Forum Discussion

GNL's avatar
GNL
Icon for New Contributor rankNew Contributor
5 years ago

DevCloud: OneAPI vector-add example parallel execution is much slower than the scalar one

Hello, I tried the BaseKit-code-samples/DPC++Compiler/vector-add example as suggested in the DeveloperZone get started page: https://devcloud.intel.com/oneapi/get-started/base-toolkit/ I measured ...
GNL's avatar
GNL
Icon for New Contributor rankNew Contributor
5 years ago

Hi,

Everything is fine for now, I hope it's the same there..

I have some experience on DSP programming and I am familiar with some of the low level optimization methods.. At the last example I gave in the prev. post you can see that the CPU exec time is about 1600+ times faster. In this case, I think our problem is not that low level. I think the main problem here is the JIT Compiler overhead.

This picture is from the book: https://www.apress.com/us/data-parallel-c-advanced-chapters-just-released/17382670

On the cloud, I will try Ahead-Of-Time compiling, as suggested in here: https://software.intel.com/en-us/oneapi-dpcpp-compiler-dev-guide-and-reference-ahead-of-time-compilation

I will notify you with the results. And maybe we need to move this subject into another topic.

GNL