feat(lite): add get model infomation before create network interface

GitOrigin-RevId: e499f3ebf8

lite/include/lite/network.h

@@ -373,6 +373,14 @@ public:
     //! dump network after global layout transform optimization
     static void dump_layout_transform_model(
             std::shared_ptr<Network> network, std::string optimized_model_path);
+
+    //! get the model io information before model loaded by model path.
+    static NetworkIO get_model_io_info(
+            const std::string& model_path, const Config& config = {});
+
+    //! get the model io information before model loaded by model memory.
+    static NetworkIO get_model_io_info(
+            const void* model_mem, size_t size, const Config& config = {});
 };
 
 }  // namespace lite

lite/lite-c/include/lite-c/network_c.h

@@ -588,6 +588,28 @@ LITE_API int LITE_enable_global_layout_transform(LiteNetwork network);
 LITE_API int LITE_dump_layout_transform_model(
         LiteNetwork network, const char* dump_file_path);
 
+/**! get the model io information before model loaded by model path.
+ * \param[in] model_path The model file path
+ * \param[in] config The model config for loading
+ * \param[out] ios The model io infermation
+ * \return int if the return is not zero, error happened, the error message
+ * can get by LITE_get_last_error
+ */
+LITE_API int LITE_get_model_io_info_by_path(
+        const char* model_path, const LiteConfig config, LiteNetworkIO* ios);
+
+/** get the model io information before model loaded by model memory.
+ * \param[in] model_mem The model memory ptr
+ * \param[in] size The model memory ptr length
+ * \param[in] config The model config for loading
+ * \param[out] ios The model io infermation
+ * \return int if the return is not zero, error happened, the error message
+ * can get by LITE_get_last_error
+ */
+LITE_API int LITE_get_model_io_info_by_memory(
+        const void* model_mem, size_t size, const LiteConfig config,
+        LiteNetworkIO* ios);
+
 #ifdef __cplusplus
 }
 #endif

lite/lite-c/src/network.cpp

@@ -167,6 +167,31 @@ lite::NetworkIO convert_to_lite_io(const LiteNetworkIO c_network_io) {
     return network_io;
 }
 
+struct InnerIO {
+    std::vector<std::string> names;
+    std::vector<LiteIO> inputs;
+    std::vector<LiteIO> outputs;
+};
+
+InnerIO convert_to_inner_io(const lite::NetworkIO& network_io) {
+    InnerIO innner_io;
+    for (size_t i = 0; i < network_io.inputs.size(); i++) {
+        lite::IO io = network_io.inputs[i];
+        innner_io.names.push_back(io.name);
+        innner_io.inputs.push_back(
+                {innner_io.names.back().c_str(), io.is_host, io.io_type,
+                 convert_to_clayout(io.config_layout)});
+    }
+    for (size_t i = 0; i < network_io.outputs.size(); i++) {
+        lite::IO io = network_io.outputs[i];
+        innner_io.names.push_back(io.name);
+        innner_io.outputs.push_back(
+                {innner_io.names.back().c_str(), io.is_host, io.io_type,
+                 convert_to_clayout(io.config_layout)});
+    }
+    return innner_io;
+}
+
 int LITE_make_default_network(LiteNetwork* network) {
     LITE_CAPI_BEGIN();
     LITE_ASSERT(network, "The network pass to LITE api is null");
@@ -665,4 +690,59 @@ int LITE_dump_layout_transform_model(LiteNetwork network, const char* dump_file_
     lite::Runtime::dump_layout_transform_model(network_shared, dump_file_path);
     LITE_CAPI_END();
 }
+
+namespace {
+static LITE_MUTEX mtx_io;
+static std::unordered_map<const void*, InnerIO>& get_global_io_holder() {
+    static std::unordered_map<const void*, InnerIO> global_holder;
+    return global_holder;
+}
+
+int write_ios_from_cpp_io(
+        const lite::NetworkIO& cpp_io, LiteNetworkIO* ios, const void* key) {
+    LITE_CAPI_BEGIN();
+    LITE_LOCK_GUARD(mtx_io);
+    get_global_io_holder()[key] = convert_to_inner_io(cpp_io);
+    auto&& inner_io = get_global_io_holder()[key];
+    ios->input_size = inner_io.inputs.size();
+    ios->output_size = inner_io.outputs.size();
+    ios->inputs = inner_io.inputs.data();
+    ios->outputs = inner_io.outputs.data();
+    size_t i = 0;
+    for (; i < ios->input_size; i++) {
+        auto io_ptr = ios->inputs + i;
+        io_ptr->name = inner_io.names[i].c_str();
+    }
+    for (; i < ios->output_size; i++) {
+        auto io_ptr = ios->outputs + i;
+        io_ptr->name = inner_io.names[i].c_str();
+    }
+    LITE_CAPI_END();
+}
+
+}  // namespace
+
+int LITE_get_model_io_info_by_path(
+        const char* model_path, const LiteConfig config, LiteNetworkIO* ios) {
+    LITE_CAPI_BEGIN();
+    LITE_ASSERT(model_path, "The model_path pass to LITE api is null");
+    auto&& cpp_ios = lite::Runtime::get_model_io_info(
+            std::string{model_path}, convert_to_lite_config(config));
+    return write_ios_from_cpp_io(
+            cpp_ios, ios, reinterpret_cast<const void*>(model_path));
+    LITE_CAPI_END();
+}
+
+int LITE_get_model_io_info_by_memory(
+        const void* model_mem, size_t size, const LiteConfig config,
+        LiteNetworkIO* ios) {
+    LITE_CAPI_BEGIN();
+    LITE_ASSERT(model_mem, "The model_mem pass to LITE api is null");
+    auto&& cpp_ios = lite::Runtime::get_model_io_info(
+            model_mem, size, convert_to_lite_config(config));
+    return write_ios_from_cpp_io(
+            cpp_ios, ios, reinterpret_cast<const void*>(model_mem));
+    LITE_CAPI_END();
+}
+
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}

