_BitScanForward on Visual C <2005 support

Makefile.PL

@@ -1,4 +1,5 @@
 use ExtUtils::MakeMaker;
+use Config;
 
 my $broken64 = (18446744073709550592 == ~0);
 if ($broken64) {
@@ -15,6 +16,24 @@ I highly recommend upgrading to a newer version of Perl.
 EOW
 }
 
+my $msvc_Oi = !!0;
+{
+    my ($cc, $ver) = $Config{cc};
+    if ($cc eq 'cl') {
+        my @ver = `$cc 2>&1`;    # Interesting output in STDERR
+        $ver = join('', @ver);
+
+        #print 'VER:'.$ver.':'."\n";
+        if ($ver =~ /Version (\d[\d\.]+)/ms) {
+            $ver = $1;
+        }
+        else {
+            die "Unknown version of Microsoft Visual C";
+        }
+        $msvc_Oi = '-Oi' if $ver < 14 && $ver >= 13; #VC2002/2003 only
+    }
+}
+
 WriteMakefile1(
     NAME         => 'Math::Prime::Util',
     ABSTRACT     => 'Utilities related to prime numbers, including fast sieves and factoring',
@@ -81,6 +100,7 @@ WriteMakefile1(
              },
 
     MIN_PERL_VERSION => 5.006002,
+    DEFINE => $msvc_Oi,
 );
 
 sub WriteMakefile1 {   # Cribbed from eumm-upgrade by Alexandr Ciornii

util.c

@@ -147,6 +147,32 @@ static UV count_zero_bits(const unsigned char* m, UV nbytes)
   return count;
 }
 
+#if defined(_MSC_VER) && _MSC_VER < 1300
+/* used for testing
+ unsigned char __fastcall _2BitScanForward(unsigned long* Index, unsigned long Mask) {
+    return _BitScanForward(Index,Mask);
+}
+unsigned char __fastcall _2BitScanReverse(unsigned long* Index, unsigned long Mask) {
+    return _BitScanReverse(Index,Mask);
+}*/
+
+unsigned char __declspec( naked ) __fastcall _BitScanForward(unsigned long* Index, unsigned long Mask) {
+    __asm {
+        bsf     eax, edx /* edx is Mask */
+        mov     [ecx], eax /*ecx is Index */
+        setnz   al
+        retn
+    };
+}
+unsigned char __declspec( naked ) __fastcall _BitScanReverse(unsigned long* Index, unsigned long Mask) {
+    __asm {
+        bsr     eax, edx /* edx is Mask */
+        mov     [ecx], eax /*ecx is Index */
+        setnz   al
+        retn
+    };
+}
+#endif
 
 /* Does trial division or prob tests, assuming x not divisible by 2, 3, or 5 */
 static int _is_prime7(UV n)

util.h

@@ -137,7 +137,26 @@ static int is_perfect_square(UV n)
   #define log2floor(n)  ((n) ? 31-__builtin_clzl(n) : 0)
  #endif
 #elif defined (_MSC_VER)
+ #if _MSC_VER >= 1400
  #include <intrin.h>
+ #elif _MSC_VER >= 1300
+/* undocumented for VC 2003, not in headers or .libs
+  -Oi required for this to work in -O1 or -Od mode
+  http://www.geoffchappell.com/studies/msvc/language/preprocessor/directives/pragma/intrinsic.htm
+*/
+  unsigned char _BitScanForward(unsigned long* Index, unsigned long Mask);
+  unsigned char _BitScanReverse(unsigned long* Index, unsigned long Mask);
+  unsigned char _BitScanForward64(unsigned long * Index,unsigned __int64 Mask);
+  unsigned char _BitScanReverse64(unsigned long * Index,unsigned __int64 Mask);
+  #pragma intrinsic(_BitScanForward)
+  #pragma intrinsic(_BitScanReverse)
+  #pragma intrinsic(_BitScanForward64)
+  #pragma intrinsic(_BitScanReverse64)
+ #else
+  unsigned char __fastcall _BitScanForward(unsigned long* Index, unsigned long Mask);
+  unsigned char __fastcall _BitScanReverse(unsigned long* Index, unsigned long Mask);
+ #endif
+
  #ifdef FUNC_ctz
   static int ctz(UV n) {
     UV tz = 0;