X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Futil%2Fbitscan.h;h=895a1e7a3721a2f90a22babbcd60eb9fabe1b034;hb=29e2a3b8f5e75508202cd651d16648256f303779;hp=4999b744d1d24141ad8e552340bef4c827f882fe;hpb=c3b66566760dd44eaeed9e4df13687dc3ee69bd9;p=mesa.git

diff --git a/src/util/bitscan.h b/src/util/bitscan.h
index 4999b744d1d..895a1e7a372 100644
--- a/src/util/bitscan.h
+++ b/src/util/bitscan.h
@@ -29,12 +29,19 @@
 #ifndef BITSCAN_H
 #define BITSCAN_H
 
+#include <assert.h>
 #include <stdint.h>
+#include <stdbool.h>
+#include <string.h>
 
 #if defined(_MSC_VER)
 #include <intrin.h>
 #endif
 
+#if defined(__POPCNT__)
+#include <popcntintrin.h>
+#endif
+
 #include "c99_compat.h"
 
 #ifdef __cplusplus
@@ -50,7 +57,7 @@ extern "C" {
 #define ffs __builtin_ffs
 #elif defined(_MSC_VER) && (_M_IX86 || _M_ARM || _M_AMD64 || _M_IA64)
 static inline
-int ffs(unsigned i)
+int ffs(int i)
 {
    unsigned long index;
    if (_BitScanForward(&index, i))
@@ -60,14 +67,14 @@ int ffs(unsigned i)
 }
 #else
 extern
-int ffs(unsigned i);
+int ffs(int i);
 #endif
 
 #ifdef HAVE___BUILTIN_FFSLL
 #define ffsll __builtin_ffsll
-#elif defined(_MSC_VER) && (_M_AMD64 || _M_ARM || _M_IA64)
+#elif defined(_MSC_VER) && (_M_AMD64 || _M_ARM64 || _M_IA64)
 static inline int
-ffsll(uint64_t i)
+ffsll(long long int i)
 {
    unsigned long index;
    if (_BitScanForward64(&index, i))
@@ -77,7 +84,7 @@ ffsll(uint64_t i)
 }
 #else
 extern int
-ffsll(uint64_t val);
+ffsll(long long int val);
 #endif
 
 
@@ -105,6 +112,52 @@ u_bit_scan64(uint64_t *mask)
    return i;
 }
 
+/* Determine if an unsigned value is a power of two.
+ *
+ * \note
+ * Zero is treated as a power of two.
+ */
+static inline bool
+util_is_power_of_two_or_zero(unsigned v)
+{
+   return (v & (v - 1)) == 0;
+}
+
+/* Determine if an uint64_t value is a power of two.
+ *
+ * \note
+ * Zero is treated as a power of two.
+ */
+static inline bool
+util_is_power_of_two_or_zero64(uint64_t v)
+{
+   return (v & (v - 1)) == 0;
+}
+
+/* Determine if an unsigned value is a power of two.
+ *
+ * \note
+ * Zero is \b not treated as a power of two.
+ */
+static inline bool
+util_is_power_of_two_nonzero(unsigned v)
+{
+   /* __POPCNT__ is different from HAVE___BUILTIN_POPCOUNT.  The latter
+    * indicates the existence of the __builtin_popcount function.  The former
+    * indicates that _mm_popcnt_u32 exists and is a native instruction.
+    *
+    * The other alternative is to use SSE 4.2 compile-time flags.  This has
+    * two drawbacks.  First, there is currently no build infrastructure for
+    * SSE 4.2 (only 4.1), so that would have to be added.  Second, some AMD
+    * CPUs support POPCNT but not SSE 4.2 (e.g., Barcelona).
+    */
+#ifdef __POPCNT__
+   return _mm_popcnt_u32(v) == 1;
+#else
+   return v != 0 && (v & (v - 1)) == 0;
+#endif
+}
+
 /* For looping over a bitmask when you want to loop over consecutive bits
  * manually, for example:
  *
@@ -134,7 +187,7 @@ u_bit_scan_consecutive_range(unsigned *mask, int *start, int *count)
 static inline void
 u_bit_scan_consecutive_range64(uint64_t *mask, int *start, int *count)
 {
-   if (*mask == ~0llu) {
+   if (*mask == ~0ull) {
       *start = 0;
       *count = 64;
       *mask = 0;
@@ -146,6 +199,126 @@ u_bit_scan_consecutive_range64(uint64_t *mask, int *start, int *count)
 }
 
