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

Automated g4 rollback of changelist 279842730.

Browse Source 
PiperOrigin-RevId: 281314244
Change-Id: I147be72280cc2f8b114437ec6917777a9739111d
This commit is contained in:
Gaurav Jain 2019-11-19 09:42:54 -08:00 committed by TensorFlower Gardener
parent 30240224dc
commit af594a2116
3 changed files with 7 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
tensorflow/python/eager/backprop.py
View File

@ -116,14 +116,13 @@ class _MockOp(object):
)
def _gradient_function(op_name, attr_tuple, device, num_inputs, inputs, outputs,
def _gradient_function(op_name, attr_tuple, num_inputs, inputs, outputs,
out_grads, skip_input_indices):
"""Calls the gradient function of the op.
Args:
op_name: the name of the op to be differentiated.
attr_tuple: the attrs, as a tuple.
device: the device of the op.
num_inputs: the number of inputs to the op.
inputs: inputs to the original operation.
outputs: outputs to the original operation.
@ -139,8 +138,7 @@ def _gradient_function(op_name, attr_tuple, device, num_inputs, inputs, outputs,
if grad_fn is None:
return [None] * num_inputs
with ops.device(device):
return grad_fn(mock_op, *out_grads)
return grad_fn(mock_op, *out_grads)
pywrap_tensorflow.TFE_Py_RegisterGradientFunction(_gradient_function)

19
tensorflow/python/eager/device_placement_test.py
View File

@ -18,7 +18,6 @@ from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.eager import backprop
from tensorflow.python.eager import context
from tensorflow.python.eager import def_function
from tensorflow.python.eager import remote
@ -28,7 +27,6 @@ from tensorflow.python.framework import constant_op
from tensorflow.python.framework import errors
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import variables
class SoftDevicePlacementTest(test.TestCase):
@ -88,23 +86,6 @@ class SoftDevicePlacementTest(test.TestCase):
# We don't support nested device placement right now.
self.assertIn('GPU:0', c.device)
@test_util.run_gpu_only
def testGradientPlacement(self):
with ops.device('GPU:0'):
x = variables.Variable(1.0)
with ops.device('CPU:0'):
y = variables.Variable(1.0)
with backprop.GradientTape() as tape:
with ops.device('GPU:0'):
x1 = constant_op.constant(2.0) * x
with ops.device('CPU:0'):
y1 = constant_op.constant(2.0) * y
z = x1 + y1
grads = tape.gradient(z, [x, y])
self.assertIn('GPU:0', grads[0].device)
self.assertIn('CPU:0', grads[1].device)
class ClusterPlacementTest(test.TestCase):

34
tensorflow/python/eager/pywrap_tfe_src.cc
View File

@ -3007,22 +3007,6 @@ PyObject* CopySequenceSettingIndicesToNull(
return result;
}
PyObject* DeviceFromTensorSeq(PyObject* seq) {
for (Py_ssize_t i = 0; i < PySequence_Size(seq); i++) {
PyObject* item = PySequence_ITEM(seq, i);
PyObject* dev = PyObject_GetAttrString(item, "device");
Py_DECREF(item);
if (dev) {
const char* devStr = TFE_GetPythonString(dev);
if (devStr && !string(devStr).empty()) {
return dev;
}
Py_DECREF(dev);
}
}
return Py_None;
}
PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
PyObject* results) {
std::vector<tensorflow::int64> input_ids = MakeTensorIDList(inputs);
@ -3049,11 +3033,6 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
string c_op_name = TFE_GetPythonString(op_name);
PyObject* device = DeviceFromTensorSeq(results);
if (device == Py_None) {
device = DeviceFromTensorSeq(inputs);
}
PyObject* op_outputs;
bool op_outputs_tuple_created = false;
std::pair<bool, tensorflow::gtl::FlatSet<int>>* outputs_not_required;
@ -3112,15 +3091,14 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
TapeSetRecordOperation(
op_name, inputs, results, input_ids, input_dtypes,
[op_name, attrs, device, num_inputs, op_inputs, op_outputs]() {
[op_name, attrs, num_inputs, op_inputs, op_outputs]() {
Py_INCREF(op_name);
Py_INCREF(attrs);
Py_INCREF(device);
Py_INCREF(num_inputs);
Py_INCREF(op_inputs);
Py_INCREF(op_outputs);
PyBackwardFunction* function = new PyBackwardFunction(
[op_name, attrs, device, num_inputs, op_inputs, op_outputs](
[op_name, attrs, num_inputs, op_inputs, op_outputs](
PyObject* output_grads,
const std::vector<tensorflow::int64>& unneeded_gradients) {
if (PyErr_Occurred()) {
@ -3140,8 +3118,8 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
skip_input_indices.reset(Py_None);
}
tensorflow::Safe_PyObjectPtr callback_args(Py_BuildValue(
"OOOOOOOO", op_name, attrs, device, num_inputs, op_inputs,
op_outputs, output_grads, skip_input_indices.get()));
"OOOOOOO", op_name, attrs, num_inputs, op_inputs, op_outputs,
output_grads, skip_input_indices.get()));
tensorflow::Safe_PyObjectPtr result(
PyObject_CallObject(gradient_function, callback_args.get()));
@ -3152,11 +3130,10 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
});
return function;
},
[op_name, attrs, device, num_inputs, op_inputs,
[op_name, attrs, num_inputs, op_inputs,
op_outputs](PyBackwardFunction* backward_function) {
Py_DECREF(op_name);
Py_DECREF(attrs);
Py_DECREF(device);
Py_DECREF(num_inputs);
Py_DECREF(op_inputs);
Py_DECREF(op_outputs);
@ -3166,7 +3143,6 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
forward_function);
Py_DECREF(num_inputs);
Py_DECREF(device);
if (op_outputs_tuple_created) Py_DECREF(op_outputs);
if (op_inputs_tuple_created) Py_DECREF(op_inputs);