| Commit message (Collapse) | Author | Age |
... | |
| |
| |
| |
| |
| |
| | |
Minor cleanup by moving the helper function ShapesEqual to GraphProperties and adding unit tests for it.
PiperOrigin-RevId: 213876779
|
| |
| |
| |
| |
| |
| | |
optimization
PiperOrigin-RevId: 213840320
|
|/ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
large number of debugging outputs in the INFO log that look like:
I0917 16:20:11.073992 9191 meta_optimizer.cc:334] Starting optimization for grappler item: tf_graph
I0917 16:20:11.079458 9191 meta_optimizer.cc:334] Starting optimization for grappler item: tf_graph
I0917 16:20:11.084827 12447 meta_optimizer.cc:334] Starting optimization for grappler item: tf_graph
I0917 16:20:11.089359 12447 meta_optimizer.cc:334] Starting optimization for grappler item: tf_graph
After this change those lines will simply no longer appear.
RELNOTES: n/a
PiperOrigin-RevId: 213690759
|
|
|
|
|
|
| |
vectorize a MapDefun function. Also implements conversion for two ops: Cast and Unpack.
PiperOrigin-RevId: 213686720
|
|\
| |
| |
| |
| |
| | |
ROCmSoftwarePlatform:upstream-staging-gpu-common-runtime-1
PiperOrigin-RevId: 213653830
|
| |
| |
| |
| | |
PiperOrigin-RevId: 213554813
|
| |
| |
| |
| | |
PiperOrigin-RevId: 213520177
|
| |
| |
| |
| |
| |
| |
| |
| | |
Stateless MapDatasets can be paralellized by switching to ParallelMapDataset. We set `num_parallel_calls` to 2 for now, but in the future a special value will be used that result in the optimal value to be selected dynamically at runtime.
This patch also exposed a memory leak which was fixed.
PiperOrigin-RevId: 213015223
|
| |
| |
| |
| |
| |
| | |
string and std::string are not necessarily the same thing in TF, but this code assumed that they are.
PiperOrigin-RevId: 212952877
|
| |
| |
| |
| | |
PiperOrigin-RevId: 212896336
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
better:
Currently, static shape inference propagates shapes of tensors, but in some cases, we do need values; for this, we use input_tensors (from Const input tensor) and input_tensors_as_shapes and output_tensors_as_shapes (these are ShapeHandle format, but has values, currently only for 1D vector).
This CL enhances propagation of input_tensors_as_shapes and output_tensors_as_shapes to improve static shape inference.
(1) forward scalar Const as input_tensors_as_shapes (currently, only 1D vector),
(2) export input_tensors_as_shapes, output const tensor, and output_tensors_as_shapes to the values of inferred input/output TensorProperties (currently, only input const tensors are exported as values),
(3) use input_tensors_as_shapes as Const tensor to function input (currently, only Const tensors),
(4) forward input_tensors_as_shapes to output_tensors_as_shapes for Identity op,
(5) when Pack op concats scalar values to form output_tensors_as_shapes, currently it uses only input_tensors (from Const input tensors), but this CL change Pack to use input_tensors_as_shapes as well.
PiperOrigin-RevId: 212696959
|
| |
| |
| |
| |
| |
| |
| |
| | |
This now matches the definition. I fixed it here rather than in the definition as it seems every call to this function names the variable "num_components".
I also tidied up the comment a little.
PiperOrigin-RevId: 212668416
|
| |
| |
| |
| |
| |
| | |
to see the total cost of running Grappler in logs. Also add a couple of VLOG(1) statements to see breakdown between main graph and function optimization.
PiperOrigin-RevId: 212531430
|
| |
| |
| |
| | |
PiperOrigin-RevId: 212465918
|
| |
| |
| |
| |
| |
| | |
have been explicitly marked to be compiled via xla.compile()
PiperOrigin-RevId: 212407112
|
| |
| |
| |
| |
| |
| | |
time.
PiperOrigin-RevId: 212321238
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211989327
|
| |
| |
| |
| |
| |
| | |
optimizers and setting optimizer names.
PiperOrigin-RevId: 211900252
|
| |
| |
| |
| |
| |
| | |
inference if possible.
PiperOrigin-RevId: 211821596
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211661670
|
| |
| |
| |
| |
| |
| | |
This patch uses take by value and move idiom to optimize copying of constructor arguments.
