rpp
rpp copied to clipboard
ROCm
AMD ROCm Performance Primitives (RPP) library is a comprehensive high-performance computer vision library for AMD processors with HIP/OpenCL/CPU back-ends.
- Add QA support for F32 bit depth for initial set of functions. - Add range checks rpp_pixel_check_0to1_avx and rpp_pixel_check_0to1_sse to ensure F32 bit depth QA test pass. - Add...
- Add tensor support for Concat Augmentation optimized using AVX2 on HOST backend. - Add tensor support for Concat Augmentation on HIP backend. - Add relevant unit and performance tests...
- Adds JPEG Compression Distortion augmentation for HOST on U8/F32/F16/I8. - Adds relevant unit/perf/QA tests.
- Adds nearest neighbors padding for box filter augmentation (Current ToT version had border pixels fading out, the PR attempts to fix the same ) - Separated the compute functions...
## Motivation Need to make clear that rpp has only been tested on a subset of Linux environments.
Fixes "Compile Warnings - Azure CI" #590 - Contains warning fixes for the HIP Backend build - Adds CHECK_RETURN_STATUS for the various hip API calls
I was helping @utk4r-sh in preparing to package rpp for Debian, but I noticed that there is a prebuilt copy of [libffts.a](https://github.com/ROCm/rpp/blob/rocm-6.4.3/libs/third_party/ffts/libffts.a) in the sources. While the FFTS license is...
- Add Solarize for both HOST and HIP Backends. - Add support for NCHW/NHWC, toggle variants and U8/F16/F32/I8 bit depths. - Add relevant QA/unit/perf tests.
### Problem Description Test package use cmake to build and run tests. On a clean install of ROCm the cmake package is missing. The test package can install cmake ###...
