Maintain code cleanlines (#1960)

CuiYifeng · web-flow · commit 728c3f1a0de1 · 2025-08-26T08:21:58.000Z
To maintain code cleanlines
diff --git a/src/ATen/native/xpu/Activation.cpp b/src/ATen/native/xpu/Activation.cpp
@@ -6,8 +6,8 @@
 
 #include <ATen/native/TensorIterator.h>
 
-#include <ATen/ops/empty_like.h>
 #include <ATen/ops/empty.h>
+#include <ATen/ops/empty_like.h>
 #include <ATen/ops/gelu_backward_native.h>
 #include <ATen/ops/gelu_native.h>
 
diff --git a/src/ATen/native/xpu/AdaptiveAveragePooling2d.cpp b/src/ATen/native/xpu/AdaptiveAveragePooling2d.cpp
@@ -5,10 +5,10 @@
 
 #include <comm/xpu_aten.h>
 
-#include <ATen/ops/mean.h>
-#include <ATen/ops/zeros_like.h>
 #include <ATen/ops/_adaptive_avg_pool2d_backward_native.h>
 #include <ATen/ops/_adaptive_avg_pool2d_native.h>
+#include <ATen/ops/mean.h>
+#include <ATen/ops/zeros_like.h>
 
 #include <ATen/native/xpu/sycl/AdaptiveAveragePooling2dKernels.h>
 
diff --git a/src/ATen/native/xpu/AdaptiveAveragePooling3d.cpp b/src/ATen/native/xpu/AdaptiveAveragePooling3d.cpp
@@ -2,10 +2,10 @@
 #include <ATen/native/AdaptivePooling.h>
 #include <ATen/native/xpu/sycl/AdaptiveAveragePooling3dKernels.h>
 
-#include <ATen/ops/empty.h>
-#include <ATen/ops/empty_like.h>
 #include <ATen/ops/adaptive_avg_pool3d_backward_native.h>
 #include <ATen/ops/adaptive_avg_pool3d_native.h>
+#include <ATen/ops/empty.h>
+#include <ATen/ops/empty_like.h>
 
 namespace at::native {
 
diff --git a/src/ATen/native/xpu/AdaptiveMaxPooling3d.cpp b/src/ATen/native/xpu/AdaptiveMaxPooling3d.cpp
@@ -3,9 +3,9 @@
 #include <ATen/native/AdaptivePooling.h>
 #include <ATen/native/xpu/sycl/AdaptiveMaxPooling3dKernels.h>
 
-#include <ATen/ops/empty.h>
 #include <ATen/ops/adaptive_max_pool3d_backward_native.h>
 #include <ATen/ops/adaptive_max_pool3d_native.h>
+#include <ATen/ops/empty.h>
 
 namespace at {
 namespace native {
diff --git a/src/ATen/native/xpu/Col2Im.cpp b/src/ATen/native/xpu/Col2Im.cpp
@@ -6,8 +6,8 @@
 
 #include <ATen/native/xpu/sycl/Col2ImKernel.h>
 
-#include <comm/xpu_aten.h>
 #include <ATen/ops/col2im_native.h>
+#include <comm/xpu_aten.h>
 
 namespace at::native {
 
diff --git a/src/ATen/native/xpu/ReflectionPad.cpp b/src/ATen/native/xpu/ReflectionPad.cpp
@@ -5,13 +5,13 @@
 #include <ATen/native/xpu/sycl/ReflectionPadKernels.h>
 
 #include <ATen/ops/empty.h>
-#include <ATen/ops/zeros_like.h>
 #include <ATen/ops/reflection_pad1d_backward_native.h>
 #include <ATen/ops/reflection_pad1d_native.h>
 #include <ATen/ops/reflection_pad2d_backward_native.h>
 #include <ATen/ops/reflection_pad2d_native.h>
 #include <ATen/ops/reflection_pad3d_backward_native.h>
 #include <ATen/ops/reflection_pad3d_native.h>
+#include <ATen/ops/zeros_like.h>
 #include "ATen/TensorMeta.h"
 
 namespace at {
diff --git a/src/ATen/native/xpu/TensorAdvancedIndexing.cpp b/src/ATen/native/xpu/TensorAdvancedIndexing.cpp
@@ -21,8 +21,8 @@
 #include <torch/library.h>
 
