COMPMID-1753: NEON: Implement Less, Greater, GreaterEqual, Equal, Not Equal

Change-Id: I6fa95badcdecb826ac5bd9113f118603d5ac2e82
Reviewed-on: https://review.mlplatform.org/393
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

5 files changed, 259 insertions, 30 deletions

diff --git a/arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h b/arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h
index 93ad437322..f02f71b50e 100644
--- a/arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h
+++ b/arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -14,9 +14,9 @@
  * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY NEAIM, DAMAGES OR OTHER
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
@@ -60,37 +60,35 @@ public:
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
 
+    /** Common signature for all the specialised arithmetic functions
+     *
+     * @param[in] input1 First tensor input. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in] input2 Second tensor input. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor. Data types supported: Dependent on subclass.
+     * @param[in] window Region on which to execute the kernel.
+     */
+    using ElementwiseFunction = void(const ITensor *input1, const ITensor *input2, ITensor *output, const Window &window);
+
 protected:
     /** Validate the argument passed to the kernel
      *
-     * @param[in] input1 First tensor input. Data types supported: S16/F16/S32/F32.
+     * @param[in] input1 First tensor input. Data types supported: QASYMM8/S16/F16/S32/F32.
      * @param[in] input2 Second tensor input. Data types supported: Same as @p input1.
-     * @param[in] output Output tensor. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor. Data types supported: Dependent on subclass.
      */
-    virtual Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) = 0;
+    static Status validate_arguments_common(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output);
 
     /** Commmon configure function for element-wise operators with no additional options (e.g. Min, Max, SquaredDiff)
      *
      */
-    template <ArithmeticOperation op>
     void configure_common(const ITensor *input1, const ITensor *input2, ITensor *output);
 
-    ArithmeticOperation _op; // Code of the operation to execute
+    /** Function to use for the particular tensor types passed to configure() */
+    std::function<void(const ITensor *input1, const ITensor *input2, ITensor *output, const Window &window)> _function;
 
-private:
-    /** Common signature for all the specialised add functions
-     *
-     * @param[in]  input1 An input tensor. Data types supported: S16/F16/S32/F32
-     * @param[in]  input2 An input tensor. Data types supported: S16/F16/S32/F32
-     * @param[out] output The output tensor. Data types supported: S16/F16/S32/F32
-     * @param[in]  window Region on which to execute the kernel.
-     */
-    using ElementwiseFunction = void(const ITensor *input1, const ITensor *input2, ITensor *output, const Window &window);
-    /** Add function to use for the particular tensor types passed to configure() */
-    ElementwiseFunction *_func;
-    const ITensor       *_input1;
-    const ITensor       *_input2;
-    ITensor             *_output;
+    const ITensor *_input1;
+    const ITensor *_input2;
+    ITensor       *_output;
 };
 
 class NEArithmeticOperationKernel : public NEElementwiseOperationKernel
@@ -125,7 +123,40 @@ public:
 
