- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
It is fairly easy to find the percentage of vectorised floating point operations in an uninstrumented code (no debug symbols) including vector length (AVX or not). How to find if a code is vectorising integer operations? Preferably using HW counters.
More generally, what to look for in a code which is computationally intensive but has relatively little FP ops? Are there suitable templates for initial analysis?
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
We don't have HW events that can calculate integer vectorized operations as we do for floating point operations.
We recommend to use Intel Advisor that shows very comprehensive information on both integer and floating point vectorization efficiency.
On this:
"More generally, what to look for in a code which is computationally intensive but has relatively little FP ops? Are there suitable templates for initial analysis?"
You can start with Microarchitecture Exploration Analysis type that will give you a pretty complete picture where you lose you performance in terms of different kind of stalls in instruction execution pipeline.
Thanks & Regards, Dmitry

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