Merge pull request #3820 from juliantaylor/microopt

A couple micro optimizations

4 files changed, 33 insertions, 17 deletions

diff --git a/numpy/core/src/multiarray/common.c b/numpy/core/src/multiarray/common.c
index 5a7a58b33..4e2d64be3 100644
--- a/numpy/core/src/multiarray/common.c
+++ b/numpy/core/src/multiarray/common.c
@@ -675,7 +675,8 @@ _zerofill(PyArrayObject *ret)
 NPY_NO_EXPORT int
 _IsAligned(PyArrayObject *ap)
 {
-    unsigned int i, aligned = 1;
+    unsigned int i;
+    npy_uintp aligned;
     const unsigned int alignment = PyArray_DESCR(ap)->alignment;
 
     /* The special casing for STRING and VOID types was removed
@@ -688,24 +689,24 @@ _IsAligned(PyArrayObject *ap)
     if (alignment == 1) {
         return 1;
     }
-    aligned = npy_is_aligned(PyArray_DATA(ap), alignment);
+    aligned = (npy_uintp)PyArray_DATA(ap);
 
     for (i = 0; i < PyArray_NDIM(ap); i++) {
 #if NPY_RELAXED_STRIDES_CHECKING
+        /* skip dim == 1 as it is not required to have stride 0 */
         if (PyArray_DIM(ap, i) > 1) {
             /* if shape[i] == 1, the stride is never used */
-            aligned &= npy_is_aligned((void*)PyArray_STRIDES(ap)[i],
-                                      alignment);
+            aligned |= (npy_uintp)PyArray_STRIDES(ap)[i];
         }
         else if (PyArray_DIM(ap, i) == 0) {
             /* an array with zero elements is always aligned */
             return 1;
         }
 #else /* not NPY_RELAXED_STRIDES_CHECKING */
-        aligned &= npy_is_aligned((void*)PyArray_STRIDES(ap)[i], alignment);
+        aligned |= (npy_uintp)PyArray_STRIDES(ap)[i];
 #endif /* not NPY_RELAXED_STRIDES_CHECKING */
     }
-    return aligned != 0;
+    return npy_is_aligned(aligned, alignment);
 }
 
 NPY_NO_EXPORT npy_bool
diff --git a/numpy/core/src/multiarray/common.h b/numpy/core/src/multiarray/common.h
index 9df644210..ffb571b2e 100644
--- a/numpy/core/src/multiarray/common.h
+++ b/numpy/core/src/multiarray/common.h
@@ -84,6 +84,29 @@ npy_is_aligned(const void * p, const npy_uintp alignment)
     }
 }
 
+/*
+ * writes result of a * b into r
+ * returns 1 if a * b overflowed else returns 0
+ */
+static NPY_INLINE int
+npy_mul_with_overflow_intp(npy_intp * r, npy_intp a, npy_intp b)
+{
+    const npy_intp half_sz = (((npy_intp)1 << (sizeof(a) * 8 / 2)) - 1);
+
+    *r = a * b;
+
+    /*
+     * avoid expensive division on common no overflow case
+     * could be improved via compiler intrinsics e.g. via clang
+     * __builtin_mul_with_overflow, gcc __int128 or cpu overflow flags
+     */
+    if (NPY_UNLIKELY((a | b) >= half_sz) &&
+        a != 0 && b > NPY_MAX_INTP / a) {
+        return 1;
+    }
+
+    return 0;
+}
 
 #include "ucsnarrow.h"
 
diff --git a/numpy/core/src/multiarray/ctors.c b/numpy/core/src/multiarray/ctors.c
index d0b75b47a..bef3feec1 100644
--- a/numpy/core/src/multiarray/ctors.c
+++ b/numpy/core/src/multiarray/ctors.c
@@ -888,7 +888,6 @@ PyArray_NewFromDescr_int(PyTypeObject *subtype, PyArray_Descr *descr, int nd,
     PyArrayObject_fields *fa;
     int i;
     size_t sd;
-    npy_intp largest;
     npy_intp size;
 
     if (descr->subarray) {
@@ -937,7 +936,6 @@ PyArray_NewFromDescr_int(PyTypeObject *subtype, PyArray_Descr *descr, int nd,
         }
     }
 
-    largest = NPY_MAX_INTP / sd;
     for (i = 0; i < nd; i++) {
         npy_intp dim = dims[i];
 
@@ -960,17 +958,14 @@ PyArray_NewFromDescr_int(PyTypeObject *subtype, PyArray_Descr *descr, int nd,
         /*
          * Care needs to be taken to avoid integer overflow when
          * multiplying the dimensions together to get the total size of the
-         * array. Hence before each multiplication we first check that the
-         * product will not exceed the maximum allowable size.
+         * array.
          */
-        if (dim > largest) {
+        if (npy_mul_with_overflow_intp(&size, size, dim)) {
             PyErr_SetString(PyExc_ValueError,
                             "array is too big.");
             Py_DECREF(descr);
             return NULL;
         }
-        size *= dim;
-        largest /= dim;
     }
 
     fa = (PyArrayObject_fields *) subtype->tp_alloc(subtype, 0);
diff --git a/numpy/core/src/multiarray/multiarraymodule.c b/numpy/core/src/multiarray/multiarraymodule.c
index 2eccd8153..f0ada8618 100644
--- a/numpy/core/src/multiarray/multiarraymodule.c
+++ b/numpy/core/src/multiarray/multiarraymodule.c
@@ -115,7 +115,6 @@ NPY_NO_EXPORT npy_intp
 PyArray_OverflowMultiplyList(npy_intp *l1, int n)
 {
     npy_intp prod = 1;
-    npy_intp imax = NPY_MAX_INTP;
     int i;
 
     for (i = 0; i < n; i++) {
@@ -124,11 +123,9 @@ PyArray_OverflowMultiplyList(npy_intp *l1, int n)
         if (dim == 0) {
             return 0;
         }
-        if (dim > imax) {
+        if (npy_mul_with_overflow_intp(&prod, prod, dim)) {
             return -1;
         }
-        imax /= dim;
-        prod *= dim;
     }
     return prod;
 }