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

Merge changes from github.

Browse Source 
Change: 142805270
This commit is contained in:
Jonathan Hseu 2016-12-22 15:38:30 -08:00 committed by TensorFlower Gardener
parent 1e5bd8cdd6
commit bed8383c27
104 changed files with 2873 additions and 1259 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
README.md
View File

@ -33,10 +33,10 @@ and discussion.**
People who are a little more adventurous can also try our nightly binaries:
* Linux CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc1-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc1-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc1-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/))
* Linux GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc1-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc1-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc1-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/))
* Mac CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc1-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc1-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/))
* Mac GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc1-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc1-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/))
* Linux CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/))
* Linux GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/))
* Mac CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/))
* Mac GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/))
* [Android](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-android/TF_BUILD_CONTAINER_TYPE=ANDROID,TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=NO_PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=android-slave/lastSuccessfulBuild/artifact/bazel-out/local_linux/bin/tensorflow/examples/android/tensorflow_demo.apk) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-android/TF_BUILD_CONTAINER_TYPE=ANDROID,TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=NO_PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=android-slave/))
#### *Try your first TensorFlow program*

14
configure vendored
View File

@ -23,9 +23,13 @@ function bazel_clean_and_fetch() {
# TODO(pcloudy): Re-enable it after bazel clean --expunge is fixed.
if ! is_windows; then
bazel clean --expunge
# TODO(https://github.com/bazelbuild/bazel/issues/2220) Remove the nested `bazel query`.
bazel fetch $(bazel query "//tensorflow/... -//tensorflow/examples/android/...")
else
# TODO(pcloudy): Also filter out //tensorflow/examples/android/... on Windows after
# https://github.com/bazelbuild/bazel/issues/2248 is fixed.
bazel fetch //tensorflow/...
fi
# TODO(https://github.com/bazelbuild/bazel/issues/2220) Remove the nested `bazel query`.
bazel fetch $(bazel query "//tensorflow/... -//tensorflow/examples/android/...")
}
## Set up python-related environment settings
@ -280,7 +284,7 @@ while true; do
TF_CUDNN_VERSION=${BASH_REMATCH[1]}
echo "libcudnn.so resolves to libcudnn${TF_CUDNN_EXT}"
elif [[ "$REALVAL" =~ ([0-9]*).dylib ]]; then
TF_CUDNN_EXT="."${BASH_REMATCH[1]}".dylib"
TF_CUDNN_EXT=${BASH_REMATCH[1]}".dylib"
TF_CUDNN_VERSION=${BASH_REMATCH[1]}
echo "libcudnn.dylib resolves to libcudnn${TF_CUDNN_EXT}"
fi
@ -387,7 +391,7 @@ if [ "$TF_NEED_OPENCL" == "1" ]; then
while true; do
fromuser=""
if [ -z "$HOST_CXX_COMPILER" ]; then
default_cxx_host_compiler=$(which g++|| true)
default_cxx_host_compiler=$(which clang++-3.6 || true)
read -p "Please specify which C++ compiler should be used as the host C++ compiler. [Default is $default_cxx_host_compiler]: " HOST_CXX_COMPILER
fromuser="1"
if [ -z "$HOST_CXX_COMPILER" ]; then
@ -410,7 +414,7 @@ done
while true; do
fromuser=""
if [ -z "$HOST_C_COMPILER" ]; then
default_c_host_compiler=$(which gcc|| true)
default_c_host_compiler=$(which clang-3.6 || true)
read -p "Please specify which C compiler should be used as the host C compiler. [Default is $default_c_host_compiler]: " HOST_C_COMPILER
fromuser="1"
if [ -z "$HOST_C_COMPILER" ]; then

6
tensorflow/c/c_api.cc
View File

@ -81,6 +81,12 @@ extern "C" {
// --------------------------------------------------------------------------
const char* TF_Version() { return TF_VERSION_STRING; }
// --------------------------------------------------------------------------
size_t TF_DataTypeSize(TF_DataType dt) {
return static_cast<size_t>(
tensorflow::DataTypeSize(static_cast<DataType>(dt)));
}
// --------------------------------------------------------------------------
struct TF_Status {
Status status;

5
tensorflow/c/c_api.h
View File

@ -100,6 +100,11 @@ typedef enum {
TF_RESOURCE = 20,
} TF_DataType;
// TF_DataTypeSize returns the sizeof() for the underlying type corresponding
// to the given TF_DataType enum value. Returns 0 for variable length types
// (eg. TF_STRING) or on failure.
extern size_t TF_DataTypeSize(TF_DataType dt);
// --------------------------------------------------------------------------
// TF_Code holds an error code. The enum values here are identical to
// corresponding values in error_codes.proto.

6
tensorflow/c/c_api_test.cc
View File

@ -211,6 +211,12 @@ TEST(CAPI, DataTypeEnum) {
EXPECT_EQ(TF_UINT16, static_cast<TF_DataType>(tensorflow::DT_UINT16));
EXPECT_EQ(TF_COMPLEX128, static_cast<TF_DataType>(tensorflow::DT_COMPLEX128));
EXPECT_EQ(TF_HALF, static_cast<TF_DataType>(tensorflow::DT_HALF));
EXPECT_EQ(TF_DataTypeSize(TF_DOUBLE),
tensorflow::DataTypeSize(tensorflow::DT_DOUBLE));
EXPECT_EQ(TF_DataTypeSize(TF_STRING),
tensorflow::DataTypeSize(tensorflow::DT_STRING));
// Test with invalid type; should always return 0 as documented
EXPECT_EQ(TF_DataTypeSize(static_cast<TF_DataType>(0)), 0);
}
TEST(CAPI, StatusEnum) {

3
tensorflow/cc/framework/scope.cc
View File

@ -29,7 +29,8 @@ Scope::Scope(Graph* graph, Status* status, Scope::NameMap* name_map,
status_(status),
name_map_(name_map),
refiner_(refiner),
scope_used_(nullptr) {}
scope_used_(nullptr),
colocation_constraints_() {}
Scope Scope::NewRootScope() {
Graph* graph = new Graph(OpRegistry::Global());

1
tensorflow/contrib/cmake/CMakeLists.txt
View File

@ -197,7 +197,6 @@ if (tensorflow_ENABLE_GPU)
endif()
include(tf_core_cpu.cmake)
include(tf_models.cmake)
include(tf_core_ops.cmake)
include(tf_core_direct_session.cmake)
include(tf_core_kernels.cmake)

2
tensorflow/contrib/cmake/setup.py
View File

@ -26,7 +26,7 @@ from setuptools import find_packages, setup, Command
from setuptools.command.install import install as InstallCommandBase
from setuptools.dist import Distribution
_VERSION = '0.12.0-rc1-cmake-experimental'
_VERSION = '0.12.0-cmake-experimental'
REQUIRED_PACKAGES = [
'numpy >= 1.11.0',

2
tensorflow/contrib/cmake/tf_cc_ops.cmake
View File

@ -93,8 +93,6 @@ file(GLOB_RECURSE tf_cc_srcs
"${tensorflow_source_dir}/tensorflow/cc/client/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/gradients/*.h"
"${tensorflow_source_dir}/tensorflow/cc/gradients/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.h"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/training/*.h"
"${tensorflow_source_dir}/tensorflow/cc/training/*.cc"
)

4
tensorflow/contrib/cmake/tf_core_cpu.cmake
View File

@ -2,6 +2,8 @@
# tf_core_cpu library
########################################################
file(GLOB_RECURSE tf_core_cpu_srcs
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.h"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.cc"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/*.h"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/*.cc"
"${tensorflow_source_dir}/tensorflow/core/graph/*.h"
@ -10,6 +12,8 @@ file(GLOB_RECURSE tf_core_cpu_srcs
)
file(GLOB_RECURSE tf_core_cpu_exclude_srcs
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*test*.h"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*test*.cc"
"${tensorflow_source_dir}/tensorflow/core/*test*.h"
"${tensorflow_source_dir}/tensorflow/core/*test*.cc"
"${tensorflow_source_dir}/tensorflow/core/*main.cc"

1
tensorflow/contrib/cmake/tf_core_ops.cmake
View File

@ -9,7 +9,6 @@ set(tf_op_lib_names
"io_ops"
"linalg_ops"
"logging_ops"
"losses"
"math_ops"
"nn_ops"
"no_op"

21
tensorflow/contrib/cmake/tf_models.cmake
View File

@ -1,21 +0,0 @@
########################################################
# tf_models_word2vec_ops library
########################################################
file(GLOB tf_models_word2vec_ops_srcs
"${tensorflow_source_dir}/tensorflow_models/tutorials/embedding/word2vec_ops.cc"
)
add_library(tf_models_word2vec_ops OBJECT ${tf_models_word2vec_ops_srcs})
add_dependencies(tf_models_word2vec_ops tf_core_framework)
########################################################
# tf_models_word2vec_kernels library
########################################################
file(GLOB tf_models_word2vec_kernels_srcs
"${tensorflow_source_dir}/tensorflow_models/tutorials/embedding/word2vec_kernels.cc"
)
add_library(tf_models_word2vec_kernels OBJECT ${tf_models_word2vec_kernels_srcs})
add_dependencies(tf_models_word2vec_kernels tf_core_cpu)

3
tensorflow/contrib/cmake/tf_python.cmake
View File

@ -310,6 +310,8 @@ add_python_module("tensorflow/contrib/learn/python/learn/tests")
add_python_module("tensorflow/contrib/learn/python/learn/tests/dataframe")
add_python_module("tensorflow/contrib/learn/python/learn/utils")
add_python_module("tensorflow/contrib/legacy_seq2seq")
add_python_module("tensorflow/contrib/legacy_seq2seq/python")
add_python_module("tensorflow/contrib/legacy_seq2seq/python/ops")
add_python_module("tensorflow/contrib/linalg")
add_python_module("tensorflow/contrib/linalg/python")
add_python_module("tensorflow/contrib/linalg/python/ops")
@ -489,7 +491,6 @@ GENERATE_PYTHON_OP_LIB("image_ops")
GENERATE_PYTHON_OP_LIB("io_ops")
GENERATE_PYTHON_OP_LIB("linalg_ops")
GENERATE_PYTHON_OP_LIB("logging_ops")
GENERATE_PYTHON_OP_LIB("losses")
GENERATE_PYTHON_OP_LIB("nn_ops")
GENERATE_PYTHON_OP_LIB("parsing_ops")
GENERATE_PYTHON_OP_LIB("random_ops")

26
tensorflow/contrib/distributions/python/kernel_tests/student_t_test.py
View File

@ -79,6 +79,32 @@ class StudentTTest(tf.test.TestCase):
self.assertAllClose(expected_pdf, pdf_values)
self.assertAllClose(np.exp(expected_log_pdf), pdf_values)
def testStudentCDFAndLogCDF(self):
with self.test_session():
batch_size = 6
df = tf.constant([3.] * batch_size)
mu = tf.constant([7.] * batch_size)
sigma = tf.constant([8.] * batch_size)
df_v = 3.
mu_v = 7.
sigma_v = 8.
t = np.array([-2.5, 2.5, 8., 0., -1., 2.], dtype=np.float32)
student = tf.contrib.distributions.StudentT(df, mu=mu, sigma=sigma)
log_cdf = student.log_cdf(t)
self.assertEquals(log_cdf.get_shape(), (6,))
log_cdf_values = log_cdf.eval()
cdf = student.cdf(t)
self.assertEquals(cdf.get_shape(), (6,))
cdf_values = cdf.eval()
expected_log_cdf = stats.t.logcdf(t, df_v, loc=mu_v, scale=sigma_v)
expected_cdf = stats.t.cdf(t, df_v, loc=mu_v, scale=sigma_v)
self.assertAllClose(expected_log_cdf, log_cdf_values, atol=0., rtol=1e-5)
self.assertAllClose(np.log(expected_cdf), log_cdf_values, atol=0., rtol=1e-5)
self.assertAllClose(expected_cdf, cdf_values, atol=0., rtol=1e-5)
self.assertAllClose(np.exp(expected_log_cdf), cdf_values, atol=0., rtol=1e-5)
def testStudentEntropy(self):
df_v = np.array([[2., 3., 7.]]) # 1x3
mu_v = np.array([[1., -1, 0]]) # 1x3

12
tensorflow/contrib/distributions/python/ops/student_t.py
View File

@ -21,6 +21,9 @@ from __future__ import print_function
import math
import numpy as np
import tensorflow as tf
from tensorflow.contrib.distributions.python.ops import distribution
from tensorflow.contrib.distributions.python.ops import distribution
from tensorflow.contrib.distributions.python.ops import distribution_util
from tensorflow.contrib.framework.python.framework import tensor_util as contrib_tensor_util
@ -217,6 +220,15 @@ class StudentT(distribution.Distribution):
(math_ops.sqrt(self.df) * math.sqrt(math.pi) * self.sigma) *
math_ops.pow(1. + math_ops.square(y) / self.df, -(0.5 + half_df)))
def _cdf(self, x):
# we use the same notation here as in wikipedia for the
t = (x - self.mu)/self.sigma
x_t = self.df / (math_ops.square(t) + self.df)
# The cdf is defined differently for positive and negative t
positive_cdf = 1. - 0.5 * math_ops.betainc(0.5 * self.df, 0.5, x_t)
negative_cdf = 0.5 * math_ops.betainc(0.5 * self.df, 0.5, x_t)
return tf.where(tf.less(t, 0), negative_cdf, positive_cdf)
def _entropy(self):
u = array_ops.expand_dims(self.df * self._ones(), -1)
v = array_ops.expand_dims(self._ones(), -1)

2
tensorflow/contrib/framework/python/ops/arg_scope.py
View File

@ -33,7 +33,7 @@
The first call to conv2d will use predefined args:
layers.conv2d(inputs, 64, [11, 11], 4, padding='VALID', ..., scope='conv1')
The second call to Conv will overwrite padding:
The second call to conv2d will overwrite padding:
layers.conv2d(inputs, 256, [5, 5], padding='SAME', ..., scope='conv2')
Example of how to reuse an arg_scope:

9
tensorflow/contrib/ios_examples/README.md
View File

@ -41,10 +41,9 @@ cp ~/graphs/inception5h/* tensorflow/contrib/ios_examples/simple/data/
If you're hitting problems, here's a checklist of common things to investigate:
- Make sure that you've run the `download_dependencies.sh` and
`compile_ios_protobuf.sh` scripts before you run `compile_ios_tensorflow`.
(These should be called by `build_all_ios.sh` if you are using it, but check
if they have run successful.)
- Make sure that you've run the `build_all_ios.sh` script
This will run `download_dependencies.sh`,`compile_ios_protobuf.sh` and `compile_ios_tensorflow.sh`.
(check each one if they have run successful.)
- Check that you have version 7.3 of Xcode.
@ -138,4 +137,4 @@ After that, you can manually look at modifying the list of kernels
included in tensorflow/contrib/makefile/tf_op_files.txt to reduce the number of
implementations to the ones you're actually using in your own model. We're
hoping to automate this step in the future, but for now manually removing them
is the best approach.
is the best approach.

11
tensorflow/contrib/learn/python/learn/README.md
View File

@ -34,7 +34,7 @@ Optionally you can install [scikit-learn](http://scikit-learn.org/stable/) and [
### Usage
Below are a few simple examples of the API. For more examples, please see [examples](https://www.tensorflow.org/code/tensorflow/examples/skflow).
Below are a few simple examples of the API. For more examples, please see [examples](https://www.tensorflow.org/code/tensorflow/examples/learn).
General tips:
@ -210,10 +210,9 @@ and then load the reported URL.
## More examples
See the [examples folder](https://www.tensorflow.org/code/tensorflow/examples/skflow) for:
See the [examples folder](https://www.tensorflow.org/code/tensorflow/examples/learn) for:
- An easy way to handle [categorical variables](https://www.tensorflow.org/code/tensorflow/examples/skflow/text_classification.py) (words are just an example of a categorical variable)
- Text Classification: see examples for [RNN](https://www.tensorflow.org/code/tensorflow/examples/skflow/text_classification_character_rnn.py) and [CNN](https://www.tensorflow.org/code/tensorflow/examples/skflow/text_classification_character_cnn.py) on characters
- [Language modeling and text sequence to sequence](https://www.tensorflow.org/code/tensorflow/examples/skflow/language_model.py)
- [Digit recognition using a CNN](https://www.tensorflow.org/code/tensorflow/examples/skflow/digits.py)
- An easy way to handle [categorical variables](https://www.tensorflow.org/code/tensorflow/examples/learn/text_classification.py) (words are just an example of a categorical variable)
- Text Classification: see examples for [RNN](https://www.tensorflow.org/code/tensorflow/examples/learn/text_classification_character_rnn.py) and [CNN](https://www.tensorflow.org/code/tensorflow/examples/learn/text_classification_character_cnn.py) on characters
- [Digit recognition using a CNN](https://www.tensorflow.org/code/tensorflow/examples/learn/mnist.py)
- And much more!

13
tensorflow/contrib/learn/python/learn/datasets/BUILD
View File

@ -18,6 +18,7 @@ py_library(
"base.py",
"mnist.py",
"produce_small_datasets.py",
"synthetic.py",
"text_datasets.py",
],
data = [":data_csv"],
@ -76,3 +77,15 @@ py_test(
"//tensorflow/python:framework_test_lib",
],
)
py_test(
name = "synthetic_test",
size = "small",
srcs = ["synthetic_test.py"],
srcs_version = "PY2AND3",
deps = [
"//tensorflow:tensorflow_py",
"//tensorflow/contrib/learn",
"//tensorflow/python:framework_test_lib",
],
)

41
tensorflow/contrib/learn/python/learn/datasets/__init__.py
View File

@ -13,7 +13,7 @@
# limitations under the License.
# ==============================================================================
"""Module includes reference datasets and utilities to load datasets."""
"""Dataset utilities and synthetic/reference datasets."""
from __future__ import absolute_import
from __future__ import division
@ -26,6 +26,7 @@ import numpy as np
from tensorflow.contrib.learn.python.learn.datasets import base
from tensorflow.contrib.learn.python.learn.datasets import mnist
from tensorflow.contrib.learn.python.learn.datasets import synthetic
from tensorflow.contrib.learn.python.learn.datasets import text_datasets
# Export load_iris and load_boston.
@ -43,6 +44,12 @@ DATASETS = {
'dbpedia': text_datasets.load_dbpedia,
}
# List of all synthetic datasets
SYNTHETIC = {
# All of these will return ['data', 'target'] -> base.Dataset
'circles': synthetic.circles,
'spirals': synthetic.spirals
}
def load_dataset(name, size='small', test_with_fake_data=False):
"""Loads dataset by name.
@ -64,3 +71,35 @@ def load_dataset(name, size='small', test_with_fake_data=False):
return DATASETS[name](size, test_with_fake_data)
else:
return DATASETS[name]()
def make_dataset(name, n_samples=100, noise=None, seed=42, *args, **kwargs):
"""Creates binary synthetic datasets
Args:
name: str, name of the dataset to generate
n_samples: int, number of datapoints to generate
noise: float or None, standard deviation of the Gaussian noise added
seed: int or None, seed for noise
Returns:
Shuffled features and labels for given synthetic dataset of type `base.Dataset`
Raises:
ValueError: Raised if `name` not found
Note:
- This is a generic synthetic data generator - individual generators might have more parameters!
See documentation for individual parameters
- Note that the `noise` parameter uses `numpy.random.normal` and depends on `numpy`'s seed
TODO:
- Support multiclass datasets
- Need shuffling routine. Currently synthetic datasets are reshuffled to avoid train/test correlation,
but that hurts reprodusability
"""
# seed = kwargs.pop('seed', None)
if name not in SYNTHETIC:
raise ValueError('Synthetic dataset not found or not implemeted: %s' % name)
else:
return SYNTHETIC[name](n_samples=n_samples, noise=noise, seed=seed, *args, **kwargs)

202
tensorflow/contrib/learn/python/learn/datasets/synthetic.py Normal file
View File

