Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Do not keep around default quantization matrices
This saves ~3Mb of static memory. Speed impact with multiple repetitions is negligible, significant speed improvement for one-off decodes. Slight variation in 6th digit of en/decoding results due to more approximate quant matrix computation. ``` Before: Encoding kPixels Bytes BPP E MP/s D MP/s Max norm pnorm BPP*pnorm Bugs ------------------------------------------------------------------------------------------------------------ jxl:d0.7 13270 3400483 2.0499581 5.466 47.067 1.50473666 0.47348297 0.970620226409 0 jxl:d1 13270 2678725 1.6148512 5.750 50.382 1.68153656 0.60257544 0.973069654737 0 jxl:d2 13270 1683788 1.0150602 6.102 53.015 3.00312042 0.95243347 0.966777268691 0 jxl:d4 13270 1004956 0.6058309 5.979 37.066 4.85261440 1.50554379 0.912104941537 0 jxl:d6 13270 718865 0.4333629 5.772 36.305 7.34840679 1.97809701 0.857233817426 0 jxl:d8 13270 576902 0.3477815 5.852 35.697 8.16245556 2.36179685 0.821389132427 0 Aggregate: 13270 1362308 0.8212584 5.817 42.674 3.60956523 1.11400975 0.914889813753 0 2268 x 1512, geomean: 90.43 MP/s [54.13, 91.49], 100 reps, 0 threads. 2268 x 1512, 56.57 MP/s [56.57, 56.57], 1 reps, 0 threads. 444 x 258, geomean: 62.33 MP/s [6.49, 63.54], 100 reps, 0 threads. 444 x 258, 7.06 MP/s [7.06, 7.06], 1 reps, 0 threads. After: Encoding kPixels Bytes BPP E MP/s D MP/s Max norm pnorm BPP*pnorm Bugs ------------------------------------------------------------------------------------------------------------ jxl:d0.7 13270 3400473 2.0499520 5.424 46.559 1.50473595 0.47348321 0.970617862124 0 jxl:d1 13270 2678723 1.6148500 5.710 49.788 1.68153679 0.60257545 0.973068940575 0 jxl:d2 13270 1683788 1.0150602 6.064 52.348 3.00312042 0.95243348 0.966777275451 0 jxl:d4 13270 1004957 0.6058315 5.942 36.558 4.85261631 1.50554828 0.912108572859 0 jxl:d6 13270 718865 0.4333629 5.736 35.850 7.34841013 1.97809768 0.857234107317 0 jxl:d8 13270 576891 0.3477748 5.811 35.189 8.16245461 2.36179583 0.821373116628 0 Aggregate: 13270 1362303 0.8212554 5.777 42.136 3.60956547 1.11401039 0.914887016940 0 2268 x 1512, geomean: 90.40 MP/s [71.12, 91.11], 100 reps, 0 threads. 2268 x 1512, 68.69 MP/s [68.69, 68.69], 1 reps, 0 threads. 444 x 258, geomean: 63.93 MP/s [37.00, 64.71], 100 reps, 0 threads. 444 x 258, 37.56 MP/s [37.56, 37.56], 1 reps, 0 threads. ```
- Loading branch information