diff options
| author | steniu01 <steven.niu@arm.com> | 2017-09-11 15:29:12 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
| commit | f81652dc970a8071f41c4138508c39684ed9111a (patch) | |
| tree | 6bbda484b46784f1e9ddf132849c15d16ee61230 /tests/validation/CL/FullyConnectedLayer.cpp | |
| parent | 9fe414430c3c989b1cdc79d41e031495aed2cb7c (diff) | |
| download | ComputeLibrary-f81652dc970a8071f41c4138508c39684ed9111a.tar.gz | |
COMPMID-516 Increase tolerance rate of Scale, Conv, fully connected and GEMM
This patch also fix the scale kernel issue where it was calcuated the
scale factor inside the gpu but now in the CPU. The GPU and CPU gave
different result for simple float division operation
Change-Id: Ib6709cb6c41dcf4fc0fa4eb79e481430695bf40e
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/87266
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>
Diffstat (limited to 'tests/validation/CL/FullyConnectedLayer.cpp')
| -rw-r--r-- | tests/validation/CL/FullyConnectedLayer.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/tests/validation/CL/FullyConnectedLayer.cpp b/tests/validation/CL/FullyConnectedLayer.cpp index 22f27e56dd..35b9d2938b 100644 --- a/tests/validation/CL/FullyConnectedLayer.cpp +++ b/tests/validation/CL/FullyConnectedLayer.cpp @@ -43,9 +43,9 @@ namespace validation namespace { /** Tolerance for float operations */ -RelativeTolerance<float> tolerance_f32(0.001f); -RelativeTolerance<half> tolerance_f16(half(0.2)); -constexpr float tolerance_num = 0.07f; /**< Tolerance number */ +RelativeTolerance<float> tolerance_f32(0.05f); +RelativeTolerance<half_float::half> tolerance_f16(half(0.2)); +constexpr float tolerance_num = 0.07f; /**< Tolerance number */ /** Tolerance for fixed point operations */ constexpr AbsoluteTolerance<float> tolerance_fixed_point(1.f); |