From: Kirill A. Korinsky <kirill@korins.ky>
Subject: misc/llama.cpp: update to b7086; devel/libggml update to 20251120
To: OpenBSD ports <ports@openbsd.org>
Date: Sun, 23 Nov 2025 14:43:57 +0100

ports@,

I'd like to update devel/libggml to commit
781baf2a14d9e0aaee542b2e1bb918bfc4132199 which is sync with misc/llama.cpp
b7086.

Tested on -current/amd64, no regression on my usecase.

I also run make test on whisper.cpp, works as before update.

check_sym reprots new symbol, so, I've increased minor version of lib.

Ok?

Index: devel/libggml/Makefile
===================================================================
RCS file: /home/cvs/ports/devel/libggml/Makefile,v
diff -u -p -r1.4 Makefile
--- devel/libggml/Makefile	5 Nov 2025 08:59:44 -0000	1.4
+++ devel/libggml/Makefile	21 Nov 2025 22:00:22 -0000
@@ -2,13 +2,12 @@ COMMENT=		tensor library for machine lea
 
 GH_ACCOUNT=		ggml-org
 GH_PROJECT=		ggml
-GH_COMMIT=		09aa758381718f7731c148238574a7e169001f13
-REVISION=		0
-DISTNAME=		ggml-0.9.4pl20251101
+GH_COMMIT=		781baf2a14d9e0aaee542b2e1bb918bfc4132199
+DISTNAME=		ggml-0.9.4pl20251120
 PKGNAME=		lib${DISTNAME}
 
-SHARED_LIBS +=		ggml 3.0
-SHARED_LIBS +=		ggml-base 3.0
+SHARED_LIBS +=		ggml 3.1
+SHARED_LIBS +=		ggml-base 3.1
 