lite/pylite/megenginelite/network.py

@@ -364,6 +364,14 @@ class _NetworkAPI(_LiteCObjBase):
         ("LITE_get_static_memory_alloc_info", [_Cnetwork, c_char_p]),
         ("LITE_enable_global_layout_transform", [_Cnetwork]),
         ("LITE_dump_layout_transform_model", [_Cnetwork, c_char_p]),
+        (
+            "LITE_get_model_io_info_by_path",
+            [c_char_p, LiteConfig, POINTER(_LiteNetworkIO)],
+        ),
+        (
+            "LITE_get_model_io_info_by_memory",
+            [c_char_p, c_size_t, LiteConfig, POINTER(_LiteNetworkIO)],
+        ),
     ]
 
 
@@ -619,3 +627,27 @@ class LiteNetwork(object):
     def dump_layout_transform_model(self, model_file):
         c_file = model_file.encode("utf-8")
         self._api.LITE_dump_layout_transform_model(self._network, c_file)
+
+
+def get_model_io_info(model_path, config=None):
+    """
+    get the model IO information before create the NetWork, this IO
+    information can be used to configuration the NetWork.
+    """
+    api = _NetworkAPI()._lib
+    c_path = c_char_p(model_path.encode("utf-8"))
+
+    ios = _LiteNetworkIO()
+
+    if config is not None:
+        api.LITE_get_model_io_info_by_path(c_path, config, byref(ios))
+    else:
+        config = LiteConfig()
+        api.LITE_get_model_io_info_by_path(c_path, config, byref(ios))
+
+    ret_ios = LiteNetworkIO()
+    for i in range(ios.input_size):
+        ret_ios.add_input(ios.inputs[i])
+    for i in range(ios.output_size):
+        ret_ios.add_output(ios.outputs[i])
+    return ret_ios

lite/pylite/test/test_utils.py

@@ -8,6 +8,7 @@
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 
 import functools
+import os
 
 import numpy as np
 
@@ -200,3 +201,20 @@ def test_tensor_collect_batch_device_numpy():
     for i in range(4):
         for j in range(48):
             assert data[i][j // 8][j % 8] == i + 1
+
+
+def test_get_model_io_ahead():
+    source_dir = os.getenv("LITE_TEST_RESOURCE")
+    model_path = os.path.join(source_dir, "shufflenet.mge")
+    ios = get_model_io_info(model_path)
+
+    assert len(ios.inputs) == 1
+    assert ios.inputs[0].name == "data"
+    assert ios.inputs[0].config_layout.shapes[1] == 3
+    assert ios.inputs[0].config_layout.shapes[2] == 224
+    assert ios.inputs[0].config_layout.shapes[3] == 224
+
+    assert len(ios.outputs) == 1
+    assert ios.outputs[0].name == "TRUE_DIV(EXP[12065],reduce0[12067])[12077]"
+    assert ios.outputs[0].config_layout.shapes[0] == 1
+    assert ios.outputs[0].config_layout.shapes[1] == 1000

lite/src/function_base.h

@@ -34,7 +34,7 @@ ADD_STATEMENT(NetworkImplDft, Dft);
 }  // namespace
 
 // if it can't find the function, ignore
-template <typename tensor_type, typename ret_type, typename... Args>
+template <typename type, typename ret_type, typename... Args>
 ret_type try_call_func(std::string func_name, Args... args) {
     mark_used_variable(func_name);
     mark_used_variable(args...);
@@ -42,10 +42,10 @@ ret_type try_call_func(std::string func_name, Args... args) {
 }
 
