refactor!: Changing the C++ api to be snake case

BREAKING CHANGE: This changes the C++ API ::ts
APIs to be snake case and for CompileModules to
become just compile

Signed-off-by: Naren Dasan <narens@nvidia.com>
Signed-off-by: Naren Dasan <naren@narendasan.com>

.gitignore

@@ -56,4 +56,5 @@ examples/int8/ptq/ptq
 examples/int8/qat/qat
 examples/int8/training/vgg16/data/*
 examples/int8/datasets/data/*
-env/**/*
+env/**/*
+bazel-Torch-TensorRT-Preview

README.md

@@ -14,7 +14,7 @@ More Information / System Architecture:
 
 ## Building a docker container for Torch-TensorRT Preview
 
-We provide a `Dockerfile` in `docker/` directory. We build `Torch-TensorRT` on top of a `Pytorch NGC container` which provide basic dependencies (like CUDA, CUDNN, CUBLAS, TensorRT, Pytorch and others) The dependency libraries in the container can be found in the <a href="https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/index.html">release notes</a>.  
+We provide a `Dockerfile` in `docker/` directory. We build `Torch-TensorRT` on top of a `Pytorch NGC container` which provide basic dependencies (like CUDA, CUDNN, CUBLAS, TensorRT, Pytorch and others) The dependency libraries in the container can be found in the <a href="https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/index.html">release notes</a>.
 
 Please follow this instruction to build a Docker container.
 
@@ -41,7 +41,7 @@ auto compile_settings = torch_tensorrt::ts::CompileSpec({input});
 // FP16 execution
 compile_settings.enabled_precisions = {torch::kHalf};
 // Compile module
-auto trt_mod = torch_tensorrt::ts::CompileModule(ts_mod, compile_settings);
+auto trt_mod = torch_tensorrt::ts::compile(ts_mod, compile_settings);
 // Run like normal
 auto results = trt_mod.forward({in_tensor});
 // Save module for later

