////////////////////////////////////////////////////////////////////////////////
//! Run the Cuda kernels
////////////////////////////////////////////////////////////////////////////////
void runCuda()
{
size_t num_bytes;
// generate wave spectrum in frequency domain
cudaGenerateSpectrumKernel(d_h0, d_ht, fftInputW, fftInputH, animTime, patchSize);
// execute inverse FFT to convert to spatial domain
// DEPRECATED: cutilSafeCall(cudaGLMapBufferObject((void**)&g_hptr, heightVertexBuffer));
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_heightVB_resource, 0));
// DEPRECATED: cutilSafeCall(cudaGLMapBufferObject((void**)&g_hptr, heightVertexBuffer));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&g_hptr, &num_bytes, cuda_heightVB_resource));
cufftSafeCall( cufftExecC2R(fftPlan, (cufftComplex *) d_ht, g_hptr) );
// calculate slope for shading
// DEPRECATED: cutilSafeCall(cudaGLMapBufferObject((void**)&g_sptr, slopeVertexBuffer));
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_slopeVB_resource, 0));
// DEPRECATED: cutilSafeCall(cudaGraphicsMapBufferObject((void**)&g_sptr, slopeVertexBuffer));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&g_sptr, &num_bytes, cuda_slopeVB_resource));
cudaCalculateSlopeKernel(g_hptr, g_sptr, meshW, meshH);
// DEPRECATED: cutilSafeCall(cudaGLUnmapBufferObject(slopeVertexBuffer));
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_slopeVB_resource, 0));
// DEPRECATED: cutilSafeCall(cudaGLUnmapBufferObject(heightVertexBuffer));
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_heightVB_resource, 0));
}
/*
* Advance the simulation by <n> generations by mapping the OpenGL pixel buffer
* objects for writing from CUDA, executing the kernel <n> times, and unmapping
* the pixel buffer object.
*/
void advance_generations(unsigned long n)
{
uint8_t* device_bufs[2];
size_t size;
DEBUG2("Mapping CUDA resources and retrieving device buffer pointers\n");
cudaGraphicsMapResources(2, cuda_graphics_resources, (cudaStream_t)0);
cudaGraphicsResourceGetMappedPointer((void**)&device_bufs[0], &size,
cuda_graphics_resources[0]);
cudaGraphicsResourceGetMappedPointer((void**)&device_bufs[1], &size,
cuda_graphics_resources[1]);
check_cuda_error();
while (n--) {
DEBUG2("Launching kernel (grid.width = %u, grid.height = %u)\n",
grid.width, grid.height);
launch_kernel(device_bufs[grid.which_buf], device_bufs[!grid.which_buf],
grid.width, grid.height);
grid.which_buf ^= 1;
}
DEBUG2("Unmapping CUDA resources\n");
cudaGraphicsUnmapResources(2, cuda_graphics_resources, (cudaStream_t)0);
cudaStreamSynchronize(0);
}
void WaterPlaneCUDA::update()
{
glBindBuffer(GL_ARRAY_BUFFER, oldVertexBuffer);
float3* verticesTest = (float3*)glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY);
for (int i = 0; i < disturbances.size();i++)
{
Disturbances *dist = disturbances.at(i);
for(int x = dist->xminW; x <= dist->xmaxW; x++)
{
for (int y = dist->zminW; y <= dist->zmaxW; y++)
{
float insideCircle = ((x-dist->centerX)*(x-dist->centerX))+((y-dist->centerZ)*(y-dist->centerZ))-dist->radiusSQ;
if (insideCircle <= 0) {
int vIndex = (y * pointsX) + x;
if (vIndex < (pointsX*pointsY)) {
verticesTest[vIndex].y = (insideCircle/dist->radiusSQ)*dist->height;
}
}
}
}
}
glUnmapBufferARB(GL_ARRAY_BUFFER);
disturbances.clear();
size_t num_bytes;
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_newVertex_resource, 0));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&gpu_newVertices, &num_bytes, cuda_newVertex_resource));
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_oldVertex_resource, 0));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&gpu_oldVertices, &num_bytes, cuda_oldVertex_resource));
