Add ctz to utils::algorithm

Use CTZ in bitset.forEachBitSet() so we can
efficiently go through them in order instead of
reverse order.

libs/utils/include/utils/algorithm.h

@@ -50,6 +50,23 @@ constexpr inline T clz(T x) noexcept {
     }
     return (sizeof(T) * CHAR_BIT) - details::popcount(x);
 }
+
+template<typename T>
+constexpr inline T ctz(T x) noexcept {
+    static_assert(sizeof(T) <= sizeof(uint64_t), "details::ctz() only support up to 64 bits");
+    T c = sizeof(T) * 8;
+    x &= -signed(x);
+    if (x) c--;
+    if (sizeof(T) * 8 > 32) { // if() only needed to quash compiler warnings
+        if (x & 0x00000000FFFFFFFF) c -= 32;
+    }
+    if (x & 0x0000FFFF) c -= 16;
+    if (x & 0x00FF00FF) c -= 8;
+    if (x & 0x0F0F0F0F) c -= 4;
+    if (x & 0x33333333) c -= 2;
+    if (x & 0x55555555) c -= 1;
+    return c;
+}
 } // namespace details
 
 constexpr inline UTILS_PUBLIC UTILS_PURE
@@ -79,6 +96,32 @@ unsigned long long UTILS_ALWAYS_INLINE clz(unsigned long long x) noexcept {
 #endif
 }
 
+constexpr inline UTILS_PUBLIC UTILS_PURE
+unsigned int UTILS_ALWAYS_INLINE ctz(unsigned int x) noexcept {
+#if __has_builtin(__builtin_ctz)
+    return __builtin_ctz(x);
+#else
+    return details::ctz(x);
+#endif
+}
+
+constexpr inline UTILS_PUBLIC UTILS_PURE
+unsigned long UTILS_ALWAYS_INLINE ctz(unsigned long x) noexcept {
+#if __has_builtin(__builtin_ctzl)
+    return __builtin_ctzl(x);
+#else
+    return details::ctz(x);
+#endif
+}
+
+constexpr inline UTILS_PUBLIC UTILS_PURE
+unsigned long long UTILS_ALWAYS_INLINE ctz(unsigned long long x) noexcept {
+#if __has_builtin(__builtin_ctzll)
+    return __builtin_ctzll(x);
+#else
+    return details::ctz(x);
+#endif
+}
 
 constexpr inline UTILS_PUBLIC UTILS_PURE
 unsigned int UTILS_ALWAYS_INLINE popcount(unsigned int x) noexcept {

libs/utils/include/utils/bitset.h

@@ -80,7 +80,7 @@ public:
         for (size_t i = 0; i < N; i++) {
             T v = storage[i];
             while (v) {
-                T k = (BITS_PER_WORD - 1) - utils::clz(v);
+                T k = utils::ctz(v);
                 v &= ~(T(1) << k);
                 exec(size_t(k + BITS_PER_WORD * i));
             }

libs/utils/test/test_algorithm.cpp

@@ -27,6 +27,11 @@ static inline T count_leading_zeros(T v) noexcept {
     return utils::clz(v);
 }
 
+template <typename T>
+static inline T count_trailing_zeros(T v) noexcept {
+    return utils::ctz(v);
+}
+
 TEST(AlgorithmTest, details_clz) {
     for (uint64_t i = 1, j = 63; i < 64; i *= 2, j--) {
         EXPECT_EQ(j, details::clz(i));
@@ -55,6 +60,28 @@ TEST(AlgorithmTest, clz) {
     }
 }
 
+TEST(AlgorithmTest, details_ctz) {
+    for (uint64_t i = 1, j = 0; i < 64; i *= 2, j++) {
+        EXPECT_EQ(j, details::ctz(i));
+    }
+    for (uint32_t i = 1, j = 0; i < 32; i *= 2, j++) {
+        EXPECT_EQ(j, details::ctz(i));
+    }
+}
+
+TEST(AlgorithmTest, ctz) {
+    for (uint64_t i = 1, j = 0; i < 64; i *= 2, j++) {
+        EXPECT_EQ(j, ctz(i));
+        EXPECT_EQ(j, details::ctz(i));
+        EXPECT_EQ(j, count_trailing_zeros(i));
+    }
+    for (uint32_t i = 1, j = 0; i < 32; i *= 2, j++) {
+        EXPECT_EQ(j, ctz(i));
+        EXPECT_EQ(j, details::ctz(i));
+        EXPECT_EQ(j, count_trailing_zeros(i));
+    }
+}
+
 TEST(AlgorithmTest, details_popcount) {
     EXPECT_EQ(0, details::popcount(0u));
     EXPECT_EQ(0, details::popcount(0lu));