From d8e765ba79772ecc3deda89ae3adab903c744296 Mon Sep 17 00:00:00 2001
From: Georgios Pinitas <georgios.pinitas@arm.com>
Date: Wed, 2 Aug 2017 13:44:33 +0100
Subject: COMPMID-472 : Implement Floor for CL and NEON.

Change-Id: I675a4545b1fe9ab665a07c834720bfe7ff589cee
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/82527
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
---
 arm_compute/core/NEON/NEMath.inl | 10 ++++++++++
 1 file changed, 10 insertions(+)

(limited to 'arm_compute/core/NEON/NEMath.inl')

diff --git a/arm_compute/core/NEON/NEMath.inl b/arm_compute/core/NEON/NEMath.inl
index 08f6749ac9..bdd747c4e9 100644
--- a/arm_compute/core/NEON/NEMath.inl
+++ b/arm_compute/core/NEON/NEMath.inl
@@ -54,6 +54,16 @@ const std::array<float32x4_t, 8> log_tab =
     }
 };
 
+inline float32x4_t vfloorq_f32(float32x4_t val)
+{
+    static const float32x4_t CONST_1 = vdupq_n_f32(1.f);
+
+    const int32x4_t   z = vcvtq_s32_f32(val);
+    const float32x4_t r = vcvtq_f32_s32(z);
+
+    return vbslq_f32(vcgtq_f32(r, val), vsubq_f32(r, CONST_1), r);
+}
+
 inline float32x4_t vinvsqrtq_f32(float32x4_t x)
 {
     float32x4_t sqrt_reciprocal = vrsqrteq_f32(x);
-- 
cgit v1.2.1