 protected:
     // Inherited methods overridden:
-    Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
+    static Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output);
+};
+
+class NEComparisonOperationKernel : public NEElementwiseOperationKernel
+{
+public:
+    NEComparisonOperationKernel()
+        : NEElementwiseOperationKernel()
+    {
+    }
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+     *
+     * @param[in] op     Comparison operation to be executed.
+     * @param[in] input1 First tensor input. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in] input2 Second tensor input. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor. Data types supported: U16/U32.
+     */
+    void configure(ComparisonOperation op, const ITensor *input1, const ITensor *input2, ITensor *output);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+     *
+     * @param[in] op     Comparison operation to be executed.
+     * @param[in] input1 First tensor input info. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor info. Data types supported: U16/U32.
+     *
+     * @return a Status
+     */
+    static Status validate(ComparisonOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
+
+protected:
+    // Inherited methods overridden:
+    static Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output);
 };
 } // namespace arm_compute
 #endif /* __ARM_COMPUTE_NEELEMENTWISEOPERATIONKERNEL_H__ */
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/cge.h b/arm_compute/core/NEON/wrapper/intrinsics/cge.h
new file mode 100644
index 0000000000..168a6f597d
--- /dev/null
+++ b/arm_compute/core/NEON/wrapper/intrinsics/cge.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_WRAPPER_CGE_H__
+#define __ARM_COMPUTE_WRAPPER_CGE_H__
+
+#include <arm_neon.h>
+
+namespace arm_compute
+{
+namespace wrapper
+{
+#define VCGE_IMPL(stype, vtype, rtype, prefix, postfix) \
+    inline rtype vcge(const vtype &a, const vtype &b)   \
+    {                                                   \
+        return prefix##_##postfix(a, b);                \
+    }
+
+VCGE_IMPL(uint8_t, uint8x8_t, uint8x8_t, vcge, u8)
+VCGE_IMPL(int8_t, int8x8_t, uint8x8_t, vcge, s8)
+VCGE_IMPL(uint16_t, uint16x4_t, uint16x4_t, vcge, u16)
+VCGE_IMPL(int16_t, int16x4_t, uint16x4_t, vcge, s16)
+VCGE_IMPL(uint32_t, uint32x2_t, uint32x2_t, vcge, u32)
+VCGE_IMPL(int32_t, int32x2_t, uint32x2_t, vcge, s32)
+VCGE_IMPL(float32x2_t, float32x2_t, uint32x2_t, vcge, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VCGE_IMPL(float16x4_t, float16x4_t, uint16x4_t, vcge, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+VCGE_IMPL(uint8_t, uint8x16_t, uint8x16_t, vcgeq, u8)
+VCGE_IMPL(int8_t, int8x16_t, uint8x16_t, vcgeq, s8)
+VCGE_IMPL(uint16_t, uint16x8_t, uint16x8_t, vcgeq, u16)
+VCGE_IMPL(int16_t, int16x8_t, uint16x8_t, vcgeq, s16)
+VCGE_IMPL(uint32_t, uint32x4_t, uint32x4_t, vcgeq, u32)
+VCGE_IMPL(int32_t, int32x4_t, uint32x4_t, vcgeq, s32)
+VCGE_IMPL(float32x4_t, float32x4_t, uint32x4_t, vcgeq, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VCGE_IMPL(float16x8_t, float16x8_t, uint16x8_t, vcgeq, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+#undef VCGE_IMPL
+} // namespace wrapper
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_WRAPPER_CGE_H__ */
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/intrinsics.h b/arm_compute/core/NEON/wrapper/intrinsics/intrinsics.h
index a0193ee3d2..c8f4a6e041 100644
--- a/arm_compute/core/NEON/wrapper/intrinsics/intrinsics.h
+++ b/arm_compute/core/NEON/wrapper/intrinsics/intrinsics.h
@@ -29,6 +29,7 @@
 #include "arm_compute/core/NEON/wrapper/intrinsics/and.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/bsl.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/ceq.h"
+#include "arm_compute/core/NEON/wrapper/intrinsics/cge.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/cgt.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/clt.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/combine.h"
@@ -48,6 +49,7 @@
 #include "arm_compute/core/NEON/wrapper/intrinsics/movn.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/mul.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/neg.h"
+#include "arm_compute/core/NEON/wrapper/intrinsics/not.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/orr.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/padd.h"
 #include "arm_compute/core/NEON/wrapper/intrinsics/pmax.h"
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/not.h b/arm_compute/core/NEON/wrapper/intrinsics/not.h
new file mode 100644
index 0000000000..33ac6b5634
--- /dev/null
+++ b/arm_compute/core/NEON/wrapper/intrinsics/not.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_WRAPPER_NOT_H__
+#define __ARM_COMPUTE_WRAPPER_NOT_H__
+
+#include <arm_neon.h>
+
+namespace arm_compute
+{
+namespace wrapper
+{
+#define VNOT_IMPL(stype, vtype, prefix, postfix) \
+    inline vtype vnot(const vtype &a)            \
+    {                                            \
+        return prefix##_##postfix(a);            \
+    }
+
+VNOT_IMPL(uint8_t, uint8x8_t, vmvn, u8)
+VNOT_IMPL(int8_t, int8x8_t, vmvn, s8)
+VNOT_IMPL(uint16_t, uint16x4_t, vmvn, u16)
+VNOT_IMPL(int16_t, int16x4_t, vmvn, s16)
+VNOT_IMPL(uint32_t, uint32x2_t, vmvn, u32)
+VNOT_IMPL(int32_t, int32x2_t, vmvn, s32)
+VNOT_IMPL(float32x2_t, float32x2_t, vinv, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VNOT_IMPL(float16x4_t, float16x4_t, vinv, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+VNOT_IMPL(uint8_t, uint8x16_t, vmvnq, u8)
+VNOT_IMPL(int8_t, int8x16_t, vmvnq, s8)
+VNOT_IMPL(uint16_t, uint16x8_t, vmvnq, u16)
+VNOT_IMPL(int16_t, int16x8_t, vmvnq, s16)
+VNOT_IMPL(uint32_t, uint32x4_t, vmvnq, u32)
+VNOT_IMPL(int32_t, int32x4_t, vmvnq, s32)
+VNOT_IMPL(float32x4_t, float32x4_t, vinvq, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VNOT_IMPL(float16x8_t, float16x8_t, vinvq, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+#undef VNOT_IMPL
+} // namespace wrapper
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_WRAPPER_NOT_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
index 5cbf1237e4..cd9ed24bee 100644
--- a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
+++ b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -14,10 +14,10 @@
  * copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY NEAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARI SING FROM,
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
@@ -33,7 +33,7 @@ class ITensor;
 
