add ops

torch_mlu_ops-v1.3.2/csrc/kernels/preload.mluh

@@ -0,0 +1,34 @@
+/*************************************************************************
+ * Copyright (C) [2023-2024] by Cambricon, Inc.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *************************************************************************/
+#ifndef CSRC_KERNELS_PRELOAD_MLUH_
+#define CSRC_KERNELS_PRELOAD_MLUH_
+
+#include "cnnl.h"
+#include "kernel_utils.h"
+namespace tmo {
+/**
+ * @brief When tp is greater than 1, while executing reducesum, the weight of ffn
+ *        or selfattention to be calculated is loaded into LLC in advance.
+ * @param queue: The queue for mlu.
+ * @param filter_ptr: Input. Pointer to the MLU memory that stores the weight of ffn or
+ * selfattention.
+ * @param filter_size: The weight size of ffn or selfattention.
+ * @param preload_size: The size of the preload weight.
+ * @note The weights of ffn or selfattention must be continuous in filter_ptr.
+ */
+KernelStatus invokePreload(cnrtQueue_t queue,
+                           void *filter_ptr,
+                           size_t filter_size,
+                           size_t preload_size);
+}  // namespace tmo
+
+#endif  // CSRC_KERNELS_PRELOAD_MLUH_