Merge pull request #26454 from fengyuentau/imgproc:update_warp_c4_kernels

imgproc: fix perf regressions on the c4 kernels of warpAffine / warpPerspective / remap #26454 ## Performance Previous performance regressions on c4 kernels are mainly on A311D https://github.com/opencv/opencv/pull/26348. Regressions on c3 kernels on intel platform will be fixed in another pull request. M2 ``` Geometric mean (ms) Name of Test base patch patch vs base (x-factor) WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.338 0.163 2.08 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.310 0.107 2.90 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.344 0.162 2.13 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.313 0.111 2.83 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.676 0.333 2.03 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.640 0.240 2.66 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.212 0.885 1.37 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 1.153 0.756 1.53 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.950 0.475 2.00 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 1.158 0.500 2.32 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 3.441 3.106 1.11 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 3.351 2.837 1.18 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.336 0.163 2.07 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.314 0.124 2.54 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.385 0.226 1.70 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.364 0.183 1.99 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.541 0.290 1.87 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.523 0.243 2.16 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.540 1.239 1.24 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 1.504 1.134 1.33 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.751 0.465 1.62 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.958 0.507 1.89 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 3.785 3.487 1.09 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 3.602 3.280 1.10 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.331 0.153 2.16 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.304 0.128 2.37 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.329 0.156 2.11 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.306 0.121 2.53 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 2.046 0.930 2.20 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 2.122 1.391 1.53 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 2.035 0.954 2.13 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 2.127 1.410 1.51 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.329 0.157 2.09 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.306 0.124 2.47 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.327 0.158 2.08 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.308 0.127 2.43 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 2.039 0.948 2.15 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 2.175 1.373 1.58 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 2.065 0.956 2.16 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 2.158 1.372 1.57 ``` Intel i7-12700K: ``` Geometric mean (ms) Name of Test base patch patch vs base (x-factor) WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.140 0.051 2.77 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.140 0.054 2.57 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.140 0.050 2.78 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.143 0.054 2.64 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.297 0.118 2.51 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.296 0.130 2.28 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.481 0.304 1.58 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.470 0.309 1.52 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.381 0.184 2.07 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.811 0.781 1.04 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.297 1.063 1.22 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 1.275 1.171 1.09 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.135 0.057 2.36 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.134 0.062 2.16 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.155 0.076 2.04 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.150 0.079 1.90 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.229 0.114 2.02 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.227 0.120 1.89 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.560 0.444 1.26 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.529 0.442 1.20 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.326 0.192 1.70 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.805 0.762 1.06 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.395 1.255 1.11 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 1.381 1.306 1.06 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.138 0.049 2.81 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.134 0.053 2.53 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.137 0.049 2.79 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.134 0.053 2.51 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 1.362 1.352 1.01 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 