codemod cuda_gpu_id to device_id (#12022)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/12022
codemod -d . --extensions h,cc,cpp,cu,py,proto,pbtxt,pb.txt,config cuda_gpu_id device_id
codemod with 'Yes to all'
Reviewed By: orionr
Differential Revision: D9986213
fbshipit-source-id: f5614a5d26078817aee8caf79a494abfd6a95ff1
diff --git a/caffe2/contrib/nccl/cuda_nccl_op_gpu.cc b/caffe2/contrib/nccl/cuda_nccl_op_gpu.cc
index ea8b349..4c5313f 100644
--- a/caffe2/contrib/nccl/cuda_nccl_op_gpu.cc
+++ b/caffe2/contrib/nccl/cuda_nccl_op_gpu.cc
@@ -11,7 +11,7 @@
// We either do an N-N op, or an N-1 op.
CAFFE_ENFORCE(op->InputSize() == op->OutputSize() || op->OutputSize() == 1);
nccl::NCCLExecution ex;
- ex.stream_gpu_id = context.cuda_gpu_id();
+ ex.stream_gpu_id = context.device_id();
ex.stream = context.cuda_stream();
ex.root = op->template GetSingleArgument<int>("root", 0);
ex.elements.resize(op->InputSize());
@@ -204,7 +204,7 @@
for (int i = 0; i < def.input().size(); ++i) {
DeviceOption dev;
dev.set_device_type(1);
- dev.set_cuda_gpu_id(i);
+ dev.set_device_id(i);
opt.push_back(dev);
}
return std::make_pair(opt, opt);
diff --git a/caffe2/contrib/nccl/nccl_ops_test.py b/caffe2/contrib/nccl/nccl_ops_test.py
index 7e8a61e..f6c22a7 100644
--- a/caffe2/contrib/nccl/nccl_ops_test.py
+++ b/caffe2/contrib/nccl/nccl_ops_test.py
@@ -21,7 +21,7 @@
def gpu_device(i):
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = i
+ device_option.device_id = i
return device_option
diff --git a/caffe2/contrib/prof/prof_dag_net.cc b/caffe2/contrib/prof/prof_dag_net.cc
index 16917dd..c867865 100644
--- a/caffe2/contrib/prof/prof_dag_net.cc
+++ b/caffe2/contrib/prof/prof_dag_net.cc
@@ -33,9 +33,9 @@
had_mismatches = true;
LOG(INFO) << "== PERFORMANCE WARNING == \n"
<< " Operator " << node.operator_->debug_def().type()
- << " expects GPU " << mismatch.second.first.cuda_gpu_id()
+ << " expects GPU " << mismatch.second.first.device_id()
<< " but tensor [" << mismatch.first << "] is on GPU "
- << mismatch.second.second.cuda_gpu_id();
+ << mismatch.second.second.device_id();
}
}
if (!had_mismatches) {
diff --git a/caffe2/contrib/tensorboard/tensorboard_exporter.py b/caffe2/contrib/tensorboard/tensorboard_exporter.py
index 93ade48..cc2c3d8 100644
--- a/caffe2/contrib/tensorboard/tensorboard_exporter.py
+++ b/caffe2/contrib/tensorboard/tensorboard_exporter.py
@@ -177,7 +177,7 @@
if device_option.device_type == caffe2_pb2.CPU:
return "/cpu:*"
if device_option.device_type == caffe2_pb2.CUDA:
- return "/gpu:{}".format(device_option.cuda_gpu_id)
+ return "/gpu:{}".format(device_option.device_id)
raise Exception("Unhandled device", device_option)
diff --git a/caffe2/contrib/warpctc/ctc_ops_test.py b/caffe2/contrib/warpctc/ctc_ops_test.py
index 25bb0a3..3b21c8b 100644
--- a/caffe2/contrib/warpctc/ctc_ops_test.py
+++ b/caffe2/contrib/warpctc/ctc_ops_test.py
@@ -79,11 +79,11 @@
def test_ctc_cost_gpu(self):
self.verify_cost(
caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA,
- cuda_gpu_id=0),
+ device_id=0),
is_test=False)
self.verify_cost(
caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA,
- cuda_gpu_id=0),
+ device_id=0),
is_test=False,
skip_input_lengths=True)
@@ -99,10 +99,10 @@
def test_ctc_forward_only_gpu(self):
self.verify_cost(
caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA,
- cuda_gpu_id=0),
+ device_id=0),
is_test=True)
self.verify_cost(
caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA,
- cuda_gpu_id=0),
+ device_id=0),
is_test=True,
skip_input_lengths=True)
diff --git a/caffe2/core/blob_gpu_test.cc b/caffe2/core/blob_gpu_test.cc
index 55eafde..8b4127e 100644
--- a/caffe2/core/blob_gpu_test.cc
+++ b/caffe2/core/blob_gpu_test.cc
@@ -195,7 +195,7 @@
}
EXPECT_TRUE(tensor_proto.has_device_detail());
EXPECT_EQ(tensor_proto.device_detail().device_type(), PROTO_CUDA);
- EXPECT_EQ(tensor_proto.device_detail().cuda_gpu_id(), gpu_id);
+ EXPECT_EQ(tensor_proto.device_detail().device_id(), gpu_id);
// Test if the restored blob is still of the same device.
blob.Reset();
EXPECT_NO_THROW(DeserializeBlob(serialized, &blob));
@@ -205,7 +205,7 @@
// Test if we force the restored blob on a different device, we
// can still get so.
