Skip to content
Closed
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion src/prefill/mmq_resident_sort.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,10 @@ inline bool mmq_resident_sort_requested() {
inline bool mmq_resident_sort_eligible(bool requested, bool use_mmq, bool native_layout, bool has_cache,
const int32_t* layer_resident, int64_t n_expert,
bool stream_all, bool layer_stream_empty) {
if (!requested || !use_mmq || native_layout || !has_cache || !layer_resident || n_expert <= 0) return false;
// native-layout packs too: slot/src/off are built the same way for both layouts, and nothing reads off[e + 1] as a
// count (4x GP100, sm_60: 1.2K prompt 7.27 -> 5.32 s, 35K 63.3 -> 54.1 s, the same bits)
(void) native_layout;
if (!requested || !use_mmq || !has_cache || !layer_resident || n_expert <= 0) return false;
// All experts, including unselected ones: leave every partial/streamed layer in ID order.
if (stream_all && !layer_stream_empty) return false;
for (int64_t e = 0; e < n_expert; ++e) if (layer_resident[e] < 0) return false;
Expand Down