Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add optimizations for AMD GPUs to VkFFT
* Do not load from shared memory inside if branches hipcc generates slower jumps and duplicated loads instead of simple conditional moves (v_cndmask_b32). * Use vector operations for better gfx90a's packed f32 math * Add input and output wrappers for faster load and store address computation * Replace blockDim with actual values
- Loading branch information