3.124 3.038 1.03 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.354 1.351 1.00 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 3.142 3.049 1.03 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 0.140 0.052 2.70 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 0.136 0.056 2.43 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 0.139 0.051 2.70 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 0.135 0.056 2.41 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 1.335 1.345 0.99 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 3.117 3.024 1.03 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.327 1.319 1.01 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 3.126 3.026 1.03 ``` A311D ``` Geometric mean (ms) Name of Test base patch patch vs base (x-factor) WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 1.762 1.361 1.29 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 2.390 2.005 1.19 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.747 1.238 1.41 WarpAffine::TestWarpAffine::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 2.399 2.016 1.19 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 3.917 3.104 1.26 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 5.995 5.172 1.16 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 6.711 5.460 1.23 WarpAffine::TestWarpAffine::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 8.017 6.890 1.16 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 6.269 5.596 1.12 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 10.301 9.507 1.08 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 18.871 17.375 1.09 WarpAffine::TestWarpAffine::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 20.365 18.227 1.12 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 2.083 1.514 1.38 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 2.966 2.309 1.28 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 2.358 1.715 1.37 WarpPerspective::TestWarpPerspective::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 3.220 2.464 1.31 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 3.763 3.014 1.25 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 5.777 4.940 1.17 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 8.791 7.819 1.12 WarpPerspective::TestWarpPerspective::(1280x720, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 10.165 8.426 1.21 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 6.047 5.293 1.14 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 9.851 9.023 1.09 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 31.739 29.323 1.08 WarpPerspective::TestWarpPerspective::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 32.439 29.236 1.11 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 1.759 1.441 1.22 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 2.681 2.270 1.18 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 1.774 1.425 1.24 map1_32fc1::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 2.672 2.252 1.19 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 14.079 9.334 1.51 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 17.770 16.155 1.10 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 15.872 11.192 1.42 map1_32fc1::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 19.167 15.342 1.25 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 2.284 1.545 1.48 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 3.040 2.231 1.36 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 2.280 1.380 1.65 map1_32fc2::TestRemap::(640x480, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 2.882 2.185 1.32 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 16UC4) 15.877 11.381 1.40 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_CONSTANT, 32FC4) 19.521 16.106 1.21 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 16UC4) 15.950 11.532 1.38 map1_32fc2::TestRemap::(1920x1080, INTER_LINEAR, BORDER_REPLICATE, 32FC4) 19.699 16.276 1.21 ``` ### Pull Request Readiness Checklist See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request - [x] I agree to contribute to the project under Apache 2 License. - [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV - [x] The PR is proposed to the proper branch - [x] There is a reference to the original bug report and related work - [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable Patch to opencv_extra has the same branch name. - [x] The feature is well documented and sample code can be built with the project CMake
1 week ago · ea0f9336e2
parent 8b84fcb376
commit ea0f9336e2
3 changed files with 400 additions and 134 deletions
--- a/modules/core/include/opencv2/core/hal/intrin_rvv_scalable.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_rvv_scalable.hpp
@ -543,6 +543,40 @@ OPENCV_HAL_IMPL_RVV_LOADSTORE_OP(v_float32, vfloat32m2_t, float, VTraits<v_float
 OPENCV_HAL_IMPL_RVV_LOADSTORE_OP(v_float64, vfloat64m2_t, double, VTraits<v_float64>::vlanes() / 2, VTraits<v_float64>::vlanes(), 64, f64)
 #endif

+template <int N = VTraits<v_uint16>::max_nlanes>
+inline void v_store(ushort* ptr, const v_uint16& a)
+{
+    ushort buf[VTraits<v_uint16>::max_nlanes];
+    v_store(buf, a);
+    for (int i = 0; i < N; i++) {
+        ptr[i] = buf[i];
+    }
+}
+template <> inline void v_store<8>(ushort* ptr, const v_uint16& a)
+{
+    ushort buf[VTraits<v_uint16>::max_nlanes];
