x86_64 assembler pack update from HEAD.

crypto/aes/asm/aes-x86_64.pl

@@ -588,6 +588,9 @@ $code.=<<___;
 .globl	AES_encrypt
 .type	AES_encrypt,\@function,3
 .align	16
+.globl	asm_AES_encrypt
+.hidden	asm_AES_encrypt
+asm_AES_encrypt:
 AES_encrypt:
 	push	%rbx
 	push	%rbp
@@ -1184,6 +1187,9 @@ $code.=<<___;
 .globl	AES_decrypt
 .type	AES_decrypt,\@function,3
 .align	16
+.globl	asm_AES_decrypt
+.hidden	asm_AES_decrypt
+asm_AES_decrypt:
 AES_decrypt:
 	push	%rbx
 	push	%rbp
@@ -1648,6 +1654,9 @@ $code.=<<___;
 .type	AES_cbc_encrypt,\@function,6
 .align	16
 .extern	OPENSSL_ia32cap_P
+.globl	asm_AES_cbc_encrypt
+.hidden	asm_AES_cbc_encrypt
+asm_AES_cbc_encrypt:
 AES_cbc_encrypt:
 	cmp	\$0,%rdx	# check length
 	je	.Lcbc_epilogue

crypto/bn/asm/x86_64-gf2m.pl

@@ -0,0 +1,389 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# May 2011
+#
+# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
+# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
+# the time being... Except that it has two code paths: code suitable
+# for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
+# later. Improvement varies from one benchmark and µ-arch to another.
+# Vanilla code path is at most 20% faster than compiler-generated code
+# [not very impressive], while PCLMULQDQ - whole 85%-160% better on
+# 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
+# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
+# all CPU time is burnt in it...
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour $output";
+
+($lo,$hi)=("%rax","%rdx");	$a=$lo;
+($i0,$i1)=("%rsi","%rdi");
+($t0,$t1)=("%rbx","%rcx");
+($b,$mask)=("%rbp","%r8");
+($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
+($R,$Tx)=("%xmm0","%xmm1");
+
+$code.=<<___;
+.text
+
+.type	_mul_1x1,\@abi-omnipotent
+.align	16
+_mul_1x1:
+	sub	\$128+8,%rsp
+	mov	\$-1,$a1
+	lea	($a,$a),$i0
+	shr	\$3,$a1
+	lea	(,$a,4),$i1
+	and	$a,$a1			# a1=a&0x1fffffffffffffff
+	lea	(,$a,8),$a8
+	sar	\$63,$a			# broadcast 63rd bit
+	lea	($a1,$a1),$a2
+	sar	\$63,$i0		# broadcast 62nd bit
+	lea	(,$a1,4),$a4
+	and	$b,$a
+	sar	\$63,$i1		# boardcast 61st bit
+	mov	$a,$hi			# $a is $lo
+	shl	\$63,$lo
+	and	$b,$i0
+	shr	\$1,$hi
+	mov	$i0,$t1
+	shl	\$62,$i0
+	and	$b,$i1
+	shr	\$2,$t1
+	xor	$i0,$lo
+	mov	$i1,$t0
+	shl	\$61,$i1
+	xor	$t1,$hi
+	shr	\$3,$t0
+	xor	$i1,$lo
+	xor	$t0,$hi
+
+	mov	$a1,$a12
+	movq	\$0,0(%rsp)		# tab[0]=0
+	xor	$a2,$a12		# a1^a2
+	mov	$a1,8(%rsp)		# tab[1]=a1
+	 mov	$a4,$a48
+	mov	$a2,16(%rsp)		# tab[2]=a2
+	 xor	$a8,$a48		# a4^a8
+	mov	$a12,24(%rsp)		# tab[3]=a1^a2
+
+	xor	$a4,$a1
