Fix-up usage of NPY_ALLOW_THREADS. Default is now WITH_THREAD but an environment variable NUMPY_NOSMP can be set which will make the default 0

9 files changed, 110 insertions, 62 deletions

diff --git a/numpy/core/blasdot/_dotblas.c b/numpy/core/blasdot/_dotblas.c
index 532c9097b..91b50b43b 100644
--- a/numpy/core/blasdot/_dotblas.c
+++ b/numpy/core/blasdot/_dotblas.c
@@ -356,13 +356,14 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 	    return PyArray_Return(ret);
     }
 
-    Py_BEGIN_ALLOW_THREADS
 
     if (ap2shape == _scalar) {
 	/* Multiplication by a scalar -- Level 1 BLAS */
 	/* if ap1shape is a matrix and we are not contiguous, then we can't
 	   just blast through the entire array using a single
 	   striding factor */
+	NPY_BEGIN_ALLOW_THREADS
+
 	if (typenum == PyArray_DOUBLE) {
 	    if (l == 1) {
 		*((double *)ret->data) = *((double *)ap2->data) * \
@@ -487,9 +488,12 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 		}
 	    }
 	}
+	NPY_END_ALLOW_THREADS
     }
     else if ((ap2shape == _column) && (ap1shape != _matrix)) {
 	int ap1s, ap2s;
+	NPY_BEGIN_ALLOW_THREADS
+
 	ap2s = ap2->strides[0] / ap2->descr->elsize;
 	if (ap1shape == _row) {
 	    ap1s = ap1->strides[1] / ap1->descr->elsize;
@@ -517,6 +521,7 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 	    cblas_cdotu_sub(l, (float *)ap1->data, ap1s, 
 			    (float *)ap2->data, ap2s, (float *)ret->data);
 	}
+	NPY_END_ALLOW_THREADS
     }
     else if (ap1shape == _matrix && ap2shape != _matrix) {
 	/* Matrix vector multiplication -- Level 2 BLAS */
@@ -531,6 +536,7 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 	    ap1 = (PyArrayObject *)new;
 	    if (new == NULL) goto fail;
 	}
+	NPY_BEGIN_ALLOW_THREADS
 	if (PyArray_ISCONTIGUOUS(ap1)) {
             Order = CblasRowMajor;
             lda = (ap1->dimensions[1] > 1 ? ap1->dimensions[1] : 1);
@@ -566,6 +572,7 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 			(float *)ap2->data, ap2s, zeroF, 
 			(float *)ret->data, 1);
 	}
+	NPY_END_ALLOW_THREADS
     }
     else if (ap1shape != _matrix && ap2shape == _matrix) {
 	/* Vector matrix multiplication -- Level 2 BLAS */
@@ -579,6 +586,7 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 	    ap2 = (PyArrayObject *)new;
 	    if (new == NULL) goto fail;
 	}
+	NPY_BEGIN_ALLOW_THREADS
 	if (PyArray_ISCONTIGUOUS(ap2)) {
 	    Order = CblasRowMajor;
             lda = (ap2->dimensions[1] > 1 ? ap2->dimensions[1] : 1);
@@ -617,6 +625,7 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 			oneF, (float *)ap2->data, lda,
 			(float *)ap1->data, ap1s, zeroF, (float *)ret->data, 1);
 	}
+	NPY_END_ALLOW_THREADS
     }
     else { /* (ap1->nd == 2 && ap2->nd == 2) */
 	/* Matrix matrix multiplication -- Level 3 BLAS */  
@@ -644,6 +653,8 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 	    ap1 = (PyArrayObject *)new;
 	    if (new == NULL) goto fail;
 	}
+
+	NPY_BEGIN_ALLOW_THREADS
 	
 	Order = CblasRowMajor;
 	Trans1 = CblasNoTrans;
@@ -682,9 +693,9 @@ dotblas_matrixproduct(PyObject *dummy, PyObject *args)
 			(float *)ap2->data, ldb,
 			zeroF, (float *)ret->data, ldc);
 	}
+	NPY_END_ALLOW_THREADS
     }
 
