From 5fbff1aa3a37761c0013af4e8c398398973bf791 Mon Sep 17 00:00:00 2001
From: chraac <chraac@gmail.com>
Date: Thu, 15 Jan 2026 11:57:37 +0800
Subject: [PATCH] feat: add OpenCL kernel support for division operations and
 update includes

---
 ggml/src/ggml-opencl/CMakeLists.txt     |  2 ++
 ggml/src/ggml-opencl/ggml-opencl.cpp    |  4 +++
 ggml/src/ggml-opencl/kernels/div.h      | 46 ++++++++++++-------------
 ggml/src/ggml-opencl/kernels/ocl_defs.h |  6 ++--
 4 files changed, 33 insertions(+), 25 deletions(-)

diff --git a/ggml/src/ggml-opencl/CMakeLists.txt b/ggml/src/ggml-opencl/CMakeLists.txt
index 2a4b79eb6a..977b945d16 100644
--- a/ggml/src/ggml-opencl/CMakeLists.txt
+++ b/ggml/src/ggml-opencl/CMakeLists.txt
@@ -22,6 +22,8 @@ if (GGML_OPENCL_USE_ADRENO_KERNELS)
     add_compile_definitions(GGML_OPENCL_USE_ADRENO_KERNELS)
 endif ()
 
+target_include_directories(${TARGET_NAME} PRIVATE "${CMAKE_CURRENT_LIST_DIR}/kernels")
+
 if (GGML_OPENCL_EMBED_KERNELS)
     add_compile_definitions(GGML_OPENCL_EMBED_KERNELS)
 
diff --git a/ggml/src/ggml-opencl/ggml-opencl.cpp b/ggml/src/ggml-opencl/ggml-opencl.cpp
index 982fa59e43..38d72cba6b 100644
--- a/ggml/src/ggml-opencl/ggml-opencl.cpp
+++ b/ggml/src/ggml-opencl/ggml-opencl.cpp
@@ -29,6 +29,10 @@
 #include <charconv>
 #include <mutex>
 
+namespace ocl_kernel_prototypes {
+    #include "div.h"
+}
+
 #undef MIN
 #undef MAX
 #define MIN(a, b) ((a) < (b) ? (a) : (b))
diff --git a/ggml/src/ggml-opencl/kernels/div.h b/ggml/src/ggml-opencl/kernels/div.h
index 75f66a0a2a..fd65b5706f 100644
--- a/ggml/src/ggml-opencl/kernels/div.h
+++ b/ggml/src/ggml-opencl/kernels/div.h
@@ -4,28 +4,28 @@
 
 #include "ocl_defs.h"
 
-kernel void kernel_div(global char * src0,
-                       ulong         offset0,
-                       global char * src1,
-                       ulong         offset1,
-                       global char * dst,
-                       ulong         offsetd,
-                       ulong         nb00,
-                       ulong         nb01,
-                       ulong         nb02,
-                       ulong         nb03,
-                       int           ne10,
-                       int           ne11,
-                       int           ne12,
-                       int           ne13,
-                       ulong         nb10,
-                       ulong         nb11,
-                       ulong         nb12,
-                       ulong         nb13,
-                       int           ne0,
-                       ulong         nb0,
-                       ulong         nb1,
-                       ulong         nb2,
-                       ulong         nb3);
+OCL_KERNEL void kernel_div(OCL_GLOBAL char * src0,
+                           ulong             offset0,
+                           OCL_GLOBAL char * src1,
+                           ulong             offset1,
+                           OCL_GLOBAL char * dst,
+                           ulong             offsetd,
+                           ulong             nb00,
+                           ulong             nb01,
+                           ulong             nb02,
+                           ulong             nb03,
+                           int               ne10,
+                           int               ne11,
+                           int               ne12,
+                           int               ne13,
+                           ulong             nb10,
+                           ulong             nb11,
+                           ulong             nb12,
+                           ulong             nb13,
+                           int               ne0,
+                           ulong             nb0,
+                           ulong             nb1,
+                           ulong             nb2,
+                           ulong             nb3);
 
 #endif  // __KERNELS_DIV_H__
diff --git a/ggml/src/ggml-opencl/kernels/ocl_defs.h b/ggml/src/ggml-opencl/kernels/ocl_defs.h
index a0b8feedb1..3faf1ecf56 100644
--- a/ggml/src/ggml-opencl/kernels/ocl_defs.h
+++ b/ggml/src/ggml-opencl/kernels/ocl_defs.h
@@ -4,10 +4,12 @@
 
 #ifdef __OPENCL_C_VERSION__
 // Device (OpenCL) Definitions
+#    define OCL_KERNEL kernel
+#    define OCL_GLOBAL global
 #else
 // Host (C++) Definitions
-#    define kernel
-#    define global
+#    define OCL_KERNEL
+#    define OCL_GLOBAL
 #    define __kernel
 #    define __global
 #    define ulong cl_ulong