- 06 Sep, 2019 4 commits
-
-
Gauri1 Deshpande authored
-
Diego Caballero authored
Initial support for float16 was done before the ngraph::float16 data type was introduced in nGraph. This PR aligns ONNX frontend with the latest float16 implementation.
-
Robert Kimball authored
* cleanup op includes in headers * fix missing include
-
Diego Caballero authored
-
- 05 Sep, 2019 6 commits
-
-
Amy Zhuang authored
* Use mkl-dnn v1.0 or v0.x depending on compilation flag. * Change cpu builder files. * Modify cmake files. Use mkldnn-v1.0 for DEX if NGRAPH_USE_MKLDNN_V1 is set to true, otherwise use mkldnn-v0.x. CODEGEN only builds with mkldnn-v1.0. * Implement mkldnn utility functions for mkldnn-v1.0. User mode scratchpad management for mkldnn-v1.0. * Query scratchpad size and allocate a buffer of max scratchpad size. * Do not create mkldnn::memory when query scratchpad size of Reorder. Modify mkldnn utility functions. Fix convolution_forward_init and inner_product_forward_init. Modify CPURuntimeContextCG. * Add user mode scratchpad to CODEGEN. * mkldnn-v1.0 splits LSTM states. Update Rnn/Lstm Op accordingly. * Address PR feedback: use MKLDNN_MAJOR_VERSION. * Modify cpu rnn fusion pass and related unit tests. * Change Rnn/Lstm arg types to Output. * Fix Lstm for CODEGEN. * Set native layout for Slice when input format is blocked. * Do not print scratchpad size. * Change external_mkldnn_v1.cmake. Fix a typo. * Add mkldnn_v1.patch for mkldnn-v1.0. * Address PR feedback. * Define MKLDNN_ERROR_MESSAGE. * Address PR feedback: change to NGRAPH_USE_LEGACY_MKLDNN. * Fix a bug. * Remove unused variable. * Fix compiler warnings. * Fix a bug for CODEGEN. * Move variable only needed for mkldnn-v0.20 inside #if. * Remove unused variables. * No in place Reshape rotation for blocked data layout with mkldnn-v1.0. * Modify mkldnn_v1.patch to force mkldnn to link to libiomp. * Fix style. * Change path for find_library and find_file. * Do not insert ConvertLayout before/after Quantize/DeQuantize for blocked data layout. * Write strides information to visualized graph. * Move variables only needed for mkldnn-v0 under #if. * Move more variables in rnn fusion. * Fix ConvertLayout constant folding for mkldnn-v1.0.
-
Robert Kimball authored
* Add missing includes * style
-
Nagy Mostafa authored
* Don't hold to shared pointers after sub-graph extraction pass is done * style
-
Adam Procter authored
* Add replace_by_friendly_name function * Add replace_nodes function * Nuke replace_by_friendly_name * Modify replace_nodes to handle parameter replacement
-
Tomasz Socha authored
* Move required reshape helpers to builders * Remove warning from matmul_factory * Rid off onnx dependency from matmul factory * Move malmul_factory into builders * Add implementation of fused MatMul * Add MatMul to serializer * Add type_prop tests * Remove reference. * Make more metods private * Use protected instead of private * Fix compilation issues * Change construction of matmul_factory in matmul op * Add MatMul operator into switch of is_supported_impl function * Change transpose flags from int to bool * Review Fix I * Update MatMul op * Use OutputVector instead of NodeVector in MatmulFactories * Fix usage of OutputVector * Convert more shared_ptrs to Outputs * Fix comments after merge * Fix comments after merge II * Fix comments after merge III
-
Denise Kutnick authored
-
- 04 Sep, 2019 6 commits
-
-
Robert Kimball authored
* wip * r50 test * update test * R50 topk calls * Faster topk * Finish topk implementation * cleanup * checkpoint * new test working * more unit test * style * wip * fix tests * add needed header * Add new TopK test to plaid manifest * Fix windows build error:
-
Amy Zhuang authored
* Fold Constant + ConvertLayout. * Address PR Feedback. * No folding if data layout is padded. * Add unit test. * Fix style error.
