diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-08-01 15:03:00 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-08-01 18:15:42 +0000 |
commit | 1c29ffc7fe02de68cf2e82709a3bc3e210cb0ba4 (patch) | |
tree | eeeb02221fe972141f54c01d33e6ed57eb0f59df /src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp | |
parent | 169cda3793ce4900b2bf103739f04bb83b1b6aae (diff) | |
download | ComputeLibrary-1c29ffc7fe02de68cf2e82709a3bc3e210cb0ba4.tar.gz |
COMPMID-2336: Fix build issues.
Change-Id: I0932dc9ca4649f0825950ed9d6d249212bc6971e
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1671
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp')
-rw-r--r-- | src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp | 11 |
1 files changed, 11 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp b/src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp index f638f0bb38..e8f44b6bfd 100644 --- a/src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp +++ b/src/core/NEON/kernels/convolution/depthwise/impl_qa8_qa8.hpp @@ -373,12 +373,23 @@ static inline void tilefn( final_accs[i] = vminq_s32(final_accs[i], vdupq_n_s32(clamp_max)); } +#ifndef __aarch64__ + const int16x8x2_t zelems = vuzpq_s16(vreinterpretq_s16_s32(final_accs[0]), + vreinterpretq_s16_s32(final_accs[1])); + const int8x16_t elems = vreinterpretq_s8_s16(zelems.val[0]); + + const int8x16x2_t zoutput = vuzpq_s8(elems, elems); + const uint8x8_t output = + vget_low_u8(vreinterpretq_u8_s8(zoutput.val[0])); + vst1_u8(get_output_ptr(oi, oj, channel), output); +#else const int8x16_t elems = vreinterpretq_s8_s16( vuzp1q_s16(vreinterpretq_s16_s32(final_accs[0]), vreinterpretq_s16_s32(final_accs[1]))); const uint8x8_t output = vget_low_u8(vreinterpretq_u8_s8(vuzp1q_s8(elems, elems))); vst1_u8(get_output_ptr(oi, oj, channel), output); +#endif // __aarch64__ } } } |