Update flat tensor ndm to account for named delegate data

Currently flat_tensor ndm only accounts for tensors in get_data, get_num_keys, get_key functions.

Add support to return named_data values as well.

TODO: consolidate tensors and named_data into one structure in the flatbuffer. This will simplify all the serialization and runtime code.

Currently, we assume that a PTD file has either tensors or named_data, not both. After the consolidation, this won't be an issue.

Differential Revision: [D73380805](https://our.internmc.facebook.com/intern/diff/D73380805/)

[ghstack-poisoned]

Author: lucylq

extension/flat_tensor/flat_tensor_data_map.cpp

@@ -65,6 +65,25 @@ Result<const flat_tensor_flatbuffer::TensorMetadata*> get_flat_tensor_metadata(
   return Error::NotFound;
 }
 
+Result<const flat_tensor_flatbuffer::NamedData*> get_named_data(
+    const char* key,
+    const flatbuffers::Vector<
+        flatbuffers::Offset<flat_tensor_flatbuffer::NamedData>>* named_data) {
+  // Linear search by name.
+  for (int i = 0; i < named_data->size(); i++) {
+    if (std::strcmp(named_data->Get(i)->key()->c_str(), key) == 0) {
+      const auto* metadata = named_data->Get(i);
+      ET_CHECK_OR_RETURN_ERROR(
+          metadata->segment_index() >= 0,
+          InvalidExternalData,
+          "Invalid segment_index %d; malformed PTD file.",
+          metadata->segment_index());
+      return metadata;
+    }
+  }
+  return Error::NotFound;
+}
+
 Result<const TensorLayout> create_tensor_layout(
     const flat_tensor_flatbuffer::TensorMetadata* tensor_metadata) {
   ScalarType scalar_type =
@@ -109,6 +128,39 @@ ET_NODISCARD Result<const TensorLayout> FlatTensorDataMap::get_metadata(
 
 ET_NODISCARD Result<FreeableBuffer> FlatTensorDataMap::get_data(
     const char* key) const {
+  // TODO(lfq): consolidate named_data and tensors.
+  // Check named data.
+  Result<const flat_tensor_flatbuffer::NamedData*> named_data =
+      get_named_data(key, flat_tensor_->named_data());
+  if (named_data.ok()) {
+    size_t segment_index = named_data.get()->segment_index();
+    ET_CHECK_OR_RETURN_ERROR(
+        segment_index < flat_tensor_->segments()->size(),
+        InvalidExternalData,
+        "Invalid segment_index %zu; malformed PTD file.",
+        segment_index);
+
+    size_t segment_offset =
+        flat_tensor_->segments()->Get(segment_index)->offset();
+    size_t segment_size = flat_tensor_->segments()->Get(segment_index)->size();
+    ET_CHECK_OR_RETURN_ERROR(
+        segment_offset <
+            header_.segment_base_offset + header_.segment_data_size,
+        InvalidExternalData,
+        "Invalid segment offset %zu is larger than the segment_base_offset + segment_data_size %" PRIu64
+        "; malformed PTD file.",
+        segment_offset,
+        header_.segment_base_offset + header_.segment_data_size);
+    return loader_->load(
+        /*offset=*/header_.segment_base_offset + segment_offset,
+        segment_size,
+        DataLoader::SegmentInfo(DataLoader::SegmentInfo::Type::External));
+  }
+  if (named_data.error() != Error::NotFound) {
+    return named_data.error();
+  }
+
+  // Check tensors, if named data is not found.
   Result<const flat_tensor_flatbuffer::TensorMetadata*> metadata =
       get_flat_tensor_metadata(key, flat_tensor_->tensors());
   if (!metadata.ok()) {
@@ -179,16 +231,28 @@ ET_NODISCARD Error FlatTensorDataMap::load_data_into(
 }
 
 ET_NODISCARD Result<size_t> FlatTensorDataMap::get_num_keys() const {
-  return flat_tensor_->tensors()->size();
+  // TODO(lfq): consolidate named_data and tensors.
+  return flat_tensor_->tensors()->size() + flat_tensor_->named_data()->size();
 }
 
 ET_NODISCARD Result<const char*> FlatTensorDataMap::get_key(
     size_t index) const {
-  if (index < 0 || index >= flat_tensor_->tensors()->size()) {
+  // TODO(lfq): consolidate named_data and tensors.
+  // Currently, this assumes we either have tensors or named_data, but not both.
+  if (flat_tensor_->tensors()->size() > 0 && flat_tensor_->named_data()->size() > 0) {
+    return Error::NotImplemented;
+  }
+  if (index < 0) {
     return Error::InvalidArgument;
   }
+  if (index < flat_tensor_->tensors()->size()) {
+    return flat_tensor_->tensors()->Get(index)->fully_qualified_name()->c_str();
+  }
+  if (index < flat_tensor_->named_data()->size()) {
+    return flat_tensor_->named_data()->Get(index)->key()->c_str();
+  }
 