-
Michał Karzyński authored
* Add opset_version field to Node * Add opset version aliases to Softmax * Add op::set1::Softmax operator * Disable opset 1 ops in INTERPRETER * Add serializer support for Softmax opset 1 * Opset1Transformation pass * Added unit tests to softmax pass * Code refactoring * Added missing virtual to set_opset_version * Clang styles applied * Update src/ngraph/pass/opset1_transform.cpp Co-Authored-By:
Adam Procter <adam.m.procter@intel.com> * Part.1 Code review remarks introduced * Part.2 Code review remarks introduced * Changed opset_version to op_version * Code review remarks introduced * Code review remarks introduced * Set Op as base class for Softmax instead of UnaryElementwiseArithmetic * Fixed unit tests * v1::Softmax::generate_adjoints mark temporarily as not supported * Fix CI. Part.2 * Fix CI. Part.3 * Code review remarks introduced * Rename Opset1Transformation to Opset1Upgrade * Fixed clag style problem with enum switch * Fixes clang compilator error * Removed unused foward declaration * Code review remarks introduced * Added checking if input rank is static
-
Amy Zhuang authored
-
Denise Kutnick authored
* clean up and triage unit tests * add back in onnx and closeness fails * Update unit_test.manifest
-
Diego Caballero authored
* [MLIR] Move MLIR compiler files to contrib/mlir/compiler We'll add contrib/mlir/tools directory in subsequent PRs for ngrpah-opt and other tools. * [MLIR] Clean up MLIR related CMake files This PR prepares the introduction of src/contrib/mlir/tools and LIT testing by refactoring and cleaning up the common part with src/contrib/mlir/compiler: - Common code for src/contrib/mlir/tools and src/contrib/mlir/compiler is moved to src/contrib/mlir/CMakeList.txt. - More generic code, common for src/contrib/mlir/* and LIT configuration files, is moved to cmake/external_mlir.cmake with pre-existing similar code. - Redundant environment var LLVM_DIR is removed in favor of LLVM_CMAKE_PATH. - Redundant cmake messages are removed. * Replacing LLVM_CMAKE_PATH with LLVM_DIR
-
- 03 Sep, 2019 1 commit
-
-
Robert Kimball authored
-
- 30 Aug, 2019 7 commits
-
-
Robert Kimball authored
-
Nagy Mostafa authored
* Op decl. No verification * WIP * WIP: Add lowerer support * Code-gen works. * Added padding support. Needs MLIR fix to work * Remove Padding support for now * Fix arrayref init. Disable tests where reshape optimization don't apply * clean up and style apply * Address PR feedback
-
GlebKazantaev authored
-
Mateusz Bencer authored
* Axes input was added to LRN * Unit tests for axes shape check were added * LRN node deserialization was updated * Fixed EOF and clang style applied * Changed Constant to Parameter type in unit tests * Expanded LRN reference ingterface * Fixed LRN assert description * Fixed passing arguments * Reference implementation for one axis * Implementation for channel * Implementation for hw * working on recurence version * Implemented recurence version for hw * Reference implementation code refactor * Fixed ref LRN implementation and added tests * Added 6D unit test * Clang styles applied * Code review remarks introduced * Support for dynamic shape of axes input * Clang styles applied * Code review remarks introduced * Added checking if axes values are in correct range * Clang styles applied * Removed redundant include * Code review remarks introduced
-
Mateusz Bencer authored
* Renamed auto broadcast name in constructors * Fixed argument doc for Or op * Changed names of fields to serialization * Changed name of auto_broadcast in XOR op * Clang styles applied
-
baojun authored
-
Diego Caballero authored
We'll add contrib/mlir/tools directory in subsequent PRs for ngrpah-opt and other tools.
