cann: fix multi-npu exec error

ggerganov · Jul 27, 2024 · 937a12c · 937a12c
1 parent 01245f5
commit 937a12c
Showing 1 changed file with 15 additions and 15 deletions.
diff --git a/ggml/src/ggml-cann.cpp b/ggml/src/ggml-cann.cpp
@@ -1559,23 +1559,18 @@ GGML_CALL static bool ggml_backend_cann_cpy_tensor_async(
             return false;
         }
 
+        // need open both directions for memcpyasync between devices.
+        ggml_cann_set_device(cann_ctx_dst->device);
+        ACL_CHECK(aclrtDeviceEnablePeerAccess(cann_ctx_src->device, 0));
         ggml_cann_set_device(cann_ctx_src->device);
         ACL_CHECK(aclrtDeviceEnablePeerAccess(cann_ctx_dst->device, 0));
+
         ACL_CHECK(aclrtMemcpyAsync(dst->data, copy_size, src->data, copy_size,
                                    ACL_MEMCPY_DEVICE_TO_DEVICE,
-                                   cann_ctx_dst->stream()));
-
-        // record event on src stream
-        if (!cann_ctx_src->copy_event) {
-            ACL_CHECK(aclrtCreateEvent(&cann_ctx_src->copy_event));
-        }
-
-        ACL_CHECK(
-            aclrtRecordEvent(cann_ctx_src->copy_event, cann_ctx_src->stream()));
+                                   cann_ctx_src->stream()));
 
-        // wait on dst stream for the copy to complete
-        ACL_CHECK(aclrtStreamWaitEvent(cann_ctx_dst->stream(),
-                                       cann_ctx_src->copy_event));
+        //TODO: workaround for Event didn`t work here.
+        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,
@@ -1773,9 +1768,14 @@ static bool ggml_backend_buft_is_cann(ggml_backend_buffer_type_t buft) {
  */
 GGML_CALL static bool ggml_backend_cann_supports_buft(
     ggml_backend_t backend, ggml_backend_buffer_type_t buft) {
-    return buft->iface.get_name == ggml_backend_cann_buffer_type_name;
-
-    GGML_UNUSED(backend);
+    if (ggml_backend_buft_is_cann(buft)) {
+        ggml_backend_cann_context * cann_ctx =
+                        (ggml_backend_cann_context *)backend->context;
+        ggml_backend_cann_buffer_type_context * buft_ctx =
+                        (ggml_backend_cann_buffer_type_context *)buft->context;
+        return buft_ctx->device == cann_ctx->device;
+    }
+    return false;
 }
 
 /**