[ROCm] Build FBGEMM_GENAI for gfx942 only (#162648)

Fixes build timeouts >4h on libtorch build jobs: 75e7f49f9c/1

Brings back code to narrow down CK compilation targets from 69a25f6888 (diff-ce80f3115ab2f6be5142f0678a1fc92c6b2d7727766ce44f48726c99e720f777)

gfx942 supports fp8

Don't enable gfx950 for now, until more optimizations are in place as per https://github.com/pytorch/pytorch/pull/162648/files#r2369588738

Validation:
[rocm6.4](https://github.com/pytorch/pytorch/actions/runs/17944766350/job/51028483128) and [rocm6.3](https://github.com/pytorch/pytorch/actions/runs/17944766350/job/51028483093) libtorch builds finished within 3.9h.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/162648
Approved by: https://github.com/jeffdaily

Co-authored-by: Jeff Daily <jeff.daily@amd.com>

aten/src/ATen/CMakeLists.txt

@@ -317,10 +317,20 @@ IF(USE_FBGEMM_GENAI)
         -greedy-reverse-local-assignment=1
         -fhip-new-launch-api)
 
+      # Only compile for gfx942 for now.
+      # This is rather hacky, I could not figure out a clean solution :(
+      set(HIP_CLANG_FLAGS_ORIGINAL ${HIP_CLANG_FLAGS})
+      string(REGEX REPLACE "--offload-arch=[^ ]*" "" FILTERED_HIP_CLANG_FLAGS "${HIP_CLANG_FLAGS}")
+      if("gfx942" IN_LIST PYTORCH_ROCM_ARCH)
+        list(APPEND FILTERED_HIP_CLANG_FLAGS --offload-arch=gfx942;)
+      endif()
+      set(HIP_CLANG_FLAGS ${FILTERED_HIP_CLANG_FLAGS})
+
       hip_add_library(
         fbgemm_genai STATIC
         ${fbgemm_genai_native_rocm_hip}
         HIPCC_OPTIONS ${HIP_HCC_FLAGS} ${FBGEMM_GENAI_EXTRA_HIPCC_FLAGS})
+      set(HIP_CLANG_FLAGS ${HIP_CLANG_FLAGS_ORIGINAL})
       set_target_properties(fbgemm_genai PROPERTIES POSITION_INDEPENDENT_CODE ON)
       target_compile_definitions(fbgemm_genai PRIVATE FBGEMM_GENAI_NO_EXTENDED_SHAPES)