diff options
author | Narumol Prangnawarat <narumol.prangnawarat@arm.com> | 2020-03-26 09:20:43 +0000 |
---|---|---|
committer | Narumol Prangnawarat <narumol.prangnawarat@arm.com> | 2020-03-26 16:16:55 +0000 |
commit | 57ef0088d20dd708ff92222d244ea02f1e1e5216 (patch) | |
tree | ae11f55f6bac939a51d5182eae441d322efb3e0e /src/armnn/NetworkUtils.hpp | |
parent | 9272f8b9050096f39796227c5d89ed7b9905146d (diff) | |
download | armnn-57ef0088d20dd708ff92222d244ea02f1e1e5216.tar.gz |
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
Diffstat (limited to 'src/armnn/NetworkUtils.hpp')
-rw-r--r-- | src/armnn/NetworkUtils.hpp | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/src/armnn/NetworkUtils.hpp b/src/armnn/NetworkUtils.hpp index 064545aac5..a922770285 100644 --- a/src/armnn/NetworkUtils.hpp +++ b/src/armnn/NetworkUtils.hpp @@ -15,6 +15,10 @@ std::vector<ConvertBf16ToFp32Layer*> InsertConvertBf16ToFp32LayersBefore(Graph& Layer& layer, bool expectCorrectInputType = true); +std::vector<ConvertFp32ToBf16Layer*> InsertConvertFp32ToBf16LayersBefore(Graph& graph, + Layer& layer, + bool expectCorrectInputType = true); + std::vector<ConvertFp32ToBf16Layer*> InsertConvertFp32ToBf16LayersAfter(Graph& graph, Layer& layer); std::vector<ConvertFp16ToFp32Layer*> InsertConvertFp16ToFp32LayersBefore(Graph& graph, |