update some changes

Author: bachelor-dou

ggml/src/ggml-cann/aclnn_ops.cpp

@@ -1209,11 +1209,6 @@ void aclnn_sin(ggml_backend_cann_context& ctx, aclTensor* acl_src,
     GGML_CANN_CALL_ACLNN_OP(Sin, acl_src, acl_dst);
 }
 
-void aclnn_geluv2(ggml_backend_cann_context& ctx, aclTensor* acl_src,
-                       aclTensor* acl_dst) {
-        GGML_CANN_CALL_ACLNN_OP(GeluV2, acl_src, 0, acl_dst);
-}
-
 void ggml_cann_timestep_embedding(ggml_backend_cann_context& ctx,
                                   ggml_tensor* dst) {
     const ggml_tensor* src = dst->src[0];

ggml/src/ggml-cann/aclnn_ops.h

@@ -593,9 +593,6 @@ void aclnn_cos(ggml_backend_cann_context& ctx, aclTensor* acl_src,
 void aclnn_sin(ggml_backend_cann_context& ctx, aclTensor* acl_src,
     aclTensor* acl_dst);
 
-void aclnn_geluv2(ggml_backend_cann_context& ctx, aclTensor* acl_src,
-    aclTensor* acl_dst);
-
 /**
  * @brief Prepares broadcast-compatible ACL tensors for two input tensors and one
  * output tensor.
@@ -843,13 +840,14 @@ void ggml_cann_unary_op(
  * @see ggml_cann_unary_op
  * @see GGML_CANN_CALL_ACLNN_OP
  */
-#define GGML_CANN_CALL_UNARY_OP(OP_NAME)                      \
-    do {                                                      \
-        ggml_tensor * src     = dst->src[0];                  \
-        aclTensor *   acl_src = ggml_cann_create_tensor(src); \
-        aclTensor *   acl_dst = ggml_cann_create_tensor(dst); \
-        GGML_CANN_CALL_ACLNN_OP(OP_NAME, acl_src, acl_dst);   \
-        ACL_CHECK(aclDestroyTensor(acl_src));                 \
-        ACL_CHECK(aclDestroyTensor(acl_dst));                 \
-    } while (0)
+#define GGML_CANN_CALL_UNARY_OP(OP_NAME)                         \
+    do {                                                         \
+        auto lambda = [](ggml_backend_cann_context& ctx,         \
+            aclTensor* acl_src,                                  \
+            aclTensor* acl_dst) {                                \
+            GGML_CANN_CALL_ACLNN_OP(OP_NAME, acl_src, acl_dst);  \
+        };                                                       \
+        ggml_cann_unary_op(lambda, ctx, dst);                    \
+    }                                                            \
+    while (0)
 #endif  // CANN_ACLNN_OPS

ggml/src/ggml-cann/ggml-cann.cpp

@@ -1602,7 +1602,12 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
                     GGML_CANN_CALL_UNARY_OP(Silu);
                     break;
                 case GGML_UNARY_OP_GELU_QUICK: {
-                    ggml_cann_unary_op<aclnn_geluv2>(ctx, dst);
+                    auto lambda = [](ggml_backend_cann_context& ctx,
+                        aclTensor* acl_src,
+                        aclTensor* acl_dst) {
+                        GGML_CANN_CALL_ACLNN_OP(GeluV2, acl_src, 0, acl_dst);
+                    };
+                    ggml_cann_unary_op(lambda, ctx, dst);
                 } break;
                 case GGML_UNARY_OP_TANH:
                     GGML_CANN_CALL_UNARY_OP(Tanh);