potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-21 19:33:03 +04:00
Code Packages Releases Wiki Activity

Merge pull request #29518 from Teddy-Yangjiale:rvv-lk

Browse Source 
video: add RVV intrinsics path for pyramidal LK optical flow iteration loop
This commit is contained in:
Alexander Smorkalov
2026-07-15 10:35:45 +03:00
committed by GitHub
parent 647b32ac7c eec2c6fd3d
commit 5c7b438295
1 changed files with 65 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/video/src/lkpyramid.cpp
+65
View File
@@ -469,6 +469,10 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
const int16x4_t d28_2 = vdup_n_s16((int16_t)iw10);
const int16x4_t d29_2 = vdup_n_s16((int16_t)iw11);
#elif CV_RVV
const size_t rvv_lanes = __riscv_vsetvlmax_e16m1();
vfloat32m2_t rvvB1 = __riscv_vfmv_v_f_f32m2(0.f, rvv_lanes);
vfloat32m2_t rvvB2 = __riscv_vfmv_v_f_f32m2(0.f, rvv_lanes);
#endif
for( y = 0; y < winSize.height; y++ )
@@ -578,6 +582,61 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
vst1q_f32(nB1, nB1v);
vst1q_f32(nB2, nB2v);
}
#elif CV_RVV
for( ; x < winSize.width*cn; )
{
const size_t vl = __riscv_vsetvl_e16m1(winSize.width*cn - x);
vuint16m1_t j00 = __riscv_vwcvtu_x_x_v_u16m1(
__riscv_vle8_v_u8mf2(Jptr + x, vl), vl);
vint32m2_t interp = __riscv_vwmul_vx_i32m2(
__riscv_vreinterpret_i16m1(j00), (int16_t)iw00, vl);
vuint16m1_t j01 = __riscv_vwcvtu_x_x_v_u16m1(
__riscv_vle8_v_u8mf2(Jptr + x + cn, vl), vl);
interp = __riscv_vwmacc_vx_i32m2(
interp, (int16_t)iw01, __riscv_vreinterpret_i16m1(j01), vl);
vuint16m1_t j10 = __riscv_vwcvtu_x_x_v_u16m1(
__riscv_vle8_v_u8mf2(Jptr + x + stepJ, vl), vl);
interp = __riscv_vwmacc_vx_i32m2(
interp, (int16_t)iw10, __riscv_vreinterpret_i16m1(j10), vl);
vuint16m1_t j11 = __riscv_vwcvtu_x_x_v_u16m1(
__riscv_vle8_v_u8mf2(Jptr + x + stepJ + cn, vl), vl);
interp = __riscv_vwmacc_vx_i32m2(
interp, (int16_t)iw11, __riscv_vreinterpret_i16m1(j11), vl);
interp = __riscv_vsra_vx_i32m2(
__riscv_vadd_vx_i32m2(
interp, 1 << (W_BITS1 - 5 - 1), vl),
W_BITS1 - 5, vl);
vint32m2_t diff = __riscv_vsub_vv_i32m2(
interp,
__riscv_vwcvt_x_x_v_i32m2(
__riscv_vle16_v_i16m1(Iptr + x, vl), vl),
vl);
vint16m1x2_t gradient = __riscv_vlseg2e16_v_i16m1x2(dIptr, vl);
vint32m2_t ix = __riscv_vwcvt_x_x_v_i32m2(
__riscv_vget_v_i16m1x2_i16m1(gradient, 0), vl);
vint32m2_t iy = __riscv_vwcvt_x_x_v_i32m2(
__riscv_vget_v_i16m1x2_i16m1(gradient, 1), vl);
rvvB1 = __riscv_vfadd_tu(
rvvB1, rvvB1,
__riscv_vfcvt_f_x_v_f32m2(
__riscv_vmul_vv_i32m2(diff, ix, vl), vl),
vl);
rvvB2 = __riscv_vfadd_tu(
rvvB2, rvvB2,
__riscv_vfcvt_f_x_v_f32m2(
__riscv_vmul_vv_i32m2(diff, iy, vl), vl),
vl);
x += (int)vl;
dIptr += vl*2;
}
#endif
for( ; x < winSize.width*cn; x++, dIptr += 2 )
@@ -601,6 +660,12 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
ib1 += (float)(nB1[0] + nB1[1] + nB1[2] + nB1[3]);
ib2 += (float)(nB2[0] + nB2[1] + nB2[2] + nB2[3]);
#elif CV_RVV
vfloat32m1_t rvvZero = __riscv_vfmv_v_f_f32m1(0.f, 1);
ib1 += __riscv_vfmv_f_s_f32m1_f32(
__riscv_vfredusum_vs_f32m2_f32m1(rvvB1, rvvZero, rvv_lanes));
ib2 += __riscv_vfmv_f_s_f32m1_f32(
__riscv_vfredusum_vs_f32m2_f32m1(rvvB2, rvvZero, rvv_lanes));
#endif
b1 = ib1*FLT_SCALE;