ENH: add NPY_LIKELY and NPY_UNLIKELY macros for branching hints

Use it for npy_is_aligned expecting alignments of the power of two.
Cuts down the time spent in _IsAligned by the testsuite relative
to the rest of multiarray.so from 0.6% to 0.4%

3 files changed, 28 insertions, 3 deletions

diff --git a/numpy/core/include/numpy/npy_common.h b/numpy/core/include/numpy/npy_common.h
index ac2e631e6..e5fd09e91 100644
--- a/numpy/core/include/numpy/npy_common.h
+++ b/numpy/core/include/numpy/npy_common.h
@@ -18,6 +18,24 @@
 #define NPY_GCC_UNROLL_LOOPS
 #endif
 
+/*
+ * give a hint to the compiler which branch is more likely or unlikely
+ * to occur, e.g. rare error cases:
+ *
+ * if (NPY_UNLIKELY(failure == 0))
+ *    return NULL;
+ *
+ * the double !! is to cast the expression (e.g. NULL) to a boolean required by
+ * the intrinsic
+ */
+#ifdef HAVE___BUILTIN_EXPECT
+#define NPY_LIKELY(x) __builtin_expect(!!(x), 1)
+#define NPY_UNLIKELY(x) __builtin_expect(!!(x), 0)
+#else
+#define NPY_LIKELY(x) (x)
+#define NPY_UNLIKELY(x) (x)
+#endif
+
 #if defined(_MSC_VER)
         #define NPY_INLINE __inline
 #elif defined(__GNUC__)
diff --git a/numpy/core/setup_common.py b/numpy/core/setup_common.py
index 9f745e965..9a630225d 100644
--- a/numpy/core/setup_common.py
+++ b/numpy/core/setup_common.py
@@ -111,6 +111,7 @@ OPTIONAL_INTRINSICS = [("__builtin_isnan", '5.'),
                        ("__builtin_isfinite", '5.'),
                        ("__builtin_bswap32", '5u'),
                        ("__builtin_bswap64", '5u'),
+                       ("__builtin_expect", '5, 0'),
                        ]
 
 # gcc function attributes
diff --git a/numpy/core/src/multiarray/common.h b/numpy/core/src/multiarray/common.h
index 636d9c67d..82b66123c 100644
--- a/numpy/core/src/multiarray/common.h
+++ b/numpy/core/src/multiarray/common.h
@@ -1,5 +1,6 @@
 #ifndef _NPY_PRIVATE_COMMON_H_
 #define _NPY_PRIVATE_COMMON_H_
+#include <numpy/npy_common.h>
 
 #define error_converting(x)  (((x) == -1) && PyErr_Occurred())
 
@@ -71,11 +72,16 @@ offset_bounds_from_strides(const int itemsize, const int nd,
 static NPY_INLINE int
 npy_is_aligned(const void * p, const npy_uintp alignment)
 {
-    /* test for the and is still faster than a direct modulo */
-    if ((alignment & (alignment - 1)) == 0)
+    /*
+     * alignment is usually a power of two
+     * the test is faster than a direct modulo
+     */
+    if (NPY_LIKELY((alignment & (alignment - 1)) == 0)) {
         return ((npy_uintp)(p) & ((alignment) - 1)) == 0;
-    else
+    }
+    else {
         return ((npy_uintp)(p) % alignment) == 0;
+    }
 }