[SYCL][E2E][Joint Matrix] Add half colmajor A colmajor B test

[YuriPlyakhin]

• Add test for half colmajor A, colmajor B load for 8x16x16
• Refactor to get rid of unnecessary SG32-specific file

[dkhaldi]

I see you deleted the SG32 test. How are we testing SG32 for this test as 2d load additions for SG32 should take care of transpose/VNNI for this test?

[YuriPlyakhin]

I see you deleted the SG32 test. How are we testing SG32 for this test

Please, check the top of the .cpp file. I'm building and running it 2 times, we don't need to keep SG32 folder to test sub-group size 32:

// RUN: %{build} -o %t.out
// RUN: %{run} %t.out
// RUN: %{build} -o %t32.out -DSG_SZ=32
// RUN: %{run} %t32.out

as 2d load additions for SG32 should take care of transpose/VNNI for this test?

SG32 with transpose/VNNI can be implemented with or without 2d block load/store. SYCL E2E test doesn't control such implementation detail. Currently SG32 implementation with transpose doesn't use 2d block load, but test will pass.

not related to your question:
I will need to update this PR due to test folders restructuring.

[dkhaldi]

LGTM

[YuriPlyakhin]

@intel/llvm-gatekeepers , failure is not related. Please, merge.

Failed Tests (1):
SYCL :: Adapters/enqueue-arg-order-buffer.cpp

I submitted #16774