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 #29430 from Teddy-Yangjiale:rvv-k1-04-resize-nn

Browse Source 
imgproc: re-enable RVV HAL resize NEAREST/NEAREST_EXACT
This commit is contained in:
Alexander Smorkalov
2026-07-09 18:05:56 +03:00
committed by GitHub
parent aa1b8a2a21 3aadaa8385
commit 4b38c5562b
1 changed files with 56 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hal/riscv-rvv/src/imgproc/resize.cpp
+56 -24
View File
@@ -6,8 +6,27 @@
#include "rvv_hal.hpp"
#include "common.hpp"
#include "opencv2/core/softfloat.hpp"
#include <list>
// Coverage matrix (depth x channels x interpolation). Entries not listed fall back to the
// generic implementation via CV_HAL_ERROR_NOT_IMPLEMENTED.
//
// | 8UC1 8UC2 8UC3 8UC4 | 16UC1 16UC2 16UC3 16UC4 | 32FC1 32FC2 32FC3 32FC4
// NEAREST | Y Y Y Y | Y* Y* - - | Y* - - -
// NEAREST_EXACT | Y Y Y Y | Y* Y* - - | Y* - - -
// LINEAR | Y Y Y Y | Y Y Y Y | Y Y Y Y
// LINEAR_EXACT | Y Y Y Y | - - - - | n/a (core uses LINEAR)
// AREA (2x2 fast) | Y Y Y Y | Y Y Y Y | - - - -
// AREA (generic dn) | Y Y Y Y | - - - - | - - - -
// AREA (upscale) | - (falls back: core maps AREA upscale to LINEAR-like path)
// CUBIC / LANCZOS4 | - (not implemented)
//
// Y* : NEAREST dispatches on element size in bytes (1..4), so 16UC1/16UC2/32FC1/32SC1 etc.
// are covered by the same gather kernels as 8UC2/8UC4.
// Additional limits: width guards require cn*src_width (resp. cn*4*src_width for AREA
// generic) to fit in ushort, since gather indices are 16-bit byte offsets.
namespace cv { namespace rvv_hal { namespace imgproc {
#if CV_HAL_RVV_1P0_ENABLED
@@ -39,16 +58,14 @@ static inline int invoke(int height, std::function<int(int, int, Args...)> func,
return func(0, 1, std::forward<Args>(args)...);
}
// cn is the element size in bytes, not the channel count; x_ofs/y_ofs hold precomputed
// clamped source offsets (x in bytes) so NEAREST and NEAREST_EXACT share the same kernel
template<int cn>
static inline int resizeNN(int start, int end, const uchar *src_data, size_t src_step, int src_height, uchar *dst_data, size_t dst_step, int dst_width, int dst_height, double scale_y, int interpolation, const ushort* x_ofs)
static inline int resizeNN(int start, int end, const uchar *src_data, size_t src_step, uchar *dst_data, size_t dst_step, int dst_width, const int* y_ofs_tab, const ushort* x_ofs)
{
const int ify = ((src_height << 16) + dst_height / 2) / dst_height;
const int ify0 = ify / 2 - src_height % 2;
for (int i = start; i < end; i++)
{
int y_ofs = interpolation == CV_HAL_INTER_NEAREST ? static_cast<int>(std::floor(i * scale_y)) : (ify * i + ify0) >> 16;
y_ofs = std::min(y_ofs, src_height - 1);
int y_ofs = y_ofs_tab[i];
int vl;
switch (cn)
@@ -739,29 +756,44 @@ static inline int resizeArea(int start, int end, const uchar *src_data, size_t s
// in the function static void resizeNN and static void resizeNN_bitexact
static inline int resizeNN(int src_type, const uchar *src_data, size_t src_step, int src_width, int src_height, uchar *dst_data, size_t dst_step, int dst_width, int dst_height, double scale_x, double scale_y, int interpolation)
{
const int cn = CV_ELEM_SIZE(src_type);
if (cn * src_width > std::numeric_limits<ushort>::max())
const int pix_size = CV_ELEM_SIZE(src_type);
if (pix_size * src_width > std::numeric_limits<ushort>::max())
return CV_HAL_ERROR_NOT_IMPLEMENTED;
std::vector<ushort> x_ofs(dst_width);
const int ifx = ((src_width << 16) + dst_width / 2) / dst_width;
