Merge pull request inducer#4 from nouiz/reorg

Numpy-like C-API interface for easier conversion in a separate header.

pygpu/elemwise.py

@@ -271,10 +271,10 @@ def __init__(self, context, arguments, operation, preamble="",
         self.preamble = preamble
 
         self.contig_src = contiguous_kernel.render(preamble=self.preamble,
-                                                   name="elemk",
+                                                   name="elem_contig",
                                                    arguments=self.arguments,
                                                    expression=self.operation)
-        self.contig_k = gpuarray.GpuKernel(self.contig_src, "elemk",
+        self.contig_k = gpuarray.GpuKernel(self.contig_src, "elem_contig",
                                            context=self.context,
                                            cluda=True, **self.flags)
         self._speckey = None
@@ -308,15 +308,16 @@ def prepare_args_contig(self, args, n, offsets):
         kernel_args.insert(0, numpy.asarray(n, dtype='uint32'))
         return kernel_args
 
-    def render_basic(self, nd):
-        return basic_kernel.render(preamble=self.preamble, name="elemk",
+    def render_basic(self, nd, name="elemk"):
+        return basic_kernel.render(preamble=self.preamble, name=name,
                                    nd=nd, arguments=self.arguments,
                                    expression=self.expression)
 
     @lfu_cache()
     def _make_basic(self, nd):
-        src = self.render_basic(nd)
-        return gpuarray.GpuKernel(src, "elemk", context=self.context,
+        name = "elem_" + str(nd)
+        src = self.render_basic(nd, name=name)
+        return gpuarray.GpuKernel(src, name, context=self.context,
                                   cluda=True, **self.flags)
 
     def prepare_args_basic(self, args, n, dims, strs, offsets):

src/CMakeLists.txt

@@ -112,6 +112,7 @@ SET(headers
   compyte/kernel.h
   compyte/types.h
   compyte/util.h
+  compyte/pygpua_cuda.h
 )
 
 INSTALL(FILES ${headers} DESTINATION include/compyte)

src/compyte/numpy_compat.h

@@ -0,0 +1,35 @@
+/*
+ * This file isn't recommanded. Using it make your code not able to work on OpenCL.
+ *
+ * But it allow faster conversion to this new library of existing code
+ */
+#include "extension.h"
+#ifndef COMPYTE_BUFFER_CUDA_H
+#define COMPYTE_BUFFER_CUDA_H
+
+int PyGpuArray_NDIM(PyGpuArrayObject *arr) {
+  return arr->ga.nd;
+}
+size_t *PyGpuArray_DIMS(PyGpuArrayObject *arr) {
+  return arr->ga.dimensions;
+}
+
+ssize_t *PyGpuArray_STRIDES(PyGpuArrayObject* arr) {
+  return arr->ga.strides;
+}
+size_t PyGpuArray_DIM(PyGpuArrayObject* arr, int n) {
+  return arr->ga.dimensions[n];
+}
+ssize_t PyGpuArray_STRIDE(PyGpuArrayObject* arr, int n) {
+  return arr->ga.strides[n];
+}
+//int PyGpuArray_ITEMSIZE(PyGpuArrayObject* arr)
+size_t PyGpuArray_SIZE(PyGpuArrayObject* arr) {
+  size_t size = 1;
+  for(int i=0; i< arr->ga.nd; i++) {
+    size *= arr->ga.dimensions[i];
+  }
+  return size;
+}
+
+#endif