#include "Fractal.h"\r
\r
#include <iostream>\r
+#include <sstream>\r
#include <assert.h>\r
using namespace std;\r
\r
+#include <omp.h>\r
+\r
#include "FractalDevice.h"\r
#include "Device.h"\r
\r
Fractal::Fractal(int w, int h) :\r
w(w), h(h),\r
dg(8, 8, 1),\r
- db(16, 16, 1),\r
- title("Fractal Cuda")\r
+ db(16, 16, 1)\r
{\r
//print(dg, db);\r
Device::assertDim(dg, db);\r
return this->ptrDomaineMathInit;\r
}\r
\r
-/**\r
- * Override\r
- */\r
-string Fractal::getTitle()\r
- {\r
- return this->title;\r
- }\r
-\r
/////\r
\r
-FractalMandelbrot::FractalMandelbrot(int w, int h, int dn) :\r
+FractalMandelbrot::FractalMandelbrot(int w, int h, int dn, bool multiGPU) :\r
Fractal(w, h),\r
variateurAnimationN(IntervalI(10, 100), dn),\r
- n(0)\r
+ n(0),\r
+ multiGPU(multiGPU)\r
{\r
+ // Constuit le titre dynamiquement.\r
+ ostringstream titleStream;\r
+ titleStream << "Fractal Mandelbrot (multi-GPU " << (this->multiGPU ? "activated" : "not activated") << ")";\r
+ this->title = titleStream.str();\r
+\r
this->ptrDomaineMathInit = new DomaineMath(-2, -1.3, 0.8, 1.3);\r
+\r
+ if (this->multiGPU)\r
+ {\r
+ const int nbDevice = Device::getDeviceCount();\r
+\r
+ this->hDevices = h / nbDevice;\r
+ this->hFirstDevice = h - ((nbDevice - 1) * this->hDevices);\r
+\r
+ // Allocation de la mémoire sur chaque GPU (sauf le premier pour lequel 'ptrDevPixels' est automatiquement alloué à l'appel de 'runGPU(..)').\r
+ this->ptrDevPixelsMultGPU = new uchar4*[nbDevice - 1];\r
+ for (int i = 0; i < nbDevice - 1; ++i)\r
+ {\r
+ HANDLE_ERROR(cudaSetDevice(i + 1));\r
+ HANDLE_ERROR(cudaMalloc(&this->ptrDevPixelsMultGPU[i], sizeof(uchar4) * w * this->hDevices));\r
+ }\r
+\r
+ HANDLE_ERROR(cudaSetDevice(0));\r
+ }\r
}\r
\r
void FractalMandelbrot::animationStep()\r
values[0] = float(this->n);\r
}\r
\r
+string FractalMandelbrot::getTitle()\r
+ {\r
+ return this->title;\r
+ }\r
+\r
void FractalMandelbrot::runGPU(uchar4* ptrDevPixels, const DomaineMath& domaineMath)\r
{\r
- fractalMandelbrot<<<dg,db>>>(ptrDevPixels, this->w, this->h, domaineMath, static_cast<int>(this->n));\r
+\r
+ if (this->multiGPU)\r
+ {\r
+ HANDLE_ERROR(cudaSetDevice(0));\r
+ fractalMandelbrot<<<dg,db>>>(ptrDevPixels, this->w, 0, this->hFirstDevice, domaineMath, this->n);\r
+\r
+ const int nbDevice = Device::getDeviceCount();\r
+\r
+ // Rend chaque tranche par un GPU différent puis copie chaque tranche dans la mémoire du premier GPU.\r
+ #pragma omp parallel for\r
+ for (int i = 0; i < nbDevice - 1; ++i)\r
+ {\r
+ HANDLE_ERROR(cudaSetDevice(i + 1));\r
+ fractalMandelbrot<<<dg,db>>>(this->ptrDevPixelsMultGPU[i], this->w, i * this->hDevices + this->hFirstDevice, (i + 1) * this->hDevices + this->hFirstDevice, domaineMath, this->n);\r
+ HANDLE_ERROR(cudaMemcpy(ptrDevPixels + this->w * this->hFirstDevice + i * this->w * this->hDevices, this->ptrDevPixelsMultGPU[i], sizeof(uchar4) * this->w * this->hDevices, cudaMemcpyDeviceToDevice));\r
+ }\r
+\r
+ HANDLE_ERROR(cudaSetDevice(0));\r
+ }\r
+ else\r
+ {\r
+ fractalMandelbrot<<<dg,db>>>(ptrDevPixels, this->w, 0, this->h, domaineMath, this->n);\r
+ }\r
}\r
\r
/////\r
values[1] = this->z_i;\r
}\r
\r
+string FractalJulia::getTitle()\r
+ {\r
+ return "Fractal Julia";\r
+ }\r
+\r
void FractalJulia::runGPU(uchar4* ptrDevPixels, const DomaineMath& domaineMath)\r
{\r
fractalJulia<<<dg,db>>>(ptrDevPixels, this->w, this->h, domaineMath, this->n, this->z_r, this->z_i);\r