Change: 150459431

Improves performance of tf.matmul(a, b, ...) for dense tensors on NVIDIA GPUs in the following cases:

a) If the inner-most dimension of b is 1, i.e. the operation is (possibly a batch of) matrix*vector multiplication(s). This is accomplished by calling Cublas GEMV rather than GEMM. This speeds up large matrix-vector products by about 4x.

b) If one or more dimensions are unknown at graph construction time but the operation is in fact either a single matrix*matrix or matrix*vector multiplication.

The following benchmark numbers illustrating the improvements for matrix * vector products
were measured on a NVIDIA Titan X (Maxwell) card.

Benchmark                                    Base (ns)  New (ns) Improvement
----------------------------------------------------------------------------
BM_Matmul_50_50_1_false_false_DT_FLOAT_gpu       18102     17056     +5.8%
BM_Matmul_50_50_1_true_false_DT_FLOAT_gpu        18108     16374     +9.6%
BM_Matmul_50_50_1_false_true_DT_FLOAT_gpu        18153     17173     +5.4%
BM_Matmul_50_50_1_true_true_DT_FLOAT_gpu         18150     15950    +12.1%
BM_Matmul_500_500_1_false_false_DT_FLOAT_gpu     64605     16874    +73.9%
BM_Matmul_500_500_1_true_false_DT_FLOAT_gpu      62810     17298    +72.5%
BM_Matmul_500_500_1_false_true_DT_FLOAT_gpu      60447     17014    +71.9%
BM_Matmul_500_500_1_true_true_DT_FLOAT_gpu       58443     16934    +71.0%
BM_Matmul_2000_2000_1_false_false_DT_FLOAT_gpu  343298     81898    +76.1%
BM_Matmul_2000_2000_1_true_false_DT_FLOAT_gpu   294738     63723    +78.4%
BM_Matmul_2000_2000_1_false_true_DT_FLOAT_gpu   300671     83650    +72.2%
BM_Matmul_2000_2000_1_true_true_DT_FLOAT_gpu    284540     63742    +77.6%
Change: 150456725

Change: 150452316

folding.
Change: 150450219

Change: 150450082

Change: 150449788

Change: 150447439

Change: 150443246

FileExistsError doesn't exist in Python 2.7.
Change: 150436736

- Removes getArgumentList() in favor of args() etc in XLA (LLVM r298010)
Change: 150427660

The main change is that RNNCells that wrap other RNNCells now override self.zero_state to call the wrapped cell's zero_state and then (maybe) perform some post-processing... instead of relying on the state_size property to provide all information about the state.

Also made zero_state calls create ops inside their own name scope.
Change: 150413265

Change: 150400474

Change: 150397471

Change: 150396376

Change: 150389857

Change: 150388263

Change: 150385615

Change: 150384503

This is a first step towards supporting functions in C++ graph construction, e.g. being able to import GraphDefs with functions.
Change: 150382046

This update contains refinements to the charts in the scalars dashboard.
Change: 150380169

This involves adding a toggle to the health pills info box in the graph visualizer. When that toggle is enabled, Tensorboard makes a request for health pills at step X when the user moves the slider.

This feature can be very slow because it requires reading from disk. Viewing health pills at say step 100,000 could take minutes to an hour. We must design ways to make this faster (for instance, have the debugger write events at a much greater frequency only after it encounters a bad value).
Change: 150379929

 - Feature in 0.4.5 is required for cuda_configure to stay fix
   #7575
Change: 150376995

Change: 150374176

Change: 150373647

Change: 150372607

Change: 150369708

Change: 150364996

Change: 150364914

Change: 150363523

- Add an "ignore y-outliers" option (default true). When true, the domain is calculated based on the middle 8 deciles of the y range, i.e. the lowest 10% of data and highest 10% of data is ignored for domain calculation purposes. Also, this is done with the smoothed data rather than the raw data. This means that brief spikes or an initially high loss value will not distort the chart. This can be disabled to view the full data domain.
- If the y values are all in the range [0, 1], then the domain is automatically set to be approximately [0, 1]. That makes it easy to compare proportional values.
- Added bold gridlines at the origin (0,0) to make the charts easier to understand at a glance.
Change: 150362432

Change: 150361253

Since there are no multi-machine tests, this is likely preferable from a
security standpoint to binding to all interfaces, and sidesteps the question of
cross-platform compatibility (some platforms may have IPV6_V6ONLY enabled by
default).
Change: 150358613

This change allows the importing from tensorflow.python.estimator.* for usage in tests.
Change: 150356830

Change: 150356236

Change: 150351473

Change: 150348366

Change: 150345708

I realized that CL/149810103 forgot to include the tfdbg binary and examples in the pip package.

This CL fixes that and adds pip test for the correct inclusion of these files.
Change: 150345489

Change: 150344647

Change: 150343323