 // if it can't find the function, throw error
-template <typename tensor_type, typename ret_type, typename... Args>
+template <typename type, typename ret_type, typename... Args>
 ret_type call_func(std::string func_name, Args... args) {
     mark_used_variable(args...);
-    auto backend_name = class_type_name<tensor_type>()();
+    auto backend_name = class_type_name<type>()();
     auto msg_info = func_name + "  is not aviliable in " + backend_name + " backend.";
     LITE_THROW(msg_info.c_str());
 }

lite/src/mge/function_dft.h

@@ -206,6 +206,26 @@ inline void call_func<NetworkImplDft, void>(
         THROW_FUNC_ERROR(func_name);
     }
 }
+
+template <>
+inline NetworkIO call_func<NetworkImplDft, NetworkIO>(
+        std::string func_name, std::string model_path, Config config) {
+    if (func_name == "get_model_io_info") {
+        return get_model_io_info_dft(model_path, config);
+    } else {
+        THROW_FUNC_ERROR(func_name);
+    }
+}
+
+template <>
+inline NetworkIO call_func<NetworkImplDft, NetworkIO>(
+        std::string func_name, const void* model_mem, size_t size, Config config) {
+    if (func_name == "get_model_io_info") {
+        return get_model_io_info_dft(model_mem, size, config);
+    } else {
+        THROW_FUNC_ERROR(func_name);
+    }
+}
 #undef THROW_FUNC_ERROR
 
 }  // namespace lite

lite/src/mge/network_impl.cpp

@@ -929,5 +929,75 @@ void NetworkImplDft::dump_layout_transform_model(std::string optimized_model_pat
                          "enable_global_layout_transform before"));
     }
 }
+
+NetworkIO lite::get_model_io_info_dft(
+        const std::string& model_path, const Config& config) {
+    FILE* fin = fopen(model_path.c_str(), "rb");
+    LITE_ASSERT(fin, "failed to open %s: %s", model_path.c_str(), strerror(errno));
+    fseek(fin, 0, SEEK_END);
+    size_t size = ftell(fin);
+    fseek(fin, 0, SEEK_SET);
+    void* ptr = malloc(size);
+    std::shared_ptr<void> buf{ptr, ::free};
+    auto nr = fread(buf.get(), 1, size, fin);
+    LITE_ASSERT(nr == size);
+    fclose(fin);
+    return get_model_io_info_dft(ptr, size, config);
+}
+
+NetworkIO lite::get_model_io_info_dft(
+        const void* model_mem, size_t size, const Config& config) {
+    std::shared_ptr<void> model{const_cast<void*>(model_mem), [](void*) {}};
+    auto input_file = mgb::serialization::InputFile::make_mem_proxy(model, size, false);
+    auto format =
+            mgb::serialization::GraphLoader::identify_graph_dump_format(*input_file);
+    if (!format.valid()) {
+        LITE_THROW("invalid model format");
+    }
+    auto loader =
+            mgb::serialization::GraphLoader::make(std::move(input_file), format.val());
+
+    mgb::serialization::GraphLoadConfig load_config;
+    load_config.comp_graph = mgb::ComputingGraph::make();
+    if (config.has_compression) {
+        load_config.tensor_value_loader = decompressed_tensor_value_loader;
+    }
+    auto compnode_locator = to_compnode_locator(config.device_type);
+    load_config.comp_node_mapper = [=](mgb::CompNode::Locator& loc) {
+        if (loc.type == mgb::CompNode::DeviceType::UNSPEC) {
+            loc.type = compnode_locator.type;
+        }
+        loc.device = compnode_locator.device;
+    };
+    auto load_result = loader->load(load_config, true);
+    NetworkIO IOs;
+    for (auto&& in_tensor_iter : load_result.tensor_map) {
+        IO in_io;
+        in_io.name = in_tensor_iter.first;
+        in_io.config_layout = to_lite_layout(in_tensor_iter.second->layout());
+        IOs.inputs.push_back(in_io);
+    }
+    auto infer_shape = [=](mgb::cg::SymbolVar var) -> const megdnn::TensorShape* {
+        auto&& static_infer_mgr = load_config.comp_graph->static_infer_manager();
+        using InferType = mgb::cg::static_infer::InferType;
+        if (static_infer_mgr.get_infer_type(var.node()).shape &
+            (InferType::CONST | InferType::RT_STATIC)) {
+            return static_infer_mgr.infer_shape_fallible(var.node());
+        } else {
+            return nullptr;
+        }
+    };
+    for (auto&& out : load_result.output_var_list) {
+        IO out_io;
+        out_io.name = out.node()->name();
+        if (auto shape = infer_shape(out)) {
+            out_io.config_layout = to_lite_layout(TensorLayout{*shape, out.dtype()});
+        } else {
+            out_io.config_layout = to_lite_layout(TensorLayout{{}, out.dtype()});
+        }
+        IOs.outputs.push_back(out_io);
+    }
+    return IOs;
+}
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}

