diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-10-16 19:21:40 +0100 |
|---|---|---|
| committer | Michele Di Giorgio <michele.digiorgio@arm.com> | 2019-11-06 15:53:24 +0000 |
| commit | dbdea0d1c025b18d4d82c278c87454427918f5b4 (patch) | |
| tree | 68bc25452f5d5b41006fb507c41516446cf8e457 /arm_compute/core/NEON/NEMath.inl | |
| parent | 75d47330e7ca0325cf5d83711452f6aeb085998f (diff) | |
| download | ComputeLibrary-dbdea0d1c025b18d4d82c278c87454427918f5b4.tar.gz | |
COMPMID-2308: NEConvolutionLayer: support QUANT8_SYMM_PER_CHANNEL filters
Change-Id: Ic1bf5f0d21ccd525f84213a360f7e199d7f50577
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2177
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/core/NEON/NEMath.inl')
| -rw-r--r-- | arm_compute/core/NEON/NEMath.inl | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/arm_compute/core/NEON/NEMath.inl b/arm_compute/core/NEON/NEMath.inl index 61315e8dbc..f1c9c2024b 100644 --- a/arm_compute/core/NEON/NEMath.inl +++ b/arm_compute/core/NEON/NEMath.inl @@ -294,6 +294,14 @@ inline float32x2_t vsin_f32(float32x2_t val) #endif /* DOXYGEN_SKIP_THIS */ +inline int32x4_t rounding_divide_by_pow2(int32x4_t x, int32x4_t exponent) +{ + const int32x4_t shift_vec = vnegq_s32(exponent); + const int32x4_t fixup = vshrq_n_s32(vandq_s32(x, shift_vec), 31); + const int32x4_t fixed_up_x = vqaddq_s32(x, fixup); + return vrshlq_s32(fixed_up_x, shift_vec); +} + inline int32x4_t rounding_divide_by_pow2(int32x4_t x, int exponent) { const int32x4_t shift_vec = vdupq_n_s32(-exponent); |