($T3,&QWP(0,$const));
+&set_label("done");
+ &pshufb ($Xi,$T3);
+ &movdqu (&QWP(0,$Xip),$Xi);
+&function_end("gcm_ghash_clmul");
+
+}
+\f
+&set_label("bswap",64);
+ &data_byte(15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0);
+ &data_byte(1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2); # 0x1c2_polynomial
+}} # $sse2
+
&set_label("rem_4bit",64);
&data_word(0,0x0000<<16,0,0x1C20<<16,0,0x3840<<16,0,0x2460<<16);
&data_word(0,0x7080<<16,0,0x6CA0<<16,0,0x48C0<<16,0,0x54E0<<16);
&data_word(0,0xE100<<16,0,0xFD20<<16,0,0xD940<<16,0,0xC560<<16);
&data_word(0,0x9180<<16,0,0x8DA0<<16,0,0xA9C0<<16,0,0xB5E0<<16);
-}
+}}} # !$x86only
+
&asciz("GHASH for x86, CRYPTOGAMS by <appro\@openssl.org>");
&asm_finish();
# Opteron 18.5 10.2 +80%
# Core2 17.5 11.0 +59%
+# May 2010
+#
+# Add PCLMULQDQ version performing at 2.07 cycles per processed byte.
+# See ghash-x86.pl for background information and details about coding
+# techniques.
+
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
sub lo() { my $r=shift; $r =~ s/%[er]([a-d])x/%\1l/;
$r =~ s/%[er]([sd]i)/%\1l/;
$r =~ s/%(r[0-9]+)[d]?/%\1b/; $r; }
-
+\f
{ my $N;
sub loop() {
my $inp = shift;
ret
.size gcm_gmult_4bit,.-gcm_gmult_4bit
___
-
-
+\f
# per-function register layout
$inp="%rdx";
$len="%rcx";
.Lghash_epilogue:
ret
.size gcm_ghash_4bit,.-gcm_ghash_4bit
+___
+\f
+######################################################################
+# PCLMULQDQ version.
+
+@_4args=$win64? ("%rcx","%rdx","%r8", "%r9") : # Win64 order
+ ("%rdi","%rsi","%rdx","%rcx"); # Unix order
+
+($Xi,$Xhi)=("%xmm0","%xmm1"); $Hkey="%xmm2";
+($T1,$T2,$T3)=("%xmm3","%xmm4","%xmm5");
+
+sub clmul64x64_T2 { # minimal register pressure
+my ($Xhi,$Xi,$Hkey,$modulo)=@_;
+$code.=<<___ if (!defined($modulo));
+ movdqa $Xi,$Xhi #
+ pshufd \$0b01001110,$Xi,$T1
+ pshufd \$0b01001110,$Hkey,$T2
+ pxor $Xi,$T1 #
+ pxor $Hkey,$T2
+___
+$code.=<<___;
+ pclmulqdq \$0x00,$Hkey,$Xi #######
+ pclmulqdq \$0x11,$Hkey,$Xhi #######
+ pclmulqdq \$0x00,$T2,$T1 #######
+ pxor $Xi,$T1 #
+ pxor $Xhi,$T1 #
+
+ movdqa $T1,$T2 #
+ psrldq \$8,$T1
+ pslldq \$8,$T2 #
+ pxor $T1,$Xhi
+ pxor $T2,$Xi #
+___
+}
+
+sub reduction_alg9 { # 17/13 times faster than Intel version
+my ($Xhi,$Xi) = @_;
+
+$code.=<<___;
+ # 1st phase
+ movdqa $Xi,$T1 #
+ psllq \$1,$Xi
+ pxor $T1,$Xi #
+ psllq \$5,$Xi #
+ pxor $T1,$Xi #
+ psllq \$57,$Xi #
+ movdqa $Xi,$T2 #
+ pslldq \$8,$Xi
+ psrldq \$8,$T2 #
+ pxor $T1,$Xi
+ pxor $T2,$Xhi #
+
+ # 2nd phase
+ movdqa $Xi,$T2
+ psrlq \$5,$Xi
+ pxor $T2,$Xi #
+ psrlq \$1,$Xi #
+ pxor $T2,$Xi #
+ pxor $Xhi,$T2
+ psrlq \$1,$Xi #