@ -0,0 +1,202 @@
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Synthetic dataset generators."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.contrib.learn.python.learn.datasets.base import Dataset
def circles(n_samples=100, noise=None, seed=None, factor=0.8, n_classes=2, *args, **kwargs):
"""Create circles separated by some value
Args:
n_samples: int, number of datapoints to generate
noise: float or None, standard deviation of the Gaussian noise added
seed: int or None, seed for the noise
factor: float, size factor of the inner circles with respect to the outer ones
n_classes: int, number of classes to generate
Returns:
Shuffled features and labels for 'circles' synthetic dataset of type `base.Dataset`
Note:
The multi-class support might not work as expected if `noise` is enabled
TODO:
- Generation of unbalanced data
Credit goes to (under BSD 3 clause):
B. Thirion,
G. Varoquaux,
A. Gramfort,
V. Michel,
O. Grisel,
G. Louppe,
J. Nothman
"""
if seed is not None:
np.random.seed(seed)
# Algo: 1) Generate initial circle, 2) For ever class generate a smaller radius circle
linspace = np.linspace(0, 2*np.pi, n_samples // n_classes)
circ_x = np.empty(0, dtype=np.int32)
circ_y = np.empty(0, dtype=np.int32)
base_cos = np.cos(linspace)
base_sin = np.sin(linspace)
y = np.empty(0, dtype=np.int32)
for label in range(n_classes):
circ_x = np.append(circ_x, base_cos)
circ_y = np.append(circ_y, base_sin)
base_cos *= factor
base_sin *= factor
y = np.append(y, label*np.ones(n_samples // n_classes, dtype=np.int32))
# Add more points if n_samples is not divisible by n_classes (unbalanced!)
extras = n_samples % n_classes
circ_x = np.append(circ_x, np.cos(np.random.rand(extras)*2*np.pi))
circ_y = np.append(circ_y, np.sin(np.random.rand(extras)*2*np.pi))
y = np.append(y, np.zeros(extras, dtype=np.int32))
# Reshape the features/labels
X = np.vstack((circ_x, circ_y)).T
y = np.hstack(y)
# Shuffle the data
indices = np.random.permutation(range(n_samples))
if noise is not None:
X += np.random.normal(scale=noise, size=X.shape)
return Dataset(data=X[indices], target=y[indices])
def spirals(n_samples=100, noise=None, seed=None,
mode = 'archimedes',
n_loops = 2,
*args, **kwargs):
"""Create spirals
Currently only binary classification is supported for spiral generation
Args:
n_samples: int, number of datapoints to generate
noise: float or None, standard deviation of the Gaussian noise added
seed: int or None, seed for the noise
n_loops: int, number of spiral loops, doesn't play well with 'bernoulli'
mode: str, how the spiral should be generated. Current implementations:
'archimedes': a spiral with equal distances between branches
'bernoulli': logarithmic spiral with branch distances increasing
'fermat': a spiral with branch distances decreasing (sqrt)
Returns:
Shuffled features and labels for 'spirals' synthetic dataset of type `base.Dataset`
Raises:
ValueError: If the generation `mode` is not valid
TODO:
- Generation of unbalanced data
"""
n_classes = 2 # I am not sure how to make it multiclass
_modes = {
'archimedes': _archimedes_spiral,
'bernoulli': _bernoulli_spiral,
'fermat': _fermat_spiral
}
if mode is None or mode not in _modes:
raise ValueError("Cannot generate spiral with mode %s"%mode)
if seed is not None:
np.random.seed(seed)
linspace = np.linspace(0, 2*n_loops*np.pi, n_samples // n_classes)
spir_x = np.empty(0, dtype=np.int32)
spir_y = np.empty(0, dtype=np.int32)
y = np.empty(0, dtype=np.int32)
for label in range(n_classes):
base_cos, base_sin = _modes[mode](linspace, label*np.pi, *args, **kwargs)
spir_x = np.append(spir_x, base_cos)
spir_y = np.append(spir_y, base_sin)
y = np.append(y, label*np.ones(n_samples // n_classes, dtype=np.int32))
# Add more points if n_samples is not divisible by n_classes (unbalanced!)
extras = n_samples % n_classes
if extras > 0:
x_exrta, y_extra = _modes[mode](np.random.rand(extras)*2*np.pi, *args, **kwargs)
spir_x = np.append(spir_x, x_extra)
spir_y = np.append(spir_y, y_extra)
y = np.append(y, np.zeros(extras, dtype=np.int32))
# Reshape the features/labels
X = np.vstack((spir_x, spir_y)).T
y = np.hstack(y)
# Shuffle the data
indices = np.random.permutation(range(n_samples))
if noise is not None:
X += np.random.normal(scale=noise, size=X.shape)
return Dataset(data=X[indices], target=y[indices])
def _archimedes_spiral(theta, theta_offset=0., *args, **kwargs):
"""Return Archimedes spiral
Args:
theta: array-like, angles from polar coordinates to be converted
theta_offset: float, angle offset in radians (2*pi = 0)
"""
x, y = theta*np.cos(theta + theta_offset), theta*np.sin(theta + theta_offset)
x_norm = np.max(np.abs(x))
y_norm = np.max(np.abs(y))
x, y = x / x_norm, y / y_norm
return x, y
def _bernoulli_spiral(theta, theta_offset=0., *args, **kwargs):
"""Return Equiangular (Bernoulli's) spiral
Args:
theta: array-like, angles from polar coordinates to be converted
theta_offset: float, angle offset in radians (2*pi = 0)
Kwargs:
exp_scale: growth rate of the exponential
"""
exp_scale = kwargs.pop('exp_scale', 0.1)
x, y = np.exp(exp_scale*theta)*np.cos(theta + theta_offset), np.exp(exp_scale*theta)*np.sin(theta + theta_offset)
x_norm = np.max(np.abs(x))
y_norm = np.max(np.abs(y))
x, y = x / x_norm, y / y_norm
return x, y
def _fermat_spiral(theta, theta_offset=0., *args, **kwargs):
"""Return Parabolic (Fermat's) spiral
Args:
theta: array-like, angles from polar coordinates to be converted
theta_offset: float, angle offset in radians (2*pi = 0)
"""
x, y = np.sqrt(theta)*np.cos(theta + theta_offset), np.sqrt(theta)*np.sin(theta + theta_offset)
x_norm = np.max(np.abs(x))
y_norm = np.max(np.abs(y))
x, y = x / x_norm, y / y_norm
return x, y

127
tensorflow/contrib/learn/python/learn/datasets/synthetic_test.py Normal file
View File

@ -0,0 +1,127 @@
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import six
import numpy as np
from tensorflow.python.platform import test
from tensorflow.contrib.learn.python.learn import datasets
from tensorflow.contrib.learn.python.learn.datasets import synthetic
class SyntheticTest(test.TestCase):
"""Test synthetic dataset generation"""
def test_make_dataset(self):
"""Test if the synthetic routine wrapper complains about the name"""
self.assertRaises(ValueError, datasets.make_dataset, name='_non_existing_name')
def test_all_datasets_callable(self):
"""Test if all methods inside the `SYNTHETIC` are callable"""
self.assertIsInstance(datasets.SYNTHETIC, dict)
if len(datasets.SYNTHETIC) > 0:
for name, method in six.iteritems(datasets.SYNTHETIC):
self.assertTrue(callable(method))
def test_circles(self):
"""Test if the circles are generated correctly
Tests:
- return type is `Dataset`
- returned `data` shape is (n_samples, n_features)
- returned `target` shape is (n_samples,)
- set of unique classes range is [0, n_classes)
TODO:
- all points have the same radius, if no `noise` specified
"""
n_samples = 100
n_classes = 2
circ = synthetic.circles(n_samples = n_samples, noise = None, n_classes = n_classes)
self.assertIsInstance(circ, datasets.base.Dataset)
self.assertTupleEqual(circ.data.shape, (n_samples,2))
self.assertTupleEqual(circ.target.shape, (n_samples,))
self.assertSetEqual(set(circ.target), set(range(n_classes)))
def test_circles_replicable(self):
"""Test if the data generation is replicable with a specified `seed`
Tests:
- return the same value if raised with the same seed
- return different values if noise or seed is different
"""
seed = 42
noise = 0.1
circ0 = synthetic.circles(n_samples = 100, noise = noise, n_classes = 2, seed = seed)
circ1 = synthetic.circles(n_samples = 100, noise = noise, n_classes = 2, seed = seed)
np.testing.assert_array_equal(circ0.data, circ1.data)
np.testing.assert_array_equal(circ0.target, circ1.target)
circ1 = synthetic.circles(n_samples = 100, noise = noise, n_classes = 2, seed = seed+1)
self.assertRaises(AssertionError, np.testing.assert_array_equal, circ0.data, circ1.data)
self.assertRaises(AssertionError, np.testing.assert_array_equal, circ0.target, circ1.target)
circ1 = synthetic.circles(n_samples = 100, noise = noise/2., n_classes = 2, seed = seed)
self.assertRaises(AssertionError, np.testing.assert_array_equal, circ0.data, circ1.data)
def test_spirals(self):
"""Test if the circles are generated correctly
Tests:
- if mode is unknown, ValueError is raised
- return type is `Dataset`
- returned `data` shape is (n_samples, n_features)
- returned `target` shape is (n_samples,)
- set of unique classes range is [0, n_classes)
"""
self.assertRaises(ValueError, synthetic.spirals, mode='_unknown_mode_spiral_')
n_samples = 100
modes = ('archimedes', 'bernoulli', 'fermat')
for mode in modes:
spir = synthetic.spirals(n_samples = n_samples, noise = None, mode = mode)
self.assertIsInstance(spir, datasets.base.Dataset)
self.assertTupleEqual(spir.data.shape, (n_samples,2))
self.assertTupleEqual(spir.target.shape, (n_samples,))
self.assertSetEqual(set(spir.target), set(range(2)))
def test_spirals_replicable(self):
"""Test if the data generation is replicable with a specified `seed`
Tests:
- return the same value if raised with the same seed
- return different values if noise or seed is different
"""
seed = 42
noise = 0.1
modes = ('archimedes', 'bernoulli', 'fermat')
for mode in modes:
spir0 = synthetic.spirals(n_samples = 1000, noise = noise, seed = seed)
spir1 = synthetic.spirals(n_samples = 1000, noise = noise, seed = seed)
np.testing.assert_array_equal(spir0.data, spir1.data)
np.testing.assert_array_equal(spir0.target, spir1.target)
spir1 = synthetic.spirals(n_samples = 1000, noise = noise, seed = seed+1)
self.assertRaises(AssertionError, np.testing.assert_array_equal, spir0.data, spir1.data)
self.assertRaises(AssertionError, np.testing.assert_array_equal, spir0.target, spir1.target)
spir1 = synthetic.spirals(n_samples = 1000, noise = noise/2., seed = seed)
self.assertRaises(AssertionError, np.testing.assert_array_equal, spir0.data, spir1.data)
if __name__ == "__main__":
test.main()

2
tensorflow/contrib/learn/python/learn/estimators/run_config.py
View File

@ -84,7 +84,7 @@ class ClusterConfig(object):
'worker': ['host3:2222', 'host4:2222', 'host5:2222']}
os.environ['TF_CONFIG'] = json.dumps({
{'cluster': cluster,
'task_id': {'type': 'worker', 'index': 1}}})
'task': {'type': 'worker', 'index': 1}}})
config = ClusterConfig()
assert config.master == 'host4:2222'
assert config.task_id == 1

4
tensorflow/contrib/losses/python/losses/loss_ops.py
View File

@ -368,7 +368,7 @@ def softmax_cross_entropy(
scope: the scope for the operations performed in computing the loss.
Returns:
A scalar `Tensor` representing the loss value.
A scalar `Tensor` representing the mean loss value.
Raises:
ValueError: If the shape of `logits` doesn't match that of `onehot_labels`
@ -410,7 +410,7 @@ def sparse_softmax_cross_entropy(logits, labels, weights=1.0, scope=None):
scope: the scope for the operations performed in computing the loss.
Returns:
A scalar `Tensor` representing the loss value.
A scalar `Tensor` representing the mean loss value.
Raises:
ValueError: If the shapes of `logits`, `labels`, and `weights` are

32
tensorflow/core/common_runtime/sycl/sycl_device.cc
View File

@ -23,7 +23,8 @@ limitations under the License.
namespace tensorflow {
static std::unordered_set<SYCLDevice *> live_devices;
static std::unordered_set<SYCLDevice*> live_devices;
static bool first_time = true;
void ShutdownSycl() {
for (auto device : live_devices) {
@ -31,7 +32,6 @@ void ShutdownSycl() {
}
live_devices.clear();
}
bool first_time = true;
void SYCLDevice::RegisterDevice() {
if (first_time) {
@ -44,17 +44,27 @@ void SYCLDevice::RegisterDevice() {
SYCLDevice::~SYCLDevice() {
device_context_->Unref();
sycl_allocator_->EnterLameDuckMode();
delete sycl_device_;
delete sycl_queue_;
if (sycl_device_) {
sycl_device_->synchronize();
delete sycl_device_;
}
if (sycl_queue_) {
delete sycl_queue_;
}
live_devices.erase(this);
}
void SYCLDevice::EnterLameDuckMode() {
sycl_allocator_->EnterLameDuckMode();
delete sycl_device_;
sycl_device_ = nullptr;
delete sycl_queue_;
sycl_queue_ = nullptr;
if (sycl_device_) {
sycl_device_->synchronize();
delete sycl_device_;
sycl_device_ = nullptr;
}
if (sycl_queue_) {
delete sycl_queue_;
sycl_queue_ = nullptr;
}
}
void SYCLDevice::Compute(OpKernel *op_kernel, OpKernelContext *context) {
@ -110,7 +120,11 @@ Status SYCLDevice::FillContextMap(const Graph *graph,
Status SYCLDevice::Sync() {
sycl_device_->synchronize();
return Status::OK();
if (sycl_device_->ok()) {
return Status::OK();
} else {
return errors::Internal("Unknown error detected on device ", name());
}
}
} // namespace tensorflow

2
tensorflow/core/common_runtime/sycl/sycl_device_context.cc
View File

@ -95,6 +95,7 @@ void SYCLDeviceContext::CopyCPUTensorToDevice(const Tensor *cpu_tensor,
assert(false && "unsupported type");
}
}
device->eigen_sycl_device()->synchronize();
done(Status::OK());
}
@ -172,6 +173,7 @@ void SYCLDeviceContext::CopyDeviceTensorToCPU(const Tensor *device_tensor,
assert(false && "unsupported type");
}
}
device->eigen_sycl_device()->synchronize();
done(Status::OK());
}

18
tensorflow/core/kernels/aggregate_ops.cc
View File

@ -28,6 +28,9 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
template <typename Device, typename T>
class AddNOp : public OpKernel {
@ -152,6 +155,21 @@ REGISTER_KERNEL_BUILDER(Name("AddN")
AddNOp<CPUDevice, int32>);
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
REGISTER_ADDN(float, SYCL);
REGISTER_ADDN(double, SYCL);
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("AddN")
.Device(DEVICE_SYCL)
.TypeConstraint<int32>("T")
.HostMemory("inputs")
.HostMemory("sum"),
AddNOp<CPUDevice, int32>);
#endif // TENSORFLOW_USE_SYCL
#undef REGISTER_ADDN
} // namespace tensorflow

113
tensorflow/core/kernels/aggregate_ops_cpu.h
View File

@ -23,6 +23,10 @@ limitations under the License.
typedef Eigen::ThreadPoolDevice CPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
namespace tensorflow {
// Partial specializations for a CPUDevice, that uses the Eigen implementation
@ -133,6 +137,115 @@ struct Add9Functor<CPUDevice, T> {
}
};
#ifdef TENSORFLOW_USE_SYCL
// Partial specializations for a SYCLDevice, that uses the Eigen implementation
// from AddNEigenImpl.
template <typename T>
struct Add2Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2) {
Add2EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2);
}
};
template <typename T>
struct Add3Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3) {
Add3EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3);
}
};
template <typename T>
struct Add4Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4) {
Add4EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4);
}
};
template <typename T>
struct Add5Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5) {
Add5EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5);
}
};
template <typename T>
struct Add6Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5,
typename TTypes<T>::ConstFlat in6) {
Add6EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6);
}
};
template <typename T>
struct Add7Functor<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1,
typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3,
typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5,
typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7) {
Add7EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7);
}
};
template <typename T>
struct Add8Functor<SYCLDevice, T> {
void operator()(
const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) {
Add8EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8);
}
};
template <typename T>
struct Add8pFunctor<SYCLDevice, T> {
void operator()(
const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8) {
Add8pEigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8);
}
};
template <typename T>
struct Add9Functor<SYCLDevice, T> {
void operator()(
const SYCLDevice& d, typename TTypes<T>::Flat out,
typename TTypes<T>::ConstFlat in1, typename TTypes<T>::ConstFlat in2,
typename TTypes<T>::ConstFlat in3, typename TTypes<T>::ConstFlat in4,
typename TTypes<T>::ConstFlat in5, typename TTypes<T>::ConstFlat in6,
typename TTypes<T>::ConstFlat in7, typename TTypes<T>::ConstFlat in8,
typename TTypes<T>::ConstFlat in9) {
Add9EigenImpl<SYCLDevice, T>::Compute(d, out, in1, in2, in3, in4, in5, in6,
in7, in8, in9);
}
};
#endif // TENSORFLOW_USE_SYCL
} // namespace functor
} // namespace tensorflow

5
tensorflow/core/kernels/constant_op.cc
View File

@ -305,4 +305,9 @@ REGISTER_KERNEL_BUILDER(Name("Placeholder").Device(DEVICE_GPU), PlaceholderOp);
REGISTER_KERNEL_BUILDER(Name("PlaceholderV2").Device(DEVICE_GPU),
PlaceholderOp);
#if TENSORFLOW_USE_SYCL
REGISTER_KERNEL_BUILDER(Name("Placeholder").Device(DEVICE_SYCL), PlaceholderOp);
REGISTER_KERNEL_BUILDER(Name("PlaceholderV2").Device(DEVICE_SYCL),
PlaceholderOp);
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

112
tensorflow/core/kernels/control_flow_ops.cc
View File

@ -121,8 +121,20 @@ REGISTER_GPU_HOST_REF_KERNEL(string);
SwitchOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#define REGISTER_SYCL_REF_SWITCH(type) \
REGISTER_KERNEL_BUILDER(Name("RefSwitch") \
.Device(DEVICE_SYCL) \
.HostMemory("pred") \
.TypeConstraint<type>("T"), \
SwitchOp)
REGISTER_SYCL_REF_SWITCH(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_REF_SWITCH);
#undef REGISTER_SYCL_KERNEL
#endif
#undef REGISTER_SYCL_REF_SWITCH
#endif // TENSORFLOW_USE_SYCL
class RefSelectOp : public OpKernel {
public:
@ -230,8 +242,18 @@ REGISTER_GPU_REF_KERNEL(bool);
MergeOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#define REGISTER_SYCL_REF_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("RefMerge") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.HostMemory("value_index"), \
MergeOp)
REGISTER_SYCL_REF_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_REF_KERNEL);
#undef REGISTER_SYCL_KERNEL
#endif
#undef REGISTER_SYCL_REF_KERNEL
#endif // TENSORFLOW_USE_SYCL
// Special GPU kernels for int32 and string.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
@ -289,7 +311,15 @@ REGISTER_GPU_REF_KERNEL(bool);
Name("Enter").Device(DEVICE_SYCL).TypeConstraint<type>("T"), EnterOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#define REGISTER_SYCL_REF_KERNEL(type) \
REGISTER_KERNEL_BUILDER( \
Name("RefEnter").Device(DEVICE_SYCL).TypeConstraint<type>("T"), EnterOp)
REGISTER_SYCL_REF_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_REF_KERNEL);
#undef REGISTER_SYCL_KERNEL
#undef REGISTER_SYCL_REF_KERNEL
#endif
// Special GPU kernels for int32 and string.
@ -349,8 +379,37 @@ REGISTER_GPU_KERNEL(bool);
Name("Exit").Device(DEVICE_SYCL).TypeConstraint<type>("T"), ExitOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#define REGISTER_SYCL_REF_KERNEL(type) \
REGISTER_KERNEL_BUILDER( \
Name("RefExit").Device(DEVICE_SYCL).TypeConstraint<type>("T"), ExitOp)
REGISTER_SYCL_REF_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_REF_KERNEL);
#undef REGISTER_SYCL_KERNEL
#endif
#undef REGISTER_SYCL_REF_KERNEL
// Special GPU kernels for int32 and string.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
#define REGISTER_SYCL_HOST_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("Exit") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
ExitOp); \
REGISTER_KERNEL_BUILDER(Name("RefExit") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
ExitOp)
REGISTER_SYCL_HOST_KERNEL(int32);
REGISTER_SYCL_HOST_KERNEL(string);
#undef REGISTER_SYCL_HOST_KERNEL
#endif // TENSORFLOW_USE_SYCL
// Special GPU kernels for int32 and string.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
@ -432,8 +491,39 @@ REGISTER_GPU_HOST_KERNEL(string);
NextIterationOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#define REGISTER_SYCL_REF_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("RefNextIteration") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
NextIterationOp)
REGISTER_SYCL_REF_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_REF_KERNEL);
#undef REGISTER_SYCL_KERNEL
#endif
#undef REGISTER_SYCL_REF_KERNEL
// Special GPU kernels for int32 and string.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
#define REGISTER_SYCL_HOST_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("NextIteration") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
NextIterationOp); \
REGISTER_KERNEL_BUILDER(Name("RefNextIteration") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
NextIterationOp)
REGISTER_SYCL_HOST_KERNEL(int32);
REGISTER_SYCL_HOST_KERNEL(string);
#undef REGISTER_SYCL_HOST_KERNEL
#endif // TENSORFLOW_USE_SYCL
// A LoopCond op has one input and one output. The input is a boolean
// scalar representing the taken branches of the "pivot" Switch that
@ -461,6 +551,14 @@ REGISTER_KERNEL_BUILDER(Name("LoopCond")
.HostMemory("output"),
LoopCondOp);
#ifdef TENSORFLOW_USE_SYCL
REGISTER_KERNEL_BUILDER(Name("LoopCond")
.Device(DEVICE_SYCL)
.HostMemory("input")
.HostMemory("output"),
LoopCondOp);
#endif // TENSORFLOW_USE_SYCL
// ControlTrigger kernels
REGISTER_KERNEL_BUILDER(Name("ControlTrigger").Device(DEVICE_CPU),
ControlTriggerOp);
@ -468,6 +566,11 @@ REGISTER_KERNEL_BUILDER(Name("ControlTrigger").Device(DEVICE_CPU),
REGISTER_KERNEL_BUILDER(Name("ControlTrigger").Device(DEVICE_GPU),
ControlTriggerOp);
#ifdef TENSORFLOW_USE_SYCL
REGISTER_KERNEL_BUILDER(Name("ControlTrigger").Device(DEVICE_SYCL),
ControlTriggerOp);
#endif // TENSORFLOW_USE_SYCL
// When called, abort op will abort the current process. This can be used to
// abort remote PSs when needed.
class AbortOp : public OpKernel {
@ -493,4 +596,5 @@ class AbortOp : public OpKernel {
};
REGISTER_KERNEL_BUILDER(Name("Abort").Device(DEVICE_CPU), AbortOp);
} // namespace tensorflow

1
tensorflow/core/kernels/cwise_op_exp.cc
View File

@ -27,6 +27,7 @@ REGISTER5(UnaryOp, CPU, "Exp", functor::exp, float, Eigen::half, double,
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::exp<TYPE>>);
REGISTER_SYCL_KERNEL(float);
REGISTER_SYCL_KERNEL(double);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL

2
tensorflow/core/kernels/cwise_op_floor_div.cc
View File

@ -27,7 +27,7 @@ REGISTER3(BinaryOp, CPU, "FloorDiv", functor::floor_div_real, float,
Name("FloorDiv") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::floor_div<TYPE>>);
BinaryOp<SYCLDevice, functor::floor_div_real<TYPE>>);
REGISTER_SYCL_KERNEL(float)
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL

