[release/2.10] [ROCm] derive the hipcub compatible cccl version from 'HIPCUB_CCCL_VERSION'

[jithunnair-amd]

Cherry-pick of pytorch#188072 (commit c8e2473) onto release/2.10.

Summary

• Introduce dependency on libhipcxx
• Guard FpLimits for c10::BFloat16 backport to pre CCCL 3.x
• Extend CUB_VERSION derivation for hipCUB using HIPCUB_CCCL_VERSION

Notes

This PR has merge conflicts that need to be resolved manually (conflict markers committed as requested).

Test plan

• Build with ROCm
• Verify hipCUB/CCCL compatibility glue compiles

Authored with assistance from Cursor

Made with Cursor

[rocm-repo-management-api]

Jenkins build for a375156b72041e35467399ffaed4f3409e28b33c commit finished as FAILURE
Links: Pipeline Overview / Build artifacts / Test Results

Detected error during Pytorch building:

      |     ^
/var/lib/jenkins/pytorch/third_party/fbgemm/external//composable_kernel/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_xdl_cshuffle_tile_loop.hpp:65:5: warning: failed to meet occupancy target given by 'amdgpu-waves-per-eu' in '_ZN2ck16tensor_operation6device34kernel_grouped_gemm_multiple_d_xdlINS_34GridwiseGemmMultiD_xdl_cshuffle_v3INS_13tensor_layout4gemm8RowMajorENS5_11ColumnMajorENS_5TupleIJS6_S7_EEES6_DB8_SA_ffNS8_IJffEEEtNS0_12element_wise11PassThroughESD_NSC_16MultiplyMultiplyELNS1_18GemmSpecializationE0ELi256ELi256ELi224ELi128ELi16ELi16ELi16ELi16ELi8ELi7ENS_8SequenceIJLi8ELi32ELi1EEEENSG_IJLi1ELi0ELi2EEEESI_Li2ELi16ELi16ELb0ELi0ESH_SI_SI_Li2ELi16ELi16ELb0ELi0ELi2ELi1ENSG_IJLi1ELi64ELi1ELi4EEEENSG_IJLi8ELi8ELi1EEEELNS_26BlockGemmPipelineSchedulerE0ELNS_24BlockGemmPipelineVersionE2ESA_SA_SA_SA_Lb0EEENS1_25GroupedGemmKernelArgumentILi2EEELSF_0ESA_SA_SB_tS6_S7_S9_S6_Li128ENS_25OffsettedBlockToCTileMap2INS_39BlockToCTileMap_Grouped_M00_N0_M01AdaptILi8ELi256ELi224EEEEESS_SD_SD_SE_LSL_0ELSM_2EEEvPU3AS4KviT13_T14_T15_': desired occupancy was 2, final occupancy is 1 [-Wpass-failed]
2 warnings generated when compiling for gfx942.
[7246/8132] Linking CXX static library lib/libfbgemm_genai.a
[7247/8132] Building HIPCC object caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/torch_hip_generated_cub-RadixSortPairs-uint32-8.hip.o
FAILED: caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/torch_hip_generated_cub-RadixSortPairs-uint32-8.hip.o /var/lib/jenkins/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/torch_hip_generated_cub-RadixSortPairs-uint32-8.hip.o 
cd /var/lib/jenkins/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip && /opt/conda/envs/py_3.12/lib/python3.12/site-packages/cmake/data/bin/cmake -E make_directory /var/lib/jenkins/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/. && /opt/conda/envs/py_3.12/lib/python3.12/site-packages/cmake/data/bin/cmake -D verbose:BOOL=OFF -D build_configuration:STRING=RELEASE -D generated_file:STRING=/var/lib/jenkins/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/./torch_hip_generated_cub-RadixSortPairs-uint32-8.hip.o -P /var/lib/jenkins/pytorch/build/caffe2/CMakeFiles/torch_hip.dir/__/aten/src/ATen/hip/torch_hip_generated_cub-RadixSortPairs-uint32-8.hip.o.cmake
In file included from /var/lib/jenkins/pytorch/aten/src/ATen/hip/cub-RadixSortPairs-uint32-8.hip:2:
In file included from /var/lib/jenkins/pytorch/aten/src/ATen/hip/cub-RadixSortPairs.cuh:6:
In file included from /var/lib/jenkins/pytorch/aten/src/ATen/hip/cub.cuh:15:
/var/lib/jenkins/pytorch/aten/src/ATen/hip/cub_definitions.cuh:31:1: error: version control conflict marker in file