+ pxor $T2,$Xi #
+___
+}
+\f
+{ my ($Htbl,$Xip)=@_4args;
+
+$code.=<<___;
+.globl gcm_init_clmul
+.type gcm_init_clmul,\@abi-omnipotent
+.align 16
+gcm_init_clmul:
+ movdqu ($Xip),$Hkey
+ pshufd \$0b01001110,$Hkey,$Hkey # dword swap
+
+ # <<1 twist
+ pshufd \$0b11111111,$Hkey,$T2 # broadcast uppermost dword
+ movdqa $Hkey,$T1
+ psllq \$1,$Hkey
+ pxor $T3,$T3 #
+ psrlq \$63,$T1
+ pcmpgtd $T2,$T3 # broadcast carry bit
+ pslldq \$8,$T1
+ por $T1,$Hkey # H<<=1
+
+ # magic reduction
+ pand .L0x1c2_polynomial(%rip),$T3
+ pxor $T3,$Hkey # if(carry) H^=0x1c2_polynomial
+
+ # calculate H^2
+ movdqa $Hkey,$Xi
+___
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey);
+ &reduction_alg9 ($Xhi,$Xi);
+$code.=<<___;
+ movdqu $Hkey,($Htbl) # save H
+ movdqu $Xi,16($Htbl) # save H^2
+ ret
+.size gcm_init_clmul,.-gcm_init_clmul
+___
+}
+
+{ my ($Xip,$Htbl)=@_4args;
+
+$code.=<<___;
+.globl gcm_gmult_clmul
+.type gcm_gmult_clmul,\@abi-omnipotent
+.align 16
+gcm_gmult_clmul:
+ movdqu ($Xip),$Xi
+ movdqa .Lbswap_mask(%rip),$T3
+ movdqu ($Htbl),$Hkey
+ pshufb $T3,$Xi
+___
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey);
+ &reduction_alg9 ($Xhi,$Xi);
+$code.=<<___;
+ pshufb $T3,$Xi
+ movdqu $Xi,($Xip)
+ ret
+.size gcm_gmult_clmul,.-gcm_gmult_clmul
+___
+}
+\f
+{ my ($Xip,$Htbl,$inp,$len)=@_4args;
+ my $Xn="%xmm6";
+ my $Xhn="%xmm7";
+ my $Hkey2="%xmm8";
+ my $T1n="%xmm9";
+ my $T2n="%xmm10";
+
+$code.=<<___;
+.globl gcm_ghash_clmul
+.type gcm_ghash_clmul,\@abi-omnipotent
+.align 16
+gcm_ghash_clmul:
+___
+$code.=<<___ if ($win64);
+.LSEH_begin_gcm_ghash_clmul:
+ # I can't trust assembler to use specific encoding:-(
+ .byte 0x48,0x83,0xec,0x58 #sub \$0x58,%rsp
+ .byte 0x0f,0x29,0x34,0x24 #movaps %xmm6,(%rsp)
+ .byte 0x0f,0x29,0x7c,0x24,0x10 #movdqa %xmm7,0x10(%rsp)
+ .byte 0x44,0x0f,0x29,0x44,0x24,0x20 #movaps %xmm8,0x20(%rsp)
+ .byte 0x44,0x0f,0x29,0x4c,0x24,0x30 #movaps %xmm9,0x30(%rsp)
+ .byte 0x44,0x0f,0x29,0x54,0x24,0x40 #movaps %xmm10,0x40(%rsp)
+___
+$code.=<<___;
+ movdqa .Lbswap_mask(%rip),$T3
+
+ movdqu ($Xip),$Xi
+ movdqu ($Htbl),$Hkey
+ pshufb $T3,$Xi
+
+ sub \$0x10,$len
+ jz .Lodd_tail
+
+ movdqu 16($Htbl),$Hkey2
+ #######
+ # Xi+2 =[H*(Ii+1 + Xi+1)] mod P =
+ # [(H*Ii+1) + (H*Xi+1)] mod P =
+ # [(H*Ii+1) + H^2*(Ii+Xi)] mod P
+ #
+ movdqu ($inp),$T1 # Ii
+ movdqu 16($inp),$Xn # Ii+1
+ pshufb $T3,$T1
+ pshufb $T3,$Xn
+ pxor $T1,$Xi # Ii+Xi
+___
+ &clmul64x64_T2 ($Xhn,$Xn,$Hkey); # H*Ii+1
+$code.=<<___;
+ movdqa $Xi,$Xhi #
+ pshufd \$0b01001110,$Xi,$T1
+ pshufd \$0b01001110,$Hkey2,$T2
+ pxor $Xi,$T1 #
+ pxor $Hkey2,$T2
+
+ lea 32($inp),$inp # i+=2