blob.Reset();
- proto.mutable_tensor()->mutable_device_detail()->set_cuda_gpu_id(0);
+ proto.mutable_tensor()->mutable_device_detail()->set_device_id(0);
EXPECT_NO_THROW(DeserializeBlob(proto.SerializeAsString(), &blob));
EXPECT_TRUE(BlobIsTensorType(blob, CUDA));
EXPECT_EQ(GetGPUIDForPointer(blob.Get<TensorCUDA>().data<float>()), 0);
diff --git a/caffe2/core/context_gpu.cu b/caffe2/core/context_gpu.cu
index 1eaa579..5ffc8c6 100644
--- a/caffe2/core/context_gpu.cu
+++ b/caffe2/core/context_gpu.cu
@@ -251,7 +251,7 @@
CUDAContext::CUDAContext(const DeviceOption& option)
: gpu_id_(
- option.has_cuda_gpu_id() ? RectifyGPUID(option.cuda_gpu_id())
+ option.has_device_id() ? RectifyGPUID(option.device_id())
: CaffeCudaGetDevice()),
random_seed_(
option.has_random_seed() ? option.random_seed()
diff --git a/caffe2/core/context_gpu.h b/caffe2/core/context_gpu.h
index 5fcdb98..afb2e93 100644
--- a/caffe2/core/context_gpu.h
+++ b/caffe2/core/context_gpu.h
@@ -182,7 +182,7 @@
}
}
- inline int cuda_gpu_id() const {
+ inline int device_id() const {
return gpu_id_;
}
@@ -281,7 +281,7 @@
}
static bool IsStreamFree(const DeviceOption& option, int stream_id) {
- auto stream = CUDAContext::cuda_stream(option.cuda_gpu_id(), stream_id);
+ auto stream = CUDAContext::cuda_stream(option.device_id(), stream_id);
return cudaStreamQuery(stream) == cudaSuccess;
}
@@ -404,7 +404,7 @@
void ExtractDeviceOption(DeviceOption* device, const void* data) override {
device->set_device_type(TypeToProto(GetDeviceType()));
- device->set_cuda_gpu_id(GetGPUIDForPointer(data));
+ device->set_device_id(GetGPUIDForPointer(data));
}
protected:
diff --git a/caffe2/core/cudnn_wrappers.h b/caffe2/core/cudnn_wrappers.h
index 1bd39fa..dea138e 100644
--- a/caffe2/core/cudnn_wrappers.h
+++ b/caffe2/core/cudnn_wrappers.h
@@ -122,9 +122,9 @@
void with_cudnn_state(size_t state_idx, F&& f) {
CAFFE_ENFORCE(
state_idx < CAFFE2_COMPILE_TIME_MAX_CUDNN_STATES, "Invalid state_idx");
- auto& sync_state = cudnn_states()[context_->cuda_gpu_id()][state_idx];
+ auto& sync_state = cudnn_states()[context_->device_id()][state_idx];
- DeviceGuard dg(context_->cuda_gpu_id());
+ DeviceGuard dg(context_->device_id());
// We need to serialize execution on the CuDNNState as we can't
// allow multiple threads to race through the cudaEventRecord
@@ -132,7 +132,7 @@
// execution)
std::lock_guard<std::mutex> g(sync_state.mutex);
if (!sync_state.state.get()) {
- sync_state.state.reset(new CuDNNState(context_->cuda_gpu_id()));
+ sync_state.state.reset(new CuDNNState(context_->device_id()));
}
CHECK_NOTNULL(sync_state.state.get())->execute(context_->cuda_stream(), f);
}
diff --git a/caffe2/core/event_gpu.cc b/caffe2/core/event_gpu.cc
index 6253ca1..44aec8d 100644
--- a/caffe2/core/event_gpu.cc
+++ b/caffe2/core/event_gpu.cc
@@ -9,21 +9,21 @@
struct CudaEventWrapper {
explicit CudaEventWrapper(const DeviceOption& option)
: cuda_stream_(nullptr),
- cuda_gpu_id_(option.cuda_gpu_id()),
+ device_id_(option.device_id()),
status_(EventStatus::EVENT_INITIALIZED) {
CAFFE_ENFORCE(option.device_type(), PROTO_CUDA);
- DeviceGuard g(cuda_gpu_id_);
+ DeviceGuard g(device_id_);
CUDA_ENFORCE(cudaEventCreate(
&cuda_event_, cudaEventDefault | cudaEventDisableTiming));
}
~CudaEventWrapper() {
- DeviceGuard g(cuda_gpu_id_);
+ DeviceGuard g(device_id_);
CUDA_CHECK(cudaEventDestroy(cuda_event_));
}
cudaEvent_t cuda_event_;
cudaStream_t cuda_stream_;
- int cuda_gpu_id_;
+ int device_id_;
std::atomic<int> status_;
std::mutex mutex_recorded_;
@@ -65,12 +65,12 @@
const auto& current_device = CaffeCudaGetDevice();
CAFFE_ENFORCE_EQ(
current_device,
- wrapper->cuda_gpu_id_,
+ wrapper->device_id_,
"When you call EventRecordCUDA, your current device should be the same "
"as the device specified by the event.");
CAFFE_ENFORCE_EQ(
current_device,
- static_cast<const CUDAContext*>(context)->cuda_gpu_id());
+ static_cast<const CUDAContext*>(context)->device_id());
CUDA_ENFORCE(cudaEventRecord(
wrapper->cuda_event_,
static_cast<const CUDAContext*>(context)->cuda_stream()));
@@ -96,7 +96,7 @@
if (wrapper->status_ == EventStatus::EVENT_SCHEDULED) {
// ok, even if event is already completed and status was not yet updated
- DeviceGuard g(wrapper->cuda_gpu_id_);
