v3d/tex: don't configure tmu config 1 if not needed

[mesa.git] / src / util / bitscan.h

diff --git a/src/util/bitscan.h b/src/util/bitscan.h
index a3f2d414bd6ec535c0f66d9d15729a2880ac2fbc..895a1e7a3721a2f90a22babbcd60eb9fabe1b034 100644 (file)
--- a/src/util/bitscan.h
+++ b/src/util/bitscan.h
@@ -38,6 +38,10 @@
 #include <intrin.h>
 #endif
 
+#if defined(__POPCNT__)
+#include <popcntintrin.h>
+#endif
+
 #include "c99_compat.h"
 
 #ifdef __cplusplus
@@ -68,7 +72,7 @@ int ffs(int i);
 
 #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(long long int i)
 {
@@ -119,6 +123,17 @@ 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
@@ -127,7 +142,20 @@ util_is_power_of_two_or_zero(unsigned v)
 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
@@ -207,7 +235,7 @@ 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_ARM || _M_IA64)
+#elif defined(_MSC_VER) && (_M_AMD64 || _M_ARM64 || _M_IA64)
    unsigned long index;
    if (_BitScanReverse64(&index, u))
       return index + 1;
@@ -258,6 +286,38 @@ u_bit_consecutive64(unsigned start, unsigned count)
    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
 }