projects / platform / upstream / armnn.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 83239f9) | patch
NNXSW-1826 Add an optimization step which combines Permute and BatchToSpace into...
authorRob Hughes <robert.hughes@arm.com>
Tue, 24 Sep 2019 15:59:56 +0000 (16:59 +0100)
committerÁron Virginás-Tar <aron.virginas-tar@arm.com>
Fri, 27 Sep 2019 10:17:26 +0000 (10:17 +0000)
commit3a7d3a70d99cbe22f5e4711d5dbbea2a245da7ed
tree445967fa35016374657b5c7e38b2715773a25e02tree | snapshot
parent83239f995e7b86062450794b85bfe4c4c387fda0commit | diff
NNXSW-1826 Add an optimization step which combines Permute and BatchToSpace into DepthToSpace

This is only possible in some limited cases, but removes an extra
layer from the graph and so should improve performance in all cases.

Change-Id: I7b3e6ba5dacb4fdb816ad270edaecda1436ab4cf
Signed-off-by: Rob Hughes <robert.hughes@arm.com>
Android.mk diff | blob | history
CMakeLists.txt diff | blob | history
src/armnn/Network.cpp diff | blob | history
src/armnn/optimizations/All.hpp diff | blob | history
src/armnn/optimizations/PermuteAndBatchToSpaceAsDepthToSpace.cpp [new file with mode: 0644] blob
src/armnn/optimizations/PermuteAndBatchToSpaceAsDepthToSpace.hpp [new file with mode: 0644] blob
src/armnn/test/optimizations/PermuteAndBatchToSpaceAsDepthToSpaceTests.cpp [new file with mode: 0644] blob
Domain: Machine Learning / ML Framework;