+    v_store(buf, a);
+    ptr[0] = buf[0]; ptr[1] = buf[1]; ptr[2] = buf[2]; ptr[3] = buf[3];
+    ptr[4] = buf[4]; ptr[5] = buf[5]; ptr[6] = buf[6]; ptr[7] = buf[7];
+}
+
+template <int N = VTraits<v_float32>::max_nlanes>
+inline void v_store(float* ptr, const v_float32& a)
+{
+    float buf[VTraits<v_float32>::max_nlanes];
+    v_store(buf, a);
+    for (int i = 0; i < N; i++) {
+        ptr[i] = buf[i];
+    }
+}
+template <> inline void v_store<4>(float* ptr, const v_float32& a)
+{
+    float buf[VTraits<v_float32>::max_nlanes];
+    v_store(buf, a);
+    ptr[0] = buf[0]; ptr[1] = buf[1];
+    ptr[2] = buf[2]; ptr[3] = buf[3];
+}
+
 ////////////// Lookup table access ////////////////////
 #define OPENCV_HAL_IMPL_RVV_LUT(_Tpvec, _Tp, suffix) \
 inline _Tpvec v_lut(const _Tp* tab, const int* idx) \
@ -1616,6 +1650,42 @@ OPENCV_HAL_IMPL_RVV_EXPAND(short, v_int32, vint32m4_t, v_int16, 16, i32, i16, __
 OPENCV_HAL_IMPL_RVV_EXPAND(uint, v_uint64, vuint64m4_t, v_uint32, 32, u64, u32, __riscv_vwcvtu_x)
 OPENCV_HAL_IMPL_RVV_EXPAND(int, v_int64, vint64m4_t, v_int32, 32, i64, i32, __riscv_vwcvt_x)

+template <int N = VTraits<v_float32>::max_nlanes>
+inline v_float32 v_load(const float* ptr)
+{
+    float buf[VTraits<v_float32>::max_nlanes];
+    v_store(buf, v_setzero_f32());
+    for (int i = 0; i < N; i++) {
+        buf[i] = ptr[i];
+    }
+    return v_load(buf);
+}
+template <> inline v_float32 v_load<4>(const float* ptr)
+{
+    float buf[VTraits<v_float32>::max_nlanes];
+    v_store(buf, v_setzero_f32());
+    buf[0] = ptr[0]; buf[1] = ptr[1]; buf[2] = ptr[2]; buf[3] = ptr[3];
+    return v_load(buf);
+}
+
+template <int N = VTraits<v_uint32>::max_nlanes>
+inline v_uint32 v_load_expand(const ushort* ptr)
+{
+    ushort buf[VTraits<v_uint16>::max_nlanes];
+    v_store(buf, v_setzero_u16());
+    for (int i = 0; i < N; i++) {
+        buf[i] = ptr[i];
+    }
+    return v_load_expand(buf);
+}
+template <> inline v_uint32 v_load_expand<4>(const ushort* ptr)
+{
+    ushort buf[VTraits<v_uint16>::max_nlanes];
+    v_store(buf, v_setzero_u16());
+    buf[0] = ptr[0]; buf[1] = ptr[1]; buf[2] = ptr[2]; buf[3] = ptr[3];
+    return v_load_expand(buf);
+}
+
 inline v_uint32 v_load_expand_q(const uchar* ptr)
 {
    return __riscv_vwcvtu_x(__riscv_vwcvtu_x(__riscv_vle8_v_u8mf2(ptr, VTraits<v_uint32>::vlanes()), VTraits<v_uint32>::vlanes()), VTraits<v_uint32>::vlanes());
@ -1627,16 +1697,16 @@ inline v_int32 v_load_expand_q(const schar* ptr)
 }

 template <int N = VTraits<v_uint32>::max_nlanes>
-inline v_uint32 v_load_expand_q(const uchar* ptr, int n = N)
+inline v_uint32 v_load_expand_q(const uchar* ptr)
 {
    uchar buf[VTraits<v_uint8>::max_nlanes];
    v_store(buf, v_setzero_u8());
-    for (int i = 0; i < n; i++) {
+    for (int i = 0; i < N; i++) {
        buf[i] = ptr[i];
    }
    return v_load_expand_q(buf);
 }
-template <> inline v_uint32 v_load_expand_q<4>(const uchar* ptr, int n)
+template <> inline v_uint32 v_load_expand_q<4>(const uchar* ptr)
 {
    uchar buf[VTraits<v_uint8>::max_nlanes];
    v_store(buf, v_setzero_u8());
@ -1714,19 +1784,48 @@ void v_rshr_pack_u_store(_Tp* ptr, const _wTpvec& a, int n = N) \
 OPENCV_HAL_IMPL_RVV_PACK_U(v_uint8, uchar, v_int16, short, 8, 16, u8, i16, __riscv_vreinterpret_v_i16m4_u16m4, VTraits<v_int16>::vlanes(), VTraits<v_uint8>::vlanes())
 OPENCV_HAL_IMPL_RVV_PACK_U(v_uint16, ushort, v_int32, int, 16, 32, u16, i32,  __riscv_vreinterpret_v_i32m4_u32m4, VTraits<v_int32>::vlanes(), VTraits<v_uint16>::vlanes())

+template <int N = VTraits<v_uint16>::max_nlanes>
+inline v_uint16 v_pack_u(const v_int32& a, const v_int32& b)
+{
+    ushort bufa[N];
+    ushort bufb[N];
+    v_pack_u_store(bufa, a);
+    v_pack_u_store(bufb, b);
+    ushort buf[N];
+    for (int i = 0; i < N; i++) {
+        buf[i] = bufa[i];
+        buf[i+N/2] = bufb[i];
+    }
+    return v_load(buf);
+}
+
+template <> inline v_uint16 v_pack_u<4>(const v_int32& a, const v_int32& b)
+{
+    constexpr int N = VTraits<v_uint16>::max_nlanes;
+    ushort bufa[N];
+    ushort bufb[N];
+    v_pack_u_store(bufa, a);
+    v_pack_u_store(bufb, b);
+
+    ushort buf[N];
+    buf[0] = bufa[0]; buf[1] = bufa[1]; buf[2] = bufa[2]; buf[3] = bufa[3];
+    buf[4] = bufb[0]; buf[5] = bufb[1]; buf[6] = bufb[2]; buf[7] = bufb[3];
+    return v_load(buf);
+}
+
 template <int N = VTraits<v_int16>::max_nlanes>
-inline void v_pack_u_store(uchar* ptr, const v_int16& a, int n = N)
+inline void v_pack_store(uchar* ptr, const v_uint16& a)
 {
    uchar buf[VTraits<v_uint8>::max_nlanes];
-    v_pack_u_store(buf, a);
-    for (int i = 0; i < n; i++) {
+    v_pack_store(buf, a);
+    for (int i = 0; i < N; i++) {
        ptr[i] = buf[i];
    }
 }
-template <> inline void v_pack_u_store<8>(uchar* ptr, const v_int16& a, int n)
+template <> inline void v_pack_store<8>(uchar* ptr, const v_uint16& a)
 {
    uchar buf[VTraits<v_uint8>::max_nlanes];
-    v_pack_u_store(buf, a);
+    v_pack_store(buf, a);
    ptr[0] = buf[0]; ptr[1] = buf[1]; ptr[2] = buf[2]; ptr[3] = buf[3];
    ptr[4] = buf[4]; ptr[5] = buf[5]; ptr[6] = buf[6]; ptr[7] = buf[7];
 }
--- a/modules/imgproc/src/warp_common.vector.hpp
+++ b/modules/imgproc/src/warp_common.vector.hpp
@ -569,16 +569,50 @@
    i##ofs##_pix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix1, i##ofs##_pix0), i##ofs##_pix0); \
