Merge pull request #4475 from juliantaylor/gil-release

Gil release for nonzero

16 files changed, 116 insertions, 112 deletions

diff --git a/numpy/core/include/numpy/ndarraytypes.h b/numpy/core/include/numpy/ndarraytypes.h
index 27970ff77..cc0206d44 100644
--- a/numpy/core/include/numpy/ndarraytypes.h
+++ b/numpy/core/include/numpy/ndarraytypes.h
@@ -928,10 +928,11 @@ typedef int (PyArray_FinalizeFunc)(PyArrayObject *, PyObject *);
 #define PyArray_IS_C_CONTIGUOUS(m) PyArray_CHKFLAGS(m, NPY_ARRAY_C_CONTIGUOUS)
 #define PyArray_IS_F_CONTIGUOUS(m) PyArray_CHKFLAGS(m, NPY_ARRAY_F_CONTIGUOUS)
 
+/* the variable is used in some places, so always define it */
+#define NPY_BEGIN_THREADS_DEF PyThreadState *_save=NULL;
 #if NPY_ALLOW_THREADS
 #define NPY_BEGIN_ALLOW_THREADS Py_BEGIN_ALLOW_THREADS
 #define NPY_END_ALLOW_THREADS Py_END_ALLOW_THREADS
-#define NPY_BEGIN_THREADS_DEF PyThreadState *_save=NULL;
 #define NPY_BEGIN_THREADS do {_save = PyEval_SaveThread();} while (0);
 #define NPY_END_THREADS   do {if (_save) PyEval_RestoreThread(_save);} while (0);
 #define NPY_BEGIN_THREADS_THRESHOLDED(loop_size) do { if (loop_size > 500) \
@@ -951,7 +952,6 @@ typedef int (PyArray_FinalizeFunc)(PyArrayObject *, PyObject *);
 #else
 #define NPY_BEGIN_ALLOW_THREADS
 #define NPY_END_ALLOW_THREADS
-#define NPY_BEGIN_THREADS_DEF
 #define NPY_BEGIN_THREADS
 #define NPY_END_THREADS
 #define NPY_BEGIN_THREADS_THRESHOLDED(loop_size)
diff --git a/numpy/core/src/multiarray/array_assign_array.c b/numpy/core/src/multiarray/array_assign_array.c
index ba18684a2..67c98d8b5 100644
--- a/numpy/core/src/multiarray/array_assign_array.c
+++ b/numpy/core/src/multiarray/array_assign_array.c
@@ -99,9 +99,7 @@ raw_array_assign_array(int ndim, npy_intp *shape,
                             dst_data, dst_strides_it,
                             src_data, src_strides_it);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     NPY_AUXDATA_FREE(transferdata);
 
@@ -194,9 +192,7 @@ raw_array_wheremasked_assign_array(int ndim, npy_intp *shape,
                             src_data, src_strides_it,
                             wheremask_data, wheremask_strides_it);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     NPY_AUXDATA_FREE(transferdata);
 
diff --git a/numpy/core/src/multiarray/array_assign_scalar.c b/numpy/core/src/multiarray/array_assign_scalar.c
index 93bc4e4b4..7c1b1f16a 100644
--- a/numpy/core/src/multiarray/array_assign_scalar.c
+++ b/numpy/core/src/multiarray/array_assign_scalar.c
@@ -86,9 +86,7 @@ raw_array_assign_scalar(int ndim, npy_intp *shape,
     } NPY_RAW_ITER_ONE_NEXT(idim, ndim, coord,
                             shape_it, dst_data, dst_strides_it);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     NPY_AUXDATA_FREE(transferdata);
 
@@ -165,9 +163,7 @@ raw_array_wheremasked_assign_scalar(int ndim, npy_intp *shape,
                             dst_data, dst_strides_it,
                             wheremask_data, wheremask_strides_it);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     NPY_AUXDATA_FREE(transferdata);
 
diff --git a/numpy/core/src/multiarray/arraytypes.c.src b/numpy/core/src/multiarray/arraytypes.c.src
index 0f75c8aa5..a430e1ee4 100644
--- a/numpy/core/src/multiarray/arraytypes.c.src
+++ b/numpy/core/src/multiarray/arraytypes.c.src
@@ -3501,15 +3501,22 @@ static int
                     NPY_CLIPMODE clipmode)
 {
     npy_intp i, j, k, tmp;
+    NPY_BEGIN_THREADS_DEF;
+
+    NPY_BEGIN_THREADS;
 
