diff --git a/docs/site_specific_config/rocm.md b/docs/site_specific_config/rocm.md index cd1ef0414e..b471243bbd 100644 --- a/docs/site_specific_config/rocm.md +++ b/docs/site_specific_config/rocm.md @@ -1,4 +1,4 @@ -# Overview of ROCm Ecosystem (v6.4.1-20250616) +# Overview of ROCm Ecosystem (v6.4.1-20250620) !!! warning "Work-in-progress" This document is a work-in-progress. @@ -6,7 +6,7 @@ This overview is being created in the context of adding support for ROCm to EESSI, the European Environment for Scientific Software Installations (). - *Last update: 16 Jun 2025* + *Last update: 20 Jun 2025* [Jump to Overview](#Introduction) | [Jump to ABC](#ABC-of-ROCm) | [Jump to Changelog](#Changelog) @@ -331,6 +331,7 @@ The "roc" variants like rocFFT are AMD's native implementations optimized specif ### ML/DL Frameworks +* MIGraphX: Graph optimization engine ([Github](https://github.com/ROCm/AMDMIGraphX)) * MIOpen: Deep learning primitives library ([Github](https://github.com/ROCm/MIOpen)) * ROCm PyTorch: PyTorch support for AMD GPUs ([Github](https://github.com/ROCm/pytorch)) * ROCm TensorFlow: TensorFlow support for AMD GPUs ([Github](https://github.com/ROCm/tensorflow-upstream)) @@ -338,10 +339,12 @@ The "roc" variants like rocFFT are AMD's native implementations optimized specif ### Communication Libraries * RCCL: Communication library for multi-GPU/multi-node training ([Github](https://github.com/ROCm/rccl)) +* rocPRIM: Provides parallel primitives used to develop GPU-accelerated code ([Github](https://github.com/ROCm/rocPRIM)) ### Marshalling Libraries * hipBLAS ([Github](https://github.com/ROCm/hipBLAS)) +* hipCUB ([Github](https://github.com/ROCm/hipCUB)) * hipFFT ([Github](https://github.com/ROCm/hipFFT)) * hipRAND ([Github](https://github.com/ROCm/hipRAND)) * hipSOLVER ([Github](https://github.com/ROCm/hipSOLVER)) @@ -358,50 +361,107 @@ graph LR; roctracer[ROC Tracer] end subgraph Libraries and Frameworks + hipblaslt[hipBLASLt] + hipsparselt[hipSPARSELt] + rocblas[rocBLAS] rocfft[rocFFT] rocrand[rocRAND] - rocblas[rocBLAS] - rocsparse[rocSPARSE] rocsolver[rocSOLVER] - hipblaslt[hipBLASLt] - hipsparselt[hipSPARSELt] + rocsparse[rocSPARSE] + migraphx[MIGraphX] miopen[MIOpen] rccl[RCCL] + rocprim[rocPRIM] + hipblas[hipBLAS] + hipcub[hipCUB] hipfft[hipFFT] hiprand[hipRAND] - hipblas[hipBLAS] - hipsparse[hipSPARSE] hipsolver[hipSOLVER] + hipsparse[hipSPARSE] end - rocfft --> rocmstack - rocrand --> rocmstack - rocblas --> rocmstack - rocsparse --> rocmstack - rocsolver --> rocmstack - rocsolver --> rocblas hipblaslt --> rocmstack hipblaslt --> roctracer hipsparselt --> rocmstack hipsparselt --> roctracer hipsparselt --> hipsparse + rocblas --> rocmstack + rocfft --> rocmstack + rocrand --> rocmstack + rocsolver --> rocmstack + rocsolver --> rocblas + rocsparse --> rocmstack + migraphx --> rocmstack + migraphx --> miopen + migraphx --> rocblas miopen --> rocmstack - miopen --> rocblas miopen --> hipblaslt + miopen --> rocblas miopen --> hipblas rccl --> rocmstack + rocprim --> rocmstack - hipfft --> rocfft - hiprand --> rocrand hipblas --> rocblas - hipblas --> rocsparse hipblas --> rocsolver - hipsparse --> rocsparse + hipblas --> rocsparse + hipcub --> rocprim + hipfft --> rocfft + hiprand --> rocrand hipsolver --> rocsolver + hipsparse --> rocsparse ``` +## Case Studies + +For some commonly used software we examined the ROCm dependencies to make our overview more comprehensive. +Specifically, we examined which ROCm libraries and frameworks they depended on. +Below you can find a list of dependencies for each case study. + +### PyTorch ROCm Dependencies + +* HIP +* hipBLAS +* hipBLASLt +* hipFFT +* hipRAND +* hipSOLVER +* hipSPARSE +* hipSPARSELt +* MIOpen +* ROCm-LLVM +* ROCm-SMI +* rocBLAS +* rocFFT +* rocRAND +* rocSOLVER +* roc-tracer + +### TensorFlow ROCm Dependencies + +* HIP +* hipBLAS +* hipCUB +* hipFFT +* hipRAND +* hipSOLVER +* hipSPARSE +* MIGraphX +* MIOpen +* RCCL +* rocBLAS +* rocFFT +* rocPRIM +* rocRAND +* rocSOLVER + +### GROMACS ROCm Dependencies + +* HIP +* ROCm-LLVM +* rocFFT + ## Compatibility Policies {: #Compatibility-Policies } [Source](https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html) @@ -579,3 +639,8 @@ Vega refers to AMD's GPU architecture that was one of the first to fully support * fixed dev tools dependencies graph * sorted libraries * removed the big dependencies graph in favour of dedicated ones + +## v.6.4.1-20250620 + +* added the MIGraphX, rocPRIM, hipCUB libraries +* added case studies for PyTorch, TensorFlow, GROMACS