12
tensorflow/core/kernels/cwise_op_neg.cc
View File

@ -27,6 +27,18 @@ REGISTER7(UnaryOp, CPU, "Neg", functor::neg, float, Eigen::half, double, int32,
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::neg<TYPE>>);
REGISTER_SYCL_KERNEL(float);
REGISTER_SYCL_KERNEL(double);
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("Neg")
.Device(DEVICE_SYCL)
.HostMemory("x")
.HostMemory("y")
.TypeConstraint<int32>("T"),
UnaryOp<CPUDevice, functor::neg<int32>>);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL

12
tensorflow/core/kernels/cwise_op_sub.cc
View File

@ -32,6 +32,18 @@ REGISTER(BinaryOp, CPU, "Sub", functor::sub, int32);
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::sub<TYPE>>);
REGISTER_SYCL_KERNEL(float);
REGISTER_SYCL_KERNEL(double);
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("Sub")
.Device(DEVICE_SYCL)
.HostMemory("x")
.HostMemory("y")
.HostMemory("z")
.TypeConstraint<int32>("T"),
BinaryOp<CPUDevice, functor::sub<int32>>);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA

24
tensorflow/core/kernels/cwise_ops_sycl_common.h
View File

@ -31,14 +31,6 @@ namespace functor {
typedef Eigen::SyclDevice SYCLDevice;
template <typename Index, int N> Eigen::array<Index, N> GenerateArrayOfOnes() {
Eigen::array<Index, N> result;
for (int i = 0; i < N; ++i) {
result[i] = 1;
}
return result;
}
template <typename OUT, typename RHS>
void Assign(const SYCLDevice& d, OUT out, RHS rhs) {
out.device(d) = rhs;
@ -67,11 +59,9 @@ struct BinaryFunctor<SYCLDevice, Functor, NDIMS, has_errors> {
typename Functor::tin_type in, bool* error) {
typedef typename Functor::func Binary;
constexpr int NumDims = Functor::tin_type::NumDimensions;
typedef typename Functor::tin_type::Scalar T;
typedef typename Functor::tin_type::Index Index;
Eigen::array<Index, NumDims> scalar_dim = GenerateArrayOfOnes<Index, NumDims>();
Eigen::TensorMap<Eigen::Tensor<T, NumDims, Eigen::RowMajor>> tmp(scalar.data(), scalar_dim);
out.device(d) = tmp.broadcast(in.dimensions()).binaryExpr(in, Binary());
static_assert(NumDims == 1, "Unexpected size");
Eigen::Sizes<1> scalar_dim;
out.device(d) = scalar.reshape(scalar_dim).broadcast(in.dimensions()).binaryExpr(in, Binary());
}
void Right(const SYCLDevice& d, typename Functor::tout_type out,
@ -79,11 +69,9 @@ struct BinaryFunctor<SYCLDevice, Functor, NDIMS, has_errors> {
typename Functor::tscalar_type scalar, bool* error) {
typedef typename Functor::func Binary;
constexpr int NumDims = Functor::tin_type::NumDimensions;
typedef typename Functor::tin_type::Scalar T;
typedef typename Functor::tin_type::Index Index;
Eigen::array<Index, NumDims> scalar_dim = GenerateArrayOfOnes<Index, NumDims>();
Eigen::TensorMap<Eigen::Tensor<T, NumDims, Eigen::RowMajor>> tmp(scalar.data(), scalar_dim);
out.device(d) = in.binaryExpr(tmp.broadcast(in.dimensions()), Binary());
static_assert(NumDims == 1, "Unexpected size");
Eigen::Sizes<1> scalar_dim;
out.device(d) = in.binaryExpr(scalar.reshape(scalar_dim).broadcast(in.dimensions()), Binary());
}
void BCast(const SYCLDevice& d,

1
tensorflow/core/kernels/debug_ops.h
View File

@ -188,7 +188,6 @@ class DebugNumericSummaryOp : public OpKernel {
const T* input_flat = input.template flat<T>().data();
element_count = input_shape.num_elements();
const double element_count_double = static_cast<double>(element_count);
for (int64 i = 0; i < element_count; ++i) {
T x = input_flat[i];
if (Eigen::numext::isnan(x)) {

25
tensorflow/core/kernels/pack_op.cc
View File

@ -34,6 +34,9 @@ typedef Eigen::ThreadPoolDevice CPUDevice;
#if GOOGLE_CUDA
typedef Eigen::GpuDevice GPUDevice;
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
// --------------------------------------------------------------------------
template <typename Device, typename T>
@ -156,4 +159,26 @@ REGISTER_KERNEL_BUILDER(Name("Pack")
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL(type) \
REGISTER_KERNEL_BUILDER( \
Name("Pack").Device(DEVICE_SYCL).TypeConstraint<type>("T"), \
PackOp<SYCLDevice, type>)
REGISTER_SYCL(float);
#undef REGISTER_SYCL
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("Pack")
.Device(DEVICE_SYCL)
.HostMemory("values")
.HostMemory("output")
.TypeConstraint<int32>("T"),
PackOp<CPUDevice, int32>);
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

29
tensorflow/core/kernels/reduction_ops_common.h
View File

@ -40,6 +40,9 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
template <typename Device>
struct Constants {
@ -60,13 +63,16 @@ struct Constants {
};
#if defined(EIGEN_HAS_INDEX_LIST)
template <>
struct Constants<CPUDevice> {
struct ConstantsBase {
const Eigen::IndexList<Eigen::type2index<0>> kZero;
const Eigen::IndexList<Eigen::type2index<1>> kOne;
const Eigen::IndexList<Eigen::type2index<0>, Eigen::type2index<2>> kZeroTwo;
};
#endif
template<> struct Constants<CPUDevice> : ConstantsBase{};
#ifdef TENSORFLOW_USE_SYCL
template<> struct Constants<SYCLDevice> : ConstantsBase{};
#endif // TENSORFLOW_USE_SYCL
#endif // EIGEN_HAS_INDEX_LIST
class ReductionHelper {
public:
@ -239,22 +245,31 @@ class ReductionOp : public OpKernel {
namespace functor {
template <typename Reducer>
struct ReduceFunctor<CPUDevice, Reducer> {
template <typename Device, typename Reducer>
struct ReduceFunctorBase {
template <typename OUT_T, typename IN_T, typename ReductionAxes>
static void Reduce(const CPUDevice& d, OUT_T out, IN_T in,
static void Reduce(const Device& d, OUT_T out, IN_T in,
const ReductionAxes& reduction_axes,
const Reducer& reducer) {
ReduceEigenImpl(d, out, in, reduction_axes, reducer);
}
template <typename OUT_T>
static void FillIdentity(const CPUDevice& d, OUT_T out,
static void FillIdentity(const Device& d, OUT_T out,
const Reducer& reducer) {
FillIdentityEigenImpl(d, out, reducer);
}
};
template <typename Reducer>
struct ReduceFunctor<CPUDevice, Reducer>
: ReduceFunctorBase<CPUDevice, Reducer>{};
#if TENSORFLOW_USE_SYCL
template <typename Reducer>
struct ReduceFunctor<SYCLDevice, Reducer>
: ReduceFunctorBase<SYCLDevice, Reducer>{};
#endif // TENSORFLOW_USE_SYCL
} // namespace functor
} // namespace tensorflow

27
tensorflow/core/kernels/reduction_ops_sum.cc
View File

@ -64,4 +64,31 @@ REGISTER_KERNEL_BUILDER(
#endif
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNELS(type) \
REGISTER_KERNEL_BUILDER( \
Name("Sum") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.TypeConstraint<int32>("Tidx") \
.HostMemory("reduction_indices"), \
ReductionOp<SYCLDevice, type, Eigen::internal::SumReducer<type>>);
REGISTER_SYCL_KERNELS(float);
REGISTER_SYCL_KERNELS(double);
#undef REGISTER_SYCL_KERNELS
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(
Name("Sum")
.Device(DEVICE_SYCL)
.TypeConstraint<int32>("T")
.TypeConstraint<int32>("Tidx")
.HostMemory("input")
.HostMemory("output")
.HostMemory("reduction_indices"),
ReductionOp<CPUDevice, int32, Eigen::internal::SumReducer<int32>>);
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

19
tensorflow/core/kernels/scatter_functor.h
View File

@ -25,6 +25,9 @@ namespace tensorflow {
class OpKernelContext;
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
namespace scatter_op {
@ -82,10 +85,9 @@ struct ScatterFunctor {
typename TTypes<Index>::ConstFlat indices);
};
// Specializations of scatter functor for CPU.
template <typename T, typename Index, scatter_op::UpdateOp op>
struct ScatterFunctor<CPUDevice, T, Index, op> {
Index operator()(OpKernelContext* c, const CPUDevice& d,
template <typename Device, typename T, typename Index, scatter_op::UpdateOp op>
struct ScatterFunctorBase {
Index operator()(OpKernelContext* c, const Device& d,
typename TTypes<T>::Matrix params,
typename TTypes<T>::ConstMatrix updates,
typename TTypes<Index>::ConstFlat indices) {
@ -106,6 +108,15 @@ struct ScatterFunctor<CPUDevice, T, Index, op> {
}
};
template <typename T, typename Index, scatter_op::UpdateOp op>
struct ScatterFunctor<CPUDevice, T, Index, op>
: ScatterFunctorBase<CPUDevice, T, Index, op>{};
#if TENSORFLOW_USE_SYCL
template<typename T, typename Index, scatter_op::UpdateOp op>
struct ScatterFunctor<SYCLDevice, T, Index, op>
: ScatterFunctorBase<SYCLDevice, T, Index, op>{};
#endif // TENSORFLOW_USE_SYCL
} // namespace functor
} // namespace tensorflow

17
tensorflow/core/kernels/scatter_op.cc
View File

@ -27,6 +27,9 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
// Check whether updates.shape = indices.shape + params.shape[1:]
static bool ValidShapes(const Tensor& params, const Tensor& updates,
@ -170,6 +173,20 @@ TF_CALL_GPU_NUMBER_TYPES_NO_HALF(REGISTER_SCATTER_UPDATE_GPU);
#endif // GOOGLE_CUDA
// Registers GPU kernels.
#if TENSORFLOW_USE_SYCL
#define REGISTER_SCATTER_ARITHEMTIC_SYCL(type) \
REGISTER_SCATTER_ARITHEMTIC(type, SYCL);
#define REGISTER_SCATTER_UPDATE_SYCL(type) REGISTER_SCATTER_UPDATE(type, SYCL);
REGISTER_SCATTER_ARITHEMTIC_SYCL(float);
REGISTER_SCATTER_UPDATE_SYCL(float);
#undef REGISTER_SCATTER_ARITHEMTIC_SYCL
#undef REGISTER_SCATTER_UPDATE_SYCL
#endif // TENSORFLOW_USE_SYCL
#undef REGISTER_SCATTER_ARITHEMTIC
#undef REGISTER_SCATTER_ARITHEMTIC_CPU
#undef REGISTER_SCATTER_ARITHEMTIC_GPU

31
tensorflow/core/kernels/session_ops.cc
View File

@ -67,6 +67,19 @@ TF_CALL_NUMBER_TYPES(REGISTER_GPU_KERNEL);
REGISTER_GPU_KERNEL(bool);
#undef REGISTER_GPU_KERNEL
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("GetSessionHandle") \
.Device(DEVICE_SYCL) \
.HostMemory("handle") \
.TypeConstraint<type>("T"), \
GetSessionHandleOp)
TF_CALL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(bool);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
class GetSessionTensorOp : public OpKernel {
public:
explicit GetSessionTensorOp(OpKernelConstruction* context)
@ -97,6 +110,19 @@ TF_CALL_NUMBER_TYPES(REGISTER_GPU_KERNEL);
REGISTER_GPU_KERNEL(bool);
#undef REGISTER_GPU_KERNEL
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("GetSessionTensor") \
.Device(DEVICE_SYCL) \
.HostMemory("handle") \
.TypeConstraint<type>("dtype"), \
GetSessionTensorOp)
TF_CALL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(bool);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
class DeleteSessionTensorOp : public OpKernel {
public:
explicit DeleteSessionTensorOp(OpKernelConstruction* context)
@ -117,4 +143,9 @@ REGISTER_KERNEL_BUILDER(
Name("DeleteSessionTensor").Device(DEVICE_GPU).HostMemory("handle"),
DeleteSessionTensorOp);
#ifdef TENSORFLOW_USE_SYCL
REGISTER_KERNEL_BUILDER(
Name("DeleteSessionTensor").Device(DEVICE_SYCL).HostMemory("handle"),
DeleteSessionTensorOp);
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

37
tensorflow/core/kernels/shape_ops.cc
View File

@ -210,6 +210,43 @@ REGISTER_KERNEL_BUILDER(Name("ShapeN")
ShapeNOp<int64>);
#endif
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("ShapeN") \
.Device(DEVICE_SYCL) \
.HostMemory("output") \
.TypeConstraint<int32>("out_type") \
.TypeConstraint<type>("T"), \
ShapeNOp<int32>); \
REGISTER_KERNEL_BUILDER(Name("ShapeN") \
.Device(DEVICE_SYCL) \
.HostMemory("output") \
.TypeConstraint<int64>("out_type") \
.TypeConstraint<type>("T"), \
ShapeNOp<int64>)
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#undef REGISTER_SYCL_KERNEL
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("ShapeN")
.Device(DEVICE_SYCL)
.HostMemory("input")
.HostMemory("output")
.TypeConstraint<int32>("T")
.TypeConstraint<int32>("out_type"),
ShapeNOp<int32>);
REGISTER_KERNEL_BUILDER(Name("ShapeN")
.Device(DEVICE_SYCL)
.HostMemory("input")
.HostMemory("output")
.TypeConstraint<int32>("T")
.TypeConstraint<int64>("out_type"),
ShapeNOp<int64>);
#endif // TENSORFLOW_USE_SYCL
class RankOp : public OpKernel {
public:
explicit RankOp(OpKernelConstruction* ctx) : OpKernel(ctx) {}

57
tensorflow/core/kernels/slice_op.cc
View File

@ -56,6 +56,9 @@ gtl::InlinedVector<int64, 4> IntTensorToInt64Vec(const Tensor& tensor) {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
// Shared code that is not dependent on the type of T. We do this to reduce
// code size by not duplicating all this for all T (float, double, int32, etc.)
@ -300,4 +303,58 @@ REGISTER_KERNEL_BUILDER(Name("Slice")
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
// Forward declarations of the functor specializations for SYCL.
namespace functor {
#define DECLARE_SYCL_SPEC(T, NDIM) \
template <> \
void Slice<SYCLDevice, T, NDIM>::operator()( \
const SYCLDevice& d, typename TTypes<T, NDIM>::Tensor output,\
typename TTypes<T, NDIM>::ConstTensor input, \
const Eigen::DSizes<Eigen::DenseIndex, NDIM>& indices, \
const Eigen::DSizes<Eigen::DenseIndex, NDIM>& sizes); \
extern template struct Slice<SYCLDevice, T, NDIM>;
#define DECLARE_FOR_N(T) \
DECLARE_SYCL_SPEC(T, 1); \
DECLARE_SYCL_SPEC(T, 2); \
DECLARE_SYCL_SPEC(T, 3); \
DECLARE_SYCL_SPEC(T, 4); \
DECLARE_SYCL_SPEC(T, 5); \
DECLARE_SYCL_SPEC(T, 6);
TF_CALL_GPU_NUMBER_TYPES(DECLARE_FOR_N);
DECLARE_FOR_N(int32);
#undef DECLARE_FOR_N
#undef DECLARE_SYCL_SPEC
} // namespace functor
#define REGISTER_SYCL(type) \
REGISTER_KERNEL_BUILDER(Name("Slice") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.HostMemory("begin") \
.HostMemory("size") \
.TypeConstraint<int32>("Index"), \
SliceOp<SYCLDevice, type>)
TF_CALL_GPU_NUMBER_TYPES(REGISTER_SYCL);
// A special GPU kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("Slice")
.Device(DEVICE_SYCL)
.TypeConstraint<int32>("T")
.TypeConstraint<int32>("Index")
.HostMemory("input")
.HostMemory("begin")
.HostMemory("size")
.HostMemory("output"),
SliceOp<CPUDevice, int32>);
#undef REGISTER_SYCL
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

12
tensorflow/core/kernels/slice_op_cpu_impl.h
View File

@ -34,6 +34,18 @@ DEFINE_CPU_KERNELS(bfloat16);
#undef DEFINE_CPU_KERNELS
#ifdef TENSORFLOW_USE_SYCL
using SyclDevice = Eigen::SyclDevice;
#define DEFINE_SYCL_KERNELS(T) \
template struct functor::Slice<SyclDevice, T, CPU_PROVIDED_IXDIM>;
TF_CALL_GPU_NUMBER_TYPES(DEFINE_SYCL_KERNELS);
DEFINE_SYCL_KERNELS(int32);
#undef DEFINE_SYCL_KERNELS
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow
#endif // THIRD_PARTY_TENSORFLOW_CORE_KERNELS_SLICE_OP_CPU_IMPL_H_

11
tensorflow/core/kernels/split_lib.h
View File

@ -48,6 +48,17 @@ struct Split<Eigen::ThreadPoolDevice, T> {
const Eigen::DSizes<Eigen::DenseIndex, 3>& slice_sizes);
};
#ifdef TENSORFLOW_USE_SYCL
template <typename T>
struct Split<Eigen::SyclDevice, T> {
void operator()(const Eigen::SyclDevice& d,
typename TTypes<T, 3>::Tensor output,
typename TTypes<T, 3>::ConstTensor input,
const Eigen::DSizes<Eigen::DenseIndex, 3>& slice_indices,
const Eigen::DSizes<Eigen::DenseIndex, 3>& slice_sizes);
};
#endif // TENSORFLOW_USE_SYCL
} // namespace functor
} // namespace tensorflow

19
tensorflow/core/kernels/split_lib_cpu.cc
View File

@ -43,5 +43,24 @@ TF_CALL_ALL_TYPES(DEFINE_CPU_KERNELS)
DEFINE_CPU_KERNELS(quint8)
DEFINE_CPU_KERNELS(bfloat16)
#ifdef TENSORFLOW_USE_SYCL
template <typename T>
void Split<Eigen::SyclDevice, T>::operator()(
const Eigen::SyclDevice& d, typename TTypes<T, 3>::Tensor output,
typename TTypes<T, 3>::ConstTensor input,
const Eigen::DSizes<Eigen::DenseIndex, 3>& slice_indices,
const Eigen::DSizes<Eigen::DenseIndex, 3>& slice_sizes) {
if (output.size() < 131072) {
output = input.slice(slice_indices, slice_sizes);
} else {
output.device(d) = input.slice(slice_indices, slice_sizes);
}
}
#define DEFINE_SYCL_KERNELS(T) template struct Split<Eigen::SyclDevice, T>;
TF_CALL_GPU_NUMBER_TYPES(DEFINE_SYCL_KERNELS)
#endif // TENSORFLOW_USE_SYCL
} // namespace functor
} // namespace tensorflow

85
tensorflow/core/kernels/split_op.cc
View File

@ -36,6 +36,9 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
template <typename Device, typename T>
class SplitOpBase : public OpKernel {
@ -243,6 +246,75 @@ class SplitOpGPU : public SplitOpBase<GPUDevice, T> {
};
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
template <typename T>
class SplitOpSYCL : public SplitOpBase<SYCLDevice, T> {
public:
typedef SplitOpBase<SYCLDevice, T> Base;
explicit SplitOpSYCL(OpKernelConstruction* c) : Base(c) {}
void Compute(OpKernelContext* context) override {
bool done = false;
Base::ComputeEasyCases(context, &done);
if (!context->status().ok() || done) {
return;
}
const int32 split_dim = context->input(0).flat<int32>()(0);
const int32 num_split = Base::num_outputs();
const Tensor& input = context->input(1);
const TensorShape& input_shape = input.shape();
// Android also uses int32 indexing, so check here also.
OP_REQUIRES(
context, FastBoundsCheck(input.NumElements(),
std::numeric_limits<Eigen::DenseIndex>::max()),
errors::InvalidArgument("Split requires input size < ",
std::numeric_limits<Eigen::DenseIndex>::max()));
Eigen::DenseIndex prefix_dim_size;
Eigen::DenseIndex split_dim_size;
Eigen::DenseIndex suffix_dim_size;
std::tie(prefix_dim_size, split_dim_size, suffix_dim_size) =
Base::template SetDims<Eigen::DenseIndex>(input_shape, split_dim);
auto input_reshaped =
input.shaped<T, 3>({prefix_dim_size, split_dim_size, suffix_dim_size});
const int64 split_dim_output_size = split_dim_size / num_split;
TensorShape output_shape(input_shape);
output_shape.set_dim(split_dim, split_dim_output_size);
Eigen::DSizes<Eigen::DenseIndex, 3> indices{0, 0, 0};
Eigen::DSizes<Eigen::DenseIndex, 3> sizes{
prefix_dim_size, split_dim_output_size, suffix_dim_size};
for (int i = 0; i < num_split; ++i) {
Tensor* result = nullptr;
OP_REQUIRES_OK(context,
context->allocate_output(i, output_shape, &result));
if (prefix_dim_size * split_dim_output_size * suffix_dim_size > 0) {
Eigen::DSizes<Eigen::DenseIndex, 3> slice_indices;
Eigen::DSizes<Eigen::DenseIndex, 3> slice_sizes;
for (int j = 0; j < 3; ++j) {
slice_indices[j] = indices[j];
slice_sizes[j] = sizes[j];
}
auto result_shaped = result->shaped<T, 3>(
{prefix_dim_size, split_dim_output_size, suffix_dim_size});
functor::Split<SYCLDevice, T>()(context->eigen_device<SYCLDevice>(),
result_shaped, input_reshaped,
slice_indices, slice_sizes);
}
indices[1] += split_dim_output_size;
}
}
};
#endif // TENSORFLOW_USE_SYCL
#define REGISTER_SPLIT(type) \
REGISTER_KERNEL_BUILDER(Name("Split") \
.Device(DEVICE_CPU) \
@ -269,4 +341,17 @@ TF_CALL_GPU_NUMBER_TYPES(REGISTER_GPU);
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL(type) \
REGISTER_KERNEL_BUILDER(Name("Split") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.HostMemory("split_dim"), \
SplitOpSYCL<type>)
TF_CALL_GPU_NUMBER_TYPES(REGISTER_SYCL);
#undef REGISTER_SYCL
#endif // TENSORFLOW_USE_SYCL
} // end namespace tensorflow