+	mov	$a4,32(%rsp)		# tab[4]=a4
+	xor	$a4,$a2
+	mov	$a1,40(%rsp)		# tab[5]=a1^a4
+	xor	$a4,$a12
+	mov	$a2,48(%rsp)		# tab[6]=a2^a4
+	 xor	$a48,$a1		# a1^a4^a4^a8=a1^a8
+	mov	$a12,56(%rsp)		# tab[7]=a1^a2^a4
+	 xor	$a48,$a2		# a2^a4^a4^a8=a1^a8
+
+	mov	$a8,64(%rsp)		# tab[8]=a8
+	xor	$a48,$a12		# a1^a2^a4^a4^a8=a1^a2^a8
+	mov	$a1,72(%rsp)		# tab[9]=a1^a8
+	 xor	$a4,$a1			# a1^a8^a4
+	mov	$a2,80(%rsp)		# tab[10]=a2^a8
+	 xor	$a4,$a2			# a2^a8^a4
+	mov	$a12,88(%rsp)		# tab[11]=a1^a2^a8
+
+	xor	$a4,$a12		# a1^a2^a8^a4
+	mov	$a48,96(%rsp)		# tab[12]=a4^a8
+	 mov	$mask,$i0
+	mov	$a1,104(%rsp)		# tab[13]=a1^a4^a8
+	 and	$b,$i0
+	mov	$a2,112(%rsp)		# tab[14]=a2^a4^a8
+	 shr	\$4,$b
+	mov	$a12,120(%rsp)		# tab[15]=a1^a2^a4^a8
+	 mov	$mask,$i1
+	 and	$b,$i1
+	 shr	\$4,$b
+
+	movq	(%rsp,$i0,8),$R		# half of calculations is done in SSE2
+	mov	$mask,$i0
+	and	$b,$i0
+	shr	\$4,$b
+___
+    for ($n=1;$n<8;$n++) {
+	$code.=<<___;
+	mov	(%rsp,$i1,8),$t1
+	mov	$mask,$i1
+	mov	$t1,$t0
+	shl	\$`8*$n-4`,$t1
+	and	$b,$i1
+	 movq	(%rsp,$i0,8),$Tx
+	shr	\$`64-(8*$n-4)`,$t0
+	xor	$t1,$lo
+	 pslldq	\$$n,$Tx
+	 mov	$mask,$i0
+	shr	\$4,$b
+	xor	$t0,$hi
+	 and	$b,$i0
+	 shr	\$4,$b
+	 pxor	$Tx,$R
+___
+    }
+$code.=<<___;
+	mov	(%rsp,$i1,8),$t1
+	mov	$t1,$t0
+	shl	\$`8*$n-4`,$t1
+	movq	$R,$i0
+	shr	\$`64-(8*$n-4)`,$t0
+	xor	$t1,$lo
+	psrldq	\$8,$R
+	xor	$t0,$hi
+	movq	$R,$i1
+	xor	$i0,$lo
+	xor	$i1,$hi
+
+	add	\$128+8,%rsp
+	ret
+.Lend_mul_1x1:
+.size	_mul_1x1,.-_mul_1x1
+___
+
+($rp,$a1,$a0,$b1,$b0) = $win64?	("%rcx","%rdx","%r8", "%r9","%r10") :	# Win64 order
+				("%rdi","%rsi","%rdx","%rcx","%r8");	# Unix order
+
+$code.=<<___;
+.extern	OPENSSL_ia32cap_P
+.globl	bn_GF2m_mul_2x2
+.type	bn_GF2m_mul_2x2,\@abi-omnipotent
+.align	16
+bn_GF2m_mul_2x2:
+	mov	OPENSSL_ia32cap_P(%rip),%rax
+	bt	\$33,%rax
+	jnc	.Lvanilla_mul_2x2
+
+	movq		$a1,%xmm0
+	movq		$b1,%xmm1
+	movq		$a0,%xmm2
+___
+$code.=<<___ if ($win64);
+	movq		40(%rsp),%xmm3
+___
+$code.=<<___ if (!$win64);
+	movq		$b0,%xmm3
+___
+$code.=<<___;
+	movdqa		%xmm0,%xmm4
+	movdqa		%xmm1,%xmm5
+	pclmulqdq	\$0,%xmm1,%xmm0	# a1·b1
+	pxor		%xmm2,%xmm4
+	pxor		%xmm3,%xmm5
+	pclmulqdq	\$0,%xmm3,%xmm2	# a0·b0
+	pclmulqdq	\$0,%xmm5,%xmm4	# (a0+a1)·(b0+b1)
+	xorps		%xmm0,%xmm4
+	xorps		%xmm2,%xmm4	# (a0+a1)·(b0+b1)-a0·b0-a1·b1
+	movdqa		%xmm4,%xmm5
+	pslldq		\$8,%xmm4
+	psrldq		\$8,%xmm5
+	pxor		%xmm4,%xmm2
+	pxor		%xmm5,%xmm0
+	movdqu		%xmm2,0($rp)
+	movdqu		%xmm0,16($rp)
+	ret
+
+.align	16
+.Lvanilla_mul_2x2:
+	lea	-8*17(%rsp),%rsp
+___
+$code.=<<___ if ($win64);
+	mov	`8*17+40`(%rsp),$b0
+	mov	%rdi,8*15(%rsp)
+	mov	%rsi,8*16(%rsp)
