JIT: refactor CSE to allow running greedy ML heuristic in release#98729
JIT: refactor CSE to allow running greedy ML heuristic in release#98729AndyAyersMS merged 2 commits intodotnet:mainfrom
Conversation
Revise things a bit so that we can run the greedy ML heuristic in release mode, with a built-in set of "good" parameters. These parameters come from a policy gradient run over a 200 method training set, on the asp.net windows x64 collection. Contributes to dotnet#92915.
ghost
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsRevise things a bit so that we can run the greedy ML heuristic in release mode, with a built-in set of "good" parameters. These parameters come from a policy gradient run over a 200 method training set, on the asp.net windows x64 collection. Contributes to #92915.
|
|
@EgorBo PTAL This will be a bit painful to review as I moved a bunch of methods to a new intermediate class, and in doing so physically moved them in the file. For the most part things did not change (much). The baked-in parameters may or may not be any good. We'll see. I will push an interim change to enable it by default. |
Diff results for #98729Throughput diffsThroughput diffs for windows/arm64 ran on windows/x64MinOpts (-0.00% to +0.01%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (+0.00% to +0.01%)
FullOpts (+0.00% to +0.01%)
Details here |
|
Looking 👀 |
Actually I'm going to hold off; let's get this bit merged and then I will put up a draft PR to flip it on by default. |
EgorBo
left a comment
EgorBo
left a comment
There was a problem hiding this comment.
As far as I can follow your changes, they look reasonable to me, LGTM!
Are the diffs huge? 🙂 |
The code size diffs don't look very good right now. One of the things I need to figure out is how to make sure that the secondary objectives don't get too crazy. |
|
This is a no diff change so failures above are unrelated. |
|
Draft PR to enable: #98776 |
3 participants
Revise things a bit so that we can run the greedy ML heuristic in release mode, with a built-in set of "good" parameters.
These parameters come from a policy gradient run over a 200 method training set, on the asp.net windows x64 collection.
Contributes to #92915.