Having "RCS usage" view open without "Multi contexts" view open

rib / gputop

A GPU profiling tool

160 stars 37 forks source link

Having "RCS usage" view open without "Multi contexts" view open #185

Closed eero-t closed 6 years ago

eero-t commented 6 years ago

"RCS usage" view can be opened from "Multi contexts" trace event view. If I close "Multi contexts" view, "RCS usage" view also closes.

I think "RCS usage" view is hierarchically at least on same level with current data in "Multi contexts" view, i.e. it should be independent.

One possibility would be to have that data fill other half of the current CPU utilization graph.

djdeath commented 6 years ago

Yeah, I guess we could have a live RCS usage over the aggregation period. That would be a new window, different from the existing one that is really tied to the timeline.

eero-t commented 6 years ago

What about just having the plain piechart (with mouse-over info) beside CPU utilization?

Best would be if contexts would use the same colors in all the views, including the "Select contexts" popup in "Per contexts" view, context names there could be prefixed with a little box of suitable color.

djdeath commented 6 years ago

So I tried to display the pie chart based on the aggregation period (which is 50ms by default) and I kind of find it fairly unusable/unreadable (click to watch video) :

Maybe we need a different representation. I think an live busyness for a main window would be good. Then somewhere else a list of the main users.

djdeath commented 6 years ago

Global busyness :

djdeath commented 6 years ago

New live view for usage (from the highlighted button on the left) : usage

eero-t commented 6 years ago

So I tried to display the pie chart based on the aggregation period (which is 50ms by default) and I kind of find it fairly unusable/unreadable (click to watch video)

Agree, that looks pretty bad, but the new live view that you came up with looks great!

Only thing that I find a bit misleading are the view titles:

"perf" is name of the kernel CPU performance counter sampling framework, which OA counter code doesn't use. IMHO you can drop that term completely, it's redundant
And "i915" is the component providing this information, I think it would be better to instead use name of the component which performance is being measured (GPU/ GT/RCS/etc)