[20/N] Fix clang-tidy warnings in jit (pytorch#133399)

Follows pytorch#133067

Pull Request resolved: pytorch#133399
Approved by: https://github.com/Skylion007

Author: cyyever

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

@@ -8,9 +8,7 @@
 #include <cstdint>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Describes the (runtime) arguments to a kernel.
 // ArgSpecs are also used as keys to lookup instantiated kernels, so
@@ -55,6 +53,4 @@ struct TORCH_API ArgSpec {
   int device_;
 };
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -18,9 +18,7 @@
 #include <tuple>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Template for computing the offset into the tensor to access a value
 static auto dim_calc = at::jit::CodeTemplate(R"(
@@ -538,7 +536,7 @@ std::string generateKernel(
   //       places where the constant None node is used
   // Note: No need to iterate over reference as n is a pointer
   for (const auto n : graph.nodes()) {
-    static_assert(std::is_pointer<decltype(n)>::value, "n must be a pointer");
+    static_assert(std::is_pointer_v<decltype(n)>, "n must be a pointer");
     // Note: FusedConcat nodes work by narrowing the output Tensors before the
     // kernel runs
     if (n->kind() == prim::FusedConcat)
@@ -680,11 +678,9 @@ std::string generateKernel(
   }
 
   if (debugFuser()) {
-    std::cerr << "fusion code:" << code_string << std::endl;
+    std::cerr << "fusion code:" << code_string << '\n';
   }
   return code_string;
 }
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -9,9 +9,7 @@
 #include <string>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Creates a CPU or CUDA kernel for the given graph.
 // Returns the C++ or CUDA string implementing the kernel.
@@ -23,6 +21,4 @@ TORCH_API std::string generateKernel(
     const std::vector<std::pair<const Value*, const TensorDesc>>& outputs,
     const bool use_cuda);
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

torch/csrc/jit/codegen/fuser/compiler.cpp

@@ -30,9 +30,7 @@ std::mutex& fusionBackendLock() {
 }
 } // namespace
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 static std::unordered_map<at::Device::Type, FusedKernelConstructor>&
 getFusionBackends() {
@@ -297,6 +295,4 @@ std::shared_ptr<FusedKernel> compileKernel(
       spec.hasRandom());
 }
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -11,9 +11,7 @@
 #include <cstdint>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Performs device-independent "upfront" compilation of the given fusion_group,
 // if it has not been registered already.
@@ -55,6 +53,4 @@ struct TORCH_API RegisterFusionBackend {
   }
 };
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -115,14 +115,12 @@ FusedKernelCUDA::FusedKernelCUDA(
   // Acquires device and NVRTC properties (for compile arch and occupancy
   // calculations)
   prop_ = at::cuda::getCurrentDeviceProperties();
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  int major, minor;
+  int major = 0, minor = 0;
   bool compile_to_sass = false;
   codegenOutputQuery(prop_, major, minor, compile_to_sass);
 
   // Creates the NVRTC program
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  nvrtcProgram program;
+  nvrtcProgram program{};
   AT_CUDA_NVRTC_CHECK(nvrtc().nvrtcCreateProgram(
       &program, code_.c_str(), nullptr, 0, nullptr, nullptr));
 
@@ -144,17 +142,14 @@ FusedKernelCUDA::FusedKernelCUDA(
       "compute_" +
 #endif
       std::to_string(major) + std::to_string(minor);
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   const std::vector<const char*> args = {
       "--std=c++17", compute.c_str(), "-default-device"};
 #endif
   const auto result =
       nvrtc().nvrtcCompileProgram(program, args.size(), args.data());
   if (result != NVRTC_SUCCESS) {
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    size_t logsize;
+    size_t logsize = 0;
     AT_CUDA_NVRTC_CHECK(nvrtc().nvrtcGetProgramLogSize(program, &logsize));
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
     std::vector<char> log(logsize);
     AT_CUDA_NVRTC_CHECK(nvrtc().nvrtcGetProgramLog(program, log.data()));
     std::stringstream cu;
@@ -164,8 +159,7 @@ FusedKernelCUDA::FusedKernelCUDA(
   ResourceGuard holdProgram(
       [&] { AT_CUDA_NVRTC_CHECK(nvrtc().nvrtcDestroyProgram(&program)); });
   AT_CUDA_NVRTC_CHECK(result);
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  size_t ptx_size;
+  size_t ptx_size = 0;
 #if defined(CUDA_VERSION) && CUDA_VERSION >= 11010
   // compile_to_sass determines whether we are generating SASS or PTX, hence
   // the different API.

