potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-31 00:03:03 +04:00
Code Packages Releases Wiki Activity

Merge branch 4.x

Browse Source
This commit is contained in:
Alexander Smorkalov
2024-09-10 10:15:22 +03:00
parent 99bc88c259 79faf857d9
commit b574db2cff
26 changed files with 237 additions and 204 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/imgproc/src/color_rgb.simd.hpp
-5
View File
@@ -1088,11 +1088,6 @@ struct mRGBA2RGBA<uchar>
uchar v3_half = v3 / 2;
dst[0] = (v3==0)? 0 : (v0 * max_val + v3_half) / v3;
dst[1] = (v3==0)? 0 : (v1 * max_val + v3_half) / v3;
dst[2] = (v3==0)? 0 : (v2 * max_val + v3_half) / v3;
dst[3] = v3;
dst[0] = (v3==0)? 0 : saturate_cast<uchar>((v0 * max_val + v3_half) / v3);
dst[1] = (v3==0)? 0 : saturate_cast<uchar>((v1 * max_val + v3_half) / v3);
dst[2] = (v3==0)? 0 : saturate_cast<uchar>((v2 * max_val + v3_half) / v3);
modules/imgproc/src/imgwarp.cpp
+33 -52
View File
@@ -1983,65 +1983,46 @@ void cv::convertMaps( InputArray _map1, InputArray _map2,
}
else if( m1type == CV_32FC2 && dstm1type == CV_16SC2 )
{
if( nninterpolate )
#if CV_TRY_SSE4_1
if( useSSE4_1 )
opt_SSE4_1::convertMaps_32f2c16s_SSE41(src1f, dst1, dst2, size.width);
else
#endif
{
#if CV_SIMD128
int span = VTraits<v_float32x4>::vlanes();
{
for( ; x <= (size.width << 1) - span * 2; x += span * 2 )
v_store(dst1 + x, v_pack(v_round(v_load(src1f + x)),
v_round(v_load(src1f + x + span))));
v_float32x4 v_scale = v_setall_f32((float)INTER_TAB_SIZE);
v_int32x4 v_mask = v_setall_s32(INTER_TAB_SIZE - 1);
v_int32x4 v_scale3 = v_setall_s32(INTER_TAB_SIZE);
int span = VTraits<v_uint16x8>::vlanes();
for (; x <= size.width - span; x += span )
{
v_float32x4 v_src0[2], v_src1[2];
v_load_deinterleave(src1f + (x << 1), v_src0[0], v_src0[1]);
v_load_deinterleave(src1f + (x << 1) + span, v_src1[0], v_src1[1]);
v_int32x4 v_ix0 = v_round(v_mul(v_src0[0], v_scale));
v_int32x4 v_ix1 = v_round(v_mul(v_src1[0], v_scale));
v_int32x4 v_iy0 = v_round(v_mul(v_src0[1], v_scale));
v_int32x4 v_iy1 = v_round(v_mul(v_src1[1], v_scale));
v_int16x8 v_dst[2];
v_dst[0] = v_pack(v_shr<INTER_BITS>(v_ix0), v_shr<INTER_BITS>(v_ix1));
v_dst[1] = v_pack(v_shr<INTER_BITS>(v_iy0), v_shr<INTER_BITS>(v_iy1));
v_store_interleave(dst1 + (x << 1), v_dst[0], v_dst[1]);
v_store(dst2 + x, v_pack_u(
v_muladd(v_scale3, (v_and(v_iy0, v_mask)), (v_and(v_ix0, v_mask))),
v_muladd(v_scale3, (v_and(v_iy1, v_mask)), (v_and(v_ix1, v_mask)))));
}
}
#endif
for( ; x < size.width; x++ )
{
dst1[x*2] = saturate_cast<short>(src1f[x*2]);
dst1[x*2+1] = saturate_cast<short>(src1f[x*2+1]);
}
}
else
{
#if CV_TRY_SSE4_1
if( useSSE4_1 )
opt_SSE4_1::convertMaps_32f2c16s_SSE41(src1f, dst1, dst2, size.width);
else
#endif
{
#if CV_SIMD128
{
v_float32x4 v_scale = v_setall_f32((float)INTER_TAB_SIZE);
v_int32x4 v_mask = v_setall_s32(INTER_TAB_SIZE - 1);
v_int32x4 v_scale3 = v_setall_s32(INTER_TAB_SIZE);
int span = VTraits<v_uint16x8>::vlanes();
for (; x <= size.width - span; x += span )
{
v_float32x4 v_src0[2], v_src1[2];
v_load_deinterleave(src1f + (x << 1), v_src0[0], v_src0[1]);
v_load_deinterleave(src1f + (x << 1) + span, v_src1[0], v_src1[1]);
v_int32x4 v_ix0 = v_round(v_mul(v_src0[0], v_scale));
v_int32x4 v_ix1 = v_round(v_mul(v_src1[0], v_scale));
v_int32x4 v_iy0 = v_round(v_mul(v_src0[1], v_scale));
v_int32x4 v_iy1 = v_round(v_mul(v_src1[1], v_scale));
