WIP - Standardize argument memory type across all RPP Tensor HIP API

ROCm / rpp

AMD ROCm Performance Primitives (RPP) library is a comprehensive high-performance computer vision library for AMD processors with HIP/OpenCL/CPU back-ends.

https://rocm.docs.amd.com/projects/rpp/en/latest/

MIT License

53 stars 38 forks source link

WIP - Standardize argument memory type across all RPP Tensor HIP API #350

Closed r-abishek closed 2 months ago

r-abishek commented 4 months ago

This PR removes inconsistencies and standardizes all array/tensor parameters to be in pinned memory across all RPP Tensor HIP functionalities to avoid copies from host to device. (src/dst large buffers in HIP device memory, and smaller function-parameter buffers in pinned memory)

kiritigowda commented 2 months ago

@r-abishek - any update on this? This has been open for a long time.

r-abishek commented 2 months ago

Closing until parallel changes ready from mivisionx/rocal