Feature request: zones with manual timestamps

[ashpil]

I've found it frustrating to integrate GPU zones in my application. Tracy wants to be very involved in the details of how GPU zones are constructed, and because of this, the interface is quite complex:

1. A Tracy GPU context must be constructed
2. Start/end queries must be recorded
3. Query times must be submitted later
4. Calibration may or may not be done

I would much rather prefer a simpler API. A single function, something like TracyManualZone(track_name, zone_name, start_time, end_time). For my use cases, this would be just as good as the GPU zones, with significantly less integration complexity (since my engine already has all the infrastructure to collect calibrated GPU timestamps -- I don't have to create an alternative just for Tracy, I can reuse all my existing infrastructure).

Furthermore, I can envision this catering to other use cases, where automatically captured CPU timestamps may be irrelevant, such as displaying events that happened on a remote server.

[slomp]

Yes, I vouch for something like this. I did experiment locally with a similar feature (I called it "virtual threads"), and they are quite handy indeed (great for tracking asynchronous task activity flowing through threads).

There are, however, a few things to note when "bringing your own timestamps":

• Tracy would either need to know about your timestamps in order to convert them to Tracy timestamps it can use to draw a consistent timeline
• You would be responsible for doing such a conversion/calibration (probably the better idea; I'd advocate for exposing the incremental linear regression utility from TracyCUDA for that)

Then, in the case of GPU zones specifically, Tracy can correlate the GPU device activity with the CPU API call that caused that activity (to measure command latencies and such). We would need a variant of TracyManualZone for that too.