Intel® C++ Compiler
Community support and assistance for creating C++ code that runs on platforms based on Intel® processors.

Question about Intel compiler(icx)

Daniil5441
Beginner
1,290 Views

I use icx compiler recently. I wonder if icx uses polly to optimize code?

0 Kudos
1 Solution
SeshaP_Intel
Moderator
841 Views

Hi,


Thank you for your patience.


Currently, there is no intention to support Polly in icx(Intel® oneAPI DPC++/C++ Compiler).

We are committed to improving icx's performance. Based on the results from the development team icx is showing much better results than clang+polly. 

Could you please confirm whether we can close this thread from our end?


Thanks and Regards,

Pendyala Sesha Srinivas


View solution in original post

0 Kudos
7 Replies
SeshaP_Intel
Moderator
1,212 Views

Hi,


Thank you for posting in Intel Communities.

We are working on this issue internally. We will get back to you soon.


Thanks and Regards,

Pendyala Sesha Srinivas


0 Kudos
SeshaP_Intel
Moderator
1,092 Views

Hi,


Could you please try the Loop Optimizations and Vectorizer in icx(Intel® oneAPI DPC++/C++ Compiler) which have good product quality in terms of performance, stability, and compile time.

Please refer to the below link for more details.

https://www.intel.com/content/www/us/en/docs/dpcpp-cpp-compiler/developer-guide-reference/2023-0/overview.html


Please let us know if you have tried and found that there are any applications in which Polly has shown better performance with the Intel LLVM compiler. 


Thanks and Regards,

Pendyala Sesha Srinivas



0 Kudos
Daniil5441
Beginner
1,068 Views

HI,

Thank you for answering me.

I have one problem. I don`t know how to use polly with Intel LLVM Compiler, but I compared Clang+polly with Intel LLVM Compiler and had got that Clang with polly creates slower code. But I am also interested in trying polly with Intel Compiler, so I will know exactly that polly works worse than native Intel LLVM Compiler.

0 Kudos
SeshaP_Intel
Moderator
994 Views

Hi,


Could you please share the source code, the steps you followed, and the results from Clang+Polly and Intel LLVM Compiler? 

Could you please let us know how you are stating that Polly creates slower code with Clang?


Thanks and Regards,

Pendyala Sesha Srinivas


0 Kudos
Daniil5441
Beginner
957 Views

Hi,

Firstly, I compiled "benchmark_decode.cpp" with "qoir.h" with Intel LLVM Compiler.

Secondly, in my code I have small benchmark that counts execution time of one decoding function called "qoir_decode" and prints "Decode time: some time" and I used Vtune to see what function takes most of the time.

Thidly, I repeat first and second steps but with Clang+Polly.

For Intel LLVM Compiler I used flags:"-fuse-ld=lld -fast -flto=thin -xHost".

For Clang+Polly I used flags:"-Ofast -march=native -flto=thin -mllvm -polly".

I had the next result:

For Intel LLVM Compiler: 72 milliseconds stable

For Clang+Polly: 79 milliseconds stable

In Vtune I got that the critical function "qoir_private_swizzle_bgrp__rgbn"  with Intel LLVM Compiler takes more time to complete than the same part code compiled with Clang+Polly that takes twice less time to complete, but other parts of code takes more time. So I had that in general Intel LLVM Compiler generates faster code.

For correct using you need type in console: "benchmark_decode path/to/file".

Qoir image that I used for benchmarking, all sources and vtune profiling files are in attachments.

I have a question: Is there a chance to use polly with Intel LLVM Compiler?

0 Kudos
SeshaP_Intel
Moderator
842 Views

Hi,


Thank you for your patience.


Currently, there is no intention to support Polly in icx(Intel® oneAPI DPC++/C++ Compiler).

We are committed to improving icx's performance. Based on the results from the development team icx is showing much better results than clang+polly. 

Could you please confirm whether we can close this thread from our end?


Thanks and Regards,

Pendyala Sesha Srinivas


0 Kudos
SeshaP_Intel
Moderator
758 Views

Hi,


Thanks for accepting our solution. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.


Thanks and Regards,

Pendyala Sesha Srinivas


0 Kudos
Reply