-    Py_END_ALLOW_THREADS
 
     Py_DECREF(ap1);
     Py_DECREF(ap2);
@@ -800,7 +811,7 @@ dotblas_innerproduct(PyObject *dummy, PyObject *args)
 				       (prior2 > prior1 ? ap2 : ap1));
     
     if (ret == NULL) goto fail;
-    Py_BEGIN_ALLOW_THREADS
+    NPY_BEGIN_ALLOW_THREADS
     memset(ret->data, 0, PyArray_NBYTES(ret));
 
     if (ap2->nd == 0) {
@@ -933,7 +944,7 @@ dotblas_innerproduct(PyObject *dummy, PyObject *args)
 			zeroF, (float *)ret->data, ldc);
 	}
     }
-    Py_END_ALLOW_THREADS
+    NPY_END_ALLOW_THREADS
     Py_DECREF(ap1);
     Py_DECREF(ap2);
     return PyArray_Return(ret);
@@ -1015,7 +1026,7 @@ static PyObject *dotblas_vdot(PyObject *dummy, PyObject *args) {
     ret = (PyArrayObject *)PyArray_SimpleNew(0, dimensions, typenum);
     if (ret == NULL) goto fail;
 
-    Py_BEGIN_ALLOW_THREADS
+    NPY_BEGIN_ALLOW_THREADS
 
     /* Dot product between two vectors -- Level 1 BLAS */
     if (typenum == PyArray_DOUBLE) {
@@ -1035,7 +1046,7 @@ static PyObject *dotblas_vdot(PyObject *dummy, PyObject *args) {
 			(float *)ap2->data, 1, (float *)ret->data);
     }
 
-    Py_END_ALLOW_THREADS
+    NPY_END_ALLOW_THREADS
 
     Py_DECREF(ap1);
     Py_DECREF(ap2);
diff --git a/numpy/core/include/numpy/arrayobject.h b/numpy/core/include/numpy/arrayobject.h
index 67f0cb321..8f4f53bf9 100644
--- a/numpy/core/include/numpy/arrayobject.h
+++ b/numpy/core/include/numpy/arrayobject.h
@@ -982,16 +982,20 @@ typedef int (PyArray_FinalizeFunc)(PyArrayObject *, PyObject *);
 #define PyArray_ISALIGNED(m) PyArray_CHKFLAGS(m, NPY_ALIGNED)
 
 
-#if defined(ALLOW_THREADS)
-#define NPY_BEGIN_THREADS_DEF PyThreadState *_save;
+#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 _save = PyEval_SaveThread();
-#define NPY_END_THREADS   PyEval_RestoreThread(_save);
+#define NPY_END_THREADS   if (_save) PyEval_RestoreThread(_save);
 #define NPY_BEGIN_THREADS_DESCR(dtype) if (!((dtype)->hasobject)) NPY_BEGIN_THREADS
 #define NPY_END_THREADS_DESCR(dtype) if (!((dtype)->hasobject)) NPY_END_THREADS
 #define NPY_ALLOW_C_API_DEF  PyGILState_STATE __save__;
 #define NPY_ALLOW_C_API      __save__ = PyGILState_Ensure();
 #define NPY_DISABLE_C_API    PyGILState_Release(__save__);
 #else
+#define NPY_BEGIN_ALLOW_THREADS 
+#define NPY_END_ALLOW_THREADS 
 #define NPY_BEGIN_THREADS_DEF
 #define NPY_BEGIN_THREADS
 #define NPY_END_THREADS
@@ -1331,7 +1335,7 @@ typedef struct {
 #define NPY_OPPBYTE NPY_BIG
 #endif
 
-#define PyArray_ISNBO(arg) ((arg) != PyArray_OPPBYTE)
+#define PyArray_ISNBO(arg) ((arg) != NPY_OPPBYTE)
 #define PyArray_IsNativeByteOrder PyArray_ISNBO
 #define PyArray_ISNOTSWAPPED(m) PyArray_ISNBO(PyArray_DESCR(m)->byteorder)
 #define PyArray_ISBYTESWAPPED(m) (!PyArray_ISNOTSWAPPED(m))
@@ -1494,7 +1498,7 @@ typedef struct {
 	} while(0)
 
 /* Copy should always return contiguous array */
-#define PyArray_Copy(obj) PyArray_NewCopy(obj, PyArray_CORDER)
+#define PyArray_Copy(obj) PyArray_NewCopy(obj, NPY_CORDER)
 
