Change: 137908539

Change: 137907844

Change: 137905523

Specifically, suppose you have a line chart with 30 runs. If you mouse over, you will have 30 tooltips, which is hard to read. Previously, if you zoom in so that only 1 run is visible on the screen, you would still get 30 tooltips. Now, if you mouse over you will only see tooltips for runs that are actually on the screen.
Change: 137905115

This is now consistent with what other file systems have.
Change: 137892031

Change: 137889948

Change: 137888607

specify additional tf_kernel_library targets that should be grouped into the
omnibus cc_library target that includes all generated rules, but should not be
considered dependencies of each individual generated tf_kernel_library target.
Use the new attribute to split out some existing deps entries and avoid
unnecessary transitive dependencies.

Split out the :l2loss_op library so that it doesn't share the same deps list
with the rest of the "nn" libraries, in particular avoiding a dependency on
:pooling_ops which in turn depends on the expensive :eigen_helpers.
Likewise, split out :lookup_table_{init_}op from the "data_flow" libraries.
Change: 137887998

Change: 137887950

Added a button below each image in the Tensorboard image panel. Clicking this new button shows an actual size version of the image.
Change: 137886562

Change: 137885496

in TensorBoard.

- Fixes GCS so that it still includes prefixes even if items is empty.
- Change list_directory to use GetChildren() instead of GetMatchingFiles.
  GetMatchingFiles in GCS has a quirk where, unlike other filesystems, it
  doesn't include directories. There's no obvious fix given the GCS API.
Change: 137885363

Change: 137883577

Change: 137881303

In GCS, a folder 'gs://bucket/path' can contain a zero-length object 'gs://bucket/path/', which primarily exists to make IsDirectory('gs://bucket/path') return 'true'.

Naturally GetChildren returned the directory marker for the specified directory whenever it existed, represented as an empty string, which violates assumptions that people normally have from this functions.
Change: 137878839

It's a weak test + causes complications if we consider both V1 & V2.
Change: 137878590

Change: 137877854

nodes and use the information provided by them to set the min/max values
on quantization-related nodes.

In eightbit mode, also changed how constant weights are quantized - instead of
doing it as a step after the main recursion, do it during the main recursion.
This allows the float inputs to FakeQuantWithMinMaxVars to be excluded from
quantization.

In eightbit mode, maintain more state in the stack during recursion.

Also change quantize reshape registration to register always and not use
TF_CALL_xyz; this matches other quantized ops.
Change: 137877226

Change: 137866950

Change: 137866409

Change: 137863303

Change: 137863207

Change: 137861834

Change: 137855838

Change: 137854718

For all but the edge tiles, we can unroll the loops that read and write
from our shared memory tile.
Change: 137853968

SparseTensor.from_value.
Change: 137850511

Change: 137850119

Change: 137849675

Change: 137849082

Change: 137848991

Change: 137847545

Change: 137845970

Change: 137843740

inference. Delegate to C++ shape function for Reshape.

Fix reshape to handle attempting to infer unknown dim when product of known
elements is 0.
Change: 137837591

Change: 137834159

Change: 137833057

depth sort, we can't do 1-bit alpha correctly. Revert to using discard in the
picking shader until we do a global camera-space z sort (if ever).

Aggregate [PointAccessor, PointAccessor, PointAccessor] as a new type,
PointAccessors3D.
Change: 137823236

Some colors are duplicated if there are more than 20.
Change: 137820753

Change: 137788894