+ DeviceGuard g(wrapper->device_id_);
auto cudaResult = cudaEventSynchronize(wrapper->cuda_event_);
if (cudaResult == cudaSuccess) {
wrapper->status_ = EventStatus::EVENT_SUCCESS;
@@ -127,7 +127,7 @@
if (context_stream != event_stream) {
// CAFFE_ENFORCE_EQ(
// CaffeCudaGetDevice(),
- // static_cast<const CUDAContext*>(context)->cuda_gpu_id());
+ // static_cast<const CUDAContext*>(context)->device_id());
CUDA_CHECK(cudaStreamWaitEvent(context_stream, wrapper->cuda_event_, 0));
}
}
diff --git a/caffe2/core/hip/event_hip.cc b/caffe2/core/hip/event_hip.cc
index 6f0db46..ebec9c5 100644
--- a/caffe2/core/hip/event_hip.cc
+++ b/caffe2/core/hip/event_hip.cc
@@ -138,7 +138,7 @@
{
// CAFFE_ENFORCE_EQ(
// CaffeCudaGetDevice(),
- // static_cast<const CUDAContext*>(context)->cuda_gpu_id());
+ // static_cast<const CUDAContext*>(context)->device_id());
HIP_CHECK(hipStreamWaitEvent(context_stream, wrapper->hip_event_, 0));
}
}
diff --git a/caffe2/core/memonger.cc b/caffe2/core/memonger.cc
index d9816e7..87633fa 100644
--- a/caffe2/core/memonger.cc
+++ b/caffe2/core/memonger.cc
@@ -176,7 +176,7 @@
// cuda device option but whose inputs/outputs are on CPU
if (net.op(op_index).type() == "CopyGPUToCPU") {
blob_device_[output].set_device_type(0);
- blob_device_[output].set_cuda_gpu_id(0);
+ blob_device_[output].set_device_id(0);
}
}
}
@@ -478,7 +478,7 @@
const DeviceOption& device_option) {
const DeviceOption& blob_device = blob_device_[blob_name];
if (device_option.device_type() != blob_device.device_type() ||
- device_option.cuda_gpu_id() != blob_device.cuda_gpu_id()) {
+ device_option.device_id() != blob_device.device_id()) {
return false;
}
for (const int token : req_tokens_[blob_name]) {
diff --git a/caffe2/core/net_async_base.cc b/caffe2/core/net_async_base.cc
index ce5fdbe..a694a48 100644
--- a/caffe2/core/net_async_base.cc
+++ b/caffe2/core/net_async_base.cc
@@ -157,7 +157,7 @@
numa_node_id);
return pool_getter(cpu_pools_, PROTO_CPU, numa_node_id, num_workers_);
} else if (device_option.device_type() == PROTO_CUDA) {
- auto gpu_id = device_option.cuda_gpu_id();
+ auto gpu_id = device_option.device_id();
CAFFE_ENFORCE(
gpu_id >= 0 && gpu_id < FLAGS_caffe2_net_async_max_gpus,
"Invalid GPU id: " + caffe2::to_string(gpu_id));
@@ -173,7 +173,7 @@
const auto& device_option = event(task_id).GetDeviceOption();
int stream_id = 0;
if (device_option.device_type() == PROTO_CUDA) {
- int gpu_id = device_option.cuda_gpu_id();
+ int gpu_id = device_option.device_id();
CAFFE_ENFORCE_GE(gpu_id, 0, "Invalid gpu id: " + caffe2::to_string(gpu_id));
if ((unsigned)gpu_id >= getStreamCounters().size()) {
getStreamCounters().resize(gpu_id + 1, 0);
diff --git a/caffe2/core/net_async_dag_gpu.cc b/caffe2/core/net_async_dag_gpu.cc
index 550a760..86d0b4d 100644
--- a/caffe2/core/net_async_dag_gpu.cc
+++ b/caffe2/core/net_async_dag_gpu.cc
@@ -112,7 +112,7 @@
int AsyncDAGNet::stream(const DeviceOption& device_option) {
int stream_id = 0;
if (device_option.device_type() == PROTO_CUDA) {
- int gpu_id = device_option.cuda_gpu_id();
+ int gpu_id = device_option.device_id();
CAFFE_ENFORCE_GE(gpu_id, 0, "Invalid gpu id: " + caffe2::to_string(gpu_id));
if ((unsigned)gpu_id >= stream_counters_.size()) {
stream_counters_.resize(gpu_id + 1, 0);
diff --git a/caffe2/core/net_gpu_test.cc b/caffe2/core/net_gpu_test.cc
index eaea937..fab5611 100644
--- a/caffe2/core/net_gpu_test.cc
+++ b/caffe2/core/net_gpu_test.cc
@@ -124,7 +124,7 @@
type: "NetTestDummy"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
)DOC";
diff --git a/caffe2/core/operator.cc b/caffe2/core/operator.cc
index 79be08c..8115ae3 100644
--- a/caffe2/core/operator.cc
+++ b/caffe2/core/operator.cc
@@ -649,7 +649,7 @@
&blob_device);
if (blob_device.device_type() == PROTO_CUDA &&
- blob_device.cuda_gpu_id() != op_device.cuda_gpu_id()) {
+ blob_device.device_id() != op_device.device_id()) {
mismatches[blob_name] = std::make_pair(op_device, blob_device);
} else if (
blob_device.device_type() == PROTO_HIP &&
diff --git a/caffe2/mkl/utils/mkl_memory.cc b/caffe2/mkl/utils/mkl_memory.cc
index 3f05f9c..9d4f347 100644
--- a/caffe2/mkl/utils/mkl_memory.cc
+++ b/caffe2/mkl/utils/mkl_memory.cc
@@ -26,7 +26,7 @@
const mkl::MKLMemory<T>* tc = static_cast<const mkl::MKLMemory<T>*>(c);
