oneAPI Deep Neural Network Library (oneDNN)
1.6.0
Performance library for Deep Learning
17 #ifndef EXAMPLE_UTILS_H
18 #define EXAMPLE_UTILS_H
29 #define COMPLAIN_DNNL_ERROR_AND_EXIT(what, status) \
31 printf("[%s:%d] `%s` returns oneDNN error: %s.\n", __FILE__, __LINE__, \
32 what, dnnl_status2str(status)); \
33 printf("Example failed.\n"); \
37 #define COMPLAIN_EXAMPLE_ERROR_AND_EXIT(complain_fmt, ...) \
39 printf("[%s:%d] Error in the example: " complain_fmt ".\n", __FILE__, \
40 __LINE__, __VA_ARGS__); \
41 printf("Example failed.\n"); \
47 dnnl_status_t s_ = f; \
48 if (s_ != dnnl_success) COMPLAIN_DNNL_ERROR_AND_EXIT(#f, s_); \
55 }
else if (argc == 2) {
57 char *engine_kind_str = argv[1];
58 if (!strcmp(engine_kind_str,
"cpu")) {
60 }
else if (!strcmp(engine_kind_str,
"gpu")) {
63 COMPLAIN_EXAMPLE_ERROR_AND_EXIT(
"%s",
64 "could not find compatible GPU\nPlease run the example "
71 COMPLAIN_EXAMPLE_ERROR_AND_EXIT(
72 "inappropriate engine kind.\n"
73 "Please run the example like this: %s [cpu|gpu].",
80 return "<Unknown engine>";
84 static inline void read_from_dnnl_memory(
void *handle,
dnnl_memory_t mem) {
98 for (
size_t i = 0; i < bytes; ++i) {
99 ((
char *)handle)[i] = ((
char *)ptr)[i];
105 #if DNNL_GPU_RUNTIME == DNNL_RUNTIME_OCL
115 cl_int ret = clEnqueueReadBuffer(
116 q, m, CL_TRUE, 0, bytes, handle, 0, NULL, NULL);
117 if (ret != CL_SUCCESS)
118 COMPLAIN_EXAMPLE_ERROR_AND_EXIT(
119 "clEnqueueReadBuffer failed (status code: %d)", ret);
127 static inline void write_to_dnnl_memory(
void *handle,
dnnl_memory_t mem) {
141 for (
size_t i = 0; i < bytes; ++i) {
142 ((
char *)ptr)[i] = ((
char *)handle)[i];
148 #if DNNL_GPU_RUNTIME == DNNL_RUNTIME_OCL
158 cl_int ret = clEnqueueWriteBuffer(
159 q, m, CL_TRUE, 0, bytes, handle, 0, NULL, NULL);
160 if (ret != CL_SUCCESS)
161 COMPLAIN_EXAMPLE_ERROR_AND_EXIT(
162 "clEnqueueWriteBuffer failed (status code: %d)", ret);
dnnl_status_t DNNL_API dnnl_memory_get_ocl_mem_object(const_dnnl_memory_t memory, cl_mem *mem_object)
Returns an OpenCL memory object associated with a memory object.
An opaque structure to describe an engine.
size_t DNNL_API dnnl_memory_desc_get_size(const dnnl_memory_desc_t *memory_desc)
Returns the size of a memory descriptor.
dnnl_status_t DNNL_API dnnl_engine_get_kind(dnnl_engine_t engine, dnnl_engine_kind_t *kind)
Returns the kind of an engine.
@ dnnl_gpu
GPU engine.
Definition: dnnl_types.h:1743
dnnl_status_t DNNL_API dnnl_stream_destroy(dnnl_stream_t stream)
Destroys an execution stream.
dnnl_engine_kind_t
Kinds of engines.
Definition: dnnl_types.h:1737
dnnl_status_t DNNL_API dnnl_memory_get_data_handle(const_dnnl_memory_t memory, void **handle)
Returns memory object's data handle.
dnnl_status_t DNNL_API dnnl_stream_create(dnnl_stream_t *stream, dnnl_engine_t engine, unsigned flags)
Creates an execution stream.
dnnl_status_t DNNL_API dnnl_memory_get_memory_desc(const_dnnl_memory_t memory, const dnnl_memory_desc_t **memory_desc)
Returns the memory descriptor for a memory object.
@ dnnl_stream_default_flags
Default stream configuration.
Definition: dnnl_types.h:2171
dnnl_status_t DNNL_API dnnl_memory_get_engine(const_dnnl_memory_t memory, dnnl_engine_t *engine)
Returns the engine of a memory object.
size_t DNNL_API dnnl_engine_get_count(dnnl_engine_kind_t kind)
Returns the number of engines of a particular kind.
dnnl_status_t DNNL_API dnnl_stream_get_ocl_command_queue(dnnl_stream_t stream, cl_command_queue *queue)
Returns the OpenCL command queue associated with an execution stream.
@ dnnl_cpu
CPU engine.
Definition: dnnl_types.h:1741
Memory descriptor.
Definition: dnnl_types.h:1140