mirrors/opencv
2
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-09-25 04:09:57 +03:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #29862 from kjg0724:variational-refinement-simd

Browse Source 
video: migrate VariationalRefinement SIMD kernels to universal/scalable intrinsics
This commit is contained in:
Alexander Smorkalov
2026-09-14 16:51:16 +03:00
committed by GitHub
parent cd3aa54bc7 2d350df3c3
commit da8527c189
1 changed files with 90 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/video/src/variational_refinement.cpp
+90 -93
View File
@@ -633,25 +633,26 @@ void VariationalRefinementImpl::ComputeDataTerm_ParBody::operator()(const Range
#undef INIT_ROW_POINTERS
int j = 0;
#if CV_SIMD128
v_float32x4 zeta_vec = v_setall_f32(zeta_squared);
v_float32x4 eps_vec = v_setall_f32(epsilon_squared);
v_float32x4 delta_vec = v_setall_f32(delta2);
v_float32x4 gamma_vec = v_setall_f32(gamma2);
v_float32x4 zero_vec = v_setall_f32(0.0f);
v_float32x4 pIx_vec, pIy_vec, pIz_vec, pdU_vec, pdV_vec;
v_float32x4 pIxx_vec, pIxy_vec, pIyy_vec, pIxz_vec, pIyz_vec;
v_float32x4 derivNorm_vec, derivNorm2_vec, weight_vec;
v_float32x4 Ik1z_vec, Ik1zx_vec, Ik1zy_vec;
v_float32x4 pa11_vec, pa12_vec, pa22_vec, pb1_vec, pb2_vec;
#if (CV_SIMD || CV_SIMD_SCALABLE)
const int vlanes = VTraits<v_float32>::vlanes();
v_float32 zeta_vec = vx_setall_f32(zeta_squared);
v_float32 eps_vec = vx_setall_f32(epsilon_squared);
v_float32 delta_vec = vx_setall_f32(delta2);
v_float32 gamma_vec = vx_setall_f32(gamma2);
v_float32 zero_vec = vx_setall_f32(0.0f);
v_float32 pIx_vec, pIy_vec, pIz_vec, pdU_vec, pdV_vec;
v_float32 pIxx_vec, pIxy_vec, pIyy_vec, pIxz_vec, pIyz_vec;
v_float32 derivNorm_vec, derivNorm2_vec, weight_vec;
v_float32 Ik1z_vec, Ik1zx_vec, Ik1zy_vec;
v_float32 pa11_vec, pa12_vec, pa22_vec, pb1_vec, pb2_vec;
for (; j < len - 3; j += 4)
for (; j <= len - vlanes; j += vlanes)
{
pIx_vec = v_load(pIx + j);
pIy_vec = v_load(pIy + j);
pIz_vec = v_load(pIz + j);
pdU_vec = v_load(pdU + j);
pdV_vec = v_load(pdV + j);
pIx_vec = vx_load(pIx + j);
pIy_vec = vx_load(pIy + j);
pIz_vec = vx_load(pIz + j);
pdU_vec = vx_load(pdU + j);
pdV_vec = vx_load(pdV + j);
derivNorm_vec = v_add(v_add(v_mul(pIx_vec, pIx_vec), v_mul(pIy_vec, pIy_vec)), zeta_vec);
Ik1z_vec = v_add(v_add(pIz_vec, v_mul(pIx_vec, pdU_vec)), v_mul(pIy_vec, pdV_vec));
@@ -663,11 +664,11 @@ void VariationalRefinementImpl::ComputeDataTerm_ParBody::operator()(const Range
pb1_vec = v_sub(zero_vec, v_mul(weight_vec, v_mul(pIz_vec, pIx_vec)));
pb2_vec = v_sub(zero_vec, v_mul(weight_vec, v_mul(pIz_vec, pIy_vec)));
pIxx_vec = v_load(pIxx + j);
pIxy_vec = v_load(pIxy + j);
pIyy_vec = v_load(pIyy + j);
pIxz_vec = v_load(pIxz + j);
pIyz_vec = v_load(pIyz + j);
pIxx_vec = vx_load(pIxx + j);
pIxy_vec = vx_load(pIxy + j);
pIyy_vec = vx_load(pIyy + j);
pIxz_vec = vx_load(pIxz + j);
pIyz_vec = vx_load(pIyz + j);
