[TF][GPU]: Enable bitwise shift test in TF. #26194
Conversation
There was a problem hiding this comment.
Please fix
RuntimeError: Exception from src/inference/src/cpp/core.cpp:104:
Exception from src/inference/src/dev/plugin.cpp:53:
Check 'false' failed at src/plugins/intel_gpu/src/plugin/program_builder.cpp:185:
[GPU] ProgramBuilder build failed!
Program build failed(0_part_0):
4:1949:2: error: invalid operands to binary expression ('float' and 'float')
DO_ELTWISE;
^~~~~~~~~~
4:1780:2: note: expanded from macro 'DO_ELTWISE'
....
----------------------------- Captured stdout call -----------------------------
OpenVINO version: 2024.4.0-16450-894fd44f4df-refs/pull/26194/head
Creating OV Core Engine...
Reading network files
Loading network
| if ie_device == 'GPU': | ||
| pytest.skip("149424: Bitwise ops are not supported on GPU") | ||
| if ie_device == 'GPU' and input_type in [np.uint64]: | ||
| pytest.skip("149424: uint64 type is not supported on GPU") |
There was a problem hiding this comment.
please create an issue ticket for GPU plugin to support uint64. And point the right ticket number here.
There was a problem hiding this comment.
@pkowalc1, please create separate ticket for uint64 support on GPU and list this ticket number in the skipping code
There was a problem hiding this comment.
@rkazants if you think there is a real need for uint64 support on GPU, please create a ticket and motivate it properly(e.g. maybe there is a business need). Currently, uint64 on GPU will be converted to int32 (info from V. Paramuzow), so there is a support but it it will not work properly in this case.
There was a problem hiding this comment.
The motivation is to be fully aligned with framework cases and capabilities. Our intention is to support not only known models but all other models about which we don't know.
Please create ticket and list here if you are responsible for resolving this task and if you want to change TF FE test sources. Otherwise, I can switch on these tests on GPU in my separate PR and create ticket on my own.
@mlukasze, fyi, I cannot allow to merge this PR without JIRA ticket about this limitation for uint64_t.
There was a problem hiding this comment.
@rkazants This PR is ready to be merged - all types except u64(as discussed above) should be working fine. However, if you want to create your own PR - no problem, please just close this one.
Details:
This PR depends on #26251, which adds support for int16 and uint16 data types in elementwise shift op in gpu plugin.
Also depends on #26383, which fixes a bug related to i16 support for elementwise ops in GPU.