cutilSafeCall(cudaMemcpyToSymbol("DIM",&pointsX,sizeof(int)));
updateWaveMapGPU1(gpu_newVertices,gpu_oldVertices);
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_newVertex_resource, 0));
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_oldVertex_resource, 0));
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_oldVertex_resource, 0));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&gpu_oldVertices, &num_bytes, cuda_oldVertex_resource));
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_normalsVB_resource, 0));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void**)&gpu_normals, &num_bytes, cuda_normalsVB_resource));
cutilSafeCall(cudaMemcpyToSymbol("DIM",&pointsX,sizeof(int)));
updateNormalsGPU1(gpu_oldVertices,gpu_normals);
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_normalsVB_resource, 0));
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_oldVertex_resource, 0));
//swap between old and new wave map
struct cudaGraphicsResource *temp = cuda_oldVertex_resource;
cuda_oldVertex_resource = cuda_newVertex_resource;
cuda_newVertex_resource = temp;
}
void runBenchmark(int iterations)
{
printf("[%s] (Benchmark Mode)\n", sSDKsample);
sdkCreateTimer(&timer);
uchar4 *d_output;
checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
size_t num_bytes;
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&d_output, &num_bytes,
cuda_pbo_resource));
sdkStartTimer(&timer);
for (int i = 0; i < iterations; ++i)
{
render(imageWidth, imageHeight, tx, ty, scale, cx, cy,
blockSize, gridSize, g_FilterMode, d_output);
}
cudaDeviceSynchronize();
sdkStopTimer(&timer);
float time = sdkGetTimerValue(&timer) / (float) iterations;
checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0));
printf("time: %0.3f ms, %f Mpixels/sec\n", time, (width*height / (time * 0.001f)) / 1e6);
}
////////////////////////////////////////////////////////////////////////////////
//! Run the Cuda part of the computation
////////////////////////////////////////////////////////////////////////////////
void runCuda()
{
cudaStream_t stream = 0;
const int nbResources = 2;
cudaGraphicsResource *ppResources[nbResources] =
{
g_histogram.cudaResource,
g_color.cudaResource,
};
// Map resources for Cuda
checkCudaErrors(cudaGraphicsMapResources(nbResources, ppResources, stream));
getLastCudaError("cudaGraphicsMapResources(2) failed");
// Get pointers
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&g_histogram.cudaBuffer, &g_histogram.size, g_histogram.cudaResource));
getLastCudaError("cudaGraphicsResourceGetMappedPointer (g_color.pBuffer) failed");
cudaGraphicsSubResourceGetMappedArray(&g_color.pCudaArray, g_color.cudaResource, 0, 0);
getLastCudaError("cudaGraphicsSubResourceGetMappedArray (g_color.pBuffer) failed");
// Execute kernel
createHistogramTex(g_histogram.cudaBuffer, g_WindowWidth, g_WindowHeight, g_color.pCudaArray);
checkCudaError();
//
// unmap the resources
//
checkCudaErrors(cudaGraphicsUnmapResources(nbResources, ppResources, stream));
getLastCudaError("cudaGraphicsUnmapResources(2) failed");
}
//-----------------------------------------------------------------------------
void QGLImageGpuWidget::fillPbo(iu::ImageGpu_8u_C4* output)
{
// map GL <-> CUDA resource
uchar4 *d_dst = NULL;
size_t start;
cudaGraphicsMapResources(1, &cuda_pbo_resource_, 0);
cudaGraphicsResourceGetMappedPointer((void**)&d_dst, &start, cuda_pbo_resource_);
// get image data
iuprivate::cuCopyImageToPbo(image_, num_channels_, bit_depth_, d_dst, min_, max_);
cudaThreadSynchronize();
// get overlays
iuprivate::OverlayList::iterator it;
