remove multiple archtictures to isa head and adding gemm tuning scripts

[scxiao]

1. remove the code to include multiple versions of arch features to isa head
2. Added a script to tune gemm performance (compatible with a script outside of this repo, may refine the script name later)

[keryell]

I do not understand the PR title (besides the typo to be fixed).
Or were the 2 other files not expected to be part of this PR?

[scxiao]

I do not understand the PR title (besides the typo to be fixed). Or were the 2 other files not expected to be part of this PR?

This PR is for the MI300X bringup and for testing the MSFT workloads. To do the testing, there is a joint (AMD/MSFT) repository that requires these two scripts. I agree the name and location of the scripts are not good, but that is required by the other repository. I will fix that (rename the script and move them to a better location) in a later PR in this repo and the other repo, so they can still work together,

[zhanglx13]

Do we need matmul_grouped.py?

[scxiao]

Do we need matmul_grouped.py?

I am not actually sure about that. Hi @alefimov-amd, what is your opinion on that?

[zhanglx13]

I asked because I saw you added it here: https://github.com/ROCmSoftwarePlatform/msft_amd_ai_operators/pull/172/files

[scxiao]

I asked because I saw you added it here: https://github.com/ROCmSoftwarePlatform/msft_amd_ai_operators/pull/172/files

I see, in the script scripts/amd/gemm/tune_gemm.sh, there is the following line

DRIVER=$(echo $DRIVER | sed -e "s/matmul_grouped.py/matmul.py/g")

to replace matmul_grouped.py with matmul.py for the driver.

So everything will actually call matmul.py.

[zhanglx13]

LGTM!