InvokeSimd: Added named barrier and SLM access tests

[fveselov]

No description provided.

[v-klochkov]

I have few non-blocking comments and one more general concern.
If I remember it correctly, slm_init() function will not work in such context and may require passing local accessor instead and re-working the test.

Sorry, I don't remember all the details, it will take some time.

[fveselov]

I have few non-blocking comments and one more general concern. If I remember it correctly, slm_init() function will not work in such context and may require passing local accessor instead and re-working the test.

Sorry, I don't remember all the details, it will take some time.

If I understand correctly, the interface for allocating SLM outside of invoke_simd isn't yet settled.

At the moment I added following line to suppress these tests:
// REQUIRES: TEMPORARY_DISABLED

@v-klochkov, Should we wait till the Jira issue resolved or PR can be merged now?

[v-klochkov]

I have few non-blocking comments and one more general concern. If I remember it correctly, slm_init() function will not work in such context and may require passing local accessor instead and re-working the test.
Sorry, I don't remember all the details, it will take some time.

If I understand correctly, the interface for allocating SLM outside of invoke_simd isn't yet settled.

At the moment I added following line to suppress these tests: // REQUIRES: TEMPORARY_DISABLED

@v-klochkov, Should we wait till the Jira issue resolved or PR can be merged now?

It probably, should not wait JIRA resolution.

Does the test pass with UseSLM = false? If Yes, then it makes sense to to enable the test for that path and add something like:
// TODO: Enable the sub-tests with UseSLM=true after the issue with SLM initialization is fixed.

[v-klochkov]

IMO, the tests are a bit raw - they may use some functionality like SLM allocation or barriers not quite correctly, but that is caused by uncertainty in how that needs to be implemented in ESIMDS. Tests are:
a) close to any user would try to do,
b) good starting point to start developing/fixing ESIMD part,
c) will be updated together with ESIMD changes implementing support in a more proper way.