 CATEGORIES=		devel
 
Index: devel/libggml/distinfo
===================================================================
RCS file: /home/cvs/ports/devel/libggml/distinfo,v
diff -u -p -r1.2 distinfo
--- devel/libggml/distinfo	4 Nov 2025 15:05:00 -0000	1.2
+++ devel/libggml/distinfo	21 Nov 2025 21:29:46 -0000
@@ -1,2 +1,2 @@
-SHA256 (ggml-0.9.4pl20251101-09aa7583.tar.gz) = fx+ZI4GhV5KlZlGM3QDWERyj1xXY8t5vh5ELtkwK15Y=
-SIZE (ggml-0.9.4pl20251101-09aa7583.tar.gz) = 2330931
+SHA256 (ggml-0.9.4pl20251120-781baf2a.tar.gz) = qu5WT9HUZ12982U0x4SWNbBDMTameA7DvFXccUbldis=
+SIZE (ggml-0.9.4pl20251120-781baf2a.tar.gz) = 2344462
Index: devel/libggml/patches/patch-src_ggml-cpu_repack_cpp
===================================================================
RCS file: devel/libggml/patches/patch-src_ggml-cpu_repack_cpp
diff -N devel/libggml/patches/patch-src_ggml-cpu_repack_cpp
--- devel/libggml/patches/patch-src_ggml-cpu_repack_cpp	4 Nov 2025 15:05:00 -0000	1.1
+++ /dev/null	1 Jan 1970 00:00:00 -0000
@@ -1,60 +0,0 @@
-Fix garbled output with REPACK at high thread counts
-https://github.com/ggml-org/llama.cpp/commit/1f5accb8d0056e6099cd5b772b1cb787dd590a13
-
-Index: src/ggml-cpu/repack.cpp
---- src/ggml-cpu/repack.cpp.orig
-+++ src/ggml-cpu/repack.cpp
-@@ -1678,10 +1678,24 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int6
-         int64_t chunk_size = (nr + nth_scaled - 1) / nth_scaled;
-         int64_t nchunk     = (nr + chunk_size - 1) / chunk_size;
- 
-+        // Ensure minimum chunk size to avoid alignment issues with high thread counts
-+        // Minimum chunk size should be at least NB_COLS to prevent overlapping chunks after alignment
-+        const int64_t min_chunk_size = NB_COLS;
-+        if (nchunk > 0 && (nr / nchunk) < min_chunk_size && nr >= min_chunk_size) {
-+            nchunk = (nr + min_chunk_size - 1) / min_chunk_size;
-+        }
-+
-         if (nth == 1 || nchunk < nth || disable_chunking) {
-             nchunk = nth;
-         }
- 
-+        // 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 = (nr + min_chunk_size - 1) / min_chunk_size;
-+        if (nchunk > max_nchunk) {
-+            nchunk = max_nchunk;
-+        }
-+
-         if (ith == 0) {
-             // Every thread starts at ith, so the first unprocessed chunk is nth.  This save a bit of coordination right at the start.
-             ggml_threadpool_chunk_set(params->threadpool, nth);
-@@ -1695,8 +1709,15 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int6
-         while (current_chunk < nchunk) {
-             int64_t src0_start = (current_chunk * ne01) / nchunk;
-             int64_t src0_end   = ((current_chunk + 1) * ne01) / nchunk;
-+
-+            // Align boundaries to NB_COLS - round up to ensure all data is included
-+            // The chunk size limiting above ensures chunks are large enough to prevent overlaps
-             src0_start = (src0_start % NB_COLS) ? src0_start + NB_COLS - (src0_start % NB_COLS) : src0_start;
-             src0_end   = (src0_end   % NB_COLS) ? src0_end   + NB_COLS - (src0_end   % NB_COLS) : src0_end;
-+            if (src0_end > ne01) {
-+                src0_end = ne01;
-+            }
-+
-             if (src0_start >= src0_end) {
-                 break;
-             }
-@@ -1808,8 +1829,12 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int6
-             int64_t src0_cur_start = (ith * ne01) / nth;
-             int64_t src0_cur_end   = ((ith + 1) * ne01) / nth;
- 
-+            // Align boundaries to NB_COLS - round up to ensure all data is included
-             src0_cur_start = (src0_cur_start % NB_COLS) ? src0_cur_start + NB_COLS - (src0_cur_start % NB_COLS) : src0_cur_start;
-             src0_cur_end   = (src0_cur_end   % NB_COLS) ? src0_cur_end   + NB_COLS - (src0_cur_end   % NB_COLS) : src0_cur_end;
-+            if (src0_cur_end > ne01) {
-+                src0_cur_end = ne01;
-+            }
- 
-             if (src0_cur_start >= src0_cur_end) {
-                 return;
Index: misc/llama.cpp/Makefile
===================================================================
RCS file: /home/cvs/ports/misc/llama.cpp/Makefile,v
diff -u -p -r1.12 Makefile
--- misc/llama.cpp/Makefile	12 Nov 2025 09:33:37 -0000	1.12
+++ misc/llama.cpp/Makefile	21 Nov 2025 23:31:00 -0000
@@ -2,8 +2,7 @@ COMMENT =		LLM inference system
 
 GH_ACCOUNT =		ggml-org
 GH_PROJECT =		llama.cpp
-GH_TAGNAME =		b6934
-REVISION =		0
+GH_TAGNAME =		b7086
 PKGNAME =		llama.cpp-0.0.${GH_TAGNAME:S/b//}
 
 SHARED_LIBS +=		llama 3.0
Index: misc/llama.cpp/distinfo
===================================================================
RCS file: /home/cvs/ports/misc/llama.cpp/distinfo,v
diff -u -p -r1.5 distinfo
--- misc/llama.cpp/distinfo	4 Nov 2025 15:05:41 -0000	1.5
+++ misc/llama.cpp/distinfo	21 Nov 2025 23:31:15 -0000
@@ -1,2 +1,2 @@
-SHA256 (llama.cpp-b6934.tar.gz) = qsr4P+8j/z/nK8k8/Iv1/QTdSFhqIshFYwCI7L5/a7k=
-SIZE (llama.cpp-b6934.tar.gz) = 26417348
+SHA256 (llama.cpp-b7086.tar.gz) = FmxxNbpcxDsrZQtbfUkFl3h1f/CbXcEjWJEKpXEszwA=
+SIZE (llama.cpp-b7086.tar.gz) = 27243237


-- 
wbr, Kirill