 #define PyArray_FromObject(op, type, min_depth, max_depth)		\
 	PyArray_FromAny(op, PyArray_DescrFromType(type), min_depth,	\
diff --git a/numpy/core/include/numpy/old_defines.h b/numpy/core/include/numpy/old_defines.h
index adb6a008d..1c270e04b 100644
--- a/numpy/core/include/numpy/old_defines.h
+++ b/numpy/core/include/numpy/old_defines.h
@@ -134,7 +134,7 @@
 #define PyArray_NSCALARKINDS   NPY_NSCALARKINDS 
 
 #define PyArray_ANYORDER     NPY_ANYORDER     
-#define PyArray_CORDER       NPY_CORDER       
+#define PyArray_CORDER       NPY_CORDER
 #define PyArray_FORTRANORDER NPY_FORTRANORDER 
 #define PyArray_ORDER        NPY_ORDER        
 
diff --git a/numpy/core/include/numpy/ufuncobject.h b/numpy/core/include/numpy/ufuncobject.h
index 0df3997e3..66162b904 100644
--- a/numpy/core/include/numpy/ufuncobject.h
+++ b/numpy/core/include/numpy/ufuncobject.h
@@ -164,12 +164,12 @@ typedef struct {
 } PyUFuncReduceObject;
 
 
-#if defined(ALLOW_THREADS)
-#define LOOP_BEGIN_THREADS if (!(loop->obj)) {_save = PyEval_SaveThread();}
-#define LOOP_END_THREADS   if (!(loop->obj)) {PyEval_RestoreThread(_save);}
+#if NPY_ALLOW_THREADS
+#define NPY_LOOP_BEGIN_THREADS if (!(loop->obj)) {_save = PyEval_SaveThread();}
+#define NPY_LOOP_END_THREADS   if (!(loop->obj)) {PyEval_RestoreThread(_save);}
 #else
-#define LOOP_BEGIN_THREADS
-#define LOOP_END_THREADS
+#define NPY_LOOP_BEGIN_THREADS
+#define NPY_LOOP_END_THREADS
 #endif
 
 #define PyUFunc_One 1
diff --git a/numpy/core/setup.py b/numpy/core/setup.py
index fa42f2aaf..aa6450207 100644
--- a/numpy/core/setup.py
+++ b/numpy/core/setup.py
@@ -35,8 +35,16 @@ def configuration(parent_package='',top_path=None):
                                         library_dirs = default_lib_dirs)
             if not result:
                 raise "ERROR: Failed to test configuration"
-            moredefs = []
-
+            # Perhaps a fancier check is in order here.
+            try:
+                nosmp = os.environ['NPY_NOSMP']
+                nosmp = 1
+            except KeyError:
+                nosmp = 0
+            if nosmp:
+                moredefs = [('NPY_ALLOW_THREADS', '0')]
+            else:
+                moredefs = [('NPY_ALLOW_THREADS','WITH_THREAD')]
             #
             mathlibs = []
             tc = testcode_mathlib()
diff --git a/numpy/core/src/arrayobject.c b/numpy/core/src/arrayobject.c
index c91d46a9e..580932869 100644
--- a/numpy/core/src/arrayobject.c
+++ b/numpy/core/src/arrayobject.c
@@ -1242,7 +1242,7 @@ PyArray_NewCopy(PyArrayObject *m1, NPY_ORDER fortran)
 	PyArrayObject *ret;
 	if (fortran == PyArray_ANYORDER) 
 		fortran = PyArray_ISFORTRAN(m1);
-        
+
 	Py_INCREF(m1->descr);
 	ret = (PyArrayObject *)PyArray_NewFromDescr(m1->ob_type,
 						    m1->descr,
@@ -5076,6 +5076,7 @@ PyArray_NewFromDescr(PyTypeObject *subtype, PyArray_Descr *descr, int nd,
                                 goto fail;
                         }
 			*/
+
 			memset(data, 0, sd);
 		}
 	}
@@ -7114,6 +7115,8 @@ _broadcast_cast(PyArrayObject *out, PyArrayObject *in,
 	PyArray_CopySwapNFunc *ocopyfunc, *icopyfunc;	
 	char *obptr;
 
+	NPY_BEGIN_THREADS_DEF
+
  	delsize = PyArray_ITEMSIZE(out);
 	selsize = PyArray_ITEMSIZE(in);
 	multi = (PyArrayMultiIterObject *)PyArray_MultiIterNew(2, out, in);
@@ -7150,6 +7153,12 @@ _broadcast_cast(PyArrayObject *out, PyArrayObject *in,
 	if (in->descr->hasobject) 
 		memset(buffers[1], 0, N*selsize);
 
