move check for AD*X to rsaz-avx2.pl.

This ensures high performance is situations when assembler supports
AVX2, but not AD*X.

(Imported from upstream's 82a9dafe32)

Change-Id: Ie67f49a1c5467807139b6a8a0d4e62162d8a974f

crypto/bn/asm/rsaz-avx2.pl

@@ -78,16 +78,19 @@ die "can't locate x86_64-xlate.pl";
 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
 		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
 	$avx = ($1>=2.19) + ($1>=2.22);
+	$addx = ($1>=2.23);
 }
 
 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
 	    `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
 	$avx = ($1>=2.09) + ($1>=2.10);
+	$addx = ($1>=2.10);
 }
 
 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
 	    `ml64 2>&1` =~ /Version ([0-9]+)\./) {
 	$avx = ($1>=10) + ($1>=11);
+	$addx = ($1>=11);
 }
 
 open OUT,"| $^X $xlate $flavour $output";
@@ -1673,6 +1676,15 @@ $code.=<<___;
 .align	32
 rsaz_avx2_eligible:
 	mov	OPENSSL_ia32cap_P+8(%rip),%eax
+___
+$code.=<<___	if ($addx);
+	mov	\$`1<<8|1<<19`,%ecx
+	mov	\$0,%edx
+	and	%eax,%ecx
+	cmp	\$`1<<8|1<<19`,%ecx	# check for BMI2+AD*X
+	cmove	%edx,%eax
+___
+$code.=<<___;
 	and	\$`1<<5`,%eax
 	shr	\$5,%eax
 	ret

crypto/bn/exponentiation.c

@@ -890,8 +890,7 @@ int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
   /* If the size of the operands allow it, perform the optimized
    * RSAZ exponentiation. For further information see
    * crypto/bn/rsaz_exp.c and accompanying assembly modules. */
-  if (((OPENSSL_ia32cap_P[2] & 0x80100) != 0x80100) /* check for MULX/AD*X */
-      && (16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) &&
+  if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) &&
       rsaz_avx2_eligible()) {
     if (NULL == bn_wexpand(rr, 16))
       goto err;