+___
+$code.=<<___;
+	mov	%r14,8*10(%rsp)
+	mov	%r13,8*11(%rsp)
+	mov	%r12,8*12(%rsp)
+	mov	%rbp,8*13(%rsp)
+	mov	%rbx,8*14(%rsp)
+.Lbody_mul_2x2:
+	mov	$rp,32(%rsp)		# save the arguments
+	mov	$a1,40(%rsp)
+	mov	$a0,48(%rsp)
+	mov	$b1,56(%rsp)
+	mov	$b0,64(%rsp)
+
+	mov	\$0xf,$mask
+	mov	$a1,$a
+	mov	$b1,$b
+	call	_mul_1x1		# a1·b1
+	mov	$lo,16(%rsp)
+	mov	$hi,24(%rsp)
+
+	mov	48(%rsp),$a
+	mov	64(%rsp),$b
+	call	_mul_1x1		# a0·b0
+	mov	$lo,0(%rsp)
+	mov	$hi,8(%rsp)
+
+	mov	40(%rsp),$a
+	mov	56(%rsp),$b
+	xor	48(%rsp),$a
+	xor	64(%rsp),$b
+	call	_mul_1x1		# (a0+a1)·(b0+b1)
+___
+	@r=("%rbx","%rcx","%rdi","%rsi");
+$code.=<<___;
+	mov	0(%rsp),@r[0]
+	mov	8(%rsp),@r[1]
+	mov	16(%rsp),@r[2]
+	mov	24(%rsp),@r[3]
+	mov	32(%rsp),%rbp
+
+	xor	$hi,$lo
+	xor	@r[1],$hi
+	xor	@r[0],$lo
+	mov	@r[0],0(%rbp)
+	xor	@r[2],$hi
+	mov	@r[3],24(%rbp)
+	xor	@r[3],$lo
+	xor	@r[3],$hi
+	xor	$hi,$lo
+	mov	$hi,16(%rbp)
+	mov	$lo,8(%rbp)
+
+	mov	8*10(%rsp),%r14
+	mov	8*11(%rsp),%r13
+	mov	8*12(%rsp),%r12
+	mov	8*13(%rsp),%rbp
+	mov	8*14(%rsp),%rbx
+___
+$code.=<<___ if ($win64);
+	mov	8*15(%rsp),%rdi
+	mov	8*16(%rsp),%rsi
+___
+$code.=<<___;
+	lea	8*17(%rsp),%rsp
+	ret
+.Lend_mul_2x2:
+.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
+.asciz	"GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align	16
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#               CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern __imp_RtlVirtualUnwind
+
+.type	se_handler,\@abi-omnipotent
+.align	16
+se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	152($context),%rax	# pull context->Rsp
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lbody_mul_2x2(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<"prologue" label
+	jb	.Lin_prologue
+
+	mov	8*10(%rax),%r14		# mimic epilogue
+	mov	8*11(%rax),%r13
+	mov	8*12(%rax),%r12
+	mov	8*13(%rax),%rbp
+	mov	8*14(%rax),%rbx
+	mov	8*15(%rax),%rdi
+	mov	8*16(%rax),%rsi
+
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+
+.Lin_prologue:
+	lea	8*17(%rax),%rax
+	mov	%rax,152($context)	# restore context->Rsp
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	se_handler,.-se_handler
+
+.section	.pdata
+.align	4
+	.rva	_mul_1x1
+	.rva	.Lend_mul_1x1
+	.rva	.LSEH_info_1x1
+
+	.rva	.Lvanilla_mul_2x2
+	.rva	.Lend_mul_2x2
+	.rva	.LSEH_info_2x2
+.section	.xdata
+.align	8
+.LSEH_info_1x1:
+	.byte	0x01,0x07,0x02,0x00
+	.byte	0x07,0x01,0x11,0x00	# sub rsp,128+8
+.LSEH_info_2x2:
+	.byte	9,0,0,0
+	.rva	se_handler
+___
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+print $code;
+close STDOUT;

