Boost logo

Boost :

From: prathamesh tagore (prathameshtagore_at_[hidden])
Date: 2021-07-05 17:41:35


Week 4(28/06/21 - 05/07/21) :
1. Investigated OpenCv's size specific modifications for kernels having
their size lesser than or equal to 5. The modification seems to be
primarily related to simd register specification (will discuss this in my
next meeting with mentors).
2. Tried using "std::transform_reduce()" instead of "std::inner_product()"
for the purpose of comparing performance. This(
https://github.com/meshtag/gil/blob/Improve_buffer_update_transform_reduce/include/boost/gil/extension/numeric/convolve_novel.hpp)
is the related source code. Note : The code works fine even without
"std::execution::par", I added it only for experimentation(tbb should be
linked for using it).
3. Benchmarked this(
https://github.com/meshtag/gil/blob/Improve_buffer_update_naive_buffer_update/include/boost/gil/extension/numeric/convolve_novel.hpp)
version which contains a seemingly naive method of buffer
update(surprisingly, the observed performance difference did not came out
to be as expected)

@lpranam @Olzhas , did you had a chance to go through this (
https://github.com/meshtag/gil/blob/Improve_buffer_update/include/boost/gil/extension/numeric/convolve_novel.hpp)
version and related benchmark file(
https://github.com/meshtag/gil/blob/Improve_buffer_update/benchmark/celero/convolve_2d.cpp).
Please comment on factors such as performance, approach, design, etc.

On Mon, Jun 14, 2021 at 10:39 PM prathamesh tagore <
prathameshtagore_at_[hidden]> wrote:

> This thread will contain weekly updates for my work during coding period
> of GSoC 2021.
>


Boost list run by Boost-Gil-Owners