*capacity = tc->size() * sizeof(T);
device->set_device_type(PROTO_MKLDNN);
- device->set_cuda_gpu_id(0);
+ device->set_device_id(0);
return tc->dims();
}
diff --git a/caffe2/observers/profile_observer_gpu.cc b/caffe2/observers/profile_observer_gpu.cc
index bf4e20b..5bd9b0a 100644
--- a/caffe2/observers/profile_observer_gpu.cc
+++ b/caffe2/observers/profile_observer_gpu.cc
@@ -70,7 +70,7 @@
int device;
cudaGetDevice(&device);
- cudaSetDevice(context->cuda_gpu_id());
+ cudaSetDevice(context->device_id());
cudaEventCreate(&start_);
cudaEventRecord(start_, context->cuda_stream());
@@ -92,7 +92,7 @@
int device;
cudaGetDevice(&device);
- cudaSetDevice(context->cuda_gpu_id());
+ cudaSetDevice(context->device_id());
cudaEventCreate(&stop_);
cudaEventRecord(stop_, context->cuda_stream());
cudaEventSynchronize(stop_);
diff --git a/caffe2/onnx/backend.cc b/caffe2/onnx/backend.cc
index 2350910..8a21fa0 100644
--- a/caffe2/onnx/backend.cc
+++ b/caffe2/onnx/backend.cc
@@ -65,7 +65,7 @@
{DeviceType::CUDA, caffe2::DeviceType::CUDA}};
caffe2::DeviceOption d;
d.set_device_type(static_cast<int32_t>(m.at(onnx_device.type)));
- d.set_cuda_gpu_id(onnx_device.device_id);
+ d.set_device_id(onnx_device.device_id);
return d;
}
diff --git a/caffe2/operators/load_save_op_gpu.cc b/caffe2/operators/load_save_op_gpu.cc
index cd70e9c..8458fab 100644
--- a/caffe2/operators/load_save_op_gpu.cc
+++ b/caffe2/operators/load_save_op_gpu.cc
@@ -8,7 +8,7 @@
if (proto->has_tensor()) {
auto* device_detail = proto->mutable_tensor()->mutable_device_detail();
device_detail->set_device_type(PROTO_CUDA);
- device_detail->set_cuda_gpu_id(CaffeCudaGetDevice());
+ device_detail->set_device_id(CaffeCudaGetDevice());
}
}
diff --git a/caffe2/operators/rnn/recurrent_network_executor_gpu.cc b/caffe2/operators/rnn/recurrent_network_executor_gpu.cc
index e16e207..061f54d 100644
--- a/caffe2/operators/rnn/recurrent_network_executor_gpu.cc
+++ b/caffe2/operators/rnn/recurrent_network_executor_gpu.cc
@@ -72,11 +72,11 @@
if (gpu_id == -1 &&
rnn_op.op->device_option().device_type() ==
DeviceTypeProto::PROTO_CUDA) {
- gpu_id = rnn_op.op->device_option().cuda_gpu_id();
+ gpu_id = rnn_op.op->device_option().device_id();
} else {
CAFFE_ENFORCE(
rnn_op.op->device_option().device_type() == 0 ||
- rnn_op.op->device_option().cuda_gpu_id() == gpu_id,
+ rnn_op.op->device_option().device_id() == gpu_id,
"RNN Executor only supports ops on one GPU");
}
diff --git a/caffe2/proto/caffe2.proto b/caffe2/proto/caffe2.proto
index 7187001..21bdec2 100644
--- a/caffe2/proto/caffe2.proto
+++ b/caffe2/proto/caffe2.proto
@@ -135,7 +135,7 @@
// optional DeviceType device_type = 1 [ default = CPU ];
optional int32 device_type = 1 [ default = 0 ]; // 0 is CPU.
// [CUDA specific] the cuda gpu id.
- optional int32 cuda_gpu_id = 2;
+ optional int32 device_id = 2;
// [general] The random seed to start the device random number generator with.
optional uint32 random_seed = 3;
// [general] What node this op should execute on.
diff --git a/caffe2/python/cnn.py b/caffe2/python/cnn.py
index f927020..f9ccf92 100644
--- a/caffe2/python/cnn.py
+++ b/caffe2/python/cnn.py
@@ -236,5 +236,5 @@
def GPU(self, gpu_id=0):
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = gpu_id
+ device_option.device_id = gpu_id
return device_option
diff --git a/caffe2/python/core.py b/caffe2/python/core.py
index 6850c02..4f683da 100644
--- a/caffe2/python/core.py
+++ b/caffe2/python/core.py
@@ -84,7 +84,7 @@
def DeviceOption(
device_type,
- cuda_gpu_id=0,
+ device_id=0,
random_seed=None,
node_name=None,
numa_node_id=None,
@@ -92,7 +92,7 @@
):
option = caffe2_pb2.DeviceOption()
option.device_type = device_type
- option.cuda_gpu_id = cuda_gpu_id
+ option.device_id = device_id
if node_name is not None:
option.node_name = node_name
if random_seed is not None:
@@ -115,7 +115,7 @@
if not opt1.device_type or not opt2.device_type:
# At least one option is for CPU, check if both are for CPU.
return not opt1.device_type and not opt2.device_type
- return opt1.cuda_gpu_id == opt2.cuda_gpu_id
+ return opt1.device_id == opt2.device_id
def InferBlobDevices(net):
@@ -2111,7 +2111,7 @@
"""A convenient function to run everything on the GPU."""
