experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Revert "fix gpu_device.cc sanity"

Browse Source 
This reverts commit 0816dc52e4.
This commit is contained in:
fo40225 2019-10-22 16:48:35 +08:00
parent bce0a57613
commit 1166a62a55
1 changed files with 29 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
tensorflow/core/common_runtime/gpu/gpu_device.cc
View File

@ -34,7 +34,6 @@ limitations under the License.
#include <vector> #include <vector>
#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
#include "tensorflow/core/common_runtime/device_factory.h" #include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/common_runtime/gpu/gpu_device.h" #include "tensorflow/core/common_runtime/gpu/gpu_device.h"
#include "tensorflow/core/common_runtime/gpu/gpu_event_mgr.h" #include "tensorflow/core/common_runtime/gpu/gpu_event_mgr.h"
@ -611,7 +610,7 @@ Status BaseGPUDevice::MaybeCopyTensorToGPU(
return err; 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()) { if (s.ok()) {
*to = std::move(*copy); *to = std::move(*copy);
} }
@ -742,8 +741,8 @@ Status ParseVisibleDeviceList(const string& visible_device_list,
if (!strings::safe_strto32(platform_gpu_id_str, &platform_gpu_id)) { if (!strings::safe_strto32(platform_gpu_id_str, &platform_gpu_id)) {
return errors::InvalidArgument( return errors::InvalidArgument(
"Could not parse entry in 'visible_device_list': '", "Could not parse entry in 'visible_device_list': '",
platform_gpu_id_str, "'. visible_device_list = ", platform_gpu_id_str,
visible_device_list); "'. visible_device_list = ", visible_device_list);
} }
if (platform_gpu_id < 0 || if (platform_gpu_id < 0 ||
platform_gpu_id >= gpu_manager->VisibleDeviceCount()) { platform_gpu_id >= gpu_manager->VisibleDeviceCount()) {
@ -1038,32 +1037,32 @@ Status BaseGPUDeviceFactory::CreateDevices(
#if GOOGLE_CUDA #if GOOGLE_CUDA
err = cudaSetDevice(platform_gpu_id.value()); err = cudaSetDevice(platform_gpu_id.value());
if (err != cudaSuccess) { if (err != cudaSuccess) {
return errors::Internal("cudaSetDevice() on GPU:", return errors::Internal(
platform_gpu_id.value(), " failed. Status: ", "cudaSetDevice() on GPU:", platform_gpu_id.value(),
cudaGetErrorString(err)); " failed. Status: ", cudaGetErrorString(err));
} }
err = cudaFree(nullptr); err = cudaFree(nullptr);
if (err != cudaSuccess) { if (err != cudaSuccess) {
return errors::Internal("CUDA runtime implicit initialization on GPU:", return errors::Internal("CUDA runtime implicit initialization on GPU:",
platform_gpu_id.value(), " failed. Status: ", platform_gpu_id.value(),
cudaGetErrorString(err)); " failed. Status: ", cudaGetErrorString(err));
} }
#elif TENSORFLOW_USE_ROCM #elif TENSORFLOW_USE_ROCM
err = hipSetDevice(platform_gpu_id.value()); err = hipSetDevice(platform_gpu_id.value());
if (err != hipSuccess) { if (err != hipSuccess) {
return errors::Internal("hipSetDevice() on GPU:", return errors::Internal(
platform_gpu_id.value(), " failed. Status: ", "hipSetDevice() on GPU:", platform_gpu_id.value(),
hipGetErrorString(err)); " failed. Status: ", hipGetErrorString(err));
} }
err = hipFree(nullptr); err = hipFree(nullptr);
if (err != hipSuccess) { if (err != hipSuccess) {
return errors::Internal("ROCm runtime implicit initialization on GPU:", return errors::Internal("ROCm runtime implicit initialization on GPU:",
platform_gpu_id.value(), " failed. Status: ", platform_gpu_id.value(),
hipGetErrorString(err)); " failed. Status: ", hipGetErrorString(err));
} }
#endif #endif
} }
// Reset to the original device. // Reset to the original device.
#if GOOGLE_CUDA #if GOOGLE_CUDA
err = cudaSetDevice(original_device); err = cudaSetDevice(original_device);
if (err != cudaSuccess) { if (err != cudaSuccess) {
@ -1174,13 +1173,15 @@ static string GetShortDeviceDescription(PlatformGpuId platform_gpu_id,
cc_minor = 0; cc_minor = 0;
} }
// LINT.IfChange // LINT.IfChange
return strings::StrCat("device: ", platform_gpu_id.value(), ", name: ", return strings::StrCat("device: ", platform_gpu_id.value(),
desc.name(), ", pci bus id: ", desc.pci_bus_id(), ", name: ", desc.name(),
", pci bus id: ", desc.pci_bus_id(),
", compute capability: ", cc_major, ".", cc_minor); ", compute capability: ", cc_major, ".", cc_minor);
// LINT.ThenChange(//tensorflow/python/platform/test.py) // LINT.ThenChange(//tensorflow/python/platform/test.py)
#elif TENSORFLOW_USE_ROCM #elif TENSORFLOW_USE_ROCM
return strings::StrCat("device: ", platform_gpu_id.value(), ", name: ", return strings::StrCat("device: ", platform_gpu_id.value(),
desc.name(), ", pci bus id: ", desc.pci_bus_id()); ", name: ", desc.name(),
", pci bus id: ", desc.pci_bus_id());
#endif #endif
} }
@ -1419,8 +1420,8 @@ struct CudaVersion {
// Initialize from version_name in the form of "3.5" // Initialize from version_name in the form of "3.5"
explicit CudaVersion(const std::string& version_name) { explicit CudaVersion(const std::string& version_name) {
size_t dot_pos = version_name.find('.'); 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); string major_str = version_name.substr(0, dot_pos);
CHECK(strings::safe_strto32(major_str, &major_part)) CHECK(strings::safe_strto32(major_str, &major_part))
<< "Illegal version name: [" << version_name << "]"; << "Illegal version name: [" << version_name << "]";
@ -1445,8 +1446,7 @@ struct CudaVersion {
}; };
std::vector<CudaVersion> supported_cuda_compute_capabilities = { std::vector<CudaVersion> supported_cuda_compute_capabilities = {
TF_CUDA_CAPABILITIES, TF_CUDA_CAPABILITIES,};
};
std::vector<CudaVersion> GetSupportedCudaComputeCapabilities() { std::vector<CudaVersion> GetSupportedCudaComputeCapabilities() {
auto cuda_caps = supported_cuda_compute_capabilities; auto cuda_caps = supported_cuda_compute_capabilities;
@ -1792,8 +1792,8 @@ void GPUKernelTracker::RecordTerminated(uint64 queued_count) {
VLOG(2) << this << " RecordTerminated queued_count=" << queued_count VLOG(2) << this << " RecordTerminated queued_count=" << queued_count
<< " first_available_=" << first_available_ << " first_available_=" << first_available_
<< " last_completed_=" << last_completed_ << " last_completed_=" << last_completed_
<< " num_pending_=" << num_pending_ << " num_pending_=" << num_pending_ << " LC="
<< " LC=" << ((last_completed_ >= 0) << ((last_completed_ >= 0)
? pending_kernels_[last_completed_].queued_count ? pending_kernels_[last_completed_].queued_count
: -1); : -1);
DCHECK_NE(first_available_, last_completed_); DCHECK_NE(first_available_, last_completed_);