+#if NPY_ALLOW_THREADS 
+	if (!in->descr->hasobject && !out->descr->hasobject) {
+		NPY_BEGIN_THREADS		       
+			}
+#endif
+
 	while(multi->index < multi->size) {
 		_strided_buffered_cast(multi->iters[0]->dataptr,
 				       ostrides,
@@ -7161,6 +7170,11 @@ _broadcast_cast(PyArrayObject *out, PyArrayObject *in,
 				       castfunc, out, in);
 		PyArray_MultiIter_NEXT(multi);
 	}
+#if NPY_ALLOW_THREADS
+	if (!in->descr->hasobject && !out->descr->hasobject) {
+		NPY_END_THREADS		       
+			}
+#endif
 	Py_DECREF(multi);
 	if (in->descr->hasobject) {
 		obptr = buffers[1];
@@ -7197,6 +7211,8 @@ PyArray_CastTo(PyArrayObject *out, PyArrayObject *mp)
 	PyArray_VectorUnaryFunc *castfunc=NULL;
 	int mpsize = PyArray_SIZE(mp);
 	int iswap, oswap;
+	
+	NPY_BEGIN_THREADS_DEF
 
 	if (mpsize == 0) return 0;
 	if (!PyArray_ISWRITEABLE(out)) {
@@ -7214,7 +7230,17 @@ PyArray_CastTo(PyArrayObject *out, PyArrayObject *mp)
 			  (PyArray_ISFARRAY_RO(mp) && PyArray_ISFARRAY(out)));
 
 	if (simple) {
+
+#if NPY_ALLOW_THREADS
+		if (!mp->descr->hasobject && !out->descr->hasobject) {
+			NPY_BEGIN_THREADS }
+#endif
 		castfunc(mp->data, out->data, mpsize, mp, out);
+
+#if NPY_ALLOW_THREADS
+		if (!mp->descr->hasobject && !out->descr->hasobject) {
+			NPY_END_THREADS   }
+#endif
 		return 0;
 	}
 
diff --git a/numpy/core/src/arraytypes.inc.src b/numpy/core/src/arraytypes.inc.src
index 1f4ee91d6..c126b8580 100644
--- a/numpy/core/src/arraytypes.inc.src
+++ b/numpy/core/src/arraytypes.inc.src
@@ -575,13 +575,12 @@ VOID_setitem(PyObject *op, char *ip, PyArrayObject *ap)
 */
 static void 
 @from@_to_@to@(register @fromtyp@ *ip, register @totyp@ *op, register intp n, 
-	       PyArrayObject *aip, PyArrayObject *aop) {
-	Py_BEGIN_ALLOW_THREADS
+	       PyArrayObject *aip, PyArrayObject *aop) 
+{
 	while (n--) {
 		*op++ = (@totyp@)*ip;
 		@incr@;
 	}
-	Py_END_ALLOW_THREADS
 }
 /**end repeat**/
 
@@ -591,12 +590,11 @@ static void
 */
 static void 
 @from@_to_BOOL(register @fromtyp@ *ip, register Bool *op, register intp n, 
-	       PyArrayObject *aip, PyArrayObject *aop) {
-	Py_BEGIN_ALLOW_THREADS
+	       PyArrayObject *aip, PyArrayObject *aop) 
+{
 	while (n--) {
 		*op++ = (Bool)(*ip++ != FALSE);
 	}
-	Py_END_ALLOW_THREADS
 }
 /**end repeat**/
 
@@ -606,13 +604,12 @@ static void
 */
 static void 
 @from@_to_BOOL(register @fromtyp@ *ip, register Bool *op, register intp n, 
-	       PyArrayObject *aip, PyArrayObject *aop) {
-	Py_BEGIN_ALLOW_THREADS
+	       PyArrayObject *aip, PyArrayObject *aop) 
+{
 	while (n--) {
 		*op = (Bool)(((*ip).real != FALSE) || ((*ip).imag != FALSE));
 		op++; ip++;
 	}
-	Py_END_ALLOW_THREADS
 }
 /**end repeat**/
 