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = gpu_id
+ device_option.device_id = gpu_id
self._net.device_option.CopyFrom(device_option)
if use_cudnn:
for op in self._net.op:
@@ -2286,7 +2286,7 @@
return None
if src.device_type == CUDA and dst.device_type == CUDA:
- if src.cuda_gpu_id == dst.cuda_gpu_id:
+ if src.device_id == dst.device_id:
return None
else:
def fun(net, *args, **kw):
@@ -2312,10 +2312,10 @@
def device_equal(src, dst):
'''
We are using this fucntion instead of == operator because optional-value
- comparison between empty device_options and {device_type:0, cuda_gpu_id:0}
+ comparison between empty device_options and {device_type:0, device_id:0}
returns not equal in some cases.
'''
- return src.device_type == dst.device_type and src.cuda_gpu_id == dst.cuda_gpu_id
+ return src.device_type == dst.device_type and src.device_id == dst.device_id
def update_placeholder_op_output(op, blob_to_device):
@@ -2429,7 +2429,7 @@
if device_option.device_type == CPU:
suffix = '_cpu'
elif device_option.device_type == CUDA:
- suffix = '_cuda_' + str(device_option.cuda_gpu_id)
+ suffix = '_cuda_' + str(device_option.device_id)
else:
raise RuntimeError(
"Unknown device type: {}".
diff --git a/caffe2/python/core_test.py b/caffe2/python/core_test.py
index 7120843..2f6dedb 100644
--- a/caffe2/python/core_test.py
+++ b/caffe2/python/core_test.py
@@ -83,17 +83,17 @@
# explicitly setting a device
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
op = core.CreateOperator("Relu", "x", "y", device_option=device_option)
self.assertTrue(op.HasField('device_option'))
self.assertEqual(op.device_option.device_type, caffe2_pb2.CUDA)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
with core.DeviceScope(device_option):
# from device scope
op = core.CreateOperator("Relu", "x", "y")
self.assertTrue(op.HasField('device_option'))
self.assertEqual(op.device_option.device_type, caffe2_pb2.CUDA)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
# from an overridden device option
override_device = caffe2_pb2.DeviceOption()
override_device.device_type = caffe2_pb2.CPU
@@ -109,13 +109,13 @@
def testNameAndDeviceScopeTogether(self):
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
with core.DeviceScope(device_option):
with core.NameScope("foo"):
op = core.CreateOperator("Relu", "x", "y")
self.assertTrue(op.HasField('device_option'))
self.assertEqual(op.device_option.device_type, caffe2_pb2.CUDA)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(len(op.input), 1)
self.assertEqual(op.input[0], "foo/x")
self.assertEqual(len(op.output), 1)
@@ -255,7 +255,7 @@
def testCreate(self):
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
op = core.CreateOperator(
"Ludicrous", "x", "y", name="ludicrous",
control_input="z", device_option=device_option,
@@ -271,7 +271,7 @@
self.assertEqual(op.control_input[0], "z")
self.assertTrue(op.HasField('device_option'))
self.assertEqual(op.device_option.device_type, caffe2_pb2.CUDA)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertTrue(len(op.arg), 3)
# can't guarantee ordering of kwargs, so generate a set of args
@@ -574,7 +574,7 @@
opt2 = caffe2_pb2.DeviceOption()
opt1.device_type = 0
self.assertTrue(core.device_option_equal(opt1, opt2))
- opt1.cuda_gpu_id = 5
+ opt1.device_id = 5
# opt1 still is on CPU, so the options should be equal
self.assertTrue(core.device_option_equal(opt1, opt2))
opt2.device_type = 0
@@ -649,7 +649,7 @@
def setUp(self):
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
self.cuda_option = device_option
self.cpu_option = caffe2_pb2.DeviceOption()
@@ -748,7 +748,7 @@
init_net = core.Net("init")
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
weight = init_net.XavierFill([], 'fc_w', shape=[10, 100])
bias = init_net.ConstantFill([], 'fc_b', shape=[10, ])
@@ -765,7 +765,7 @@
self.assertEqual(op.input[1], "fc_w_cuda_1")
self.assertEqual(op.input[2], "fc_b_cuda_1")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(new_net._net.op[-2].type, "CopyCPUToGPU")
self.assertEqual(new_net._net.op[0].type, "CopyCPUToGPU")
self.assertNotEqual(blob_to_device["fc_w"], device_option)
@@ -775,7 +775,7 @@
init_net = core.Net("init")
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
weight = init_net.XavierFill([], 'fc_w', shape=[10, 100])
bias = init_net.ConstantFill([], 'fc_b', shape=[10, ])
const = init_net.ConstantFill([], 'const', shape=[], value=1.)