lite/src/mge/network_impl.h

@@ -262,6 +262,13 @@ private:
 #endif
     std::unique_ptr<mgb::OprIODumpBase> m_iodump;
 };
+//! get the model information before model loaded by Network
+NetworkIO get_model_io_info_dft(const std::string& model_path, const Config& config);
+
+//! get the model information before model loaded by Network by model memory and
+//! size
+NetworkIO get_model_io_info_dft(
+        const void* model_mem, size_t size, const Config& config);
 
 }  // namespace lite
 

lite/src/network.cpp

@@ -534,4 +534,26 @@ void Runtime::dump_layout_transform_model(
     LITE_THROW("dump_layout_transform_model is not aviliable in the backend.");
     LITE_ERROR_HANDLER_END
 }
+
+NetworkIO Runtime::get_model_io_info(
+        const std::string& model_path, const Config& config) {
+    LITE_ERROR_HANDLER_BEGIN
+    if (config.backend == LiteBackend::LITE_DEFAULT) {
+        return call_func<NetworkImplDft, NetworkIO>(
+                "get_model_io_info", model_path, config);
+    }
+    LITE_THROW("get_model_io_info is not aviliable in the backend.");
+    LITE_ERROR_HANDLER_END
+}
+
+NetworkIO Runtime::get_model_io_info(
+        const void* model_mem, size_t size, const Config& config) {
+    LITE_ERROR_HANDLER_BEGIN
+    if (config.backend == LiteBackend::LITE_DEFAULT) {
+        return call_func<NetworkImplDft, NetworkIO>(
+                "get_model_io_info", model_mem, size, config);
+    }
+    LITE_THROW("get_model_io_info is not aviliable in the backend.");
+    LITE_ERROR_HANDLER_END
+}
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}

lite/test/test_network.cpp

@@ -106,6 +106,54 @@ TEST(TestNetWork, GetAllName) {
     ASSERT_TRUE(output_names[0] == "TRUE_DIV(EXP[12065],reduce0[12067])[12077]");
 }
 
+TEST(TestNetWork, GetAllIoInfoAhead) {
+    Config config;
+    std::string model_path = "./shufflenet.mge";
+
+    auto ios = Runtime::get_model_io_info(model_path);
+
+    FILE* fin = fopen(model_path.c_str(), "rb");
+    ASSERT_TRUE(fin);
+    fseek(fin, 0, SEEK_END);
+    size_t size = ftell(fin);
+    fseek(fin, 0, SEEK_SET);
+    void* ptr = malloc(size);
+    std::shared_ptr<void> buf{ptr, ::free};
+    auto nr = fread(buf.get(), 1, size, fin);
+    LITE_ASSERT(nr == size);
+    fclose(fin);
+
+    auto ios_mem = Runtime::get_model_io_info(ptr, size);
+
+    ASSERT_EQ(ios.inputs.size(), ios_mem.inputs.size());
+    ASSERT_EQ(ios.inputs.size(), 1);
+
+    ASSERT_EQ(ios.outputs.size(), ios_mem.outputs.size());
+    ASSERT_EQ(ios.outputs.size(), 1);
+
+    ASSERT_TRUE(ios.inputs[0].name == "data");
+    ASSERT_TRUE(ios.outputs[0].name == "TRUE_DIV(EXP[12065],reduce0[12067])[12077]");
+
+    ASSERT_TRUE(ios_mem.inputs[0].name == "data");
+    ASSERT_TRUE(
+            ios_mem.outputs[0].name == "TRUE_DIV(EXP[12065],reduce0[12067])[12077]");
+    ASSERT_EQ(ios.inputs[0].config_layout.ndim, 4);
+    ASSERT_EQ(ios.inputs[0].config_layout.shapes[1], 3);
+    ASSERT_EQ(ios.inputs[0].config_layout.shapes[2], 224);
+
+    ASSERT_EQ(ios.outputs[0].config_layout.ndim, 2);
+    ASSERT_EQ(ios.outputs[0].config_layout.shapes[0], 1);
+    ASSERT_EQ(ios.outputs[0].config_layout.shapes[1], 1000);
+
+    ASSERT_EQ(ios_mem.inputs[0].config_layout.ndim, 4);
+    ASSERT_EQ(ios_mem.inputs[0].config_layout.shapes[1], 3);
+    ASSERT_EQ(ios_mem.inputs[0].config_layout.shapes[2], 224);
+
+    ASSERT_EQ(ios_mem.outputs[0].config_layout.ndim, 2);
+    ASSERT_EQ(ios_mem.outputs[0].config_layout.shapes[0], 1);
+    ASSERT_EQ(ios_mem.outputs[0].config_layout.shapes[1], 1000);
+}
+
 TEST(TestNetWork, LoadFBSModel) {
     Config config;
     std::string model_path = "./ax.mge";

