Increase tolerance for NEMeanStdDevNormalization
A reduction step is performed with 2x2 reduction vectors for fp32 that
can lead to different results. Increasing the tolerance to accommodate.
Resolves: COMPMID-4644
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I53380bbd64e18efdaace9d0e96bb35cbfcea8f04
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5947
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/NEON/MeanStdDevNormalizationLayer.cpp b/tests/validation/NEON/MeanStdDevNormalizationLayer.cpp
index 90d3d05..dee8f78 100644
--- a/tests/validation/NEON/MeanStdDevNormalizationLayer.cpp
+++ b/tests/validation/NEON/MeanStdDevNormalizationLayer.cpp
@@ -47,7 +47,7 @@
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
RelativeTolerance<half> tolerance_f16(half(0.2f));
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-RelativeTolerance<float> tolerance_f32(1e-8f);
+RelativeTolerance<float> tolerance_f32(1e-4f);
} // namespace
TEST_SUITE(NEON)