From 72eb4ba2f59eb8f7579d263d8e32a0866d21aabd Mon Sep 17 00:00:00 2001
From: "Meng, Hengyu" <hengyu.meng@intel.com>
Date: Fri, 22 Mar 2024 10:13:29 +0800
Subject: [PATCH] add offload_op in sycl

---
 ggml-sycl.cpp | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/ggml-sycl.cpp b/ggml-sycl.cpp
index cc9ee0762..d6737c0e3 100644
--- a/ggml-sycl.cpp
+++ b/ggml-sycl.cpp
@@ -17392,7 +17392,7 @@ static ggml_backend_i ggml_backend_sycl_interface = {
     /* .graph_plan_compute      = */ NULL,
     /* .graph_compute           = */ ggml_backend_sycl_graph_compute,
     /* .supports_op             = */ ggml_backend_sycl_supports_op,
-    /* .offload_op              = */ NULL,
+    /* .offload_op              = */ ggml_backend_sycl_offload_op,
     /* .event_new               = */ NULL,
     /* .event_free              = */ NULL,
     /* .event_record            = */ NULL,
@@ -17400,6 +17400,12 @@ static ggml_backend_i ggml_backend_sycl_interface = {
     /* .event_synchronize       = */ NULL,
 };
 
+GGML_CALL static bool ggml_backend_sycl_offload_op(ggml_backend_t backend, const ggml_tensor * op) {
+    const int min_batch_size = 32;
+    return op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS;
+    GGML_UNUSED(backend);
+}
+
 static ggml_guid_t ggml_backend_sycl_guid() {
     static ggml_guid guid = { 0x58, 0x05, 0x13, 0x8f, 0xcd, 0x3a, 0x61, 0x9d, 0xe7, 0xcd, 0x98, 0xa9, 0x03, 0xfd, 0x7c, 0x53 };
     return &guid;