25
tensorflow/core/kernels/tile_ops.cc
View File

@ -40,6 +40,9 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
// Forward declarations of functors that will be defined in
// tile_ops_cpu_impl*.cc and tile_ops_gpu.cu.cc.
@ -225,6 +228,11 @@ inline void TileOp<Device>::HandleCase(
#define HANDLE_TYPE_NAME_GPU(T) \
HANDLE_CASE_DIM(GPUDevice, T, DataTypeToEnum<T>::value);
#ifdef TENSORFLOW_USE_SYCL
#define HANDLE_TYPE_NAME_SYCL(T) \
HANDLE_CASE_DIM(SYCLDevice, T, DataTypeToEnum<T>::value);
#endif // TENSORFLOW_USE_SYCL
TF_CALL_bool(HANDLE_TYPE_NAME_CPU);
TF_CALL_float(HANDLE_TYPE_NAME_CPU);
TF_CALL_double(HANDLE_TYPE_NAME_CPU);
@ -248,8 +256,15 @@ TF_CALL_complex64(HANDLE_TYPE_NAME_GPU);
TF_CALL_complex128(HANDLE_TYPE_NAME_GPU);
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
TF_CALL_float(HANDLE_TYPE_NAME_SYCL);
#endif // TENSORFLOW_USE_SYCL
#undef HANDLE_TYPE_NAME_CPU
#undef HANDLE_TYPE_NAME_GPU
#ifdef TENSORFLOW_USE_SYCL
#undef HANDLE_TYPE_NAME_SYCL
#endif // TENSORFLOW_USE_SYCL
#undef HANDLE_CASE_DIM
#undef HANDLE_CASE
@ -578,4 +593,14 @@ REGISTER_KERNEL_BUILDER(Name("TileGrad")
TileGradientOp<GPUDevice>);
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
REGISTER_KERNEL_BUILDER(Name("Tile")
.Device(DEVICE_SYCL)
.TypeConstraint<float>("T")
.TypeConstraint<int32>("Tmultiples")
.HostMemory("multiples"),
TileOp<SYCLDevice>);
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

24
tensorflow/core/kernels/tile_ops_cpu_impl.h
View File

@ -62,6 +62,30 @@ TF_CALL_complex128(DEFINE_TYPE);
#undef DEFINE_DIM
#undef DEFINE_TYPE
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
// Register functors used for TileOp.
#define DEFINE_DIM(T, NDIM) template struct Tile<SYCLDevice, T, NDIM>;
#define DEFINE_TYPE(T) DEFINE_DIM(T, CPU_PROVIDED_IXDIM)
TF_CALL_float(DEFINE_TYPE);
#undef DEFINE_DIM
#undef DEFINE_TYPE
// Register functors used for TileGradientOp.
#define DEFINE_DIM(T, NDIM) \
template struct TileGrad<SYCLDevice, T, NDIM>; \
template struct ReduceAndReshape<SYCLDevice, T, NDIM, 1>;
#define DEFINE_TYPE(T) DEFINE_DIM(T, CPU_PROVIDED_IXDIM)
TF_CALL_float(DEFINE_TYPE);
#undef DEFINE_DIM
#undef DEFINE_TYPE
#endif // TENSORFLOW_USE_SYCL
} // end namespace functor
} // end namespace tensorflow

17
tensorflow/core/kernels/training_ops.cc
View File

@ -46,6 +46,17 @@ struct ApplyGradientDescent<CPUDevice, T> {
}
};
#ifdef TENSORFLOW_USE_SYCL
template <typename T>
struct ApplyGradientDescent<SYCLDevice, T> {
void operator()(const SYCLDevice& d, typename TTypes<T>::Flat var,
typename TTypes<T>::ConstScalar lr,
typename TTypes<T>::ConstFlat grad) {
var.device(d) -= grad * lr();
}
};
#endif
template <typename T>
struct ApplyAdadelta<CPUDevice, T> {
void operator()(const CPUDevice& d, typename TTypes<T>::Flat var,
@ -357,6 +368,12 @@ TF_CALL_half(REGISTER_CPU_KERNELS);
TF_CALL_float(REGISTER_CPU_KERNELS);
TF_CALL_double(REGISTER_CPU_KERNELS);
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNELS(T) REGISTER_KERNELS(SYCL, T);
TF_CALL_float(REGISTER_SYCL_KERNELS);
#undef REGISTER_SYCL_KERNELS
#endif
#if GOOGLE_CUDA
// Forward declarations of the functor specializations for GPU.
namespace functor {

24
tensorflow/core/kernels/transpose_functor_cpu.cc
View File

@ -114,4 +114,28 @@ Status DoTranspose<Device>(const Device& d, const Tensor& in,
return Status::OK();
}
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
template <>
Status DoTranspose<SYCLDevice>(const SYCLDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
CHECK_GE(in.dims(), 2);
CHECK_EQ(in.dims(), out->dims());
CHECK_EQ(in.dims(), perm.size());
CHECK_EQ(in.dtype(), out->dtype());
switch (in.dtype()) {
case DT_FLOAT:
case DT_INT32:
internal::Transpose<SYCLDevice, uint32>(d, in, perm, out);
break;
default:
return errors::Unimplemented("Unsupported dtype on SYCL: ", in.dtype());
}
return Status::OK();
}
#endif // TENSORFLOW_USE_SYCL
} // namespace tensorflow

25
tensorflow/core/kernels/unpack_op.cc
View File

@ -32,6 +32,10 @@ namespace tensorflow {
typedef Eigen::ThreadPoolDevice CPUDevice;
typedef Eigen::GpuDevice GPUDevice;
#ifdef TENSORFLOW_USE_SYCL
typedef Eigen::SyclDevice SYCLDevice;
#endif // TENSORFLOW_USE_SYCL
template <typename Device, typename T>
class UnpackOp : public OpKernel {
public:
@ -149,4 +153,25 @@ REGISTER_KERNEL_BUILDER(Name("Unpack")
#endif // GOOGLE_CUDA
#ifdef TENSORFLOW_USE_SYCL
#define REGISTER_SYCL(type) \
REGISTER_KERNEL_BUILDER( \
Name("Unpack").Device(DEVICE_SYCL).TypeConstraint<type>("T"), \
UnpackOp<SYCLDevice, type>)
REGISTER_SYCL(float);
#undef REGISTER_SYCL
// A special SYCL kernel for int32.
// TODO(b/25387198): Also enable int32 in device memory. This kernel
// registration requires all int32 inputs and outputs to be in host memory.
REGISTER_KERNEL_BUILDER(Name("Unpack")
.Device(DEVICE_SYCL)
.HostMemory("value")
.HostMemory("output")
.TypeConstraint<int32>("T"),
UnpackOp<CPUDevice, int32>);
#endif // TENSORFLOW_USE_SYCL
} // end namespace tensorflow

47
tensorflow/core/ops/array_ops.cc
View File

@ -416,24 +416,49 @@ REGISTER_OP("SplitV")
.Attr("T: type")
.Attr("Tlen: {int32, int64} = DT_INT64")
.SetShapeFn([](InferenceContext* c) {
ShapeHandle unused;
DimensionHandle split_dimension;
TF_RETURN_IF_ERROR(c->MakeDimForScalarInput(2, &split_dimension));
int32 num_outputs = c->num_outputs();
// Return unknown shapes with the same rank as the input
// or unknown rank if input's rank isn't known
// can't determine exact shapes until runtime because
// we don't know where the tensor containing the split sizes
// is located
int32 rank = c->Rank(c->input(0));
ShapeHandle input = c->input(0);
int32 rank = c->Rank(input);
ShapeHandle output_shape;
const Tensor* size_splits = c->input_tensor(1);
if (rank == InferenceContext::kUnknownRank) {
// If the rank of input tensor is unknown, then return unkown shapes.
output_shape = c->UnknownShape();
for (int i = 0; i < num_outputs; ++i) {
c->set_output(i, output_shape);
}
} else if (rank == 0) {
// Throw error if input is a scalar.
return errors::InvalidArgument("Can't split scalars");
} else {
} else if (size_splits == nullptr || !c->ValueKnown(split_dimension)) {
// If split dimension or tensor containing the split sizes is unkown,
// then return unknown shapes of same rank as input.
output_shape = c->UnknownShapeOfRank(rank);
}
for (int i = 0; i < num_outputs; ++i) {
c->set_output(i, output_shape);
for (int i = 0; i < num_outputs; ++i) {
c->set_output(i, output_shape);
}
} else {
// Determine the output shape if split dimension and split sizes are known
int64 split_dim = c->Value(split_dimension);
TF_RETURN_IF_ERROR(c->WithRankAtLeast(input, split_dim + 1, &input));
std::vector<int64> data;
if (size_splits->dtype() == DT_INT32) {
data = AsInt64<int32>(size_splits, size_splits->shape().dim_size(0));
} else {
data = AsInt64<int64>(size_splits, size_splits->shape().dim_size(0));
}
if (num_outputs != data.size()) {
return errors::InvalidArgument(
"Length of size_splits should be equal to num_outputs");
}
for (int i = 0; i < num_outputs; ++i) {
output_shape = c->UnknownShapeOfRank(rank);
TF_RETURN_IF_ERROR(
c->ReplaceDim(input, split_dim, c->MakeDim(data[i]), &output_shape));
c->set_output(i, output_shape);
}
}
return Status::OK();

42
tensorflow/core/platform/default/logging.cc
View File

@ -84,39 +84,49 @@ void LogMessage::GenerateLogMessage() {
namespace {
int64 MinLogLevel() {
const char* tf_env_var_val = getenv("TF_CPP_MIN_LOG_LEVEL");
// Parse log level (int64) from environment variable (char*)
int64 LogLevelStrToInt(const char* tf_env_var_val) {
if (tf_env_var_val == nullptr) {
return 0;
}
// Ideally we would use env_var / safe_strto64, but it is
// hard to use here without pulling in a lot of dependencies,
// so we do a poor-man's parsing.
// so we use std:istringstream instead
string min_log_level(tf_env_var_val);
if (min_log_level == "1") {
// Maps to WARNING
return 1;
} else if (min_log_level == "2") {
// Maps to ERROR
return 2;
} else if (min_log_level == "3") {
// Maps to FATAL
return 3;
} else {
// Maps to INFO (the default).
return 0;
std::istringstream ss(min_log_level);
int64 level;
if (!(ss >> level)) {
// Invalid vlog level setting, set level to default (0)
level = 0;
}
return level;
}
int64 MinLogLevelFromEnv() {
const char* tf_env_var_val = getenv("TF_CPP_MIN_LOG_LEVEL");
return LogLevelStrToInt(tf_env_var_val);
}
int64 MinVLogLevelFromEnv() {
const char* tf_env_var_val = getenv("TF_CPP_MIN_VLOG_LEVEL");
return LogLevelStrToInt(tf_env_var_val);
}
} // namespace
LogMessage::~LogMessage() {
// Read the min log level once during the first call to logging.
static int64 min_log_level = MinLogLevel();
static int64 min_log_level = MinLogLevelFromEnv();
if (TF_PREDICT_TRUE(severity_ >= min_log_level)) GenerateLogMessage();
}
int64 LogMessage::MinVLogLevel() {
static int64 min_vlog_level = MinVLogLevelFromEnv();
return min_vlog_level;
}
LogMessageFatal::LogMessageFatal(const char* file, int line)
: LogMessage(file, line, FATAL) {}
LogMessageFatal::~LogMessageFatal() {

16
tensorflow/core/platform/default/logging.h
View File

@ -41,6 +41,11 @@ class LogMessage : public std::basic_ostringstream<char> {
LogMessage(const char* fname, int line, int severity);
~LogMessage();
// Returns the minimum log level for VLOG statements.
// E.g., if MinVLogLevel() is 2, then VLOG(2) statements will produce output,
// but VLOG(3) will not. Defaults to 0.
static int64 MinVLogLevel();
protected:
void GenerateLogMessage();
@ -71,11 +76,18 @@ class LogMessageFatal : public LogMessage {
#define LOG(severity) _TF_LOG_##severity
// TODO(jeff): Define a proper implementation of VLOG_IS_ON
#ifdef IS_MOBILE_PLATFORM
// Turn VLOG off when under mobile devices for considerations of binary size.
#define VLOG_IS_ON(lvl) ((lvl) <= 0)
#else
// Otherwise, Set TF_CPP_MIN_VLOG_LEVEL environment to update minimum log level
// of VLOG
#define VLOG_IS_ON(lvl) \
((lvl) <= ::tensorflow::internal::LogMessage::MinVLogLevel())
#endif
#define VLOG(lvl) \
if (VLOG_IS_ON(lvl)) \
if (TF_PREDICT_FALSE(VLOG_IS_ON(lvl))) \
::tensorflow::internal::LogMessage(__FILE__, __LINE__, tensorflow::INFO)
// CHECK dies with a fatal error if condition is not true. It is *not*

3
tensorflow/examples/how_tos/__init__.py Normal file
View File

@ -0,0 +1,3 @@
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

3
tensorflow/examples/udacity/README.md
View File

@ -54,6 +54,9 @@ to get the ip of the new virtual machine. To switch from default virtual machine
Note that `docker-machine env tensorflow` outputs some environment variables such like `DOCKER_HOST`. Then your docker client is now connected to the docker host in virtual machine `tensorflow`
* **I'm getting the error - docker: Cannot connect to the Docker daemon. Is the docker daemon running on this host? - when I run 'docker run'.**
This is a permissions issue, and a popular answer is provided for Linux and Max OSX [here](http://stackoverflow.com/questions/21871479/docker-cant-connect-to-docker-daemon) on StackOverflow.
Notes for anyone needing to build their own containers (mostly instructors)
===========================================================================

244
tensorflow/g3doc/api_docs/python/functions_and_classes/shard0/tf.summary.TaggedRunMetadata.md
View File

@ -1,4 +1,185 @@
- - -
#### `tf.summary.TaggedRunMetadata.ByteSize()` {#TaggedRunMetadata.ByteSize}
- - -
#### `tf.summary.TaggedRunMetadata.Clear()` {#TaggedRunMetadata.Clear}
- - -
#### `tf.summary.TaggedRunMetadata.ClearExtension(extension_handle)` {#TaggedRunMetadata.ClearExtension}
- - -
#### `tf.summary.TaggedRunMetadata.ClearField(field_name)` {#TaggedRunMetadata.ClearField}
- - -
#### `tf.summary.TaggedRunMetadata.CopyFrom(other_msg)` {#TaggedRunMetadata.CopyFrom}
Copies the content of the specified message into the current message.
The method clears the current message and then merges the specified
message using MergeFrom.
##### Args:
* <b>`other_msg`</b>: Message to copy into the current one.
- - -
#### `tf.summary.TaggedRunMetadata.DiscardUnknownFields()` {#TaggedRunMetadata.DiscardUnknownFields}
- - -
#### `tf.summary.TaggedRunMetadata.FindInitializationErrors()` {#TaggedRunMetadata.FindInitializationErrors}
Finds required fields which are not initialized.
##### Returns:
A list of strings. Each string is a path to an uninitialized field from
the top-level message, e.g. "foo.bar[5].baz".
- - -
#### `tf.summary.TaggedRunMetadata.FromString(s)` {#TaggedRunMetadata.FromString}
- - -
#### `tf.summary.TaggedRunMetadata.HasExtension(extension_handle)` {#TaggedRunMetadata.HasExtension}
- - -
#### `tf.summary.TaggedRunMetadata.HasField(field_name)` {#TaggedRunMetadata.HasField}
- - -
#### `tf.summary.TaggedRunMetadata.IsInitialized(errors=None)` {#TaggedRunMetadata.IsInitialized}
Checks if all required fields of a message are set.
##### Args:
* <b>`errors`</b>: A list which, if provided, will be populated with the field
paths of all missing required fields.
##### Returns:
True iff the specified message has all required fields set.
- - -
#### `tf.summary.TaggedRunMetadata.ListFields()` {#TaggedRunMetadata.ListFields}
- - -
#### `tf.summary.TaggedRunMetadata.MergeFrom(msg)` {#TaggedRunMetadata.MergeFrom}
- - -
#### `tf.summary.TaggedRunMetadata.MergeFromString(serialized)` {#TaggedRunMetadata.MergeFromString}
- - -
#### `tf.summary.TaggedRunMetadata.ParseFromString(serialized)` {#TaggedRunMetadata.ParseFromString}
Parse serialized protocol buffer data into this message.
Like MergeFromString(), except we clear the object first and
do not return the value that MergeFromString returns.
- - -
#### `tf.summary.TaggedRunMetadata.RegisterExtension(extension_handle)` {#TaggedRunMetadata.RegisterExtension}
- - -
#### `tf.summary.TaggedRunMetadata.SerializePartialToString()` {#TaggedRunMetadata.SerializePartialToString}
- - -
#### `tf.summary.TaggedRunMetadata.SerializeToString()` {#TaggedRunMetadata.SerializeToString}
- - -
#### `tf.summary.TaggedRunMetadata.SetInParent()` {#TaggedRunMetadata.SetInParent}
Sets the _cached_byte_size_dirty bit to true,
and propagates this to our listener iff this was a state change.
- - -
#### `tf.summary.TaggedRunMetadata.WhichOneof(oneof_name)` {#TaggedRunMetadata.WhichOneof}
Returns the name of the currently set field inside a oneof, or None.
- - -
#### `tf.summary.TaggedRunMetadata.__deepcopy__(memo=None)` {#TaggedRunMetadata.__deepcopy__}
- - -
#### `tf.summary.TaggedRunMetadata.__eq__(other)` {#TaggedRunMetadata.__eq__}
- - -
#### `tf.summary.TaggedRunMetadata.__getstate__()` {#TaggedRunMetadata.__getstate__}
@ -6,3 +187,66 @@
Support the pickle protocol.
- - -
#### `tf.summary.TaggedRunMetadata.__hash__()` {#TaggedRunMetadata.__hash__}
- - -
#### `tf.summary.TaggedRunMetadata.__init__(**kwargs)` {#TaggedRunMetadata.__init__}
- - -
#### `tf.summary.TaggedRunMetadata.__ne__(other_msg)` {#TaggedRunMetadata.__ne__}
- - -
#### `tf.summary.TaggedRunMetadata.__repr__()` {#TaggedRunMetadata.__repr__}
- - -
#### `tf.summary.TaggedRunMetadata.__setstate__(state)` {#TaggedRunMetadata.__setstate__}
Support the pickle protocol.
- - -
#### `tf.summary.TaggedRunMetadata.__str__()` {#TaggedRunMetadata.__str__}
- - -
#### `tf.summary.TaggedRunMetadata.__unicode__()` {#TaggedRunMetadata.__unicode__}
- - -
#### `tf.summary.TaggedRunMetadata.run_metadata` {#TaggedRunMetadata.run_metadata}
Magic attribute generated for "run_metadata" proto field.
- - -
#### `tf.summary.TaggedRunMetadata.tag` {#TaggedRunMetadata.tag}
Magic attribute generated for "tag" proto field.

17
tensorflow/g3doc/api_docs/python/functions_and_classes/shard1/tf.merge_all_summaries.md Normal file
View File

@ -0,0 +1,17 @@
### `tf.merge_all_summaries(*args, **kwargs)` {#merge_all_summaries}
Merges all summaries collected in the default graph. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.merge_all.
Args:
key: `GraphKey` used to collect the summaries. Defaults to
`GraphKeys.SUMMARIES`.
Returns:
If no summaries were collected, returns None. Otherwise returns a scalar
`Tensor` of type `string` containing the serialized `Summary` protocol
buffer resulting from the merging.

49
tensorflow/g3doc/api_docs/python/functions_and_classes/shard2/tf.image_summary.md Normal file
View File

@ -0,0 +1,49 @@
### `tf.image_summary(*args, **kwargs)` {#image_summary}
Outputs a `Summary` protocol buffer with images. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.image. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, the max_images argument was renamed to max_outputs.
The summary has up to `max_images` summary values containing images. The
images are built from `tensor` which must be 4-D with shape `[batch_size,
height, width, channels]` and where `channels` can be:
* 1: `tensor` is interpreted as Grayscale.
* 3: `tensor` is interpreted as RGB.
* 4: `tensor` is interpreted as RGBA.
The images have the same number of channels as the input tensor. For float
input, the values are normalized one image at a time to fit in the range
`[0, 255]`. `uint8` values are unchanged. The op uses two different
normalization algorithms:
* If the input values are all positive, they are rescaled so the largest one
is 255.
* If any input value is negative, the values are shifted so input value 0.0
is at 127. They are then rescaled so that either the smallest value is 0,
or the largest one is 255.
The `tag` argument is a scalar `Tensor` of type `string`. It is used to
build the `tag` of the summary values:
* If `max_images` is 1, the summary value tag is '*tag*/image'.
* If `max_images` is greater than 1, the summary value tags are
generated sequentially as '*tag*/image/0', '*tag*/image/1', etc.
Args:
tag: A scalar `Tensor` of type `string`. Used to build the `tag`
of the summary values.
tensor: A 4-D `uint8` or `float32` `Tensor` of shape `[batch_size, height,
width, channels]` where `channels` is 1, 3, or 4.
max_images: Max number of batch elements to generate images for.
collections: Optional list of ops.GraphKeys. The collections to add the
summary to. Defaults to [ops.GraphKeys.SUMMARIES]
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`. The serialized `Summary` protocol
buffer.

