MLBEDSW-5209 Vela: output diff depthwise with non-zero zero points
Fixed by adjusting zero points for ops with int8 IFM and asymmetric weights
since the reference does not support asymmetric weights for int8 IFM and
ignores the zero points.
Signed-off-by: Fredrik Svedberg <fredrik.svedberg@arm.com>
Change-Id: I2a206a01a471a53aa864a6a3616aa23d2a5a23c8
diff --git a/ethosu/vela/tflite_writer.py b/ethosu/vela/tflite_writer.py
index e6dd85b..d134c07 100644
--- a/ethosu/vela/tflite_writer.py
+++ b/ethosu/vela/tflite_writer.py
@@ -236,6 +236,8 @@
QuantizationParameters.QuantizationParametersAddScale(builder, scale)
if zero_point is not None:
QuantizationParameters.QuantizationParametersAddZeroPoint(builder, zero_point)
+ if quant.quant_dim is not None:
+ QuantizationParameters.QuantizationParametersAddQuantizedDimension(builder, quant.quant_dim)
qp = QuantizationParameters.QuantizationParametersEnd(builder)
return qp