     switch(clipmode) {
     case NPY_RAISE:
         for (i = 0; i < n_outer; i++) {
             for (j = 0; j < m_middle; j++) {
                 tmp = indarray[j];
-                /* We don't know what axis we're operating on, so don't report it in case of an error. */
-                if (check_and_adjust_index(&tmp, nindarray, -1) < 0)
+                /*
+                 * We don't know what axis we're operating on,
+                 * so don't report it in case of an error.
+                 */
+                if (check_and_adjust_index(&tmp, nindarray, -1, _save) < 0) {
                     return 1;
+                }
                 if (NPY_LIKELY(nelem == 1)) {
                     *dest++ = *(src + tmp);
                 }
@@ -3571,6 +3578,8 @@ static int
         }
         break;
     }
+
+    NPY_END_THREADS;
     return 0;
 }
 /**end repeat**/
diff --git a/numpy/core/src/multiarray/common.c b/numpy/core/src/multiarray/common.c
index 48d13cd6e..0e8a21394 100644
--- a/numpy/core/src/multiarray/common.c
+++ b/numpy/core/src/multiarray/common.c
@@ -618,7 +618,7 @@ index2ptr(PyArrayObject *mp, npy_intp i)
         return NULL;
     }
     dim0 = PyArray_DIMS(mp)[0];
-    if (check_and_adjust_index(&i, dim0, 0) < 0)
+    if (check_and_adjust_index(&i, dim0, 0, NULL) < 0)
         return NULL;
     if (i == 0) {
         return PyArray_DATA(mp);
diff --git a/numpy/core/src/multiarray/common.h b/numpy/core/src/multiarray/common.h
index ff1d4ec77..d64411115 100644
--- a/numpy/core/src/multiarray/common.h
+++ b/numpy/core/src/multiarray/common.h
@@ -2,9 +2,21 @@
 #define _NPY_PRIVATE_COMMON_H_
 #include <numpy/npy_common.h>
 #include <numpy/npy_cpu.h>
+#include <numpy/ndarraytypes.h>
 
 #define error_converting(x)  (((x) == -1) && PyErr_Occurred())
 
+#ifdef NPY_ALLOW_THREADS
+#define NPY_BEGIN_THREADS_NDITER(iter) \
+        do { \
+            if (!NpyIter_IterationNeedsAPI(iter)) { \
+                NPY_BEGIN_THREADS_THRESHOLDED(NpyIter_GetIterSize(iter)); \
+            } \
+        } while(0)
+#else
+#define NPY_BEGIN_THREADS_NDITER(iter)
+#endif
+
 /*
  * Recursively examines the object to determine an appropriate dtype
  * to use for converting to an ndarray.
@@ -70,12 +82,17 @@ convert_shape_to_string(npy_intp n, npy_intp *vals, char *ending);
  * 0 <= *index < max_item, and returns 0.
  * 'axis' should be the array axis that is being indexed over, if known. If
  * unknown, use -1.
+ * If _save is NULL it is assumed the GIL is taken
+ * If _save is not NULL it is assumed the GIL is not taken and it
+ * is acquired in the case of an error
  */
 static NPY_INLINE int
-check_and_adjust_index(npy_intp *index, npy_intp max_item, int axis)
+check_and_adjust_index(npy_intp *index, npy_intp max_item, int axis,
+                       PyThreadState * _save)
 {
     /* Check that index is valid, taking into account negative indices */
     if (NPY_UNLIKELY((*index < -max_item) || (*index >= max_item))) {
+        NPY_END_THREADS;
         /* Try to be as clear as possible about what went wrong. */
         if (axis >= 0) {
             PyErr_Format(PyExc_IndexError,
@@ -85,8 +102,7 @@ check_and_adjust_index(npy_intp *index, npy_intp max_item, int axis)
         } else {
             PyErr_Format(PyExc_IndexError,
                          "index %"NPY_INTP_FMT" is out of bounds "
-                         "for size %"NPY_INTP_FMT,
-                         *index, max_item);
+                         "for size %"NPY_INTP_FMT, *index, max_item);
         }
         return -1;
     }
diff --git a/numpy/core/src/multiarray/ctors.c b/numpy/core/src/multiarray/ctors.c
index 5bdd5b552..2ab46d258 100644
--- a/numpy/core/src/multiarray/ctors.c
+++ b/numpy/core/src/multiarray/ctors.c
@@ -2635,9 +2635,7 @@ PyArray_CopyAsFlat(PyArrayObject *dst, PyArrayObject *src, NPY_ORDER order)
         }
     }
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     NPY_AUXDATA_FREE(transferdata);
     NpyIter_Deallocate(dst_iter);
@@ -2845,6 +2843,7 @@ PyArray_Arange(double start, double stop, double step, int type_num)
     PyArray_ArrFuncs *funcs;
     PyObject *obj;
     int ret;
+    NPY_BEGIN_THREADS_DEF;
 
