ggml-alloc: avoid return silently

In certain cases, the allocate_node() function may silently return
without performing any memory allocation.

Author: lshzh-ww

ggml-alloc.c

@@ -441,8 +441,8 @@ static void allocate_node(struct ggml_allocr * alloc, struct ggml_tensor * node)
                         else {
                             AT_PRINTF("reusing parent %s for %s\n", parent->name, node->name);
                             node->data = parent->data;
+                            return;
                         }
-                        return;
                     }
                 }
             }
@@ -528,7 +528,7 @@ static size_t ggml_allocator_alloc_graph_tensors_n(
                 }
                 AT_PRINTF("\n");
             }
-            
+
 
             // update parents
             // update immediately if there is no parse_seq

llama.cpp

@@ -2704,11 +2704,6 @@ static struct ggml_cgraph * llm_build_falcon(
             struct ggml_tensor * inpFF = attn_norm;
 
             cur = ggml_mul_mat(ctx0, model.layers[il].w3, inpFF);
-
-            // TODO: this is temporary needed to introduce artificial dependency between FF and ATTN
-            //       adding this, because there seems to be a bug in the Metal concurrency optimization
-            //       without this line, the results are non-deterministic and wrong
-            cur->src[2] = attn_out;
             offload_func(cur);
 
             cur = ggml_gelu(ctx0, cur);