 #include <ATen/ops/index_add_meta.h>
-#include <ATen/ops/index_reduce_meta.h>
 #include <ATen/ops/index_add_native.h>
+#include <ATen/ops/index_reduce_meta.h>
 #include <ATen/ops/index_reduce_native.h> //generated
 //#include <ATen/ops/index_reduce_prod_native.h> //generated
 
diff --git a/src/ATen/native/xpu/TensorFactories.cpp b/src/ATen/native/xpu/TensorFactories.cpp
@@ -6,7 +6,6 @@
 #include <ATen/ops/_efficientzerotensor_native.h>
 #include <ATen/ops/empty_native.h>
 #include <ATen/ops/empty_strided_native.h>
-#include <ATen/ops/_efficientzerotensor_native.h>
 
 #include <ATen/native/xpu/sycl/ComplexKernels.h>
 #include <ATen/native/xpu/sycl/RandpermKernel.h>
diff --git a/src/ATen/native/xpu/TensorShape.cpp b/src/ATen/native/xpu/TensorShape.cpp
@@ -8,10 +8,10 @@
 #include <ATen/native/TypeProperties.h>
 #include <ATen/native/xpu/sycl/ShapeKernels.h>
 #include <ATen/native/xpu/sycl/TensorShapeKernels.h>
-#include <comm/RegisterUtils.h>
 #include <ATen/ops/as_strided_copy_native.h>
 #include <ATen/ops/as_strided_native.h>
 #include <ATen/ops/cat_native.h>
+#include <comm/RegisterUtils.h>
 
 namespace at {
 
diff --git a/src/ATen/native/xpu/XPUScalar.cpp b/src/ATen/native/xpu/XPUScalar.cpp
@@ -2,8 +2,8 @@
 #include <ATen/Dispatch_v2.h>
 #include <ATen/EmptyTensor.h>
 #include <ATen/core/Tensor.h>
-#include <comm/SYCLContext.h>
 #include <ATen/ops/_local_scalar_dense_native.h>
+#include <comm/SYCLContext.h>
 
 namespace at::native {
 
diff --git a/src/ATen/native/xpu/sycl/AdaptiveAveragePooling2dKernels.cpp b/src/ATen/native/xpu/sycl/AdaptiveAveragePooling2dKernels.cpp
@@ -301,8 +301,8 @@ struct AdaptiveAvgPool2dBwdSLMChannelsLastKernelFunctor
     index_t istartW = item.get_local_id(1) + item.get_group(1) * iW;
     index_t iendW = std::min(istartW + iW, isizeW_);
 
-    // Stride for threads, each subgroup can reuse L1 as they go. So theoretically
-    // better chance to survive cache eviction.
+    // Stride for threads, each subgroup can reuse L1 as they go. So
+    // theoretically better chance to survive cache eviction.
     for (index_t ih = istartH; ih < iendH; ih += item.get_local_range(0)) {
       index_t ostartH = START_IND_INT(ih, isizeH_, osizeH_);
       index_t oendH = END_IND_INT(ih, isizeH_, osizeH_);
diff --git a/src/ATen/native/xpu/sycl/DilatedMaxPool2d.cpp b/src/ATen/native/xpu/sycl/DilatedMaxPool2d.cpp
@@ -187,7 +187,8 @@ struct MaxPool2dChannelLastVec {
         StartW += dilationW_;
       for (int h = StartH; h < EndH; h += dilationH_) {
         for (int w = StartW; w < EndW; w += dilationW_) {
-          load_offset = batch * inputSizeH_*inputSizeW_*numPlane_ / vec_size + plane +
+          load_offset =
+              batch * inputSizeH_ * inputSizeW_ * numPlane_ / vec_size + plane +
               h * inputSizeW_ * numPlane_ / vec_size + w * numPlane_ / vec_size;
           vec_t val_vec = input_vec_[load_offset];
 #pragma unroll
@@ -511,7 +512,7 @@ struct MaxPool2dBackwardDeterministicKernelFunctor {
         dilationH,                                                  \
         dilationW,                                                  \
         stride);                                                    \
-    sycl_kernel_submit(num_wg * wg_size, wg_size, queue, kfn);      \
+    sycl_kernel_submit(num_wg* wg_size, wg_size, queue, kfn);       \
   }
 