@@ -791,12 +791,12 @@
op = nets[1]._net.op[0]
self.assertEqual(op.type, "CopyCPUToGPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.output[0], "fc_w_cuda_1")
op = nets[1]._net.op[1]
self.assertEqual(op.type, "CopyCPUToGPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.output[0], "fc_b_cuda_1")
op = nets[1]._net.op[2]
self.assertEqual(op.type, "FC")
@@ -804,7 +804,7 @@
self.assertEqual(op.input[1], "fc_w_cuda_1")
self.assertEqual(op.input[2], "fc_b_cuda_1")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
op = nets[1]._net.op[3]
self.assertEqual(op.type, "Add")
self.assertEqual(op.input[0], "fc1")
@@ -822,7 +822,7 @@
type: "CopyCPUToGPU"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -832,7 +832,7 @@
type: "CopyCPUToGPU"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -844,7 +844,7 @@
type: "FC"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -855,7 +855,7 @@
type: "Add"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
external_input: "data"
@@ -870,7 +870,7 @@
init_net = core.Net("init")
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
with core.DeviceScope(device_option):
weight = init_net.XavierFill([], 'fc_w', shape=[10, 100])
@@ -887,7 +887,7 @@
self.assertEqual(op.input[1], "fc_w")
self.assertEqual(op.input[2], "fc_b")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
"""
For reference, net.Proto() should be like:
name: ""
@@ -900,7 +900,7 @@
type: "FC"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
external_input: "data"
@@ -912,7 +912,7 @@
net = core.Net("test")
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
with core.DeviceScope(device_option):
net.Relu("data", "relu1")
@@ -920,10 +920,10 @@
with core.DeviceScope(device_option):
net.Relu("data", "relu3")
net.Relu("data", "relu4")
- device_option.cuda_gpu_id = 0
+ device_option.device_id = 0
with core.DeviceScope(device_option):
net.Relu("data", "relu5")
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
with core.DeviceScope(device_option):
net.Relu("data", "relu6")
@@ -931,12 +931,12 @@
op = new_net._net.op[0]
self.assertEqual(op.type, "CopyCPUToGPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.output[0], "data_cuda_1")
op = new_net._net.op[1]
self.assertEqual(op.type, "Relu")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.output[0], "relu1")
op = new_net._net.op[2]
self.assertEqual(op.type, "Relu")
@@ -945,7 +945,7 @@
op = new_net._net.op[3]
self.assertEqual(op.type, "Relu")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.input[0], "data_cuda_1")
self.assertEqual(op.output[0], "relu3")
op = new_net._net.op[4]
@@ -955,18 +955,18 @@
op = new_net._net.op[5]
self.assertEqual(op.type, "CopyCPUToGPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 0)
+ self.assertEqual(op.device_option.device_id, 0)
self.assertEqual(op.output[0], "data_cuda_0")
op = new_net._net.op[6]
self.assertEqual(op.type, "Relu")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 0)
+ self.assertEqual(op.device_option.device_id, 0)
self.assertEqual(op.input[0], "data_cuda_0")
self.assertEqual(op.output[0], "relu5")
op = new_net._net.op[7]
self.assertEqual(op.type, "Relu")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 1)
+ self.assertEqual(op.device_option.device_id, 1)
self.assertEqual(op.input[0], "data_cuda_1")
self.assertEqual(op.output[0], "relu6")
"""
@@ -979,7 +979,7 @@
type: "CopyCPUToGPU"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -989,7 +989,7 @@
type: "Relu"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -1005,7 +1005,7 @@
type: "Relu"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
op {
@@ -1021,7 +1021,7 @@
type: "CopyCPUToGPU"
device_option {
device_type: 1
- cuda_gpu_id: 0
+ device_id: 0
}
}
op {
@@ -1031,7 +1031,7 @@
type: "Relu"
device_option {
device_type: 1
- cuda_gpu_id: 0
+ device_id: 0
}
}
op {
@@ -1041,7 +1041,7 @@
type: "Relu"
device_option {
device_type: 1
- cuda_gpu_id: 1
+ device_id: 1
}
}
external_input: "data"
@@ -1060,7 +1060,7 @@
cpu_device[i].node_name = 'node:' + str(i)
gpu_device.append(caffe2_pb2.DeviceOption())
gpu_device[i].device_type = caffe2_pb2.CUDA
- gpu_device[i].cuda_gpu_id = 0
+ gpu_device[i].device_id = 0
gpu_device[i].node_name = 'node:' + str(i)
send_node = 'node:0'
recv_node = 'node:1'
@@ -1100,12 +1100,12 @@
op = init_net._net.op[2]
self.assertEqual(op.type, "CopyGPUToCPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 0)
+ self.assertEqual(op.device_option.device_id, 0)
self.assertEqual(op.output[0], "fc_w_cpu")
op = init_net._net.op[3]
self.assertEqual(op.type, "CopyGPUToCPU")
self.assertEqual(op.device_option.device_type, 1)
- self.assertEqual(op.device_option.cuda_gpu_id, 0)
+ self.assertEqual(op.device_option.device_id, 0)
self.assertEqual(op.output[0], "fc_b_cpu")
op = init_net._net.op[4]
self.assertEqual(op.type, placeholder_send)
@@ -1128,7 +1128,7 @@
net = core.Net("test")
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = 1
+ device_option.device_id = 1
net.Adagrad(['param', 'moment', 'grad', 'lr'], ['param', 'moment'])
with core.DeviceScope(device_option):
diff --git a/caffe2/python/data_parallel_model.py b/caffe2/python/data_parallel_model.py
index 89770dc..749c8b1 100644
--- a/caffe2/python/data_parallel_model.py
+++ b/caffe2/python/data_parallel_model.py
@@ -813,7 +813,7 @@
device_prefix = "gpu" if device.device_type == caffe2_pb2.CUDA else "cpu"