     if (_safe_ceil_to_intp((stop - start)/step, &length)) {
         PyErr_SetString(PyExc_OverflowError,
@@ -2892,7 +2891,9 @@ PyArray_Arange(double start, double stop, double step, int type_num)
         Py_DECREF(range);
         return NULL;
     }
+    NPY_BEGIN_THREADS_DESCR(PyArray_DESCR(range));
     funcs->fill(PyArray_DATA(range), length, range);
+    NPY_END_THREADS;
     if (PyErr_Occurred()) {
         goto fail;
     }
@@ -2989,6 +2990,7 @@ PyArray_ArangeObj(PyObject *start, PyObject *stop, PyObject *step, PyArray_Descr
     npy_intp length;
     PyArray_Descr *native = NULL;
     int swap;
+    NPY_BEGIN_THREADS_DEF;
 
     /* Datetime arange is handled specially */
     if ((dtype != NULL && (dtype->type_num == NPY_DATETIME ||
@@ -3105,7 +3107,9 @@ PyArray_ArangeObj(PyObject *start, PyObject *stop, PyObject *step, PyArray_Descr
         Py_DECREF(range);
         goto fail;
     }
+    NPY_BEGIN_THREADS_DESCR(PyArray_DESCR(range));
     funcs->fill(PyArray_DATA(range), length, range);
+    NPY_END_THREADS;
     if (PyErr_Occurred()) {
         goto fail;
     }
diff --git a/numpy/core/src/multiarray/einsum.c.src b/numpy/core/src/multiarray/einsum.c.src
index ebd381b06..b3148f573 100644
--- a/numpy/core/src/multiarray/einsum.c.src
+++ b/numpy/core/src/multiarray/einsum.c.src
@@ -22,6 +22,7 @@
 #include <ctype.h>
 
 #include "convert.h"
+#include "common.h"
 
 #ifdef NPY_HAVE_SSE_INTRINSICS
 #define EINSUM_USE_SSE1 1
@@ -2969,7 +2970,6 @@ PyArray_EinsteinSum(char *subscripts, npy_intp nop,
         char **dataptr;
         npy_intp *stride;
         npy_intp *countptr;
-        int needs_api = NpyIter_IterationNeedsAPI(iter);
         NPY_BEGIN_THREADS_DEF;
 
         iternext = NpyIter_GetIterNext(iter, NULL);
@@ -2982,19 +2982,15 @@ PyArray_EinsteinSum(char *subscripts, npy_intp nop,
         stride = NpyIter_GetInnerStrideArray(iter);
         countptr = NpyIter_GetInnerLoopSizePtr(iter);
 
-        if (!needs_api) {
-            NPY_BEGIN_THREADS;
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
         NPY_EINSUM_DBG_PRINT("Einsum loop\n");
         do {
             sop(nop, dataptr, stride, *countptr);
         } while(iternext(iter));
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
 
         /* If the API was needed, it may have thrown an error */
-        if (needs_api && PyErr_Occurred()) {
+        if (NpyIter_IterationNeedsAPI(iter) && PyErr_Occurred()) {
             Py_DECREF(ret);
             ret = NULL;
         }
diff --git a/numpy/core/src/multiarray/item_selection.c b/numpy/core/src/multiarray/item_selection.c
index ebd69fe3b..71c17c9c9 100644
--- a/numpy/core/src/multiarray/item_selection.c
+++ b/numpy/core/src/multiarray/item_selection.c
@@ -129,12 +129,15 @@ PyArray_TakeFrom(PyArrayObject *self0, PyObject *indices0, int axis,
 
