[SYCL][Joint Matrix] Update apply to make both matrices read/write

[dkhaldi]

Spec change was added in #13153
It states that the overload of joint_matrix_apply that takes two matrices can modify both matrices.
I also updated the test to reflect the change.

[YuriPlyakhin]

I suggest moving joint_matrix_activation test to IGC repo, since it is regression test intended for bug fix in IGC. It doesn't increase test coverage for SYCL compiler implementation of SYCL Joint Matrix, but rather generates yet another variant of IR, which IGC should be able to handle.

[dkhaldi]

I suggest moving joint_matrix_activation test to IGC repo, since it is regression test intended for bug fix in IGC. It doesn't increase test coverage for SYCL compiler implementation of SYCL Joint Matrix, but rather generates yet another variant of IR, which IGC should be able to handle.

Since when we add SYCL tests to IGC. IGC takes binaries from this SYCL suite.

[YuriPlyakhin]

I suggest moving joint_matrix_activation test to IGC repo, since it is regression test intended for bug fix in IGC. It doesn't increase test coverage for SYCL compiler implementation of SYCL Joint Matrix, but rather generates yet another variant of IR, which IGC should be able to handle.

Since when we add SYCL tests to IGC. IGC takes binaries from this SYCL suite.

The test in IGC would not be using "SYCL" language. It would be LLVM IR or SPIRV running through ocloc. This process is simpler/faster than add tests to SYCL repo, then build binaries, then add to IGC testing. Also, it would run as pre-checkin for each IGC PR, which makes it faster to catch regressions. I suggest we discuss the details in the call. Also, maybe the test is not necessary at all, since there is a lit test together with IGC bug fix, which maybe enough already. Again, let's discuss it on the call.

[YuriPlyakhin]

LGTM
nit: PR description needs update to remove the line about adding new test.

[dkhaldi]

@intel/llvm-gatekeepers, please help merge this PR.