-  return flat_tensor_->tensors()->Get(index)->fully_qualified_name()->c_str();
+  return Error::InvalidArgument;
 }
 
 /* static */ Result<FlatTensorDataMap> FlatTensorDataMap::load(

extension/flat_tensor/test/flat_tensor_data_map_test.cpp

@@ -28,32 +28,36 @@ using torch::executor::util::FileDataLoader;
 
 class FlatTensorDataMapTest : public ::testing::Test {
  protected:
+  void create_loader(const char* path, const char* module_name) {
+    // Create a loader for the serialized data map.
+    Result<FileDataLoader> loader = FileDataLoader::from(path);
+    ASSERT_EQ(loader.error(), Error::Ok);
+    loaders_.insert(
+        {module_name,
+         std::make_unique<FileDataLoader>(std::move(loader.get()))});
+  }
   void SetUp() override {
     // Since these tests cause ET_LOG to be called, the PAL must be initialized
     // first.
     executorch::runtime::runtime_init();
 
-    // Load data map. The eager linear model is defined at:
-    // //executorch/test/models/linear_model.py
-    const char* path = std::getenv("ET_MODULE_LINEAR_DATA_PATH");
-    Result<FileDataLoader> loader = FileDataLoader::from(path);
-    ASSERT_EQ(loader.error(), Error::Ok);
-
-    data_map_loader_ =
-        std::make_unique<FileDataLoader>(std::move(loader.get()));
+    // Model defined in //executorch/test/models/linear_model.py
+    create_loader(std::getenv("ET_MODULE_LINEAR_DATA_PATH"), "linear");
+    // Model defined in //executorch/test/models/export_delegated_program.py
+    create_loader(std::getenv("ET_MODULE_LINEAR_XNN_DATA_PATH"), "linear_xnn");
   }
-  std::unique_ptr<FileDataLoader> data_map_loader_;
+  std::unordered_map<std::string, std::unique_ptr<FileDataLoader>> loaders_;
 };
 
 TEST_F(FlatTensorDataMapTest, LoadFlatTensorDataMap) {
   Result<FlatTensorDataMap> data_map =
-      FlatTensorDataMap::load(data_map_loader_.get());
+      FlatTensorDataMap::load(loaders_["linear"].get());
   EXPECT_EQ(data_map.error(), Error::Ok);
 }
 
 TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_GetMetadata) {
   Result<FlatTensorDataMap> data_map =
-      FlatTensorDataMap::load(data_map_loader_.get());
+      FlatTensorDataMap::load(loaders_["linear"].get());
   EXPECT_EQ(data_map.error(), Error::Ok);
 
   // Check tensor layouts are correct.
@@ -95,7 +99,7 @@ TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_GetMetadata) {
 
 TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_GetData) {
   Result<FlatTensorDataMap> data_map =
-      FlatTensorDataMap::load(data_map_loader_.get());
+      FlatTensorDataMap::load(loaders_["linear"].get());
   EXPECT_EQ(data_map.error(), Error::Ok);
 
   // Check tensor data sizes are correct.
@@ -116,7 +120,7 @@ TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_GetData) {
 
 TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_Keys) {
   Result<FlatTensorDataMap> data_map =
-      FlatTensorDataMap::load(data_map_loader_.get());
+      FlatTensorDataMap::load(loaders_["linear"].get());
   EXPECT_EQ(data_map.error(), Error::Ok);
 
   // Check num tensors is 2.
@@ -140,7 +144,7 @@ TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_Keys) {
 
 TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_LoadInto) {
   Result<FlatTensorDataMap> data_map =
-      FlatTensorDataMap::load(data_map_loader_.get());
+      FlatTensorDataMap::load(loaders_["linear"].get());
   EXPECT_EQ(data_map.error(), Error::Ok);
 
