From 9ee28672473d65d5f0054ded91ab1137aa3722f9 Mon Sep 17 00:00:00 2001
From: Nekotekina <nekotekina@gmail.com>
Date: Mon, 4 Feb 2019 17:00:38 +0300
Subject: [PATCH] Improve count leading/trailing zeros implementation

Use x86 intrinsics if compiled with appropriate instruction support
---
 Utilities/asm.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/Utilities/asm.h b/Utilities/asm.h
index f1c81e37bb..61999060ba 100644
--- a/Utilities/asm.h
+++ b/Utilities/asm.h
@@ -9,6 +9,8 @@ namespace utils
 #ifdef _MSC_VER
 		ulong res;
 		return _BitScanReverse(&res, arg) || nonzero ? res ^ 31 : 32;
+#elif __LZCNT__
+		return _lzcnt_u32(arg);
 #else
 		return arg || nonzero ? __builtin_clz(arg) : 32;
 #endif
@@ -19,6 +21,8 @@ namespace utils
 #ifdef _MSC_VER
 		ulong res;
 		return _BitScanReverse64(&res, arg) || nonzero ? res ^ 63 : 64;
+#elif __LZCNT__
+		return _lzcnt_u64(arg);
 #else
 		return arg || nonzero ? __builtin_clzll(arg) : 64;
 #endif
@@ -29,6 +33,8 @@ namespace utils
 #ifdef _MSC_VER
 		ulong res;
 		return _BitScanForward(&res, arg) || nonzero ? res : 32;
+#elif __BMI__
+		return _tzcnt_u32(arg);
 #else
 		return arg || nonzero ? __builtin_ctz(arg) : 32;
 #endif
@@ -39,6 +45,8 @@ namespace utils
 #ifdef _MSC_VER
 		ulong res;
 		return _BitScanForward64(&res, arg) || nonzero ? res : 64;
+#elif __BMI__
+		return _tzcnt_u64(arg);
 #else
 		return arg || nonzero ? __builtin_ctzll(arg) : 64;
 #endif