[Code Clean] Replace std::runtime_error with TORCH_CHECK (#165305)

Fixes part of #148114

Including:

- torch/csrc/distributed

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165305
Approved by: https://github.com/FFFrog, https://github.com/albanD

torch/csrc/distributed/c10d/TCPStoreLibUvBackend.cpp

@@ -361,13 +361,12 @@ class UvTcpServer : public UvTcpSocket {
 
     int addr_len = sizeof(addr_s);
 
-    if (uv_tcp_getsockname(
+    TORCH_CHECK(
+        uv_tcp_getsockname(
             (uv_tcp_t*)unsafeGetStream(),
             reinterpret_cast<::sockaddr*>(&addr_s),
-            &addr_len) != 0) {
-      throw std::runtime_error(
+            &addr_len) == 0,
         "The port number of the socket cannot be retrieved.");
-    }
 
     if (addr_s.ss_family == AF_INET) {
       portNum_ = ntohs(reinterpret_cast<sockaddr_in*>(&addr_s)->sin_port);

torch/csrc/distributed/c10d/control_collectives/StoreCollectives.cpp

@@ -49,7 +49,7 @@ void StoreCollectives::barrier(
           msg += fmt::format("{}, ", i);
         }
       }
-      throw std::runtime_error(msg + e.what());
+      TORCH_CHECK(false, msg, e.what());
     }
   }
 }
@@ -118,7 +118,7 @@ std::vector<std::vector<uint8_t>> StoreCollectives::gatherRecv(
         msg += fmt::format("{}, ", i);
       }
     }
-    throw std::runtime_error(msg + e.what());
+    TORCH_CHECK(false, msg, e.what());
   }
 
   // insert local data
@@ -194,7 +194,7 @@ std::vector<std::vector<uint8_t>> StoreCollectives::allGather(
         msg += fmt::format("{}, ", i);
       }
     }
-    throw std::runtime_error(msg + e.what());
+    TORCH_CHECK(false, msg, e.what());
   }
 }
 

torch/csrc/distributed/c10d/control_plane/PythonHandlers.cpp

@@ -4,6 +4,7 @@
 #include <fstream>
 #include <string>
 
+#include <c10/util/Exception.h>
 #include <c10/util/tempfile.h>
 #include <torch/csrc/distributed/c10d/exception.h>
 #include <torch/csrc/utils/pybind.h>
@@ -17,9 +18,7 @@ RegisterHandler tracebackHandler{
       auto tmpfile = c10::make_tempfile("torch-dump_traceback");
 
       auto cfile = ::fopen(tmpfile.name.c_str(), "w");
-      if (!cfile) {
-        throw std::runtime_error("failed to open file for writing");
-      }
+      TORCH_CHECK(cfile, "failed to open file for writing");
 
       {
         py::gil_scoped_acquire guard{};

torch/csrc/distributed/c10d/control_plane/WorkerServer.cpp

@@ -2,6 +2,7 @@
 #include <unordered_map>
 
 #include <ATen/core/interned_strings.h>
+#include <c10/util/Exception.h>
 #include <c10/util/FileSystem.h>
 #include <c10/util/thread_name.h>
 #include <caffe2/utils/threadpool/WorkersPool.h>
@@ -144,22 +145,20 @@ WorkerServer::WorkerServer(const std::string& hostOrFile, int port) {
   if (port == -1) {
     // using unix sockets
     server_.set_address_family(AF_UNIX);
-
-    if (c10::filesystem::exists(hostOrFile)) {
-      throw std::runtime_error(fmt::format("{} already exists", hostOrFile));
-    }
+    TORCH_CHECK(
+        !c10::filesystem::exists(hostOrFile),
+        fmt::format("{} already exists", hostOrFile));
 
     C10D_WARNING("Server listening to UNIX {}", hostOrFile);
-    if (!server_.bind_to_port(hostOrFile, 80)) {
-      throw std::runtime_error(fmt::format("Error binding to {}", hostOrFile));
-    }
+    TORCH_CHECK(
+        server_.bind_to_port(hostOrFile, 80),
+        fmt::format("Error binding to {}", hostOrFile));
   } else {
     C10D_WARNING("Server listening to TCP {}:{}", hostOrFile, port);
-    if (!server_.bind_to_port(hostOrFile, port)) {
-      throw std::runtime_error(
+    TORCH_CHECK(
+        server_.bind_to_port(hostOrFile, port),
         fmt::format("Error binding to {}:{}", hostOrFile, port));
   }
-  }
 
   serverThread_ = std::thread([this]() {
     c10::setThreadName("pt_workerserver");