Merge pull request #4179 from NervanaSystems/cyphers/m28

Cyphers/m28

Author: diyessi

CMakeLists.txt

@@ -169,7 +169,7 @@ option(NGRAPH_USE_LEGACY_MKLDNN "Use legacy MKLDNN" FALSE)
 option(NGRAPH_MLIR_ENABLE "Control the building of MLIR backend" FALSE)
 option(NGRAPH_INTERPRETER_ENABLE "Control the building of the INTERPRETER backend" TRUE)
 option(NGRAPH_NOP_ENABLE "Control the building of the NOP backend" TRUE)
-option(NGRAPH_GENERIC_CPU_ENABLE "Enable build nGraph for generic CPU backend" FALSE)
+option(NGRAPH_GENERIC_CPU_ENABLE "Enable build nGraph for generic CPU backend" TRUE)
 option(NGRAPH_DEBUG_ENABLE "Enable output for NGRAPH_DEBUG statements" FALSE)
 option(NGRAPH_DEPRECATED_ENABLE "Enable compiler deprecation pragmas for deprecated APIs (recommended only for development use)" FALSE)
 option(NGRAPH_ONNX_IMPORT_ENABLE "Enable ONNX importer" FALSE)
@@ -199,10 +199,7 @@ if (NGRAPH_STATIC_LIB_ENABLE)
     set(NGRAPH_EXPORT_TARGETS_ENABLE OFF)
 endif()
 
-if (NGRAPH_CPU_ENABLE
-        AND
-    (NOT NGRAPH_GENERIC_CPU_ENABLE)
-   )
+if (NGRAPH_CPU_ENABLE)
     set(NGRAPH_INTEL_CPU_ONLY_ENABLE ON)
 endif()
 

doc/sphinx/source/conf.py

@@ -73,11 +73,11 @@
 # built documents.
 #
 # The short X.Y version.
-version = '0.27'
+version = '0.28'
 
 # The Documentation full version, including alpha/beta/rc tags. Some features
 # available in the latest code will not necessarily be documented first
-release = '0.27.1'
+release = '0.28.0'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.
@@ -143,6 +143,7 @@
     ]
 }
 
+html_last_updated_fmt= ''
 
 # -- Options for HTMLHelp output ------------------------------------------
 

doc/sphinx/source/ops/constant.rst

@@ -43,6 +43,6 @@ Outputs
 C++ Interface
 =============
 
-.. doxygenclass:: ngraph::op::Constant
+.. doxygenclass:: ngraph::op::v0::Constant
    :project: ngraph
    :members:

doc/sphinx/source/ops/parameter.rst

@@ -4,7 +4,7 @@
 Parameter
 #########
 
-.. code-block: cpp
+.. code-block:: cpp
 
    Parameter // A function parameter.
 
@@ -38,7 +38,9 @@ Outputs
 | ``output`` | ``element_type`` | ``shape``  |
 +------------+------------------+------------+
 
-A ``Parameter`` produces the value of the tensor passed to the function in the position of the parameter in the function's arguments. The passed tensor must have the element type and shape specified by the parameter.
+A ``Parameter`` produces the value of the tensor passed to the function 
+in the position of the parameter in the function's arguments. The passed 
+tensor must have the element type and shape specified by the parameter.
 
 Backprop
 ========
@@ -51,6 +53,6 @@ Backprop
 C++ Interface
 =============
 
-.. doxygenclass:: ngraph::op::Parameter
+.. doxygenclass:: ngraph::op::v0::Parameter
    :project: ngraph
    :members:

doc/sphinx/source/ops/result.rst

@@ -45,6 +45,6 @@ Mathematical Definition
 C++ Interface
 =============
 
-.. doxygenclass:: ngraph::op::Result
+.. doxygenclass:: ngraph::op::v0::Result
    :project: ngraph
    :members:

doc/sphinx/source/project/release-notes.rst

@@ -21,19 +21,15 @@ We are pleased to announce the release of version |version|.
 Core updates for |version|
 --------------------------
 
-+ New ops
-+ Provenance improvements from 0.25.1
-+ More dynamic shape ops
-+ More informative errors
-
 
 
 Latest documentation updates
 ----------------------------
 
-+ Additional details on quantization
-+ Index updates
-+ API updates
++ Dynamic Shapes and APIs
++ Provenance
++ Add linkages and overview for quantization APIs
++ New ngraph.ai themed illustrations
 
 .. important:: Pre-releases (``-rc-0.*``) have newer features, and are less stable.  
 