- namescope = "{}_{}/".format(device_prefix, device.cuda_gpu_id)
+ namescope = "{}_{}/".format(device_prefix, device.device_id)
for op in mnet.Proto().op:
if "RecurrentNetwork" in op.type:
raise("RecurrentNetwork conversion not yet supported")
@@ -1540,7 +1540,7 @@
continue
op_dev = op.device_option
- op_gpu = op_dev.cuda_gpu_id
+ op_gpu = op_dev.device_id
# This avoids failing on operators that are only for CPU
if op_dev.device_type != caffe2_pb2.CUDA:
@@ -1904,7 +1904,7 @@
new_ops = []
ops = {d: [] for d in range(num_devices)}
for op in orig_ops:
- ops[op.device_option.cuda_gpu_id].append(op)
+ ops[op.device_option.device_id].append(op)
for j in range(num_ops_per_dev):
tp = None
diff --git a/caffe2/python/hypothesis_test_util.py b/caffe2/python/hypothesis_test_util.py
index 5cc18f9..8470df1 100644
--- a/caffe2/python/hypothesis_test_util.py
+++ b/caffe2/python/hypothesis_test_util.py
@@ -259,7 +259,7 @@
# Include device option for each GPU
expanded_device_options = [cpu_do] + (
- [caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA, cuda_gpu_id=i)
+ [caffe2_pb2.DeviceOption(device_type=caffe2_pb2.CUDA, device_id=i)
for i in range(workspace.NumCudaDevices())]
if workspace.has_gpu_support else [])
diff --git a/caffe2/python/model_helper.py b/caffe2/python/model_helper.py
index f8e3f32..1e881d2 100644
--- a/caffe2/python/model_helper.py
+++ b/caffe2/python/model_helper.py
@@ -596,7 +596,7 @@
rename_list(step_op.output)
if device is not None:
step_op.device_option.device_type = device.device_type
- step_op.device_option.cuda_gpu_id = device.cuda_gpu_id
+ step_op.device_option.device_id = device.device_id
rename_list(arg.n.external_input)
rename_list(arg.n.external_output)
@@ -610,7 +610,7 @@
if device is not None:
op.device_option.device_type = device.device_type
- op.device_option.cuda_gpu_id = device.cuda_gpu_id
+ op.device_option.device_id = device.device_id
validate_op(op)
predict_proto.op.extend([op])
known_blobs.update(op.output)
diff --git a/caffe2/python/muji.py b/caffe2/python/muji.py
index b407f96..2f2b5ac 100644
--- a/caffe2/python/muji.py
+++ b/caffe2/python/muji.py
@@ -26,7 +26,7 @@
"""
device_option = caffe2_pb2.DeviceOption()
device_option.device_type = caffe2_pb2.CUDA
- device_option.cuda_gpu_id = gpu_id
+ device_option.device_id = gpu_id
return device_option
diff --git a/caffe2/python/net_printer.py b/caffe2/python/net_printer.py
index 4b5cddb..7583f86 100644
--- a/caffe2/python/net_printer.py
+++ b/caffe2/python/net_printer.py
@@ -268,11 +268,11 @@
def format_device_option(dev_opt):
if not dev_opt or not (
- dev_opt.device_type or dev_opt.cuda_gpu_id or dev_opt.node_name):
+ dev_opt.device_type or dev_opt.device_id or dev_opt.node_name):
return None
return call(
'DeviceOption',
- [dev_opt.device_type, dev_opt.cuda_gpu_id, "'%s'" % dev_opt.node_name])
+ [dev_opt.device_type, dev_opt.device_id, "'%s'" % dev_opt.node_name])
@Printer.register(OperatorDef)
diff --git a/caffe2/python/numa_test.py b/caffe2/python/numa_test.py
index 8d3a362..3178345 100644
--- a/caffe2/python/numa_test.py
+++ b/caffe2/python/numa_test.py
@@ -27,7 +27,7 @@
gpu_device_option = caffe2_pb2.DeviceOption()
gpu_device_option.device_type = caffe2_pb2.CUDA
- gpu_device_option.cuda_gpu_id = 0
+ gpu_device_option.device_id = 0
net.CopyCPUToGPU("output_blob_0", "output_blob_0_gpu",
device_option=gpu_device_option)
diff --git a/caffe2/python/onnx/backend_rep.py b/caffe2/python/onnx/backend_rep.py
index 8cc3f9e..5802e49 100644
--- a/caffe2/python/onnx/backend_rep.py
+++ b/caffe2/python/onnx/backend_rep.py
@@ -24,7 +24,7 @@
@property
def _name_scope(self):
if self.predict_net.device_option.device_type == caffe2_pb2.CUDA:
- return 'gpu_{}'.format(self.predict_net.device_option.cuda_gpu_id)
+ return 'gpu_{}'.format(self.predict_net.device_option.device_id)
return ''
def run(self, inputs, **kwargs):
diff --git a/caffe2/python/operator_test/load_save_test.py b/caffe2/python/operator_test/load_save_test.py
index 07f378b..b90a7f8 100644
--- a/caffe2/python/operator_test/load_save_test.py
+++ b/caffe2/python/operator_test/load_save_test.py
@@ -89,7 +89,7 @@
self.assertEqual(proto.tensor.device_detail.device_type,
device_type)
if device_type == caffe2_pb2.CUDA:
- self.assertEqual(proto.tensor.device_detail.cuda_gpu_id,
+ self.assertEqual(proto.tensor.device_detail.device_id,
gpu_id)
blobs = [str(i) for i in range(len(arrays))]
diff --git a/caffe2/python/operator_test/rnn_cell_test.py b/caffe2/python/operator_test/rnn_cell_test.py
index 9d9bb38..66ac07d 100644
--- a/caffe2/python/operator_test/rnn_cell_test.py
+++ b/caffe2/python/operator_test/rnn_cell_test.py
@@ -1216,7 +1216,7 @@
if arg.name == "step_net":
for step_op in arg.n.op:
self.assertEqual(0, step_op.device_option.device_type)
- self.assertEqual(1, step_op.device_option.cuda_gpu_id)
+ self.assertEqual(1, step_op.device_option.device_id)
elif arg.name == 'backward_step_net':
self.assertEqual(caffe2_pb2.NetDef(), arg.n)
diff --git a/caffe2/python/optimizer.py b/caffe2/python/optimizer.py
index 482d16a..5454b8c 100644
--- a/caffe2/python/optimizer.py
+++ b/caffe2/python/optimizer.py
@@ -81,7 +81,7 @@
if current_scope.device_type == caffe2_pb2.CUDA:
return self.get_gpu_blob_name(
- base_str, current_scope.cuda_gpu_id, current_scope.node_name
+ base_str, current_scope.device_id, current_scope.node_name
)
else:
return self.get_cpu_blob_name(base_str, current_scope.node_name)