    i##ofs##_pix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix3, i##ofs##_pix2), i##ofs##_pix2); \
    i##ofs##_pix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_pix2, i##ofs##_pix0), i##ofs##_pix0);
-#define CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4() \
+#define CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_16UC4_I(ofs) \
+    const uint16_t *srcptr##ofs = src + addr[i+ofs]; \
+    v_float32 i##ofs##_pix0 = v_cvt_f32(v_reinterpret_as_s32(vx_load_expand(srcptr##ofs))); \
+    v_float32 i##ofs##_pix1 = v_cvt_f32(v_reinterpret_as_s32(vx_load_expand(srcptr##ofs+4))); \
+    v_float32 i##ofs##_pix2 = v_cvt_f32(v_reinterpret_as_s32(vx_load_expand(srcptr##ofs+srcstep))); \
+    v_float32 i##ofs##_pix3 = v_cvt_f32(v_reinterpret_as_s32(vx_load_expand(srcptr##ofs+srcstep+4))); \
+    v_float32 i##ofs##_alpha = vx_setall_f32(valpha[i+ofs]), \
+              i##ofs##_beta  = vx_setall_f32(vbeta[i+ofs]);  \
+    i##ofs##_pix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix1, i##ofs##_pix0), i##ofs##_pix0); \
+    i##ofs##_pix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix3, i##ofs##_pix2), i##ofs##_pix2); \
+    i##ofs##_pix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_pix2, i##ofs##_pix0), i##ofs##_pix0);
+#define CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_32FC4_I(ofs) \
+    const float *srcptr##ofs = src + addr[i+ofs]; \
+    v_float32 i##ofs##_pix0 = vx_load(srcptr##ofs); \
+    v_float32 i##ofs##_pix1 = vx_load(srcptr##ofs+4); \
+    v_float32 i##ofs##_pix2 = vx_load(srcptr##ofs+srcstep); \
+    v_float32 i##ofs##_pix3 = vx_load(srcptr##ofs+srcstep+4); \
+    v_float32 i##ofs##_alpha = vx_setall_f32(valpha[i+ofs]), \
+              i##ofs##_beta  = vx_setall_f32(vbeta[i+ofs]);  \
+    i##ofs##_pix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix1, i##ofs##_pix0), i##ofs##_pix0); \
+    i##ofs##_pix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_pix3, i##ofs##_pix2), i##ofs##_pix2); \
+    i##ofs##_pix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_pix2, i##ofs##_pix0), i##ofs##_pix0);
+#define CV_WARP_SIMD128_STORE_8UC4_I() \
+    v_uint16 i01_pix = v_pack_u(v_round(i0_pix0), v_round(i1_pix0)); \
+    v_uint16 i23_pix = v_pack_u(v_round(i2_pix0), v_round(i3_pix0)); \
+    v_pack_store(dstptr + 4*(x+i),   i01_pix); \
+    v_pack_store(dstptr + 4*(x+i+2), i23_pix);
+#define CV_WARP_SIMD128_STORE_16UC4_I() \
+    v_uint16 i01_pix = v_pack_u(v_round(i0_pix0), v_round(i1_pix0)); \
+    v_uint16 i23_pix = v_pack_u(v_round(i2_pix0), v_round(i3_pix0)); \
+    vx_store(dstptr + 4*(x+i),   i01_pix); \
+    vx_store(dstptr + 4*(x+i+2), i23_pix);
+#define CV_WARP_SIMD128_STORE_32FC4_I() \
+    vx_store(dstptr + 4*(x+i),    i0_pix0); \
+    vx_store(dstptr + 4*(x+i)+4,  i1_pix0); \
+    vx_store(dstptr + 4*(x+i)+8,  i2_pix0); \
+    vx_store(dstptr + 4*(x+i)+12, i3_pix0);
+#define CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(DEPTH) \
    for (int i = 0; i < uf; i+=vlanes_32) { \
-        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4_I(0); \
-        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4_I(1); \
-        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4_I(2); \
-        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4_I(3); \
-        auto i01_pix = v_pack_u(v_round(i0_pix0), v_round(i1_pix0)), \
-             i23_pix = v_pack_u(v_round(i2_pix0), v_round(i3_pix0)); \
-        v_pack_store(dstptr + 4*(x+i), i01_pix); \
-        v_pack_store(dstptr + 4*(x+i+2), i23_pix); \
+        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(0); \
+        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(1); \
+        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(2); \
+        CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(3); \
+        CV_WARP_SIMD128_STORE_##DEPTH##C4_I(); \
    }
 #define CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4_I(ofs0, ofs1) \
    const uint8_t *srcptr##ofs0 = src + addr[i+ofs0]; \
@ -602,16 +636,70 @@
    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix11, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00); \
    i##ofs0##ofs1##_fpix22 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix33, i##ofs0##ofs1##_fpix22), i##ofs0##ofs1##_fpix22); \
    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_beta,  v_sub(i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00);
-#define CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4() \
+#define CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_16UC4_I(ofs0, ofs1) \
+    const uint16_t *srcptr##ofs0 = src + addr[i+ofs0]; \
+    const uint16_t *srcptr##ofs1 = src + addr[i+ofs1]; \
+    v_int32 i##ofs0##_pix01 = v_reinterpret_as_s32(v256_load_expand(srcptr##ofs0)), \
+            i##ofs0##_pix23 = v_reinterpret_as_s32(v256_load_expand(srcptr##ofs0+srcstep)); \
+    v_int32 i##ofs1##_pix01 = v_reinterpret_as_s32(v256_load_expand(srcptr##ofs1)), \
+            i##ofs1##_pix23 = v_reinterpret_as_s32(v256_load_expand(srcptr##ofs1+srcstep)); \
+    v_float32 i##ofs0##_fpix01 = v_cvt_f32(i##ofs0##_pix01), i##ofs0##_fpix23 = v_cvt_f32(i##ofs0##_pix23); \
+    v_float32 i##ofs1##_fpix01 = v_cvt_f32(i##ofs1##_pix01), i##ofs1##_fpix23 = v_cvt_f32(i##ofs1##_pix23); \
+    v_float32 i##ofs0##ofs1##_fpix00, i##ofs0##ofs1##_fpix11, \
+              i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix33; \
