[PATCH] D53562: [ARM] Use the Cortex-A57 sched model for Cortex-A72

[Sam Parker via Phabricator via llvm-commits]

samparker added a comment.

Ok, fair enough, my LNT numbers show that the MISched results are more variable:

Regressions (%):

| benchmark                | Model | Model + MISched |
| Stanford/FloatMM         |       | 40.1            |
| McGill/queens            | 29.6  | 29.4            |
| Stanford/Puzzle          | 16.05 | 15.6            |
| FreeBench/mason/mason    |       | 9.27            |
| Olden/power/power        |       | 7.61            |
| VersaBench/ecbdes/ecbdes |       | 5.25            |
| BenchmarkGame/fannkuch   | 6.55  |                 |
| Fhourstones/fhourstones  | 5.75  |                 |
|

Improvements (%):

| benchmark                                          | Model | Model + MISched |
| Stanford/Perm                                      | 21.7  | 22.53           |
| Olden/mst/mst                                      | 22.05 | 22.21           |
| FreeBench/fourinarow/fourinarow                    | 22.83 | 21.21           |
| TSVC/CrossingThresholds-dbl/CrossingThresholds-dbl | 14.17 | 19.52           |
| TSVC/Searching-flt/Searching-flt                   | 16.26 | 19.23           |
| TSVC/Searching-dbl/Searching-dbl                   | 16.28 | 19.15           |
|

I will remove the use of the feature for now.

cheers,


https://reviews.llvm.org/D53562