     func = PyArray_DESCR(self)->f->fasttake;
     if (func == NULL) {
+        NPY_BEGIN_THREADS_DEF;
+        NPY_BEGIN_THREADS_DESCR(PyArray_DESCR(self));
         switch(clipmode) {
         case NPY_RAISE:
             for (i = 0; i < n; i++) {
                 for (j = 0; j < m; j++) {
                     tmp = ((npy_intp *)(PyArray_DATA(indices)))[j];
-                    if (check_and_adjust_index(&tmp, max_item, axis) < 0) {
+                    if (check_and_adjust_index(&tmp, max_item, axis,
+                                               _save) < 0) {
                         goto fail;
                     }
                     tmp_src = src + tmp * chunk;
@@ -216,8 +219,10 @@ PyArray_TakeFrom(PyArrayObject *self0, PyObject *indices0, int axis,
             }
             break;
         }
+        NPY_END_THREADS;
     }
     else {
+        /* no gil release, need it for error reporting */
         err = func(dest, src, (npy_intp *)(PyArray_DATA(indices)),
                     max_item, n, m, nelem, clipmode);
         if (err) {
@@ -300,7 +305,7 @@ PyArray_PutTo(PyArrayObject *self, PyObject* values0, PyObject *indices0,
             for (i = 0; i < ni; i++) {
                 src = PyArray_BYTES(values) + chunk*(i % nv);
                 tmp = ((npy_intp *)(PyArray_DATA(indices)))[i];
-                if (check_and_adjust_index(&tmp, max_item, 0) < 0) {
+                if (check_and_adjust_index(&tmp, max_item, 0, NULL) < 0) {
                     goto fail;
                 }
                 PyArray_Item_INCREF(src, PyArray_DESCR(self));
@@ -350,7 +355,7 @@ PyArray_PutTo(PyArrayObject *self, PyObject* values0, PyObject *indices0,
             for (i = 0; i < ni; i++) {
                 src = PyArray_BYTES(values) + chunk * (i % nv);
                 tmp = ((npy_intp *)(PyArray_DATA(indices)))[i];
-                if (check_and_adjust_index(&tmp, max_item, 0) < 0) {
+                if (check_and_adjust_index(&tmp, max_item, 0, NULL) < 0) {
                     goto fail;
                 }
                 memmove(dest + tmp * chunk, src, chunk);
@@ -2455,12 +2460,15 @@ PyArray_Nonzero(PyArrayObject *self)
         char * data = PyArray_BYTES(self);
         npy_intp stride = (ndim == 0) ? 0 : PyArray_STRIDE(self, 0);
         npy_intp count = (ndim == 0) ? 1 : PyArray_DIM(self, 0);
+        NPY_BEGIN_THREADS_DEF;
 
         /* nothing to do */
         if (nonzero_count == 0) {
             goto finish;
         }
 
+        NPY_BEGIN_THREADS_THRESHOLDED(count);
+
         /* avoid function call for bool */
         if (PyArray_ISBOOL(self)) {
             /*
@@ -2501,6 +2509,8 @@ PyArray_Nonzero(PyArrayObject *self)
             }
         }
 
+        NPY_END_THREADS;
+
         goto finish;
     }
 
@@ -2521,6 +2531,7 @@ PyArray_Nonzero(PyArrayObject *self)
 
     if (NpyIter_GetIterSize(iter) != 0) {
         npy_intp * multi_index;
+        NPY_BEGIN_THREADS_DEF;
         /* Get the pointers for inner loop iteration */
         iternext = NpyIter_GetIterNext(iter, NULL);
         if (iternext == NULL) {
@@ -2534,6 +2545,9 @@ PyArray_Nonzero(PyArrayObject *self)
             Py_DECREF(ret);
             return NULL;
         }
+
+        NPY_BEGIN_THREADS_NDITER(iter);
+
         dataptr = NpyIter_GetDataPtrArray(iter);
 
         multi_index = (npy_intp *)PyArray_DATA(ret);
@@ -2556,6 +2570,8 @@ PyArray_Nonzero(PyArrayObject *self)
                 }
             } while(iternext(iter));
         }
+
+        NPY_END_THREADS;
     }
 
     NpyIter_Deallocate(iter);
@@ -2624,7 +2640,7 @@ PyArray_MultiIndexGetItem(PyArrayObject *self, npy_intp *multi_index)
         npy_intp shapevalue = shape[idim];
         npy_intp ind = multi_index[idim];
 
-        if (check_and_adjust_index(&ind, shapevalue, idim) < 0) {
+        if (check_and_adjust_index(&ind, shapevalue, idim, NULL) < 0) {
           return NULL;
         }
         data += ind * strides[idim];
@@ -2653,7 +2669,7 @@ PyArray_MultiIndexSetItem(PyArrayObject *self, npy_intp *multi_index,
         npy_intp shapevalue = shape[idim];
         npy_intp ind = multi_index[idim];
 
