[SYCL] Test indirect access memory tracking in the L0 plugin

[againull]

No description provided.

[rdeodhar]

To improve readability I suggest

1. writing the accessors as: "accessor acc { b_d3_data, cgh, read_write };"
2. removing "cl::sycl::" throughout since you already have "using namespace sycl;"

But, what is the test checking exactly? Is indirect access tracking on by default?

[againull]

To improve readability I suggest

1. writing the accessors as: "accessor acc { b_d3_data, cgh, read_write };"
2. removing "cl::sycl::" throughout since you already have "using namespace sycl;"

But, what is the test checking exactly? Is indirect access tracking on by default?

Addressed comments and added description of the test.

[rdeodhar]

I ran the test on Gen9 with/without the env var on a recent build of intel/llvm.
Once, with the env var I got this error:
$ SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 ./a.out
pass
double free or corruption (out)

There may still be a bug in the implementation, but I'm approving this PR for the test.

[againull]

I ran the test on Gen9 with/without the env var on a recent build of intel/llvm. Once, with the env var I got this error: $ SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 ./a.out pass double free or corruption (out)

There may still be a bug in the implementation, but I'm approving this PR for the test.

@rdeodhar I can't reproduce this error with several thousand runs. Did you change the test or compiler somehow?

[rdeodhar]

I didn't change the test.
I ran it with an intel/llvm compiler as in PR 4891, which has just been merged. So effectively, with the latest intel/llvm compiler.

[againull]

I didn't change the test. I ran it with an intel/llvm compiler as in PR 4891, which has just been merged. So effectively, with the latest intel/llvm compiler.

Still can't reproduce neither with build from PR 4891 nor the latest intel/llvm or nightly build.
Could you please create a tracker with detailed instructions.

[vladimirlaz]

is this PR still needed? Should it be merged?

[againull]

is this PR still needed? Should it be merged?

It was waiting for gpu runtime version to be uplifted because old driver had an issue. GPU driver version was uplifted 4 days ago, so this can be merged now.