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

@@ -12,10 +12,7 @@
 #include <string>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
-namespace cuda {
+namespace torch::jit::fuser::cuda {
 
 // query codegen output arch and target
 TORCH_CUDA_CU_API void codegenOutputQuery(
@@ -60,7 +57,4 @@ struct TORCH_CUDA_CU_API FusedKernelCUDA
   CUfunction function_;
 };
 
-} // namespace cuda
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser::cuda

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

@@ -3,10 +3,7 @@
 #include <ATen/code_template.h>
 #include <torch/csrc/Export.h>
 
-namespace torch {
-namespace jit {
-namespace fuser {
-namespace cuda {
+namespace torch::jit::fuser::cuda {
 
 /*with type_as not checking type of its input, a fusion group can have non-fp32
 tensor as input. Correct code for this case is generated, however, nvrtc does
@@ -405,7 +402,4 @@ __device__ float __bfloat162float(const __nv_bfloat16 a) {
 )";
 #endif
 
-} // namespace cuda
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser::cuda

torch/csrc/jit/codegen/fuser/executor.cpp

@@ -14,15 +14,9 @@
 #include <optional>
 
 #include <algorithm>
-#include <iostream> // TODO: remove, debugging only
-#include <map>
-#include <stdexcept>
-#include <tuple>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Returns the "map size" for this run, which is the common size for all
 // intermediate tensors.
@@ -215,8 +209,7 @@ static void launchFusion(
 
   // Computes map_size, numel from the first input
   at::IntArrayRef map_size;
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  uint32_t numel;
+  uint32_t numel = 0;
   std::vector<int64_t> keep_alive_size;
   if (fusion.chunkDesc()[0].isNoop()) {
     map_size = inputs[0].sizes();
@@ -409,6 +402,4 @@ bool runFusion(const int64_t key, Stack& stack, std::string* code_out) {
   return true;
 }
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -7,9 +7,7 @@
 
 #include <cstdint>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 // Runs the fusion associated with the key (see registerFusion() in interface.h)
 // on the inputs taken from the given Stack.
@@ -18,6 +16,4 @@ TORCH_API bool runFusion(
     Stack& stack,
     std::string* code_out = nullptr);
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

torch/csrc/jit/codegen/fuser/fallback.cpp

@@ -9,9 +9,7 @@
 
 #include <stdexcept>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 namespace {
 c10::AliasAnalysisKind aliasAnalysisIsSpecialCase() {
@@ -46,6 +44,4 @@ void runFallback(int64_t key, Stack& stack) {
   InterpreterState{(*maybe_spec)->code()}.run(stack);
 }
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -4,12 +4,8 @@
 
 #include <cstdlib>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 void runFallback(int64_t key, Stack& stack);
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

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

@@ -9,14 +9,11 @@
 #include <string>
 #include <vector>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 struct FusedKernel {
   AT_DISALLOW_COPY_AND_ASSIGN(FusedKernel);
 
-  // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
   FusedKernel(
       std::string name,
       std::string code,
@@ -98,6 +95,4 @@ struct FusedKernel {
   const bool has_random_;
 };
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser

torch/csrc/jit/codegen/fuser/interface.cpp

@@ -8,8 +8,7 @@
 #include <c10/util/Flags.h>
 #include <stdexcept>
 
-namespace torch {
-namespace jit {
+namespace torch::jit {
 
 namespace detail {
 
@@ -105,5 +104,4 @@ size_t nCompiledKernels() {
   return fuser::nCompiledKernels();
 }
 
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit

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

@@ -9,8 +9,7 @@
 #include <memory>
 #include <vector>
 
-namespace torch {
-namespace jit {
+namespace torch::jit {
 
 constexpr int kCPUDevice = -1;
 
@@ -52,5 +51,4 @@ TORCH_API std::string debugGetFusedKernelCode(
 
 TORCH_API size_t nCompiledKernels();
 
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit

torch/csrc/jit/codegen/fuser/kernel_cache.cpp

@@ -6,9 +6,7 @@
 #include <mutex>
 #include <unordered_map>
 
-namespace torch {
-namespace jit {
-namespace fuser {
+namespace torch::jit::fuser {
 
 struct KernelCacheImpl {
   // Note: std::unordered_map does not invalidate references even if rehashing
@@ -76,7 +74,7 @@ std::optional<KernelSpec*> retrieve(const int64_t key) {
 }
 
 // precondition: graph has been normalized via normalizeGraphForCache
-std::optional<KernelSpec*> lookupGraph(std::shared_ptr<Graph> graph) {
+std::optional<KernelSpec*> lookupGraph(const std::shared_ptr<Graph>& graph) {
   auto& cache = getKernelCache();
   std::string repr = graph->toString(false);
 
@@ -87,6 +85,4 @@ std::optional<KernelSpec*> lookupGraph(std::shared_ptr<Graph> graph) {
   return nolock_retrieve(cache, it->second);
 }
 
-} // namespace fuser
-} // namespace jit
-} // namespace torch
+} // namespace torch::jit::fuser