x86_64 assembly pack: improve masm support.

6 files changed, 23 insertions, 13 deletions

diff --git a/crypto/aes/asm/aesni-sha256-x86_64.pl b/crypto/aes/asm/aesni-sha256-x86_64.pl
index 6abb5e542e..c1fce89834 100644
--- a/crypto/aes/asm/aesni-sha256-x86_64.pl
+++ b/crypto/aes/asm/aesni-sha256-x86_64.pl
@@ -56,7 +56,7 @@ if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 
 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	   `ml64 2>&1` =~ /Version ([0-9]+)\./) {
-	$avx = ($1>=10) + ($1>=11);
+	$avx = ($1>=10) + ($1>=12);
 }
 
 if (!$avx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/) {
diff --git a/crypto/bn/asm/rsaz-x86_64.pl b/crypto/bn/asm/rsaz-x86_64.pl
index cd651f7e78..3a4f0c59a5 100755
--- a/crypto/bn/asm/rsaz-x86_64.pl
+++ b/crypto/bn/asm/rsaz-x86_64.pl
@@ -110,7 +110,7 @@ if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 
 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	    `ml64 2>&1` =~ /Version ([0-9]+)\./) {
-	$addx = ($1>=11);
+	$addx = ($1>=12);
 }
 
 if (!$addx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9])\.([0-9]+)/) {
diff --git a/crypto/bn/asm/x86_64-mont.pl b/crypto/bn/asm/x86_64-mont.pl
index 38039286be..2989b58f25 100755
--- a/crypto/bn/asm/x86_64-mont.pl
+++ b/crypto/bn/asm/x86_64-mont.pl
@@ -65,7 +65,7 @@ if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 
 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	    `ml64 2>&1` =~ /Version ([0-9]+)\./) {
-	$addx = ($1>=11);
+	$addx = ($1>=12);
 }
 
 # int bn_mul_mont(
@@ -742,9 +742,11 @@ my @A0=("%r10","%r11");
 my @A1=("%r12","%r13");
 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
 
+$code.=<<___	if ($addx);
+.extern	bn_sqrx8x_internal		# see x86_64-mont5 module
+___
 $code.=<<___;
 .extern	bn_sqr8x_internal		# see x86_64-mont5 module
-.extern	bn_sqrx8x_internal		# see x86_64-mont5 module
 
 .type	bn_sqr8x_mont,\@function,6
 .align	32
diff --git a/crypto/bn/asm/x86_64-mont5.pl b/crypto/bn/asm/x86_64-mont5.pl
index c107df95f5..fa22c30b1f 100755
--- a/crypto/bn/asm/x86_64-mont5.pl
+++ b/crypto/bn/asm/x86_64-mont5.pl
@@ -50,7 +50,7 @@ if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 
 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	    `ml64 2>&1` =~ /Version ([0-9]+)\./) {
-	$addx = ($1>=11);
+	$addx = ($1>=12);
 }
 
 # int bn_mul_mont_gather5(
diff --git a/crypto/modes/asm/aesni-gcm-x86_64.pl b/crypto/modes/asm/aesni-gcm-x86_64.pl
index 4c423b2303..cfc856cf35 100644
--- a/crypto/modes/asm/aesni-gcm-x86_64.pl
+++ b/crypto/modes/asm/aesni-gcm-x86_64.pl
@@ -92,7 +92,7 @@ _aesni_ctr32_ghash_6x:
 
 .align	32
 .Loop6x:
-	add		\$6<<24,$counter
+	add		\$`6<<24`,$counter
 	jc		.Lhandle_ctr32		# discard $inout[1-5]?
 	vmovdqu		0x00-0x20($Xip),$Hkey	# $Hkey^1
 	  vpaddb	$T2,$inout5,$T1		# next counter value
@@ -520,7 +520,7 @@ _aesni_ctr32_6x:
 	vmovups		0x10-0x80($key),$rndkey
 	lea		0x20-0x80($key),%r12
 	vpxor		$Z0,$T1,$inout0
-	add		\$6<<24,$counter
+	add		\$`6<<24`,$counter
 	jc		.Lhandle_ctr32_2
 	vpaddb		$T2,$T1,$inout1
 	vpaddb		$T2,$inout1,$inout2
diff --git a/crypto/perlasm/x86_64-xlate.pl b/crypto/perlasm/x86_64-xlate.pl
index 4dd7f5252d..84745f600a 100755
--- a/crypto/perlasm/x86_64-xlate.pl
+++ b/crypto/perlasm/x86_64-xlate.pl
@@ -270,15 +270,20 @@ my %globals;
 		sprintf "%s%s(%%%s)",	$self->{asterisk},$self->{label},$self->{base};
 	    }
 	} else {
-	    %szmap = (	b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR",
-	    		q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR",
-			y=>"" );
+	    %szmap = (	b=>"BYTE$PTR",  w=>"WORD$PTR",
+			l=>"DWORD$PTR", d=>"DWORD$PTR",
+	    		q=>"QWORD$PTR", o=>"OWORD$PTR",
+			x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" );
 
 	    $self->{label} =~ s/\./\$/g;
 	    $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
 	    $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
-	    $sz="q" if ($self->{asterisk} || opcode->mnemonic() =~ /^v?movq$/);
-	    $sz="l" if (opcode->mnemonic() =~ /^v?movd$/);
+
+	    ($self->{asterisk})					&& ($sz="q") ||
+	    (opcode->mnemonic() =~ /^v?mov([qd])$/)		&& ($sz=$1)  ||
+	    (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/)		&& ($sz=$1)  ||
+	    (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/)	&& ($sz=$1)  ||
+	    (opcode->mnemonic() =~ /^vinsert[fi]128$/)		&& ($sz="x");
 
 	    if (defined($self->{index})) {
 		sprintf "%s[%s%s*%d%s]",$szmap{$sz},
@@ -536,7 +541,7 @@ my %globals;
 					$v="$current_segment\tENDS\n" if ($current_segment);
 					$current_segment = ".text\$";
 					$v.="$current_segment\tSEGMENT ";
-					$v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE";
+					$v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
 					$v.=" 'CODE'";
 				    }
 				    $self->{value} = $v;
@@ -834,6 +839,8 @@ if ($nasm) {
     print <<___;
 default	rel
 %define XMMWORD
+%define YMMWORD
+%define ZMMWORD
 ___
 } elsif ($masm) {
     print <<___;
@@ -897,6 +904,7 @@ while($line=<>) {
 		    # $insn.=$sz compensates for movq, pinsrw, ...
 		    if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
 		    if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
+		    if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
 		    if ($arg =~ /^mm[0-9]+$/)  { $insn.=$sz; $sz="q" if(!$sz); last; }
 		}
 		@args = reverse(@args);