From: Alberto Cabrera Pérez <redacted>
Date: Wed, 26 Nov 2025 21:14:54 +0000 (+0000)
Subject: Fix chunks being too small with small matrix sizes (llama/17526)
X-Git-Tag: upstream/0.9.4.395~113
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=c6b3a42cd02d5b248e35fb524c1a2a294e95e2f8;p=pkg%2Fggml%2Fsources%2Fggml

Fix chunks being too small with small matrix sizes (llama/17526)
---

diff --git a/src/ggml-cpu/repack.cpp b/src/ggml-cpu/repack.cpp
index d1321191..875faedf 100644
--- a/src/ggml-cpu/repack.cpp
+++ b/src/ggml-cpu/repack.cpp
@@ -1731,12 +1731,13 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
             nchunk0 = (nr0 + min_chunk_size - 1) / min_chunk_size;
         }
 
-        if (nth == 1 || nchunk0 < nth || disable_chunking) {
+        int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
+        // Only increase nchunk0 to nth if it won't make chunks too small
+        if (nth == 1 || ((nchunk0 < nth || disable_chunking) && (nr0 + nth - 1) / nth >= min_chunk_size)) {
             nchunk0 = nth;
+            dr0 = (nr0 + nchunk0 - 1) / nchunk0;
         }
 
-        const int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
-
         // Ensure nchunk doesn't exceed the number of rows divided by minimum chunk size
         // This prevents creating too many tiny chunks that could overlap after alignment
         const int64_t max_nchunk = (nr0 + min_chunk_size - 1) / min_chunk_size;