COMPMID-3069: Resolve Bfloat16 validation issues.

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: Ic1c610cc01064071a7ff8b69ea598c4b0ddbe0ff
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2953
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 6 insertions, 2 deletions

diff --git a/tests/validation/reference/GEMMLowp.cpp b/tests/validation/reference/GEMMLowp.cpp
index 61617c8aae..36d86d1532 100644
--- a/tests/validation/reference/GEMMLowp.cpp
+++ b/tests/validation/reference/GEMMLowp.cpp
@@ -26,6 +26,8 @@
 #include "arm_compute/core/Types.h"
 #include "tests/validation/reference/UtilsQuantizedAsymm.h"
 
+#include "support/ToolchainSupport.h"
+
 #include <limits>
 
 namespace arm_compute
@@ -152,7 +154,7 @@ void quantize_down_scale_by_float(const SimpleTensor<TIn> *in, const SimpleTenso
         const float_t multiplier = (is_per_channel) ? result_real_multiplier[i % cols_in] : result_real_multiplier[0];
 
         float_t result_f = static_cast<float_t>(result) * multiplier + static_cast<float_t>(result_offset);
-        result           = static_cast<TIn>(std::round(result_f));
+        result           = static_cast<TIn>(support::cpp11::round(result_f));
 
         // Bounded ReLu
         if(min != max)
diff --git a/tests/validation/reference/QLSTMLayerNormalization.cpp b/tests/validation/reference/QLSTMLayerNormalization.cpp
index 6764a81617..90d59b93ad 100644
--- a/tests/validation/reference/QLSTMLayerNormalization.cpp
+++ b/tests/validation/reference/QLSTMLayerNormalization.cpp
@@ -28,6 +28,8 @@
 #include "PixelWiseMultiplication.h"
 #include "src/core/utils/quantization/AsymmHelpers.cpp"
 
+#include "support/ToolchainSupport.h"
+
 namespace arm_compute
 {
 namespace test
@@ -79,7 +81,7 @@ SimpleTensor<int16_t> qlstm_layer_normalization(const SimpleTensor<int16_t> &src
 
     for(int i = 0; i < output.num_elements(); i++)
     {
-        const auto output_val_s32 = static_cast<int32_t>(std::round(output_float[i] * std::pow(2, 12)));
+        const auto output_val_s32 = static_cast<int32_t>(support::cpp11::round(output_float[i] * std::pow(2, 12)));
         output[i]                 = utility::clamp<int32_t, int16_t>(output_val_s32, std::numeric_limits<int16_t>::min());
     }