for ( it=overlay_list_.begin() ; it != overlay_list_.end(); it++ )
if ((*it)->isActive())
cuCopyOverlayToPbo((*it), d_dst, image_->size());
cudaThreadSynchronize();
if (output != NULL)
{
// copy final pbo to output
iu::ImageGpu_8u_C4 temp(d_dst, image_->width(), image_->height(),
image_->width()*sizeof(uchar4), true);
iu::copy(&temp, output);
}
// unmap GL <-> CUDA resource
cudaGraphicsUnmapResources(1, &cuda_pbo_resource_, 0);
}
/*Private Methods*/
void Screen::render()
{
iteration_count++;
uchar4 * out_data = 0;
cudaError_t cudaStatus;
cudaStatus = cudaGraphicsMapResources(1, &cuda_pbo_resource, 0);
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "Call to cudaGraphicsMapResources failed.\n");
}
cudaStatus = cudaGraphicsResourceGetMappedPointer((void **)&out_data, NULL, cuda_pbo_resource);
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "Call to cudaGraphicsResourceGetMappedPointer failed.\n");
}
for (int i = 0; i < ITERATIONS_PER_RENDER; i++)
{
kernel_ptr->launchCalculations(out_data, MAX_WIN_WIDTH, MAX_WIN_HEIGHT);
}
std::cout << "Iteration number: " << iteration_count*ITERATIONS_PER_RENDER << std::endl;
cudaStatus = cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0);
if (cudaStatus != cudaSuccess) {
fprintf(stderr, "Call to cudaGraphicsUnmapResources failed.\n");
}
}
void displayFunc(void)
{
sdkStartTimer(&timer);
TColor *d_dst = NULL;
size_t num_bytes;
if (frameCounter++ == 0)
{
sdkResetTimer(&timer);
}
// DEPRECATED: checkCudaErrors(cudaGLMapBufferObject((void**)&d_dst, gl_PBO));
checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
getLastCudaError("cudaGraphicsMapResources failed");
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&d_dst, &num_bytes, cuda_pbo_resource));
getLastCudaError("cudaGraphicsResourceGetMappedPointer failed");
checkCudaErrors(CUDA_Bind2TextureArray());
runImageFilters(d_dst);
checkCudaErrors(CUDA_UnbindTexture());
// DEPRECATED: checkCudaErrors(cudaGLUnmapBufferObject(gl_PBO));
checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0));
// Common display code path
{
glClear(GL_COLOR_BUFFER_BIT);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, imageW, imageH, GL_RGBA, GL_UNSIGNED_BYTE, BUFFER_DATA(0));
glBegin(GL_TRIANGLES);
glTexCoord2f(0, 0);
glVertex2f(-1, -1);
glTexCoord2f(2, 0);
glVertex2f(+3, -1);
glTexCoord2f(0, 2);
glVertex2f(-1, +3);
glEnd();
glFinish();
}
if (frameCounter == frameN)
{
frameCounter = 0;
if (g_FPS)
{
printf("FPS: %3.1f\n", frameN / (sdkGetTimerValue(&timer) * 0.001));
g_FPS = false;
}
}
glutSwapBuffers();
glutReportErrors();
sdkStopTimer(&timer);
computeFPS();
}
void VBO::map()
{
size_t num_bytes;
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_vbo_resource));
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void **)&dev_vbo, &num_bytes,
cuda_vbo_resource));
}
// This is the normal display path
void display(void)
{
cutilCheckError(cutStartTimer(timer));
// Sobel operation
Pixel *data = NULL;
// map PBO to get CUDA device pointer
cutilSafeCall(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
size_t num_bytes;
cutilSafeCall(cudaGraphicsResourceGetMappedPointer((void **)&data, &num_bytes,
cuda_pbo_resource));
//printf("CUDA mapped PBO: May access %ld bytes\n", num_bytes);
sobelFilter(data, imWidth, imHeight, g_SobelDisplayMode, imageScale, blockOp, pointOp );
cutilSafeCall(cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0));
glClear(GL_COLOR_BUFFER_BIT);
glBindTexture(GL_TEXTURE_2D, texid);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo_buffer);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, imWidth, imHeight,
GL_LUMINANCE, GL_UNSIGNED_BYTE, OFFSET(0));
