SHA-1 for x86_64.

Configure

@@ -118,7 +118,7 @@ my $x86_elf_asm="x86cpuid-elf.o:bn86-elf.o co86-elf.o mo86-elf.o:dx86-elf.o yx86
 my $x86_coff_asm="x86cpuid-cof.o:bn86-cof.o co86-cof.o mo86-cof.o:dx86-cof.o yx86-cof.o:ax86-cof.o:bx86-cof.o:mx86-cof.o:sx86-cof.o s512sse2-cof.o:cx86-cof.o:rx86-cof.o:rm86-cof.o:r586-cof.o:wp_block.o w86mmx-cof.o";
 my $x86_out_asm="x86cpuid-out.o:bn86-out.o co86-out.o mo86-out.o:dx86-out.o yx86-out.o:ax86-out.o:bx86-out.o:mx86-out.o:sx86-out.o s512sse2-out.o:cx86-out.o:rx86-out.o:rm86-out.o:r586-out.o:wp_block.o w86mmx-out.o";
 
-my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o";
+my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o::aes-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o:::wp-x86_64.o";
 my $ia64_asm="ia64cpuid.o:bn-ia64.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o:::";
 my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o:des_enc-sparc.o fcrypt_b.o:aes_core.o aes_cbc.o aes-sparcv9.o::md5-sparcv9.o::::::";
 my $no_asm=":::::::::::";

crypto/sha/Makefile

@@ -64,6 +64,8 @@ sha256-ia64.s: asm/sha512-ia64.pl
 	(cd asm; $(PERL) sha512-ia64.pl ../$@ $(CFLAGS))
 sha512-ia64.s: asm/sha512-ia64.pl
 	(cd asm; $(PERL) sha512-ia64.pl ../$@ $(CFLAGS))
+sha1-x86_64.s: asm/sha1-x86_64.pl
+	$(PERL) asm/sha1-x86_64.pl $@
 sha256-x86_64.s: asm/sha512-x86_64.pl
 	$(PERL) asm/sha512-x86_64.pl $@
 sha512-x86_64.s: asm/sha512-x86_64.pl