v_int16x8 v_dst[2];
v_dst[0] = v_pack(v_shr<INTER_BITS>(v_ix0), v_shr<INTER_BITS>(v_ix1));
v_dst[1] = v_pack(v_shr<INTER_BITS>(v_iy0), v_shr<INTER_BITS>(v_iy1));
v_store_interleave(dst1 + (x << 1), v_dst[0], v_dst[1]);
v_store(dst2 + x, v_pack_u(
v_muladd(v_scale3, (v_and(v_iy0, v_mask)), (v_and(v_ix0, v_mask))),
v_muladd(v_scale3, (v_and(v_iy1, v_mask)), (v_and(v_ix1, v_mask)))));
}
}
#endif
for( ; x < size.width; x++ )
{
int ix = saturate_cast<int>(src1f[x*2]*INTER_TAB_SIZE);
int iy = saturate_cast<int>(src1f[x*2+1]*INTER_TAB_SIZE);
dst1[x*2] = saturate_cast<short>(ix >> INTER_BITS);
dst1[x*2+1] = saturate_cast<short>(iy >> INTER_BITS);
dst2[x] = (ushort)((iy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (ix & (INTER_TAB_SIZE-1)));
}
int ix = saturate_cast<int>(src1f[x*2]*INTER_TAB_SIZE);
int iy = saturate_cast<int>(src1f[x*2+1]*INTER_TAB_SIZE);
dst1[x*2] = saturate_cast<short>(ix >> INTER_BITS);
dst1[x*2+1] = saturate_cast<short>(iy >> INTER_BITS);
dst2[x] = (ushort)((iy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (ix & (INTER_TAB_SIZE-1)));
}
}
}
modules/imgproc/src/resize.cpp
+56
View File
@@ -94,6 +94,10 @@ static void hlineResize(ET* src, int cn, int *ofst, FT* m, FT* dst, int dst_min,
}
}
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
ET* src_last = src + cn*ofst[dst_width - 1];
for (; i < dst_width; i++) // Points that fall right from src image so became equal to rightmost src point
{
@@ -125,6 +129,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 2, true, 1>
ET* px = src + ofst[i];
*(dst++) = m[0] * px[0] + m[1] * px[1];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + ofst[dst_width - 1])[0];
for (; i < dst_width; i++) // Points that fall right from src image so became equal to rightmost src point
{
@@ -149,6 +157,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 2, true, 2>
*(dst++) = m[0] * px[0] + m[1] * px[2];
*(dst++) = m[0] * px[1] + m[1] * px[3];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 2*ofst[dst_width - 1])[0];
src1 = (src + 2*ofst[dst_width - 1])[1];
for (; i < dst_width; i++) // Points that fall right from src image so became equal to rightmost src point
@@ -177,6 +189,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 2, true, 3>
*(dst++) = m[0] * px[1] + m[1] * px[4];
*(dst++) = m[0] * px[2] + m[1] * px[5];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 3*ofst[dst_width - 1])[0];
src1 = (src + 3*ofst[dst_width - 1])[1];
src2 = (src + 3*ofst[dst_width - 1])[2];
@@ -209,6 +225,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 2, true, 4>
*(dst++) = m[0] * px[2] + m[1] * px[6];
*(dst++) = m[0] * px[3] + m[1] * px[7];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 4*ofst[dst_width - 1])[0];
src1 = (src + 4*ofst[dst_width - 1])[1];
src2 = (src + 4*ofst[dst_width - 1])[2];
@@ -237,6 +257,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 4, true, 1>
ET* px = src + ofst[i];
*(dst++) = m[0] * src[0] + m[1] * src[1] + m[2] * src[2] + m[3] * src[3];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + ofst[dst_width - 1])[0];
for (; i < dst_width; i++) // Points that fall right from src image so became equal to rightmost src point
{
@@ -261,6 +285,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 4, true, 2>
*(dst++) = m[0] * src[0] + m[1] * src[2] + m[2] * src[4] + m[3] * src[6];
