On Fri, Oct 30, 2026 at 12:00:41AM -0400, Matt Turner wrote:
On Thu, Oct 29, 2026 at 11:34 AM Matt Turner [off-list ref] wrote:
quoted
On Thu, Oct 29, 2026 at 5:11 AM Peter Zijlstra [off-list ref] wrote:
quoted
On Wed, Oct 28, 2026 at 02:09:29PM -0400, Matt Turner wrote:
quoted
The 21264 from EV67 on, and the 21364, have a second performance counter
mode, ProfileMe, alongside the Aggregate mode that the driver uses today.
An Aggregate mode sample reports the PC where the overflow interrupt was
taken. On an EV7 that is anywhere from 15 instructions before to 12
after the one that overflowed the counter, which makes perf annotate
output close to useless. In ProfileMe mode the CPU picks one fetched
instruction, follows it down the pipeline, and reports its PC along with
what happened to it: whether it retired, whether it trapped and why,
whether it was a mispredicted branch, stalled before map, or was killed
on a wrong path, and what PCTR1 counted while it was in flight.
Isn't this the thing that ended up as AMD's IBS?
I have no idea, but it wouldn't surprise me too much :)
Looking a bit further, it does seem so. ProfileMe was published by Dean
et al. at MICRO-30 in 1997, and IBS works the same way: tag one
instruction, follow it down the pipeline, and report what happened to
it.
Right, I thought I remembered something like that.
Anyway, I was thinking that perhaps there was something to share with
the existing IBS work, but the AI overlords abused me of that idea quite
quickly. While they are very similar in spirit, they are very distinct
in practice.
Carry on the legacy compute :-)