[3/N] Use internal linkage in C++ files (#151297)

Follows #151070.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/151297
Approved by: https://github.com/Skylion007

benchmarks/static_runtime/test_static_module.cc

@@ -16,8 +16,6 @@ using namespace torch;
 using namespace torch::jit;
 using namespace torch::jit::test;
 
-C10_DECLARE_bool(static_runtime_disable_debug_memory_overlap_check);
-
 namespace {
 
 StaticModule makeStaticModuleFromScript(const std::string& script) {

caffe2/utils/threadpool/ThreadPool.h

@@ -9,6 +9,7 @@
 #include <mutex>
 #include <vector>
 
+#include "c10/util/Flags.h"
 #include "caffe2/core/common.h"
 
 //
@@ -65,4 +66,14 @@ class TORCH_API /*alignas(kCacheLineSize)*/ ThreadPool {
 size_t getDefaultNumThreads();
 } // namespace caffe2
 
+C10_DECLARE_bool(caffe2_threadpool_force_inline);
+
+// Whether or not threadpool caps apply to Android
+C10_DECLARE_int(caffe2_threadpool_android_cap);
+
+// Whether or not threadpool caps apply to iOS and MacOS
+C10_DECLARE_int(caffe2_threadpool_ios_cap);
+C10_DECLARE_int(caffe2_threadpool_macos_cap);
+
+C10_DECLARE_int(pthreadpool_size);
 #endif // CAFFE2_UTILS_THREADPOOL_H_

caffe2/utils/threadpool/pthreadpool-cpp.cc

@@ -1,9 +1,8 @@
 #include <caffe2/utils/threadpool/pthreadpool-cpp.h>
 #include <caffe2/utils/threadpool/thread_pool_guard.h>
+#include <caffe2/utils/threadpool/ThreadPool.h>
 #include <c10/util/Exception.h>
 
-#include <atomic>
-
 namespace {
 // After fork, the child process inherits the data-structures of the parent
 // process' thread-pool, but since those threads don't exist, the thread-pool
@@ -102,9 +101,6 @@ PThreadPool* pthreadpool(size_t thread_count) {
   return threadpool.get();
 }
 
-// Forward declaration
-size_t getDefaultNumThreads();
-
 PThreadPool* pthreadpool() {
   return pthreadpool(getDefaultNumThreads());
 }

caffe2/utils/threadpool/thread_pool_guard.cpp

@@ -2,7 +2,7 @@
 
 namespace caffe2 {
 
-thread_local bool _NoPThreadPoolGuard_enabled = false;
+static thread_local bool _NoPThreadPoolGuard_enabled = false;
 
 bool _NoPThreadPoolGuard::is_enabled() {
   return _NoPThreadPoolGuard_enabled;

tools/autograd/gen_autograd_functions.py

@@ -272,7 +272,7 @@ PY_RAW_GETSETDEF_STRUCT = CodeTemplate(
 # Getter templates
 GETTER_DEFINITION = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   auto prop = static_cast<${op}*>(self->cdata.get())->${name};
   ${body}
@@ -283,7 +283,7 @@ PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_SAVEDVAR = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   const auto& prop = static_cast<${op}*>(self->cdata.get())->${name}_;
   ${body}
@@ -294,7 +294,7 @@ PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_RAW_SAVEDVAR = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   const auto& prop = static_cast<${op}*>(self->cdata.get())->${name}_;
   ${body}
@@ -305,7 +305,7 @@ PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_VEC_SAVEDVAR = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   const auto *node = static_cast<${op}*>(self->cdata.get());
   const auto& prop = node->${name}_;
@@ -321,7 +321,7 @@ PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_RAW_VEC_SAVEDVAR = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   const auto *node = static_cast<${op}*>(self->cdata.get());
   const auto& prop = node->${name}_;
@@ -337,7 +337,7 @@ PyObject* THP${op}_${name}_raw_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_OPT = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   auto opt_prop = static_cast<${op}*>(self->cdata.get())->${name};
   if (!opt_prop.has_value()) {
@@ -352,7 +352,7 @@ PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
 