const int ifx0 = ifx / 2 - src_width % 2;
for (int i = 0; i < dst_width; i++)
std::vector<int> y_ofs(dst_height);
if (interpolation == CV_HAL_INTER_NEAREST)
{
x_ofs[i] = interpolation == CV_HAL_INTER_NEAREST ? static_cast<ushort>(std::floor(i * scale_x)) : (ifx * i + ifx0) >> 16;
x_ofs[i] = std::min(x_ofs[i], static_cast<ushort>(src_width - 1)) * cn;
for (int i = 0; i < dst_width; i++)
x_ofs[i] = static_cast<ushort>(std::min(cvFloor(i * scale_x), src_width - 1) * pix_size);
for (int i = 0; i < dst_height; i++)
y_ofs[i] = std::min(cvFloor(i * scale_y), src_height - 1);
}
else
{
// NEAREST_EXACT must reproduce resizeNN_bitexact_tab: Pillow-compatible pixel-center
// mapping accumulated in softdouble, otherwise results diverge from the core path
const softdouble fx = softdouble(src_width) / softdouble(dst_width);
softdouble vx = fx * softdouble(0.5);
for (int i = 0; i < dst_width; i++, vx += fx)
x_ofs[i] = static_cast<ushort>(std::min(cvFloor(vx), src_width - 1) * pix_size);
const softdouble fy = softdouble(src_height) / softdouble(dst_height);
softdouble vy = fy * softdouble(0.5);
for (int i = 0; i < dst_height; i++, vy += fy)
y_ofs[i] = std::min(cvFloor(vy), src_height - 1);
}
switch (src_type)
// gather kernels only depend on the element size, so e.g. 16UC1 shares the 8UC2 path
switch (pix_size)
{
case CV_8UC1:
return invoke(dst_height, {resizeNN<1>}, src_data, src_step, src_height, dst_data, dst_step, dst_width, dst_height, scale_y, interpolation, x_ofs.data());
case CV_8UC2:
return invoke(dst_height, {resizeNN<2>}, src_data, src_step, src_height, dst_data, dst_step, dst_width, dst_height, scale_y, interpolation, x_ofs.data());
case CV_8UC3:
return invoke(dst_height, {resizeNN<3>}, src_data, src_step, src_height, dst_data, dst_step, dst_width, dst_height, scale_y, interpolation, x_ofs.data());
case CV_8UC4:
return invoke(dst_height, {resizeNN<4>}, src_data, src_step, src_height, dst_data, dst_step, dst_width, dst_height, scale_y, interpolation, x_ofs.data());
case 1:
return invoke(dst_height, {resizeNN<1>}, src_data, src_step, dst_data, dst_step, dst_width, y_ofs.data(), x_ofs.data());
case 2:
return invoke(dst_height, {resizeNN<2>}, src_data, src_step, dst_data, dst_step, dst_width, y_ofs.data(), x_ofs.data());
case 3:
return invoke(dst_height, {resizeNN<3>}, src_data, src_step, dst_data, dst_step, dst_width, y_ofs.data(), x_ofs.data());
case 4:
return invoke(dst_height, {resizeNN<4>}, src_data, src_step, dst_data, dst_step, dst_width, y_ofs.data(), x_ofs.data());
}
return CV_HAL_ERROR_NOT_IMPLEMENTED;
}
@@ -990,8 +1022,8 @@ int resize(int src_type, const uchar *src_data, size_t src_step, int src_width,
{
inv_scale_x = 1 / inv_scale_x;
inv_scale_y = 1 / inv_scale_y;
//if (interpolation == CV_HAL_INTER_NEAREST || interpolation == CV_HAL_INTER_NEAREST_EXACT)
// return resizeNN(src_type, src_data, src_step, src_width, src_height, dst_data, dst_step, dst_width, dst_height, inv_scale_x, inv_scale_y, interpolation);
if (interpolation == CV_HAL_INTER_NEAREST || interpolation == CV_HAL_INTER_NEAREST_EXACT)
return resizeNN(src_type, src_data, src_step, src_width, src_height, dst_data, dst_step, dst_width, dst_height, inv_scale_x, inv_scale_y, interpolation);
if (interpolation == CV_HAL_INTER_LINEAR || interpolation == CV_HAL_INTER_LINEAR_EXACT)
return resizeLinear(src_type, src_data, src_step, src_width, src_height, dst_data, dst_step, dst_width, dst_height, inv_scale_x, inv_scale_y, interpolation);
if (interpolation == CV_HAL_INTER_AREA)