aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorAria Shrimpton <me@aria.rip>2024-01-20 00:18:23 +0000
committerAria Shrimpton <me@aria.rip>2024-01-20 00:18:23 +0000
commit9bde0fdfea3fb3da1f9e5a53f436a00ba8947a58 (patch)
tree3097925a1664710ef6bdf3bf854b9543f97d9b31
parent43b346ef8ab91dc13b773dda871433cf98adade8 (diff)
chore: update tasks
-rw-r--r--Tasks.org21
1 files changed, 13 insertions, 8 deletions
diff --git a/Tasks.org b/Tasks.org
index b4ca56d..f31d35b 100644
--- a/Tasks.org
+++ b/Tasks.org
@@ -108,17 +108,22 @@ Ideally, we could use information about the cargo project the file resides in to
We also need to figure out which operations we want to bother counting, and how we can get an 'allocation context'/callstack.
-* DOING Integration
+* DONE Integration
-We have the whole pipeline:
- - Use primrose to get candidates
- - Get/retrieve cached cost models
- - Estimate cost of each candidate
- - TODO: Pick a candidate, and generate code using that candidate
+We have the code to do all this end-to-end, and to run a 'brute force' for comparison.
-* TODO Fix fitting / cost estimation
+* TODO It's fucking broken
-Our cost estimates are broken - probably a bug in the fitting code.
+We currently pick the wrong option for our single (trivial) test case (linkedlist instead of vec).
+Possible cause: The Vec benchmarks give really weird outputs:
+
+#+begin_src text
+insert/8192 time: [794.01 ns 857.54 ns 905.72 ns]
+insert/16384 time: [347.55 ns 648.94 ns 940.90 ns]
+insert/24576 time: [58.427 ns 58.572 ns 58.731 ns]
+insert/32768 time: [6.3390 µs 6.3430 µs 6.3475 µs]
+insert/40960 time: [72.887 ns 73.165 ns 73.482 ns]
+#+end_src
* BLOCKED Benchmarks & Evaluation