+    v_recombine(i##ofs0##_fpix01, i##ofs1##_fpix01, i##ofs0##ofs1##_fpix00, i##ofs0##ofs1##_fpix11); \
+    v_recombine(i##ofs0##_fpix23, i##ofs1##_fpix23, i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix33); \
+    v_float32 i##ofs0##_alpha = vx_setall_f32(valpha[i+ofs0]), \
+              i##ofs1##_alpha = vx_setall_f32(valpha[i+ofs1]), \
+              i##ofs0##_beta  = vx_setall_f32(vbeta[i+ofs0]), \
+              i##ofs1##_beta  = vx_setall_f32(vbeta[i+ofs1]); \
+    v_float32 i##ofs0##ofs1##_alpha = v_combine_low(i##ofs0##_alpha, i##ofs1##_alpha), \
+              i##ofs0##ofs1##_beta  = v_combine_low(i##ofs0##_beta,  i##ofs1##_beta); \
+    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix11, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00); \
+    i##ofs0##ofs1##_fpix22 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix33, i##ofs0##ofs1##_fpix22), i##ofs0##ofs1##_fpix22); \
+    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_beta,  v_sub(i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00);
+#define CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_32FC4_I(ofs0, ofs1) \
+    const float *srcptr##ofs0 = src + addr[i+ofs0]; \
+    const float *srcptr##ofs1 = src + addr[i+ofs1]; \
+    v_float32 i##ofs0##_fpix01 = v256_load(srcptr##ofs0), \
+              i##ofs0##_fpix23 = v256_load(srcptr##ofs0+srcstep); \
+    v_float32 i##ofs1##_fpix01 = v256_load(srcptr##ofs1), \
+              i##ofs1##_fpix23 = v256_load(srcptr##ofs1+srcstep); \
+    v_float32 i##ofs0##ofs1##_fpix00, i##ofs0##ofs1##_fpix11, \
+              i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix33; \
+    v_recombine(i##ofs0##_fpix01, i##ofs1##_fpix01, i##ofs0##ofs1##_fpix00, i##ofs0##ofs1##_fpix11); \
+    v_recombine(i##ofs0##_fpix23, i##ofs1##_fpix23, i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix33); \
+    v_float32 i##ofs0##_alpha = vx_setall_f32(valpha[i+ofs0]), \
+              i##ofs1##_alpha = vx_setall_f32(valpha[i+ofs1]), \
+              i##ofs0##_beta  = vx_setall_f32(vbeta[i+ofs0]), \
+              i##ofs1##_beta  = vx_setall_f32(vbeta[i+ofs1]); \
+    v_float32 i##ofs0##ofs1##_alpha = v_combine_low(i##ofs0##_alpha, i##ofs1##_alpha), \
+              i##ofs0##ofs1##_beta  = v_combine_low(i##ofs0##_beta,  i##ofs1##_beta); \
+    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix11, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00); \
+    i##ofs0##ofs1##_fpix22 = v_fma(i##ofs0##ofs1##_alpha, v_sub(i##ofs0##ofs1##_fpix33, i##ofs0##ofs1##_fpix22), i##ofs0##ofs1##_fpix22); \
+    i##ofs0##ofs1##_fpix00 = v_fma(i##ofs0##ofs1##_beta,  v_sub(i##ofs0##ofs1##_fpix22, i##ofs0##ofs1##_fpix00), i##ofs0##ofs1##_fpix00);
+#define CV_WARP_SIMD256_STORE_8UC4_I() \
+    auto i01_pix = v_round(i01_fpix00), i23_pix = v_round(i23_fpix00); \
+    v_pack_store(dstptr + 4*(x+i), v_pack_u(i01_pix, i23_pix)); \
+    auto i45_pix = v_round(i45_fpix00), i67_pix = v_round(i67_fpix00); \
+    v_pack_store(dstptr + 4*(x+i+4), v_pack_u(i45_pix, i67_pix));
+#define CV_WARP_SIMD256_STORE_16UC4_I() \
+    auto i01_pix = v_round(i01_fpix00), i23_pix = v_round(i23_fpix00); \
+    vx_store(dstptr + 4*(x+i), v_pack_u(i01_pix, i23_pix)); \
+    auto i45_pix = v_round(i45_fpix00), i67_pix = v_round(i67_fpix00); \
+    vx_store(dstptr + 4*(x+i+4), v_pack_u(i45_pix, i67_pix));
+#define CV_WARP_SIMD256_STORE_32FC4_I() \
+    vx_store(dstptr + 4*(x+i),    i01_fpix00); \
+    vx_store(dstptr + 4*(x+i)+8,  i23_fpix00); \
+    vx_store(dstptr + 4*(x+i)+16, i45_fpix00); \
+    vx_store(dstptr + 4*(x+i)+24, i67_fpix00);
+#define CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(DEPTH) \
    for (int i = 0; i < uf; i+=vlanes_32) { \
-        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4_I(0, 1); \
-        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4_I(2, 3); \
-        auto i01_pix = v_round(i01_fpix00), i23_pix = v_round(i23_fpix00); \
-        v_pack_store(dstptr + 4*(x+i), v_pack_u(i01_pix, i23_pix)); \
-        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4_I(4, 5); \
-        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4_I(6, 7); \
-        auto i45_pix = v_round(i45_fpix00), i67_pix = v_round(i67_fpix00); \
-        v_pack_store(dstptr + 4*(x+i+4), v_pack_u(i45_pix, i67_pix)); \
+        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(0, 1); \
+        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(2, 3); \
+        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(4, 5); \
+        CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(6, 7); \
+        CV_WARP_SIMD256_STORE_##DEPTH##C4_I(); \
    }
 #define CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4_I(ofs) \
    const uint8_t *srcptr##ofs = src + addr[i+ofs]; \
@ -624,14 +712,48 @@
    i##ofs##_fpix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix1, i##ofs##_fpix0), i##ofs##_fpix0); \
    i##ofs##_fpix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix3, i##ofs##_fpix2), i##ofs##_fpix2); \
    i##ofs##_fpix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_fpix2, i##ofs##_fpix0), i##ofs##_fpix0);
-#define CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4() \
+#define CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_16UC4_I(ofs) \
+    const uint16_t *srcptr##ofs = src + addr[i+ofs]; \
