SYCL kernels 2

[ayarkhan]

Use SYCL kernels in SLATE (generated by dpct tool) for a SYCL-OneMKL hardware.

[ayarkhan]

On sunspot, testing using 1 process, 8-threads, 1 GPU.
( cd ~/icl/slate/test/; env ZE_AFFINITY_MASK=0.0 OMP_NUM_THREADS=8 ./run_tests.py --quick --target d --origin s --timeout 300 --type s,d,c,z --check y --ref n )
Gives inconsistent results on different runs (different qsub allocations)

% 2023-11-16 21:31:31, 1 MPI ranks, CPU-only MPI, 8 OpenMP threads, 1 GPU devices per MPI rank
16 routines FAILED: gbmm, gemm, gemmA, gemmC, hemmC, herk, her2k, her2k, syr2k, syr2k, syrk, trsmA, heev, stedc, hegv, scale_row_col
% 2023-11-16 23:08:37, 1 MPI ranks, CPU-only MPI, 8 OpenMP threads, 1 GPU devices per MPI rank
12 routines FAILED: gbmm, gemmC, her2k, syrk, trmm, trsmA, gbtrs, heev, stedc, hegv, svd, scale_row_col

One unit test fails (test_norm, synorm_dev, fro (upper and lower)

1 unit tests FAILED: test_norm
synorm_dev( one, lower )                                    pass,  0.0752 sec
synorm_dev( fro, lower )                                    FAILED:
    error < 5*eps at unit_test/test_norm.cc:479
synorm_dev( one, upper )                                    pass,  0.0685 sec
synorm_dev( fro, upper )                                    FAILED:
    error < 5*eps at unit_test/test_norm.cc:479
synorm_offdiag( one )                                       pass,  0.0000 sec

[mgates3]

In Makefiles, use spaces for if-else-endif blocks. Use tabs only for rules.

[mgates3]

Spaces.