Merge pull request #4729 from xavery/bitcount-use-cpu

Use POPCNT, if available, to count the number of set bits

src/platform/platform.h

@@ -196,6 +196,9 @@ bool platform_check_steam_overlay_attached();
 
 datetime64 platform_get_datetime_now_utc();
 
+// Called very early in the program before parsing commandline arguments.
+void core_init();
+
 // Windows specific definitions
 #ifdef __WINDOWS__
 	#ifndef WIN32_LEAN_AND_MEAN

src/platform/posix.c

@@ -51,6 +51,8 @@ utf8 _openrctDataDirectoryPath[MAX_PATH] = { 0 };
  */
 int main(int argc, const char **argv)
 {
+	core_init();
+	
 	int run_game = cmdline_run(argv, argc);
 	if (run_game == 1)
 	{

src/platform/shared.c

@@ -782,3 +782,8 @@ uint8 platform_get_currency_value(const char *currCode) {
 	
 	return CURRENCY_POUNDS;
 }
+
+void core_init()
+{
+	bitcount_init();
+}

src/platform/windows.c

@@ -61,6 +61,8 @@ int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpCmdLine
 {
 	_dllModule = hInstance;
 
+	core_init();
+
 	int argc;
 	char ** argv = (char**)windows_get_command_line_args(&argc);
 	int runGame = cmdline_run((const char **)argv, argc);
@@ -86,6 +88,8 @@ int main(int argc, char *argv[])
 	HINSTANCE hInstance = GetModuleHandle(NULL);
 	_dllModule = hInstance;
 
+	core_init();
+
 	int runGame = cmdline_run((const char **)argv, argc);
 	if (runGame == 1) {
 		openrct2_launch();
@@ -124,6 +128,8 @@ __declspec(dllexport) int StartOpenRCT(HINSTANCE hInstance, HINSTANCE hPrevInsta
 		_dllModule = GetModuleHandleA(OPENRCT2_DLL_MODULE_NAME);
 	}
 
+	core_init();
+
 	// argv = CommandLineToArgvA(lpCmdLine, &argc);
 	argv = (char**)windows_get_command_line_args(&argc);
 	runGame = cmdline_run((const char **)argv, argc);

src/util/util.c

@@ -195,15 +195,74 @@ int bitscanforward(int source)
 	#endif
 }
 
-int bitcount(int source)
+#if defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
+	#include <cpuid.h>
+	#define OpenRCT2_POPCNT_GNUC
+#elif defined(_MSC_VER) && (_MSC_VER >= 1500) && (defined(_M_X64) || defined(_M_IX86)) // VS2008
+	#include <nmmintrin.h>
+	#define OpenRCT2_POPCNT_MSVC
+#endif
+
+static bool bitcount_popcnt_available()
 {
-	int result = 0;
-	for (int i = 0; i < 32; i++) {
-		if (source & (1u << i)) {
-			result++;
-		}
-	}
-	return result;
+	// POPCNT support is declared as the 23rd bit of ECX with CPUID(EAX = 1).
+	#if defined(OpenRCT2_POPCNT_GNUC)
+		// we could use __builtin_cpu_supports, but it requires runtime support from
+		// the compiler's library, which clang doesn't have yet.
+		unsigned int eax, ebx, ecx = 0, edx; // avoid "maybe uninitialized"
+		__get_cpuid(1, &eax, &ebx, &ecx, &edx);
+		return (ecx & (1 << 23));
+	#elif defined(OpenRCT2_POPCNT_MSVC)
+		int regs[4];
+		__cpuid(regs, 1);
+		return (regs[2] & (1 << 23));
+	#else
+		return false;
+	#endif
+}
+
+static int bitcount_popcnt(uint32 source)
+{
+	#if defined(OpenRCT2_POPCNT_GNUC)
+		// use asm directly in order to actually emit the instruction : using
+		// __builtin_popcount results in an extra call to a library function.
+		int rv;
+		asm volatile ("popcnt %1,%0" : "=r"(rv) : "rm"(source) : "cc");
+		return rv;
+	#elif defined(OpenRCT2_POPCNT_MSVC)
+		return _mm_popcnt_u32(source);
+	#else
+		openrct2_assert(false, "bitcount_popcnt() called, without support compiled in");
+		return INT_MAX;
+	#endif
+}
+
+static int bitcount_lut(uint32 source)
+{
+	// https://graphics.stanford.edu/~seander/bithacks.html
+	static const unsigned char BitsSetTable256[256] = 
+	{
+	#define B2(n) n,     n+1,     n+1,     n+2
+	#define B4(n) B2(n), B2(n+1), B2(n+1), B2(n+2)
+	#define B6(n) B4(n), B4(n+1), B4(n+1), B4(n+2)
+	B6(0), B6(1), B6(1), B6(2)
+	};
+	return BitsSetTable256[source & 0xff] +
+		BitsSetTable256[(source >> 8) & 0xff] +
+		BitsSetTable256[(source >> 16) & 0xff] +
+		BitsSetTable256[source >> 24];
+}
+
+static int(*bitcount_fn)(uint32);
+
+void bitcount_init()
+{
+	bitcount_fn = bitcount_popcnt_available() ? bitcount_popcnt : bitcount_lut;
+}
+
+int bitcount(uint32 source)
+{
+	return bitcount_fn(source);
 }
 
 bool strequals(const char *a, const char *b, int length, bool caseInsensitive)

src/util/util.h

@@ -38,7 +38,8 @@ void path_end_with_separator(utf8 *path, size_t size);
 bool readentirefile(const utf8 *path, void **outBuffer, size_t *outLength);
 
 int bitscanforward(int source);
-int bitcount(int source);
+void bitcount_init();
+int bitcount(uint32 source);
 bool strequals(const char *a, const char *b, int length, bool caseInsensitive);
 int strcicmp(char const *a, char const *b);
 int strlogicalcmp(char const *a, char const *b);