@@ -624,11 +621,9 @@ static void
 BOOL_to_@to@(register Bool *ip, register @totyp@ *op, register intp n, 
 	     PyArrayObject *aip, PyArrayObject *aop) 
 {
-	Py_BEGIN_ALLOW_THREADS
 	while (n--) {
 		*op++ = (@totyp@)(*ip++ != FALSE);
 	}
-	Py_END_ALLOW_THREADS
 }
 /**end repeat**/
 
@@ -643,12 +638,11 @@ static void
 @from@_to_@to@(register @fromtyp@ *ip, register @totyp@ *op, register intp n, 
 	       PyArrayObject *aip, PyArrayObject *aop) 
 {	
-	Py_BEGIN_ALLOW_THREADS
 	while (n--) {
 		*op++ = (@totyp@)*ip++;
 		*op++ = 0.0;
 	}
-	Py_END_ALLOW_THREADS
+
 }
 /**end repeat**/
 
@@ -661,13 +655,13 @@ static void
 */
 static void 
 @from@_to_@to@(register @fromtyp@ *ip, register @totyp@ *op, register intp n, 
-	       PyArrayObject *aip, PyArrayObject *aop) {
-	Py_BEGIN_ALLOW_THREADS
+	       PyArrayObject *aip, PyArrayObject *aop) 
+{
 	n <<= 1;
 	while (n--) {
 		*op++ = (@totyp@)*ip++;
 	}
-	Py_END_ALLOW_THREADS
+
 }
 /**end repeat**/
 
diff --git a/numpy/core/src/multiarraymodule.c b/numpy/core/src/multiarraymodule.c
index 90553bc6c..f2368ba88 100644
--- a/numpy/core/src/multiarraymodule.c
+++ b/numpy/core/src/multiarraymodule.c
@@ -2819,7 +2819,7 @@ PyArray_CopyAndTranspose(PyObject *op)
 		return NULL;
 	}
 	/* do 2-d loop */
-	Py_BEGIN_ALLOW_THREADS
+	NPY_BEGIN_ALLOW_THREADS
 	optr = PyArray_DATA(ret);
 	str2 = elsize*dims[0];
 	for (i=0; i<dims[0]; i++) {
@@ -2831,7 +2831,7 @@ PyArray_CopyAndTranspose(PyObject *op)
 			iptr += str2;
 		}
 	}
-	Py_END_ALLOW_THREADS
+	NPY_END_ALLOW_THREADS
 	Py_DECREF(arr);
 	return ret;
 }
@@ -4982,7 +4982,7 @@ PyArray_FromString(char *data, intp slen, PyArray_Descr *dtype,
 						     dtype, 1, &n, NULL,
 						     NULL, 0, NULL);
 			if (ret == NULL) return NULL;
-			Py_BEGIN_ALLOW_THREADS
+			NPY_BEGIN_ALLOW_THREADS
 			ptr = data;
 			dptr = ret->data;
 			for (index=0; index < n; index++) {
@@ -5004,7 +5004,7 @@ PyArray_FromString(char *data, intp slen, PyArray_Descr *dtype,
 				PyArray_DIM(ret,0) = nread;
 			