 GETTER_DEFINITION_OPT_ARRAYREF = CodeTemplate(
     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   auto opt_prop = static_cast<${op}*>(self->cdata.get())->${name};
   if (!opt_prop.list.has_value()) {
@@ -832,7 +832,7 @@ def process_function(info: DifferentiabilityInfo, template: CodeTemplate) -> str
             getter_definitions.append(
                 CodeTemplate(
                     """\
-PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
+static PyObject* THP${op}_${name}_getter(THPCppFunction *self, void *_unused) {
   HANDLE_TH_ERRORS
   const auto *node = static_cast<${op}*>(self->cdata.get());
   const auto& prop = node->${name};

tools/autograd/templates/python_variable_methods.cpp

@@ -842,7 +842,7 @@ static PyObject * THPVariable_requires_grad_(PyObject* self, PyObject* args, PyO
   END_HANDLE_TH_ERRORS
 }
 
-inline bool dispatch_is_contiguous(const Tensor & self, MemoryFormat memory_format) {
+static inline bool dispatch_is_contiguous(const Tensor & self, MemoryFormat memory_format) {
   return self.is_contiguous(memory_format);
 }
 

torch/csrc/jit/api/function_impl.h

@@ -176,5 +176,5 @@ struct TORCH_API GraphFunction : public Function {
 TORCH_API GraphFunction* tryToGraphFunction(Function&) noexcept;
 TORCH_API GraphFunction& toGraphFunction(Function&);
 TORCH_API const GraphFunction& toGraphFunction(const Function&);
-
 } // namespace torch::jit
+C10_DECLARE_bool(torch_jit_do_not_store_optimized_graph);

torch/csrc/jit/mobile/train/export_data.cpp

@@ -18,8 +18,6 @@
 namespace torch::jit {
 namespace mobile {
 
-char const* toString(OpCode op);
-
 namespace {
 
 /**

torch/csrc/jit/python/init.cpp

@@ -163,7 +163,8 @@ std::optional<IValue> toTypeInferredIValueOptional(py::handle input) {
 }
 } // anonymous namespace
 
-#if !defined(USE_ROCM)
+#if defined(BUILDING_TESTS) && !defined(USE_ROCM)
+// NOLINTNEXTLINE(misc-use-internal-linkage)
 TORCH_API void runJITCPPTests();
 #endif
 

torch/csrc/jit/runtime/profiling_graph_executor_impl.h

@@ -7,6 +7,11 @@ TORCH_DECLARE_bool(torch_jit_static_then_dynamic);
 
 TORCH_DECLARE_bool(torch_jit_always_dynamic);
 
+C10_DECLARE_bool(torch_jit_release_profiling_graph_after_optimization);
+C10_DECLARE_int32(torch_jit_release_profiling_graph_delay_in_seconds);
+C10_DECLARE_int64(torch_jit_num_profiled_runs);
+C10_DECLARE_int64(torch_jit_bailout_depth);
+
 namespace torch::jit {
 
 TORCH_API void runNooptPassPipeline(std::shared_ptr<Graph>& graph);

torch/csrc/jit/runtime/static/impl.h

@@ -1145,3 +1145,4 @@ class TORCH_API StaticRuntime {
 };
 
 } // namespace torch::jit
+C10_DECLARE_bool(static_runtime_disable_debug_memory_overlap_check);

torch/csrc/jit/runtime/static/ops.h

@@ -182,5 +182,6 @@ bool sr_schema_check(
 }
 
 bool sr_schema_check_kind(torch::jit::Node* node, c10::Symbol node_kind);
-
 } // namespace torch::jit
+
+C10_DECLARE_bool(static_runtime_enable_fast_math);

torch/csrc/jit/runtime/static/passes.h

@@ -86,4 +86,5 @@ TORCH_API void UseInPlaceGetRealInputsFromOptionalInputsV2(
 
 TORCH_API void PrepackWeights(std::shared_ptr<Graph>& graph);
 
+C10_DECLARE_bool(enable_clip_ranges_gather_fusions);
 } // namespace torch::jit

torchgen/operator_versions/gen_mobile_upgraders.py

@@ -129,12 +129,9 @@ UPGRADER_CPP_SRC = CodeTemplate(
     MOBILE_UPGRADERS_HEADER_DESCRIPTION
     + """
 #include <caffe2/serialize/versions.h>
+#include <torch/csrc/jit/mobile/type_parser.h>
 #include <torch/csrc/jit/mobile/upgrader_mobile.h>
 
-namespace c10 {
-TypePtr parseType(const std::string& pythonStr);
-} // namespace c10
-
 namespace torch {
 namespace jit {