[21/N] Fix clang-tidy warnings in jit (pytorch#134537)

Follows pytorch#133399

Pull Request resolved: pytorch#134537
Approved by: https://github.com/Skylion007

Author: cyyever

torch/csrc/jit/api/function_impl.h

@@ -7,7 +7,6 @@
 namespace torch::jit {
 
 struct TORCH_API GraphFunction : public Function {
-  // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
   GraphFunction(
       c10::QualifiedName name,
       std::shared_ptr<Graph> graph,

torch/csrc/jit/codegen/fuser/cuda/fused_kernel.cpp

@@ -16,13 +16,9 @@
 #include <cmath>
 #include <sstream>
 #include <stdexcept>
-#include <tuple>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
-namespace cuda {
+namespace torch::jit::fuser::cuda {
 
 // See NOTE [ USE OF NVRTC AND DRIVER API ]
 const at::cuda::NVRTC& nvrtc() {
@@ -85,7 +81,6 @@ void codegenOutputQuery(
 }
 
 // Compiles the specified kernel and stores the metadata required to run it
-// NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
 FusedKernelCUDA::FusedKernelCUDA(
     at::DeviceIndex device,
     std::string name,
@@ -114,6 +109,7 @@ FusedKernelCUDA::FusedKernelCUDA(
 
   // Acquires device and NVRTC properties (for compile arch and occupancy
   // calculations)
+  // NOLINTNEXTLINE(cppcoreguidelines-prefer-member-initializer)
   prop_ = at::cuda::getCurrentDeviceProperties();
   int major = 0, minor = 0;
   bool compile_to_sass = false;
@@ -197,8 +193,7 @@ static int ceilDiv(const int a, const int b) {
 void FusedKernelCUDA::launch_raw(
     const uint32_t numel,
     std::vector<void*>& arguments) const {
-  // NOLINTNEXTLINE(bugprone-unused-raii)
-  at::cuda::CUDAGuard{device_};
+  at::cuda::CUDAGuard guard{device_};
   // Hacked at::DeviceGuard (see note above)
   const auto prior_device = at::cuda::current_device();
   at::cuda::set_device(device_);
@@ -269,7 +264,4 @@ static std::shared_ptr<FusedKernel> createFusionKernel(
 
 RegisterFusionBackend reg(DeviceType::CUDA, createFusionKernel);
 
-} // namespace cuda
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser::cuda

torch/csrc/jit/codegen/fuser/cuda/fused_kernel.h

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <ATen/ATen.h>
 #include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 
@@ -50,11 +49,11 @@ struct TORCH_CUDA_CU_API FusedKernelCUDA
   // Note: per device to store device properties and compute launch heuristics
   //  Acquiring these values at launch time would be too slow
   at::DeviceIndex device_;
-  int maxBlocks_;
-  cudaDeviceProp* prop_;
+  int maxBlocks_{};
+  cudaDeviceProp* prop_{};
   std::vector<char> ptx_;
-  CUmodule module_;
-  CUfunction function_;
+  CUmodule module_{};
+  CUfunction function_{};
 };
 
 } // namespace torch::jit::fuser::cuda

torch/csrc/jit/codegen/fuser/partition_desc.h

@@ -28,8 +28,7 @@ struct TORCH_API PartitionDesc {
       // so dim - 1 is no longer contiguous
       cont[dim_ - 1] = false;
     }
-    // NOLINTNEXTLINE(modernize-make-shared)
-    subTensorDesc_.reset(new TensorDesc(_desc.scalar_type, cont));
+    subTensorDesc_ = std::make_shared<TensorDesc>(_desc.scalar_type, cont);
   }
 
   bool isNoop() const {

torch/csrc/jit/runtime/jit_trace.cpp

@@ -56,8 +56,8 @@ Node* traceNode(Node* node, TracingData& td, Stack& stack) {
 }
 
 void eraseAllOutputs(Node* opt_pn) {
-  // NOLINTNEXTLINE
-  for (int i = opt_pn->outputs().size() - 1; i >= 0; i--) {
+  for (auto i = static_cast<int64_t>(opt_pn->outputs().size()) - 1; i >= 0;
+       i--) {
     opt_pn->eraseOutput(i);
   }
 }
@@ -275,10 +275,12 @@ void insertTracingNodes(Block* block, ProfilingRecord* pr, TracingData& td) {
 // nodes and the outputs of the node in the scripted graph.
 // There are a few subtleties with tracing Ifs and Loops
 // discussed above
-std::shared_ptr<Graph> TraceGraph(std::shared_ptr<Graph> graph, Stack& stack) {
+std::shared_ptr<Graph> TraceGraph(
+    const std::shared_ptr<Graph>& graph,
+    Stack& stack) {
   TracingData td;
   GRAPH_DUMP("Before Inline:", graph);
-  Inline(*graph.get());
+  Inline(*graph);
   EliminateDeadCode(graph);
   GRAPH_DUMP("After Inline:", graph);
   auto pr = ProfilingRecord::instrumentGraph(graph);