 			}
-			Py_END_ALLOW_THREADS
+			NPY_END_ALLOW_THREADS
 		}
 		else {
 #define _FILEBUFNUM 4096
@@ -5022,7 +5022,7 @@ PyArray_FromString(char *data, intp slen, PyArray_Descr *dtype,
 						     NULL, NULL, 
 						     0, NULL);
 			if (ret==NULL) return NULL;
-			Py_BEGIN_ALLOW_THREADS
+			NPY_BEGIN_ALLOW_THREADS
 			totalbytes = bytes = size * dtype->elsize;
 			dptr = ret->data;
 			ptr = data;
@@ -5048,7 +5048,7 @@ PyArray_FromString(char *data, intp slen, PyArray_Descr *dtype,
 						    nread*ret->descr->elsize);
 			PyArray_DIM(ret,0) = nread;
 #undef _FILEBUFNUM
-			Py_END_ALLOW_THREADS
+			NPY_END_ALLOW_THREADS
 		}
 	}
 	return (PyObject *)ret;
@@ -5230,9 +5230,9 @@ PyArray_FromFile(FILE *fp, PyArray_Descr *typecode, intp num, char *sep)
 							  NULL, NULL, 
 							  0, NULL);
 		if (r==NULL) return NULL;
-		Py_BEGIN_ALLOW_THREADS
+		NPY_BEGIN_ALLOW_THREADS
 		nread = fread(r->data, typecode->elsize, num, fp);
-		Py_END_ALLOW_THREADS
+		NPY_END_ALLOW_THREADS
 	}
 	else {  /* character reading */
 		intp i;
@@ -5257,7 +5257,7 @@ PyArray_FromFile(FILE *fp, PyArray_Descr *typecode, intp num, char *sep)
 						     NULL, NULL, 
 						     0, NULL);
 			if (r==NULL) return NULL;
-			Py_BEGIN_ALLOW_THREADS			
+			NPY_BEGIN_ALLOW_THREADS			
 			dptr = r->data;
 			for (i=0; i < num; i++) {
 				if (done) break;
@@ -5266,7 +5266,7 @@ PyArray_FromFile(FILE *fp, PyArray_Descr *typecode, intp num, char *sep)
 				nread += 1;
 				dptr += r->descr->elsize;
 			}
-			Py_END_ALLOW_THREADS			
+			NPY_END_ALLOW_THREADS			
 			if (PyErr_Occurred()) {
 				Py_DECREF(r);
 				return NULL;
@@ -5287,7 +5287,7 @@ PyArray_FromFile(FILE *fp, PyArray_Descr *typecode, intp num, char *sep)
 						     NULL, NULL, 
 						     0, NULL);
 			if (r==NULL) return NULL;
-			Py_BEGIN_ALLOW_THREADS	
+			NPY_BEGIN_ALLOW_THREADS	
 			totalbytes = bytes = size * typecode->elsize;
 			dptr = r->data;
 			while (!done) {
@@ -5313,7 +5313,7 @@ PyArray_FromFile(FILE *fp, PyArray_Descr *typecode, intp num, char *sep)
 			r->data = PyDataMem_RENEW(r->data, nread*r->descr->elsize);
 			PyArray_DIM(r,0) = nread;
 			num = nread;
-			Py_END_ALLOW_THREADS
+			NPY_END_ALLOW_THREADS
 #undef _FILEBUFNUM
 		}
 		if (PyErr_Occurred()) {
@@ -6380,6 +6380,9 @@ PyMODINIT_FUNC initmultiarray(void) {
 	s = PyString_FromString("3.0");
 	PyDict_SetItemString(d, "__version__", s);
 	Py_DECREF(s);
+	s = PyInt_FromLong(NPY_ALLOW_THREADS);
+	PyDict_SetItemString(d, "ALLOW_THREADS", s);
+	Py_DECREF(s);
         Py_INCREF(&PyArray_Type);
 	PyDict_SetItemString(d, "ndarray", (PyObject *)&PyArray_Type);
         Py_INCREF(&PyArrayIter_Type);
diff --git a/numpy/core/src/ufuncobject.c b/numpy/core/src/ufuncobject.c
index 8e7e2a159..573bff639 100644
--- a/numpy/core/src/ufuncobject.c
+++ b/numpy/core/src/ufuncobject.c
@@ -1430,12 +1430,12 @@ PyUFunc_GenericFunction(PyUFuncObject *self, PyObject *args,
 {
 	PyUFuncLoopObject *loop;
 	int i;
-        BEGIN_THREADS_DEF
+        NPY_BEGIN_THREADS_DEF
 
 	if (!(loop = construct_loop(self, args, mps))) return -1;
         if (loop->notimplemented) {ufuncloop_dealloc(loop); return -2;}
 
-	LOOP_BEGIN_THREADS
+	NPY_LOOP_BEGIN_THREADS
 
 	switch(loop->meth) {
 	case ONE_UFUNCLOOP:
@@ -1685,13 +1685,13 @@ PyUFunc_GenericFunction(PyUFuncObject *self, PyObject *args,
         }
         }
                 
-        LOOP_END_THREADS
+        NPY_LOOP_END_THREADS
                 
         ufuncloop_dealloc(loop);
 	return 0;
 
  fail:
-        LOOP_END_THREADS
+        NPY_LOOP_END_THREADS
 
 	if (loop) ufuncloop_dealloc(loop);
 	return -1;
@@ -1988,14 +1988,14 @@ PyUFunc_Reduce(PyUFuncObject *self, PyArrayObject *arr, int axis, int otype)
         PyUFuncReduceObject *loop;
         intp i, n;
         char *dptr;
-        BEGIN_THREADS_DEF
+        NPY_BEGIN_THREADS_DEF
         	