 template <typename scalar_t, bool is_channels_last>
diff --git a/src/ATen/native/xpu/sycl/LayerNormKernels.cpp b/src/ATen/native/xpu/sycl/LayerNormKernels.cpp
@@ -599,8 +599,8 @@ void _layer_norm_kernel(
       beta.defined() ? can_vectorize(beta_data, alignment) : true;
 
   if ((std::is_same_v<T, float> || std::is_same_v<T, at::Half> ||
-       std::is_same_v<T, at::BFloat16>) &&
-      N <= static_cast<int64_t>(1ULL << std::numeric_limits<float>::digits) &&
+       std::is_same_v<T, at::BFloat16>)&&N <=
+          static_cast<int64_t>(1ULL << std::numeric_limits<float>::digits) &&
       N % num_vec_elems == 0 && can_vec_X && can_vec_Y && can_vec_gamma &&
       can_vec_beta) {
     launch_vectorized_layer_norm_kernel(
@@ -1077,7 +1077,8 @@ void _layer_norm_backward_kernel(
     int num_tile_n = (N + tile_size_n - 1) / tile_size_n;
     bool adjust_m = true;
     // for M = 64*1024, N = 1, we choose tile size (256, 16) on pvc
-    // TODO: Consider tuning the tile size selection logic (tile_size_m, tile_size_n) and occupancy calculation
+    // TODO: Consider tuning the tile size selection logic (tile_size_m,
+    // tile_size_n) and occupancy calculation
     for (auto i = 0; i < 3; i++) {
       // occupancy <= 50%
       if (num_tile_m * num_tile_n * local_size_x * SIMD /
@@ -1097,8 +1098,9 @@ void _layer_norm_backward_kernel(
       }
     }
     // tile size can be (1024,32), (512,32), (512,16), (256, 16)
-    // Modifying these parameters (num_subgroup, workgroup_size, tile_size, elements_per_thread)
-    // will alter the kernel configuration, potentially affecting performance and behavior.
+    // Modifying these parameters (num_subgroup, workgroup_size, tile_size,
+    // elements_per_thread) will alter the kernel configuration, potentially
+    // affecting performance and behavior.
     const scalar_t* dY_data = dY.const_data_ptr<scalar_t>();
     const scalar_t* X_data = X.const_data_ptr<scalar_t>();
     weight_t* dg_data =
@@ -1120,8 +1122,8 @@ void _layer_norm_backward_kernel(
       dbeta_blocks_ptr = dbeta_blocks.data_ptr<weight_t>();
     }
 
-    size_t num_workgroup =
-        std::min(num_tile_m * num_tile_n, static_cast<int>(thread_slots / local_size_x));
+    size_t num_workgroup = std::min(
+        num_tile_m * num_tile_n, static_cast<int>(thread_slots / local_size_x));
     if (dgamma.defined() && dbeta.defined()) {
       GammaBetaReduceFunctor<
           scalar_t,
diff --git a/src/xccl/FlightRecorderXCCL.cpp b/src/xccl/FlightRecorderXCCL.cpp
@@ -1,7 +1,7 @@
 #ifdef USE_C10D_XCCL
 
-#include <torch/csrc/distributed/c10d/FlightRecorderDetail.hpp>
 #include <ATen/xpu/XPUEvent.h>
+#include <torch/csrc/distributed/c10d/FlightRecorderDetail.hpp>
 #include <xccl/ProcessGroupXCCL.hpp>
 
 namespace c10d {
diff --git a/src/xccl/ProcessGroupXCCL.cpp b/src/xccl/ProcessGroupXCCL.cpp
@@ -1,7 +1,7 @@
 #ifdef USE_C10D_XCCL
 
-#include <torch/csrc/distributed/c10d/ParamCommsUtils.hpp>
 #include <torch/csrc/distributed/c10d/FlightRecorderDetail.hpp>
+#include <torch/csrc/distributed/c10d/ParamCommsUtils.hpp>
 #include <xccl/NanCheck_XPU.hpp>
 #include <xccl/ProcessGroupXCCL.hpp>
 