 /** Basic function to run @ref NEArithmeticOperationKernel for max
  *
- * @note The tensor data type for the inputs must be S16/F16/S32/F32.
+ * @note The tensor data type for the inputs must be QASYMM8/S16/F16/S32/F32.
  * @note The function performs a max operation between two tensors.
  */
 class NEElementwiseMax : public INESimpleFunction
@@ -59,7 +59,7 @@ public:
 
 /** Basic function to run @ref NEArithmeticOperationKernel for min
  *
- * @note The tensor data type for the inputs must be S16/F16/S32/F32.
+ * @note The tensor data type for the inputs must be QASYMM8/S16/F16/S32/F32.
  * @note The function performs a max operation between two tensors.
  */
 class NEElementwiseMin : public INESimpleFunction
@@ -85,7 +85,7 @@ public:
 
 /** Basic function to run @ref NEArithmeticOperationKernel for squared difference
  *
- * @note The tensor data type for the inputs must be S16/F16/S32/F32.
+ * @note The tensor data type for the inputs must be QASYMM8/S16/F16/S32/F32.
  * @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2
  */
 class NEElementwiseSquaredDiff : public INESimpleFunction
@@ -108,5 +108,73 @@ public:
      */
     static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
 };
+
+/** Basic function to run @ref NEComparisonOperationKernel
+ *
+ * @note The tensor data type for the inputs must be QASYMM8/S16/F16/S32/F32.
+ * @note The function performs a comparison operation between two tensors.
+ */
+class NEElementwiseComparison : public INESimpleFunction
+{
+public:
+    /** Initialise the kernel's inputs, output and conversion policy.
+     *
+     * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
+     * @param[out]     output Output tensor. Data types supported: U16/U32.
+     * @param[in]      op     Comparison Operation to be performed.
+     */
+    void configure(ITensor *input1, ITensor *input2, ITensor *output, ComparisonOperation op);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+     *
+     * @param[in] input1 First tensor input info. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor info. Data types supported: U16/U32.
+     * @param[in] op     Comparison Operation to be performed.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation op);
+};
+
+/** Basic function to run @ref NEComparisonOperationKernel
+ *
+ * @note The tensor data type for the inputs must be QASYMM8/S16/F16/S32/F32.
+ * @note The function performs a comparison operation between two tensors.
+ */
+template <ComparisonOperation op>
+class NEElementwiseComparisonStatic : public INESimpleFunction
+{
+public:
+    /** Initialise the kernel's inputs, output and conversion policy.
+     *
+     * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
+     * @param[out]     output Output tensor. Data types supported: U16/U32.
+     */
+    void configure(ITensor *input1, ITensor *input2, ITensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+     *
+     * @param[in] input1 First tensor input info. Data types supported: QASYMM8/S16/F16/S32/F32.
+     * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
+     * @param[in] output Output tensor info. Data types supported: U16/U32.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
+};
+
+/** Basic function to run equal comparison. */
+using NEEqual = NEElementwiseComparisonStatic<ComparisonOperation::Equal>;
+/** Basic function to run not equal comparison. */
+using NENotEqual = NEElementwiseComparisonStatic<ComparisonOperation::NotEqual>;
+/** Basic function to run greater comparison. */
+using NEGreater = NEElementwiseComparisonStatic<ComparisonOperation::Greater>;
+/** Basic function to run greater-equal comparison. */
+using NEGreaterEqual = NEElementwiseComparisonStatic<ComparisonOperation::GreaterEqual>;
+/** Basic function to run less comparison. */
+using NELess = NEElementwiseComparisonStatic<ComparisonOperation::Less>;
+/** Basic function to run less-equal comparison. */
+using NELessEqual = NEElementwiseComparisonStatic<ComparisonOperation::LessEqual>;
 } // namespace arm_compute
 #endif /* __ARM_COMPUTE_NEELEMENTWISEOPERATIONS_H__ */