   // get the metadata
@@ -160,3 +164,62 @@ TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_LoadInto) {
   }
   free(data);
 }
+
+TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_GetData_Xnnpack) {
+  Result<FlatTensorDataMap> data_map =
+      FlatTensorDataMap::load(loaders_["linear_xnn"].get());
+  EXPECT_EQ(data_map.error(), Error::Ok);
+
+  // Check tensor data sizes are correct.
+  // 64eec129c8d3f58ee6b7ca145b25e312fa82d3d276db5adaedb59aaebb824885 is the
+  // hash of the 3*3 identity matrix
+  Result<FreeableBuffer> data_weight_res = data_map->get_data(
+      "64eec129c8d3f58ee6b7ca145b25e312fa82d3d276db5adaedb59aaebb824885");
+  ASSERT_EQ(Error::Ok, data_weight_res.error());
+  FreeableBuffer data_a = std::move(data_weight_res.get());
+  EXPECT_EQ(data_a.size(), 36); // 3*3*4 (3*3 matrix, 4 bytes per float)
+
+  // 15ec7bf0b50732b49f8228e07d24365338f9e3ab994b00af08e5a3bffe55fd8b is the
+  // hash of the 3*1 vector [1, 1, 1]
+  Result<FreeableBuffer> data_bias_res = data_map->get_data(
+      "15ec7bf0b50732b49f8228e07d24365338f9e3ab994b00af08e5a3bffe55fd8b");
+  ASSERT_EQ(Error::Ok, data_bias_res.error());
+  FreeableBuffer data_b = std::move(data_bias_res.get());
+  EXPECT_EQ(data_b.size(), 12); // 3*4 (3*1 vector, 4 bytes per float)
+
+  // Check get_data fails when key is not found.
+  Result<FreeableBuffer> data_c_res = data_map->get_data("c");
+  EXPECT_EQ(data_c_res.error(), Error::NotFound);
+}
+
+TEST_F(FlatTensorDataMapTest, FlatTensorDataMap_Keys_Xnnpack) {
+  Result<FlatTensorDataMap> data_map =
+      FlatTensorDataMap::load(loaders_["linear_xnn"].get());
+  EXPECT_EQ(data_map.error(), Error::Ok);
+
+  // Check num tensors is 2.
+  Result<size_t> num_tensors_res = data_map->get_num_keys();
+  ASSERT_EQ(Error::Ok, num_tensors_res.error());
+  EXPECT_EQ(num_tensors_res.get(), 2);
+
+  // Check get_key returns the correct keys.
+  Result<const char*> key0_res = data_map->get_key(0);
+  ASSERT_EQ(Error::Ok, key0_res.error());
+  EXPECT_EQ(
+      strcmp(
+          key0_res.get(),
+          "64eec129c8d3f58ee6b7ca145b25e312fa82d3d276db5adaedb59aaebb824885"),
+      0);
+
+  Result<const char*> key1_res = data_map->get_key(1);
+  ASSERT_EQ(Error::Ok, key1_res.error());
+  EXPECT_EQ(
+      strcmp(
+          key1_res.get(),
+          "15ec7bf0b50732b49f8228e07d24365338f9e3ab994b00af08e5a3bffe55fd8b"),
+      0);
+
+  // Check get_key fails when out of bounds.
+  Result<const char*> key2_res = data_map->get_key(2);
+  EXPECT_EQ(key2_res.error(), Error::InvalidArgument);
+}

extension/flat_tensor/test/targets.bzl

@@ -35,8 +35,8 @@ def define_common_targets(is_fbcode=False):
             # The tests use this var to find the program file to load. This uses
             # an fbcode target path because the authoring/export tools
             # intentionally don't work in xplat (since they're host-only tools).
-            "ET_MODULE_LINEAR_PROGRAM_PATH": "$(location fbcode//executorch/test/models:exported_program_and_data[ModuleLinear.pte])",
             "ET_MODULE_LINEAR_DATA_PATH": "$(location fbcode//executorch/test/models:exported_program_and_data[ModuleLinear.ptd])",
+            "ET_MODULE_LINEAR_XNN_DATA_PATH": "$(location fbcode//executorch/test/models:exported_program_data[ModuleLinear-e.ptd])",
         }
 
         runtime.cxx_test(

test/models/export_delegated_program.py

@@ -99,6 +99,11 @@ class ModuleLinear(torch.nn.Module):
     def __init__(self):
         super().__init__()
         self.linear = torch.nn.Linear(3, 3)
+        # Make the linear deterministic.
+        self.linear.weight.data = torch.tensor(
+            [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]]
+        )  # 3x3 identity matrix
+        self.linear.bias.data = torch.tensor([0.0, 0.0, 0.0])
 
     def forward(self, x: torch.Tensor):
         return self.linear(x)

test/models/targets.bzl

@@ -222,6 +222,7 @@ def define_common_targets():
         default_outs = ["."],
         visibility = [
             "//executorch/runtime/executor/test/...",
+            "//executorch/extension/flat_tensor/test/...",
             "//executorch/test/...",
         ],
     )