release tensorlist outside aclnn_concat

Author: hipudding

ggml-cann/aclnn_ops.cpp

@@ -137,13 +137,9 @@ void ggml_cann_leaky_relu(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
     ACL_CHECK(aclDestroyTensor(acl_dst));
 }
 
-void aclnn_concat(ggml_backend_cann_context& ctx, aclTensor *acl_src0,
-                  aclTensor *acl_src1, aclTensor *acl_dst, int64_t concat_dim,
+void aclnn_concat(ggml_backend_cann_context& ctx, aclTensorList* tensorList,
+                  aclTensor* acl_dst, int64_t concat_dim,
                   ggml_tensor* bind_tensor) {
-
-    aclTensor* tensors[] = {acl_src0, acl_src1};
-    aclTensorList* tensorList = aclCreateTensorList(tensors, 2);
-
     uint64_t workspaceSize = 0;
     aclOpExecutor* executor;
     void* workspaceAddr = nullptr;
@@ -157,12 +153,6 @@ void aclnn_concat(ggml_backend_cann_context& ctx, aclTensor *acl_src0,
 
     aclrtStream main_stream = ctx.stream();
     ACL_CHECK(aclnnCat(workspaceAddr, workspaceSize, executor, main_stream));
-
-    //ACL_CHECK(aclDestroyTensor(acl_src0));
-    //ACL_CHECK(aclDestroyTensor(acl_src1));
-    ACL_CHECK(aclDestroyTensorList(tensorList));
-    ACL_CHECK(aclDestroyTensor(acl_dst));
-
 }
 
 void ggml_cann_concat(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
@@ -173,14 +163,11 @@ void ggml_cann_concat(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
     aclTensor* acl_dst = create_acl_tensor(dst);
 
     int64_t concat_dim = 1;
+    aclTensor* tensors[] = {acl_src0, acl_src1};
+    aclTensorList* tensorList = aclCreateTensorList(tensors, 2);
+    aclnn_concat(ctx, tensorList, acl_dst, concat_dim, dst);
 
-    aclnn_concat(ctx, acl_src0, acl_src1, acl_dst, concat_dim, dst);
-
-    // release acl_src0, acl_src1 in aclnn_concat
-    // ACL_CHECK(aclDestroyTensor(acl_src0));
-    // ACL_CHECK(aclDestroyTensor(acl_src1));
-    // ->
-    // ACL_CHECK(aclDestroyTensorList(tensorList));
+    ACL_CHECK(aclDestroyTensorList(tensorList));
     ACL_CHECK(aclDestroyTensor(acl_dst));
 }
 
@@ -1331,9 +1318,12 @@ void ggml_cann_timestep_embedding(ggml_backend_cann_context& ctx, ggml_tensor* d
     // concat
     int64_t concat_dim = 3;
     aclTensor* acl_dst = create_acl_tensor(dst);
-    aclnn_concat(ctx, tmp_cos_tensor, tmp_sin_tensor, acl_dst, concat_dim, dst);
+    aclTensor* tensors[] = {tmp_cos_tensor, tmp_sin_tensor};
+    aclTensorList* tensorList = aclCreateTensorList(tensors, 2);
+    aclnn_concat(ctx, tensorList, acl_dst, concat_dim, dst);
 
     // release
+    ACL_CHECK(aclDestroyTensorList(tensorList));
     ACL_CHECK(aclDestroyTensor(acl_src));
     ACL_CHECK(aclDestroyTensor(tmp_arange_tensor));
     ACL_CHECK(aclDestroyTensor(tmp_permute_tenosr));