derivNorm_vec = v_add(v_add(v_mul(pIxx_vec, pIxx_vec), v_mul(pIxy_vec, pIxy_vec)), zeta_vec);
derivNorm2_vec = v_add(v_add(v_mul(pIyy_vec, pIyy_vec), v_mul(pIxy_vec, pIxy_vec)), zeta_vec);
@@ -687,6 +688,7 @@ void VariationalRefinementImpl::ComputeDataTerm_ParBody::operator()(const Range
v_store(pb1 + j, pb1_vec);
v_store(pb2 + j, pb2_vec);
}
vx_cleanup();
#endif
for (; j < len; j++)
{
@@ -840,38 +842,40 @@ void VariationalRefinementImpl::ComputeSmoothnessTermHorPass_ParBody::operator()
pA_v_next[j] += pWeight[j];
int j = 0;
#if CV_SIMD128
v_float32x4 alpha2_vec = v_setall_f32(alpha2);
v_float32x4 eps_vec = v_setall_f32(epsilon_squared);
v_float32x4 cW_u_vec, cW_v_vec;
v_float32x4 pWeight_vec, ux_vec, vx_vec, uy_vec, vy_vec;
#if (CV_SIMD || CV_SIMD_SCALABLE)
const int vlanes = VTraits<v_float32>::vlanes();
v_float32 alpha2_vec = vx_setall_f32(alpha2);
v_float32 eps_vec = vx_setall_f32(epsilon_squared);
v_float32 cW_u_vec, cW_v_vec;
v_float32 pWeight_vec, ux_vec, vx_vec, uy_vec, vy_vec;
for (; j < len - 4; j += 4)
for (; j < len - vlanes; j += vlanes) // leave the rightmost element to the scalar code below (no UPDATE across the right border)
{
cW_u_vec = v_load(cW_u + j);
cW_v_vec = v_load(cW_v + j);
cW_u_vec = vx_load(cW_u + j);
cW_v_vec = vx_load(cW_v + j);
ux_vec = v_sub(v_load(cW_u_next + j), cW_u_vec);
vx_vec = v_sub(v_load(cW_v_next + j), cW_v_vec);
uy_vec = v_sub(v_load(cW_u_next_row + j), cW_u_vec);
vy_vec = v_sub(v_load(cW_v_next_row + j), cW_v_vec);
ux_vec = v_sub(vx_load(cW_u_next + j), cW_u_vec);
vx_vec = v_sub(vx_load(cW_v_next + j), cW_v_vec);
uy_vec = v_sub(vx_load(cW_u_next_row + j), cW_u_vec);
vy_vec = v_sub(vx_load(cW_v_next_row + j), cW_v_vec);
pWeight_vec =
v_div(alpha2_vec, v_sqrt(v_add(v_add(v_add(v_add(v_mul(ux_vec, ux_vec), v_mul(vx_vec, vx_vec)), v_mul(uy_vec, uy_vec)), v_mul(vy_vec, vy_vec)), eps_vec)));
v_store(pWeight + j, pWeight_vec);
ux_vec = v_mul(pWeight_vec, v_sub(v_load(pW_u_next + j), v_load(pW_u + j)));
vx_vec = v_mul(pWeight_vec, v_sub(v_load(pW_v_next + j), v_load(pW_v + j)));
ux_vec = v_mul(pWeight_vec, v_sub(vx_load(pW_u_next + j), vx_load(pW_u + j)));
vx_vec = v_mul(pWeight_vec, v_sub(vx_load(pW_v_next + j), vx_load(pW_v + j)));
v_store(pA_u + j, v_add(v_load(pA_u + j), pWeight_vec));
v_store(pA_v + j, v_add(v_load(pA_v + j), pWeight_vec));
v_store(pB_u + j, v_add(v_load(pB_u + j), ux_vec));
v_store(pB_v + j, v_add(v_load(pB_v + j), vx_vec));
v_store(pA_u + j, v_add(vx_load(pA_u + j), pWeight_vec));
v_store(pA_v + j, v_add(vx_load(pA_v + j), pWeight_vec));
v_store(pB_u + j, v_add(vx_load(pB_u + j), ux_vec));
v_store(pB_v + j, v_add(vx_load(pB_v + j), vx_vec));
v_store(pA_u_next + j, v_add(v_load(pA_u_next + j), pWeight_vec));
v_store(pA_v_next + j, v_add(v_load(pA_v_next + j), pWeight_vec));
v_store(pB_u_next + j, v_sub(v_load(pB_u_next + j), ux_vec));
v_store(pB_v_next + j, v_sub(v_load(pB_v_next + j), vx_vec));
