diff options
author | Aron Virginas-Tar <Aron.Virginas-Tar@arm.com> | 2019-08-01 11:24:08 +0100 |
---|---|---|
committer | Áron Virginás-Tar <aron.virginas-tar@arm.com> | 2019-08-01 10:26:21 +0000 |
commit | a0baa17e501970f5a60227a9ccaa95ab0b9730c9 (patch) | |
tree | 6fcecd6c836e3559ca75de5c61e371cf5b518814 /ConversionUtils.hpp | |
parent | b8805204fb0ea64079735921bcc0cc2b1aedfcf6 (diff) | |
download | android-nn-driver-a0baa17e501970f5a60227a9ccaa95ab0b9730c9.tar.gz |
Fix explanatory comment around SanitizeBiasQuantizationScale()
Signed-off-by: Aron Virginas-Tar <Aron.Virginas-Tar@arm.com>
Change-Id: I92700da149a800f1d8215e964d1e67056628f067
Diffstat (limited to 'ConversionUtils.hpp')
-rw-r--r-- | ConversionUtils.hpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/ConversionUtils.hpp b/ConversionUtils.hpp index 2fa8a072..18a65413 100644 --- a/ConversionUtils.hpp +++ b/ConversionUtils.hpp @@ -337,8 +337,8 @@ Shape GetOperandShape(const V1_0::Operand& operand) // ArmNN requires the bias scale to be equal to the product of the weight and input scales, which is also // what AndroidNN requires. However for some of the AndroidNN tests the values don't exactly match so -// we accept some tolerance. We don't want to ArmNN itself to accept these inconsistencies as it is up to the user -// (us, in this case) to ensure they match. +// we accept some tolerance. We don't want ArmNN itself to accept these inconsistencies as it is up to the +// user (us, in this case) to ensure they match. void SanitizeBiasQuantizationScale(armnn::TensorInfo& biasInfo, const armnn::TensorInfo& weightInfo, const armnn::TensorInfo& inputInfo) { |