- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi - I am running the Dhrystone 2.1 benchmark on an Atom Z530 running Linux. I am using the ICC V10.1 compiler. What ICC options should I use to maximize results? I am currently using O3, xL, ipo.
Using ipo is debatable as inlining is generally considered a no-no wrt Dhrystone.
I have tried PGO but results were about 10% worse.
Any suggestions would be greatly appreciated.
Using ipo is debatable as inlining is generally considered a no-no wrt Dhrystone.
I have tried PGO but results were about 10% worse.
Any suggestions would be greatly appreciated.
Link Copied
3 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Why not set -ansi-alias? Is there a bug in the code?
If you want to disable in-lining, you set -fno-inline-functions as well as removing -ipo options.
icc 11.1 has fixed some performance regressions of 10.x and will release soon.
Although -xL would most often perform better than -xW or -xT, that wouldn't be true of every benchmark.
If you want to disable in-lining, you set -fno-inline-functions as well as removing -ipo options.
icc 11.1 has fixed some performance regressions of 10.x and will release soon.
Although -xL would most often perform better than -xW or -xT, that wouldn't be true of every benchmark.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Quoting - tim18
Why not set -ansi-alias? Is there a bug in the code?
If you want to disable in-lining, you set -fno-inline-functions as well as removing -ipo options.
icc 11.1 has fixed some performance regressions of 10.x and will release soon.
Although -xL would most often perform better than -xW or -xT, that wouldn't be true of every benchmark.
If you want to disable in-lining, you set -fno-inline-functions as well as removing -ipo options.
icc 11.1 has fixed some performance regressions of 10.x and will release soon.
Although -xL would most often perform better than -xW or -xT, that wouldn't be true of every benchmark.
Thanks. -ansi-alias did not make a difference. -xL results were better than -xW and only slightly better than -xT.
Do you expect the 11.1 improvements to be significant over V10.1 ? Thanks.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I'm also suggesting you to use the 11.0.074 or 11.1 coming out soon. Each release there're many improvements.
With 11.x, use /QxSSE3_ATOM.
Jennifer
With 11.x, use /QxSSE3_ATOM.
Jennifer
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