@@ -207,11 +207,11 @@ std::string dump_xccl_trace(
     bool includeCollectives,
     bool includeStackTraces,
     bool onlyActive) {
-    auto xcclDumpMap = std::unordered_map<
-        std::string,
-        std::unordered_map<std::string, std::string>>();
-    return FlightRecorderXCCL::get()->dump(
-        xcclDumpMap, includeCollectives, includeStackTraces, onlyActive);
+  auto xcclDumpMap = std::unordered_map<
+      std::string,
+      std::unordered_map<std::string, std::string>>();
+  return FlightRecorderXCCL::get()->dump(
+      xcclDumpMap, includeCollectives, includeStackTraces, onlyActive);
 }
 
 constexpr int64_t kSynchronizeBusyWaitMillis = 10;
@@ -317,9 +317,7 @@ bool ProcessGroupXCCL::WorkXCCL::wait(std::chrono::milliseconds timeout) {
   return true;
 }
 
-ProcessGroupXCCL::Options::Options()
-    : Backend::Options(XCCL_BACKEND_NAME) {}
-
+ProcessGroupXCCL::Options::Options() : Backend::Options(XCCL_BACKEND_NAME) {}
 
 static std::atomic<size_t> process_group_id = 0;
 
@@ -362,7 +360,8 @@ ProcessGroupXCCL::ProcessGroupXCCL(
   traceBufferSize_ = getCvarInt({"TORCH_FR_BUFFER_SIZE"}, 2000);
 
   this->setGroupUid(options_->group_name);
-  // In PGNCCL, the pg ranks are recorded on comm setup in each op, but we just do it here.
+  // In PGNCCL, the pg ranks are recorded on comm setup in each op, but we just
+  // do it here.
   const auto XcclVersion = getXcclVersion();
   FlightRecorderXCCL::get()->record_pg_ranks(
       std::make_tuple(pg_uid_, pg_desc_), groupRanks());
@@ -435,7 +434,8 @@ void ProcessGroupXCCL::setCompletedPgStatus(
   pgStatus_->lastCompletedNumelIn = work->numelIn_;
   pgStatus_->lastCompletedNumelOut = work->numelOut_;
   // To avoid complexity, we're not computing duration.
-  FlightRecorderXCCL::get()->retire_id(work->trace_id_, /*compute_duration*/false);
+  FlightRecorderXCCL::get()->retire_id(
+      work->trace_id_, /*compute_duration*/ false);
 }
 
 void ProcessGroupXCCL::setSequenceNumberForGroup() {}
@@ -768,9 +768,7 @@ c10::intrusive_ptr<Work> ProcessGroupXCCL::collective(
       c10::ListType::create(c10::TensorType::get()), devices);
   work->future_->markCompleted(at::IValue(*work->outputs_));
   work->future_->addCallback(
-      [this, work](at::ivalue::Future&) {
-          this->setCompletedPgStatus(work);
-      });
+      [this, work](at::ivalue::Future&) { this->setCompletedPgStatus(work); });
   work->blockingWait_ = blockingWait_;
 
   work->numelIn_ = 0;
@@ -881,10 +879,9 @@ c10::intrusive_ptr<Work> ProcessGroupXCCL::pointToPoint(
     work->future_ = c10::make_intrusive<at::ivalue::Future>(
         c10::ListType::create(c10::TensorType::get()), devices);
     work->future_->markCompleted(at::IValue(*work->outputs_));
-    work->future_->addCallback(
-        [this, work](at::ivalue::Future&) {
-            this->setCompletedPgStatus(work);
-        });
+    work->future_->addCallback([this, work](at::ivalue::Future&) {
+      this->setCompletedPgStatus(work);
+    });
 
     work->numelIn_ = work->numelOut_ = tensor.numel();
     setEnqueuedPgStatus(work);
diff --git a/src/xccl/ProcessGroupXCCL.hpp b/src/xccl/ProcessGroupXCCL.hpp
@@ -129,10 +129,10 @@ class TORCH_API ProcessGroupXCCL : public Backend {
   };
 
