commit e3fad1913ea83789d1a026c6fb7829fd30e42f6c
author Marat Dukhan <maratek@google.com> Fri Nov 22 13:01:42 2019 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Fri Nov 22 13:12:15 2019 -0800
tree 32a0e349b6db80f95422a1ac0a6c9b0aa6684cb5
parent 8c3b028228ba9fdcb2f0a6e109e460aaae49fda1

Fix wrong vreinterpret intrinsics in NEONFMA SpCHW DWCONV micro-kernels

Fix compilation errors on ARM64 gcc

PiperOrigin-RevId: 282025289

src/f32-dwconv-spchw/5x5s2p2-neonfma.c

diff --git a/src/f32-dwconv-spchw/5x5s2p2-neonfma.c b/src/f32-dwconv-spchw/5x5s2p2-neonfma.c
index b955004..7d48e69 100644
--- a/src/f32-dwconv-spchw/5x5s2p2-neonfma.c
+++ b/src/f32-dwconv-spchw/5x5s2p2-neonfma.c
@@ -253,11 +253,11 @@
       float32x4_t vi4x468A = vuzp1q_f32(vi4x4567, vi4x89AB);
       float32x4_t vi4x579B = vuzp2q_f32(vi4x4567, vi4x89AB);
 
-      vi0x468A = vreinterpretq_u32_f32(vandq_u32(vmask_even, vreinterpretq_f32_u32(vi0x468A)));
-      vi1x468A = vreinterpretq_u32_f32(vandq_u32(vmask_even, vreinterpretq_f32_u32(vi1x468A)));
-      vi2x468A = vreinterpretq_u32_f32(vandq_u32(vmask_even, vreinterpretq_f32_u32(vi2x468A)));
-      vi3x468A = vreinterpretq_u32_f32(vandq_u32(vmask_even, vreinterpretq_f32_u32(vi3x468A)));
-      vi4x468A = vreinterpretq_u32_f32(vandq_u32(vmask_even, vreinterpretq_f32_u32(vi4x468A)));
+      vi0x468A = vreinterpretq_f32_u32(vandq_u32(vmask_even, vreinterpretq_u32_f32(vi0x468A)));
+      vi1x468A = vreinterpretq_f32_u32(vandq_u32(vmask_even, vreinterpretq_u32_f32(vi1x468A)));
+      vi2x468A = vreinterpretq_f32_u32(vandq_u32(vmask_even, vreinterpretq_u32_f32(vi2x468A)));
+      vi3x468A = vreinterpretq_f32_u32(vandq_u32(vmask_even, vreinterpretq_u32_f32(vi3x468A)));
+      vi4x468A = vreinterpretq_f32_u32(vandq_u32(vmask_even, vreinterpretq_u32_f32(vi4x468A)));
 
       vi0x579B = vreinterpretq_u32_f32(vandq_u32(vmask_odd, vreinterpretq_f32_u32(vi0x579B)));
       vi1x579B = vreinterpretq_u32_f32(vandq_u32(vmask_odd, vreinterpretq_f32_u32(vi1x579B)));