- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I've been trying to understand a performance bug involving the Linux kernel. I rebuilt the MIC's kernel planning to make changes in order to explore the problem. I've used the GCC cross-compiler and sources from the MPSS distribution.
Unfortunately, when re-running my problematic scenarios on the rebuilt kernel, the performance problems become inexplicably vastly worse. Since I'm just trying to understand the problem, the performance decrease alone is not my issue: my issue is that the rebuilt kernel's behavior has changed so much from the distributed kernel's. The huge changes lead me to believe that I might be building the kernel with different options or a different compiler.
How is the distributed kernel built? Is it built with the same GCC compiler distributed with MPSS? Is it built using the same options and source provided in http://registrationcenter.intel.com/irc_nas/3778/mpss-src-3.1.2.tar ?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The Intel Xeon Phi's Linux kernel is indeed built using the same GCC cross compiler and source code provided on the MPSS release site. (Contrary to bustaf's speculation, the object code produced when cross-compiling is identical to what would be produced using a native GCC running on Intel Xeon Phi itself.)
I've replied to your earlier thread with the details.
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Tim,
I'm forwarding this and your other issues to the MPSS team. Hopefully, they can address some of the issues.
Regards
--
Taylor
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Ok, thanks.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Basically, I believe your suggestions should be correct; it should build with the tools provided in the distribution. I haven't built the current version myself, but this was so in the past.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi
The kernel by default in package is probably compiled by the system hosted on the card Phi.
When you use the Crossed compile on you Xeon system the compiler is unable to take precisely characteristics of
processor on phi card.
No relation with options of kernel or with type compiler choice.
Unfortunately I doubt that even if you add the better flags possible he will not solve this problem.
The solution is an little complex,require you compile source GNU compiler on the system hosted by phi card
before you recompile natively a kernel with him.
Regards
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The Intel Xeon Phi's Linux kernel is indeed built using the same GCC cross compiler and source code provided on the MPSS release site. (Contrary to bustaf's speculation, the object code produced when cross-compiling is identical to what would be produced using a native GCC running on Intel Xeon Phi itself.)
I've replied to your earlier thread with the details.
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page