@@ -277,7 +277,7 @@
# to include device information.
ONE = param_init_net.ConstantFill(
[],
- "ONE_{}_{}{}".format(dev.device_type, dev.cuda_gpu_id, dev.node_name),
+ "ONE_{}_{}{}".format(dev.device_type, dev.device_id, dev.node_name),
shape=[1],
value=1.0
)
@@ -486,12 +486,12 @@
ONE = param_init_net.ConstantFill(
[],
- "ONE_{}_{}".format(dev.device_type, dev.cuda_gpu_id),
+ "ONE_{}_{}".format(dev.device_type, dev.device_id),
shape=[1],
value=1.0
)
WD = param_init_net.ConstantFill(
- [], "wd_{}_{}".format(dev.device_type, dev.cuda_gpu_id),
+ [], "wd_{}_{}".format(dev.device_type, dev.device_id),
shape=[1], value=self.weight_decay
)
@@ -1158,7 +1158,7 @@
ONE = param_init_net.ConstantFill(
[],
- "ONE_{}_{}".format(dev.device_type, dev.cuda_gpu_id),
+ "ONE_{}_{}".format(dev.device_type, dev.device_id),
shape=[1],
value=1.0
)
diff --git a/caffe2/python/predictor/predictor_exporter_test.py b/caffe2/python/predictor/predictor_exporter_test.py
index b4c7153..ef11246 100644
--- a/caffe2/python/predictor/predictor_exporter_test.py
+++ b/caffe2/python/predictor/predictor_exporter_test.py
@@ -193,7 +193,7 @@
# check device options
for op in list(init_net.Proto().op) + list(predict_init_net.Proto().op):
- self.assertEqual(1, op.device_option.cuda_gpu_id)
+ self.assertEqual(1, op.device_option.device_id)
self.assertEqual(caffe2_pb2.CPU, op.device_option.device_type)
def test_db_fails_without_params(self):
diff --git a/caffe2/python/pybind_state_dlpack.h b/caffe2/python/pybind_state_dlpack.h
index 679152c..6db4ae4 100644
--- a/caffe2/python/pybind_state_dlpack.h
+++ b/caffe2/python/pybind_state_dlpack.h
@@ -34,7 +34,7 @@
"Unsupported device type: ",
device_option.device_type());
tensor_context.device_type = *device_type_ptr;
- tensor_context.device_id = device_option.cuda_gpu_id();
+ tensor_context.device_id = device_option.device_id();
if (tensor->size() <= 0) {
tensor->Resize(0);
@@ -87,7 +87,7 @@
int dlpack_device_id = dlTensor->ctx.device_id;
CAFFE_ENFORCE_EQ(
dlpack_device_id,
- device_option.cuda_gpu_id(),
+ device_option.device_id(),
"Expected same device id for DLPack and C2 tensors");
std::vector<int64_t> dims;
diff --git a/caffe2/utils/proto_utils.cc b/caffe2/utils/proto_utils.cc
index dc8e088..dd80282 100644
--- a/caffe2/utils/proto_utils.cc
+++ b/caffe2/utils/proto_utils.cc
@@ -30,7 +30,7 @@
case PROTO_CPU:
return option.numa_node_id();
case PROTO_CUDA:
- return option.cuda_gpu_id();
+ return option.device_id();
case PROTO_MKLDNN:
return option.numa_node_id();
case PROTO_HIP:
@@ -43,7 +43,7 @@
C10_EXPORT bool IsSameDevice(const DeviceOption& lhs, const DeviceOption& rhs) {
return (
lhs.device_type() == rhs.device_type() &&
- lhs.cuda_gpu_id() == rhs.cuda_gpu_id() &&
+ lhs.device_id() == rhs.device_id() &&
lhs.hip_gpu_id() == rhs.hip_gpu_id() &&
lhs.node_name() == rhs.node_name() &&
lhs.numa_node_id() == rhs.numa_node_id());
diff --git a/caffe2/utils/proto_utils_test.cc b/caffe2/utils/proto_utils_test.cc
index c9f37f4..5d8fb86 100644
--- a/caffe2/utils/proto_utils_test.cc
+++ b/caffe2/utils/proto_utils_test.cc
@@ -11,9 +11,9 @@
EXPECT_FALSE(IsSameDevice(a, b));
b.set_node_name("my_node");
EXPECT_TRUE(IsSameDevice(a, b));
- b.set_cuda_gpu_id(2);
+ b.set_device_id(2);
EXPECT_FALSE(IsSameDevice(a, b));
- a.set_cuda_gpu_id(2);
+ a.set_device_id(2);
EXPECT_TRUE(IsSameDevice(a, b));
a.set_device_type(DeviceTypeProto::PROTO_CUDA);
b.set_device_type(DeviceTypeProto::PROTO_CPU);
diff --git a/tools/amd_build/pyHIPIFY/cuda_to_hip_mappings.py b/tools/amd_build/pyHIPIFY/cuda_to_hip_mappings.py
index 113403f..3a98a4c 100644
--- a/tools/amd_build/pyHIPIFY/cuda_to_hip_mappings.py
+++ b/tools/amd_build/pyHIPIFY/cuda_to_hip_mappings.py
@@ -2216,7 +2216,7 @@
"CURAND_ENFORCE" :("HIPRAND_ENFORCE", API_CAFFE2),
"curandGenerateUniform" : ("hiprandGenerateUniform", API_CAFFE2),
"curand_generator" : ("hiprand_generator", API_CAFFE2),
- "cuda_gpu_id" : ("hip_gpu_id", API_CAFFE2),
+ "device_id" : ("hip_gpu_id", API_CAFFE2),
"CaffeCudaGetDevice" : ("CaffeHipGetDevice", API_CAFFE2),
}
