mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 13:47:32 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #14649 from savuor:fix/luv_hls_read_oob

Browse Source
This commit is contained in:
Alexander Alekhin 2019-05-27 16:24:55 +00:00
commit aaf56c2839
2 changed files with 38 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
modules/imgproc/src/color_hsv.simd.hpp
View File

@ -724,7 +724,7 @@ struct RGB2HLS_b
{ {
CV_INSTRUMENT_REGION(); CV_INSTRUMENT_REGION();
int i, j, scn = srccn; int scn = srccn;
#if CV_SIMD #if CV_SIMD
float CV_DECL_ALIGNED(CV_SIMD_WIDTH) buf[bufChannels*BLOCK_SIZE]; float CV_DECL_ALIGNED(CV_SIMD_WIDTH) buf[bufChannels*BLOCK_SIZE];
@ -744,15 +744,15 @@ struct RGB2HLS_b
} }
#endif #endif
for( i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*3 ) for(int i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*3 )
{ {
int dn = std::min(n - i, (int)BLOCK_SIZE); int dn = std::min(n - i, (int)BLOCK_SIZE);
j = 0;
#if CV_SIMD #if CV_SIMD
v_float32 v255inv = vx_setall_f32(1.f/255.f); v_float32 v255inv = vx_setall_f32(1.f/255.f);
if (scn == 3) if (scn == 3)
{ {
int j = 0;
static const int nBlock = fsize*2; static const int nBlock = fsize*2;
for ( ; j <= (dn * bufChannels - nBlock); for ( ; j <= (dn * bufChannels - nBlock);
j += nBlock, src += nBlock) j += nBlock, src += nBlock)
@ -763,9 +763,14 @@ struct RGB2HLS_b
v_store_aligned(buf + j + 0*fsize, v_cvt_f32(qrgb0)*v255inv); v_store_aligned(buf + j + 0*fsize, v_cvt_f32(qrgb0)*v255inv);
v_store_aligned(buf + j + 1*fsize, v_cvt_f32(qrgb1)*v255inv); v_store_aligned(buf + j + 1*fsize, v_cvt_f32(qrgb1)*v255inv);
} }
for( ; j < dn*3; j++, src++ )
{
buf[j] = src[0]*(1.f/255.f);
}
} }
else // if (scn == 4) else // if (scn == 4)
{ {
int j = 0;
static const int nBlock = fsize*4; static const int nBlock = fsize*4;
for ( ; j <= dn*bufChannels - nBlock*bufChannels; for ( ; j <= dn*bufChannels - nBlock*bufChannels;
j += nBlock*bufChannels, src += nBlock*4) j += nBlock*bufChannels, src += nBlock*4)
@ -795,17 +800,24 @@ struct RGB2HLS_b
v_store_interleave(buf + j + k*bufChannels*fsize, f[0*4+k], f[1*4+k], f[2*4+k]); v_store_interleave(buf + j + k*bufChannels*fsize, f[0*4+k], f[1*4+k], f[2*4+k]);
} }
} }
} for( ; j < dn*3; j += 3, src += 4 )
#endif
for( ; j < dn*3; j += 3, src += scn )
{ {
buf[j+0] = src[0]*(1.f/255.f); buf[j+0] = src[0]*(1.f/255.f);
buf[j+1] = src[1]*(1.f/255.f); buf[j+1] = src[1]*(1.f/255.f);
buf[j+2] = src[2]*(1.f/255.f); buf[j+2] = src[2]*(1.f/255.f);
} }
}
#else
for(int j = 0; j < dn*3; j += 3, src += scn )
{
buf[j+0] = src[0]*(1.f/255.f);
buf[j+1] = src[1]*(1.f/255.f);
buf[j+2] = src[2]*(1.f/255.f);
}
#endif
cvt(buf, buf, dn); cvt(buf, buf, dn);
j = 0; int j = 0;
#if CV_SIMD #if CV_SIMD
for( ; j <= dn*3 - fsize*3*4; j += fsize*3*4) for( ; j <= dn*3 - fsize*3*4; j += fsize*3*4)
{ {

