Add a better performing config in the matmul example

[fcharras]

It has been reported in #1122 that the performance in the matmul tutorial is way below torch.matmul performance.

After playing with the parameters I found that the current grid search does not seem adapted to the max series gpu.

Adding this set of parameters to the grid search (basically changing num_warps from 2 to 16 to the config that I found is selected as the best config) gives a big (3 times) speedup on the 512 * 512 matmul:

In [4]: %time matmul(a, b)[0,0].cpu()
CPU times: user 343 µs, sys: 674 µs, total: 1.02 ms
Wall time: 965 µs
Out[4]: tensor(12.5781, dtype=torch.float16)

In [5]: %time torch.matmul(a, b)[0,0].cpu()
CPU times: user 262 µs, sys: 513 µs, total: 775 µs
Wall time: 715 µs
Out[5]: tensor(12.5781, dtype=torch.float16)

for comparison, those are the performances I get from this tutorial on the current main branch:

In [41]: %time matmul(a, b)[0,0].cpu()
CPU times: user 2.58 ms, sys: 0 ns, total: 2.58 ms
Wall time: 2.51 ms
Out[41]: tensor(12.5781, dtype=torch.float16)

In [42]: %time torch.matmul(a, b)[0,0].cpu()
CPU times: user 0 ns, sys: 900 µs, total: 900 µs
Wall time: 826 µs
Out[42]: tensor(12.5781, dtype=torch.float16)

I didn't go further in depth in trying to change the grid search. This is just a single change that I noticed improves a lot the performance for max series. Maybe the grid search can be tweaked even further to achieve other speedups.

[whitneywhtsang]

Added get_xpu_autotune_config in #1147, we can add this config for xpu backend in that function.

[whitneywhtsang]

We can copy the autotune configs from https://github.com/intel/intel-xpu-backend-for-triton/blob/llvm-target/python/tutorials/09-experimental-block-pointer.py#L102.

[whitneywhtsang]

We can copy the autotune configs from https://github.com/intel/intel-xpu-backend-for-triton/blob/llvm-target/python/tutorials/09-experimental-block-pointer.py#L102.

Done in #1185.

[whitneywhtsang]

@fcharras Please reopen this PR or create a new PR if there are any additional configs you would like to be added. Thanks for your contribution.