COMPMID-3805: Fix SQRT non-zero output for zero input

- For AArch64, NEActivationLayerKernel uses vsqrt rather than
  vinvsqrt.
- For non-AArch64, it masks values to ensure zero input
  results in zero output without producing NaN.
- Test cases for FP16 and FP32's positive boundary values
  are added.

Change-Id: Ic0104ee5d7045059c2e9bd052616a4a3b43a315d
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4150
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

2 files changed, 57 insertions, 0 deletions

diff --git a/src/core/NEON/wrapper/intrinsics/intrinsics.h b/src/core/NEON/wrapper/intrinsics/intrinsics.h
index 495321a6a1..070f3c7065 100644
--- a/src/core/NEON/wrapper/intrinsics/intrinsics.h
+++ b/src/core/NEON/wrapper/intrinsics/intrinsics.h
@@ -66,6 +66,7 @@
 #include "src/core/NEON/wrapper/intrinsics/round.h"
 #include "src/core/NEON/wrapper/intrinsics/setlane.h"
 #include "src/core/NEON/wrapper/intrinsics/sin.h"
+#include "src/core/NEON/wrapper/intrinsics/sqrt.h"
 #include "src/core/NEON/wrapper/intrinsics/store.h"
 #include "src/core/NEON/wrapper/intrinsics/sub.h"
 #include "src/core/NEON/wrapper/intrinsics/tanh.h"
diff --git a/src/core/NEON/wrapper/intrinsics/sqrt.h b/src/core/NEON/wrapper/intrinsics/sqrt.h
new file mode 100644
index 0000000000..11954cf6c9
--- /dev/null
+++ b/src/core/NEON/wrapper/intrinsics/sqrt.h
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2020 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_SQRT_H
+#define ARM_COMPUTE_WRAPPER_SQRT_H
+
+#ifdef __aarch64__
+
+#include <arm_neon.h>
+
+namespace arm_compute
+{
+namespace wrapper
+{
+#define VSQRT_IMPL(type, prefix, postfix) \
+    inline type vsqrt(const type &a)      \
+    {                                     \
+        return prefix##_##postfix(a);     \
+    }
+
+VSQRT_IMPL(float32x2_t, vsqrt, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VSQRT_IMPL(float16x4_t, vsqrt, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+VSQRT_IMPL(float32x4_t, vsqrtq, f32)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+VSQRT_IMPL(float16x8_t, vsqrtq, f16)
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+} // namespace wrapper
+} // namespace arm_compute
+
+#endif // __aarch64__
+
+#endif /* ARM_COMPUTE_WRAPPER_SQRT_H */
\ No newline at end of file