- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I'm running a program in 2 different platform which have same instruction set(Atom and Xeon)As my prediction, regardless of branches, it should have same instruction count but in my result
the value of "INST_RETIRED.ANY - BR_INST_RETIRED.ANY" is not equal and the difference is over 10% which I thought it's not under the error tolerance.
Therefore, is there any event I didn't considered but I should?
THX!
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
Please consider below factors in cross platforms:
1. Source file is same, but execution binaries may not be same.
2. Execution paths are same? (same test case?)
3. If binaries are same for Intel Atomprocessorand Intel Xeon processor, use amplxe-cl to test
a) amplxe-cl -collect lightweight-hotspots -r r001lh -- app
b) amplxe-cl -report pmu-events -r r001lh/ -filter module=app
Will you get same event counts for INST_RETIRED.ANY for app?
Regards, Peter
Please consider below factors in cross platforms:
1. Source file is same, but execution binaries may not be same.
2. Execution paths are same? (same test case?)
3. If binaries are same for Intel Atomprocessorand Intel Xeon processor, use amplxe-cl to test
a) amplxe-cl -collect lightweight-hotspots -r r001lh -- app
b) amplxe-cl -report pmu-events -r r001lh/ -filter module=app
Will you get same event counts for INST_RETIRED.ANY for app?
Regards, Peter

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