My working on projects are developed on Visual Studio 2022 laster version (17.10.4) and Intel DPC++ 2024.2. Recently I found that the debug build can run successfully on both CPU and GPU, however, the release build with -O2 optimization can ONLY run on GPU and is problematic on CPU.
When runing on CPU, it seems that the program got stuck on the first queue parallel_for and the memory usage increases drastically without bound. I'm new to DPC++ programming. I'm just wondering if there is any hint that could help me fix this issue?
I tried to comment out the part of codes developed recently, and found that there is a line of code, by commenting it out the release build can run on CPU (but to have a correct result that line of code is needed). In fact, as long as that code appears, not even really executed ( I even tried to comment out the function calling that code), the release build fails on CPU. I guess it is related to reorders during -O2 optimization. Again I'm new to DPC++ programming and not very familiar with -O2 optimization processes.
BTW, the release build without -O2 is good. However, I really need to use -O2 optimization to improve the performance of the code.
My project contains 10k+lines of code with multiple parallel_for s, so it's not realistic to post it here.
Thanks so much!
PC