@@ -42,6 +38,16 @@ Latest documentation updates
 Changelog on Previous Releases
 ==============================
 
+0.27.1
+
++ Fixes broken serializer for Sum and Product
++ New ops
++ Provenance improvements from 0.25.1
++ More dynamic shape ops
++ More informative errors
++ Additional details on quantization
++ Index updates
++ API updates
 + All ops support ``Output<Node>`` arguments
 + Additional ops
 + ONNX handling unknown domains

doc/sphinx/source/training/index.rst

@@ -5,10 +5,14 @@
 Distributed Training
 ####################
 
+.. important:: Distributed training for CPU backend is not supported. Distributed 
+   training support is provided only with the Intel® Nervana™ Neural Network Processor
+   for Training (NNP-T).
 
 .. toctree::
    :maxdepth: 1
 
-   overview.rst
    data_ingest.rst
 
+
+

doc/sphinx/source/training/overview.rst

@@ -1,32 +1,9 @@
+:orphan:
+
 .. training/overview.rst:
 
 .. _overview:
 
 Basic Concepts
 ==============
 
-.. important:: Distributed training is not officially supported as of version
-   |version|; however, some configuration options have worked for nGraph 
-   devices in testing environments.
-
-
-Data scientists with locally-scalable rack or cloud-based resources will likely
-find it worthwhile to experiment with different modes or variations of
-distributed training. Deployments using nGraph Library with supported backends
-can be configured to train with data parallelism and will soon work with model
-parallelism. Distributing workloads is increasingly important, as more data and
-bigger models mean the ability to :doc:`../core/constructing-graphs/distribute-train`
-work with larger and larger datasets, or to work with models having many layers
-that aren't designed to fit to a single device.
-
-Distributed training with data parallelism splits the data and each worker
-node has the same model; during each iteration, the gradients are aggregated
-across all workers with an op that performs "allreduce", and applied to update
-the weights.
-
-Using multiple machines helps to scale and speed up deep learning. With large 
-mini-batch training, one could train ResNet-50 with Imagenet-1k data to the
-*Top 5* classifier in minutes using thousands of CPU nodes. See
-`arxiv.org/abs/1709.05011`_.
-
-.. _arxiv.org/abs/1709.05011: https://arxiv.org/format/1709.05011

python/setup.py

@@ -26,6 +26,7 @@
 PYNGRAPH_ROOT_DIR = os.path.abspath(os.path.dirname(__file__))
 NGRAPH_DEFAULT_INSTALL_DIR = os.environ.get('HOME')
 NGRAPH_ONNX_IMPORT_ENABLE = os.environ.get('NGRAPH_ONNX_IMPORT_ENABLE')
+NGRAPH_PYTHON_DEBUG = os.environ.get('NGRAPH_PYTHON_DEBUG')
 
 
 def find_ngraph_dist_dir():
@@ -367,6 +368,13 @@ def _add_extra_compile_arg(self, flag, compile_args):
             return True
         return False
 
+    def add_debug_or_release_flags(self):
+        """Return compiler flags for Release and Debug build types."""
+        if NGRAPH_PYTHON_DEBUG in ['TRUE', 'ON', True]:
+            return ['-O0', '-g']
+        else:
+            return ['-O2', '-D_FORTIFY_SOURCE=2']
+
     def build_extensions(self):
         """Build extension providing extra compiler flags."""
         if sys.platform == 'win32':
@@ -388,7 +396,8 @@ def build_extensions(self):
             add_platform_specific_link_args(ext.extra_link_args)
 
             ext.extra_compile_args += ['-Wformat', '-Wformat-security']
-            ext.extra_compile_args += ['-O2', '-D_FORTIFY_SOURCE=2']
+            ext.extra_compile_args += self.add_debug_or_release_flags()
+
             if sys.platform == 'darwin':
                 ext.extra_compile_args += ['-stdlib=libc++']
         build_ext.build_extensions(self)

src/contrib/mlir/CMakeLists.txt

@@ -25,7 +25,7 @@ add_subdirectory(tools/ngraph-opt)
 set(SRC
     backend/cpu/cpu_backend.cpp
     backend/pass/affine_lowerer.cpp
-    backend/pass/memory_optimization.cpp
+    backend/analysis/memory_analysis.cpp
     core/compiler.cpp
     core/ngraph_dialect/dialect.cpp
     core/ngraph_dialect/type.cpp