   ProcessGroupXCCL(
-    const c10::intrusive_ptr<Store>& store,
-    int rank,
-    int size,
-    c10::intrusive_ptr<Options> options = Options::create());
+      const c10::intrusive_ptr<Store>& store,
+      int rank,
+      int size,
+      c10::intrusive_ptr<Options> options = Options::create());
 
   C10_DEPRECATED ProcessGroupXCCL(
       const c10::intrusive_ptr<Store>& store,
@@ -415,7 +415,8 @@ class TORCH_API ProcessGroupXCCL : public Backend {
 
   const std::vector<uint64_t>& groupRanks() const;
   void setEnqueuedPgStatus(c10::intrusive_ptr<ProcessGroupXCCL::WorkXCCL> work);
-  void setCompletedPgStatus(c10::intrusive_ptr<ProcessGroupXCCL::WorkXCCL> work);
+  void setCompletedPgStatus(
+      c10::intrusive_ptr<ProcessGroupXCCL::WorkXCCL> work);
   bool dumpDebuggingInfo(bool includeStackTrace = true);
 
  protected:
diff --git a/src/xccl/ProcessGroupXCCLMonitor.cpp b/src/xccl/ProcessGroupXCCLMonitor.cpp
@@ -7,9 +7,8 @@ namespace c10d {
 HeartbeatMonitorXCCL::HeartbeatMonitorXCCL(ProcessGroupXCCL* pg) {
   pg_ = pg;
   coordCheckIntervalMilSec_ = getCvarInt(TORCH_XCCL_COORD_CHECK_MILSEC, 1000);
-  LOG(INFO)
-      << pg_->logPrefix() << "HeartbeatMonitor environments: "
-      << "TORCH_XCCL_COORD_CHECK_MILSEC: " << coordCheckIntervalMilSec_;
+  LOG(INFO) << pg_->logPrefix() << "HeartbeatMonitor environments: "
+            << "TORCH_XCCL_COORD_CHECK_MILSEC: " << coordCheckIntervalMilSec_;
 }
 
 void HeartbeatMonitorXCCL::stop() {
diff --git a/src/xccl/ProcessGroupXCCLMonitor.hpp b/src/xccl/ProcessGroupXCCLMonitor.hpp
@@ -8,14 +8,13 @@
 #ifdef USE_C10D_XCCL
 namespace c10d {
 
-// This definition will later be moved to a common header for ProcessGroups NCCL/Gloo/XCCL
+// This definition will later be moved to a common header for ProcessGroups
+// NCCL/Gloo/XCCL
 #if defined(__linux__)
 struct DumpPipe {
   DumpPipe(int rank) {
-    std::string fileStem =
-        getCvarString({"TORCH_FR_DEBUG_INFO_PIPE_FILE"}, "");
-    if (fileStem.empty() ||
-        getCvarInt({"TORCH_FR_BUFFER_SIZE"}, 0) <= 0) {
+    std::string fileStem = getCvarString({"TORCH_FR_DEBUG_INFO_PIPE_FILE"}, "");
+    if (fileStem.empty() || getCvarInt({"TORCH_FR_BUFFER_SIZE"}, 0) <= 0) {
       return;
     }
     TORCH_CHECK(!fileStem.empty(), "TORCH_FR_DEBUG_INFO_PIPE_FILE is empty");
@@ -33,8 +32,9 @@ struct DumpPipe {
         ", Error: ",
         std::strerror(errno));
     fd_ = open(filename.c_str(), O_RDONLY | O_NONBLOCK);
-    LOG(INFO) << "Pipe file " << filename
-              << " has been opened, write to it to trigger ProcessGroup Debug Dump.";
+    LOG(INFO)
+        << "Pipe file " << filename
+        << " has been opened, write to it to trigger ProcessGroup Debug Dump.";
     TORCH_CHECK(fd_ != -1, "Error opening named pipe ", filename);
   }
   bool shouldDump() {
@@ -89,5 +89,5 @@ class HeartbeatMonitorXCCL {
   std::thread xcclHeartbeatMonitorThread_;
   std::atomic<bool> terminateHeartbeatMonitorThread_{false};
 };
-}
+} // namespace c10d
 #endif // USE_C10D_XCCL