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glDisable(GL_DEPTH_TEST);
glEnable(GL_TEXTURE_2D);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
glBegin(GL_QUADS);
glVertex2f(0, 0); glTexCoord2f(0, 0);
glVertex2f(0, 1); glTexCoord2f(1, 0);
glVertex2f(1, 1); glTexCoord2f(1, 1);
glVertex2f(1, 0); glTexCoord2f(0, 1);
glEnd();
glBindTexture(GL_TEXTURE_2D, 0);
if (g_CheckRender && g_CheckRender->IsQAReadback() && g_Verify) {
printf("> (Frame %d) readback BackBuffer\n", frameCount);
g_CheckRender->readback( imWidth, imHeight );
g_CheckRender->savePPM ( sOriginal_ppm[g_Index], true, NULL );
if (!g_CheckRender->PPMvsPPM(sOriginal_ppm[g_Index], sReference_ppm[g_Index], MAX_EPSILON_ERROR, 0.15f)) {
g_TotalErrors++;
}
g_Verify = false;
}
glutSwapBuffers();
cutilCheckError(cutStopTimer(timer));
computeFPS();
glutPostRedisplay();
}
std::pair<float *, unsigned int> VBOCudaMapper::mapVBOBuffer()
{
cudaCheck(cudaGraphicsMapResources(1, &cuda_graphics_resource_, nullptr));
float * cuda_data_ptr = nullptr;
size_t byte_size;
cudaCheck(cudaGraphicsResourceGetMappedPointer((void **)&cuda_data_ptr, &byte_size, cuda_graphics_resource_));
return{ cuda_data_ptr, byte_size };
}
void *
OsdCudaVertexBuffer::Map() {
size_t num_bytes;
void *ptr;
cudaGraphicsMapResources(1, &_cudaResource, 0);
cudaGraphicsResourceGetMappedPointer(&ptr, &num_bytes, _cudaResource);
return ptr;
}
void Canvas::paintGL()
{
glClear(GL_COLOR_BUFFER_BIT);
if(!ready) return;
size_t size;
checkCudaErrors(cudaGraphicsMapResources(1, &resource, 0));
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void**)&img, &size, resource));
if(renderMode == RENDER_MODE_RAYCASTING)
{
/*glEnable(GL_BLEND);
glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
glBegin(GL_QUADS);
glColor4f(0.4745f, 0.9294f, 0.8901f, 1.f);
glVertex2f(1.f, 1.f);
glColor4f(0.4745f, 0.9294f, 0.8901f, 1.f);
glVertex2f(-1.f, 1.f);
glColor4f(0.9490f, 0.9647f, 0.9803f, 1.f);
glVertex2f(-1.f, -1.f);
glColor4f(0.9490f, 0.9647f, 0.9803f, 1.f);
glVertex2f(1.f, -1.f);
glEnd();*/
render_raycasting(img, deviceVolume, transferFunction, camera, volumeReader.GetElementBoundingSphereRadius());
}
else
{
render_pathtracer(img, renderParams);
if(renderParams.frameNo == 0)
{
char* data = new char[WIDTH * HEIGHT * 4];
cudaMemcpy(data, img, sizeof(glm::u8vec4) * WIDTH * HEIGHT, cudaMemcpyDeviceToHost);
stbi_write_tga("0.tga", WIDTH, HEIGHT, 4, data);
delete []data;
}
}
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaGraphicsUnmapResources(1, &resource, 0));
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo);
glDrawPixels(WIDTH, HEIGHT, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glDisable(GL_BLEND);
renderParams.frameNo++;
}
void
OsdCudaD3D11VertexBuffer::map() {
if (_cudaBuffer) return;
size_t num_bytes;
void *ptr;
cudaGraphicsMapResources(1, &_cudaResource, 0);
cudaGraphicsResourceGetMappedPointer(&ptr, &num_bytes, _cudaResource);
_cudaBuffer = ptr;
}
// display results using OpenGL
void display()
{
sdkStartTimer(&timer);
// execute filter, writing results to pbo
unsigned int *dResult;
//DEPRECATED: checkCudaErrors( cudaGLMapBufferObject((void**)&d_result, pbo) );
checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
size_t num_bytes;
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&dResult, &num_bytes, cuda_pbo_resource));
bilateralFilterRGBA(dResult, width, height, euclidean_delta, filter_radius, iterations, kernel_timer);
// DEPRECATED: checkCudaErrors(cudaGLUnmapBufferObject(pbo));
checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0));
// Common display code path
{
glClear(GL_COLOR_BUFFER_BIT);
// load texture from pbo
