liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Vectorize 49 and 81 in ImProcFunctions::Median_Denoise() (#3346)

Browse Source
This commit is contained in:
Flössie
2016-06-30 18:18:13 +02:00
parent e2deb16403
commit 1f63ab85fa
1 changed files with 28 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
rtengine/FTblockDN.cc
View File

@@ -376,10 +376,21 @@ void ImProcFunctions::Median_Denoise( float **src, float **dst, const int width,
medianOut[i][j] = medianIn[i][j]; medianOut[i][j] = medianIn[i][j];
} }
for (; j < width - border; ++j) { #ifdef __SSE2__
int kk = 0; std::array<vfloat, 49> vpp ALIGNED16;
for (; j < width - border - 3; j += 4) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
vpp[kk] = LVFU(medianIn[i + ii][j + jj]);
}
}
for (int ii = -border; ii <= border; ++ii) { STVFU(medianOut[i][j], median(vpp));
}
#endif
for (; j < width - border; ++j) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) { for (int jj = -border; jj <= border; ++jj, ++kk) {
pp[kk] = medianIn[i + ii][j + jj]; pp[kk] = medianIn[i + ii][j + jj];
} }
@@ -399,10 +410,21 @@ void ImProcFunctions::Median_Denoise( float **src, float **dst, const int width,
medianOut[i][j] = medianIn[i][j]; medianOut[i][j] = medianIn[i][j];
} }
for (; j < width - border; ++j) { #ifdef __SSE2__
int kk = 0; std::array<vfloat, 81> vpp ALIGNED16;
for (; j < width - border - 3; j += 4) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
vpp[kk] = LVFU(medianIn[i + ii][j + jj]);
}
}
for (int ii = -border; ii <= border; ++ii) { STVFU(medianOut[i][j], median(vpp));
}
#endif
for (; j < width - border; ++j) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) { for (int jj = -border; jj <= border; ++jj, ++kk) {
pp[kk] = medianIn[i + ii][j + jj]; pp[kk] = medianIn[i + ii][j + jj];
} }