v_store(pA_u_next + j, v_add(vx_load(pA_u_next + j), pWeight_vec));
v_store(pA_v_next + j, v_add(vx_load(pA_v_next + j), pWeight_vec));
v_store(pB_u_next + j, v_sub(vx_load(pB_u_next + j), ux_vec));
v_store(pB_v_next + j, v_sub(vx_load(pB_v_next + j), vx_vec));
}
vx_cleanup();
#endif
for (; j < len - 1; j++)
{
@@ -952,24 +956,26 @@ void VariationalRefinementImpl::ComputeSmoothnessTermVertPass_ParBody::operator(
#undef INIT_ROW_POINTERS
int j = 0;
#if CV_SIMD128
v_float32x4 pWeight_vec, uy_vec, vy_vec;
for (; j < len - 3; j += 4)
#if (CV_SIMD || CV_SIMD_SCALABLE)
const int vlanes = VTraits<v_float32>::vlanes();
v_float32 pWeight_vec, uy_vec, vy_vec;
for (; j <= len - vlanes; j += vlanes)
{
pWeight_vec = v_load(pWeight + j);
uy_vec = v_mul(pWeight_vec, v_sub(v_load(pW_u_next_row + j), v_load(pW_u + j)));
vy_vec = v_mul(pWeight_vec, v_sub(v_load(pW_v_next_row + j), v_load(pW_v + j)));
pWeight_vec = vx_load(pWeight + j);
uy_vec = v_mul(pWeight_vec, v_sub(vx_load(pW_u_next_row + j), vx_load(pW_u + j)));
vy_vec = v_mul(pWeight_vec, v_sub(vx_load(pW_v_next_row + j), vx_load(pW_v + j)));
v_store(pA_u + j, v_add(v_load(pA_u + j), pWeight_vec));
v_store(pA_v + j, v_add(v_load(pA_v + j), pWeight_vec));
v_store(pB_u + j, v_add(v_load(pB_u + j), uy_vec));
v_store(pB_v + j, v_add(v_load(pB_v + j), vy_vec));
v_store(pA_u + j, v_add(vx_load(pA_u + j), pWeight_vec));
v_store(pA_v + j, v_add(vx_load(pA_v + j), pWeight_vec));
v_store(pB_u + j, v_add(vx_load(pB_u + j), uy_vec));
v_store(pB_v + j, v_add(vx_load(pB_v + j), vy_vec));
v_store(pA_u_next_row + j, v_add(v_load(pA_u_next_row + j), pWeight_vec));
v_store(pA_v_next_row + j, v_add(v_load(pA_v_next_row + j), pWeight_vec));
v_store(pB_u_next_row + j, v_sub(v_load(pB_u_next_row + j), uy_vec));
v_store(pB_v_next_row + j, v_sub(v_load(pB_v_next_row + j), vy_vec));
v_store(pA_u_next_row + j, v_add(vx_load(pA_u_next_row + j), pWeight_vec));
v_store(pA_v_next_row + j, v_add(vx_load(pA_v_next_row + j), pWeight_vec));
v_store(pB_u_next_row + j, v_sub(vx_load(pB_u_next_row + j), uy_vec));
v_store(pB_v_next_row + j, v_sub(vx_load(pB_v_next_row + j), vy_vec));
}
vx_cleanup();
#endif
for (; j < len; j++)
{
@@ -1056,49 +1062,40 @@ void VariationalRefinementImpl::RedBlackSOR_ParBody::operator()(const Range &ran
#undef INIT_ROW_POINTERS
j = 0;
#if CV_SIMD128
v_float32x4 pW_prev_vec = v_setall_f32(pW_next[-1]);
v_float32x4 pdu_prev_vec = v_setall_f32(pdu_next[-1]);
v_float32x4 pdv_prev_vec = v_setall_f32(pdv_next[-1]);
v_float32x4 omega_vec = v_setall_f32(var->omega);
v_float32x4 pW_vec, pW_next_vec, pW_prev_row_vec;
v_float32x4 pdu_next_vec, pdu_prev_row_vec, pdu_next_row_vec;
v_float32x4 pdv_next_vec, pdv_prev_row_vec, pdv_next_row_vec;
v_float32x4 pW_shifted_vec, pdu_shifted_vec, pdv_shifted_vec;
v_float32x4 pa12_vec, sigmaU_vec, sigmaV_vec, pdu_vec, pdv_vec;
for (; j < len - 3; j += 4)
#if (CV_SIMD || CV_SIMD_SCALABLE)
const int vlanes = VTraits<v_float32>::vlanes();
