Guard _BitScanReverse on MSVC so that MSVC 6 doesn't break.

This also adds some extra casts to shut up compiler warnings
 reported on MSVC 6 where there is implicit truncation for the
 arguments of bitexact_cos().
Lacking access to CLZ/BSR will make the code a fair bit slower but
 that is better than failing to compile.

celt/bands.c

@@ -923,8 +923,8 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c
          fill &= ((1<<B)-1)<<B;
          delta = 16384;
       } else {
-         imid = bitexact_cos(itheta);
-         iside = bitexact_cos(16384-itheta);
+         imid = bitexact_cos((opus_int16)itheta);
+         iside = bitexact_cos((opus_int16)(16384-itheta));
          /* This is the mid vs side allocation that minimizes squared error
             in that band. */
          delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid));

celt/ecintrin.h

@@ -48,7 +48,7 @@
 /*Count leading zeros.
   This macro should only be used for implementing ec_ilog(), if it is defined.
   All other code should use EC_ILOG() instead.*/
-#if defined(_MSC_VER)
+#if defined(_MSC_VER) && (_MSC_VER >= 1400)
 # include <intrin.h>
 /*In _DEBUG mode this is not an intrinsic by default.*/
 # pragma intrinsic(_BitScanReverse)

celt/entcode.c

@@ -33,6 +33,11 @@
 #include "arch.h"
 
 #if !defined(EC_CLZ)
+/*This is a fallback for systems where we don't know how to access
+   a BSR or CLZ instruction (see ecintrin.h).
+  If you are optimizing Opus on a new platform and it has a native CLZ or
+   BZR (e.g. cell, MIPS, x86, etc) then making it available to Opus will be
+   an easy performance win.*/
 int ec_ilog(opus_uint32 _v){
   /*On a Pentium M, this branchless version tested as the fastest on
      1,000,000,000 random 32-bit integers, edging out a similar version with