-
- 29 Aug, 2019 5 commits
-
-
Scott Cyphers authored
* Fix crash when NGRAPH_ENABLE_{VISUALIZE,SERIALIZE}_TRACING=1 * Initialization for kw (#3413) * Fix sum reference to handle corner cases with +-inf (#3412) * Fix sum reference to handle corner cases with +-inf * Review comments, and try to make Windows happy * Update GPU unit_test.manifest * More template grindery, to make macOS happy * Update jenkins-trigger.groovy (#3428) * Cyphers/exec can create tensors (#3445) * Add method to check if an Executable can create tensors * Add a Backend check to see if Executable can create tensors * Update per review comment * style * Possible fix so that backend does not keep holding on to the dummy ng_exec * Separate dynamic loading from static linking (#3456) * Separate dynamic loading from static linking * Missed find_my_pathname * Update jenkins-trigger.groovy (#3518) * add rank id to trace file name (#3525)
-
pawelpiotrowicz authored
* no_fork : Remove allication for empty vector * style
-
pawelpiotrowicz authored
* no_fork: Performance issue vector of string to T * style
-
Nagy Mostafa authored
* WIP * Fix incorrect CK output adjustment * Bug fix and enroce sanity check * Change cycle search depth, and fix sanity check * cpu_fusion.validate_fuse_gru_inputs passes. * Fix as_single_output to be able to always create a GOE * minor fix. style-apply * Clean up debug msgs * Switch to backward cycle check * Enable failing test * PR fixes * Address feedback: Add fwd cycle checks. Make cycle checking depth configurable
-
Scott Cyphers authored
-
- 28 Aug, 2019 4 commits
-
-
Scott Cyphers authored
-
Scott Cyphers authored
* Initialize CPU transformer the same way for static and non-static linking. * Initialize earlier
-
Scott Cyphers authored
Set up visibility support for all the library backends.
-
Diego Caballero authored
-
- 27 Aug, 2019 4 commits
-
-
Adam Procter authored
* New clang-format rules for src/ngraph/op/util * Restore root .clang-format * Remove redundant .clang-formats * Revert "Remove redundant .clang-formats" This reverts commit e158df1e507570b55348ce326fd7e8b4ae819f36. * Remove redundant .clang-formats * New clang-format rules for src/ngraph/op/fused * New clang-format rules for src/ngraph/op/experimental * New clang-format rules for src/ngraph/op/.
-
Adam Procter authored
* New clang-format rules for src/ngraph/frontend * Remove now-redundant .clang-format
-
Adam Procter authored
-
Diego Caballero authored
This PR is a stepping stone towards unifying nGraph MLIRCompiler printing flags with those used in MLIR. It enables flag -print-ir-after-all implemented in MLIR pass manager and adds flag -print-ngraph-ir-after-all to MLIRCompiler so that we can use it to dump IR for those transformations that we do directly in MLIRCompiler without using a proper pass. Eventually, everything should be working as a pass and the nGraph variant of the flag shouldn't be needed. NGRAPH_MLIR_DUMP_ALL macro is no longer needed.
-
- 26 Aug, 2019 1 commit
-
-
Nagy Mostafa authored
* Print CK info in visualizer * style-apply * Fix parameters printing * Fixes * typo
-
- 23 Aug, 2019 2 commits
-
-
Diego Caballero authored
* [MLIR] Bump MLIR repo 8/20/2019 MLIR_ commit 0cdb20a6add19bc96c20dad28589a1e54e4d8469 Author: Lei Zhang <antiagainst@google.com> Date: Tue Aug 20 13:33:41 2019 -0700 Add spv.specConstant and spv._reference_of LLVM: commit 3b9a27b6908040881dad394022f8c472c15c0784 Author: Simon Pilgrim <llvm-dev@redking.me.uk> Date: Tue Aug 20 17:54:37 2019 +0000 Fix typo in comment. NFCI. * Address Bob's feedback
-
Nishant Patel authored
* Segregate builders op wise * Style * Update ngraph.hpp
-