torch/csrc/jit/runtime/jit_trace.h

@@ -3,6 +3,6 @@
 
 namespace torch::jit {
 TORCH_API std::shared_ptr<Graph> TraceGraph(
-    std::shared_ptr<Graph> graph,
+    const std::shared_ptr<Graph>& graph,
     Stack& stack);
 } // namespace torch::jit

torch/csrc/jit/runtime/register_ops_utils.cpp

@@ -287,20 +287,20 @@ void listAdd(Stack& stack) {
   c10::List<IValue> ret = make_result_list<IValue>(a.elementType());
 
   if (a.use_count() == 1) {
-    ret = std::move(a);
+    ret = a;
   } else {
     ret = a.copy();
   }
 
-  ret.append(std::move(b));
+  ret.append(b);
 
   push(stack, std::move(ret));
 }
 
 void listInplaceAdd(Stack& stack) {
   c10::List<IValue> b = pop(stack).to<c10::List<IValue>>();
   c10::List<IValue> a = pop(stack).to<c10::List<IValue>>();
-  a.append(std::move(b));
+  a.append(b);
   push(stack, std::move(a));
 }
 

torch/csrc/jit/runtime/register_prim_ops.cpp

@@ -830,7 +830,7 @@ static const std::vector<OperatorGeneratorArgs> opGenArgs{
             ss << i;
           }
           drop(stack, num_inputs);
-          ss << std::endl;
+          ss << '\n';
           auto* handler = getPrintHandler();
           TORCH_INTERNAL_ASSERT(handler);
           handler(ss.str());

torch/csrc/jit/runtime/script_profile.cpp

@@ -147,7 +147,7 @@ const ScriptProfile::SourceMap& ScriptProfile::dumpStats() {
   for (const auto& datapoint : datapoints_) {
     if (const auto& source = datapoint->sourceRange.source()) {
       if (auto fileLineCol = datapoint->sourceRange.file_line_col()) {
-        auto it = sourceMap_.find(*source.get());
+        auto it = sourceMap_.find(*source);
         if (it == sourceMap_.end()) {
           it = sourceMap_.emplace(SourceRef{source}, LineMap{}).first;
         }

torch/csrc/jit/runtime/symbolic_shape_registry.cpp

@@ -219,7 +219,7 @@ void checkInputAndOutputTypes(
 
 void transformShapeFunction(
     const FunctionSchema* schema_string,
-    std::shared_ptr<Graph> graph) {
+    const std::shared_ptr<Graph>& graph) {
   Inline(*graph);
 
   // ATEN operators can return multiple unboxed values, this in contrast to
@@ -411,7 +411,7 @@ TORCH_API std::optional<BoundedShapeGraphs> boundedGraphsForSchema(
 
 void RegisterShapeComputeGraphForSchema(
     const FunctionSchema& schema,
-    std::shared_ptr<Graph> g) {
+    const std::shared_ptr<Graph>& g) {
   std::lock_guard<std::mutex> guard(lock);
   if (cached_schema_to_graph.empty()) {
     loadFunctions();

torch/csrc/jit/runtime/symbolic_shape_registry.h

@@ -52,7 +52,7 @@ struct BoundedShapeGraphs {
 
 TORCH_API void RegisterShapeComputeGraphForSchema(
     const FunctionSchema& schema,
-    std::shared_ptr<Graph> g);
+    const std::shared_ptr<Graph>& g);
 
 TORCH_API std::optional<std::shared_ptr<Graph>> shapeComputeGraphForSchema(
     const FunctionSchema& schema);

torch/csrc/jit/serialization/pickler.h

@@ -94,7 +94,6 @@ enum class PickleOpCode : char {
 
 using ::c10::IValue;
 
-// NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
 struct WriteableTensorData {
   const char* data() const {
     return static_cast<const char*>(tensor_.storage().data());
@@ -140,7 +139,6 @@ class TORCH_API Pickler {
         memoized_class_types_(memoized_class_types),
         get_tensor_id_(std::move(get_tensor_id)),
         tag_aggregates_(tag_aggregates) {}
-  // NOLINTNEXTLINE(bugprone-exception-escape)
   ~Pickler();
 
   // Push protocol onto the stack