core/partitioning/README.md

@@ -62,6 +62,6 @@ torchtrt::ts::CompileSpec cfg(input_sizes);
 cfg.torch_fallback = torchtrt::CompileSpec::TorchFallback(true);
 cfg.torch_fallback.min_block_size = 2;
 cfg.torch_fallback.forced_fallback_ops.push_back("aten::relu");
-auto trt_mod = torchtrt::ts::CompileModule(mod, cfg);
+auto trt_mod = torchtrt::ts::compile(mod, cfg);
 auto out = trt_mod.forward({in});
 ```

cpp/bin/torchtrtc/main.cpp

@@ -600,7 +600,7 @@ int main(int argc, char** argv) {
   // Instead of compiling, just embed engine in a PyTorch module
   if (embed_engine) {
     std::string serialized_engine = read_buf(real_input_path);
-    auto trt_mod = torchtrt::ts::EmbedEngineInNewModule(serialized_engine, compile_settings.device);
+    auto trt_mod = torchtrt::ts::embed_engine_in_new_module(serialized_engine, compile_settings.device);
     trt_mod.save(real_output_path);
     return 0;
   }
@@ -622,12 +622,12 @@ int main(int argc, char** argv) {
   }
 
   if (save_engine) {
-    auto engine = torchtrt::ts::ConvertMethodToTRTEngine(mod, "forward", compile_settings);
+    auto engine = torchtrt::ts::convert_method_to_trt_engine(mod, "forward", compile_settings);
     std::ofstream out(real_output_path);
     out << engine;
     out.close();
   } else {
-    auto trt_mod = torchtrt::ts::CompileModule(mod, compile_settings);
+    auto trt_mod = torchtrt::ts::compile(mod, compile_settings);
 
     if (!no_threshold_check &&
         (compile_settings.enabled_precisions.size() == 1 &&

cpp/include/torch_tensorrt/torch_tensorrt.h

@@ -701,7 +701,7 @@ struct TORCHTRT_API CompileSpec {
  *
  * @returns bool: Method is supported by Torch-TensorRT.TorchScript
  */
-TORCHTRT_API bool CheckMethodOperatorSupport(const torch::jit::Module& module, std::string method_name);
+TORCHTRT_API bool check_method_operator_support(const torch::jit::Module& module, std::string method_name);
 
 /**
  * @brief Compile a TorchScript module for NVIDIA GPUs using TensorRT
@@ -717,7 +717,7 @@ TORCHTRT_API bool CheckMethodOperatorSupport(const torch::jit::Module& module, s
  *
  * @return: A new module trageting a TensorRT engine
  */
-TORCHTRT_API torch::jit::Module CompileModule(const torch::jit::Module& module, CompileSpec info);
+TORCHTRT_API torch::jit::Module compile(const torch::jit::Module& module, CompileSpec info);
 
 /**
  * @brief Compile a TorchScript method for NVIDIA GPUs using TensorRT
@@ -733,7 +733,7 @@ TORCHTRT_API torch::jit::Module CompileModule(const torch::jit::Module& module,
  * @return: std::string: Serialized TensorRT engine equivilant to the method
  * graph
  */
-TORCHTRT_API std::string ConvertMethodToTRTEngine(
+TORCHTRT_API std::string convert_method_to_trt_engine(
     const torch::jit::Module& module,
     std::string method_name,
     CompileSpec info);
@@ -751,6 +751,6 @@ TORCHTRT_API std::string ConvertMethodToTRTEngine(
  *
  * @return: A new module trageting a TensorRT engine
  */
-TORCHTRT_API torch::jit::Module EmbedEngineInNewModule(const std::string& engine, Device device);
+TORCHTRT_API torch::jit::Module embed_engine_in_new_module(const std::string& engine, Device device);
 } // namespace torchscript
 } // namespace torch_tensorrt

cpp/src/torch_tensorrt.cpp

@@ -12,11 +12,11 @@ namespace torchscript {
 // Defined in compile_spec.cpp
 torch_tensorrt::core::CompileSpec to_internal_compile_spec(CompileSpec external);
 
-bool CheckMethodOperatorSupport(const torch::jit::script::Module& module, std::string method_name) {
+bool check_method_operator_support(const torch::jit::script::Module& module, std::string method_name) {
   return torch_tensorrt::core::CheckMethodOperatorSupport(module, method_name);
 }
 
-std::string ConvertMethodToTRTEngine(
+std::string convert_method_to_trt_engine(
     const torch::jit::script::Module& module,
     std::string method_name,
     CompileSpec info) {
@@ -26,14 +26,14 @@ std::string ConvertMethodToTRTEngine(
   return torch_tensorrt::core::ConvertGraphToTRTEngine(module, method_name, to_internal_compile_spec(info));
