From: Chenguang Li <redacted>
Date: Mon, 8 Sep 2025 02:03:29 +0000 (+0800)
Subject: CANN: Stream sync between devices for acl_graph (#15809)
X-Git-Tag: upstream/0.0.6527~119
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=85ca66a74676e6d5df4433016488e039a4b464ae;p=pkg%2Fggml%2Fsources%2Fllama.cpp

CANN: Stream sync between devices for acl_graph (#15809)

* CANN: Switch to stream synchronization

Switch to stream synchronization because events are not effective.

Co-authored-by: hipudding <redacted>

* CANN: add Comments

---------

Co-authored-by: hipudding <redacted>
---

diff --git a/ggml/src/ggml-cann/ggml-cann.cpp b/ggml/src/ggml-cann/ggml-cann.cpp
index 756ad8df..2f9f373f 100755
--- a/ggml/src/ggml-cann/ggml-cann.cpp
+++ b/ggml/src/ggml-cann/ggml-cann.cpp
@@ -2092,16 +2092,17 @@ static bool ggml_backend_cann_cpy_tensor_async(
         ACL_CHECK(aclrtMemcpyAsync(dst->data, copy_size, src->data, copy_size,
                                    ACL_MEMCPY_DEVICE_TO_DEVICE,
                                    cann_ctx_src->stream()));
-
         // record event on src stream after the copy
-        if (!cann_ctx_src->copy_event) {
-            ACL_CHECK(aclrtCreateEventWithFlag(&cann_ctx_src->copy_event, ACL_EVENT_SYNC));
-        }
-        ACL_CHECK(aclrtRecordEvent(cann_ctx_src->copy_event, cann_ctx_src->stream()));
-
-        // wait on dst stream for the copy to complete
-        ggml_cann_set_device(cann_ctx_dst->device);
-        ACL_CHECK(aclrtStreamWaitEvent(cann_ctx_dst->stream(), cann_ctx_src->copy_event));
+        // TODO: this event is not effective with acl graph mode, change to use aclrtSynchronizeStream
+        // if (!cann_ctx_src->copy_event) {
+        //     ACL_CHECK(aclrtCreateEventWithFlag(&cann_ctx_src->copy_event, ACL_EVENT_SYNC));
+        // }
+        // ACL_CHECK(aclrtRecordEvent(cann_ctx_src->copy_event, cann_ctx_src->stream()));
+
+        // // wait on dst stream for the copy to complete
+        // ggml_cann_set_device(cann_ctx_dst->device);
+        // ACL_CHECK(aclrtStreamWaitEvent(cann_ctx_dst->stream(), cann_ctx_src->copy_event));
+        ACL_CHECK(aclrtSynchronizeStream(cann_ctx_src->stream()));
     } else {
         // src and dst are on the same backend
         ACL_CHECK(aclrtMemcpyAsync(dst->data, copy_size, src->data, copy_size,