diff options
author | 2017-08-22 23:37:36 -0700 | |
---|---|---|
committer | 2017-08-22 23:41:55 -0700 | |
commit | 33e34ef79f66743e44ced81ed9cf182e00ad5d13 (patch) | |
tree | 2450fdfb1282bebfe4a4995757fbe4b71a2ba764 /tensorflow/stream_executor/stream_executor_pimpl.cc | |
parent | 1e849fed6032015eb25149b801de0f7be2d87026 (diff) |
[SE] Change comments/identifiers/logs from when StreamExecutor was GPUExecutor.
PiperOrigin-RevId: 166163619
Diffstat (limited to 'tensorflow/stream_executor/stream_executor_pimpl.cc')
-rw-r--r-- | tensorflow/stream_executor/stream_executor_pimpl.cc | 98 |
1 files changed, 49 insertions, 49 deletions
diff --git a/tensorflow/stream_executor/stream_executor_pimpl.cc b/tensorflow/stream_executor/stream_executor_pimpl.cc index 4cac6f6c96..e23d3ddace 100644 --- a/tensorflow/stream_executor/stream_executor_pimpl.cc +++ b/tensorflow/stream_executor/stream_executor_pimpl.cc @@ -36,7 +36,7 @@ limitations under the License. #include "tensorflow/stream_executor/stream_executor_internal.h" namespace { -bool FLAGS_check_gpu_leaks = false; +bool FLAGS_check_device_leaks = false; } // namespace namespace perftools { @@ -81,7 +81,7 @@ internal::StreamExecutorInterface *StreamExecutorImplementationFromPlatformKind( } if (factory == nullptr) { LOG(FATAL) - << "cannot create GPU executor implementation for platform kind: " + << "cannot create StreamExecutor implementation for platform kind: " << PlatformKindString(platform_kind); } return factory(plugin_config); @@ -192,7 +192,7 @@ StreamExecutor::~StreamExecutor() { << "especially if any stream is still active!"; } - if (FLAGS_check_gpu_leaks) { + if (FLAGS_check_device_leaks) { for (auto it : mem_allocs_) { LOG(INFO) << "Memory alloced at executor exit: addr: " << port::Printf("%p", it.first) @@ -499,17 +499,17 @@ bool StreamExecutor::SynchronousMemSet(DeviceMemoryBase *location, int value, return implementation_->SynchronousMemSet(location, value, size); } -bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *gpu_dst, +bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *device_dst, const void *host_src, uint64 size) { - VLOG(1) << "Called StreamExecutor::SynchronousMemcpy(gpu_dst=" - << gpu_dst->opaque() << ", host_src=" << host_src << ", size=" << size - << ") H2D" << StackTraceIfVLOG10(); + VLOG(1) << "Called StreamExecutor::SynchronousMemcpy(device_dst=" + << device_dst->opaque() << ", host_src=" << host_src + << ", size=" << size << ") H2D" << StackTraceIfVLOG10(); // Tracing overloaded methods is very difficult due to issues with type // inference on template args. Since use of these overloaded methods is // discouraged anyway, this isn't a huge deal. port::Status status = - implementation_->SynchronousMemcpy(gpu_dst, host_src, size); + implementation_->SynchronousMemcpy(device_dst, host_src, size); if (!status.ok()) { LOG(ERROR) << "synchronous memcpy: " << status; } @@ -517,29 +517,29 @@ bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *gpu_dst, } bool StreamExecutor::SynchronousMemcpy(void *host_dst, - const DeviceMemoryBase &gpu_src, + const DeviceMemoryBase &device_src, uint64 size) { VLOG(1) << "Called StreamExecutor::SynchronousMemcpy(host_dst=" << host_dst - << ", gpu_src=" << gpu_src.opaque() << ", size=" << size << ") D2H" - << StackTraceIfVLOG10(); + << ", device_src=" << device_src.opaque() << ", size=" << size + << ") D2H" << StackTraceIfVLOG10(); port::Status status = - implementation_->SynchronousMemcpy(host_dst, gpu_src, size); + implementation_->SynchronousMemcpy(host_dst, device_src, size); if (!status.ok()) { LOG(ERROR) << "synchronous memcpy: " << status; } return status.ok(); } -bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *gpu_dst, - const DeviceMemoryBase &gpu_src, +bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *device_dst, + const DeviceMemoryBase &device_src, uint64 size) { - VLOG(1) << "Called StreamExecutor::SynchronousMemcpy(gpu_dst=" - << gpu_dst->opaque() << ", gpu_src=" << gpu_src.opaque() + VLOG(1) << "Called StreamExecutor::SynchronousMemcpy(device_dst=" + << device_dst->opaque() << ", device_src=" << device_src.opaque() << ", size=" << size << ") D2D" << StackTraceIfVLOG10(); - port::Status status = - implementation_->SynchronousMemcpyDeviceToDevice(gpu_dst, gpu_src, size); + port::Status status = implementation_->SynchronousMemcpyDeviceToDevice( + device_dst, device_src, size); if (!status.ok()) { LOG(ERROR) << "synchronous memcpy: " << status; } @@ -547,48 +547,47 @@ bool StreamExecutor::SynchronousMemcpy(DeviceMemoryBase *gpu_dst, } port::Status StreamExecutor::SynchronousMemcpyD2H( - const DeviceMemoryBase &gpu_src, int64 size, void *host_dst) { - VLOG(1) << "Called StreamExecutor::SynchronousMemcpyD2H(gpu_src=" - << gpu_src.opaque() << ", size=" << size << ", host_dst=" << host_dst - << ")" << StackTraceIfVLOG10(); + const DeviceMemoryBase &device_src, int64 size, void *host_dst) { + VLOG(1) << "Called StreamExecutor::SynchronousMemcpyD2H(device_src=" + << device_src.opaque() << ", size=" << size + << ", host_dst=" << host_dst << ")" << StackTraceIfVLOG10(); port::Status result{port::Status::OK()}; - SCOPED_TRACE(TraceListener::SynchronousMemcpyD2H, - &result, gpu_src, size, host_dst); + SCOPED_TRACE(TraceListener::SynchronousMemcpyD2H, &result, device_src, size, + host_dst); port::Status status = - implementation_->SynchronousMemcpy(host_dst, gpu_src, size); + implementation_->SynchronousMemcpy(host_dst, device_src, size); if (!status.ok()) { - return port::Status{ - port::error::INTERNAL, - port::Printf( - "failed to synchronously memcpy device-to-host: GPU %p to host %p " - "size %lld: %s", - gpu_src.opaque(), host_dst, size, status.ToString().c_str())}; + return port::Status{port::error::INTERNAL, + port::Printf("failed to synchronously memcpy " + "device-to-host: device %p to host %p " + "size %lld: %s", + device_src.opaque(), host_dst, size, + status.ToString().c_str())}; } return result; } -port::Status StreamExecutor::SynchronousMemcpyH2D(const void *host_src, - int64 size, - DeviceMemoryBase *gpu_dst) { +port::Status StreamExecutor::SynchronousMemcpyH2D( + const void *host_src, int64 size, DeviceMemoryBase *device_dst) { VLOG(1) << "Called StreamExecutor::SynchronousMemcpyH2D(host_src=" << host_src - << ", size=" << size << ", gpu_dst" << gpu_dst->opaque() << ")" + << ", size=" << size << ", device_dst" << device_dst->opaque() << ")" << StackTraceIfVLOG10(); port::Status result{port::Status::OK()}; - SCOPED_TRACE(TraceListener::SynchronousMemcpyH2D, - &result, host_src, size, gpu_dst); + SCOPED_TRACE(TraceListener::SynchronousMemcpyH2D, &result, host_src, size, + device_dst); port::Status status = - implementation_->SynchronousMemcpy(gpu_dst, host_src, size); + implementation_->SynchronousMemcpy(device_dst, host_src, size); if (!status.ok()) { result = port::Status{ port::error::INTERNAL, port::Printf("failed to synchronously memcpy host-to-device: host " - "%p to GPU %p size %lld: %s", - host_src, gpu_dst->opaque(), size, + "%p to device %p size %lld: %s", + host_src, device_dst->opaque(), size, status.ToString().c_str())}; } @@ -596,20 +595,21 @@ port::Status StreamExecutor::SynchronousMemcpyH2D(const void *host_src, } bool StreamExecutor::Memcpy(Stream *stream, void *host_dst, - const DeviceMemoryBase &gpu_src, uint64 size) { - return implementation_->Memcpy(stream, host_dst, gpu_src, size); + const DeviceMemoryBase &device_src, uint64 size) { + return implementation_->Memcpy(stream, host_dst, device_src, size); } -bool StreamExecutor::Memcpy(Stream *stream, DeviceMemoryBase *gpu_dst, +bool StreamExecutor::Memcpy(Stream *stream, DeviceMemoryBase *device_dst, const void *host_src, uint64 size) { - return implementation_->Memcpy(stream, gpu_dst, host_src, size); + return implementation_->Memcpy(stream, device_dst, host_src, size); } bool StreamExecutor::MemcpyDeviceToDevice(Stream *stream, - DeviceMemoryBase *gpu_dst, - const DeviceMemoryBase &gpu_src, + DeviceMemoryBase *device_dst, + const DeviceMemoryBase &device_src, uint64 size) { - return implementation_->MemcpyDeviceToDevice(stream, gpu_dst, gpu_src, size); + return implementation_->MemcpyDeviceToDevice(stream, device_dst, device_src, + size); } bool StreamExecutor::MemZero(Stream *stream, DeviceMemoryBase *location, @@ -700,7 +700,7 @@ void StreamExecutor::EnqueueOnBackgroundThread(std::function<void()> task) { } void StreamExecutor::CreateAllocRecord(void *opaque, uint64 bytes) { - if (FLAGS_check_gpu_leaks && opaque != nullptr && bytes != 0) { + if (FLAGS_check_device_leaks && opaque != nullptr && bytes != 0) { mutex_lock lock{mu_}; mem_allocs_[opaque] = AllocRecord{ bytes, ""}; @@ -708,7 +708,7 @@ void StreamExecutor::CreateAllocRecord(void *opaque, uint64 bytes) { } void StreamExecutor::EraseAllocRecord(void *opaque) { - if (FLAGS_check_gpu_leaks && opaque != nullptr) { + if (FLAGS_check_device_leaks && opaque != nullptr) { mutex_lock lock{mu_}; if (mem_allocs_.find(opaque) == mem_allocs_.end()) { LOG(ERROR) << "Deallocating unknown pointer: " |