Effectfully/inline fix #6842
Effectfully/inline fix #6842
Conversation
effectfully
added
Do not merge
EXPERIMENT
effectfully
temporarily deployed
to
github-pages
GitHub Actions
Inactive
| ({cpu: 397650662016 | ||
| | mem: 1503390947}) No newline at end of file | ||
| ({cpu: 409445366016 | ||
| | mem: 1577107847}) |
There was a problem hiding this comment.
+3% CPU, +5% MEM
| 395 |
There was a problem hiding this comment.
... but smaller size. And the best part is that the size even smaller than without inlining fix at all, i.e. this is smaller than on master.
Here's a three-way merge screenshot of CPU/MEM/size for "8 queens" (first @zliu41's branch, then master, then this branch):
| ({cpu: 5476180 | ||
| | mem: 28020}) No newline at end of file | ||
| ({cpu: 5428180 | ||
| | mem: 27720}) |
There was a problem hiding this comment.
A slight cost reduction.
| ({cpu: 304890223 | ||
| | mem: 1188386}) No newline at end of file | ||
| ({cpu: 304602223 | ||
| | mem: 1186586}) |
There was a problem hiding this comment.
a slight reduction
| ({cpu: 88899135 | ||
| | mem: 351394}) No newline at end of file | ||
| ({cpu: 88611135 | ||
| | mem: 349594}) |
There was a problem hiding this comment.
a slight reduction
| ({cpu: 174115055 | ||
| | mem: 815970}) No newline at end of file | ||
| ({cpu: 174099055 | ||
| | mem: 815870}) |
There was a problem hiding this comment.
Then a bunch of reductions.
| ({cpu: 243481580 | ||
| | mem: 1195798}) No newline at end of file | ||
| ({cpu: 243561580 | ||
| | mem: 1196298}) |
There was a problem hiding this comment.
Then increases again.
| (\xxx -> 1) ((\s -> s s) (\s -> force trace "hello" (\z -> s s z))) No newline at end of file | ||
| 1 |
There was a problem hiding this comment.
A semantics change.
| ({cpu: 5252180 | ||
| | mem: 26620}) No newline at end of file | ||
| ({cpu: 5300180 | ||
| | mem: 26920}) |
There was a problem hiding this comment.
A slight increase.
| (program | ||
| 1.1.0 | ||
| ((\s -> s s) | ||
| (\s ds -> force (case ds [(delay 42), (\x xs -> delay (s s xs))])) | ||
| ((\x0 -> | ||
| (\s -> s s) | ||
| (\s ds -> force (case ds [(delay 42), (\x xs -> delay (s s xs))])) | ||
| x0) | ||
| ((\s -> s s) | ||
| (\s n -> | ||
| force |
There was a problem hiding this comment.
This I guess show the deficiency of the approach. It's nice that we can inline recursive functions better, but we still can't inline fully applied recursive functions.
This shows what happens if we add #6839 to #6841 (i.e. the base branch is
zliu41/inline--fix, notmaster). See the comments.