crypto/sha/asm/sha512-x86_64.pl

@@ -95,50 +95,44 @@ sub ROUND_00_15()
 { my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
 
 $code.=<<___;
-	mov	$e,$a0
+	ror	\$`$Sigma1[2]-$Sigma1[1]`,$a0
-	mov	$e,$a1
 	mov	$f,$a2
-
-	ror	\$$Sigma1[0],$a0
-	ror	\$$Sigma1[1],$a1
-	xor	$g,$a2			# f^g
-
-	xor	$a1,$a0
-	ror	\$`$Sigma1[2]-$Sigma1[1]`,$a1
-	and	$e,$a2			# (f^g)&e
 	mov	$T1,`$SZ*($i&0xf)`(%rsp)
 
-	xor	$a1,$a0			# Sigma1(e)
+	ror	\$`$Sigma0[2]-$Sigma0[1]`,$a1
-	xor	$g,$a2			# Ch(e,f,g)=((f^g)&e)^g
+	xor	$e,$a0
+	xor	$g,$a2			# f^g
+
+	ror	\$`$Sigma1[1]-$Sigma1[0]`,$a0
 	add	$h,$T1			# T1+=h
+	xor	$a,$a1
 
-	mov	$a,$h
-	add	$a0,$T1			# T1+=Sigma1(e)
-
-	add	$a2,$T1			# T1+=Ch(e,f,g)
-	mov	$a,$a0
-	mov	$a,$a1
-
-	ror	\$$Sigma0[0],$h
-	ror	\$$Sigma0[1],$a0
-	mov	$a,$a2
 	add	($Tbl,$round,$SZ),$T1	# T1+=K[round]
+	and	$e,$a2			# (f^g)&e
+	mov	$b,$h
 
-	xor	$a0,$h
+	ror	\$`$Sigma0[1]-$Sigma0[0]`,$a1
-	ror	\$`$Sigma0[2]-$Sigma0[1]`,$a0
+	xor	$e,$a0
-	or	$c,$a1			# a|c
+	xor	$g,$a2			# Ch(e,f,g)=((f^g)&e)^g
+
+	xor	$c,$h			# b^c
+	xor	$a,$a1
+	add	$a2,$T1			# T1+=Ch(e,f,g)
+	mov	$b,$a2
+
+	ror	\$$Sigma1[0],$a0	# Sigma1(e)
+	and	$a,$h			# h=(b^c)&a
+	and	$c,$a2			# b&c
+
+	ror	\$$Sigma0[0],$a1	# Sigma0(a)
+	add	$a0,$T1			# T1+=Sigma1(e)
+	add	$a2,$h			# h+=b&c (completes +=Maj(a,b,c)
 
-	xor	$a0,$h			# h=Sigma0(a)
-	and	$c,$a2			# a&c
 	add	$T1,$d			# d+=T1
-
-	and	$b,$a1			# (a|c)&b
 	add	$T1,$h			# h+=T1
-
-	or	$a2,$a1			# Maj(a,b,c)=((a|c)&b)|(a&c)
 	lea	1($round),$round	# round++
+	add	$a1,$h			# h+=Sigma0(a)
 
-	add	$a1,$h			# h+=Maj(a,b,c)
 ___
 }
 
@@ -147,32 +141,30 @@ sub ROUND_16_XX()
 
 $code.=<<___;
 	mov	`$SZ*(($i+1)&0xf)`(%rsp),$a0
-	mov	`$SZ*(($i+14)&0xf)`(%rsp),$T1
+	mov	`$SZ*(($i+14)&0xf)`(%rsp),$a1
-
+	mov	$a0,$T1
-	mov	$a0,$a2
+	mov	$a1,$a2
 
+	ror	\$`$sigma0[1]-$sigma0[0]`,$T1
+	xor	$a0,$T1
 	shr	\$$sigma0[2],$a0
-	ror	\$$sigma0[0],$a2
 
-	xor	$a2,$a0
+	ror	\$$sigma0[0],$T1
-	ror	\$`$sigma0[1]-$sigma0[0]`,$a2
+	xor	$T1,$a0			# sigma0(X[(i+1)&0xf])
+	mov	`$SZ*(($i+9)&0xf)`(%rsp),$T1
 
-	xor	$a2,$a0			# sigma0(X[(i+1)&0xf])
+	ror	\$`$sigma1[1]-$sigma1[0]`,$a2
-	mov	$T1,$a1
+	xor	$a1,$a2
-
+	shr	\$$sigma1[2],$a1
-	shr	\$$sigma1[2],$T1
-	ror	\$$sigma1[0],$a1
-
-	xor	$a1,$T1
-	ror	\$`$sigma1[1]-$sigma1[0]`,$a1
-
-	xor	$a1,$T1			# sigma1(X[(i+14)&0xf])
 
+	ror	\$$sigma1[0],$a2
 	add	$a0,$T1
-
+	xor	$a2,$a1			# sigma1(X[(i+14)&0xf])
-	add	`$SZ*(($i+9)&0xf)`(%rsp),$T1
 
 	add	`$SZ*($i&0xf)`(%rsp),$T1
+	mov	$e,$a0
+	add	$a1,$T1
+	mov	$a,$a1
 ___
 	&ROUND_00_15(@_);
 }
@@ -219,6 +211,8 @@ $func:
 ___
 	for($i=0;$i<16;$i++) {
 		$code.="	mov	$SZ*$i($inp),$T1\n";
+		$code.="	mov	@ROT[4],$a0\n";
+		$code.="	mov	@ROT[0],$a1\n";
 		$code.="	bswap	$T1\n";
 		&ROUND_00_15($i,@ROT);
 		unshift(@ROT,pop(@ROT));

crypto/x86_64cpuid.pl

@@ -23,7 +23,7 @@ print<<___;
 	call	OPENSSL_cpuid_setup
 
 .hidden	OPENSSL_ia32cap_P
-.comm	OPENSSL_ia32cap_P,8
+.comm	OPENSSL_ia32cap_P,8,4
 
 .text