Improve the performance of bit set searches with a compiler intrinsic.

runtime/platform/utils.h

Index: runtime/platform/utils.h
diff --git a/runtime/platform/utils.h b/runtime/platform/utils.h
index 22ee772b7de01c5976c433766821b16cc2f9928b..4cd71913f7ed85987e60790b95836037cc81929c 100644
--- a/runtime/platform/utils.h
+++ b/runtime/platform/utils.h
@@ -79,6 +79,9 @@ class Utils {
   static uint32_t RoundUpToPowerOfTwo(uint32_t x);
   static int CountOneBits(uint32_t x);
 
+  static int CountTrailingZeros(uint32_t x);
+  static int CountTrailingZeros(uint64_t x);

[siva, 2012/07/24 01:07:43]

I am wondering if it makes sense to just have one version of
this 
static int CountTrailingZeros(uword x);
and in the implementation use
#if defined(ARCH_IS_64_BIT)
#elif defined(ARCH_IS_32_BIT)
#else
#error ...
#endif

[cshapiro, 2012/07/24 02:43:50]

Originally, I wrote this method as a template like so

template<typename T>
int CountTrailingZeros(T x) {
  if (sizeof(x) == 4) // 32-bit intrinsic
  else if (sizeof(x) == 8) // 64 bit intrinsic
  else // assert
}

This looked a bit ugly and seemed like overkill so I just punted on it and
settled on separate versions for 32-bit and 64-bit because the other two bit
operations on lines 79 and 80 were specified for 32-bit words and not a uword.

However, this might just be shortsightedness and maybe all of these functions
should be defined for uword only.  Let me know what you think and I will happily
fix this and do a cleanup for the other 2 definitions as well.

+
   // Computes a hash value for the given string.
   static uint32_t StringHash(const char* data, int length);
 
@@ -170,4 +173,14 @@ class Utils {
 
 }  // namespace dart
 
+#if defined(TARGET_OS_LINUX)
+#include "platform/utils_linux.h"
+#elif defined(TARGET_OS_MACOS)
+#include "platform/utils_macos.h"
+#elif defined(TARGET_OS_WINDOWS)
+#include "platform/utils_win.h"
+#else
+#error Unknown target os.
+#endif
+
 #endif  // PLATFORM_UTILS_H_