Move TZCNT and BSR intrinsics to platform.h, add MSVC versions (#636)

3 files changed, 38 insertions, 7 deletions

diff --git a/c/common/platform.h b/c/common/platform.h
index 2633e04..ca79359 100755
--- a/c/common/platform.h
+++ b/c/common/platform.h
@@ -41,6 +41,10 @@
 #define BROTLI_X_BIG_ENDIAN BIG_ENDIAN
 #endif
 
+#if BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+#include <intrin.h>
+#endif
+
 #if defined(BROTLI_ENABLE_LOG) || defined(BROTLI_DEBUG)
 #include <assert.h>
 #include <stdio.h>
@@ -522,6 +526,36 @@ BROTLI_MIN_MAX(size_t) BROTLI_MIN_MAX(uint32_t) BROTLI_MIN_MAX(uint8_t)
   (A)[(J)] = __brotli_swap_tmp;   \
 }
 
+#if BROTLI_64_BITS
+#if BROTLI_GNUC_HAS_BUILTIN(__builtin_ctzll, 3, 4, 0) || \
+    BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
+#define BROTLI_TZCNT64 __builtin_ctzll
+#elif BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+#if defined(BROTLI_TARGET_X64)
+#define BROTLI_TZCNT64 _tzcnt_u64
+#else /* BROTLI_TARGET_X64 */
+static BROTLI_INLINE uint32_t BrotliBsf64Msvc(uint64_t x) {
+  uint32_t lsb;
+  _BitScanForward64(&lsb, x);
+  return lsb;
+}
+#define BROTLI_TZCNT64 BrotliBsf64Msvc
+#endif /* BROTLI_TARGET_X64 */
+#endif /* __builtin_ctzll */
+#endif /* BROTLI_64_BITS */
+
+#if BROTLI_GNUC_HAS_BUILTIN(__builtin_clz, 3, 4, 0) || \
+    BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
+#define BROTLI_BSR32(x) (31u ^ (uint32_t)__builtin_clz(x))
+#elif BROTLI_MSVC_VERSION_CHECK(12, 0, 0)
+static BROTLI_INLINE uint32_t BrotliBsr32Msvc(uint32_t x) {
+  uint32_t msb;
+  _BitScanReverse(&msb, x);
+  return msb;
+}
+#define BROTLI_BSR32 BrotliBsr32Msvc
+#endif /* __builtin_clz */
+
 /* Default brotli_alloc_func */
 static void* BrotliDefaultAllocFunc(void* opaque, size_t size) {
   BROTLI_UNUSED(opaque);
diff --git a/c/enc/fast_log.h b/c/enc/fast_log.h
index b1268e0..eca58e8 100644
--- a/c/enc/fast_log.h
+++ b/c/enc/fast_log.h
@@ -19,10 +19,8 @@ extern "C" {
 #endif
 
 static BROTLI_INLINE uint32_t Log2FloorNonZero(size_t n) {
-  /* TODO: generalize and move to platform.h */
-#if BROTLI_GNUC_HAS_BUILTIN(__builtin_clz, 3, 4, 0) || \
-    BROTLI_INTEL_VERSION_CHECK(16, 0, 0)
-  return 31u ^ (uint32_t)__builtin_clz((uint32_t)n);
+#if defined(BROTLI_BSR32)
+  return BROTLI_BSR32((uint32_t)n);
 #else
   uint32_t result = 0;
   while (n >>= 1) result++;
diff --git a/c/enc/find_match_length.h b/c/enc/find_match_length.h
index bc428cf..f8853a7 100644
--- a/c/enc/find_match_length.h
+++ b/c/enc/find_match_length.h
@@ -17,8 +17,7 @@ extern "C" {
 #endif
 
 /* Separate implementation for little-endian 64-bit targets, for speed. */
-#if defined(__GNUC__) && defined(_LP64) && defined(BROTLI_LITTLE_ENDIAN)
-
+#if defined(BROTLI_TZCNT64) && BROTLI_64_BITS && BROTLI_LITTLE_ENDIAN
 static BROTLI_INLINE size_t FindMatchLengthWithLimit(const uint8_t* s1,
                                                      const uint8_t* s2,
                                                      size_t limit) {
@@ -32,7 +31,7 @@ static BROTLI_INLINE size_t FindMatchLengthWithLimit(const uint8_t* s1,
     } else {
       uint64_t x = BROTLI_UNALIGNED_LOAD64LE(s2) ^
           BROTLI_UNALIGNED_LOAD64LE(s1 + matched);
-      size_t matching_bits = (size_t)__builtin_ctzll(x);
+      size_t matching_bits = (size_t)BROTLI_TZCNT64(x);
       matched += matching_bits >> 3;
       return matched;
     }