 
+/**
+ * Find last bit set in a word.  The least significant bit is 1.
+ * Return 0 if no bits are set.
+ * Essentially ffs() in the reverse direction.
+ */
+static inline unsigned
+util_last_bit(unsigned u)
+{
+#if defined(HAVE___BUILTIN_CLZ)
+   return u == 0 ? 0 : 32 - __builtin_clz(u);
+#elif defined(_MSC_VER) && (_M_IX86 || _M_ARM || _M_AMD64 || _M_IA64)
+   unsigned long index;
+   if (_BitScanReverse(&index, u))
+      return index + 1;
+   else
+      return 0;
+#else
+   unsigned r = 0;
+   while (u) {
+      r++;
+      u >>= 1;
+   }
+   return r;
+#endif
+}
+
+/**
+ * Find last bit set in a word.  The least significant bit is 1.
+ * Return 0 if no bits are set.
+ * Essentially ffsll() in the reverse direction.
+ */
+static inline unsigned
+util_last_bit64(uint64_t u)
+{
+#if defined(HAVE___BUILTIN_CLZLL)
+   return u == 0 ? 0 : 64 - __builtin_clzll(u);
+#elif defined(_MSC_VER) && (_M_AMD64 || _M_ARM64 || _M_IA64)
+   unsigned long index;
+   if (_BitScanReverse64(&index, u))
+      return index + 1;
+   else
+      return 0;
+#else
+   unsigned r = 0;
+   while (u) {
+      r++;
+      u >>= 1;
+   }
+   return r;
+#endif
+}
+
+/**
+ * Find last bit in a word that does not match the sign bit. The least
+ * significant bit is 1.
+ * Return 0 if no bits are set.
+ */
+static inline unsigned
+util_last_bit_signed(int i)
+{
+   if (i >= 0)
+      return util_last_bit(i);
+   else
+      return util_last_bit(~(unsigned)i);
+}
+
+/* Returns a bitfield in which the first count bits starting at start are
+ * set.
+ */
+static inline unsigned
+u_bit_consecutive(unsigned start, unsigned count)
+{
+   assert(start + count <= 32);
+   if (count == 32)
+      return ~0;
+   return ((1u << count) - 1) << start;
+}
+
+static inline uint64_t
+u_bit_consecutive64(unsigned start, unsigned count)
+{
+   assert(start + count <= 64);
+   if (count == 64)
+      return ~(uint64_t)0;
+   return (((uint64_t)1 << count) - 1) << start;
+}
+
+/**
+ * Return number of bits set in n.
+ */
+static inline unsigned
+util_bitcount(unsigned n)
+{
+#if defined(HAVE___BUILTIN_POPCOUNT)
+   return __builtin_popcount(n);
+#else
+   /* K&R classic bitcount.
+    *
+    * For each iteration, clear the LSB from the bitfield.
+    * Requires only one iteration per set bit, instead of
+    * one iteration per bit less than highest set bit.
+    */
+   unsigned bits;
+   for (bits = 0; n; bits++) {
+      n &= n - 1;
+   }
+   return bits;
+#endif
+}
+
+static inline unsigned
+util_bitcount64(uint64_t n)
+{
+#ifdef HAVE___BUILTIN_POPCOUNTLL
+   return __builtin_popcountll(n);
+#else
+   return util_bitcount(n) + util_bitcount(n >> 32);
+#endif
+}
+
 #ifdef __cplusplus
 }
 #endif