- 02 Nov, 2017 16 commits
-
-
Jaikrishnan Menon authored
Conflicts: src/ngraph/codegen/compiler.cpp
-
Jai Menon authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
the other backend flag
-
Jaikrishnan Menon authored
The CPU backend is disabled by default and must be enabled with the CMake variable NGRAPH_CPU_ENABLE set to true
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
Conflicts: src/ngraph/runtime/cpu/call_frame.cpp src/ngraph/runtime/cpu/call_frame.hpp src/ngraph/runtime/cpu/external_function.cpp test/cpu.cpp
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Adam Procter authored
Fix bug with multiple updates to adjoint map
-
Adam Procter authored
-
Adam Procter authored
-
- 01 Nov, 2017 6 commits
-
-
Scott Cyphers authored
-
Adam Procter authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
the LLVM integration requires tar.xz decompression support which was added in 3.1
-
Jaikrishnan Menon authored
Conflicts: test/CMakeLists.txt test/execute.cpp
-
Jaikrishnan Menon authored
-
- 31 Oct, 2017 6 commits
-
-
Jai Menon authored
-
Jaikrishnan Menon authored
Also, add more codegen options
-
Jaikrishnan Menon authored
-
Adam Procter authored
-
Jaikrishnan Menon authored
-
Jai Menon authored
-
- 30 Oct, 2017 12 commits
-
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
Ops are now being implemented in priority order based on MNIST MLP requirements
-
Christian Convey authored
- Adds Sphinx docs support. - Sphinx and Doxygen docs now built by CI. - Sphinx and Doxygen docs now installed by `make install`.
-
Robert Kimball authored
-
Robert Kimball authored
-
Adam Procter authored
This adds support for > < >= <= on bool, which by the type checker's standards should have been allowed anyway. Also, this bug had presenting in a different way from expected (should have thrown "unhandled element type" from external_function, instead bonked because it incorrectly dispatched on int8_t). This was due to element::Type::operator== comparing only size, signedness, and floatiness, meaning that char and signed char (aka int8_t) look the same. Fixed by adding a test for m_cname equality.
-
Robert Kimball authored
-
Chris Lishka authored
-
Jaikrishnan Menon authored
-
Jaikrishnan Menon authored
-
Robert Kimball authored
Bob/test cleanup
-
Robert Kimball authored
-