Commit message (Collapse) | Author | Age | |
---|---|---|---|
* | Update links to tf lite site. | Billy Lamberta | 2018-09-20 |
| | | | | PiperOrigin-RevId: 213917881 | ||
* | Remove all_opensource_files. It's not needed any more. | Martin Wicke | 2018-03-28 |
| | | | | PiperOrigin-RevId: 190878279 | ||
* | Add documentation to contrib/quantization to reduce confusion with ↵ | Suharsh Sivakumar | 2018-02-20 |
| | | | | | | contrib/quantize. PiperOrigin-RevId: 186358131 | ||
* | Selected BUILD cleanup in tensorflow/contrib/... | A. Unique TensorFlower | 2017-06-18 |
| | | | | PiperOrigin-RevId: 159373397 | ||
* | Remove hourglass imports from kernel_tests | Justine Tunney | 2016-12-14 |
| | | | | Change: 142080137 | ||
* | Merge changes from github. | Benoit Steiner | 2016-11-09 |
| | | | | Change: 138675832 | ||
* | Move contrib/quantization ops to tensorflow/core | Andrew Harp | 2016-10-17 |
| | | | | Change: 136410307 | ||
* | Rename gemm_wrapper.h to soc_interface.h | A. Unique TensorFlower | 2016-10-12 |
| | | | | Change: 135940182 | ||
* | Add a placeholder of utility to transfer a graph to SoC | A. Unique TensorFlower | 2016-10-11 |
| | | | | Change: 135834593 | ||
* | Replace references to deprecated eight_bit_int_gemm with gemmlowp | Andrew Harp | 2016-10-10 |
| | | | | Change: 135719801 | ||
* | Merge changes from github. | A. Unique TensorFlower | 2016-10-10 |
| | | | | Change: 135698415 | ||
* | In quantize_graph.py, reset output_nodes after set_input_graph is called, in | A. Unique TensorFlower | 2016-10-07 |
| | | | | | case an output node was rewritten because its input was removed. Change: 135479283 | ||
* | Automated rollback of change 134501895 | A. Unique TensorFlower | 2016-09-28 |
| | | | | Change: 134506649 | ||
* | Move contrib/quantization ops to tensorflow/core | Andrew Harp | 2016-09-28 |
| | | | | Change: 134501895 | ||
* | Fix cpu utils for android armv7a | A. Unique TensorFlower | 2016-09-22 |
| | | | | Change: 133983864 | ||
* | Merge changes from github. | Martin Wicke | 2016-09-21 |
| | | | | Change: 133874452 | ||
* | Implement a function to get frequency for android armv7a | A. Unique TensorFlower | 2016-09-16 |
| | | | | Change: 133434099 | ||
* | Add an adapter class of CpuUtilsHelper for Android armv7a | A. Unique TensorFlower | 2016-09-12 |
| | | | | Change: 132911631 | ||
* | Merge changes from github. | Dan Mané | 2016-09-09 |
| | | | | Change: 132733397 | ||
* | To prepare for removal of cuda_cc_test(s) build rule, make all build rules | Gunhan Gulsoy | 2016-09-09 |
| | | | | | consistent by making all rules take srcs arg. Change: 132710890 | ||
* | Switch ops in functional_ops, math_ops, state_ops, string_ops, | A. Unique TensorFlower | 2016-09-09 |
| | | | | | | | | | | | | | | | | | contrib/quantization, and contrib/layers to use C++ shape inference functions. Implement Betainc C++ shape inference function; it's a little different from the python one, mainly because propagating an unknown input (a_shape in this case) when all are unknown is not correct for C++ (in case we later backwards bind the value), since it could end up being a broadcasted scalar. Change ReduceJoin's C++ shape inference function to match python. Fix SymbolicGradient's shape function in C++ - there are actually more inputs than outputs. Changes QuantizedBiasAdd to be more precise (the C++ shape fn invokes the common bias_add to compute the shape of the first output). Change: 132688147 | ||
* | Convert ops in contrib, parsing_ops, and random_ops to use C++ shape functions. | A. Unique TensorFlower | 2016-09-08 |
| | | | | Change: 132565142 | ||
* | Move CpuUtils to more appropriate place, core/platform/profile_utils | A. Unique TensorFlower | 2016-09-02 |
| | | | | Change: 132095035 | ||
* | Add a test to verify overhead of hexagon rpc calls | A. Unique TensorFlower | 2016-09-01 |
| | | | | Change: 131987004 | ||
* | Add a way to link and call hexagon shared library APIs | A. Unique TensorFlower | 2016-08-26 |
| | | | | Change: 131413727 | ||
* | Add shape_inference::ShapeHandle and shape_inference::DimensionHandle to | A. Unique TensorFlower | 2016-08-23 |
