There is no conversion of data permitted by load_matrix_sync.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| recast nvcuda::wmma::fragment from unsigned char to half | 0 | 654 | December 23, 2019 | |
| Bank Conflicts When Using wmma::load_matrix in CUDA without Swizzle? | 0 | 149 | September 12, 2024 | |
| load int8 shared memory data into fp16 wmma::fragment | 0 | 509 | August 7, 2019 | |
| Question about api load_matrix_sync | 0 | 59 | September 26, 2024 | |
| About compute accuracy | 22 | 123 | February 10, 2025 | |
| How to get one load operation to load values of different types? | 3 | 34 | February 24, 2025 | |
| Is loading the matrices in like this good practice for WMMA instructions in C++ CUDA? | 0 | 41 | December 30, 2024 | |
| How to Load 4 Consecutive Values from Shared Memory into uint MultiA for MMA? | 4 | 16 | November 25, 2024 | |
| coalescing memory in short to float conversion | 3 | 4497 | January 23, 2009 | |
| Fastest Tiled WMMA for Matrices of Any Size? | 3 | 246 | October 26, 2024 |