|
@@ -3935,12 +3935,20 @@ Sparse Row Representation) sparse matrix interface.
|
|
|
TODO
|
|
TODO
|
|
|
@end deftypefun
|
|
@end deftypefun
|
|
|
|
|
|
|
|
-@deftypefun void starpu_multiformat_data_register(starpu_data_handle *handle, uint32_t home_node, void *ptr, uint32_t nobjects, struct starpu_multiformat_data_interface_ops *format_ops);
|
|
|
|
|
|
|
+@deftypefun void starpu_multiformat_data_register(starpu_data_handle *@var{handle}, uint32_t @var{home_node}, void *@var{ptr}, uint32_t @var{nobjects}, struct starpu_multiformat_data_interface_ops *@var{format_ops});
|
|
|
Register a piece of data that can be represented in different ways, depending upon
|
|
Register a piece of data that can be represented in different ways, depending upon
|
|
|
the processing unit that manipulates it. It allows the programmer, for instance, to
|
|
the processing unit that manipulates it. It allows the programmer, for instance, to
|
|
|
use an array of structures when working on a CPU, and a structure of arrays when
|
|
use an array of structures when working on a CPU, and a structure of arrays when
|
|
|
working on a GPU.
|
|
working on a GPU.
|
|
|
|
|
|
|
|
|
|
+@var{nobjects} is the number of elements in the data. @var{format_ops} describes
|
|
|
|
|
+the format itself: @code{cpu_elemsize} is the size of each element on CPUs,
|
|
|
|
|
+@code{opencl_elemsize} is the size of each element on OpenCL devices,
|
|
|
|
|
+@code{cuda_elemsize} is the size of each element on CUDA devices.
|
|
|
|
|
+@code{cpu_to_opencl_cl}, @code{opencl_to_cpu_cl}, @code{cpu_to_cuda_cl}, and
|
|
|
|
|
+@code{cuda_to_cpu_cl} are pointers to codelets which convert between the various
|
|
|
|
|
+formats.
|
|
|
|
|
+
|
|
|
@example
|
|
@example
|
|
|
#define NX 1024
|
|
#define NX 1024
|
|
|
struct point array_of_structs[NX];
|
|
struct point array_of_structs[NX];
|
|
@@ -3950,8 +3958,8 @@ starpu_data_handle handle;
|
|
|
* The conversion of a piece of data is itself a task, though it is created,
|
|
* The conversion of a piece of data is itself a task, though it is created,
|
|
|
* submitted and destroyed by StarPU internals and not by the user. Therefore,
|
|
* submitted and destroyed by StarPU internals and not by the user. Therefore,
|
|
|
* we have to define two codelets.
|
|
* we have to define two codelets.
|
|
|
- * Note that the conversion from the CPU format to the GPU format has to be
|
|
|
|
|
- * executed on the GPU, and the conversion from the GPU to the CPU has to be
|
|
|
|
|
|
|
+ * Note that for now the conversion from the CPU format to the GPU format has to
|
|
|
|
|
+ * be executed on the GPU, and the conversion from the GPU to the CPU has to be
|
|
|
* executed on the CPU.
|
|
* executed on the CPU.
|
|
|
*/
|
|
*/
|
|
|
#ifdef STARPU_USE_OPENCL
|
|
#ifdef STARPU_USE_OPENCL
|
|
@@ -3976,7 +3984,8 @@ struct starpu_multiformat_data_interface_ops format_ops = @{
|
|
|
.cpu_to_opencl_cl = &cpu_to_opencl_cl,
|
|
.cpu_to_opencl_cl = &cpu_to_opencl_cl,
|
|
|
.opencl_to_cpu_cl = &opencl_to_cpu_cl,
|
|
.opencl_to_cpu_cl = &opencl_to_cpu_cl,
|
|
|
#endif
|
|
#endif
|
|
|
- .cpu_elemsize = ...
|
|
|
|
|
|
|
+ .cpu_elemsize = 2 * sizeof(float),
|
|
|
|
|
+ ...
|
|
|
@};
|
|
@};
|
|
|
starpu_multiformat_data_register(handle, 0, &array_of_structs, NX, &format_ops);
|
|
starpu_multiformat_data_register(handle, 0, &array_of_structs, NX, &format_ops);
|
|
|
@end example
|
|
@end example
|