| | | | | | | | | replace uses of const Shape* and const Dimension*. This change only adds a typedef and updates references. A later change will make DimensionHandle and ShapeHandle real types instead of typedefs (to further hide the pointer access). Change: 131118981 | ||
* | Add matmul test palceholder for hexagon | A. Unique TensorFlower | 2016-08-19 |
| | | | | Change: 130781400 | ||
* | Automated rollback of change 130687505 | A. Unique TensorFlower | 2016-08-18 |
| | | | | Change: 130694711 | ||
* | Add matmul test palceholder for hexagon | A. Unique TensorFlower | 2016-08-18 |
| | | | | Change: 130687505 | ||
* | Add a makefile for tests in contrib/quantization | A. Unique TensorFlower | 2016-08-18 |
| | | | | Change: 130665817 | ||
* | Merge changes from github. | Benoit Steiner | 2016-08-12 |
| | | | | Change: 130150683 | ||
* | Add shape function for QuantizedConcat, moving ConcatShape into | Vijay Vasudevan | 2016-08-10 |
| | | | | | a common shape fn. Change: 129928393 | ||
* | Fix race in tensorflow's use of gemmlowp contexts. | A. Unique TensorFlower | 2016-08-09 |
| | | | | Change: 129782045 | ||
* | Optimizing graphs for inference. | Pete Warden | 2016-08-07 |
| | | | | Change: 129581148 | ||
* | Remove "external/gemmlowp" from gemmlowp #includes since it's not needed and ↵ | Kiril Gorovoy | 2016-08-04 |
| | | | | | | fails to build when compiling TensorFlow as a submodule. Change: 129359615 | ||
* | Ensure custom GPU kernels are properly registered. | Eugene Brevdo | 2016-08-03 |
| | | | | | This change ensures that GOOGLE_CUDA=1 when compiling .cc files with --config=cuda enabled (also includes other important copts like -fno-exceptions.) Change: 129242223 | ||
* | TensorFlow: Finish off quantization shape functions in contrib | Vijay Vasudevan | 2016-08-02 |
| | | | | Change: 129143268 | ||
* | TensorFlow: implement MaxPoolShape common function + tests, add to | Vijay Vasudevan | 2016-07-29 |
| | | | | | | | | ops that need it. Moved one of the implementation-specific checks in the shape function to the kernel instead, where it belongs. Change: 128834923 | ||
* | Add C++ shape inference for quantizev2 and dequantize. | Suharsh Sivakumar | 2016-07-28 |
| | | | | Change: 128768449 | ||
* | Add shape functions for some of the quantization ops. | Vijay Vasudevan | 2016-07-26 |
| | | | | Change: 128520578 | ||
* | Add existing common shape function uses to MatMuls, Conv2Ds, AvgPool. | Vijay Vasudevan | 2016-07-25 |
| | | | | | | | | Since common shape functions are already tested, and the additions here are pretty straight-forward extensions that use the common shapes, I've elided tests for them, but could add them if we thought it was useful. Change: 128418673 | ||
* | Merge changes from github. | Martin Wicke | 2016-07-25 |
| | | | | Change: 128401884 | ||
* | In contrib/quantization, use eigen threadpool when calling gemmlowp to | A. Unique TensorFlower | 2016-07-16 |
| | | | | | avoid creating a new one each time. Change: 127624630 | ||
* | Fix divide-by-zero in specialized version of RequantizeMany...UsingEigen. | A. Unique TensorFlower | 2016-07-13 |
| | | | | Change: 127374168 | ||
* | Move computation in quantized_bias_add_op.cc to quantization_utils.h. | A. Unique TensorFlower | 2016-07-13 |
| | | | | Change: 127370125 | ||
* | Fixes a bug in TensorFlow quantization, where a negative constant node | A. Unique TensorFlower | 2016-07-13 |
| | | | | | | had its min and max values erroneously calculated, causing an exception to be raised. Change: 127362133 | ||
* | Avoid overflow in float-int32 cast when making test case for quantization. | A. Unique TensorFlower | 2016-07-13 |
| | | | | | Also fix bug in creation of test for 1-byte values. Change: 127344565 | ||
* | Change QuantizedBiasAdd op to use eigen floating point computation | A. Unique TensorFlower | 2016-07-13 |
| | | | | | instead of reference computation. Change: 127319811 | ||
* | Enable optimized requantization function. | A. Unique TensorFlower | 2016-07-12 |
| | | | | | | | | Fix rezeroing of input range in optimized requantize function to account for signed input. Change some computations to be in scaled-up fixed point instead of after scaling back down. Make rounding delta always be positive. Change: 127272485 | ||
* | Change DequantizeOp to use eigen for MIN_FIRST case. | A. Unique TensorFlower | 2016-07-11 |
| | | | | Change: 127155464 |