When I run the encode_bench example in release mode, the GPU timings are wildly wrong:
=== Testing H264 with Default tuning ===
Config: Cbr, bitrate=8000000, GOP=120, B-frames=0
Encoded 300 frames, 300677 bytes, 3103.4:1 compression
GPU Encode timings: Min: 1.213583ms, Max: 5509.59579136s, Avg: 3666.656801518s, P99: 5509.5853056s, Total: 1096330383ms
Frame Latency Encode timings: Min: 2.006408ms, Max: 8.098792ms, Avg: 3.588754ms, P99: 4.868367ms, Total: 1073ms
Wall Latency Encode timings: Min: 1.956413ms, Max: 7.935761ms, Avg: 3.531354ms, P99: 4.729837ms, Total: 1055ms
In debug mode everything is a bit slower and consumes more CPU, but the GPU timings are correct:
=== Testing H264 with Default tuning ===
Config: Cbr, bitrate=8000000, GOP=120, B-frames=0
Encoded 300 frames, 300677 bytes, 3103.4:1 compression
GPU Encode timings: Min: 2.9995ms, Max: 7.195667ms, Avg: 4.952396ms, P99: 6.743583ms, Total: 1480ms
Frame Latency Encode timings: Min: 3.139149ms, Max: 7.440555ms, Avg: 5.142626ms, P99: 7.002523ms, Total: 1537ms
Wall Latency Encode timings: Min: 3.072359ms, Max: 7.321645ms, Avg: 5.06514ms, P99: 6.930739ms, Total: 1514ms
This is with the anv driver, on a Kaby Lake GPU, from Mesa master (commit 2b7a72457a55b80e02700debdbc11037db7201e2), with ANV_DEBUG=video-encode set, on pixelforge master.
When I run the encode_bench example in release mode, the GPU timings are wildly wrong:
In debug mode everything is a bit slower and consumes more CPU, but the GPU timings are correct:
This is with the anv driver, on a Kaby Lake GPU, from Mesa master (commit 2b7a72457a55b80e02700debdbc11037db7201e2), with ANV_DEBUG=video-encode set, on pixelforge master.