Post Description

This is a small fix version to improve performance of GPU acceleration. It will become faster on AMD Radeon graphics boards. It may be slightly faster on Nvidia GeForce graphics boards. There is no difference in CPU calculation. Because this isn't tested so much, there may be a bug, failure, or mistake. If you see a problem, please report the incident. I will try to solve as possible as I can.

I changed 3 points in my OpenCL implementation. It's possible to test them by lc option at command-line. Thanks cavalia88, Slava46, and Anime Tosho for many tests and wonderful idea. OpenCL perfomance is varied in every graphics boards. If you have a fast graphics board, enabling "GPU acceleration" would be faster. If it's not so fast (or is slow) on your PC, just un-check the feature.

Data transfur between PC's RAM and GPU's VRAM
Calculation over GPU
Calculate 2 blocks at once to reduce number of table lookup

[ Changes from 1.3.3.1 to 1.3.3.2 ]

PAR2 client update

Improvement
GPU acceleration will work well on AMD graphics boards.

[ Changes from 1.3.3.0 to 1.3.3.1 ]

Installer update

It shows dialog to select "per user" or "per machine" installation.
PAR2 client update

Change

Max number of threads to read files on SSD was increased to 6.
Improvement

GPU acceleration would become faster.

[ Changes from 1.3.2.9 to 1.3.3.0 ]

GUI update

Change
Option adapted to new "lc" settings.
It's possible to add 5th item in "Media size" on Create window.
PAR2 client update

Change

Max number of using threads is increased to 32.
Threshold to use GPU was decreased.
Improvement

Matrix inversion may use more threads.
L3 cache optimization was improved for recent CPUs.