-        if (check_and_adjust_index(&ind, shapevalue, idim) < 0) {
+        if (check_and_adjust_index(&ind, shapevalue, idim, NULL) < 0) {
             return -1;
         }
         data += ind * strides[idim];
diff --git a/numpy/core/src/multiarray/iterators.c b/numpy/core/src/multiarray/iterators.c
index e04ac3bec..81c6b2a8e 100644
--- a/numpy/core/src/multiarray/iterators.c
+++ b/numpy/core/src/multiarray/iterators.c
@@ -71,7 +71,7 @@ parse_index_entry(PyObject *op, npy_intp *step_size,
         *n_steps = SINGLE_INDEX;
         *step_size = 0;
         if (check_index) {
-            if (check_and_adjust_index(&i, max, axis) < 0) {
+            if (check_and_adjust_index(&i, max, axis, NULL) < 0) {
             goto fail;
             }
         }
@@ -668,7 +668,7 @@ iter_subscript_int(PyArrayIterObject *self, PyArrayObject *ind)
     itemsize = PyArray_DESCR(self->ao)->elsize;
     if (PyArray_NDIM(ind) == 0) {
         num = *((npy_intp *)PyArray_DATA(ind));
-        if (check_and_adjust_index(&num, self->size, -1) < 0) {
+        if (check_and_adjust_index(&num, self->size, -1, NULL) < 0) {
             PyArray_ITER_RESET(self);
             return NULL;
         }
@@ -702,7 +702,7 @@ iter_subscript_int(PyArrayIterObject *self, PyArrayObject *ind)
     swap = (PyArray_ISNOTSWAPPED(ret) != PyArray_ISNOTSWAPPED(self->ao));
     while (counter--) {
         num = *((npy_intp *)(ind_it->dataptr));
-        if (check_and_adjust_index(&num, self->size, -1) < 0) {
+        if (check_and_adjust_index(&num, self->size, -1, NULL) < 0) {
             Py_DECREF(ind_it);
             Py_DECREF(ret);
             PyArray_ITER_RESET(self);
@@ -926,7 +926,7 @@ iter_ass_sub_int(PyArrayIterObject *self, PyArrayObject *ind,
     copyswap = PyArray_DESCR(self->ao)->f->copyswap;
     if (PyArray_NDIM(ind) == 0) {
         num = *((npy_intp *)PyArray_DATA(ind));
-        if (check_and_adjust_index(&num, self->size, -1) < 0) {
+        if (check_and_adjust_index(&num, self->size, -1, NULL) < 0) {
             return -1;
         }
         PyArray_ITER_GOTO1D(self, num);
@@ -940,7 +940,7 @@ iter_ass_sub_int(PyArrayIterObject *self, PyArrayObject *ind,
     counter = ind_it->size;
     while (counter--) {
         num = *((npy_intp *)(ind_it->dataptr));
-        if (check_and_adjust_index(&num, self->size, -1) < 0) {
+        if (check_and_adjust_index(&num, self->size, -1, NULL) < 0) {
             Py_DECREF(ind_it);
             return -1;
         }
@@ -1017,7 +1017,7 @@ iter_ass_subscript(PyArrayIterObject *self, PyObject *ind, PyObject *val)
         PyErr_Clear();
     }
     else {
-        if (check_and_adjust_index(&start, self->size, -1) < 0) {
+        if (check_and_adjust_index(&start, self->size, -1, NULL) < 0) {
             goto finish;
         }
         retval = 0;
diff --git a/numpy/core/src/multiarray/lowlevel_strided_loops.c.src b/numpy/core/src/multiarray/lowlevel_strided_loops.c.src
index 042aa2a62..bd69c5ca9 100644
--- a/numpy/core/src/multiarray/lowlevel_strided_loops.c.src
+++ b/numpy/core/src/multiarray/lowlevel_strided_loops.c.src
@@ -1396,7 +1396,7 @@ mapiter_trivial_@name@(PyArrayObject *self, PyArrayObject *ind,
     /* Check the indices beforehand */
     while (itersize--) {
         indval = *((npy_intp*)ind_ptr);
-        if (check_and_adjust_index(&indval, fancy_dim, 1) < 0 ) {
+        if (check_and_adjust_index(&indval, fancy_dim, 1, NULL) < 0 ) {
             return -1;
         }
         ind_ptr += ind_stride;
@@ -1426,7 +1426,7 @@ mapiter_trivial_@name@(PyArrayObject *self, PyArrayObject *ind,
         while (itersize--) {
             indval = *((npy_intp*)ind_ptr);
 #if @isget@
-            if (check_and_adjust_index(&indval, fancy_dim, 1) < 0 ) {
+            if (check_and_adjust_index(&indval, fancy_dim, 1, NULL) < 0 ) {
                 return -1;
             }
 #else
@@ -1560,7 +1560,7 @@ mapiter_@name@(PyArrayMapIterObject *mit)
 
