Compare commits

...

3 Commits

Author SHA1 Message Date
Mike Griese
4e82c9accf Revert "Only flush during Present()"
This reverts commit 74181014cf.
2023-07-07 06:28:30 -05:00
Mike Griese
74181014cf Only flush during Present()
This is interesting, but didn't meaningfully move the needle

* one line flush change

32.443MB, 3.905s, 8.307MB/s
32.443MB, 3.938s, 8.238MB/s
32.443MB, 3.915s, 8.288MB/s

* with lock the buffer and flush on present

32.443MB, 3.891s, 8.339MB/s
32.443MB, 3.872s, 8.379MB/s
32.443MB, 3.877s, 8.368MB/s
2023-07-07 06:27:36 -05:00
Mike Griese
1a9612775e This one line change seems to improve perf ~30%. Doesn't _seem_ to break things...
* 1.17

32.443MB, 5.329s, 6.088MB/s
32.443MB, 5.335s, 6.081MB/s

* 1.18

32.443MB, 5.475s, 5.926MB/s
32.443MB, 5.509s, 5.889MB/s
32.443MB, 5.530s, 5.867MB/s

* main

6.489MB, 1.186s, 5.473MB/s
6.489MB, 1.203s, 5.396MB/s
6.489MB, 1.182s, 5.489MB/s
6.489MB, 1.182s, 5.489MB/s
6.489MB, 1.118s, 5.804MB/s
6.489MB, 1.117s, 5.807MB/s
32.443MB, 5.308s, 6.112MB/s

* one line flush change

32.443MB, 3.905s, 8.307MB/s
32.443MB, 3.938s, 8.238MB/s
32.443MB, 3.915s, 8.288MB/s
2023-07-07 06:01:21 -05:00

View File

@@ -122,7 +122,7 @@ CATCH_RETURN();
// If we flushed for any reason other than circling (i.e, a sequence that we
// didn't understand), we don't need to push the buffer out on EndPaint.
_noFlushOnEnd = !circled;
_noFlushOnEnd = true; // !circled;
_trace.TraceTriggerCircling(*pForcePaint);