*(dst++) = m[0] * src[1] + m[1] * src[3] + m[2] * src[5] + m[3] * src[7];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 2*ofst[dst_width - 1])[0];
src1 = (src + 2*ofst[dst_width - 1])[1];
for (; i < dst_width; i++) // Points that fall right from src image so became equal to rightmost src point
@@ -289,6 +317,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 4, true, 3>
*(dst++) = m[0] * src[1] + m[1] * src[4] + m[2] * src[7] + m[3] * src[10];
*(dst++) = m[0] * src[2] + m[1] * src[5] + m[2] * src[8] + m[3] * src[11];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 3*ofst[dst_width - 1])[0];
src1 = (src + 3*ofst[dst_width - 1])[1];
src2 = (src + 3*ofst[dst_width - 1])[2];
@@ -321,6 +353,10 @@ template <typename ET, typename FT> struct hline<ET, FT, 4, true, 4>
*(dst++) = m[0] * src[2] + m[1] * src[6] + m[2] * src[10] + m[3] * src[14];
*(dst++) = m[0] * src[3] + m[1] * src[7] + m[2] * src[11] + m[3] * src[15];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src0 = (src + 4*ofst[dst_width - 1])[0];
src1 = (src + 4*ofst[dst_width - 1])[1];
src2 = (src + 4*ofst[dst_width - 1])[2];
@@ -382,6 +418,10 @@ void hlineResizeCn<uint8_t, ufixedpoint16, 2, true, 1>(uint8_t* src, int, int *o
uint8_t* px = src + ofst[i];
*(dst++) = m[0] * px[0] + m[1] * px[1];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src_0 = (src + ofst[dst_width - 1])[0];
#if (CV_SIMD || CV_SIMD_SCALABLE)
v_src_0 = vx_setall_u16(*((uint16_t*)&src_0));
@@ -438,6 +478,10 @@ void hlineResizeCn<uint8_t, ufixedpoint16, 2, true, 2>(uint8_t* src, int, int *o
*(dst++) = m[0] * px[0] + m[1] * px[2];
*(dst++) = m[0] * px[1] + m[1] * px[3];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
((ufixedpoint16*)(srccn.w))[0] = (src + 2 * ofst[dst_width - 1])[0]; ((ufixedpoint16*)(srccn.w))[1] = (src + 2 * ofst[dst_width - 1])[1];
#if (CV_SIMD || CV_SIMD_SCALABLE)
v_srccn = v_reinterpret_as_u16(vx_setall_u32(srccn.d));
@@ -510,6 +554,10 @@ void hlineResizeCn<uint8_t, ufixedpoint16, 2, true, 3>(uint8_t* src, int, int *o
*(dst++) = m[0] * px[1] + m[1] * px[4];
*(dst++) = m[0] * px[2] + m[1] * px[5];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
((ufixedpoint16*)(srccn.w))[0] = (src + 3*ofst[dst_width - 1])[0];
((ufixedpoint16*)(srccn.w))[1] = (src + 3*ofst[dst_width - 1])[1];
((ufixedpoint16*)(srccn.w))[2] = (src + 3*ofst[dst_width - 1])[2];
@@ -583,6 +631,10 @@ void hlineResizeCn<uint8_t, ufixedpoint16, 2, true, 4>(uint8_t* src, int, int *o
*(dst++) = m[0] * px[2] + m[1] * px[6];
*(dst++) = m[0] * px[3] + m[1] * px[7];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
((ufixedpoint16*)(srccn.w))[0] = (src + 4 * ofst[dst_width - 1])[0]; ((ufixedpoint16*)(srccn.w))[1] = (src + 4 * ofst[dst_width - 1])[1];
((ufixedpoint16*)(srccn.w))[2] = (src + 4 * ofst[dst_width - 1])[2]; ((ufixedpoint16*)(srccn.w))[3] = (src + 4 * ofst[dst_width - 1])[3];
#if (CV_SIMD || CV_SIMD_SCALABLE)
@@ -634,6 +686,10 @@ void hlineResizeCn<uint16_t, ufixedpoint32, 2, true, 1>(uint16_t* src, int, int
uint16_t* px = src + ofst[i];
*(dst++) = m[0] * px[0] + m[1] * px[1];
}
// Avoid reading a potentially unset ofst, leading to a random memory read.
if (i >= dst_width) {
return;
}
src_0 = (src + ofst[dst_width - 1])[0];
#if (CV_SIMD || CV_SIMD_SCALABLE)
v_src_0 = vx_setall_u32(*((uint32_t*)&src_0));