+    v_float32 i##ofs##_fpix0 = v_cvt_f32(v_reinterpret_as_s32(v_load_expand<4>(srcptr##ofs))), \
+              i##ofs##_fpix1 = v_cvt_f32(v_reinterpret_as_s32(v_load_expand<4>(srcptr##ofs+4))), \
+              i##ofs##_fpix2 = v_cvt_f32(v_reinterpret_as_s32(v_load_expand<4>(srcptr##ofs+srcstep))), \
+              i##ofs##_fpix3 = v_cvt_f32(v_reinterpret_as_s32(v_load_expand<4>(srcptr##ofs+srcstep+4))); \
+    v_float32 i##ofs##_alpha = vx_setall_f32(valpha[i+ofs]), \
+              i##ofs##_beta  = vx_setall_f32(vbeta[i+ofs]); \
+    i##ofs##_fpix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix1, i##ofs##_fpix0), i##ofs##_fpix0); \
+    i##ofs##_fpix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix3, i##ofs##_fpix2), i##ofs##_fpix2); \
+    i##ofs##_fpix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_fpix2, i##ofs##_fpix0), i##ofs##_fpix0);
+#define CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_32FC4_I(ofs) \
+    const float *srcptr##ofs = src + addr[i+ofs]; \
+    v_float32 i##ofs##_fpix0 = v_load<4>(srcptr##ofs), \
+              i##ofs##_fpix1 = v_load<4>(srcptr##ofs+4), \
+              i##ofs##_fpix2 = v_load<4>(srcptr##ofs+srcstep), \
+              i##ofs##_fpix3 = v_load<4>(srcptr##ofs+srcstep+4); \
+    v_float32 i##ofs##_alpha = vx_setall_f32(valpha[i+ofs]), \
+              i##ofs##_beta  = vx_setall_f32(vbeta[i+ofs]); \
+    i##ofs##_fpix0 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix1, i##ofs##_fpix0), i##ofs##_fpix0); \
+    i##ofs##_fpix2 = v_fma(i##ofs##_alpha, v_sub(i##ofs##_fpix3, i##ofs##_fpix2), i##ofs##_fpix2); \
+    i##ofs##_fpix0 = v_fma(i##ofs##_beta,  v_sub(i##ofs##_fpix2, i##ofs##_fpix0), i##ofs##_fpix0);
+#define CV_WARP_SIMDX_STORE_8UC4_I() \
+    auto i01_pix = v_pack_u<4>(v_round(i0_fpix0), v_round(i1_fpix0)), \
+         i23_pix = v_pack_u<4>(v_round(i2_fpix0), v_round(i3_fpix0)); \
+    v_pack_store<8>(dstptr + 4*(x+i),   i01_pix); \
+    v_pack_store<8>(dstptr + 4*(x+i+2), i23_pix);
+#define CV_WARP_SIMDX_STORE_16UC4_I() \
+    auto i01_pix = v_pack_u<4>(v_round(i0_fpix0), v_round(i1_fpix0)), \
+         i23_pix = v_pack_u<4>(v_round(i2_fpix0), v_round(i3_fpix0)); \
+    v_store<8>(dstptr + 4*(x+i),   i01_pix); \
+    v_store<8>(dstptr + 4*(x+i+2), i23_pix);
+#define CV_WARP_SIMDX_STORE_32FC4_I() \
+    v_store<4>(dstptr + 4*(x+i),    i0_fpix0); \
+    v_store<4>(dstptr + 4*(x+i)+4,  i1_fpix0); \
+    v_store<4>(dstptr + 4*(x+i)+8,  i2_fpix0); \
+    v_store<4>(dstptr + 4*(x+i)+12, i3_fpix0);
+#define CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(DEPTH) \
    for (int i = 0; i < uf; i+=4) { \
-        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4_I(0); \
-        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4_I(1); \
-        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4_I(2); \
-        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4_I(3); \
-        auto i01_pix = v_pack(v_round(i0_fpix0), v_round(i1_fpix0)), \
-             i23_pix = v_pack(v_round(i2_fpix0), v_round(i3_fpix0)); \
-        v_pack_u_store<8>(dstptr + 4*(x+i), i01_pix); \
-        v_pack_u_store<8>(dstptr + 4*(x+i+2), i23_pix); \
+        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(0); \
+        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(1); \
+        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(2); \
+        CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_##DEPTH##C4_I(3); \
+        CV_WARP_SIMDX_STORE_##DEPTH##C4_I(); \
    }
--- a/modules/imgproc/src/warp_kernels.simd.hpp
+++ b/modules/imgproc/src/warp_kernels.simd.hpp
@ -298,7 +298,7 @@ void warpAffineLinearInvoker_8UC1(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -418,7 +418,7 @@ void warpAffineLinearInvoker_8UC3(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -547,7 +547,7 @@ void warpAffineLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -590,11 +590,11 @@ void warpAffineLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step, int
                    vx_store(vbeta, src_y0);
                    vx_store(vbeta+vlanes_32, src_y1);
    #if CV_SIMD256
-                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD128
-                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD_SCALABLE
-                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(8U);
    #endif
                } else {
                    uint8_t pixbuf[max_uf*4*4];
@ -660,7 +660,7 @@ void warpAffineLinearInvoker_16UC1(const uint16_t *src_data, size_t src_step, in
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -761,7 +761,7 @@ void warpAffineLinearInvoker_16UC3(const uint16_t *src_data, size_t src_step, in
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -866,7 +866,7 @@ void warpAffineLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, in
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -876,7 +876,6 @@ void warpAffineLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, in
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        uint16_t pixbuf[max_uf*4*4];

        uint16_t bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -904,18 +903,26 @@ void warpAffineLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, in
                CV_WARPAFFINE_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 16U);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(16U);
+    #endif
                } else {