crypto/sha/asm/sha1-x86_64.pl

+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. Rights for redistribution and usage in source and binary
+# forms are granted according to the OpenSSL license.
+# ====================================================================
+#
+# sha1_block procedure for x86_64.
+#
+# It was brought to my attention that on EM64T compiler-generated code
+# was far behind 32-bit assembler implementation. This is unlike on
+# Opteron where compiler-generated code was only 15% behind 32-bit
+# assembler, which originally made it hard to motivate the effort.
+# There was suggestion to mechanically translate 32-bit code, but I
+# dismissed it, reasoning that x86_64 offers enough register bank
+# capacity to fully utilize SHA-1 parallelism. Therefore this fresh
+# implementation:-) However! While 64-bit code does performs better
+# on Opteron, I failed to beat 32-bit assembler on EM64T core. Well,
+# x86_64 does offer larger *addressable* bank, but out-of-order core
+# reaches for even more registers through dynamic aliasing, and EM64T
+# core must have managed to run-time optimize even 32-bit code just as
+# good as 64-bit one. Performance improvement is summarized in the
+# following table:
+#
+#		gcc 3.4		32-bit asm	cycles/byte
+# Opteron	+45%		+20%		6.8
+# Xeon		+65%		+0%		9.9
+
+$output=shift;
+open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
+
+$ctx="%rdi";	# 1st arg
+$inp="%rsi";	# 2nd arg
+$num="%rdx";	# 3rd arg
+
+# reassign arguments in order to produce more compact code
+$ctx="%r8";
+$inp="%r9";
+$num="%r10";
+
+$xi="%eax";
+$t0="%ebx";
+$t1="%ecx";
+$A="%edx";
+$B="%esi";
+$C="%edi";
+$D="%ebp";
+$E="%r11d";
+$T="%r12d";
+
+@V=($A,$B,$C,$D,$E,$T);
+
+sub PROLOGUE {
+my $func=shift;
+$code.=<<___;
+.globl	$func
+.type	$func,\@function,3
+.align	16
+$func:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	mov	%rsp,%rax
+	mov	%rdi,$ctx	# reassigned argument
+	sub	\$`8+16*4`,%rsp
+	mov	%rsi,$inp	# reassigned argument
+	and	\$-64,%rsp
+	mov	%rdx,$num	# reassigned argument
+	mov	%rax,`16*4`(%rsp)
+
+	mov	0($ctx),$A
+	mov	4($ctx),$B
+	mov	8($ctx),$C
+	mov	12($ctx),$D
+	mov	16($ctx),$E
+___
+}
+
+sub EPILOGUE {
+my $func=shift;
+$code.=<<___;
+	mov	`16*4`(%rsp),%rsp
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	ret
+.size	$func,.-$func
+___
+}
+
+sub BODY_00_19 {
+my ($i,$a,$b,$c,$d,$e,$f,$host)=@_;
+my $j=$i+1;
+$code.=<<___ if ($i==0);
+	mov	`4*$i`($inp),$xi	
+	`"bswap	$xi"	if(!defined($host))`
+	mov	$xi,`4*$i`(%rsp)
+___
+$code.=<<___ if ($i<15);
+	lea	0x5a827999($xi,$e),$f
+	mov	$c,$t0
+	mov	`4*$j`($inp),$xi
+	mov	$a,$e
+	xor	$d,$t0
+	`"bswap	$xi"	if(!defined($host))`	
+	rol	\$5,$e
+	and	$b,$t0
+	mov	$xi,`4*$j`(%rsp)
+	add	$e,$f
+	xor	$d,$t0
+	rol	\$30,$b
+	add	$t0,$f
+___
+$code.=".Lshortcut:\n" if ($i==15);
+$code.=<<___ if ($i>=15);
+	lea	0x5a827999($xi,$e),$f
+	mov	`4*($j%16)`(%rsp),$xi
+	mov	$c,$t0
+	mov	$a,$e
+	xor	`4*(($j+2)%16)`(%rsp),$xi
+	xor	$d,$t0
+	rol	\$5,$e
+	xor	`4*(($j+8)%16)`(%rsp),$xi
+	and	$b,$t0
+	add	$e,$f
+	xor	`4*(($j+13)%16)`(%rsp),$xi
+	xor	$d,$t0
+	rol	\$30,$b
+	add	$t0,$f
+	rol	\$1,$xi
+	mov	$xi,`4*($j%16)`(%rsp)
+___
+}
+
+sub BODY_20_39 {
+my ($i,$a,$b,$c,$d,$e,$f)=@_;
+my $j=$i+1;
+my $K=($i<40)?0x6ed9eba1:0xca62c1d6;
+$code.=<<___ if ($i<79);
+	lea	$K($xi,$e),$f
+	mov	`4*($j%16)`(%rsp),$xi
+	mov	$c,$t0
+	mov	$a,$e
+	xor	`4*(($j+2)%16)`(%rsp),$xi
+	xor	$b,$t0
+	rol	\$5,$e
+	xor	`4*(($j+8)%16)`(%rsp),$xi
+	xor	$d,$t0
+	add	$e,$f
+	xor	`4*(($j+13)%16)`(%rsp),$xi
+	rol	\$30,$b
+	add	$t0,$f
+	rol	\$1,$xi
+	mov	$xi,`4*($j%16)`(%rsp)
+___
+$code.=<<___ if ($i==79);
+	lea	$K($xi,$e),$f
+	mov	$c,$t0
+	mov	$a,$e
+	xor	$b,$t0
+	rol	\$5,$e
+	xor	$d,$t0
+	add	$e,$f
+	rol	\$30,$b
+	add	$t0,$f
+___
+}
+
+sub BODY_40_59 {
+my ($i,$a,$b,$c,$d,$e,$f)=@_;
+my $j=$i+1;
+$code.=<<___;
+	lea	0x8f1bbcdc($xi,$e),$f
+	mov	`4*($j%16)`(%rsp),$xi
+	mov	$b,$t0
+	mov	$b,$t1
+	xor	`4*(($j+2)%16)`(%rsp),$xi
+	mov	$a,$e
+	and	$c,$t0
+	xor	`4*(($j+8)%16)`(%rsp),$xi
+	or	$c,$t1
+	rol	\$5,$e
+	xor	`4*(($j+13)%16)`(%rsp),$xi
+	and	$d,$t1
+	add	$e,$f
+	rol	\$1,$xi
+	or	$t1,$t0
+	rol	\$30,$b
+	mov	$xi,`4*($j%16)`(%rsp)
+	add	$t0,$f
+___
+}
+
+$code=".text\n";
+
+&PROLOGUE("sha1_block_asm_data_order");
+$code.=".align	4\n.Lloop:\n";
+for($i=0;$i<20;$i++)	{ &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
+for(;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+for(;$i<60;$i++)	{ &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
+for(;$i<80;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+	add	0($ctx),$E
+	add	4($ctx),$T
+	add	8($ctx),$A
+	add	12($ctx),$B
+	add	16($ctx),$C
+	mov	$E,0($ctx)
+	mov	$T,4($ctx)
+	mov	$A,8($ctx)
+	mov	$B,12($ctx)
+	mov	$C,16($ctx)
+
+	xchg	$E,$A	# mov	$E,$A
+	xchg	$T,$B	# mov	$T,$B
+	xchg	$E,$C	# mov	$A,$C
+	xchg	$T,$D	# mov	$B,$D
+			# mov	$C,$E
+	lea	`16*4`($inp),$inp
+	sub	\$1,$num
+	jnz	.Lloop
+___
+&EPILOGUE("sha1_block_asm_data_order");
+
+####################################################################
+
+@V=($A,$B,$C,$D,$E,$T);
+
+&PROLOGUE("sha1_block_asm_host_order");
+for($i=0;$i<15;$i++)	{ &BODY_00_19($i,@V,1); unshift(@V,pop(@V)); }
+$code.=<<___;
+	jmp	.Lshortcut
+.size	sha1_block_asm_host_order,.-sha1_block_asm_host_order
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+print $code;
+close STDOUT;

crypto/sha/sha_locl.h

@@ -115,7 +115,8 @@
 # endif
 
 # ifdef SHA1_ASM
-#  if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__)
+#  if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) \
+   || defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64)
 #   define sha1_block_host_order		sha1_block_asm_host_order
 #   define DONT_IMPLEMENT_BLOCK_HOST_ORDER
 #   define sha1_block_data_order		sha1_block_asm_data_order