237
tensorflow/g3doc/api_docs/python/functions_and_classes/shard2/tf.summary.SummaryDescription.md
View File

@ -1,4 +1,185 @@
- - -
#### `tf.summary.SummaryDescription.ByteSize()` {#SummaryDescription.ByteSize}
- - -
#### `tf.summary.SummaryDescription.Clear()` {#SummaryDescription.Clear}
- - -
#### `tf.summary.SummaryDescription.ClearExtension(extension_handle)` {#SummaryDescription.ClearExtension}
- - -
#### `tf.summary.SummaryDescription.ClearField(field_name)` {#SummaryDescription.ClearField}
- - -
#### `tf.summary.SummaryDescription.CopyFrom(other_msg)` {#SummaryDescription.CopyFrom}
Copies the content of the specified message into the current message.
The method clears the current message and then merges the specified
message using MergeFrom.
##### Args:
* <b>`other_msg`</b>: Message to copy into the current one.
- - -
#### `tf.summary.SummaryDescription.DiscardUnknownFields()` {#SummaryDescription.DiscardUnknownFields}
- - -
#### `tf.summary.SummaryDescription.FindInitializationErrors()` {#SummaryDescription.FindInitializationErrors}
Finds required fields which are not initialized.
##### Returns:
A list of strings. Each string is a path to an uninitialized field from
the top-level message, e.g. "foo.bar[5].baz".
- - -
#### `tf.summary.SummaryDescription.FromString(s)` {#SummaryDescription.FromString}
- - -
#### `tf.summary.SummaryDescription.HasExtension(extension_handle)` {#SummaryDescription.HasExtension}
- - -
#### `tf.summary.SummaryDescription.HasField(field_name)` {#SummaryDescription.HasField}
- - -
#### `tf.summary.SummaryDescription.IsInitialized(errors=None)` {#SummaryDescription.IsInitialized}
Checks if all required fields of a message are set.
##### Args:
* <b>`errors`</b>: A list which, if provided, will be populated with the field
paths of all missing required fields.
##### Returns:
True iff the specified message has all required fields set.
- - -
#### `tf.summary.SummaryDescription.ListFields()` {#SummaryDescription.ListFields}
- - -
#### `tf.summary.SummaryDescription.MergeFrom(msg)` {#SummaryDescription.MergeFrom}
- - -
#### `tf.summary.SummaryDescription.MergeFromString(serialized)` {#SummaryDescription.MergeFromString}
- - -
#### `tf.summary.SummaryDescription.ParseFromString(serialized)` {#SummaryDescription.ParseFromString}
Parse serialized protocol buffer data into this message.
Like MergeFromString(), except we clear the object first and
do not return the value that MergeFromString returns.
- - -
#### `tf.summary.SummaryDescription.RegisterExtension(extension_handle)` {#SummaryDescription.RegisterExtension}
- - -
#### `tf.summary.SummaryDescription.SerializePartialToString()` {#SummaryDescription.SerializePartialToString}
- - -
#### `tf.summary.SummaryDescription.SerializeToString()` {#SummaryDescription.SerializeToString}
- - -
#### `tf.summary.SummaryDescription.SetInParent()` {#SummaryDescription.SetInParent}
Sets the _cached_byte_size_dirty bit to true,
and propagates this to our listener iff this was a state change.
- - -
#### `tf.summary.SummaryDescription.WhichOneof(oneof_name)` {#SummaryDescription.WhichOneof}
Returns the name of the currently set field inside a oneof, or None.
- - -
#### `tf.summary.SummaryDescription.__deepcopy__(memo=None)` {#SummaryDescription.__deepcopy__}
- - -
#### `tf.summary.SummaryDescription.__eq__(other)` {#SummaryDescription.__eq__}
- - -
#### `tf.summary.SummaryDescription.__getstate__()` {#SummaryDescription.__getstate__}
@ -6,3 +187,59 @@
Support the pickle protocol.
- - -
#### `tf.summary.SummaryDescription.__hash__()` {#SummaryDescription.__hash__}
- - -
#### `tf.summary.SummaryDescription.__init__(**kwargs)` {#SummaryDescription.__init__}
- - -
#### `tf.summary.SummaryDescription.__ne__(other_msg)` {#SummaryDescription.__ne__}
- - -
#### `tf.summary.SummaryDescription.__repr__()` {#SummaryDescription.__repr__}
- - -
#### `tf.summary.SummaryDescription.__setstate__(state)` {#SummaryDescription.__setstate__}
Support the pickle protocol.
- - -
#### `tf.summary.SummaryDescription.__str__()` {#SummaryDescription.__str__}
- - -
#### `tf.summary.SummaryDescription.__unicode__()` {#SummaryDescription.__unicode__}
- - -
#### `tf.summary.SummaryDescription.type_hint` {#SummaryDescription.type_hint}
Magic attribute generated for "type_hint" proto field.

521
tensorflow/g3doc/api_docs/python/functions_and_classes/shard2/tf.test.TestCase.md
View File

@ -173,125 +173,6 @@ Checks that for all elements of farray1 and farray2
* <b>`err`</b>: a float value.
- - -
#### `tf.test.TestCase.assertBetween(value, minv, maxv, msg=None)` {#TestCase.assertBetween}
Asserts that value is between minv and maxv (inclusive).
- - -
#### `tf.test.TestCase.assertCommandFails(command, regexes, env=None, close_fds=True, msg=None)` {#TestCase.assertCommandFails}
Asserts a shell command fails and the error matches a regex in a list.
##### Args:
* <b>`command`</b>: List or string representing the command to run.
* <b>`regexes`</b>: the list of regular expression strings.
* <b>`env`</b>: Dictionary of environment variable settings.
* <b>`close_fds`</b>: Whether or not to close all open fd's in the child after
forking.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertCommandSucceeds(command, regexes=('',), env=None, close_fds=True, msg=None)` {#TestCase.assertCommandSucceeds}
Asserts that a shell command succeeds (i.e. exits with code 0).
##### Args:
* <b>`command`</b>: List or string representing the command to run.
* <b>`regexes`</b>: List of regular expression byte strings that match success.
* <b>`env`</b>: Dictionary of environment variable settings.
* <b>`close_fds`</b>: Whether or not to close all open fd's in the child after
forking.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsExactSubsequence(container, subsequence, msg=None)` {#TestCase.assertContainsExactSubsequence}
Assert that "container" contains "subsequence" as an exact subsequence.
Asserts that "container" contains all the elements of "subsequence", in
order, and without other elements interspersed. For example, [1, 2, 3] is an
exact subsequence of [0, 0, 1, 2, 3, 0] but not of [0, 0, 1, 2, 0, 3, 0].
##### Args:
* <b>`container`</b>: the list we're testing for subsequence inclusion.
* <b>`subsequence`</b>: the list we hope will be an exact subsequence of container.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsInOrder(strings, target, msg=None)` {#TestCase.assertContainsInOrder}
Asserts that the strings provided are found in the target in order.
This may be useful for checking HTML output.
##### Args:
* <b>`strings`</b>: A list of strings, such as [ 'fox', 'dog' ]
* <b>`target`</b>: A target string in which to look for the strings, such as
'The quick brown fox jumped over the lazy dog'.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsSubsequence(container, subsequence, msg=None)` {#TestCase.assertContainsSubsequence}
Assert that "container" contains "subsequence" as a subsequence.
Asserts that "container" contains all the elements of "subsequence", in
order, but possibly with other elements interspersed. For example, [1, 2, 3]
is a subsequence of [0, 0, 1, 2, 0, 3, 0] but not of [0, 0, 1, 3, 0, 2, 0].
##### Args:
* <b>`container`</b>: the list we're testing for subsequence inclusion.
* <b>`subsequence`</b>: the list we hope will be a subsequence of container.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsSubset(expected_subset, actual_set, msg=None)` {#TestCase.assertContainsSubset}
Checks whether actual iterable is a superset of expected iterable.
- - -
#### `tf.test.TestCase.assertCountEqual(*args, **kwargs)` {#TestCase.assertCountEqual}
An unordered sequence specific comparison.
Equivalent to assertItemsEqual(). This method is a compatibility layer
for Python 3k, since 2to3 does not convert assertItemsEqual() calls into
assertCountEqual() calls.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertDeviceEqual(device1, device2)` {#TestCase.assertDeviceEqual}
@ -314,48 +195,9 @@ Checks whether actual is a superset of expected.
- - -
#### `tf.test.TestCase.assertDictEqual(a, b, msg=None)` {#TestCase.assertDictEqual}
Raises AssertionError if a and b are not equal dictionaries.
##### Args:
#### `tf.test.TestCase.assertDictEqual(d1, d2, msg=None)` {#TestCase.assertDictEqual}
* <b>`a`</b>: A dict, the expected value.
* <b>`b`</b>: A dict, the actual value.
* <b>`msg`</b>: An optional str, the associated message.
##### Raises:
* <b>`AssertionError`</b>: if the dictionaries are not equal.
- - -
#### `tf.test.TestCase.assertEmpty(container, msg=None)` {#TestCase.assertEmpty}
Assert that an object has zero length.
##### Args:
* <b>`container`</b>: Anything that implements the collections.Sized interface.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertEndsWith(actual, expected_end, msg=None)` {#TestCase.assertEndsWith}
Assert that actual.endswith(expected_end) is True.
##### Args:
* <b>`actual`</b>: str
* <b>`expected_end`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
@ -440,11 +282,10 @@ Included for symmetry with assertIsNone.
- - -
#### `tf.test.TestCase.assertItemsEqual(*args, **kwargs)` {#TestCase.assertItemsEqual}
#### `tf.test.TestCase.assertItemsEqual(expected_seq, actual_seq, msg=None)` {#TestCase.assertItemsEqual}
An unordered sequence specific comparison.
It asserts that actual_seq and expected_seq have the same element counts.
An unordered sequence specific comparison. It asserts that
actual_seq and expected_seq have the same element counts.
Equivalent to::
self.assertEqual(Counter(iter(actual_seq)),
@ -457,30 +298,6 @@ Asserts that each element has the same count in both sequences.
- [0, 1, 1] and [1, 0, 1] compare equal.
- [0, 0, 1] and [0, 1] compare unequal.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertJsonEqual(first, second, msg=None)` {#TestCase.assertJsonEqual}
Asserts that the JSON objects defined in two strings are equal.
A summary of the differences will be included in the failure message
using assertSameStructure.
##### Args:
* <b>`first`</b>: A string contining JSON to decode and compare to second.
* <b>`second`</b>: A string contining JSON to decode and compare to first.
* <b>`msg`</b>: Additional text to include in the failure message.
- - -
@ -550,13 +367,6 @@ if not.
* <b>`msg`</b>: An optional string message to append to the failure message.
- - -
#### `tf.test.TestCase.assertNoCommonElements(expected_seq, actual_seq, msg=None)` {#TestCase.assertNoCommonElements}
Checks whether actual iterable and expected iterable are disjoint.
- - -
#### `tf.test.TestCase.assertNotAlmostEqual(first, second, places=None, msg=None, delta=None)` {#TestCase.assertNotAlmostEqual}
@ -587,33 +397,6 @@ as significant digits (measured from the most signficant digit).
Objects that are equal automatically fail.
- - -
#### `tf.test.TestCase.assertNotEmpty(container, msg=None)` {#TestCase.assertNotEmpty}
Assert that an object has non-zero length.
##### Args:
* <b>`container`</b>: Anything that implements the collections.Sized interface.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertNotEndsWith(actual, unexpected_end, msg=None)` {#TestCase.assertNotEndsWith}
Assert that actual.endswith(unexpected_end) is False.
##### Args:
* <b>`actual`</b>: str
* <b>`unexpected_end`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertNotEqual(first, second, msg=None)` {#TestCase.assertNotEqual}
@ -651,20 +434,6 @@ Included for symmetry with assertIsInstance.
Fail the test if the text matches the regular expression.
- - -
#### `tf.test.TestCase.assertNotStartsWith(actual, unexpected_start, msg=None)` {#TestCase.assertNotStartsWith}
Assert that actual.startswith(unexpected_start) is False.
##### Args:
* <b>`actual`</b>: str
* <b>`unexpected_start`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertProtoEquals(expected_message_maybe_ascii, message)` {#TestCase.assertProtoEquals}
@ -739,38 +508,6 @@ Asserts that the message in a raised exception matches a regexp.
* <b>`kwargs`</b>: Extra kwargs.
- - -
#### `tf.test.TestCase.assertRaisesWithLiteralMatch(expected_exception, expected_exception_message, callable_obj=None, *args, **kwargs)` {#TestCase.assertRaisesWithLiteralMatch}
Asserts that the message in a raised exception equals the given string.
Unlike assertRaisesRegexp, this method takes a literal string, not
a regular expression.
with self.assertRaisesWithLiteralMatch(ExType, 'message'):
DoSomething()
##### Args:
* <b>`expected_exception`</b>: Exception class expected to be raised.
* <b>`expected_exception_message`</b>: String message expected in the raised
exception. For a raise exception e, expected_exception_message must
equal str(e).
* <b>`callable_obj`</b>: Function to be called, or None to return a context.
* <b>`args`</b>: Extra args.
* <b>`kwargs`</b>: Extra kwargs.
##### Returns:
A context manager if callable_obj is None. Otherwise, None.
##### Raises:
self.failureException if callable_obj does not raise a macthing exception.
- - -
#### `tf.test.TestCase.assertRaisesWithPredicateMatch(exception_type, expected_err_re_or_predicate)` {#TestCase.assertRaisesWithPredicateMatch}
@ -795,71 +532,6 @@ predicate search.
exception.
- - -
#### `tf.test.TestCase.assertRaisesWithRegexpMatch(expected_exception, expected_regexp, callable_obj=None, *args, **kwargs)` {#TestCase.assertRaisesWithRegexpMatch}
Asserts that the message in a raised exception matches the given regexp.
This is just a wrapper around assertRaisesRegexp. Please use
assertRaisesRegexp instead of assertRaisesWithRegexpMatch.
##### Args:
* <b>`expected_exception`</b>: Exception class expected to be raised.
* <b>`expected_regexp`</b>: Regexp (re pattern object or string) expected to be
found in error message.
* <b>`callable_obj`</b>: Function to be called, or None to return a context.
* <b>`args`</b>: Extra args.
* <b>`kwargs`</b>: Extra keyword args.
##### Returns:
A context manager if callable_obj is None. Otherwise, None.
##### Raises:
self.failureException if callable_obj does not raise a macthing exception.
- - -
#### `tf.test.TestCase.assertRegexMatch(actual_str, regexes, message=None)` {#TestCase.assertRegexMatch}
Asserts that at least one regex in regexes matches str.
If possible you should use assertRegexpMatches, which is a simpler
version of this method. assertRegexpMatches takes a single regular
expression (a string or re compiled object) instead of a list.
Notes:
1. This function uses substring matching, i.e. the matching
succeeds if *any* substring of the error message matches *any*
regex in the list. This is more convenient for the user than
full-string matching.
2. If regexes is the empty list, the matching will always fail.
3. Use regexes=[''] for a regex that will always pass.
4. '.' matches any single character *except* the newline. To
match any character, use '(.|
)'.
5. '^' matches the beginning of each line, not just the beginning
of the string. Similarly, '$' matches the end of each line.
6. An exception will be thrown if regexes contains an invalid
regex.
Args:
actual_str: The string we try to match with the items in regexes.
regexes: The regular expressions we want to match against str.
See "Notes" above for detailed notes on how this is interpreted.
message: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertRegexpMatches(text, expected_regexp, msg=None)` {#TestCase.assertRegexpMatches}
@ -867,79 +539,6 @@ Asserts that at least one regex in regexes matches str.
Fail the test unless the text matches the regular expression.
- - -
#### `tf.test.TestCase.assertSameElements(expected_seq, actual_seq, msg=None)` {#TestCase.assertSameElements}
Assert that two sequences have the same elements (in any order).
This method, unlike assertItemsEqual, doesn't care about any
duplicates in the expected and actual sequences.
>> assertSameElements([1, 1, 1, 0, 0, 0], [0, 1])
# Doesn't raise an AssertionError
If possible, you should use assertItemsEqual instead of
assertSameElements.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertSameStructure(a, b, aname='a', bname='b', msg=None)` {#TestCase.assertSameStructure}
Asserts that two values contain the same structural content.
The two arguments should be data trees consisting of trees of dicts and
lists. They will be deeply compared by walking into the contents of dicts
and lists; other items will be compared using the == operator.
If the two structures differ in content, the failure message will indicate
the location within the structures where the first difference is found.
This may be helpful when comparing large structures.
##### Args:
* <b>`a`</b>: The first structure to compare.
* <b>`b`</b>: The second structure to compare.
* <b>`aname`</b>: Variable name to use for the first structure in assertion messages.
* <b>`bname`</b>: Variable name to use for the second structure.
* <b>`msg`</b>: Additional text to include in the failure message.
- - -
#### `tf.test.TestCase.assertSequenceAlmostEqual(expected_seq, actual_seq, places=None, msg=None, delta=None)` {#TestCase.assertSequenceAlmostEqual}
An approximate equality assertion for ordered sequences.
Fail if the two sequences are unequal as determined by their value
differences rounded to the given number of decimal places (default 7) and
comparing to zero, or by comparing that the difference between each value
in the two sequences is more than the given delta.
Note that decimal places (from zero) are usually not the same as significant
digits (measured from the most signficant digit).
If the two sequences compare equal then they will automatically compare
almost equal.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`places`</b>: The number of decimal places to compare.
* <b>`msg`</b>: The message to be printed if the test fails.
* <b>`delta`</b>: The OK difference between compared values.
- - -
#### `tf.test.TestCase.assertSequenceEqual(seq1, seq2, msg=None, seq_type=None)` {#TestCase.assertSequenceEqual}
@ -960,26 +559,6 @@ which can be indexed, has a length, and has an equality operator.
differences.
- - -
#### `tf.test.TestCase.assertSequenceStartsWith(prefix, whole, msg=None)` {#TestCase.assertSequenceStartsWith}
An equality assertion for the beginning of ordered sequences.
If prefix is an empty sequence, it will raise an error unless whole is also
an empty sequence.
If prefix is not a sequence, it will raise an error if the first element of
whole does not match.
##### Args:
* <b>`prefix`</b>: A sequence expected at the beginning of the whole parameter.
* <b>`whole`</b>: The sequence in which to look for prefix.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertSetEqual(set1, set2, msg=None)` {#TestCase.assertSetEqual}
@ -1031,51 +610,6 @@ Assert that actual.startswith(expected_start) is True.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertTotallyOrdered(*groups, **kwargs)` {#TestCase.assertTotallyOrdered}
Asserts that total ordering has been implemented correctly.
For example, say you have a class A that compares only on its attribute x.
Comparators other than __lt__ are omitted for brevity.
class A(object):
def __init__(self, x, y):
self.x = x
self.y = y
def __hash__(self):
return hash(self.x)
def __lt__(self, other):
try:
return self.x < other.x
except AttributeError:
return NotImplemented
assertTotallyOrdered will check that instances can be ordered correctly.
For example,
self.assertTotallyOrdered(
[None], # None should come before everything else.
[1], # Integers sort earlier.
[A(1, 'a')],
[A(2, 'b')], # 2 is after 1.
[A(3, 'c'), A(3, 'd')], # The second argument is irrelevant.
[A(4, 'z')],
['foo']) # Strings sort last.
##### Args:
* <b>`*groups`</b>: A list of groups of elements. Each group of elements is a list
of objects that are equal. The elements in each group must be less than
the elements in the group after it. For example, these groups are
totally ordered: [None], [1], [2, 2], [3].
* <b>`**kwargs`</b>: optional msg keyword argument can be passed.
- - -
#### `tf.test.TestCase.assertTrue(expr, msg=None)` {#TestCase.assertTrue}
@ -1098,13 +632,6 @@ A tuple-specific equality assertion.
differences.
- - -
#### `tf.test.TestCase.assertUrlEqual(a, b, msg=None)` {#TestCase.assertUrlEqual}
Asserts that urls are equal, ignoring ordering of query params.
- - -
#### `tf.test.TestCase.assert_(expr, msg=None)` {#TestCase.assert_}
@ -1166,9 +693,9 @@ tearDown.
- - -
#### `tf.test.TestCase.fail(msg=None, prefix=None)` {#TestCase.fail}
#### `tf.test.TestCase.fail(msg=None)` {#TestCase.fail}
Fail immediately with the given message, optionally prefixed.
Fail immediately, with the given message.
- - -
@ -1220,13 +747,6 @@ Fail immediately with the given message, optionally prefixed.
- - -
#### `tf.test.TestCase.getRecordedProperties()` {#TestCase.getRecordedProperties}
Return any properties that the user has recorded.
- - -
#### `tf.test.TestCase.get_temp_dir()` {#TestCase.get_temp_dir}
@ -1241,20 +761,6 @@ Return any properties that the user has recorded.
- - -
#### `tf.test.TestCase.recordProperty(property_name, property_value)` {#TestCase.recordProperty}
Record an arbitrary property for later use.
##### Args:
* <b>`property_name`</b>: str, name of property to record; must be a valid XML
attribute name
* <b>`property_value`</b>: value of property; must be valid XML attribute value
- - -
#### `tf.test.TestCase.run(result=None)` {#TestCase.run}
@ -1280,18 +786,11 @@ Hook method for setting up class fixture before running tests in the class.
#### `tf.test.TestCase.shortDescription()` {#TestCase.shortDescription}
Format both the test method name and the first line of its docstring.
Returns a one-line description of the test, or None if no
description has been provided.
If no docstring is given, only returns the method name.
This method overrides unittest.TestCase.shortDescription(), which
only returns the first line of the docstring, obscuring the name
of the test upon failure.
##### Returns:
* <b>`desc`</b>: A short description of a test method.
The default implementation of this method returns the first line of
the specified test method's docstring.
- - -

22
tensorflow/g3doc/api_docs/python/functions_and_classes/shard3/tf.scalar_summary.md Normal file
View File

@ -0,0 +1,22 @@
### `tf.scalar_summary(*args, **kwargs)` {#scalar_summary}
Outputs a `Summary` protocol buffer with scalar values. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.scalar. Note that tf.summary.scalar uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in. Also, passing a tensor or list of tags to a scalar summary op is no longer supported.
The input `tags` and `values` must have the same shape. The generated
summary has a summary value for each tag-value pair in `tags` and `values`.
Args:
tags: A `string` `Tensor`. Tags for the summaries.
values: A real numeric Tensor. Values for the summaries.
collections: Optional list of graph collections keys. The new summary op is
added to these collections. Defaults to `[GraphKeys.SUMMARIES]`.
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`. The serialized `Summary` protocol
buffer.

