perf: avoid copying of creating memory dist calculator #2219
Conversation
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2219 +/- ##
==========================================
- Coverage 81.06% 81.01% -0.05%
==========================================
Files 186 186
Lines 53745 53746 +1
Branches 53745 53746 +1
==========================================
- Hits 43566 43545 -21
- Misses 7704 7710 +6
- Partials 2475 2491 +16
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
| metric_type: self.metric_type, | ||
| }) | ||
| } | ||
| } | ||
|
|
||
| struct InMemoryDistanceCal { | ||
| vectors: Arc<MatrixView<Float32Type>>, | ||
| query: Vec<f32>, | ||
| query: *const f32, |
There was a problem hiding this comment.
Can we use &[f32] with lifetime? Using raw pointer means that the lifetime of query is not guaranteed when the distance() method is called.
There was a problem hiding this comment.
i tried this but found that the dist_calc is created by the trait method of VectorStorage, so it's impossible to add lifetime for this.
any better way?
There was a problem hiding this comment.
if we don't care about locality
Arc<[f32]> works
There was a problem hiding this comment.
We should probably make queries some sort of Arc<[T]> early in the query, so it's cheaply clone-able.
There was a problem hiding this comment.
a bit more pedantic. we probably want something like this https://github.com/pytorch/pytorch/blob/main/c10/util/intrusive_ptr.h
but I'm not sure if there is an equivalent that is easily implementable in rust
There was a problem hiding this comment.
the vector in Query is Arc<dyn Array>, changed the type from &[f32] to ArrayRef
| pub fn row(&self, i: usize) -> Option<ArrayRef> { | ||
| assert!( | ||
| !self.transpose, | ||
| "Centroid is not defined for transposed matrix." | ||
| ); | ||
| if i >= self.num_rows() { | ||
| None | ||
| } else { | ||
| let dim = self.num_columns(); | ||
| Some(self.data.slice(i * dim, dim)) | ||
| } | ||
| } | ||
|
|
There was a problem hiding this comment.
yes, see
pub fn vector(&self, id: u32) -> ArrayRef {
self.vectors.row(id as usize).unwrap()
}
d1da882 to
7257e5b
Compare
No description provided.