Open githubmanticore opened 1 year ago
➤ Ilya Kuznetsov commented:
Regarding class members vs stack variables. This optimization still works, and its worth about 5% overall performance. Need to re-implement it.
➤ Ilya Kuznetsov commented:
Benchmarks on hetzner servers show that stack var optimization still works. Reimplemented it in a537b16f
➤ Ilya Kuznetsov commented:
With recent changes, 'new' 4gb-v2 is 7% slower than the 'old' 4gb-v2 running the use around
query.
➤ Ilya Kuznetsov commented:
Reverted max_block_docs to 512 in 0336d10f. Now 'new' 4gb-v2 is 4% slower than the 'old' 4gb-v2 running the use around
query.
➤ Ilya Kuznetsov commented:
Note: keywords
query is 6% slower.
➤ Ilya Kuznetsov commented:
All is done except subtree caching. It is commented out, need to benchmark it.
Streamed hits code is now in 4gb-v2, but it needs some improvements: