From a1e7e2818ab282e4d3b707feb5783b4bd4fbe45b Mon Sep 17 00:00:00 2001
From: George Wort <george.wort@arm.com>
Date: Tue, 15 Jan 2019 11:00:29 +0000
Subject: COMPMID-1768: NEON: Implement RealDiv

Change-Id: I0868669f7b733df141794fba1d79436e7581bd3a
Reviewed-on: https://review.mlplatform.org/426
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
---
 .../NEON/kernels/NEElementwiseOperationKernel.cpp  | 42 +++++++++++++++++++++-
 1 file changed, 41 insertions(+), 1 deletion(-)

(limited to 'src/core/NEON/kernels/NEElementwiseOperationKernel.cpp')

diff --git a/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp b/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp
index 88fd730554..99a3b5ac66 100644
--- a/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp
+++ b/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp
@@ -123,6 +123,11 @@ inline ScalarType elementwise_arithm_op_scalar(const ScalarType &a, const Scalar
             res = (a - b) * (a - b);
             break;
         }
+        case ArithmeticOperation::DIV:
+        {
+            res = a / b;
+            break;
+        }
         default:
             ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
     }
@@ -154,7 +159,6 @@ inline VectorType elementwise_arithm_op(const VectorType &a, const VectorType &b
             res                  = wrapper::vmul(tmp, tmp);
             break;
         }
-
         default:
             ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
     }
@@ -162,6 +166,20 @@ inline VectorType elementwise_arithm_op(const VectorType &a, const VectorType &b
     return res;
 }
 
+template <>
+inline float32x4_t elementwise_arithm_op<ArithmeticOperation::DIV, float32x4_t>(const float32x4_t &a, const float32x4_t &b)
+{
+    return wrapper::vdiv(a, b);
+}
+
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+template <>
+inline float16x8_t elementwise_arithm_op<ArithmeticOperation::DIV, float16x8_t>(const float16x8_t &a, const float16x8_t &b)
+{
+    return wrapper::vdiv(a, b);
+}
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
 template <ArithmeticOperation op>
 inline float32x4x4_t elementwise_arithm_op(const float32x4x4_t &a, const float32x4x4_t &b)
 {
@@ -833,6 +851,28 @@ Status NEArithmeticOperationKernel::validate(ArithmeticOperation op, const ITens
     return Status{};
 }
 
+/** The division operator */
+
+void NEDivisionOperationKernel::configure(const ITensor *input1, const ITensor *input2, ITensor *output)
+{
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input1->info(), *input2->info(), *output->info()));
+    configure_common(input1, input2, output);
+    _function = configure_arithm_func<ArithmeticOperation::DIV>(input1, input2, output);
+}
+
+Status NEDivisionOperationKernel::validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::F16, DataType::F32);
+    return NEArithmeticOperationKernel::validate_arguments(input1, input2, output);
+}
+
+Status NEDivisionOperationKernel::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input1, *input2, *output));
+    return Status{};
+}
+
 /** Comparison operators (equal, not equal, less than, greater than, less than or equal, greater than or equal) */
 
 void NEComparisonOperationKernel::configure(ComparisonOperation op, const ITensor *input1, const ITensor *input2, ITensor *output)
-- 
cgit v1.2.1