 #if @isget@ && @one_iter@
                             if (check_and_adjust_index(&indval, fancy_dims[i],
-                                                       iteraxis) < 0 ) {
+                                                       iteraxis, NULL) < 0 ) {
                                 return -1;
                             }
 #else
@@ -1662,7 +1662,7 @@ mapiter_@name@(PyArrayMapIterObject *mit)
 
 #if @isget@ && @one_iter@
                     if (check_and_adjust_index(&indval, fancy_dims[i],
-                                               iteraxis) < 0 ) {
+                                               iteraxis, NULL) < 0 ) {
                         NPY_AUXDATA_FREE(transferdata);
                         return -1;
                     }
diff --git a/numpy/core/src/multiarray/mapping.c b/numpy/core/src/multiarray/mapping.c
index 6d7e4ffe3..cafec7689 100644
--- a/numpy/core/src/multiarray/mapping.c
+++ b/numpy/core/src/multiarray/mapping.c
@@ -799,7 +799,7 @@ get_item_pointer(PyArrayObject *self, char **ptr,
     *ptr = PyArray_BYTES(self);
     for (i=0; i < index_num; i++) {
         if ((check_and_adjust_index(&(indices[i].value),
-                               PyArray_DIMS(self)[i], i)) < 0) {
+                               PyArray_DIMS(self)[i], i, NULL)) < 0) {
             return -1;
         }
         *ptr += PyArray_STRIDE(self, i) * indices[i].value;
@@ -842,7 +842,8 @@ get_view_from_index(PyArrayObject *self, PyArrayObject **view,
         switch (indices[i].type) {
             case HAS_INTEGER:
                 if ((check_and_adjust_index(&indices[i].value,
-                                PyArray_DIMS(self)[orig_dim], orig_dim)) < 0) {
+                                PyArray_DIMS(self)[orig_dim], orig_dim,
+                                NULL)) < 0) {
                     return -1;
                 }
                 data_ptr += PyArray_STRIDE(self, orig_dim) * indices[i].value;
@@ -2366,7 +2367,7 @@ PyArray_MapIterCheckIndices(PyArrayMapIterObject *mit)
             while (itersize--) {
                 indval = *((npy_intp*)data);
                 if (check_and_adjust_index(&indval,
-                                           outer_dim, outer_axis) < 0) {
+                                           outer_dim, outer_axis, NULL) < 0) {
                     return -1;
                 }
                 data += stride;
@@ -2400,7 +2401,7 @@ PyArray_MapIterCheckIndices(PyArrayMapIterObject *mit)
             while (itersize--) {
                 indval = *((npy_intp*)*iterptr);
                 if (check_and_adjust_index(&indval,
-                                           outer_dim, outer_axis) < 0) {
+                                           outer_dim, outer_axis, NULL) < 0) {
 
                     Py_DECREF(intp_type);
                     NpyIter_Deallocate(op_iter);
diff --git a/numpy/core/src/multiarray/methods.c b/numpy/core/src/multiarray/methods.c
index 91f639650..b68745085 100644
--- a/numpy/core/src/multiarray/methods.c
+++ b/numpy/core/src/multiarray/methods.c
@@ -647,7 +647,7 @@ array_toscalar(PyArrayObject *self, PyObject *args)
             return NULL;
         }
 
-        if (check_and_adjust_index(&value, size, -1) < 0) {
+        if (check_and_adjust_index(&value, size, -1, NULL) < 0) {
             return NULL;
         }
 
@@ -724,7 +724,7 @@ array_setscalar(PyArrayObject *self, PyObject *args)
             return NULL;
         }
 
-        if (check_and_adjust_index(&value, size, -1) < 0) {
+        if (check_and_adjust_index(&value, size, -1, NULL) < 0) {
             return NULL;
         }
 
diff --git a/numpy/core/src/multiarray/multiarraymodule.c b/numpy/core/src/multiarray/multiarraymodule.c
index cfaf7799a..2b8fedf9a 100644
--- a/numpy/core/src/multiarray/multiarraymodule.c
+++ b/numpy/core/src/multiarray/multiarraymodule.c
@@ -2774,9 +2774,7 @@ PyArray_Where(PyObject *condition, PyObject *x, PyObject *y)
 
         needs_api = NpyIter_IterationNeedsAPI(iter);
 
