Hi, I want to know how this function cublasLtMatmulAlgoGetHeuristic gets the best performing operator like ampere_sgemm_32x128__nt based on the GPU architecture and parameters like shape(m,n,k). Is there a formula? Also, is this function not open source?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Questions about algo paramter in cublasGemmEx, is there any detailed description? | 1 | 430 | August 29, 2023 | |
| Autotuning for GEMM kernel and combination with other kernels | 1 | 429 | December 1, 2022 | |
| cublasSgemm - is there a way to choose algorithm | 6 | 1559 | August 15, 2022 | |
| Using gcgemm from CuBLAS | 1 | 707 | March 23, 2020 | |
| How to judge whether IMMA kernel is chosed by cublasLt? | 0 | 648 | July 18, 2022 | |
| Is it necessary to tune cublas to get the best performance? | 3 | 73 | July 17, 2024 | |
| Question about CLBlast | 0 | 479 | February 20, 2019 | |
| Support fp16 for more cublas/cusolver? | 1 | 375 | March 8, 2021 | |
| Using cudaFuncSetAttribute for functions | 0 | 407 | March 12, 2020 | |
| Where to find detailed information about the algorithm type when using cublasGemmEx | 1 | 477 | October 27, 2023 |