fix: make the immediate LoRA apply mode work better when using Vulkan (leejet#1021)

Author: leejet

ggml_extend.hpp

@@ -1141,14 +1141,23 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_ones(struct ggml_context* ctx,
 }
 
 __STATIC_INLINE__ ggml_tensor* ggml_ext_cast_f32(ggml_context* ctx, ggml_tensor* a) {
+#ifdef SD_USE_VULKAN
+    auto zero_index = ggml_get_tensor(ctx, "ggml_runner_build_in_tensor:zero_int");
+    auto out        = ggml_reshape_1d(ctx, a, ggml_nelements(a));
+    out             = ggml_get_rows(ctx, out, zero_index);
+    out             = ggml_reshape(ctx, out, a);
+    // auto out = ggml_cast(ctx, a, GGML_TYPE_F32);
+    return out;
+#else
     auto out         = ggml_reshape_2d(ctx, a, 1, ggml_nelements(a));
     ggml_tensor* one = ggml_ext_ones(ctx, 1, 1, 1, 1);  // [1,]
     if (ggml_is_transposed(out)) {
         out = ggml_mul_mat(ctx, one, out);
     } else {
         out = ggml_mul_mat(ctx, out, one);
     }
-    out = ggml_reshape(ctx, out, a);
+    out                    = ggml_reshape(ctx, out, a);
+#endif
     return out;
 }
 
@@ -1556,6 +1565,9 @@ struct GGMLRunner {
     std::vector<float> one_vec = {1.f};
     ggml_tensor* one_tensor    = nullptr;
 
+    std::vector<int> zero_int_vec = {0};
+    ggml_tensor* zero_int_tensor  = nullptr;
+
     std::map<struct ggml_tensor*, const void*> backend_tensor_data_map;
     std::map<std::string, struct ggml_tensor*> cache_tensor_map;  // name -> tensor
     const std::string final_result_name = "ggml_runner_final_result_tensor";
@@ -1626,10 +1638,15 @@ struct GGMLRunner {
         one_tensor = ggml_new_tensor_1d(compute_ctx, GGML_TYPE_F32, 1);
         ggml_set_name(one_tensor, "ggml_runner_build_in_tensor:one");
         set_backend_tensor_data(one_tensor, one_vec.data());
+
+        zero_int_tensor = ggml_new_tensor_1d(compute_ctx, GGML_TYPE_I32, 1);
+        ggml_set_name(zero_int_tensor, "ggml_runner_build_in_tensor:zero_int");
+        set_backend_tensor_data(zero_int_tensor, zero_int_vec.data());
     }
 
     void prepare_build_in_tensor_after(struct ggml_cgraph* gf) {
         ggml_build_forward_expand(gf, one_tensor);
+        ggml_build_forward_expand(gf, zero_int_tensor);
     }
 
     struct ggml_cgraph* new_graph_custom(size_t graph_size) {