xaptronic
diff --git a/‎ggml-opencl.cpp
Copy file name to clipboardExpand all lines: ggml-opencl.cpp
+9Lines changed: 9 additions & 0 deletions b/‎ggml-opencl.cpp
Copy file name to clipboardExpand all lines: ggml-opencl.cpp
+9Lines changed: 9 additions & 0 deletions
diff --git a/‎ggml-opencl.h
Copy file name to clipboardExpand all lines: ggml-opencl.h
+2Lines changed: 2 additions & 0 deletions b/‎ggml-opencl.h
Copy file name to clipboardExpand all lines: ggml-opencl.h
+2Lines changed: 2 additions & 0 deletions
diff --git a/‎llama.cpp
Copy file name to clipboardExpand all lines: llama.cpp
+5-1Lines changed: 5 additions & 1 deletion b/‎llama.cpp
Copy file name to clipboardExpand all lines: llama.cpp
+5-1Lines changed: 5 additions & 1 deletion
@@ -662,6 +662,15 @@ static void ggml_cl_pool_free(cl_mem mem, size_t size) {
     clReleaseMemObject(mem);
 }
 
+void ggml_cl_free_data(const struct ggml_tensor* tensor) {
+    if (tensor->backend != GGML_BACKEND_GPU) {
+        return;
+    }
+
+    cl_mem mem = (cl_mem)tensor->data;
+    clReleaseMemObject(mem);
+}
+
 static cl_int ggml_cl_h2d_tensor_2d(cl_command_queue queue, cl_mem dst, size_t offset, const struct ggml_tensor * src, uint64_t i3, uint64_t i2, cl_event* ev) {
     cl_int err;
     const uint64_t ne0 = src->ne[0];
 
@@ -16,6 +16,8 @@ void   ggml_cl_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor
 void * ggml_cl_host_malloc(size_t size);
 void   ggml_cl_host_free(void * ptr);
 
+void ggml_cl_free_data(const struct ggml_tensor* tensor);
+
 void ggml_cl_transform_tensor(struct ggml_tensor * tensor);
 void ggml_cl_load_data(const char * fname, struct ggml_tensor * tensor, size_t offset);
 
 
@@ -210,7 +210,11 @@ struct llama_model {
         for (size_t i = 0; i < tensors_by_name.size(); ++i) {
             ggml_cuda_free_data(tensors_by_name[i].second);
         }
-#endif // GGML_USE_CUBLAS
+#elif defined(GGML_USE_CLBLAST)
+        for (size_t i = 0; i < tensors_by_name.size(); ++i) {
+            ggml_cl_free_data(tensors_by_name[i].second);
+        }
+#endif
     }
 };
Original file line number	Diff line number	Diff line change
`@@ -210,7 +210,11 @@ struct llama_model {`
`210`	`210`	`for (size_t i = 0; i < tensors_by_name.size(); ++i) {`
`211`	`211`	`ggml_cuda_free_data(tensors_by_name[i].second);`
`212`	`212`	`}`
`213`		`-#endif // GGML_USE_CUBLAS`
	`213`	`+#elif defined(GGML_USE_CLBLAST)`
	`214`	`+ for (size_t i = 0; i < tensors_by_name.size(); ++i) {`
	`215`	`+ ggml_cl_free_data(tensors_by_name[i].second);`
	`216`	`+ }`
	`217`	`+#endif`
`214`	`218`	`}`
`215`	`219`	`};`
`216`	`220`