27
modules/imgproc/src/color_lab.cpp
View File

@ -3266,7 +3266,7 @@ struct RGB2Luv_b
return; return;
} }
int i, j, scn = srccn; int scn = srccn;
#if CV_SIMD #if CV_SIMD
float CV_DECL_ALIGNED(CV_SIMD_WIDTH) buf[bufChannels*BLOCK_SIZE]; float CV_DECL_ALIGNED(CV_SIMD_WIDTH) buf[bufChannels*BLOCK_SIZE];
#else #else
@ -3295,16 +3295,16 @@ struct RGB2Luv_b
} }
#endif #endif
for( i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*bufChannels ) for(int i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*bufChannels )
{ {
int dn = std::min(n - i, (int)BLOCK_SIZE); int dn = std::min(n - i, (int)BLOCK_SIZE);
j = 0;
static const softfloat f255inv = softfloat::one()/f255; static const softfloat f255inv = softfloat::one()/f255;
#if CV_SIMD #if CV_SIMD
v_float32 v255inv = vx_setall_f32((float)f255inv); v_float32 v255inv = vx_setall_f32((float)f255inv);
if(scn == 4) if(scn == 4)
{ {
int j = 0;
static const int nBlock = fsize*4; static const int nBlock = fsize*4;
for( ; j <= dn*bufChannels - nBlock*3; for( ; j <= dn*bufChannels - nBlock*3;
j += nBlock*3, src += nBlock*4) j += nBlock*3, src += nBlock*4)
@ -3334,9 +3334,16 @@ struct RGB2Luv_b
v_store_interleave(buf + j + k*3*fsize, f[0*4+k], f[1*4+k], f[2*4+k]); v_store_interleave(buf + j + k*3*fsize, f[0*4+k], f[1*4+k], f[2*4+k]);
} }
} }
for( ; j < dn*bufChannels; j += bufChannels, src += 4 )
{
buf[j ] = (float)(src[0]*((float)f255inv));
buf[j+1] = (float)(src[1]*((float)f255inv));
buf[j+2] = (float)(src[2]*((float)f255inv));
}
} }
else // scn == 3 else // scn == 3
{ {
int j = 0;
static const int nBlock = fsize*2; static const int nBlock = fsize*2;
for( ; j <= dn*bufChannels - nBlock; for( ; j <= dn*bufChannels - nBlock;
j += nBlock, src += nBlock) j += nBlock, src += nBlock)
@ -3348,17 +3355,23 @@ struct RGB2Luv_b
v_store_aligned(buf + j + 0*fsize, v_cvt_f32(q0)*v255inv); v_store_aligned(buf + j + 0*fsize, v_cvt_f32(q0)*v255inv);
v_store_aligned(buf + j + 1*fsize, v_cvt_f32(q1)*v255inv); v_store_aligned(buf + j + 1*fsize, v_cvt_f32(q1)*v255inv);
} }
for( ; j < dn*bufChannels; j++, src++ )
{
buf[j] = (float)(src[0]*((float)f255inv));
} }
#endif }
for( ; j < dn*bufChannels; j += bufChannels, src += scn ) #else
for(int j = 0; j < dn*bufChannels; j += bufChannels, src += scn )
{ {
buf[j ] = (float)(src[0]*((float)f255inv)); buf[j ] = (float)(src[0]*((float)f255inv));
buf[j+1] = (float)(src[1]*((float)f255inv)); buf[j+1] = (float)(src[1]*((float)f255inv));
buf[j+2] = (float)(src[2]*((float)f255inv)); buf[j+2] = (float)(src[2]*((float)f255inv));
} }
#endif
fcvt(buf, buf, dn); fcvt(buf, buf, dn);
j = 0; int j = 0;
#if CV_SIMD #if CV_SIMD
for( ; j <= dn*3 - fsize*3*4; j += fsize*3*4) for( ; j <= dn*3 - fsize*3*4; j += fsize*3*4)
@ -3389,7 +3402,7 @@ struct RGB2Luv_b
#endif #endif
for( ; j < dn*3; j += 3 ) for( ; j < dn*3; j += 3 )
{ {
dst[j] = saturate_cast<uchar>(buf[j]*(float)fL); dst[j+0] = saturate_cast<uchar>(buf[j+0]*(float)fL);
dst[j+1] = saturate_cast<uchar>(buf[j+1]*(float)fu + (float)su); dst[j+1] = saturate_cast<uchar>(buf[j+1]*(float)fu + (float)su);
dst[j+2] = saturate_cast<uchar>(buf[j+2]*(float)fv + (float)sv); dst[j+2] = saturate_cast<uchar>(buf[j+2]*(float)fv + (float)sv);
} }