GpuArray sort operation

src/CMakeLists.txt

@@ -55,6 +55,7 @@ gpuarray_collectives_cuda_nccl.c
 gpuarray_buffer_opencl.c
 gpuarray_blas_opencl_clblas.c
 gpuarray_blas_opencl_clblast.c
+gpuarray_sort.c
 )
 
 check_function_exists(strlcat HAVE_STRL)
@@ -126,6 +127,7 @@ set(headers
   gpuarray/kernel.h
   gpuarray/types.h
   gpuarray/util.h
+  gpuarray/sort.h
 )
 
 install(FILES ${headers} DESTINATION include/gpuarray)

src/gpuarray/sort.h

@@ -0,0 +1,64 @@
+#ifndef GPUARRAY_SORT_H
+#define GPUARRAY_SORT_H
+/** \file sort.h
+ *  \brief Sort operations generator.
+ */
+
+#include <gpuarray/buffer.h>
+#include <gpuarray/array.h>
+#include <gpuarray/kernel.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+#ifdef CONFUSE_EMACS
+}
+#endif
+
+#define SHARED_SIZE_LIMIT 	1024U
+#define SAMPLE_STRIDE 		128
+
+typedef struct _GpuSortData {
+  GpuArray       BufKey;
+  GpuArray       BufArg;
+  GpuArray       d_RanksA;
+  GpuArray       d_RanksB;
+  GpuArray       d_LimitsA;
+  GpuArray       d_LimitsB;
+} GpuSortData;
+
+typedef struct _GpuSortConfig {
+  unsigned int   dims;
+  unsigned int   Nfloor;
+  int            Nleft;
+  unsigned int   sortDirFlg;
+  unsigned int   argSortFlg;
+  int            typecodeKey;
+  size_t         typesizeKey;
+  int            typecodeArg;
+  size_t         typesizeArg;
+} GpuSortConfig;
+
+typedef struct _GpuSortBuffers {
+  GpuArray       BufKey;
+  GpuArray       BufArg;
+} GpuSortBuff;
+
+typedef struct _GpuSortKernels {
+  GpuKernel      k_bitonic;
+  GpuKernel      k_ranks;
+  GpuKernel      k_ranks_idxs;
+  GpuKernel      k_merge;
+  GpuKernel      k_merge_global;
+} GpuSortKernels;
+
+
+int GpuArray_sort(GpuArray *r, GpuArray *a, unsigned int sortDir, GpuArray *dstArg);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif