commit 20cfa45faefbf56f62c8b1aa95dfd0b4f52e5641
author Pablo Marquez Tello <pablo.tello@arm.com> Mon Mar 20 16:29:21 2023 +0000
committer Pablo Marquez Tello <pablo.tello@arm.com> Thu Mar 23 12:01:05 2023 +0000
tree 7c8b7d17ce04714a8214b89adac5e25cafd2c730
parent 251b5145309e941a32c6c7691062544f2fd243fb

Round to nearest with ties to away from zero in Relu

* This patch adds support for rounding modes in vmlaq_qasymm8_signed
  which is used to compute Relu for quantized types

* Partially resolves MLCE-1018

Change-Id: I2a267b84745430e1ffe92b8bc79828a39332db18
Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9354
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

src/core/NEON/NEAsymm.h

diff --git a/src/core/NEON/NEAsymm.h b/src/core/NEON/NEAsymm.h
index 9b92a86..5b8d2be 100644
--- a/src/core/NEON/NEAsymm.h
+++ b/src/core/NEON/NEAsymm.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2020, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -715,6 +715,12 @@
     const uint16x8_t pb = vcombine_u16(vqmovun_s32(rf.val[2]), vqmovun_s32(rf.val[3]));
     return { pa, pb };
 }
+
+template <RoundingPolicy round_policy = RoundingPolicy::TO_ZERO>
+qasymm8x16_signed_t vmlaq_qasymm8(qasymm8x16_signed_t vd, float32x4_t vs, float32x4_t vo);
+
+template <RoundingPolicy round_policy = RoundingPolicy::TO_ZERO>
+qasymm8x16_signed_t vmlaq_qasymm8_signed(qasymm8x16_signed_t vd, float32x4_t vs, float32x4_t vo);
 } // namespace arm_compute
 #include "src/core/NEON/NEAsymm.inl"
 #endif // ARM_COMPUTE_NEASYMM_H