         /* Construct loop object */
         loop = construct_reduce(self, &arr, axis, otype, UFUNC_REDUCE, 0,
 				"reduce");
 	if (!loop) return NULL;
 
-        LOOP_BEGIN_THREADS
+        NPY_LOOP_BEGIN_THREADS
         switch(loop->meth) {
         case ZERODIM_REDUCELOOP:
 		/* fprintf(stderr, "ZERO..%d\n", loop->size); */
@@ -2106,7 +2106,7 @@ PyUFunc_Reduce(PyUFuncObject *self, PyArrayObject *arr, int axis, int otype)
                 }
         }
 
-        LOOP_END_THREADS
+        NPY_LOOP_END_THREADS
 
         ret = loop->ret;
 	/* Hang on to this reference -- will be decref'd with loop */
@@ -2115,7 +2115,7 @@ PyUFunc_Reduce(PyUFuncObject *self, PyArrayObject *arr, int axis, int otype)
         return (PyObject *)ret;
 
  fail:
-        LOOP_END_THREADS
+        NPY_LOOP_END_THREADS
 
         if (loop) ufuncreduce_dealloc(loop);
         return NULL;
@@ -2130,13 +2130,14 @@ PyUFunc_Accumulate(PyUFuncObject *self, PyArrayObject *arr, int axis,
         PyUFuncReduceObject *loop;
         intp i, n;
         char *dptr;
+	NPY_BEGIN_THREADS_DEF
         
         /* Construct loop object */
         loop = construct_reduce(self, &arr, axis, otype, UFUNC_ACCUMULATE, 0,
 				"accumulate");
 	if (!loop) return NULL;
 
-	LOOP_BEGIN_THREADS
+	NPY_LOOP_BEGIN_THREADS
         switch(loop->meth) {
         case ZERODIM_REDUCELOOP: /* Accumulate */
 		/* fprintf(stderr, "ZERO..%d\n", loop->size); */
@@ -2251,7 +2252,7 @@ PyUFunc_Accumulate(PyUFuncObject *self, PyArrayObject *arr, int axis,
                 }
         }
 
-	LOOP_END_THREADS
+	NPY_LOOP_END_THREADS
         ret = loop->ret;
 	/* Hang on to this reference -- will be decref'd with loop */
         Py_INCREF(ret);
@@ -2259,7 +2260,7 @@ PyUFunc_Accumulate(PyUFuncObject *self, PyArrayObject *arr, int axis,
         return (PyObject *)ret;
 
  fail:
-	LOOP_END_THREADS
+	NPY_LOOP_END_THREADS
 
         if (loop) ufuncreduce_dealloc(loop);
         return NULL;
@@ -2295,6 +2296,7 @@ PyUFunc_Reduceat(PyUFuncObject *self, PyArrayObject *arr, PyArrayObject *ind,
 	intp mm=arr->dimensions[axis]-1;
 	intp n, i, j;
 	char *dptr;
+	NPY_BEGIN_THREADS_DEF
 
 	/* Check for out-of-bounds values in indices array */		
 	for (i=0; i<nn; i++) {
@@ -2312,7 +2314,7 @@ PyUFunc_Reduceat(PyUFuncObject *self, PyArrayObject *arr, PyArrayObject *ind,
 				"reduceat");
 	if (!loop) return NULL;
 
-	LOOP_BEGIN_THREADS
+	NPY_LOOP_BEGIN_THREADS
 	switch(loop->meth) {
 	/* zero-length index -- return array immediately */
 	case ZERODIM_REDUCELOOP:
@@ -2401,7 +2403,7 @@ PyUFunc_Reduceat(PyUFuncObject *self, PyArrayObject *arr, PyArrayObject *ind,
 		break;
 	}
 
-	LOOP_END_THREADS
+	NPY_LOOP_END_THREADS
 	
         ret = loop->ret;
 	/* Hang on to this reference -- will be decref'd with loop */
@@ -2410,7 +2412,7 @@ PyUFunc_Reduceat(PyUFuncObject *self, PyArrayObject *arr, PyArrayObject *ind,
         return (PyObject *)ret;
 	
  fail:
-	LOOP_END_THREADS
+	NPY_LOOP_END_THREADS
 
         if (loop) ufuncreduce_dealloc(loop);
 	return NULL;