+ sub \$0x20,$len
+ jbe .Leven_tail
+
+.Lmod_loop:
+___
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey2,1); # H^2*(Ii+Xi)
+$code.=<<___;
+ movdqu ($inp),$T1 # Ii
+ pxor $Xn,$Xi # (H*Ii+1) + H^2*(Ii+Xi)
+ pxor $Xhn,$Xhi
+
+ movdqu 16($inp),$Xn # Ii+1
+ pshufb $T3,$T1
+ pshufb $T3,$Xn
+
+ movdqa $Xn,$Xhn #
+ pshufd \$0b01001110,$Xn,$T1n
+ pshufd \$0b01001110,$Hkey,$T2n
+ pxor $Xn,$T1n #
+ pxor $Hkey,$T2n
+ pxor $T1,$Xhi # "Ii+Xi", consume early
+
+ movdqa $Xi,$T1 # 1st phase
+ psllq \$1,$Xi
+ pxor $T1,$Xi #
+ psllq \$5,$Xi #
+ pxor $T1,$Xi #
+ pclmulqdq \$0x00,$Hkey,$Xn #######
+ psllq \$57,$Xi #
+ movdqa $Xi,$T2 #
+ pslldq \$8,$Xi
+ psrldq \$8,$T2 #
+ pxor $T1,$Xi
+ pxor $T2,$Xhi #
+
+ pclmulqdq \$0x11,$Hkey,$Xhn #######
+ movdqa $Xi,$T2 # 2nd phase
+ psrlq \$5,$Xi
+ pxor $T2,$Xi #
+ psrlq \$1,$Xi #
+ pxor $T2,$Xi #
+ pxor $Xhi,$T2
+ psrlq \$1,$Xi #
+ pxor $T2,$Xi #
+
+ pclmulqdq \$0x00,$T2n,$T1n #######
+ movdqa $Xi,$Xhi #
+ pshufd \$0b01001110,$Xi,$T1
+ pshufd \$0b01001110,$Hkey2,$T2
+ pxor $Xi,$T1 #
+ pxor $Hkey2,$T2
+
+ pxor $Xn,$T1n #
+ pxor $Xhn,$T1n #
+ movdqa $T1n,$T2n #
+ psrldq \$8,$T1n
+ pslldq \$8,$T2n #
+ pxor $T1n,$Xhn
+ pxor $T2n,$Xn #
+
+ lea 32($inp),$inp
+ sub \$0x20,$len
+ ja .Lmod_loop
+
+.Leven_tail:
+___
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey2,1); # H^2*(Ii+Xi)
+$code.=<<___;
+ pxor $Xn,$Xi # (H*Ii+1) + H^2*(Ii+Xi)
+ pxor $Xhn,$Xhi
+___
+ &reduction_alg9 ($Xhi,$Xi);
+$code.=<<___;
+ test $len,$len
+ jnz .Ldone
+
+.Lodd_tail:
+ movdqu ($inp),$T1 # Ii
+ pshufb $T3,$T1
+ pxor $T1,$Xi # Ii+Xi
+___
+ &clmul64x64_T2 ($Xhi,$Xi,$Hkey); # H*(Ii+Xi)
+ &reduction_alg9 ($Xhi,$Xi);
+$code.=<<___;
+.Ldone:
+ pshufb $T3,$Xi
+ movdqu $Xi,($Xip)
+___
+$code.=<<___ if ($win64);
+ movaps (%rsp),%xmm6
+ movaps 0x10(%rsp),%xmm7
+ movaps 0x20(%rsp),%xmm8
+ movaps 0x30(%rsp),%xmm9
+ movaps 0x40(%rsp),%xmm10
+ add \$0x58,%rsp
+___
+$code.=<<___;
+ ret
+.LSEH_end_gcm_ghash_clmul:
+.size gcm_ghash_clmul,.-gcm_ghash_clmul
+___
+}
+
+$code.=<<___;
+.align 64
+.Lbswap_mask:
+ .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
+.L0x1c2_polynomial:
+ .byte 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
.align 64
-.type rem_4bit,\@object
+.type .Lrem_4bit,\@object
.Lrem_4bit:
.long 0,`0x0000<<16`,0,`0x1C20<<16`,0,`0x3840<<16`,0,`0x2460<<16`
.long 0,`0x7080<<16`,0,`0x6CA0<<16`,0,`0x48C0<<16`,0,`0x54E0<<16`
.asciz "GHASH for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
.align 64
___
-
+\f
# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
# CONTEXT *context,DISPATCHER_CONTEXT *disp)
if ($win64) {