v_float32 omega_vec = vx_setall_f32(var->omega);
v_float32 pW_vec, pW_prev_row_vec;
v_float32 pdu_next_vec, pdu_prev_row_vec, pdu_next_row_vec;
v_float32 pdv_next_vec, pdv_prev_row_vec, pdv_next_row_vec;
v_float32 pW_shifted_vec, pdu_shifted_vec, pdv_shifted_vec;
v_float32 pa12_vec, sigmaU_vec, sigmaV_vec, pdu_vec, pdv_vec;
for (; j <= len - vlanes; j += vlanes)
{
pW_vec = v_load(pW + j);
pW_next_vec = v_load(pW_next + j);
pW_prev_row_vec = v_load(pW_prev_row + j);
pdu_next_vec = v_load(pdu_next + j);
pdu_prev_row_vec = v_load(pdu_prev_row + j);
pdu_next_row_vec = v_load(pdu_next_row + j);
pdv_next_vec = v_load(pdv_next + j);
pdv_prev_row_vec = v_load(pdv_prev_row + j);
pdv_next_row_vec = v_load(pdv_next_row + j);
pa12_vec = v_load(pa12 + j);
pW_shifted_vec = v_reinterpret_as_f32(
v_extract<3>(v_reinterpret_as_s32(pW_prev_vec), v_reinterpret_as_s32(pW_next_vec)));
pdu_shifted_vec = v_reinterpret_as_f32(
v_extract<3>(v_reinterpret_as_s32(pdu_prev_vec), v_reinterpret_as_s32(pdu_next_vec)));
pdv_shifted_vec = v_reinterpret_as_f32(
v_extract<3>(v_reinterpret_as_s32(pdv_prev_vec), v_reinterpret_as_s32(pdv_next_vec)));
pW_vec = vx_load(pW + j);
pW_prev_row_vec = vx_load(pW_prev_row + j);
pdu_next_vec = vx_load(pdu_next + j);
pdu_prev_row_vec = vx_load(pdu_prev_row + j);
pdu_next_row_vec = vx_load(pdu_next_row + j);
pdv_next_vec = vx_load(pdv_next + j);
pdv_prev_row_vec = vx_load(pdv_prev_row + j);
pdv_next_row_vec = vx_load(pdv_next_row + j);
pa12_vec = vx_load(pa12 + j);
pW_shifted_vec = vx_load(pW_next + j - 1);
pdu_shifted_vec = vx_load(pdu_next + j - 1);
pdv_shifted_vec = vx_load(pdv_next + j - 1);
sigmaU_vec = v_add(v_add(v_add(v_mul(pW_shifted_vec, pdu_shifted_vec), v_mul(pW_vec, pdu_next_vec)), v_mul(pW_prev_row_vec, pdu_prev_row_vec)), v_mul(pW_vec, pdu_next_row_vec));
sigmaV_vec = v_add(v_add(v_add(v_mul(pW_shifted_vec, pdv_shifted_vec), v_mul(pW_vec, pdv_next_vec)), v_mul(pW_prev_row_vec, pdv_prev_row_vec)), v_mul(pW_vec, pdv_next_row_vec));
pdu_vec = v_load(pdu + j);
pdv_vec = v_load(pdv + j);
pdu_vec = v_add(pdu_vec, v_mul(omega_vec, v_sub(v_div(v_sub(v_add(sigmaU_vec, v_load(pb1 + j)), v_mul(pdv_vec, pa12_vec)), v_load(pa11 + j)), pdu_vec)));
pdv_vec = v_add(pdv_vec, v_mul(omega_vec, v_sub(v_div(v_sub(v_add(sigmaV_vec, v_load(pb2 + j)), v_mul(pdu_vec, pa12_vec)), v_load(pa22 + j)), pdv_vec)));
pdu_vec = vx_load(pdu + j);
pdv_vec = vx_load(pdv + j);
pdu_vec = v_add(pdu_vec, v_mul(omega_vec, v_sub(v_div(v_sub(v_add(sigmaU_vec, vx_load(pb1 + j)), v_mul(pdv_vec, pa12_vec)), vx_load(pa11 + j)), pdu_vec)));
pdv_vec = v_add(pdv_vec, v_mul(omega_vec, v_sub(v_div(v_sub(v_add(sigmaV_vec, vx_load(pb2 + j)), v_mul(pdu_vec, pa12_vec)), vx_load(pa22 + j)), pdv_vec)));
v_store(pdu + j, pdu_vec);
v_store(pdv + j, pdv_vec);
pW_prev_vec = pW_next_vec;
pdu_prev_vec = pdu_next_vec;
pdv_prev_vec = pdv_next_vec;
}
vx_cleanup();
#endif
for (; j < len; j++)
{