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
glBindTexture(GL_TEXTURE_2D, texid);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, 0);
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
// fragment program is required to display floating point texture
glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, shader);
glEnable(GL_FRAGMENT_PROGRAM_ARB);
glDisable(GL_DEPTH_TEST);
glBegin(GL_QUADS);
{
glTexCoord2f(0, 0);
glVertex2f(0, 0);
glTexCoord2f(1, 0);
glVertex2f(1, 0);
glTexCoord2f(1, 1);
glVertex2f(1, 1);
glTexCoord2f(0, 1);
glVertex2f(0, 1);
}
glEnd();
glBindTexture(GL_TEXTURE_TYPE, 0);
glDisable(GL_FRAGMENT_PROGRAM_ARB);
}
glutSwapBuffers();
glutReportErrors();
sdkStopTimer(&timer);
computeFPS();
}
GLuint CGLUtil::gpuMapRgb2PixelBufferObj(const cv::cuda::GpuMat& cvgmRGB_ ){
//http://rickarkin.blogspot.co.uk/2012/03/use-pbo-to-share-buffer-between-cuda.html
int nPyrLevel_ = getLevel( cvgmRGB_.cols );
GLuint uTexture;
// map OpenGL buffer object for writing from CUDA
if (cvgmRGB_.channels() == 3) {
uTexture = _auTexture[nPyrLevel_];
void *pDev;
cudaSafeCall( cudaGraphicsMapResources(1, &_apResourceRGBPxielBO[nPyrLevel_], 0));
size_t nSize;
cudaSafeCall( cudaGraphicsResourceGetMappedPointer((void **)&pDev, &nSize , _apResourceRGBPxielBO[nPyrLevel_]));
cv::cuda::GpuMat cvgmRGBA( cvgmRGB_.size(), CV_8UC3, pDev);
cvgmRGB_.copyTo(cvgmRGBA);
cudaSafeCall( cudaGraphicsUnmapResources(1, &_apResourceRGBPxielBO[nPyrLevel_], 0) );
//texture mapping
glBindTexture( GL_TEXTURE_2D, uTexture);
glBindBuffer ( GL_PIXEL_UNPACK_BUFFER_ARB, _auRGBPixelBO[nPyrLevel_]);
glTexImage2D( GL_TEXTURE_2D, 0, GL_RGB, cvgmRGB_.cols, cvgmRGB_.rows, 0, GL_RGB, GL_UNSIGNED_BYTE, NULL);
errorDetectorGL();
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
glBindTexture(GL_TEXTURE_2D, 0);
}
else if (cvgmRGB_.channels()==1) {
uTexture = _auGrayTexture[nPyrLevel_];
void *pDev;
cudaSafeCall( cudaGraphicsMapResources(1, &_apResourceGrayPxielBO[nPyrLevel_], 0));
size_t nSize;
cudaSafeCall( cudaGraphicsResourceGetMappedPointer((void **)&pDev, &nSize , _apResourceGrayPxielBO[nPyrLevel_]));
cv::cuda::GpuMat cvgmRGBA( cvgmRGB_.size(), CV_8UC1, pDev);
cvgmRGB_.copyTo(cvgmRGBA);
cudaSafeCall( cudaGraphicsUnmapResources(1, &_apResourceGrayPxielBO[nPyrLevel_], 0) );
//texture mapping
glBindTexture(GL_TEXTURE_2D, uTexture);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, _auGrayPixelBO[nPyrLevel_]);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RED, cvgmRGB_.cols, cvgmRGB_.rows, 0, GL_RED, GL_UNSIGNED_BYTE, NULL);
errorDetectorGL();
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
glBindTexture(GL_TEXTURE_2D, 0);
}
return uTexture;
}//gpuMapRgb2PixelBufferObj
void render() {
uchar4 *d_out = 0;
cudaGraphicsMapResources(1, &cuda_pbo_resource, 0);
cudaGraphicsResourceGetMappedPointer((void **)&d_out, NULL,
cuda_pbo_resource);
kernelLauncher(d_out, W, H, param, sys);
cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0);
// update contents of the title bar
char title[64];
sprintf(title, "Stability: param = %.1f, sys = %d", param, sys);
glutSetWindowTitle(title);
}
void RunCuda(struct cudaGraphicsResource **resource)
{
// map OpenGL buffer object for writing from CUDA
checkCudaErrors(cudaGraphicsMapResources(1, resource, 0), exit(0));
float4 *devPtr;
size_t size;
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&devPtr, &size, *resource), exit(0));
//printf("CUDA mapped VBO: May access %ld bytes\n", size);
launch_kernel(devPtr, MeshWidth, MeshHeight, _anim);
// unmap buffer object
checkCudaErrors(cudaGraphicsUnmapResources(1, resource, 0), exit(0));
}
T* BodySystemGPU<T>::getArray(BodyArray array)