 }
 
-torch::jit::script::Module CompileModule(const torch::jit::script::Module& module, CompileSpec info) {
+torch::jit::script::Module compile(const torch::jit::script::Module& module, CompileSpec info) {
   LOG_DEBUG(get_build_info());
   // Want to export a much simpler (non TRT header dependent) API so doing the
   // type conversion here
   return torch_tensorrt::core::CompileGraph(module, to_internal_compile_spec(info));
 }
 
-torch::jit::Module EmbedEngineInNewModule(const std::string& engine, Device device) {
+torch::jit::Module embed_engine_in_new_module(const std::string& engine, Device device) {
   return torch_tensorrt::core::EmbedEngineInNewModule(engine, to_internal_cuda_device(device));
 }
 

examples/benchmark/main.cpp

@@ -127,11 +127,11 @@ int main(int argc, const char* argv[]) {
   compile_spec.enabled_precisions.insert(torch::kF16);
 #endif
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
 #ifdef SAVE_ENGINE
   std::cout << "Compiling graph to save as TRT engine (/tmp/engine_converted_from_jit.trt)" << std::endl;
-  auto engine = torch_tensorrt::ts::ConvertMethodToTRTEngine(mod, "forward", compile_spec);
+  auto engine = torch_tensorrt::ts::convert_method_to_trt_engine(mod, "forward", compile_spec);
   std::ofstream out("/tmp/engine_converted_from_jit.trt");
   out << engine;
   out.close();

examples/int8/ptq/main.cpp

@@ -56,14 +56,14 @@ torch::jit::Module compile_int8_model(const std::string& data_dir, torch::jit::M
 
 #ifdef SAVE_ENGINE
   std::cout << "Compiling graph to save as TRT engine (/tmp/engine_converted_from_jit.trt)" << std::endl;
-  auto engine = torch_tensorrt::ts::ConvertMethodToTRTEngine(mod, "forward", compile_spec);
+  auto engine = torch_tensorrt::ts::convert_method_to_trt_engine(mod, "forward", compile_spec);
   std::ofstream out("/tmp/int8_engine_converted_from_jit.trt");
   out << engine;
   out.close();
 #endif
 
   std::cout << "Compiling and quantizing module" << std::endl;
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
   return std::move(trt_mod);
 }
 

examples/int8/qat/main.cpp

@@ -40,14 +40,14 @@ torch::jit::Module compile_int8_qat_model(const std::string& data_dir, torch::ji
 
 #ifdef SAVE_ENGINE
   std::cout << "Compiling graph to save as TRT engine (/tmp/engine_converted_from_jit.trt)" << std::endl;
-  auto engine = torch_tensorrt::ts::ConvertMethodToTRTEngine(mod, "forward", compile_spec);
+  auto engine = torch_tensorrt::ts::convert_method_to_trt_engine(mod, "forward", compile_spec);
   std::ofstream out("/tmp/int8_engine_converted_from_jit.trt");
   out << engine;
   out.close();
 #endif
 
   std::cout << "Compiling and quantizing module" << std::endl;
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
   return std::move(trt_mod);
 }
 

tests/accuracy/test_dla_fp16_accuracy.cpp

@@ -34,7 +34,7 @@ TEST_P(AccuracyTests, DLAFP16AccuracyIsClose) {
   compile_spec.device.allow_gpu_fallback = true;
   compile_spec.workspace_size = 1 << 28;
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
   torch::Tensor trt_correct = torch::zeros({1}, {torch::kCUDA}), trt_total = torch::zeros({1}, {torch::kCUDA});
   for (auto batch : *eval_dataloader) {

tests/accuracy/test_dla_int8_accuracy.cpp

@@ -61,7 +61,7 @@ TEST_P(AccuracyTests, DLAINT8AccuracyIsClose) {
   torch::Tensor jit_accuracy = (jit_correct / jit_total) * 100;
 
   // Compile Graph
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
   // Check the INT8 accuracy in TRT
   torch::Tensor trt_correct = torch::zeros({1}, {torch::kCUDA}), trt_total = torch::zeros({1}, {torch::kCUDA});

tests/accuracy/test_fp16_accuracy.cpp

@@ -29,7 +29,7 @@ TEST_P(AccuracyTests, FP16AccuracyIsClose) {
   auto compile_spec = torch_tensorrt::ts::CompileSpec({input_shape});
   compile_spec.enabled_precisions.insert(torch::kF16);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
   torch::Tensor trt_correct = torch::zeros({1}, {torch::kCUDA}), trt_total = torch::zeros({1}, {torch::kCUDA});
   for (auto batch : *eval_dataloader) {

