fmha v3 API only generate for supported targeting GPU arch

[HollowMan6]

Motivation

Refactor to resolve the building issue for gfx940, and also fix symbol not found if "gfx942" or "gfx950" is not the targeting compiling GPU arch:

  In file included from TransformerEngine/build/cmake/ck_fused_attn/gen_src/mha_bwd.cpp:1:
  In file included from TransformerEngine/transformer_engine/common/ck_fused_attn/../../../3rdparty/aiter/csrc/include/mha_bwd.h:7:
  TransformerEngine/transformer_engine/common/ck_fused_attn/../../../3rdparty/aiter/csrc/include/aiter_hip_common.h:147:5: warning: ignoring return value of function declared with 'nodiscard' attribute [-Wunused-result]
    147 |     hipGetDeviceProperties(&prop, 0);
        |     ^~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~
  /appl/lumi/SW/CrayEnv/EB/rocm/6.2.2/include/hip/hip_runtime_api.h:91:32: note: expanded from macro 'hipGetDeviceProperties'
     91 | #define hipGetDeviceProperties hipGetDevicePropertiesR0600
        |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~
  TransformerEngine/build/cmake/ck_fused_attn/gen_src/mha_bwd.cpp:69:9: error: use of undeclared identifier 'gfx90a'
     69 |     t = gfx90a::fmha_bwd_v3(traits, args, stream_config, seqlen_q_padded, seqlen_k_padded, is_v3_api_check);
        |         ^
  1 warning and 1 error generated when compiling for gfx90a.

Technical Details

Use a list of supported GPU arch to determine before generation.

Test Plan

Test Result

Submission Checklist

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

[Copilot]

Pull Request Overview

Restrict fmha v3 code generation to known-supported GPU architectures to avoid compile errors on unsupported arch targets (e.g., gfx90a/gfx940).

• Introduce a supported-arch allowlist for v3 API generation.
• Gate single-target v3 generation on the allowlist; otherwise fall back to multi-target path.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
csrc/cpp_itfs/mha_fwd_generate.py	Adds V3_SUPPORTED_ARCH and updates get_v3_api to only emit single-target v3 calls for supported arches.
csrc/cpp_itfs/mha_bwd_generate.py	Mirrors the fwd changes for the backward path.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Pull Request Overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[HollowMan6]

Looks like this has already been solved by #1318, so I'm closing this