Transitioning from Intel MKL-DNN to oneDNN

Intel® oneAPI Deep Neural Network Developer Guide and Reference

Download PDF

ID 768875

Date 2/28/2024

Version

Public

A newer version of this document is available. Customers should click here to go to the newest version.

Visible to Intel only — GUID: GUID-07A2CF88-0B99-458F-981E-3CEACE8CAC6D

View Details

Transitioning from Intel MKL-DNN to oneDNN

To simplify library naming and differentiate it from Intel MKL, starting with version 1.1 the library name is changed to Deep Neural Network Library (DNNL).

NOTE:

The subsequent library name change to oneAPI Deep Neural Network Library (oneDNN) does not impact API, environment variables, or build options.

There are some incompatibilities, described in the Broken Compatibility with Intel MKL-DNN section below.

Summary of Changes

In short, the migration can be as simple as just replacing all MKLDNN/mkldnn substrings with DNNL/dnnl.

Source Code Changes

All headers, functions, types, and namespaces are renamed by replacing mkldnn with dnnl. The macros with MKLDNN are replaced with DNNL counterparts.

An example of code with Intel MKL-DNN v1.0:

#include "mkldnn.hpp"

using namespace mkldnn;

mkldnn_memory_desc_t md;
if (md.format_kind == mkldnn_blocked) {}
conv.exec(stream, {{MKLDNN_ARGS_SRC, src}, ...});

The updated example with DNNL v1.1:

#include "dnnl.hpp"

using namespace dnnl;

dnnl_memory_desc_t md;
if (md.format_kind == dnnl_blocked) {}
conv.exec(stream, {{DNNL_ARGS_SRC, src}, ...});

To API compatibility with Intel MKL-DNN is based on include/mkldnn_dnnl_mangling.h header file that maps all Intel MKL-DNN symbols to DNNL ones using C preprocessor:

// ...
#define mkldnn_memory_desc_t           dnnl_memory_desc_t
#define mkldnn_memory_desc_init_by_tag dnnl_memory_desc_init_by_tag
// ...

This file is included to every former Intel MKL-DNN header files (for instance, see mkldnn.h) along with the DNNL counterpart.

Build Process

The changes to the build options are similar to the ones in the source code. All the options and namespaces with MKLDNN are replaced with DNNL :

Intel MKL-DNN	DNNL
MKLDNN (namespace)	DNNL (namespace)
MKLDNN_ARCH_OPT_FLAGS	DNNL_ARCH_OPT_FLAGS
MKLDNN_BUILD_EXAMPLES	DNNL_BUILD_EXAMPLES
MKLDNN_BUILD_FOR_CI	DNNL_BUILD_FOR_CI
MKLDNN_BUILD_TESTS	DNNL_BUILD_TESTS
MKLDNN_CPU_RUNTIME	DNNL_CPU_RUNTIME
MKLDNN_ENABLE_CONCURRENT_EXEC	DNNL_ENABLE_CONCURRENT_EXEC
MKLDNN_ENABLE_JIT_PROFILING	DNNL_ENABLE_JIT_PROFILING
MKLDNN_GPU_BACKEND	DNNL_GPU_BACKEND
MKLDNN_GPU_RUNTIME	DNNL_GPU_RUNTIME
MKLDNN_INSTALL_MODE	DNNL_INSTALL_MODE
MKLDNN_LIBRARY_TYPE	DNNL_LIBRARY_TYPE
MKLDNN_THREADING	DNNL_THREADING
MKLDNN_USE_CLANG_SANITIZER	DNNL_USE_CLANG_SANITIZER
MKLDNN_VERBOSE	DNNL_VERBOSE
MKLDNN_WERROR	DNNL_WERROR

You must switch to DNNL build options as well:

# Through find package
find_package(dnnl DNNL CONFIG REQUIRED)
target_link_libraries(project_app DNNL::dnnl)

# Or direct sub-project inclusion
add_subdirectory(${DNNL_DIR} DNNL)
include_directories(${DNNL_DIR}/include)
target_link_libraries(project_app dnnl)

Broken Compatibility with Intel MKL-DNN

Unfortunately, full compatibility after renaming is not implemented. DNNL is not compatible with Intel MKL-DNN in the following things:

ABI: An application or a library built with Intel MKL-DNN cannot switch on using DNNL without recompiling.
Microsoft* Visual Studio Solution files that are generated by cmake will be based on DNNL name only (e.g. MKLDNN.sln becomes DNNL.sln, and the former is no more generated).

Information for Developers

The implementation of renaming (several patches and scripts that rename the library) can be found here.

Select Your Language

Using Intel.com Search

Quick Links

Recent Searches

Advanced Search

Only search in

Intel® oneAPI Deep Neural Network Developer Guide and Reference

Transitioning from Intel MKL-DNN to oneDNN

Summary of Changes

Broken Compatibility with Intel MKL-DNN

Information for Developers