-
Notifications
You must be signed in to change notification settings - Fork 130
[SYCL] Remove host from supported device types and backends #1224
[SYCL] Remove host from supported device types and backends #1224
Conversation
This commit removes the host as both a valid backend and a valid device type in the LIT configuration file. Co-authored-by: Sachkov, Alexey <alexey.sachkov@intel.com> Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
@@ -336,7 +314,7 @@ | |||
if 'cpu' in config.target_devices.split(','): | |||
found_at_least_one_device = True | |||
lit_config.note("Test CPU device") | |||
cpu_run_substitute = "env SYCL_DEVICE_FILTER={SYCL_PLUGIN}:cpu,host ".format(SYCL_PLUGIN=config.sycl_be) | |||
cpu_run_substitute = "env SYCL_DEVICE_FILTER={SYCL_PLUGIN}:cpu ".format(SYCL_PLUGIN=config.sycl_be) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should outline those small ,host
changes into a separate PR: I'm concerned that in some cases tests might be launched on host while it is still there. In combination with removal of code pieces like:
if (Queue.get_device().is_host()) {
std::cout << "Skipping test\n";
return 0;
}
That could introduce new failures while we removing tests.
UPD: for example, in #1210 a test simply crashed under environment "SYCL_DEVICE_FILTER=ext_oneapi_level_zero:gpu,host"
. Note: I'm not entirely sure that it is because it attempted to launch the test on host
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I suggest we move the tests that depend on this to be part of this PR to avoid further inter-dependencies.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I suggest we move the tests that depend on this to be part of this PR to avoid further inter-dependencies.
It is hard to grep for such. Actually, I think that the situation I described won't occur in our CI, but it is theoretically possible
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Most cases should work as long as the host device isn't the only device. Even if that happens, it should only happen between those tests being merged and this being merged.
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Changes to subdevice.cpp and subsubdevice.cpp were moved from #1198 to this as the host will be picked up if it is not filtered out, causing the tests to fail. |
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
* [SYCL] Add MAJOR_VERSION to the name of the sycl library on Win (intel#1237) * Update version of sycl library on Win (intel#1285) Co-authored-by: aelovikov-intel <andrei.elovikov@intel.com> Co-authored-by: Nikita <nikita.kornev@intel.com>
This commit removes the host as both a valid backend and a valid device type in the LIT configuration file. Co-authored-by: Sachkov, Alexey <alexey.sachkov@intel.com> Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
This commit removes the host as both a valid backend and a valid device type in the LIT configuration file. Co-authored-by: Sachkov, Alexey <alexey.sachkov@intel.com> Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
…vm-test-suite#1224) This commit removes the host as both a valid backend and a valid device type in the LIT configuration file. Co-authored-by: Sachkov, Alexey <alexey.sachkov@intel.com> Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
This commit removes the host as both a valid backend and a valid device type in the LIT configuration file.