Apple M1 aeson performance regression
Hello there!
I find aeson(620e2d6) regression between 8.10.7 and 9.4.2 GHC versions on my Apple M1 Macbook.
It can be reproduced using internal criterion benchmark with -O2 -fexpose-all-unfoldings -fspecialise-aggressively GHC flags. Same version of text used for both benchmarks.
If you need my bench.sh or results of other benchmarks -- let me know, i would like to help:)
| Name | Old | New | Ratio |
|---|---|---|---|
| Twitter/decode/Generic/direct/twitter100 | 1.1543492e-3 | 5.680091e-4 | 2.0322723 |
| Twitter/decode/Manual/direct/twitter100 | 1.0893879e-3 | 5.620007e-4 | 1.9384102 |
| Twitter/decode/TH/direct/twitter100 | 1.2560157e-3 | 5.5859325e-4 | 2.2485337 |
| Twitter/encode/Generic/direct/twitter100 | 1.1647143e-4 | 3.084223e-4 | 0.3776362 |
| Twitter/encode/Generic/viaValue/twitter100 | 3.6394934e-4 | 4.864574e-4 | 0.7481628 |
| Twitter/encode/Manual/direct/twitter100 | 9.9359924e-5 | 3.055389e-4 | 0.32519564 |
| Twitter/encode/Manual/viaValue/twitter100 | 3.5252149e-4 | 5.158261e-4 | 0.6834115 |
| Twitter/encode/TH/direct/twitter100 | 7.874909e-5 | 2.142956e-4 | 0.36747882 |
| Twitter/encode/TH/viaValue/twitter100 | 3.59988e-4 | 4.8431472e-4 | 0.7432935 |
Thanks!