projects / platform / upstream / armnn.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 9272f8b) | patch
IVGCVSW-4597 Modify BF16 optimizer to Convert only inputs and weights of
authorNarumol Prangnawarat <narumol.prangnawarat@arm.com>
Thu, 26 Mar 2020 09:20:43 +0000 (09:20 +0000)
committerNarumol Prangnawarat <narumol.prangnawarat@arm.com>
Thu, 26 Mar 2020 16:16:55 +0000 (16:16 +0000)
commit57ef0088d20dd708ff92222d244ea02f1e1e5216
treeae11f55f6bac939a51d5182eae441d322efb3e0etree | snapshot
parent9272f8b9050096f39796227c5d89ed7b9905146dcommit | diff
IVGCVSW-4597 Modify BF16 optimizer to Convert only inputs and weights of
Convolution2d and FullyConnected layers

 * Add InsertConvertFp32ToBf16LayersBefore
 * Add ConvertWeight to ConvertFp32NetworkToBf16Impl for Conv2d and FullyConnected
 * Allow different input and output when input is BF16 and output is FP32
Conv2d and FullyConnected layers
 * Unit tests

Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: Ic8f92ff28edcae08a72a3114a28f50c4619f919b
src/armnn/Network.cpp diff | blob | history
src/armnn/NetworkUtils.cpp diff | blob | history
src/armnn/NetworkUtils.hpp diff | blob | history
src/armnn/optimizations/ConvertFp32NetworkToBf16.hpp diff | blob | history
src/armnn/test/optimizations/Fp32NetworkToBf16ConverterTests.cpp diff | blob | history
src/backends/backendsCommon/WorkloadData.cpp diff | blob | history
src/backends/reference/RefLayerSupport.cpp diff | blob | history
Domain: Machine Learning / ML Framework;