diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-08-23 10:25:06 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:54 +0000 |
| commit | 41acb76af9c8512ac39121103b21ce2aafbcbfe8 (patch) | |
| tree | 788106d83c95c88954698a3f7d25d02db1cfe024 /arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h | |
| parent | 02baf01d75dc639440cf6a3196162f02413661dc (diff) | |
| download | ComputeLibrary-41acb76af9c8512ac39121103b21ce2aafbcbfe8.tar.gz | |
COMPMID-1534 - Fixing FP16 tests on NEON
- Fixed GEMMConvolutionLayer test. The issue was related to the tolerance
- Fixed DirectConvolutioNLayer test. The issue was in the convolver_3x3
Change-Id: I9d5b906d7e5e32a0a34300d529d6edb804ac1c4e
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/145377
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h')
| -rw-r--r-- | arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h | 10 |
1 files changed, 8 insertions, 2 deletions
diff --git a/arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h b/arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h index d56fd44700..b245505ac6 100644 --- a/arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h +++ b/arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h @@ -460,8 +460,12 @@ inline float16x8x2_t convolve_3x3<2>(const float16_t *in_top, const float16_t *i { float16x8x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 2), out.val[0], 1); - out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 0), out.val[0], 2); - out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 2), out.val[0], 3); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 4), out.val[0], 2); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 6), out.val[0], 3); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 0), out.val[0], 4); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 2), out.val[0], 5); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 4), out.val[0], 6); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 6), out.val[0], 7); return out; } @@ -470,6 +474,8 @@ inline float16x8x2_t convolve_3x3<3>(const float16_t *in_top, const float16_t *i { float16x8x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 3), out.val[0], 1); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 6), out.val[0], 2); + out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 1), out.val[0], 3); return out; } |