TF Lite kernel tests cross-compilation enablement using CMake #52110
Conversation
| @@ -41,18 +41,37 @@ build_flatbuffers( | |||
| "" | |||
| ) | |||
|
|
|||
| set(DELEGATE_PROVIDERS_SUPP | |||
| ${TFLITE_SOURCE_DIR}/nnapi/sl/SupportLibrary.cc | |||
|
|
||
| set(DELEGATE_PROVIDERS | ||
| ${DELEGATE_PROVIDERS_SUPP} | ||
| ${TFLITE_SOURCE_DIR}/tools/delegates/gpu_delegate_provider.cc |
| @@ -29,6 +29,8 @@ constexpr char KernelTestDelegateProviders::kUseSimpleAllocator[]; | |||
|
|
|||
| KernelTestDelegateProviders::KernelTestDelegateProviders() | |||
| : delegate_list_util_(¶ms_) { | |||
| TFLITE_TOOLS_CHECK(delegate_list_util_.GetCount() > 0) | |||
There was a problem hiding this comment.
what's happening if we don't have this?
There was a problem hiding this comment.
If there is no delegate available, calling delegate_list_util.AppendCmdlineFlags(&flags) (line 46) appends no command line flags. Since at least --help flag is expected to be available (line 49), the condition returns false and an error is reported (specifically: "help was not found", as implemented in tensorflow/lite/tools/tool_params.cc:53). So this check aims to provide the user with more instructive/helpful error message, should this happen again in the future.
There was a problem hiding this comment.
It turns out this change breaks many tests. You'd better revert it.
bazel test tensorflow/lite/kernels:optional_tensor_test
There was a problem hiding this comment.
The check has been reverted in the latest commits.
| to build the flatc compiler with CMake in advance using the host toolchain: | ||
|
|
||
| ```sh | ||
| cmake ../tensorflow_src/tensorflow/lite/tools/cmake/native_tools/flatbuffers |
There was a problem hiding this comment.
shouldn't we use a separate directory for native_tools and tflite build? I think we'd better mention it.
There was a problem hiding this comment.
The documentation has been updated to make this more obvious.
|
|
||
| ```sh | ||
| cmake -DCMAKE_INSTALL_PREFIX=<native_tools_dir> ../tensorflow_src/tensorflow/lite/tools/cmake/native_tools/flatbuffers | ||
| cmake --build . |
There was a problem hiding this comment.
don't we need to run install command?
There was a problem hiding this comment.
No, running cmake --build . installs the flatc automatically.
google-ml-butler
bot
added
kokoro:force-run
| ##### Specifics of kernel (unit) tests cross-compilation | ||
|
|
||
| Cross-compilation of the unit tests requires flatc compiler for the host architecture. | ||
| For this purpose, there is a CMakeLists located in _tensorflow/lite/tools/cmake/native_tools/flatbuffers_ |
There was a problem hiding this comment.
Is "xxx" valid markdown format?
https://www.markdownguide.org/basic-syntax/
I think it's better to use xxx instead.
There was a problem hiding this comment.
True - updated in the latest commit.
| # purposes the FLATC_INSTALL_PREFIX is already in cache and is just used in this module. | ||
| # In case of standard flatbuffers build (as a dependency) the variable needs to be set. | ||
| if(NOT DEFINED FLATC_INSTALL_PREFIX) | ||
| set(FLATC_INSTALL_PREFIX <INSTALL_DIR> CACHE PATH "Flatc installation directory") |
There was a problem hiding this comment.
There is a space at the ending. Plz remove it.
There was a problem hiding this comment.
Space removed in the latest commit.
| @@ -29,6 +29,8 @@ constexpr char KernelTestDelegateProviders::kUseSimpleAllocator[]; | |||
|
|
|||
| KernelTestDelegateProviders::KernelTestDelegateProviders() | |||
| : delegate_list_util_(¶ms_) { | |||
| TFLITE_TOOLS_CHECK(delegate_list_util_.GetCount() > 0) | |||
There was a problem hiding this comment.
It turns out this change breaks many tests. You'd better revert it.
bazel test tensorflow/lite/kernels:optional_tensor_test
This PR implements the enablement of the TF Lite kernel tests CMake cross-compilation using the natively built flatc compiler (for details please see the updated build_cmake.md file).
As it currently stands in the master branch, the native CMake compilation produces malformed TF Lite kernel tests. Therefore, for testing purposes, this PR also encompasses changes implemented by the currently reviewed PR which fixes it.