projects / platform / core / ml / nnfw.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 95b203f) | patch
[nnc] Do not transpose weights when serializing FullyConnected operation for x86...
authorСергей Баранников/AI Tools Lab /SRR/Engineer/삼성전자 <s.barannikov@samsung.com>
Thu, 20 Dec 2018 17:36:02 +0000 (20:36 +0300)
committerРоман Михайлович Русяев/AI Tools Lab /SRR/Staff Engineer/삼성전자 <r.rusyaev@samsung.com>
Thu, 20 Dec 2018 17:36:02 +0000 (20:36 +0300)
commit3df55fb2c051db5570522e4a59b104952c4ebb10
tree166d9c2ba9986a3520f576e3e1d0276d82b010eatree | snapshot
parent95b203f674f7f77836fb14cf1f559cb6e1a32fddcommit | diff
[nnc] Do not transpose weights when serializing FullyConnected operation for x86-spp back end (#2736)

Remove call to transposeTensor<1, 0> when serializing weights tensor for FullyConnected op for x86-cpp backend.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
contrib/nnc/passes/soft_backend/SBSerializer.cpp diff | blob | history
contrib/nnc/passes/soft_backend/code_snippets/cpp_fully_connected.def diff | blob | history
contrib/nnc/unittests/soft_backend/CPPOperations.cpp diff | blob | history
Domain: Machine Learning / ML Framework;