aboutsummaryrefslogtreecommitdiff
path: root/arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h
diff options
context:
space:
mode:
Diffstat (limited to 'arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h')
-rw-r--r--arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h14
1 files changed, 6 insertions, 8 deletions
diff --git a/arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h b/arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h
index fee206638b..fd0c0f0c34 100644
--- a/arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h
+++ b/arm_compute/core/NEON/kernels/detail/NEDirectConvolution3x3.h
@@ -45,13 +45,11 @@ inline float32x4x3_t load_matrix_row(const float *ptr)
}
template <unsigned int stridex>
-float32x4x2_t convolve_3x3(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, int fixed_point_position);
+float32x4x2_t convolve_3x3(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2);
template <>
-inline float32x4x2_t convolve_3x3<1>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, int fixed_point_position)
+inline float32x4x2_t convolve_3x3<1>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2)
{
- ARM_COMPUTE_UNUSED(fixed_point_position);
-
const float32x4x3_t vtop =
{
{
@@ -108,9 +106,9 @@ inline float32x4x2_t convolve_3x3<1>(const float *in_top, const float *in_mid, c
}
template <>
-inline float32x4x2_t convolve_3x3<2>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, int fixed_point_position)
+inline float32x4x2_t convolve_3x3<2>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2)
{
- float32x4x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2, fixed_point_position);
+ float32x4x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2);
out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[0], 2), out.val[0], 1);
out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[1], 0), out.val[0], 2);
out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[1], 2), out.val[0], 3);
@@ -118,9 +116,9 @@ inline float32x4x2_t convolve_3x3<2>(const float *in_top, const float *in_mid, c
}
template <>
-inline float32x4x2_t convolve_3x3<3>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, int fixed_point_position)
+inline float32x4x2_t convolve_3x3<3>(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2)
{
- float32x4x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2, fixed_point_position);
+ float32x4x2_t out = convolve_3x3<1>(in_top, in_mid, in_low, m0, m1, m2);
out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[0], 3), out.val[0], 1);
return out;
}