Mirrors/ebsynth
1
0
Fork 0
mirror of https://github.com/jamriska/ebsynth.git synced 2025-12-16 16:37:50 +01:00
Code Issues Packages Projects Releases Wiki Activity

use GCD instead of OpenMP on Mac

Browse Source
This commit is contained in:
Ondrej Jamriska
2018-09-14 16:29:56 +02:00
parent f5b08897b3
commit 66db8eddeb
1 changed files with 32 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
src/ebsynth_cpu.cpp
View File

@@ -4,12 +4,17 @@
#include "ebsynth.h" #include "ebsynth.h"
#include "jzq.h" #include "jzq.h"
#include "omp.h"
#include <cmath> #include <cmath>
#include <cfloat> #include <cfloat>
#include <cstring> #include <cstring>
#ifdef __APPLE__
#include <dispatch/dispatch.h>
#else
#include <omp.h>
#endif
#define FOR(A,X,Y) for(int Y=0;Y<A.height();Y++) for(int X=0;X<A.width();X++) #define FOR(A,X,Y) for(int Y=0;Y<A.height();Y++) for(int X=0;X<A.width();X++)
A2V2i nnfInit(const V2i& sizeA, A2V2i nnfInit(const V2i& sizeA,
@@ -531,7 +536,8 @@ void patchmatch(const V2i& sizeA,
const int numIters, const int numIters,
const int numThreads, const int numThreads,
A2V2i& N, A2V2i& N,
A2f& E) A2f& E,
A2i& Omega)
{ {
const int w = patchWidth; const int w = patchWidth;
@@ -547,14 +553,20 @@ void patchmatch(const V2i& sizeA,
const int nir = int(irad.size()); const int nir = int(irad.size());
#ifdef __APPLE__
dispatch_queue_t gcdq = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH,0);
const int numThreads_ = 8;
#else
const int numThreads_ = numThreads<1 ? omp_get_max_threads() : numThreads; const int numThreads_ = numThreads<1 ? omp_get_max_threads() : numThreads;
#endif
const int minTileHeight = 8; const int minTileHeight = 8;
const int numTiles = int(ceil(float(sizeA(1))/float(numThreads_))) > minTileHeight ? numThreads_ : std::max(int(ceil(float(sizeA(1))/float(minTileHeight))),1); const int numTiles = int(ceil(float(sizeA(1))/float(numThreads_))) > minTileHeight ? numThreads_ : std::max(int(ceil(float(sizeA(1))/float(minTileHeight))),1);
const int tileHeight = sizeA(1)/numTiles; const int tileHeight = sizeA(1)/numTiles;
const float omegaBest = (float(sizeA(0)*sizeA(1)) / const float omegaBest = (float(sizeA(0)*sizeA(1)) /
float(sizeB(0)*sizeB(1))) * float(patchWidth*patchWidth); float(sizeB(0)*sizeB(1))) * float(patchWidth*patchWidth);
A2i Omega(sizeB);
fill(&Omega,(int)0); fill(&Omega,(int)0);
for(int y=0;y<sizeA(1);y++) for(int y=0;y<sizeA(1);y++)
for(int x=0;x<sizeA(0);x++) for(int x=0;x<sizeA(0);x++)
@@ -566,11 +578,19 @@ void patchmatch(const V2i& sizeA,
{ {
const int iter_seed = rand(); const int iter_seed = rand();
#ifdef __APPLE__
dispatch_apply(numTiles,gcdq,^(size_t blockIdx)
#else
#pragma omp parallel num_threads(numTiles) #pragma omp parallel num_threads(numTiles)
#endif
{ {
const bool odd = (iter%2 == 0); const bool odd = (iter%2 == 0);
#ifdef __APPLE__
const int threadId = blockIdx;
#else
const int threadId = omp_get_thread_num(); const int threadId = omp_get_thread_num();
#endif
const int _y0 = threadId*tileHeight; const int _y0 = threadId*tileHeight;
const int _y1 = threadId==numTiles-1 ? sizeA(1) : std::min(_y0+tileHeight,sizeA(1)); const int _y1 = threadId==numTiles-1 ? sizeA(1) : std::min(_y0+tileHeight,sizeA(1));
@@ -635,6 +655,9 @@ void patchmatch(const V2i& sizeA,
#undef RANDI #undef RANDI
} }
} }
#ifdef __APPLE__
);
#endif
} }
} }
@@ -683,7 +706,7 @@ void ebsynthCpu(int numStyleChannels,
Array2<Vec<2,int>> NNF; Array2<Vec<2,int>> NNF;
//Array2<Vec<2,int>> NNF2; //Array2<Vec<2,int>> NNF2;
Array2<float> E; Array2<float> E;
//Array2<int> Omega; Array2<int> Omega;
}; };
std::vector<PyramidLevel> pyramid(levelCount); std::vector<PyramidLevel> pyramid(levelCount);
@@ -727,7 +750,7 @@ void ebsynthCpu(int numStyleChannels,
//pyramid[level].mask2 = Array2<unsigned char>(levelTargetSize); //pyramid[level].mask2 = Array2<unsigned char>(levelTargetSize);
pyramid[level].NNF = Array2<Vec<2,int>>(levelTargetSize); pyramid[level].NNF = Array2<Vec<2,int>>(levelTargetSize);
//pyramid[level].NNF2 = Array2<Vec<2,int>>(levelTargetSize); //pyramid[level].NNF2 = Array2<Vec<2,int>>(levelTargetSize);
//pyramid[level].Omega = Array2<int>(levelSourceSize); pyramid[level].Omega = Array2<int>(levelSourceSize);
pyramid[level].E = Array2<float>(levelTargetSize); pyramid[level].E = Array2<float>(levelTargetSize);
if (level<levelCount-1) if (level<levelCount-1)
@@ -860,7 +883,8 @@ void ebsynthCpu(int numStyleChannels,
numPatchMatchItersPerLevel[level], numPatchMatchItersPerLevel[level],
-1, -1,
pyramid[level].NNF, pyramid[level].NNF,
pyramid[level].E); pyramid[level].E,
pyramid[level].Omega);
} }
} }
/* /*
@@ -946,7 +970,7 @@ void ebsynthCpu(int numStyleChannels,
//pyramid[level].mask = Array2<unsigned char>(); //pyramid[level].mask = Array2<unsigned char>();
//pyramid[level].mask2 = Array2<unsigned char>(); //pyramid[level].mask2 = Array2<unsigned char>();
//pyramid[level].NNF2 = Array2<Vec<2,int>>(); //pyramid[level].NNF2 = Array2<Vec<2,int>>();
//pyramid[level].Omega = Array2<int>(); pyramid[level].Omega = Array2<int>();
pyramid[level].E = Array2<float>(); pyramid[level].E = Array2<float>();
if (targetModulationData) { pyramid[level].targetModulation = Array2<Vec<NG,unsigned char>>(); } if (targetModulationData) { pyramid[level].targetModulation = Array2<Vec<NG,unsigned char>>(); }
} }