[MHA V3] Remove uses of AITER_ASM_DIR and embed code objects

[draganmladjenovic]

Motivation

Remove the need for users to set AITER_ASM_DIR

Technical Details

Adapted generate.py scripts to embed code object into source files

Test Plan

Build check and manual testing of aiter/op_tests/cpp/mha on gfx942 + TE tests/jax/test_fused_attn.py + op_tests/test_mha.py

Test Result

Passing op_tests/test_mha.py on gfx942 and gfx950.

Submission Checklist

• [] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[valarLip]

looks not works well ..
/workspace/aiter/jit/build/module_fmha_v3_fwd/blob/asm_fmha_fwd_v3_gfx950.cpp:46:186: error: use of undeclared identifier 'fwd_hd192_hd128_bf16_causal'
46 | template<> struct FmhaFwdV3Buf<fmha_fwd_kernel_selector<FmhaFwdBf16, 192, 1, false, false, 1, GPUArch::gfx950>> { static constexpr const void * fwd_v3_buf = fwd_hd192_hd128_bf16_causal; };

[draganmladjenovic]

looks not works well .. /workspace/aiter/jit/build/module_fmha_v3_fwd/blob/asm_fmha_fwd_v3_gfx950.cpp:46:186: error: use of undeclared identifier 'fwd_hd192_hd128_bf16_causal' 46 | template<> struct FmhaFwdV3Buf<fmha_fwd_kernel_selector<FmhaFwdBf16, 192, 1, false, false, 1, GPUArch::gfx950>> { static constexpr const void * fwd_v3_buf = fwd_hd192_hd128_bf16_causal; };

Where do you observe this? I see fwd_hd192_hd128_bf16_causal.co exists.

[draganmladjenovic]

@valarLip testing passed. Can we merge this. I have rebase this daily due to new change to main.

[draganmladjenovic]

@ valarLip can we merge this before I get another merge conflict?