tests/accuracy/test_fp32_accuracy.cpp

@@ -29,7 +29,7 @@ TEST_P(AccuracyTests, FP32AccuracyIsClose) {
   auto compile_spec = torch_tensorrt::ts::CompileSpec({input_shape});
   compile_spec.enabled_precisions.insert(torch::kF32);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
   torch::Tensor trt_correct = torch::zeros({1}, {torch::kCUDA}), trt_total = torch::zeros({1}, {torch::kCUDA});
   for (auto batch : *eval_dataloader) {

tests/accuracy/test_int8_accuracy.cpp

@@ -58,7 +58,7 @@ TEST_P(AccuracyTests, INT8AccuracyIsClose) {
   torch::Tensor jit_accuracy = (jit_correct / jit_total) * 100;
 
   // Compile Graph
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_spec);
 
   // Check the INT8 accuracy in TRT
   torch::Tensor trt_correct = torch::zeros({1}, {torch::kCUDA}), trt_total = torch::zeros({1}, {torch::kCUDA});

tests/cpp/test_compiled_modules.cpp

@@ -13,7 +13,7 @@ TEST_P(CppAPITests, CompiledModuleIsClose) {
   std::vector<at::Tensor> jit_results;
   jit_results.push_back(jit_results_ivalues.toTensor());
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, input_shapes);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, input_shapes);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());

tests/cpp/test_default_input_types.cpp

@@ -15,7 +15,7 @@ TEST_P(CppAPITests, InputsUseDefaultFP32) {
   auto spec = torch_tensorrt::ts::CompileSpec({in});
   spec.enabled_precisions.insert(torch_tensorrt::DataType::kHalf);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());
@@ -38,7 +38,7 @@ TEST_P(CppAPITests, InputsUseDefaultFP16) {
 
   mod.to(torch::kHalf);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());
@@ -60,7 +60,7 @@ TEST_P(CppAPITests, InputsUseDefaultFP16WithoutFP16Enabled) {
 
   mod.to(torch::kHalf);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());
@@ -84,7 +84,7 @@ TEST_P(CppAPITests, InputsRespectUserSettingFP16WeightsFP32In) {
 
   mod.to(torch::kHalf);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());
@@ -106,7 +106,7 @@ TEST_P(CppAPITests, InputsRespectUserSettingFP32WeightsFP16In) {
   auto spec = torch_tensorrt::ts::CompileSpec({in});
   spec.enabled_precisions.insert(torch_tensorrt::DataType::kHalf);
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());

tests/cpp/test_example_tensors.cpp

@@ -11,7 +11,7 @@ TEST_P(CppAPITests, InputsFromTensors) {
 
   auto spec = torch_tensorrt::ts::CompileSpec({trt_inputs_ivalues[0].toTensor()});
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, spec);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, spec);
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, trt_inputs_ivalues);
   std::vector<at::Tensor> trt_results;
   trt_results.push_back(trt_results_ivalues.toTensor());

tests/cpp/test_module_fallback.cpp

@@ -26,7 +26,7 @@ TEST(CppAPITest, ResNetModuleFallbacksCorrectly) {
   cfg.torch_executed_modules.push_back("torchvision.models.resnet.BasicBlock");
 
   auto jit_results = mod.forward(jit_inputs_ivalues).toTensor();
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, cfg);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, cfg);
   auto trt_results = trt_mod.forward(trt_inputs_ivalues).toTensor();
   ASSERT_TRUE(torch_tensorrt::tests::util::almostEqual(jit_results, trt_results, 2e-6));
 }
@@ -54,7 +54,7 @@ TEST(CppAPITest, MobileNetModuleFallbacksCorrectlyWithOneEngine) {
   cfg.torch_executed_modules.push_back("torchvision.models.mobilenetv2.ConvBNActivation");
 
   auto jit_results = mod.forward(jit_inputs_ivalues).toTensor();
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, cfg);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, cfg);
 
   auto g = trt_mod.get_method("forward").graph();
   auto nodes = g->block()->nodes();

tests/cpp/test_modules_as_engines.cpp

@@ -40,8 +40,8 @@ TEST_P(CppAPITests, ModuleToEngineToModuleIsClose) {
   cudaGetDevice(&device_id);
   compile_spec.device.device_type = torch_tensorrt::Device::DeviceType::kGPU;
   compile_spec.device.gpu_id = device_id;
