perf: optimize H3 inference and segmented weight prefetch (#2240)

This commit is contained in:
leejet authored and GitHub committed 2026-10-30 00:38:14 +08:00
1 parent e7b7ec6d6f
commit 37bc88b5e9
17 files changed
+367 -210

No files matched your search

+9 -1
View File
@@ -178,7 +178,7 @@ namespace sd {
disable();
return;
}
switch (manager->prefetch_params(owner_id_, params)) {
switch (manager->prepare_prefetch_params(owner_id_, params)) {
case WeightPrefetchResult::Scheduled:
queued_params_ = std::move(params);
queued_segment_ = next_segment;
@@ -195,6 +195,14 @@ namespace sd {
disable();
}
void SegmentWeightPipeline::submit_next() {
if (queued_segment_ != SIZE_MAX) {
if (auto manager = residency_manager_.lock()) {
manager->submit_prefetched_params(owner_id_);
}
}
}
void SegmentWeightPipeline::clear() {
if (auto manager = residency_manager_.lock()) {
manager->clear_prefetched_params(owner_id_);