PiperOrigin-RevId: 211553877
|
| |
| |
| |
| |
| |
| |
| |
| | |
monotonic unary functions.
Add the ability to flip Max <-> Min if the function is non-increasing, e.g. Max(Neg(x)) => Neg(Min(x)).
PiperOrigin-RevId: 211490436
|
| |
| |
| |
| |
| |
| |
| |
| | |
Reinstate the use of integral-exponent power function MathUtil::IPow, but make sure to use a floating point base, so as to compute the result using floating point arithmetic. This behaviour is equivalent to, but faster than, std::pow.
Note that care must be taken to convert the base to double, which we effect by providing an explicit template type argument for MathUtil::IPow.
PiperOrigin-RevId: 211290304
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211204708
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211162510
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211161172
|
| |
| |
| |
| | |
PiperOrigin-RevId: 211137964
|
|/
|
|
|
| |
Rename CUDA GPU ID to platform GPU ID so the notion is applicable on both CUDA
and ROCm platform.
|
|
|
|
|
|
|
|
| |
value of nodes in the preserve set, e.g. fetch nodes.
I simplified the rewiring logic a tad.
PiperOrigin-RevId: 211017989
|
|
|
|
|
|
| |
list.
PiperOrigin-RevId: 210602483
|
|
|
|
| |
PiperOrigin-RevId: 210596417
|
|
|
|
|
|
| |
This will make it easier to replace tensorflow::StringPiece with absl::string_view, as absl::string_view does not contain a ToString method.
PiperOrigin-RevId: 210550029
|
|
|
|
| |
PiperOrigin-RevId: 210439649
|
|
|
|
| |
PiperOrigin-RevId: 210404649
|
|
|
|
| |
PiperOrigin-RevId: 210402159
|
|
|
|
|
|
|
|
|
|
| |
conversions.
That is, instances of sp.ToString() are replaced with string(sp).
This will allow tensorflow::StringPiece::ToString to be removed, which is necessary before it can be replaced with absl::string_view.
PiperOrigin-RevId: 210394878
|
|
|
|
|
|
|
| |
This patch introduces FilterFusion optimization which can fuse
multiple FilterDataset operations.
PiperOrigin-RevId: 210189643
|
|
|
|
| |
PiperOrigin-RevId: 210187033
|
|\
| |
| |
| | |
PiperOrigin-RevId: 210127837
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
tensor.h soon)
We plan to remove the import variant.h from tensor.h; and variant.h brings in a lot
of transitive imports (including protos like tensor.proto.h). To prepare, we're
updating folks who this will break.
PiperOrigin-RevId: 210043667
|
| |
| |
| |
| | |
PiperOrigin-RevId: 210007888
|
| |
| |
| |
| |
| |
| |
| |
| | |
This solves te problem when passing a scalar tensor to function op input, as
Placeholer shape inference outputs unknown shape for scalar if graphdef version
is < 24.
PiperOrigin-RevId: 210007276
|
| |
| |
| |
| |
| |
| | |
Otherwise executing the op behaves differently from constant folding it.
PiperOrigin-RevId: 209949852
|
|/
|
|
| |
per_process_gpu_memory_fraction take effect
|
|
|
|
|
|
|
| |
- Use graph_utils::GetInputNode in optimizers
- Moved python optimization tests into their own files from optimize_dataset_op_test
PiperOrigin-RevId: 209819734
|
|
|
|
|
|
|
|
| |
That is, instances of sp.ToString() are replaced with string(sp).
This will allow tensorflow::StringPiece::ToString to be removed, which is necessary before it can be replaced with absl::string_view.
PiperOrigin-RevId: 209806694
|
|
|
|
|
|
| |
order of Map->Batch dataset transformations to Batch->Map
PiperOrigin-RevId: 209674669
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(1) previously, it set unknown shape flag for scalar input, but now it
returns TensorShapeProto with rank equal to the expected and all dims set to 1,
and unknown shape flag is not set.
(2) Also, fixed a bug; when a rank is known, but dim_size() < rank (note that
dim_size() may be non-zero), we previously called add_dim() with dim 1 rank
times, which then makes dim_size() is incremented by rank, but we expect
dim_size() equal to rank.
(3) Added test for MaybeGetMinimumShape().
PiperOrigin-RevId: 208845501
|
|
|
|
|
|
| |
total, and number of ops predicted with unknown shapes
PiperOrigin-RevId: 208274158
|