4
tensorflow/g3doc/api_docs/python/functions_and_classes/shard4/tf.summary.SummaryDescription.RegisterExtension.md Normal file
View File

@ -0,0 +1,4 @@
#### `tf.summary.SummaryDescription.RegisterExtension(extension_handle)` {#SummaryDescription.RegisterExtension}

26
tensorflow/g3doc/api_docs/python/functions_and_classes/shard5/tf.histogram_summary.md Normal file
View File

@ -0,0 +1,26 @@
### `tf.histogram_summary(*args, **kwargs)` {#histogram_summary}
Outputs a `Summary` protocol buffer with a histogram. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.histogram. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on their scope.
The generated
[`Summary`](https://www.tensorflow.org/code/tensorflow/core/framework/summary.proto)
has one summary value containing a histogram for `values`.
This op reports an `InvalidArgument` error if any value is not finite.
Args:
tag: A `string` `Tensor`. 0-D. Tag to use for the summary value.
values: A real numeric `Tensor`. Any shape. Values to use to
build the histogram.
collections: Optional list of graph collections keys. The new summary op is
added to these collections. Defaults to `[GraphKeys.SUMMARIES]`.
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`. The serialized `Summary` protocol
buffer.

27
tensorflow/g3doc/api_docs/python/functions_and_classes/shard5/tf.merge_summary.md Normal file
View File

@ -0,0 +1,27 @@
### `tf.merge_summary(*args, **kwargs)` {#merge_summary}
Merges summaries. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.merge.
This op creates a
[`Summary`](https://www.tensorflow.org/code/tensorflow/core/framework/summary.proto)
protocol buffer that contains the union of all the values in the input
summaries.
When the Op is run, it reports an `InvalidArgument` error if multiple values
in the summaries to merge use the same tag.
Args:
inputs: A list of `string` `Tensor` objects containing serialized `Summary`
protocol buffers.
collections: Optional list of graph collections keys. The new summary op is
added to these collections. Defaults to `[GraphKeys.SUMMARIES]`.
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`. The serialized `Summary` protocol
buffer resulting from the merging.

4
tensorflow/g3doc/api_docs/python/functions_and_classes/shard5/tf.summary.SummaryDescription.FromString.md Normal file
View File

@ -0,0 +1,4 @@
#### `tf.summary.SummaryDescription.FromString(s)` {#SummaryDescription.FromString}

4
tensorflow/g3doc/api_docs/python/functions_and_classes/shard6/tf.summary.TaggedRunMetadata.RegisterExtension.md Normal file
View File

@ -0,0 +1,4 @@
#### `tf.summary.TaggedRunMetadata.RegisterExtension(extension_handle)` {#TaggedRunMetadata.RegisterExtension}

207
tensorflow/g3doc/api_docs/python/functions_and_classes/shard7/tf.train.SummaryWriter.md Normal file
View File

@ -0,0 +1,207 @@
- - -
#### `tf.train.SummaryWriter.__init__(*args, **kwargs)` {#SummaryWriter.__init__}
Creates a `SummaryWriter` and an event file. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.FileWriter. The interface and behavior is the same; this is just a rename.
This class is deprecated, and should be replaced with tf.summary.FileWriter.
On construction the summary writer creates a new event file in `logdir`.
This event file will contain `Event` protocol buffers constructed when you
call one of the following functions: `add_summary()`, `add_session_log()`,
`add_event()`, or `add_graph()`.
If you pass a `Graph` to the constructor it is added to
the event file. (This is equivalent to calling `add_graph()` later).
TensorBoard will pick the graph from the file and display it graphically so
you can interactively explore the graph you built. You will usually pass
the graph from the session in which you launched it:
```python
...create a graph...
# Launch the graph in a session.
sess = tf.Session()
# Create a summary writer, add the 'graph' to the event file.
writer = tf.train.SummaryWriter(<some-directory>, sess.graph)
```
The other arguments to the constructor control the asynchronous writes to
the event file:
* `flush_secs`: How often, in seconds, to flush the added summaries
and events to disk.
* `max_queue`: Maximum number of summaries or events pending to be
written to disk before one of the 'add' calls block.
Args:
logdir: A string. Directory where event file will be written.
graph: A `Graph` object, such as `sess.graph`.
max_queue: Integer. Size of the queue for pending events and summaries.
flush_secs: Number. How often, in seconds, to flush the
pending events and summaries to disk.
graph_def: DEPRECATED: Use the `graph` argument instead.
- - -
#### `tf.train.SummaryWriter.add_event(event)` {#SummaryWriter.add_event}
Adds an event to the event file.
##### Args:
* <b>`event`</b>: An `Event` protocol buffer.
- - -
#### `tf.train.SummaryWriter.add_graph(graph, global_step=None, graph_def=None)` {#SummaryWriter.add_graph}
Adds a `Graph` to the event file.
The graph described by the protocol buffer will be displayed by
TensorBoard. Most users pass a graph in the constructor instead.
##### Args:
* <b>`graph`</b>: A `Graph` object, such as `sess.graph`.
* <b>`global_step`</b>: Number. Optional global step counter to record with the
graph.
* <b>`graph_def`</b>: DEPRECATED. Use the `graph` parameter instead.
##### Raises:
* <b>`ValueError`</b>: If both graph and graph_def are passed to the method.
- - -
#### `tf.train.SummaryWriter.add_meta_graph(meta_graph_def, global_step=None)` {#SummaryWriter.add_meta_graph}
Adds a `MetaGraphDef` to the event file.
The `MetaGraphDef` allows running the given graph via
`saver.import_meta_graph()`.
##### Args:
* <b>`meta_graph_def`</b>: A `MetaGraphDef` object, often as retured by
`saver.export_meta_graph()`.
* <b>`global_step`</b>: Number. Optional global step counter to record with the
graph.
##### Raises:
* <b>`TypeError`</b>: If both `meta_graph_def` is not an instance of `MetaGraphDef`.
- - -
#### `tf.train.SummaryWriter.add_run_metadata(run_metadata, tag, global_step=None)` {#SummaryWriter.add_run_metadata}
Adds a metadata information for a single session.run() call.
##### Args:
* <b>`run_metadata`</b>: A `RunMetadata` protobuf object.
* <b>`tag`</b>: The tag name for this metadata.
* <b>`global_step`</b>: Number. Optional global step counter to record with the
StepStats.
##### Raises:
* <b>`ValueError`</b>: If the provided tag was already used for this type of event.
- - -
#### `tf.train.SummaryWriter.add_session_log(session_log, global_step=None)` {#SummaryWriter.add_session_log}
Adds a `SessionLog` protocol buffer to the event file.
This method wraps the provided session in an `Event` protocol buffer
and adds it to the event file.
##### Args:
* <b>`session_log`</b>: A `SessionLog` protocol buffer.
* <b>`global_step`</b>: Number. Optional global step value to record with the
summary.
- - -
#### `tf.train.SummaryWriter.add_summary(summary, global_step=None)` {#SummaryWriter.add_summary}
Adds a `Summary` protocol buffer to the event file.
This method wraps the provided summary in an `Event` protocol buffer
and adds it to the event file.
You can pass the result of evaluating any summary op, using
[`Session.run()`](client.md#Session.run) or
[`Tensor.eval()`](framework.md#Tensor.eval), to this
function. Alternatively, you can pass a `tf.Summary` protocol
buffer that you populate with your own data. The latter is
commonly done to report evaluation results in event files.
##### Args:
* <b>`summary`</b>: A `Summary` protocol buffer, optionally serialized as a string.
* <b>`global_step`</b>: Number. Optional global step value to record with the
summary.
- - -
#### `tf.train.SummaryWriter.close()` {#SummaryWriter.close}
Flushes the event file to disk and close the file.
Call this method when you do not need the summary writer anymore.
- - -
#### `tf.train.SummaryWriter.flush()` {#SummaryWriter.flush}
Flushes the event file to disk.
Call this method to make sure that all pending events have been written to
disk.
- - -
#### `tf.train.SummaryWriter.get_logdir()` {#SummaryWriter.get_logdir}
Returns the directory where event file will be written.
- - -
#### `tf.train.SummaryWriter.reopen()` {#SummaryWriter.reopen}
Reopens the EventFileWriter.
Can be called after `close()` to add more events in the same directory.
The events will go into a new events file.
Does nothing if the EventFileWriter was not closed.

37
tensorflow/g3doc/api_docs/python/functions_and_classes/shard9/tf.audio_summary.md Normal file
View File

@ -0,0 +1,37 @@
### `tf.audio_summary(*args, **kwargs)` {#audio_summary}
Outputs a `Summary` protocol buffer with audio. (deprecated)
THIS FUNCTION IS DEPRECATED. It will be removed after 2016-11-30.
Instructions for updating:
Please switch to tf.summary.audio. Note that tf.summary.histogram uses the node name instead of the tag. This means that TensorFlow will automatically de-duplicate summary names based on the scope they are created in.
The summary has up to `max_outputs` summary values containing audio. The
audio is built from `tensor` which must be 3-D with shape `[batch_size,
frames, channels]` or 2-D with shape `[batch_size, frames]`. The values are
assumed to be in the range of `[-1.0, 1.0]` with a sample rate of
`sample_rate`.
The `tag` argument is a scalar `Tensor` of type `string`. It is used to
build the `tag` of the summary values:
* If `max_outputs` is 1, the summary value tag is '*tag*/audio'.
* If `max_outputs` is greater than 1, the summary value tags are
generated sequentially as '*tag*/audio/0', '*tag*/audio/1', etc.
Args:
tag: A scalar `Tensor` of type `string`. Used to build the `tag`
of the summary values.
tensor: A 3-D `float32` `Tensor` of shape `[batch_size, frames, channels]`
or a 2-D `float32` `Tensor` of shape `[batch_size, frames]`.
sample_rate: A Scalar `float32` `Tensor` indicating the sample rate of the
signal in hertz.
max_outputs: Max number of batch elements to generate audio for.
collections: Optional list of ops.GraphKeys. The collections to add the
summary to. Defaults to [ops.GraphKeys.SUMMARIES]
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`. The serialized `Summary` protocol
buffer.

4
tensorflow/g3doc/api_docs/python/functions_and_classes/shard9/tf.summary.TaggedRunMetadata.FromString.md Normal file
View File

@ -0,0 +1,4 @@
#### `tf.summary.TaggedRunMetadata.FromString(s)` {#TaggedRunMetadata.FromString}

481
tensorflow/g3doc/api_docs/python/summary.md
View File

@ -485,6 +485,187 @@ metadata is stored in its NodeDef. This method retrieves the description.
### `class tf.summary.SummaryDescription` {#SummaryDescription}
- - -
#### `tf.summary.SummaryDescription.ByteSize()` {#SummaryDescription.ByteSize}
- - -
#### `tf.summary.SummaryDescription.Clear()` {#SummaryDescription.Clear}
- - -
#### `tf.summary.SummaryDescription.ClearExtension(extension_handle)` {#SummaryDescription.ClearExtension}
- - -
#### `tf.summary.SummaryDescription.ClearField(field_name)` {#SummaryDescription.ClearField}
- - -
#### `tf.summary.SummaryDescription.CopyFrom(other_msg)` {#SummaryDescription.CopyFrom}
Copies the content of the specified message into the current message.
The method clears the current message and then merges the specified
message using MergeFrom.
##### Args:
* <b>`other_msg`</b>: Message to copy into the current one.
- - -
#### `tf.summary.SummaryDescription.DiscardUnknownFields()` {#SummaryDescription.DiscardUnknownFields}
- - -
#### `tf.summary.SummaryDescription.FindInitializationErrors()` {#SummaryDescription.FindInitializationErrors}
Finds required fields which are not initialized.
##### Returns:
A list of strings. Each string is a path to an uninitialized field from
the top-level message, e.g. "foo.bar[5].baz".
- - -
#### `tf.summary.SummaryDescription.FromString(s)` {#SummaryDescription.FromString}
- - -
#### `tf.summary.SummaryDescription.HasExtension(extension_handle)` {#SummaryDescription.HasExtension}
- - -
#### `tf.summary.SummaryDescription.HasField(field_name)` {#SummaryDescription.HasField}
- - -
#### `tf.summary.SummaryDescription.IsInitialized(errors=None)` {#SummaryDescription.IsInitialized}
Checks if all required fields of a message are set.
##### Args:
* <b>`errors`</b>: A list which, if provided, will be populated with the field
paths of all missing required fields.
##### Returns:
True iff the specified message has all required fields set.
- - -
#### `tf.summary.SummaryDescription.ListFields()` {#SummaryDescription.ListFields}
- - -
#### `tf.summary.SummaryDescription.MergeFrom(msg)` {#SummaryDescription.MergeFrom}
- - -
#### `tf.summary.SummaryDescription.MergeFromString(serialized)` {#SummaryDescription.MergeFromString}
- - -
#### `tf.summary.SummaryDescription.ParseFromString(serialized)` {#SummaryDescription.ParseFromString}
Parse serialized protocol buffer data into this message.
Like MergeFromString(), except we clear the object first and
do not return the value that MergeFromString returns.
- - -
#### `tf.summary.SummaryDescription.RegisterExtension(extension_handle)` {#SummaryDescription.RegisterExtension}
- - -
#### `tf.summary.SummaryDescription.SerializePartialToString()` {#SummaryDescription.SerializePartialToString}
- - -
#### `tf.summary.SummaryDescription.SerializeToString()` {#SummaryDescription.SerializeToString}
- - -
#### `tf.summary.SummaryDescription.SetInParent()` {#SummaryDescription.SetInParent}
Sets the _cached_byte_size_dirty bit to true,
and propagates this to our listener iff this was a state change.
- - -
#### `tf.summary.SummaryDescription.WhichOneof(oneof_name)` {#SummaryDescription.WhichOneof}
Returns the name of the currently set field inside a oneof, or None.
- - -
#### `tf.summary.SummaryDescription.__deepcopy__(memo=None)` {#SummaryDescription.__deepcopy__}
- - -
#### `tf.summary.SummaryDescription.__eq__(other)` {#SummaryDescription.__eq__}
- - -
#### `tf.summary.SummaryDescription.__getstate__()` {#SummaryDescription.__getstate__}
@ -492,12 +673,249 @@ metadata is stored in its NodeDef. This method retrieves the description.
Support the pickle protocol.
- - -
#### `tf.summary.SummaryDescription.__hash__()` {#SummaryDescription.__hash__}
- - -
#### `tf.summary.SummaryDescription.__init__(**kwargs)` {#SummaryDescription.__init__}
- - -
#### `tf.summary.SummaryDescription.__ne__(other_msg)` {#SummaryDescription.__ne__}
- - -
#### `tf.summary.SummaryDescription.__repr__()` {#SummaryDescription.__repr__}
- - -
#### `tf.summary.SummaryDescription.__setstate__(state)` {#SummaryDescription.__setstate__}
Support the pickle protocol.
- - -
#### `tf.summary.SummaryDescription.__str__()` {#SummaryDescription.__str__}
- - -
#### `tf.summary.SummaryDescription.__unicode__()` {#SummaryDescription.__unicode__}
- - -
#### `tf.summary.SummaryDescription.type_hint` {#SummaryDescription.type_hint}
Magic attribute generated for "type_hint" proto field.
- - -
### `class tf.summary.TaggedRunMetadata` {#TaggedRunMetadata}
- - -
#### `tf.summary.TaggedRunMetadata.ByteSize()` {#TaggedRunMetadata.ByteSize}
- - -
#### `tf.summary.TaggedRunMetadata.Clear()` {#TaggedRunMetadata.Clear}
- - -
#### `tf.summary.TaggedRunMetadata.ClearExtension(extension_handle)` {#TaggedRunMetadata.ClearExtension}
- - -
#### `tf.summary.TaggedRunMetadata.ClearField(field_name)` {#TaggedRunMetadata.ClearField}
- - -
#### `tf.summary.TaggedRunMetadata.CopyFrom(other_msg)` {#TaggedRunMetadata.CopyFrom}
Copies the content of the specified message into the current message.
The method clears the current message and then merges the specified
message using MergeFrom.
##### Args:
* <b>`other_msg`</b>: Message to copy into the current one.
- - -
#### `tf.summary.TaggedRunMetadata.DiscardUnknownFields()` {#TaggedRunMetadata.DiscardUnknownFields}
- - -
#### `tf.summary.TaggedRunMetadata.FindInitializationErrors()` {#TaggedRunMetadata.FindInitializationErrors}
Finds required fields which are not initialized.
##### Returns:
A list of strings. Each string is a path to an uninitialized field from
the top-level message, e.g. "foo.bar[5].baz".
- - -
#### `tf.summary.TaggedRunMetadata.FromString(s)` {#TaggedRunMetadata.FromString}
- - -
#### `tf.summary.TaggedRunMetadata.HasExtension(extension_handle)` {#TaggedRunMetadata.HasExtension}
- - -
#### `tf.summary.TaggedRunMetadata.HasField(field_name)` {#TaggedRunMetadata.HasField}
- - -
#### `tf.summary.TaggedRunMetadata.IsInitialized(errors=None)` {#TaggedRunMetadata.IsInitialized}
Checks if all required fields of a message are set.
##### Args:
* <b>`errors`</b>: A list which, if provided, will be populated with the field
paths of all missing required fields.
##### Returns:
True iff the specified message has all required fields set.
- - -
#### `tf.summary.TaggedRunMetadata.ListFields()` {#TaggedRunMetadata.ListFields}
- - -
#### `tf.summary.TaggedRunMetadata.MergeFrom(msg)` {#TaggedRunMetadata.MergeFrom}
- - -
#### `tf.summary.TaggedRunMetadata.MergeFromString(serialized)` {#TaggedRunMetadata.MergeFromString}
- - -
#### `tf.summary.TaggedRunMetadata.ParseFromString(serialized)` {#TaggedRunMetadata.ParseFromString}
Parse serialized protocol buffer data into this message.
Like MergeFromString(), except we clear the object first and
do not return the value that MergeFromString returns.
- - -
#### `tf.summary.TaggedRunMetadata.RegisterExtension(extension_handle)` {#TaggedRunMetadata.RegisterExtension}
- - -
#### `tf.summary.TaggedRunMetadata.SerializePartialToString()` {#TaggedRunMetadata.SerializePartialToString}
- - -
#### `tf.summary.TaggedRunMetadata.SerializeToString()` {#TaggedRunMetadata.SerializeToString}
- - -
#### `tf.summary.TaggedRunMetadata.SetInParent()` {#TaggedRunMetadata.SetInParent}
Sets the _cached_byte_size_dirty bit to true,
and propagates this to our listener iff this was a state change.
- - -
#### `tf.summary.TaggedRunMetadata.WhichOneof(oneof_name)` {#TaggedRunMetadata.WhichOneof}
Returns the name of the currently set field inside a oneof, or None.
- - -
#### `tf.summary.TaggedRunMetadata.__deepcopy__(memo=None)` {#TaggedRunMetadata.__deepcopy__}
- - -
#### `tf.summary.TaggedRunMetadata.__eq__(other)` {#TaggedRunMetadata.__eq__}
- - -
#### `tf.summary.TaggedRunMetadata.__getstate__()` {#TaggedRunMetadata.__getstate__}
@ -505,4 +923,67 @@ Support the pickle protocol.
Support the pickle protocol.
- - -
#### `tf.summary.TaggedRunMetadata.__hash__()` {#TaggedRunMetadata.__hash__}
- - -
#### `tf.summary.TaggedRunMetadata.__init__(**kwargs)` {#TaggedRunMetadata.__init__}
- - -
#### `tf.summary.TaggedRunMetadata.__ne__(other_msg)` {#TaggedRunMetadata.__ne__}
- - -
#### `tf.summary.TaggedRunMetadata.__repr__()` {#TaggedRunMetadata.__repr__}
- - -
#### `tf.summary.TaggedRunMetadata.__setstate__(state)` {#TaggedRunMetadata.__setstate__}
Support the pickle protocol.
- - -
#### `tf.summary.TaggedRunMetadata.__str__()` {#TaggedRunMetadata.__str__}
- - -
#### `tf.summary.TaggedRunMetadata.__unicode__()` {#TaggedRunMetadata.__unicode__}
- - -
#### `tf.summary.TaggedRunMetadata.run_metadata` {#TaggedRunMetadata.run_metadata}
Magic attribute generated for "run_metadata" proto field.
- - -
#### `tf.summary.TaggedRunMetadata.tag` {#TaggedRunMetadata.tag}
Magic attribute generated for "tag" proto field.