+                    uint16_t pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 16U);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -972,7 +979,7 @@ void warpAffineLinearInvoker_32FC1(const float *src_data, size_t src_step, int s
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1071,7 +1078,7 @@ void warpAffineLinearInvoker_32FC3(const float *src_data, size_t src_step, int s
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1176,7 +1183,7 @@ void warpAffineLinearInvoker_32FC4(const float *src_data, size_t src_step, int s
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1186,7 +1193,6 @@ void warpAffineLinearInvoker_32FC4(const float *src_data, size_t src_step, int s
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        float pixbuf[max_uf*4*4];

        float bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -1218,16 +1224,25 @@ void warpAffineLinearInvoker_32FC4(const float *src_data, size_t src_step, int s
                CV_WARPAFFINE_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 32F);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(32F);
+    #endif
                } else {
+                    float pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 32F);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -1284,7 +1299,7 @@ void warpAffineLinearApproxInvoker_8UC1(const uint8_t *src_data, size_t src_step
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1391,7 +1406,7 @@ void warpAffineLinearApproxInvoker_8UC3(const uint8_t *src_data, size_t src_step
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1505,7 +1520,7 @@ void warpAffineLinearApproxInvoker_8UC4(const uint8_t *src_data, size_t src_step
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1622,7 +1637,7 @@ void warpPerspectiveLinearInvoker_8UC1(const uint8_t *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1744,7 +1759,7 @@ void warpPerspectiveLinearInvoker_8UC3(const uint8_t *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1874,7 +1889,7 @@ void warpPerspectiveLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -1917,11 +1932,11 @@ void warpPerspectiveLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step,
                    vx_store(vbeta, src_y0);
                    vx_store(vbeta+vlanes_32, src_y1);
    #if CV_SIMD256
-                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD128
-                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD_SCALABLE
-                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(8U);
    #endif
                } else {
                    uint8_t pixbuf[max_uf*4*4];
@ -1988,7 +2003,7 @@ void warpPerspectiveLinearInvoker_16UC1(const uint16_t *src_data, size_t src_ste
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2089,7 +2104,7 @@ void warpPerspectiveLinearInvoker_16UC3(const uint16_t *src_data, size_t src_ste
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2194,7 +2209,7 @@ void warpPerspectiveLinearInvoker_16UC4(const uint16_t *src_data, size_t src_ste
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2204,7 +2219,6 @@ void warpPerspectiveLinearInvoker_16UC4(const uint16_t *src_data, size_t src_ste
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        uint16_t pixbuf[max_uf*4*4];

        uint16_t bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -2232,18 +2246,26 @@ void warpPerspectiveLinearInvoker_16UC4(const uint16_t *src_data, size_t src_ste
                CV_WARPPERSPECTIVE_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 16U);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(16U);
+    #endif
                } else {
+                    uint16_t pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 16U);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -2301,7 +2323,7 @@ void warpPerspectiveLinearInvoker_32FC1(const float *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2401,7 +2423,7 @@ void warpPerspectiveLinearInvoker_32FC3(const float *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2507,7 +2529,7 @@ void warpPerspectiveLinearInvoker_32FC4(const float *src_data, size_t src_step,
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2517,7 +2539,6 @@ void warpPerspectiveLinearInvoker_32FC4(const float *src_data, size_t src_step,
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        float pixbuf[max_uf*4*4];

        float bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -2549,16 +2570,25 @@ void warpPerspectiveLinearInvoker_32FC4(const float *src_data, size_t src_step,
                CV_WARPPERSPECTIVE_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 32F);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(32F);
+    #endif
                } else {
+                    float pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 32F);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -2616,7 +2646,7 @@ void warpPerspectiveLinearApproxInvoker_8UC1(const uint8_t *src_data, size_t src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2724,7 +2754,7 @@ void warpPerspectiveLinearApproxInvoker_8UC3(const uint8_t *src_data, size_t src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2838,7 +2868,7 @@ void warpPerspectiveLinearApproxInvoker_8UC4(const uint8_t *src_data, size_t src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -2961,7 +2991,7 @@ void remapLinearInvoker_8UC1(const uint8_t *src_data, size_t src_step, int src_r