-        if (!needs_api) {
-            NPY_BEGIN_THREADS_THRESHOLDED(NpyIter_GetIterSize(iter));
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
 
         if (NpyIter_GetIterSize(iter) != 0) {
             NpyIter_IterNextFunc *iternext = NpyIter_GetIterNext(iter, NULL);
@@ -2836,9 +2834,7 @@ PyArray_Where(PyObject *condition, PyObject *x, PyObject *y)
             } while (iternext(iter));
         }
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
 
         /* Get the result from the iterator object array */
         ret = (PyObject*)NpyIter_GetOperandArray(iter)[0];
diff --git a/numpy/core/src/umath/ufunc_object.c b/numpy/core/src/umath/ufunc_object.c
index 96eb600e1..233855460 100644
--- a/numpy/core/src/umath/ufunc_object.c
+++ b/numpy/core/src/umath/ufunc_object.c
@@ -1153,9 +1153,7 @@ trivial_two_operand_loop(PyArrayObject **op,
 
     innerloop(data, count, stride, innerloopdata);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 }
 
 static void
@@ -1186,9 +1184,7 @@ trivial_three_operand_loop(PyArrayObject **op,
 
     innerloop(data, count, stride, innerloopdata);
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 }
 
 /*
@@ -1270,7 +1266,6 @@ iterator_loop(PyUFuncObject *ufunc,
     npy_uint32 op_flags[NPY_MAXARGS];
     NpyIter *iter;
     char *baseptrs[NPY_MAXARGS];
-    int needs_api;
 
     NpyIter_IterNextFunc *iternext;
     char **dataptr;
@@ -1325,8 +1320,6 @@ iterator_loop(PyUFuncObject *ufunc,
         return -1;
     }
 
-    needs_api = NpyIter_IterationNeedsAPI(iter);
-
     /* Copy any allocated outputs */
     op_it = NpyIter_GetOperandArray(iter);
     for (i = nin; i < nop; ++i) {
@@ -1370,9 +1363,7 @@ iterator_loop(PyUFuncObject *ufunc,
         stride = NpyIter_GetInnerStrideArray(iter);
         count_ptr = NpyIter_GetInnerLoopSizePtr(iter);
 
-        if (!needs_api) {
-            NPY_BEGIN_THREADS_THRESHOLDED(NpyIter_GetIterSize(iter));
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
 
         /* Execute the loop */
         do {
@@ -1380,9 +1371,7 @@ iterator_loop(PyUFuncObject *ufunc,
             innerloop(dataptr, count_ptr, stride, innerloopdata);
         } while (iternext(iter));
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
     }
 
     NpyIter_Deallocate(iter);
@@ -1579,8 +1568,6 @@ execute_fancy_ufunc_loop(PyUFuncObject *ufunc,
     PyArrayObject **op_it;
     npy_uint32 iter_flags;
 
-    NPY_BEGIN_THREADS_DEF;
-
     if (wheremask != NULL) {
         if (nop + 1 > NPY_MAXARGS) {
             PyErr_SetString(PyExc_ValueError,
@@ -1669,6 +1656,7 @@ execute_fancy_ufunc_loop(PyUFuncObject *ufunc,
         NpyAuxData *innerloopdata;
         npy_intp fixed_strides[2*NPY_MAXARGS];
         PyArray_Descr **iter_dtypes;
+        NPY_BEGIN_THREADS_DEF;
 
         /* Validate that the prepare_ufunc_output didn't mess with pointers */
         for (i = nin; i < nop; ++i) {
@@ -1708,9 +1696,7 @@ execute_fancy_ufunc_loop(PyUFuncObject *ufunc,
         strides = NpyIter_GetInnerStrideArray(iter);
         countptr = NpyIter_GetInnerLoopSizePtr(iter);
 
-        if (!needs_api) {
-            NPY_BEGIN_THREADS;
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
 
         NPY_UF_DBG_PRINT("Actual inner loop:\n");
         /* Execute the loop */
@@ -1721,9 +1707,7 @@ execute_fancy_ufunc_loop(PyUFuncObject *ufunc,
                         *countptr, innerloopdata);
         } while (iternext(iter));
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
 
         NPY_AUXDATA_FREE(innerloopdata);
     }
@@ -2749,9 +2733,7 @@ reduce_loop(NpyIter *iter, char **dataptrs, npy_intp *strides,
         return -1;
     }
 