lite/test/test_network_c.cpp

@@ -252,6 +252,55 @@ TEST(TestCapiNetWork, GetAllName) {
     LITE_destroy_network(c_network);
 }
 
+TEST(TestCapiNetWork, GetAllNameAhead) {
+    std::string model_path = "./shufflenet.mge";
+    LiteNetworkIO ios, ios_mem;
+    LITE_CAPI_CHECK(LITE_get_model_io_info_by_path(
+            model_path.c_str(), *default_config(), &ios));
+    FILE* fin = fopen(model_path.c_str(), "rb");
+    ASSERT_TRUE(fin);
+    fseek(fin, 0, SEEK_END);
+    size_t size = ftell(fin);
+    fseek(fin, 0, SEEK_SET);
+    void* ptr = malloc(size);
+    std::shared_ptr<void> buf{ptr, ::free};
+    auto nr = fread(buf.get(), 1, size, fin);
+    LITE_ASSERT(nr == size);
+    fclose(fin);
+
+    LITE_CAPI_CHECK(
+            LITE_get_model_io_info_by_memory(ptr, size, *default_config(), &ios_mem));
+
+    ASSERT_EQ(ios.input_size, 1);
+    ASSERT_EQ(ios.output_size, 1);
+    ASSERT_EQ(ios_mem.input_size, 1);
+    ASSERT_EQ(ios_mem.output_size, 1);
+
+    ASSERT_TRUE(std::string(ios.inputs->name) == "data");
+    ASSERT_TRUE(ios.inputs->config_layout.ndim == 4);
+    ASSERT_TRUE(ios.inputs->config_layout.shapes[1] == 3);
+    ASSERT_TRUE(ios.inputs->config_layout.shapes[2] == 224);
+    ASSERT_TRUE(ios.inputs->config_layout.shapes[3] == 224);
+    ASSERT_TRUE(
+            std::string(ios.outputs->name) ==
+            "TRUE_DIV(EXP[12065],reduce0[12067])[12077]");
+    ASSERT_TRUE(ios.outputs->config_layout.ndim == 2);
+    ASSERT_TRUE(ios.outputs->config_layout.shapes[0] == 1);
+    ASSERT_TRUE(ios.outputs->config_layout.shapes[1] == 1000);
+
+    ASSERT_TRUE(std::string(ios_mem.inputs->name) == "data");
+    ASSERT_TRUE(ios_mem.inputs->config_layout.ndim == 4);
+    ASSERT_TRUE(ios_mem.inputs->config_layout.shapes[1] == 3);
+    ASSERT_TRUE(ios_mem.inputs->config_layout.shapes[2] == 224);
+    ASSERT_TRUE(ios_mem.inputs->config_layout.shapes[3] == 224);
+    ASSERT_TRUE(
+            std::string(ios_mem.outputs->name) ==
+            "TRUE_DIV(EXP[12065],reduce0[12067])[12077]");
+    ASSERT_TRUE(ios_mem.outputs->config_layout.ndim == 2);
+    ASSERT_TRUE(ios_mem.outputs->config_layout.shapes[0] == 1);
+    ASSERT_TRUE(ios_mem.outputs->config_layout.shapes[1] == 1000);
+}
+
 #if LITE_BUILD_WITH_RKNPU
 
 static int GetTop(