This page contains the release notes for AMD ROCm Software.
| Library | Version |
|---|---|
| AMDMIGraphX | 2.8 |
| hipBLAS | 2.0.0 |
| hipCUB | 3.0.0 |
| hipFFT | 1.0.13 |
| hipRAND | 2.10.17 |
| hipSOLVER | 2.0.0 |
| hipSPARSE | 3.0.0 |
| hipSPARSELt | ⇒ 0.1.0 |
| hipTensor | 1.1.0 |
| MIOpen | 2.19.0 |
| rccl | 2.15.5 |
| rocALUTION | 3.0.3 |
| rocBLAS | 4.0.0 |
| rocFFT | 1.0.25 |
| rocm-cmake | 0.11.0 |
| rocPRIM | 3.0.0 |
| rocRAND | 2.10.17 |
| rocSOLVER | 3.24.0 |
| rocSPARSE | 3.0.2 |
| rocThrust | 3.0.0 |
| rocWMMA | 1.3.0 |
| Tensile | 4.39.0 |
hipSPARSELt 0.1.0 for ROCm 6.0.1
- Enable hipSPARSELt APIs
- Support platform: gfx940, gfx941, gfx942
- Support problem type: fp16, bf16, int8
- Support activation: relu, gelu, abs, sigmod, tanh
- Support gelu scaling
- Support bias vector
- Support batched computation (single sparse x multiple dense, multiple sparse x single dense)
- Support cuSPARSELt v0.4 backend
- Integreate with tensilelite kernel generator
- Add Gtest: hipsparselt-test
- Add benchmarking tool: hipsparselt-bench
- Add sample app: example_spmm_strided_batched, example_prune, example_compress