.rva .LSEH_end_gcm_ghash_4bit
.rva .LSEH_info_gcm_ghash_4bit
+ .rva .LSEH_begin_gcm_ghash_clmul
+ .rva .LSEH_end_gcm_ghash_clmul
+ .rva .LSEH_info_gcm_ghash_clmul
+
.section .xdata
.align 8
.LSEH_info_gcm_gmult_4bit:
.byte 9,0,0,0
.rva se_handler
.rva .Lghash_prologue,.Lghash_epilogue # HandlerData
+.LSEH_info_gcm_ghash_clmul:
+ .byte 0x01,0x1f,0x0b,0x00
+ .byte 0x1f,0xa8,0x04,0x00 #movaps 0x40(rsp),xmm10
+ .byte 0x19,0x98,0x03,0x00 #movaps 0x30(rsp),xmm9
+ .byte 0x13,0x88,0x02,0x00 #movaps 0x20(rsp),xmm8
+ .byte 0x0d,0x78,0x01,0x00 #movaps 0x10(rsp),xmm7
+ .byte 0x08,0x68,0x00,0x00 #movaps (rsp),xmm6
+ .byte 0x04,0xa2,0x00,0x00 #sub rsp,0x58
___
}
+\f
+sub rex {
+ local *opcode=shift;
+ my ($dst,$src)=@_;
+
+ if ($dst>=8 || $src>=8) {
+ $rex=0x40;
+ $rex|=0x04 if($dst>=8);
+ $rex|=0x01 if($src>=8);
+ push @opcode,$rex;
+ }
+}
+
+sub pclmulqdq {
+ my $arg=shift;
+ my @opcode=(0x66);
+
+ if ($arg=~/\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
+ rex(\@opcode,$3,$2);
+ push @opcode,0x0f,0x3a,0x44;
+ push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
+ my $c=$1;
+ push @opcode,$c=~/^0/?oct($c):$c;
+ return ".byte\t".join(',',@opcode);
+ }
+ return "pclmulqdq\t".$arg;
+}
+
$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+$code =~ s/\bpclmulqdq\s+(\$.*%xmm[0-9]+).*$/pclmulqdq($1)/gem;
print $code;
#define PUTU32(p,v) *(u32 *)(p) = BSWAP4(v)
#endif
-#define PACK(s) ((size_t)(s)<<(sizeof(size_t)*8-16))
+#define PACK(s) ((size_t)(s)<<(sizeof(size_t)*8-16))
+#define REDUCE1BIT(V) do { \
+ if (sizeof(size_t)==8) { \
+ u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \
+ V.lo = (V.hi<<63)|(V.lo>>1); \
+ V.hi = (V.hi>>1 )^T; \
+ } \
+ else { \
+ u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \
+ V.lo = (V.hi<<63)|(V.lo>>1); \
+ V.hi = (V.hi>>1 )^((u64)T<<32); \
+ } \
+} while(0)
+
#ifdef TABLE_BITS
#undef TABLE_BITS
#endif
* Even though permitted values for TABLE_BITS are 8, 4 and 1, it should
* never be set to 8. 8 is effectively reserved for testing purposes.
* Under ideal conditions "8-bit" version should be twice as fast as
- * "4-bit" one. But world is far from ideal. For gcc-generated x86 code,
- * "8-bit" was observed to run only ~50% faster. On x86_64 observed
- * improvement was ~75%, much closer to optimal, but the fact of
- * deviation means that references to pre-computed tables end up on
- * critical path and as tables are pretty big, 4KB per key+1KB shared,
- * execution time is sensitive to cache timing. It's not actually
- * proven, but 4-bit procedure is believed to provide adequate
- * all-round performance...