-  auto engine = torch_tensorrt::ts::ConvertMethodToTRTEngine(mod, "forward", input_ranges);
-  auto trt_mod = torch_tensorrt::ts::EmbedEngineInNewModule(engine, compile_spec.device);
+  auto engine = torch_tensorrt::ts::convert_method_to_trt_engine(mod, "forward", input_ranges);
+  auto trt_mod = torch_tensorrt::ts::embed_engine_in_new_module(engine, compile_spec.device);
 
   torch::jit::IValue trt_results_ivalues = torch_tensorrt::tests::util::RunModuleForward(trt_mod, inputs_ivalues);
   std::vector<at::Tensor> trt_results;

tests/cpp/test_multi_gpu_serde.cpp

@@ -14,7 +14,7 @@ TEST_P(CppAPITests, CompiledModuleIsClose) {
   std::vector<at::Tensor> jit_results;
   jit_results.push_back(jit_results_ivalues.toTensor());
 
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, input_shapes);
+  auto trt_mod = torch_tensorrt::ts::compile(mod, input_shapes);
 
   // Deliberately changing the device ID. torch_tensorrt runtime should correct the Device ID internally
   torch_tensorrt::set_device(1);

tests/cpp/test_multiple_registered_engines.cpp

@@ -41,13 +41,13 @@ TEST(CppAPITest, CanRunMultipleEngines) {
   std::vector<at::Tensor> jit2_results;
   jit2_results.push_back(jit2_results_ivalues.toTensor());
 
-  auto trt_mod1 = torch_tensorrt::ts::CompileModule(mod1, input_shapes);
+  auto trt_mod1 = torch_tensorrt::ts::compile(mod1, input_shapes);
   torch::jit::IValue trt1_results_ivalues =
       torch_tensorrt::tests::util::RunModuleForward(trt_mod1, trt1_inputs_ivalues);
   std::vector<at::Tensor> trt1_results;
   trt1_results.push_back(trt1_results_ivalues.toTensor());
 
-  auto trt_mod2 = torch_tensorrt::ts::CompileModule(mod2, input_shapes);
+  auto trt_mod2 = torch_tensorrt::ts::compile(mod2, input_shapes);
   torch::jit::IValue trt2_results_ivalues =
       torch_tensorrt::tests::util::RunModuleForward(trt_mod2, trt2_inputs_ivalues);
   std::vector<at::Tensor> trt2_results;

tests/cpp/test_runtime_thread_safety.cpp

@@ -52,8 +52,8 @@ TEST(CppAPITests, RuntimeThreadSafety) {
   // FP32 execution
   compile_settings.enabled_precisions = {torch::kFloat};
   compile_settings.strict_types = true;
-  auto trt_mod = torch_tensorrt::ts::CompileModule(mod, compile_settings);
-  std::cout << "torch_tensorrt::ts::CompileModule" << std::endl;
+  auto trt_mod = torch_tensorrt::ts::compile(mod, compile_settings);
+  std::cout << "torch_tensorrt::ts::compile" << std::endl;
 
   int num_threads = 10;
   std::vector<torch::jit::IValue> out_vec(num_threads), trt_out_vec(num_threads);

tests/cpp/test_serialization.cpp

@@ -27,7 +27,7 @@ TEST_P(CppAPITests, SerializedModuleIsStillCorrect) {
     pre_serialized_inputs_ivalues.push_back(in.clone());
   }
 
-  auto pre_serialized_mod = torch_tensorrt::ts::CompileModule(mod, input_shapes);
+  auto pre_serialized_mod = torch_tensorrt::ts::compile(mod, input_shapes);
   torch::jit::IValue pre_serialized_results_ivalues =
       torch_tensorrt::tests::util::RunModuleForward(pre_serialized_mod, pre_serialized_inputs_ivalues);
   std::vector<at::Tensor> pre_serialized_results;
@@ -57,7 +57,7 @@ TEST_P(CppAPITests, SerializedDynamicModuleIsStillCorrect) {
   }
 
   auto pre_serialized_mod =
-      torch_tensorrt::ts::CompileModule(mod, torch_tensorrt::ts::CompileSpec(toInputRangesDynamic(input_shapes)));
+      torch_tensorrt::ts::compile(mod, torch_tensorrt::ts::CompileSpec(toInputRangesDynamic(input_shapes)));
   torch::jit::IValue pre_serialized_results_ivalues =
       torch_tensorrt::tests::util::RunModuleForward(pre_serialized_mod, pre_serialized_inputs_ivalues);
   std::vector<at::Tensor> pre_serialized_results;

tests/util/run_forward.cpp

@@ -18,7 +18,7 @@ std::vector<at::Tensor> RunModuleForwardAsEngine(torch::jit::Module& mod, std::v
     input_ranges.push_back(in.sizes());
   }
 
-  auto engine = torch_tensorrt::ts::ConvertMethodToTRTEngine(mod, "forward", input_ranges);
+  auto engine = torch_tensorrt::ts::convert_method_to_trt_engine(mod, "forward", input_ranges);
   return RunEngine(engine, inputs);
 }