521
tensorflow/g3doc/api_docs/python/test.md
View File

@ -213,125 +213,6 @@ Checks that for all elements of farray1 and farray2
* <b>`err`</b>: a float value.
- - -
#### `tf.test.TestCase.assertBetween(value, minv, maxv, msg=None)` {#TestCase.assertBetween}
Asserts that value is between minv and maxv (inclusive).
- - -
#### `tf.test.TestCase.assertCommandFails(command, regexes, env=None, close_fds=True, msg=None)` {#TestCase.assertCommandFails}
Asserts a shell command fails and the error matches a regex in a list.
##### Args:
* <b>`command`</b>: List or string representing the command to run.
* <b>`regexes`</b>: the list of regular expression strings.
* <b>`env`</b>: Dictionary of environment variable settings.
* <b>`close_fds`</b>: Whether or not to close all open fd's in the child after
forking.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertCommandSucceeds(command, regexes=('',), env=None, close_fds=True, msg=None)` {#TestCase.assertCommandSucceeds}
Asserts that a shell command succeeds (i.e. exits with code 0).
##### Args:
* <b>`command`</b>: List or string representing the command to run.
* <b>`regexes`</b>: List of regular expression byte strings that match success.
* <b>`env`</b>: Dictionary of environment variable settings.
* <b>`close_fds`</b>: Whether or not to close all open fd's in the child after
forking.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsExactSubsequence(container, subsequence, msg=None)` {#TestCase.assertContainsExactSubsequence}
Assert that "container" contains "subsequence" as an exact subsequence.
Asserts that "container" contains all the elements of "subsequence", in
order, and without other elements interspersed. For example, [1, 2, 3] is an
exact subsequence of [0, 0, 1, 2, 3, 0] but not of [0, 0, 1, 2, 0, 3, 0].
##### Args:
* <b>`container`</b>: the list we're testing for subsequence inclusion.
* <b>`subsequence`</b>: the list we hope will be an exact subsequence of container.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsInOrder(strings, target, msg=None)` {#TestCase.assertContainsInOrder}
Asserts that the strings provided are found in the target in order.
This may be useful for checking HTML output.
##### Args:
* <b>`strings`</b>: A list of strings, such as [ 'fox', 'dog' ]
* <b>`target`</b>: A target string in which to look for the strings, such as
'The quick brown fox jumped over the lazy dog'.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsSubsequence(container, subsequence, msg=None)` {#TestCase.assertContainsSubsequence}
Assert that "container" contains "subsequence" as a subsequence.
Asserts that "container" contains all the elements of "subsequence", in
order, but possibly with other elements interspersed. For example, [1, 2, 3]
is a subsequence of [0, 0, 1, 2, 0, 3, 0] but not of [0, 0, 1, 3, 0, 2, 0].
##### Args:
* <b>`container`</b>: the list we're testing for subsequence inclusion.
* <b>`subsequence`</b>: the list we hope will be a subsequence of container.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertContainsSubset(expected_subset, actual_set, msg=None)` {#TestCase.assertContainsSubset}
Checks whether actual iterable is a superset of expected iterable.
- - -
#### `tf.test.TestCase.assertCountEqual(*args, **kwargs)` {#TestCase.assertCountEqual}
An unordered sequence specific comparison.
Equivalent to assertItemsEqual(). This method is a compatibility layer
for Python 3k, since 2to3 does not convert assertItemsEqual() calls into
assertCountEqual() calls.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertDeviceEqual(device1, device2)` {#TestCase.assertDeviceEqual}
@ -354,48 +235,9 @@ Checks whether actual is a superset of expected.
- - -
#### `tf.test.TestCase.assertDictEqual(a, b, msg=None)` {#TestCase.assertDictEqual}
Raises AssertionError if a and b are not equal dictionaries.
##### Args:
#### `tf.test.TestCase.assertDictEqual(d1, d2, msg=None)` {#TestCase.assertDictEqual}
* <b>`a`</b>: A dict, the expected value.
* <b>`b`</b>: A dict, the actual value.
* <b>`msg`</b>: An optional str, the associated message.
##### Raises:
* <b>`AssertionError`</b>: if the dictionaries are not equal.
- - -
#### `tf.test.TestCase.assertEmpty(container, msg=None)` {#TestCase.assertEmpty}
Assert that an object has zero length.
##### Args:
* <b>`container`</b>: Anything that implements the collections.Sized interface.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertEndsWith(actual, expected_end, msg=None)` {#TestCase.assertEndsWith}
Assert that actual.endswith(expected_end) is True.
##### Args:
* <b>`actual`</b>: str
* <b>`expected_end`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
@ -480,11 +322,10 @@ Included for symmetry with assertIsNone.
- - -
#### `tf.test.TestCase.assertItemsEqual(*args, **kwargs)` {#TestCase.assertItemsEqual}
#### `tf.test.TestCase.assertItemsEqual(expected_seq, actual_seq, msg=None)` {#TestCase.assertItemsEqual}
An unordered sequence specific comparison.
It asserts that actual_seq and expected_seq have the same element counts.
An unordered sequence specific comparison. It asserts that
actual_seq and expected_seq have the same element counts.
Equivalent to::
self.assertEqual(Counter(iter(actual_seq)),
@ -497,30 +338,6 @@ Asserts that each element has the same count in both sequences.
- [0, 1, 1] and [1, 0, 1] compare equal.
- [0, 0, 1] and [0, 1] compare unequal.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertJsonEqual(first, second, msg=None)` {#TestCase.assertJsonEqual}
Asserts that the JSON objects defined in two strings are equal.
A summary of the differences will be included in the failure message
using assertSameStructure.
##### Args:
* <b>`first`</b>: A string contining JSON to decode and compare to second.
* <b>`second`</b>: A string contining JSON to decode and compare to first.
* <b>`msg`</b>: Additional text to include in the failure message.
- - -
@ -590,13 +407,6 @@ if not.
* <b>`msg`</b>: An optional string message to append to the failure message.
- - -
#### `tf.test.TestCase.assertNoCommonElements(expected_seq, actual_seq, msg=None)` {#TestCase.assertNoCommonElements}
Checks whether actual iterable and expected iterable are disjoint.
- - -
#### `tf.test.TestCase.assertNotAlmostEqual(first, second, places=None, msg=None, delta=None)` {#TestCase.assertNotAlmostEqual}
@ -627,33 +437,6 @@ as significant digits (measured from the most signficant digit).
Objects that are equal automatically fail.
- - -
#### `tf.test.TestCase.assertNotEmpty(container, msg=None)` {#TestCase.assertNotEmpty}
Assert that an object has non-zero length.
##### Args:
* <b>`container`</b>: Anything that implements the collections.Sized interface.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertNotEndsWith(actual, unexpected_end, msg=None)` {#TestCase.assertNotEndsWith}
Assert that actual.endswith(unexpected_end) is False.
##### Args:
* <b>`actual`</b>: str
* <b>`unexpected_end`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertNotEqual(first, second, msg=None)` {#TestCase.assertNotEqual}
@ -691,20 +474,6 @@ Included for symmetry with assertIsInstance.
Fail the test if the text matches the regular expression.
- - -
#### `tf.test.TestCase.assertNotStartsWith(actual, unexpected_start, msg=None)` {#TestCase.assertNotStartsWith}
Assert that actual.startswith(unexpected_start) is False.
##### Args:
* <b>`actual`</b>: str
* <b>`unexpected_start`</b>: str
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertProtoEquals(expected_message_maybe_ascii, message)` {#TestCase.assertProtoEquals}
@ -779,38 +548,6 @@ Asserts that the message in a raised exception matches a regexp.
* <b>`kwargs`</b>: Extra kwargs.
- - -
#### `tf.test.TestCase.assertRaisesWithLiteralMatch(expected_exception, expected_exception_message, callable_obj=None, *args, **kwargs)` {#TestCase.assertRaisesWithLiteralMatch}
Asserts that the message in a raised exception equals the given string.
Unlike assertRaisesRegexp, this method takes a literal string, not
a regular expression.
with self.assertRaisesWithLiteralMatch(ExType, 'message'):
DoSomething()
##### Args:
* <b>`expected_exception`</b>: Exception class expected to be raised.
* <b>`expected_exception_message`</b>: String message expected in the raised
exception. For a raise exception e, expected_exception_message must
equal str(e).
* <b>`callable_obj`</b>: Function to be called, or None to return a context.
* <b>`args`</b>: Extra args.
* <b>`kwargs`</b>: Extra kwargs.
##### Returns:
A context manager if callable_obj is None. Otherwise, None.
##### Raises:
self.failureException if callable_obj does not raise a macthing exception.
- - -
#### `tf.test.TestCase.assertRaisesWithPredicateMatch(exception_type, expected_err_re_or_predicate)` {#TestCase.assertRaisesWithPredicateMatch}
@ -835,71 +572,6 @@ predicate search.
exception.
- - -
#### `tf.test.TestCase.assertRaisesWithRegexpMatch(expected_exception, expected_regexp, callable_obj=None, *args, **kwargs)` {#TestCase.assertRaisesWithRegexpMatch}
Asserts that the message in a raised exception matches the given regexp.
This is just a wrapper around assertRaisesRegexp. Please use
assertRaisesRegexp instead of assertRaisesWithRegexpMatch.
##### Args:
* <b>`expected_exception`</b>: Exception class expected to be raised.
* <b>`expected_regexp`</b>: Regexp (re pattern object or string) expected to be
found in error message.
* <b>`callable_obj`</b>: Function to be called, or None to return a context.
* <b>`args`</b>: Extra args.
* <b>`kwargs`</b>: Extra keyword args.
##### Returns:
A context manager if callable_obj is None. Otherwise, None.
##### Raises:
self.failureException if callable_obj does not raise a macthing exception.
- - -
#### `tf.test.TestCase.assertRegexMatch(actual_str, regexes, message=None)` {#TestCase.assertRegexMatch}
Asserts that at least one regex in regexes matches str.
If possible you should use assertRegexpMatches, which is a simpler
version of this method. assertRegexpMatches takes a single regular
expression (a string or re compiled object) instead of a list.
Notes:
1. This function uses substring matching, i.e. the matching
succeeds if *any* substring of the error message matches *any*
regex in the list. This is more convenient for the user than
full-string matching.
2. If regexes is the empty list, the matching will always fail.
3. Use regexes=[''] for a regex that will always pass.
4. '.' matches any single character *except* the newline. To
match any character, use '(.|
)'.
5. '^' matches the beginning of each line, not just the beginning
of the string. Similarly, '$' matches the end of each line.
6. An exception will be thrown if regexes contains an invalid
regex.
Args:
actual_str: The string we try to match with the items in regexes.
regexes: The regular expressions we want to match against str.
See "Notes" above for detailed notes on how this is interpreted.
message: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertRegexpMatches(text, expected_regexp, msg=None)` {#TestCase.assertRegexpMatches}
@ -907,79 +579,6 @@ Asserts that at least one regex in regexes matches str.
Fail the test unless the text matches the regular expression.
- - -
#### `tf.test.TestCase.assertSameElements(expected_seq, actual_seq, msg=None)` {#TestCase.assertSameElements}
Assert that two sequences have the same elements (in any order).
This method, unlike assertItemsEqual, doesn't care about any
duplicates in the expected and actual sequences.
>> assertSameElements([1, 1, 1, 0, 0, 0], [0, 1])
# Doesn't raise an AssertionError
If possible, you should use assertItemsEqual instead of
assertSameElements.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`msg`</b>: The message to be printed if the test fails.
- - -
#### `tf.test.TestCase.assertSameStructure(a, b, aname='a', bname='b', msg=None)` {#TestCase.assertSameStructure}
Asserts that two values contain the same structural content.
The two arguments should be data trees consisting of trees of dicts and
lists. They will be deeply compared by walking into the contents of dicts
and lists; other items will be compared using the == operator.
If the two structures differ in content, the failure message will indicate
the location within the structures where the first difference is found.
This may be helpful when comparing large structures.
##### Args:
* <b>`a`</b>: The first structure to compare.
* <b>`b`</b>: The second structure to compare.
* <b>`aname`</b>: Variable name to use for the first structure in assertion messages.
* <b>`bname`</b>: Variable name to use for the second structure.
* <b>`msg`</b>: Additional text to include in the failure message.
- - -
#### `tf.test.TestCase.assertSequenceAlmostEqual(expected_seq, actual_seq, places=None, msg=None, delta=None)` {#TestCase.assertSequenceAlmostEqual}
An approximate equality assertion for ordered sequences.
Fail if the two sequences are unequal as determined by their value
differences rounded to the given number of decimal places (default 7) and
comparing to zero, or by comparing that the difference between each value
in the two sequences is more than the given delta.
Note that decimal places (from zero) are usually not the same as significant
digits (measured from the most signficant digit).
If the two sequences compare equal then they will automatically compare
almost equal.
##### Args:
* <b>`expected_seq`</b>: A sequence containing elements we are expecting.
* <b>`actual_seq`</b>: The sequence that we are testing.
* <b>`places`</b>: The number of decimal places to compare.
* <b>`msg`</b>: The message to be printed if the test fails.
* <b>`delta`</b>: The OK difference between compared values.
- - -
#### `tf.test.TestCase.assertSequenceEqual(seq1, seq2, msg=None, seq_type=None)` {#TestCase.assertSequenceEqual}
@ -1000,26 +599,6 @@ which can be indexed, has a length, and has an equality operator.
differences.
- - -
#### `tf.test.TestCase.assertSequenceStartsWith(prefix, whole, msg=None)` {#TestCase.assertSequenceStartsWith}
An equality assertion for the beginning of ordered sequences.
If prefix is an empty sequence, it will raise an error unless whole is also
an empty sequence.
If prefix is not a sequence, it will raise an error if the first element of
whole does not match.
##### Args:
* <b>`prefix`</b>: A sequence expected at the beginning of the whole parameter.
* <b>`whole`</b>: The sequence in which to look for prefix.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertSetEqual(set1, set2, msg=None)` {#TestCase.assertSetEqual}
@ -1071,51 +650,6 @@ Assert that actual.startswith(expected_start) is True.
* <b>`msg`</b>: Optional message to report on failure.
- - -
#### `tf.test.TestCase.assertTotallyOrdered(*groups, **kwargs)` {#TestCase.assertTotallyOrdered}
Asserts that total ordering has been implemented correctly.
For example, say you have a class A that compares only on its attribute x.
Comparators other than __lt__ are omitted for brevity.
class A(object):
def __init__(self, x, y):
self.x = x
self.y = y
def __hash__(self):
return hash(self.x)
def __lt__(self, other):
try:
return self.x < other.x
except AttributeError:
return NotImplemented
assertTotallyOrdered will check that instances can be ordered correctly.
For example,
self.assertTotallyOrdered(
[None], # None should come before everything else.
[1], # Integers sort earlier.
[A(1, 'a')],
[A(2, 'b')], # 2 is after 1.
[A(3, 'c'), A(3, 'd')], # The second argument is irrelevant.
[A(4, 'z')],
['foo']) # Strings sort last.
##### Args:
* <b>`*groups`</b>: A list of groups of elements. Each group of elements is a list
of objects that are equal. The elements in each group must be less than
the elements in the group after it. For example, these groups are
totally ordered: [None], [1], [2, 2], [3].
* <b>`**kwargs`</b>: optional msg keyword argument can be passed.
- - -
#### `tf.test.TestCase.assertTrue(expr, msg=None)` {#TestCase.assertTrue}
@ -1138,13 +672,6 @@ A tuple-specific equality assertion.
differences.
- - -
#### `tf.test.TestCase.assertUrlEqual(a, b, msg=None)` {#TestCase.assertUrlEqual}
Asserts that urls are equal, ignoring ordering of query params.
- - -
#### `tf.test.TestCase.assert_(expr, msg=None)` {#TestCase.assert_}
@ -1206,9 +733,9 @@ tearDown.
- - -
#### `tf.test.TestCase.fail(msg=None, prefix=None)` {#TestCase.fail}
#### `tf.test.TestCase.fail(msg=None)` {#TestCase.fail}
Fail immediately with the given message, optionally prefixed.
Fail immediately, with the given message.
- - -
@ -1260,13 +787,6 @@ Fail immediately with the given message, optionally prefixed.
- - -
#### `tf.test.TestCase.getRecordedProperties()` {#TestCase.getRecordedProperties}
Return any properties that the user has recorded.
- - -
#### `tf.test.TestCase.get_temp_dir()` {#TestCase.get_temp_dir}
@ -1281,20 +801,6 @@ Return any properties that the user has recorded.
- - -
#### `tf.test.TestCase.recordProperty(property_name, property_value)` {#TestCase.recordProperty}
Record an arbitrary property for later use.
##### Args:
* <b>`property_name`</b>: str, name of property to record; must be a valid XML
attribute name
* <b>`property_value`</b>: value of property; must be valid XML attribute value
- - -
#### `tf.test.TestCase.run(result=None)` {#TestCase.run}
@ -1320,18 +826,11 @@ Hook method for setting up class fixture before running tests in the class.
#### `tf.test.TestCase.shortDescription()` {#TestCase.shortDescription}
Format both the test method name and the first line of its docstring.
Returns a one-line description of the test, or None if no
description has been provided.
If no docstring is given, only returns the method name.
This method overrides unittest.TestCase.shortDescription(), which
only returns the first line of the docstring, obscuring the name
of the test upon failure.
##### Returns:
* <b>`desc`</b>: A short description of a test method.
The default implementation of this method returns the first line of
the specified test method's docstring.
- - -

99
tensorflow/g3doc/get_started/os_setup.md
View File

@ -61,58 +61,54 @@ $ sudo easy_install pip
$ sudo easy_install --upgrade six
```
We have also uploaded the CPU version of the binaries to Pypi, so you can
simply install on Linux, Mac or Windows with:
We have also uploaded the binaries to Pypi, so you can
simply install tensorflow on Linux, Mac or Windows with pip install. Note you will need pip version 8.1 or later for the following commands to work on Linux :
```bash
$ pip install tensorflow
```
Note that you will need pip version 8.1 or later for the above command to work on Linux.
For installing the version with GPU support, please use:
For Windows users, you can also install the GPU version of the binary with:
```bash
$ pip install tensorflow-gpu
```
Unfortunately, this command is not yet available for Linux or Mac GPU binaries
due to their sizes exceeding the Pypi limit.
If the above commands do not work on your system or you want to install the GPU version
of the binary on Linux or Mac, you can follow these instructions:
If the above commands do not work on your system, you can follow these instructions:
```bash
# Ubuntu/Linux 64-bit, CPU only, Python 2.7
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp27-none-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp27-none-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 2.7
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp27-none-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp27-none-linux_x86_64.whl
# Mac OS X, CPU only, Python 2.7:
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py2-none-any.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py2-none-any.whl
# Mac OS X, GPU enabled, Python 2.7:
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py2-none-any.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py2-none-any.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.4
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.4
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.5
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp35-cp35m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.5
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp35-cp35m-linux_x86_64.whl
# Mac OS X, CPU only, Python 3.4 or 3.5:
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py3-none-any.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py3-none-any.whl
# Mac OS X, GPU enabled, Python 3.4 or 3.5:
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py3-none-any.whl
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py3-none-any.whl
```
Install TensorFlow:
@ -154,14 +150,14 @@ Both distributions include pip. To install the CPU-only version of
TensorFlow, enter the following command at a command prompt:
```bat
C:\> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-0.12.0rc1-cp35-cp35m-win_amd64.whl
C:\> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-0.12.0-cp35-cp35m-win_amd64.whl
```
To install the GPU version of TensorFlow, enter the following command
at a command prompt:
```bat
C:\> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/gpu/tensorflow_gpu-0.12.0rc1-cp35-cp35m-win_amd64.whl
C:\> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/gpu/tensorflow_gpu-0.12.0-cp35-cp35m-win_amd64.whl
```
You can now [test your installation](#test-the-tensorflow-installation).
@ -216,37 +212,37 @@ Now, install TensorFlow just as you would for a regular Pip installation. First
```bash
# Ubuntu/Linux 64-bit, CPU only, Python 2.7
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp27-none-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp27-none-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 2.7
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp27-none-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp27-none-linux_x86_64.whl
# Mac OS X, CPU only, Python 2.7:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py2-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py2-none-any.whl
# Mac OS X, GPU enabled, Python 2.7:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py2-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py2-none-any.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.4
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.4
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.5
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp35-cp35m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.5
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp35-cp35m-linux_x86_64.whl
# Mac OS X, CPU only, Python 3.4 or 3.5:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py3-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py3-none-any.whl
# Mac OS X, GPU enabled, Python 3.4 or 3.5:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py3-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py3-none-any.whl
```
Finally install TensorFlow:
@ -368,37 +364,37 @@ select the correct binary to install:
```bash
# Ubuntu/Linux 64-bit, CPU only, Python 2.7
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp27-none-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp27-none-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 2.7
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp27-none-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp27-none-linux_x86_64.whl
# Mac OS X, CPU only, Python 2.7:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py2-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py2-none-any.whl
# Mac OS X, GPU enabled, Python 2.7:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py2-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py2-none-any.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.4
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.4
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp34-cp34m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.5
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.0-cp35-cp35m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, GPU enabled, Python 3.5
# Requires CUDA toolkit 8.0 and CuDNN v5. For other versions, see "Installing from sources" below.
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0rc1-cp35-cp35m-linux_x86_64.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-0.12.0-cp35-cp35m-linux_x86_64.whl
# Mac OS X, CPU only, Python 3.4 or 3.5:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0rc1-py3-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.0-py3-none-any.whl
# Mac OS X, GPU enabled, Python 3.4 or 3.5:
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0rc1-py3-none-any.whl
(tensorflow)$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/gpu/tensorflow_gpu-0.12.0-py3-none-any.whl
```
Finally install TensorFlow:
@ -466,7 +462,7 @@ code.
code.
We also have tags with `latest` replaced by a released version (e.g.,
`0.12.0-rc1-gpu`).
`0.12.0-gpu`).
With Docker the installation is as follows:
@ -651,6 +647,12 @@ sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64/
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
```
##### Install other dependencies
```bash
$ sudo apt-get install libcupti-dev
```
#### Optional: Install OpenCL (Experimental, Linux only)
In order to build or run TensorFlow with OpenCL support, both OpenCL (>= 1.2)
@ -755,7 +757,7 @@ export PATH="$CUDA_HOME/bin:$PATH"
```
Finally, you will also want to install the [CUDA Deep Neural
Network](https://developer.nvidia.com/cudnn) (cuDNN v5) library which currently
Network](https://developer.nvidia.com/cudnn) (cuDNN v5.1) library which currently
requires an [Accelerated Computing Developer
Program](https://developer.nvidia.com/accelerated-computing-developer) account.
Once you have it downloaded locally, you can unzip and move the header and
@ -868,7 +870,7 @@ $ bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_pack
$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
# The name of the .whl file will depend on your platform.
$ sudo pip install /tmp/tensorflow_pkg/tensorflow-0.12.0rc1-py2-none-any.whl
$ sudo pip install /tmp/tensorflow_pkg/tensorflow-0.12.0-py2-none-any.whl
```
## Optimizing CPU performance
@ -1112,6 +1114,17 @@ Step #1 might already solve the problem, however if it still persists, execute s
This issue occurs with new Anaconda installations when `pip` tries to remove `easy-install.pth`.
This file is not included in Anaconda packages, which causes the `pip` installation to fail.
#### Cupti_wrapper.cc: Could not find cuptiActivityRegisterCallbacksin libcupti DSO
If, when running a TensorFlow Python script, you encounter the following error:
```
c:\tf_jenkins\home\workspace\nightly-win\device\gpu\os\windows\tensorflow\core\platform\default\gpu\cupti_wrapper.cc:59] Check failed: ::tensorflow::Status::OK() == (::tensorflow::Env::Default()->GetSymbolFromLibrary( GetDsoHandle(), kName, &f )) (OK vs. Not found: cuptiActivityRegisterCallbacks not found)could not find cuptiActivityRegisterCallbacksin libcupti DSO
```
Add `<path-to-cuda-folder>\NVIDIA GPU Computing Toolkit\CUDA\v8.0\extras\CUPTI\libx64` to your `PATH`.
This issue occurs because on CUDA 8.0 the location of the file `cupti64_80.dll` is not on `PATH` by default.
### Linux issues

2
tensorflow/g3doc/how_tos/embedding_viz/index.md
View File

@ -166,7 +166,7 @@ to find it:
```python
embedding.sprite.image_path = PATH_TO_SPRITE_IMAGE
# Specify the width and height of a single thumbnail.
embedding.single_image_dim.extend([w, h])
embedding.sprite.single_image_dim.extend([w, h])
```
## Interaction

30
tensorflow/g3doc/how_tos/hadoop/index.md
View File

@ -28,26 +28,38 @@ be set:
* **HADOOP_HDFS_HOME**: The location of your HDFS installation. You can also
set this environment variable by running:
```shell
source ${HADOOP_HOME}/libexec/hadoop-config.sh
```
```shell
source ${HADOOP_HOME}/libexec/hadoop-config.sh
```
* **LD_LIBRARY_PATH**: To include the path to libjvm.so, and optionally the path
to libhdfs.so if your Hadoop distribution does not install libhdfs.so in
`$HADOOP_HDFS_HOME/lib/native`. On Linux:
```shell
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${JAVA_HOME}/jre/lib/amd64/server
```
```shell
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${JAVA_HOME}/jre/lib/amd64/server
```
* **CLASSPATH**: The Hadoop jars must be added prior to running your
TensorFlow program. The CLASSPATH set by
`${HADOOP_HOME}/libexec/hadoop-config.sh` is insufficient. Globs must be
expanded as described in the libhdfs documentation:
```shell
CLASSPATH=$($HADOOP_HDFS_HOME}/bin/hadoop classpath --glob) python your_script.py
```
```shell
CLASSPATH=$($HADOOP_HDFS_HOME}/bin/hadoop classpath --glob) python your_script.py
```
For older version of Hadoop/libhdfs (older than 2.6.0), you have to expand the
classpath wildcard manually. For more details, see
[HADOOP-10903](https://issues.apache.org/jira/browse/HADOOP-10903).
If the Hadoop cluster is in secure mode, the following environment variable must
be set:
* **KERB_TICKET_CACHE_PATH**: The path of Kerberos ticket cache file. For example:
```shell
export KERB_TICKET_CACHE_PATH=/tmp/krb5cc_10002
```
If you are running [Distributed TensorFlow](../distributed/index.md), then all
workers must have the environment variables set and Hadoop installed.