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3100,7 +3130,7 @@ void remapLinearInvoker_8UC3(const uint8_t *src_data, size_t src_step, int src_r
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3247,7 +3277,7 @@ void remapLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step, int src_r
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3292,11 +3322,11 @@ void remapLinearInvoker_8UC4(const uint8_t *src_data, size_t src_step, int src_r
                    vx_store(vbeta, src_y0);
                    vx_store(vbeta+vlanes_32, src_y1);
    #if CV_SIMD256
-                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD128
-                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(8U);
    #elif CV_SIMD_SCALABLE
-                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_8UC4();
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(8U);
    #endif
                } else {
                    uint8_t pixbuf[max_uf*4*4];
@ -3378,7 +3408,7 @@ void remapLinearInvoker_16UC1(const uint16_t *src_data, size_t src_step, int src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3496,7 +3526,7 @@ void remapLinearInvoker_16UC3(const uint16_t *src_data, size_t src_step, int src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3618,7 +3648,7 @@ void remapLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, int src
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3628,7 +3658,6 @@ void remapLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, int src
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        uint16_t pixbuf[max_uf*4*4];

        uint16_t bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -3658,18 +3687,26 @@ void remapLinearInvoker_16UC4(const uint16_t *src_data, size_t src_step, int src
                CV_REMAP_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 16U);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(16U);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(16U);
+    #endif
                } else {
+                    uint16_t pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 16U);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_U16(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CONVERT_U16F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32U16(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -3742,7 +3779,7 @@ void remapLinearInvoker_32FC1(const float *src_data, size_t src_step, int src_ro
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3859,7 +3896,7 @@ void remapLinearInvoker_32FC3(const float *src_data, size_t src_step, int src_ro
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3982,7 +4019,7 @@ void remapLinearInvoker_32FC4(const float *src_data, size_t src_step, int src_ro
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -3992,7 +4029,6 @@ void remapLinearInvoker_32FC4(const float *src_data, size_t src_step, int src_ro
        int32_t addr[max_uf],
                src_ix[max_uf],
                src_iy[max_uf];
-        float pixbuf[max_uf*4*4];

        float bvalbuf[max_uf*4];
        for (int i = 0; i < uf; i++) {
@ -4026,16 +4062,25 @@ void remapLinearInvoker_32FC4(const float *src_data, size_t src_step, int src_ro
                CV_REMAP_LINEAR_VECTOR_COMPUTE_MAPPED_COORD2(C4);

                if (v_reduce_min(inner_mask) != 0) { // all loaded pixels are completely inside the image
-                    CV_WARP_LINEAR_VECTOR_SHUFFLE_ALLWITHIN(C4, 32F);
+                    float valpha[max_uf], vbeta[max_uf];
+                    vx_store(valpha, src_x0);
+                    vx_store(valpha+vlanes_32, src_x1);
+                    vx_store(vbeta, src_y0);
+                    vx_store(vbeta+vlanes_32, src_y1);
+    #if CV_SIMD256
+                    CV_WARP_SIMD256_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD128
+                    CV_WARP_SIMD128_LOAD_SHUFFLE_INTER_C4(32F);
+    #elif CV_SIMD_SCALABLE
+                    CV_WARP_SIMDX_LOAD_SHUFFLE_INTER_C4(32F);
+    #endif
                } else {
+                    float pixbuf[max_uf*4*4];
                    CV_WARP_LINEAR_VECTOR_SHUFFLE_NOTALLWITHIN(C4, 32F);
+                    CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
+                    CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
                }
-
-                CV_WARP_LINEAR_VECTOR_INTER_LOAD_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_CALC_F32(C4);
-
-                CV_WARP_LINEAR_VECTOR_INTER_STORE_F32F32(C4);
            }
 #endif // (CV_SIMD || CV_SIMD_SCALABLE)

@ -4107,7 +4152,7 @@ void remapLinearApproxInvoker_8UC1(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -4229,7 +4274,7 @@ void remapLinearApproxInvoker_8UC3(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);
@ -4359,7 +4404,7 @@ void remapLinearApproxInvoker_8UC4(const uint8_t *src_data, size_t src_step, int
        std::array<float, max_vlanes_32> start_indices;
        std::iota(start_indices.data(), start_indices.data() + max_vlanes_32, 0.f);

-        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 2),
+        v_uint32 inner_srows = vx_setall_u32((unsigned)srcrows - 1),
                 inner_scols = vx_setall_u32((unsigned)srccols - 1),
                 outer_srows = vx_setall_u32((unsigned)srcrows + 1),
                 outer_scols = vx_setall_u32((unsigned)srccols + 1);