From 1166a62a552ae43301b38ddfd8d6aa8aac6a2824 Mon Sep 17 00:00:00 2001 From: fo40225 Date: Tue, 22 Oct 2019 16:48:35 +0800 Subject: [PATCH] Revert "fix gpu_device.cc sanity" This reverts commit 0816dc52e439d190df5950860438e3dd05c76f41. --- .../core/common_runtime/gpu/gpu_device.cc | 58 +++++++++---------- 1 file changed, 29 insertions(+), 29 deletions(-) diff --git a/tensorflow/core/common_runtime/gpu/gpu_device.cc b/tensorflow/core/common_runtime/gpu/gpu_device.cc index 6a74d49818f..46586b018b6 100644 --- a/tensorflow/core/common_runtime/gpu/gpu_device.cc +++ b/tensorflow/core/common_runtime/gpu/gpu_device.cc @@ -34,7 +34,6 @@ limitations under the License. #include #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" - #include "tensorflow/core/common_runtime/device_factory.h" #include "tensorflow/core/common_runtime/gpu/gpu_device.h" #include "tensorflow/core/common_runtime/gpu/gpu_event_mgr.h" @@ -611,7 +610,7 @@ Status BaseGPUDevice::MaybeCopyTensorToGPU( return err; } - auto wrapped_done = [ to, copy, done = std::move(done) ](const Status& s) { + auto wrapped_done = [to, copy, done = std::move(done)](const Status& s) { if (s.ok()) { *to = std::move(*copy); } @@ -651,7 +650,7 @@ Status BaseGPUDevice::MakeTensorFromProto(const TensorProto& tensor_proto, std::list notifications; Status copy_status; auto copier = [this, &alloc_attrs, ¬ifications, ©_status]( - const Tensor& from, Tensor* to) { + const Tensor& from, Tensor* to) { // Copier isn't run in a multithreaded environment, so we don't // have to worry about the notifications list being modified in parallel. notifications.emplace_back(); @@ -742,8 +741,8 @@ Status ParseVisibleDeviceList(const string& visible_device_list, if (!strings::safe_strto32(platform_gpu_id_str, &platform_gpu_id)) { return errors::InvalidArgument( "Could not parse entry in 'visible_device_list': '", - platform_gpu_id_str, "'. visible_device_list = ", - visible_device_list); + platform_gpu_id_str, + "'. visible_device_list = ", visible_device_list); } if (platform_gpu_id < 0 || platform_gpu_id >= gpu_manager->VisibleDeviceCount()) { @@ -1038,32 +1037,32 @@ Status BaseGPUDeviceFactory::CreateDevices( #if GOOGLE_CUDA err = cudaSetDevice(platform_gpu_id.value()); if (err != cudaSuccess) { - return errors::Internal("cudaSetDevice() on GPU:", - platform_gpu_id.value(), " failed. Status: ", - cudaGetErrorString(err)); + return errors::Internal( + "cudaSetDevice() on GPU:", platform_gpu_id.value(), + " failed. Status: ", cudaGetErrorString(err)); } err = cudaFree(nullptr); if (err != cudaSuccess) { return errors::Internal("CUDA runtime implicit initialization on GPU:", - platform_gpu_id.value(), " failed. Status: ", - cudaGetErrorString(err)); + platform_gpu_id.value(), + " failed. Status: ", cudaGetErrorString(err)); } #elif TENSORFLOW_USE_ROCM err = hipSetDevice(platform_gpu_id.value()); if (err != hipSuccess) { - return errors::Internal("hipSetDevice() on GPU:", - platform_gpu_id.value(), " failed. Status: ", - hipGetErrorString(err)); + return errors::Internal( + "hipSetDevice() on GPU:", platform_gpu_id.value(), + " failed. Status: ", hipGetErrorString(err)); } err = hipFree(nullptr); if (err != hipSuccess) { return errors::Internal("ROCm runtime implicit initialization on GPU:", - platform_gpu_id.value(), " failed. Status: ", - hipGetErrorString(err)); + platform_gpu_id.value(), + " failed. Status: ", hipGetErrorString(err)); } #endif } -// Reset to the original device. + // Reset to the original device. #if GOOGLE_CUDA err = cudaSetDevice(original_device); if (err != cudaSuccess) { @@ -1174,13 +1173,15 @@ static string GetShortDeviceDescription(PlatformGpuId platform_gpu_id, cc_minor = 0; } // LINT.IfChange - return strings::StrCat("device: ", platform_gpu_id.value(), ", name: ", - desc.name(), ", pci bus id: ", desc.pci_bus_id(), + return strings::StrCat("device: ", platform_gpu_id.value(), + ", name: ", desc.name(), + ", pci bus id: ", desc.pci_bus_id(), ", compute capability: ", cc_major, ".", cc_minor); -// LINT.ThenChange(//tensorflow/python/platform/test.py) + // LINT.ThenChange(//tensorflow/python/platform/test.py) #elif TENSORFLOW_USE_ROCM - return strings::StrCat("device: ", platform_gpu_id.value(), ", name: ", - desc.name(), ", pci bus id: ", desc.pci_bus_id()); + return strings::StrCat("device: ", platform_gpu_id.value(), + ", name: ", desc.name(), + ", pci bus id: ", desc.pci_bus_id()); #endif } @@ -1419,8 +1420,8 @@ struct CudaVersion { // Initialize from version_name in the form of "3.5" explicit CudaVersion(const std::string& version_name) { size_t dot_pos = version_name.find('.'); - CHECK(dot_pos != string::npos) << "Illegal version name: [" << version_name - << "]"; + CHECK(dot_pos != string::npos) + << "Illegal version name: [" << version_name << "]"; string major_str = version_name.substr(0, dot_pos); CHECK(strings::safe_strto32(major_str, &major_part)) << "Illegal version name: [" << version_name << "]"; @@ -1445,8 +1446,7 @@ struct CudaVersion { }; std::vector supported_cuda_compute_capabilities = { - TF_CUDA_CAPABILITIES, -}; + TF_CUDA_CAPABILITIES,}; std::vector GetSupportedCudaComputeCapabilities() { auto cuda_caps = supported_cuda_compute_capabilities; @@ -1792,10 +1792,10 @@ void GPUKernelTracker::RecordTerminated(uint64 queued_count) { VLOG(2) << this << " RecordTerminated queued_count=" << queued_count << " first_available_=" << first_available_ << " last_completed_=" << last_completed_ - << " num_pending_=" << num_pending_ - << " LC=" << ((last_completed_ >= 0) - ? pending_kernels_[last_completed_].queued_count - : -1); + << " num_pending_=" << num_pending_ << " LC=" + << ((last_completed_ >= 0) + ? pending_kernels_[last_completed_].queued_count + : -1); DCHECK_NE(first_available_, last_completed_); DCHECK_GT(num_pending_, 0); // Starting just past the last completed entry, find the entry with