1
tensorflow/g3doc/resources/index.md
View File

@ -43,6 +43,7 @@ something amazing with TensorFlow, we'd like to hear about it!
The TensorFlow community has created many great projects around TensorFlow, including:
* [Machine Learning with TensorFlow (Book & Code)](http://tensorflowbook.com)
* [@jtoy's awesome "Awesome TensorFlow" list of awesome things](https://github.com/jtoy/awesome-tensorflow)
* [TensorFlow tutorials](https://github.com/pkmital/tensorflow_tutorials)
* [Scikit Flow - Simplified Interface for TensorFlow](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/learn/python/learn)

4
tensorflow/g3doc/tutorials/image_recognition/index.md
View File

@ -42,8 +42,8 @@ For example, here are the results from [AlexNet] classifying some images:
To compare models, we examine how often the model fails to predict the
correct answer as one of their top 5 guesses -- termed "top-5 error rate".
[AlexNet] achieved by setting a top-5 error rate of 15.3% on the 2012
validation data set; [BN-Inception-v2] achieved 6.66%;
[Inception-v3] reaches 3.46%.
validation data set; [Inception (GoogLeNet)] achieved 6.67%;
[BN-Inception-v2] achieved 4.9%; [Inception-v3] reaches 3.46%.
> How well do humans do on ImageNet Challenge? There's a [blog post] by
Andrej Karpathy who attempted to measure his own performance. He reached

2
tensorflow/g3doc/tutorials/recurrent/index.md
View File

@ -11,7 +11,7 @@ In this tutorial we will show how to train a recurrent neural network on
a challenging task of language modeling. The goal of the problem is to fit a
probabilistic model which assigns probabilities to sentences. It does so by
predicting next words in a text given a history of previous words. For this
purpose we will use the [Penn Tree Bank](http://www.cis.upenn.edu/~treebank/)
purpose we will use the [Penn Tree Bank](https://catalog.ldc.upenn.edu/ldc99t42)
(PTB) dataset, which is a popular benchmark for measuring quality of these
models, whilst being small and relatively fast to train.

2
tensorflow/java/src/main/java/org/tensorflow/Session.java
View File

@ -225,7 +225,7 @@ public final class Session implements AutoCloseable {
public Reference() {
synchronized (nativeHandleLock) {
if (nativeHandle == 0) {
throw new IllegalStateException("run() called after the Session was close()d");
throw new IllegalStateException("run() cannot be called on the Session after close()");
}
++numActiveRuns;
}

19
tensorflow/python/kernel_tests/control_flow_ops_py_test.py
View File

@ -255,7 +255,7 @@ class ControlFlowTest(test.TestCase):
enter_one = control_flow_ops.enter(one, "foo", True)
enter_n = control_flow_ops.enter(n, "foo", True)
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
merge_i = control_flow_ops.merge([enter_i, enter_i])[0]
less_op = math_ops.less(merge_i, enter_n)
@ -289,7 +289,7 @@ class ControlFlowTest(test.TestCase):
add_i = math_ops.add(switch_i[1], enter_one)
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
next_i = control_flow_ops.next_iteration(add_i)
merge_i.op._update_input(1, next_i)
@ -567,7 +567,7 @@ class ControlFlowTest(test.TestCase):
def testCondRecvIdentity(self):
# Make sure the switch identity is not removed by optimization.
with session.Session(config=opt_cfg()) as sess:
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
pred = constant_op.constant(True)
def fn1():
@ -1341,12 +1341,15 @@ class ControlFlowTest(test.TestCase):
self.assertEqual(45, rx.eval())
def _testWhileGrad_ColocateGradients(self, colocate):
gpu_dev_name = test.gpu_device_name() if test.is_gpu_available() else "/gpu:0"
gpu_short_name = gpu_dev_name.split('/')[-1]
with self.test_session(graph=ops.Graph()) as sess:
v = constant_op.constant(2.0, name="v")
c = lambda v: math_ops.less(v, 100.0)
def b(x):
with ops.device("/gpu:0"):
with ops.device(gpu_dev_name):
return math_ops.square(x)
loop = control_flow_ops.while_loop(c, b, [v], parallel_iterations=1)
@ -1360,12 +1363,12 @@ class ControlFlowTest(test.TestCase):
for (name, dev) in r_devices:
if not colocate and name.endswith("Square"):
# Only forward graph contain gpu in Square device
self.assertTrue("gpu:0" in dev)
self.assertTrue(gpu_short_name in dev)
elif colocate and "Square" in name:
# Forward and backward graphs contain gpu in Square/Square_grad devices
self.assertTrue("gpu:0" in dev)
self.assertTrue(gpu_short_name in dev)
else:
self.assertFalse("gpu:0" in dev)
self.assertFalse(gpu_short_name in dev)
self.assertAllClose(1024.0, sess.run(r))
def testWhileGrad_ColocateGradients(self):
@ -2566,7 +2569,7 @@ class AssertTest(test.TestCase):
def testGuardedAssertDoesNotCopyWhenTrue(self):
with self.test_session(use_gpu=True) as sess:
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
value = constant_op.constant(1.0)
with ops.device("/cpu:0"):
true = constant_op.constant(True)

4
tensorflow/python/kernel_tests/conv_ops_test.py
View File

@ -181,8 +181,8 @@ def GetTestConfigs():
all the valid test configs as tuples of data_format and use_gpu.
"""
test_configs = [("NHWC", False), ("NHWC", True)]
if test.is_gpu_available():
# "NCHW" format is not currently supported on CPU.
if test.is_gpu_available(cuda_only=True):
# "NCHW" format is only supported on CUDA.
test_configs += [("NCHW", True)]
return test_configs

2
tensorflow/python/kernel_tests/decode_image_op_test.py
View File

@ -28,7 +28,7 @@ from tensorflow.python.ops import io_ops
import tensorflow.python.ops.nn_grad # pylint: disable=unused-import
from tensorflow.python.platform import test
prefix_path = "tensorflow/core/lib"
prefix_path = os.path.join("tensorflow", "core", "lib")
class DecodeImageOpTest(test.TestCase):

16
tensorflow/python/kernel_tests/pooling_ops_test.py
View File

@ -598,8 +598,8 @@ class PoolingTest(test.TestCase):
self.assertAllClose(cpu_val, gpu_val)
def testMaxPoolingWithArgmax(self):
# MaxPoolWithArgMax is implemented only on GPU.
if not test.is_gpu_available():
# MaxPoolWithArgMax is implemented only on CUDA.
if not test.is_gpu_available(cuda_only=True):
return
tensor_input = [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0]
with self.test_session(use_gpu=True) as sess:
@ -617,8 +617,8 @@ class PoolingTest(test.TestCase):
self.assertAllEqual(argmax.ravel(), [0, 1, 3, 5])
def testMaxPoolingGradWithArgmax(self):
# MaxPoolWithArgMax is implemented only on GPU.
if not test.is_gpu_available():
# MaxPoolWithArgMax is implemented only on CUDA.
if not test.is_gpu_available(cuda_only=True):
return
orig_input = [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0]
tensor_input = [11.0, 12.0, 13.0, 14.0]
@ -1250,8 +1250,8 @@ class PoolingTest(test.TestCase):
def GetMaxPoolFwdTest(input_size, filter_size, strides, padding):
def Test(self):
# MaxPoolWithArgMax is implemented only on GPU.
if not test.is_gpu_available():
# MaxPoolWithArgMax is implemented only on CUDA.
if not test.is_gpu_available(cuda_only=True):
return
self._CompareMaxPoolingFwd(input_size, filter_size, strides, padding)
@ -1261,8 +1261,8 @@ def GetMaxPoolFwdTest(input_size, filter_size, strides, padding):
def GetMaxPoolGradTest(input_size, filter_size, output_size, strides, padding):
def Test(self):
# MaxPoolWithArgMax is implemented only on GPU.
if not test.is_gpu_available():
# MaxPoolWithArgMax is implemented only on CUDA.
if not test.is_gpu_available(cuda_only=True):
return
self._CompareMaxPoolingBk(input_size, output_size, filter_size, strides,
padding)

6
tensorflow/python/kernel_tests/session_ops_test.py
View File

@ -138,7 +138,7 @@ class SessionOpsTest(test.TestCase):
self.assertEqual(500, sess.run(y, feed_dict={f: h.handle}))
# Feed another tensor handle.
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
a = constant_op.constant(10)
h = session_ops.get_session_handle(a)
h = sess.run(h)
@ -169,7 +169,7 @@ class SessionOpsTest(test.TestCase):
def testMultiDevices(self):
with self.test_session() as sess:
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
a = constant_op.constant(1.0)
a_handle = sess.run(session_ops.get_session_handle(a))
with ops.device("/cpu:0"):
@ -194,7 +194,7 @@ class SessionOpsTest(test.TestCase):
x_handle = sess.run(session_ops.get_session_handle(one))
# addition lives on GPU
with ops.device("/gpu:0"):
with ops.device(test.gpu_device_name()):
add_h1, add_t1 = session_ops.get_session_tensor(one_handle.handle,
dtypes.float32)
add_h2, add_t2 = session_ops.get_session_tensor(x_handle.handle,

8
tensorflow/python/kernel_tests/split_op_test.py
View File

@ -128,6 +128,14 @@ class SplitOpTest(test.TestCase):
self.assertAllEqual(result[:, 0:1], inp_grads[0])
self.assertAllEqual(result[:, 1:4], inp_grads[1])
def testOutputShape(self):
with self.test_session(use_gpu=False):
tensor = array_ops.placeholder(dtypes.float32, shape=[None, 12])
size_splits = [3, 7, 2]
outputs = array_ops.split(tensor, size_splits, 1)
for i, output in enumerate(outputs):
self.assertEqual(output.get_shape().as_list(), [None, size_splits[i]])
def _compare(self, x, dim, num, use_gpu):
np_ans = np.split(x, num, dim)
with self.test_session(use_gpu=use_gpu) as sess:

4
tensorflow/python/ops/array_ops.py
View File

@ -1176,8 +1176,8 @@ def boolean_mask(tensor, mask, name="boolean_mask"):
name: A name for this operation (optional).
Returns:
Tensor populated by entries in `tensor` corresponding to `True` values in
`mask`.
(N-K+1)-dimensional tensor populated by entries in `tensor` corresponding
to `True` values in `mask`.
Raises:
ValueError: If shapes do not conform.

4
tensorflow/python/ops/losses/losses.py
View File

@ -603,7 +603,7 @@ def softmax_cross_entropy(
loss_collection: collection to which the loss will be added.
Returns:
A scalar `Tensor` representing the loss value.
A scalar `Tensor` representing the mean loss value.
Raises:
ValueError: If the shape of `logits` doesn't match that of `onehot_labels`
@ -652,7 +652,7 @@ def sparse_softmax_cross_entropy(labels, logits, weights=1.0, scope=None,
loss_collection: collection to which the loss will be added.
Returns:
A scalar `Tensor` representing the loss value.
A scalar `Tensor` representing the mean loss value.
Raises:
ValueError: If the shapes of logits, labels, and weight are incompatible, or

6
tensorflow/python/ops/math_ops.py
View File

@ -1673,13 +1673,15 @@ def matmul(a,
# 3-D tensor `a`
a = tf.constant(np.arange(1,13), shape=[2, 2, 3]) => [[[ 1. 2. 3.]
a = tf.constant(np.arange(1, 13, dtype=np.int32),
shape=[2, 2, 3]) => [[[ 1. 2. 3.]
[ 4. 5. 6.]],
[[ 7. 8. 9.]
[10. 11. 12.]]]
# 3-D tensor `b`
b = tf.constant(np.arange(13,25), shape=[2, 3, 2]) => [[[13. 14.]
b = tf.constant(np.arange(13, 25, dtype=np.int32),
shape=[2, 3, 2]) => [[[13. 14.]
[15. 16.]
[17. 18.]],
[[19. 20.]

15
tensorflow/python/ops/nn.py
View File

@ -68,12 +68,15 @@ For the `'SAME'` padding, the output height and width are computed as:
and the padding on the top and left are computed as:
pad_along_height = ((out_height - 1) * strides[1] +
filter_height - in_height)
pad_along_width = ((out_width - 1) * strides[2] +
filter_width - in_width)
pad_top = pad_along_height / 2
pad_left = pad_along_width / 2
pad_along_height = max((out_height - 1) * strides[1] +
filter_height - in_height, 0)
pad_along_width = max((out_width - 1) * strides[2] +
filter_width - in_width, 0)
pad_top = pad_along_height // 2
pad_bottom = pad_along_height - pad_top
pad_left = pad_along_width // 2
pad_right = pad_along_width - pad_left
Note that the division by 2 means that there might be cases when the padding on
both sides (top vs bottom, right vs left) are off by one. In this case, the

1
tensorflow/python/ops/rnn_cell_impl.py
View File

@ -138,4 +138,3 @@ class _RNNCell(object):
zeros.set_shape(_state_size_with_prefix(state_size, prefix=[None]))
return zeros

3
tensorflow/python/platform/test.py
View File

@ -48,6 +48,7 @@ methods. We will document these methods soon.
@@get_temp_dir
@@is_built_with_cuda
@@is_gpu_available
@@gpu_device_name
## Gradient checking
@ -141,7 +142,7 @@ def gpu_device_name():
"""Returns the name of a GPU device if available or the empty string."""
for x in _device_lib.list_local_devices():
if x.device_type == 'GPU' or x.device_type == 'SYCL':
return x.name()
return x.name
return ''

7
tensorflow/python/training/saver_test.py
View File

@ -413,20 +413,17 @@ class SaverTest(test.TestCase):
return
save_path = os.path.join(self.get_temp_dir(), "gpu")
with session.Session("", graph=ops_lib.Graph()) as sess:
with sess.graph.device("/gpu:0"):
with sess.graph.device(test.gpu_device_name()):
v0_1 = variables.Variable(123.45)
save = saver_module.Saver({"v0": v0_1})
variables.global_variables_initializer().run()
save.save(sess, save_path)
with session.Session("", graph=ops_lib.Graph()) as sess:
with sess.graph.device("/gpu:0"):
with sess.graph.device(test.gpu_device_name()):
v0_2 = variables.Variable(543.21)
save = saver_module.Saver({"v0": v0_2})
variables.global_variables_initializer().run()
self.assertAllClose(543.21, v0_2.eval())
save.restore(sess, save_path)
self.assertAllClose(123.45, v0_2.eval())
def testVariables(self):
save_path = os.path.join(self.get_temp_dir(), "variables")

4
tensorflow/tools/ci_build/builds/test_tutorials.sh
View File

@ -230,7 +230,9 @@ test_ptb_word_lm() {
PTB_DATA_URL="http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz"
DATA_DIR="${TUT_TEST_DATA_DIR}/ptb"
if [[ ! -d "${DATA_DIR}/simple-examples/data" ]]; then
if [[ ! -f "${DATA_DIR}/simple-examples/data/ptb.train.txt" ]] || \
[[ ! -f "${DATA_DIR}/simple-examples/data/ptb.valid.txt" ]] || \
[[ ! -f "${DATA_DIR}/simple-examples/data/ptb.test.txt" ]]; then
# Download and extract data
echo "Downloading and extracting PTB data from \"${PTB_DATA_URL}\" to "\
"${DATA_DIR}"

3
tensorflow/tools/ci_build/protobuf/protobuf_optimized_pip.sh
View File

@ -15,6 +15,7 @@
# ==============================================================================
PROTOBUF_VERSION="3.1.0"
PYTHON_BIN=${PYTHON_BIN:-python}
DIR=${PWD}/protobuf
set -ex
@ -29,4 +30,4 @@ CXXFLAGS="-fPIC -g -O2" ./configure
make -j8
export PROTOC=$DIR/src/protoc
cd python
python setup.py bdist_wheel --cpp_implementation --compile_static_extension
$PYTHON_BIN setup.py bdist_wheel --cpp_implementation --compile_static_extension

2
tensorflow/tools/pip_package/setup.py
View File

@ -29,7 +29,7 @@ from setuptools.dist import Distribution
# This version string is semver compatible, but incompatible with pip.
# For pip, we will remove all '-' characters from this string, and use the
# result for pip.
_VERSION = '0.12.0-rc1'
_VERSION = '0.12.0'
REQUIRED_PACKAGES = [
'numpy >= 1.11.0',

8
tensorflow/workspace.bzl
View File

@ -17,11 +17,11 @@ def tf_workspace(path_prefix = "", tf_repo_name = ""):
native.new_http_archive(
name = "eigen_archive",
urls = [
"https://bitbucket.org/eigen/eigen/get/c362228c7087.tar.gz",
"http://bazel-mirror.storage.googleapis.com/bitbucket.org/eigen/eigen/get/c362228c7087.tar.gz",
"http://bazel-mirror.storage.googleapis.com/bitbucket.org/eigen/eigen/get/60578b474802.tar.gz",
"https://bitbucket.org/eigen/eigen/get/60578b474802.tar.gz",
],
sha256 = "ea091e0a1d78b0c6714f7109a6a80af30c4745f3ff8c398675cf42271acc0c77",
strip_prefix = "eigen-eigen-c362228c7087",
sha256 = "7527cda827aff351981ebd910012e16be4d899c28a9ae7f143ae60e7f3f7b83d",
strip_prefix = "eigen-eigen-60578b474802",
build_file = str(Label("//:eigen.BUILD")),
)

9
third_party/eigen3/unsupported/Eigen/CXX11/FixedPoint vendored
View File

@ -31,8 +31,15 @@
#include "src/FixedPoint/FixedPointTypes.h"
// Use optimized implementations whenever available
#ifdef EIGEN_VECTORIZE_AVX2
#ifdef EIGEN_VECTORIZE_AVX512
#include "src/Tensor/TensorContractionThreadPool.h"
#include "src/FixedPoint/PacketMathAVX512.h"
#include "src/FixedPoint/TypeCastingAVX512.h"
#elif defined EIGEN_VECTORIZE_AVX2
#define EIGEN_USE_OPTIMIZED_INT8_UINT8_MAT_MAT_PRODUCT
#define EIGEN_USE_OPTIMIZED_INT16_INT16_MAT_MAT_PRODUCT
#include "src/Tensor/TensorContractionThreadPool.h"
#include "src/FixedPoint/PacketMathAVX2.h"
#include "src/FixedPoint/MatMatProductAVX2.h"
#include "src/FixedPoint/TypeCastingAVX2.h"

2
third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/PacketMathAVX2.h vendored
View File

@ -46,6 +46,7 @@ typedef struct Packet4q32i {
Packet4q32i(__m128i val) : val(val) {}
} Packet4q32i;
#ifndef EIGEN_VECTORIZE_AVX512
template <>
struct packet_traits<QInt8> : default_packet_traits {
typedef Packet32q8i type;
@ -112,6 +113,7 @@ struct packet_traits<QInt32> : default_packet_traits {
HasSetLinear = 0
};
};
#endif
template <>
struct unpacket_traits<Packet32q8i> {

Some files were not shown because too many files have changed in this diff Show More