- */
+ * "4-bit" one. For gcc-generated x86[_64] code, "8-bit" was observed to
+ * run ~75% faster, closer to 100% for commercial compilers... But the
+ * catch is that "8-bit" procedure consumes 16 times more memory, 4KB
+ * per indivudual key + 1KB shared, and as access to these tables end up
+ * on critical path, real-life execution time would be sensitive to
+ * cache timing. It's not actually proven, but "4-bit" procedure is
+ * believed to provide adequate all-round performance...
+ */
#define TABLE_BITS 4
#if TABLE_BITS==8
V.lo = H[1];
for (Htable[128]=V, i=64; i>0; i>>=1) {
- if (sizeof(size_t)==8) {
- u64 T = U64(0xe100000000000000) & (0-(V.lo&1));
- V.lo = (V.hi<<63)|(V.lo>>1);
- V.hi = (V.hi>>1 )^T;
- }
- else {
- u32 T = 0xe1000000U & (0-(u32)(V.lo&1));
- V.lo = (V.hi<<63)|(V.lo>>1);
- V.hi = (V.hi>>1 )^((u64)T<<32);
- }
+ REDUCE1BIT(V);
Htable[i] = V;
}
#if defined(OPENSSL_SMALL_FOOTPRINT)
int i;
#endif
-#define REDUCE(V) do { \
- if (sizeof(size_t)==8) { \
- u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \
- V.lo = (V.hi<<63)|(V.lo>>1); \
- V.hi = (V.hi>>1 )^T; \
- } \
- else { \
- u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \
- V.lo = (V.hi<<63)|(V.lo>>1); \
- V.hi = (V.hi>>1 )^((u64)T<<32); \
- } \
-} while(0)
Htable[0].hi = 0;
Htable[0].lo = 0;
#if defined(OPENSSL_SMALL_FOOTPRINT)
for (Htable[8]=V, i=4; i>0; i>>=1) {
- REDUCE(V);
+ REDUCE1BIT(V);
Htable[i] = V;
}
}
#else
Htable[8] = V;
- REDUCE(V);
+ REDUCE1BIT(V);
Htable[4] = V;
- REDUCE(V);
+ REDUCE1BIT(V);
Htable[2] = V;
- REDUCE(V);
+ REDUCE1BIT(V);
Htable[1] = V;
Htable[3].hi = V.hi^Htable[2].hi, Htable[3].lo = V.lo^Htable[2].lo;
V=Htable[4];
}
}
#endif
-#undef REDUCE
}
#ifndef GHASH_ASM
#define GCM_MUL(ctx,Xi) gcm_gmult_4bit(ctx->Xi.u,ctx->Htable)
#if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT)
-#define GHASH(in,len,ctx) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len)
+#define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len)
/* GHASH_CHUNK is "stride parameter" missioned to mitigate cache
* trashing effect. In other words idea is to hash data while it's
* still in L1 cache after encryption pass... */
Z.hi ^= V.hi&M;
Z.lo ^= V.lo&M;
- if (sizeof(size_t)==8) {
- u64 T = U64(0xe100000000000000) & (0-(V.lo&1));
- V.lo = (V.hi<<63)|(V.lo>>1);
- V.hi = (V.hi>>1 )^T;
- }
- else {
- u32 T = 0xe1000000U & (0-(u32)(V.lo&1));
- V.lo = (V.hi<<63)|(V.lo>>1);
- V.hi = (V.hi>>1 )^((u64)T<<32);
- }
-
+ REDUCE1BIT(V);
}
}
u128 Htable[256];
#else
u128 Htable[16];
+ void (*gmult)(u64 Xi[2],const u128 Htable[16]);
+ void (*ghash)(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
#endif
unsigned int res, pad;
block128_f block;
void *key;
};
+#if TABLE_BITS==4 && defined(GHASH_ASM) && !defined(I386_ONLY) && \
+ (defined(__i386) || defined(__i386__) || \
+ defined(__x86_64) || defined(__x86_64__) || \
+ defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64))
+# define GHASH_ASM_IAX
