There seems to be a bubble waking up from waitForFences, which costs a few percent performance and also increased variance in performance. This change inserts an "almost_ready" fence when the graph is about 80% complete and we waitForFences for the almost_ready fence and then spin (with _mm_pauses) waiting for the final fence to be signaled. |
||
|---|---|---|
| .. | ||
| cmake | ||
| include | ||
| src | ||
| .gitignore | ||
| CMakeLists.txt | ||