-    if (!needs_api) {
-        NPY_BEGIN_THREADS_THRESHOLDED(NpyIter_GetIterSize(iter));
-    }
+    NPY_BEGIN_THREADS_NDITER(iter);
 
     if (skip_first_count > 0) {
         do {
@@ -2804,9 +2786,7 @@ reduce_loop(NpyIter *iter, char **dataptrs, npy_intp *strides,
     } while (iternext(iter));
 
 finish_loop:
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     return (needs_api && PyErr_Occurred()) ? -1 : 0;
 }
@@ -3133,9 +3113,7 @@ PyUFunc_Accumulate(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *out,
 
         needs_api = NpyIter_IterationNeedsAPI(iter);
 
-        if (!needs_api) {
-            NPY_BEGIN_THREADS_THRESHOLDED(NpyIter_GetIterSize(iter));
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
 
         do {
 
@@ -3165,9 +3143,7 @@ PyUFunc_Accumulate(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *out,
             }
         } while (iternext(iter));
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
     }
     else if (iter == NULL) {
         char *dataptr_copy[3];
@@ -3228,9 +3204,7 @@ PyUFunc_Accumulate(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *out,
             innerloop(dataptr_copy, &count,
                         stride_copy, innerloopdata);
 
-            if (!needs_api) {
-                NPY_END_THREADS;
-            }
+            NPY_END_THREADS;
         }
     }
 
@@ -3281,7 +3255,7 @@ PyUFunc_Reduceat(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *ind,
                             op_axes_arrays[2]};
     npy_uint32 op_flags[3];
     int i, idim, ndim, otype_final;
-    int needs_api, need_outer_iterator;
+    int need_outer_iterator;
 
     NpyIter *iter = NULL;
 
@@ -3526,11 +3500,7 @@ PyUFunc_Reduceat(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *ind,
         stride_copy[1] = stride1;
         stride_copy[2] = stride0;
 
-        needs_api = NpyIter_IterationNeedsAPI(iter);
-
-        if (!needs_api) {
-            NPY_BEGIN_THREADS;
-        }
+        NPY_BEGIN_THREADS_NDITER(iter);
 
         do {
 
@@ -3567,9 +3537,7 @@ PyUFunc_Reduceat(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *ind,
             }
         } while (iternext(iter));
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
     }
     else if (iter == NULL) {
         char *dataptr_copy[3];
@@ -3582,7 +3550,7 @@ PyUFunc_Reduceat(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *ind,
         /* Execute the loop with no iterators */
         npy_intp stride0 = 0, stride1 = PyArray_STRIDE(op[1], axis);
 
-        needs_api = PyDataType_REFCHK(op_dtypes[0]);
+        int needs_api = PyDataType_REFCHK(op_dtypes[0]);
 
         NPY_UF_DBG_PRINT("UFunc: Reduce loop with no iterators\n");
 
@@ -3626,9 +3594,7 @@ PyUFunc_Reduceat(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *ind,
             }
         }
 
-        if (!needs_api) {
-            NPY_END_THREADS;
-        }
+        NPY_END_THREADS;
     }
 
 finish:
@@ -5197,9 +5163,7 @@ ufunc_at(PyUFuncObject *ufunc, PyObject *args)
         i--;
     }
 
-    if (!needs_api) {
-        NPY_END_THREADS;
-    }
+    NPY_END_THREADS;
 
     if (err_msg) {
         PyErr_SetString(PyExc_ValueError, err_msg);
diff --git a/numpy/lib/src/_compiled_base.c b/numpy/lib/src/_compiled_base.c
index 652268f24..f70cd2bab 100644
--- a/numpy/lib/src/_compiled_base.c
+++ b/numpy/lib/src/_compiled_base.c
@@ -1417,6 +1417,9 @@ _packbits( void *In,
     npy_intp out_Nm1;
     int maxi, remain, nonzero, j;
     char *outptr,*inptr;
+    NPY_BEGIN_THREADS_DEF;
+
+    NPY_BEGIN_THREADS_THRESHOLDED(out_N);
 
     outptr = Out;    /* pointer to output buffer */
     inptr  = In;     /* pointer to input buffer */
@@ -1451,6 +1454,8 @@ _packbits( void *In,
         *outptr = build;
         outptr += out_stride;
     }
+
+    NPY_END_THREADS;
     return;
 }
 
@@ -1468,6 +1473,9 @@ _unpackbits(void *In,
     unsigned char mask;
     int i, index;
     char *inptr, *outptr;
+    NPY_BEGIN_THREADS_DEF;
+
+    NPY_BEGIN_THREADS_THRESHOLDED(in_N);
 
     outptr = Out;
     inptr  = In;
@@ -1480,6 +1488,8 @@ _unpackbits(void *In,
         }
         inptr += in_stride;
     }
+
+    NPY_END_THREADS;
     return;
 }