- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
ldfp instruction could load two contiguous single or double precision valuse from memory in one move,
DoesICC9.0 provideany option to turn this optimization off?
In the other hand, I could not find any option to turn off SWP
Software Pipelining in ICC9.0 .
Thanks
Message Edited by George_Her on 09-22-2005 01:50 AM
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
icc equivalent of ifort directive
#pragma noswp !DIR$ NOSWP
Look them up in the docs. If you want further help with C, use the C++ branch of the forum.
When the compiler decides to use load-pair, it usually makes a very large improvement in performance, much more than the schedule would indicate. You can prevent it by using -O1, or the pragma.
Searching the documents is cumbersome. icc/ifort 9.0 documentation expects you to copy the html files over to Windows. In the 8.1 documents, pragma noswp is covered in Compiler User's Guide > Vol II: Optimizing Applications > Optimization support features > Compiler Directives > Pipelining for Itanium
#pragma noswp !DIR$ NOSWP
Look them up in the docs. If you want further help with C, use the C++ branch of the forum.
When the compiler decides to use load-pair, it usually makes a very large improvement in performance, much more than the schedule would indicate. You can prevent it by using -O1, or the pragma.
Searching the documents is cumbersome. icc/ifort 9.0 documentation expects you to copy the html files over to Windows. In the 8.1 documents, pragma noswp is covered in Compiler User's Guide > Vol II: Optimizing Applications > Optimization support features > Compiler Directives > Pipelining for Itanium
Message Edited by tim18 on 09-22-2005 09:19 AM

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page