+extern unsigned int OPENSSL_ia32cap_P[2];
+
+void gcm_init_clmul(u128 Htable[16],const u64 Xi[2]);
+void gcm_gmult_clmul(u64 Xi[2],const u128 Htable[16]);
+void gcm_ghash_clmul(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
+
+# if defined(__i386) || defined(__i386__) || defined(_M_IX86)
+# define GHASH_ASM_X86
+void gcm_gmult_4bit_mmx(u64 Xi[2],const u128 Htable[16]);
+void gcm_ghash_4bit_mmx(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
+
+void gcm_gmult_4bit_x86(u64 Xi[2],const u128 Htable[16]);
+void gcm_ghash_4bit_x86(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
+# endif
+
+# undef GCM_MUL
+# define GCM_MUL(ctx,Xi) (*((ctx)->gmult))(ctx->Xi.u,ctx->Htable)
+# undef GHASH
+# define GHASH(ctx,in,len) (*((ctx)->ghash))((ctx)->Xi.u,(ctx)->Htable,in,len)
+#endif
+
void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx,void *key,block128_f block)
{
const union { long one; char little; } is_endian = {1};
#if TABLE_BITS==8
gcm_init_8bit(ctx->Htable,ctx->H.u);
#elif TABLE_BITS==4
+# if defined(GHASH_ASM_IAX)
+ if (OPENSSL_ia32cap_P[1]&(1<<1)) {
+ gcm_init_clmul(ctx->Htable,ctx->H.u);
+ ctx->gmult = gcm_gmult_clmul;
+ ctx->ghash = gcm_ghash_clmul;
+ return;
+ }
gcm_init_4bit(ctx->Htable,ctx->H.u);
+# if defined(GHASH_ASM_X86)
+ if (OPENSSL_ia32cap_P[0]&(1<<23)) {
+ ctx->gmult = gcm_gmult_4bit_mmx;
+ ctx->ghash = gcm_ghash_4bit_mmx;
+ } else {
+ ctx->gmult = gcm_gmult_4bit_x86;
+ ctx->ghash = gcm_ghash_4bit_x86;
+ }
+# else
+ ctx->gmult = gcm_gmult_4bit;
+ ctx->ghash = gcm_ghash_4bit;
+# endif
+# else
+ gcm_init_4bit(ctx->Htable,ctx->H.u);
+# endif
#endif
}
#ifdef GHASH
if ((i = (len&(size_t)-16))) {
- GHASH(aad,i,ctx);
+ GHASH(ctx,aad,i);
aad += i;
len -= i;
}
in += 16;
j -= 16;
}
- GHASH(out-GHASH_CHUNK,GHASH_CHUNK,ctx);
+ GHASH(ctx,out-GHASH_CHUNK,GHASH_CHUNK);
len -= GHASH_CHUNK;
}
if ((i = (len&(size_t)-16))) {
in += 16;
len -= 16;
}
- GHASH(out-j,j,ctx);
+ GHASH(ctx,out-j,j);
}
#else
while (len>=16) {
while (len>=GHASH_CHUNK) {
size_t j=GHASH_CHUNK;
- GHASH(in,GHASH_CHUNK,ctx);
+ GHASH(ctx,in,GHASH_CHUNK);
while (j) {
(*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
++ctr;
len -= GHASH_CHUNK;
}
if ((i = (len&(size_t)-16))) {
- GHASH(in,i,ctx);
+ GHASH(ctx,in,i);
while (len>=16) {
(*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
++ctr;
{
size_t start,stop,gcm_t,ctr_t,OPENSSL_rdtsc();
union { u64 u; u8 c[1024]; } buf;
+ int i;
AES_set_encrypt_key(K1,sizeof(K1)*8,&key);
CRYPTO_gcm128_init(&ctx,&key,(block128_f)AES_encrypt);
ctr_t/(double)sizeof(buf),
(gcm_t-ctr_t)/(double)sizeof(buf));
#ifdef GHASH
- GHASH(buf.c,sizeof(buf),&ctx);
+ GHASH(&ctx,buf.c,sizeof(buf));
start = OPENSSL_rdtsc();
- GHASH(buf.c,sizeof(buf),&ctx);
+ for (i=0;i<100;++i) GHASH(&ctx,buf.c,sizeof(buf));
gcm_t = OPENSSL_rdtsc() - start;
- printf("%.2f\n",gcm_t/(double)sizeof(buf));
+ printf("%.2f\n",gcm_t/(double)sizeof(buf)/(double)i);
#endif
}
#endif
projects / oweals / openssl.git / commitdiff