volk
volk copied to clipboard
gnuradio
The Vector Optimized Library of Kernels
Doyxgen includes build paths quite excessively. First, these lines in `docs/Doxyfile.in` include absolute paths after the CMake `configure_file` command in `Doxyfile`. ``` PROJECT_LOGO = @CMAKE_SOURCE_DIR@/docs/volk_logo_small.png LAYOUT_FILE = @CMAKE_CURRENT_SOURCE_DIR@/DoxygenLayout.xml INPUT =...
Previously, this part of our CI wasn't run because it always failed. However, we've seen recent changes that may allow it to run again.
Added AVX512 kernels and some minor cleanup. Using AVX512F yields 40% speedup over the AVX2_FMA implementation on my 7950X3D. Compared to the generic atan2 implementation this is a 65x speedup....
In cases where we don't cross-compile, we might want to detect if a CPU is 32bit or 64bit. CMake provides functionality for this case starting in CMake 3.10. Let's use...
I think this is a better implementation of the reciprocal kernel as it uses the new _mm512_rcp14_ps intrinsic that handles exceptions correctly. It's accurate to tol < 6.2e-5. On a...
Cross compiling for iOS on macOS x86/64 just isn't working. I don't have an M1 Mac. Maybe I should try cross compiling on linux to produce the assembly files that...
In #77, the range of integers used when testing 16i inputs was reduced to -7 .. +7, to accommodate the `volk_16ic_x2_dot_prod_16ic` kernel which does not handle integer overflows consistently between...
Coverage
This adds support for Coverage reporting. Requirements: - gcc or clang - gcov - gcovr Usage: ```shell mkdir build-coverage && cd build-coverage cmake -DCMAKE_BUILD_TYPE=Coverage .. make -j make coverage firefox...
In a lot of issues and PRs we discuss problems with our current tests. We need to discuss a way forward to improve this situation. One option would be to...
UBSAN reports undefined behaviour in this code: https://github.com/gnuradio/volk/blob/42f57cd67506e7fb6a7795af3948b803de0085f4/kernels/volk/volk_8u_x4_conv_k7_r2_8u.h#L112-L113 Here `w` is an `unsigned int[2]`, and the `s * (sizeof(decision_t) / sizeof(unsigned int))` calculation intentionally reaches outside the bounds of the...