{
assert(m_bInitialized);
T *hdata = 0;
T *ddata = 0;
cudaGraphicsResource *pgres = nullptr;
int currentReadHost = m_bUseSysMem ? m_currentRead : 0;
switch (array)
{
default:
case BODYSYSTEM_POSITION:
hdata = m_hPos[currentReadHost];
ddata = m_deviceData[0].dPos[m_currentRead];
if (m_bUsePBO)
{
pgres = m_pGRes[m_currentRead];
}
break;
case BODYSYSTEM_VELOCITY:
hdata = m_hVel;
ddata = m_deviceData[0].dVel;
break;
}
if (!m_bUseSysMem)
{
if (pgres)
{
checkCudaErrors(cudaGraphicsResourceSetMapFlags(pgres, cudaGraphicsMapFlagsReadOnly));
checkCudaErrors(cudaGraphicsMapResources(1, &pgres, 0));
size_t bytes;
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&ddata, &bytes, pgres));
}
checkCudaErrors(cudaMemcpy(hdata, ddata,
m_numBodies*4*sizeof(T), cudaMemcpyDeviceToHost));
if (pgres)
{
checkCudaErrors(cudaGraphicsUnmapResources(1, &pgres, 0));
}
}
return hdata;
}
void TriangleMesh::bind(TriangleMeshData& data)
{
data.numVertices = 0;
data.numIndices = 0;
data.maxNumVertices = static_cast<unsigned int>(maxNumVertices);
data.maxNumIndices = static_cast<unsigned int>(maxNumIndices);
if (!bound)
{
if (!registered)
{
PGA_CUDA_checkedCall(cudaGraphicsD3D11RegisterResource(&cudaVertexAttributesBuffer, vertexAttributesBuffer, cudaGraphicsRegisterFlagsNone /* the only flag working for now */));
PGA_CUDA_checkedCall(cudaGraphicsD3D11RegisterResource(&cudaIndexBuffer, indexBuffer, cudaGraphicsRegisterFlagsNone /* the only flag working for now */));
registered = true;
}
PGA_CUDA_checkedCall(cudaGraphicsMapResources(1, &cudaVertexAttributesBuffer, 0));
PGA_CUDA_checkedCall(cudaGraphicsMapResources(1, &cudaIndexBuffer, 0));
bound = true;
}
size_t size;
PGA_CUDA_checkedCall(cudaGraphicsResourceGetMappedPointer((void**)&data.verticesAttributes, &size, cudaVertexAttributesBuffer));
PGA_CUDA_checkedCall(cudaGraphicsResourceGetMappedPointer((void**)&data.indices, &size, cudaIndexBuffer));
}
void TriangleMesh::bind(TriangleMeshData& data)
{
data.numVertices = 0;
data.numIndices = 0;
data.maxNumVertices = static_cast<unsigned int>(maxNumVertices);
data.maxNumIndices = static_cast<unsigned int>(maxNumIndices);
if (!bound)
{
if (!registered)
{
PGA_CUDA_checkedCall(cudaGraphicsGLRegisterBuffer(&cudaVertexAttributesBuffer, vertexAttributesBuffer, cudaGraphicsMapFlagsReadOnly /* cudaGraphicsMapFlagsWriteDiscard */));
PGA_CUDA_checkedCall(cudaGraphicsGLRegisterBuffer(&cudaIndexBuffer, indexBuffer, cudaGraphicsMapFlagsReadOnly /* cudaGraphicsMapFlagsWriteDiscard */));
registered = true;
}
PGA_CUDA_checkedCall(cudaGraphicsMapResources(1, &cudaVertexAttributesBuffer, 0));
PGA_CUDA_checkedCall(cudaGraphicsMapResources(1, &cudaIndexBuffer, 0));
bound = true;
}
size_t size;
PGA_CUDA_checkedCall(cudaGraphicsResourceGetMappedPointer((void**)&data.verticesAttributes, &size, cudaVertexAttributesBuffer));
PGA_CUDA_checkedCall(cudaGraphicsResourceGetMappedPointer((void**)&data.indices, &size, cudaIndexBuffer));
}
void copy(Renderable& out, const T * devicePtr)
{
cudaGraphicsResource *cudaVBOResource;
CUDA_ERROR_CHECK(cudaGraphicsGLRegisterBuffer(&cudaVBOResource, out.vbo(), cudaGraphicsMapFlagsWriteDiscard));
size_t num_bytes;
T* vboDevicePtr = NULL;
CUDA_ERROR_CHECK(cudaGraphicsMapResources(1, &cudaVBOResource, 0));
CUDA_ERROR_CHECK(cudaGraphicsResourceGetMappedPointer((void **)&vboDevicePtr, &num_bytes, cudaVBOResource));
CUDA_ERROR_CHECK(cudaMemcpy(vboDevicePtr, devicePtr, num_bytes, cudaMemcpyDeviceToDevice));
CUDA_ERROR_CHECK(cudaGraphicsUnmapResources(1, &cudaVBOResource, 0));
CUDA_ERROR_CHECK(cudaGraphicsUnregisterResource(cudaVBOResource));
}
void render() {
uchar4 *d_out = 0;
cudaGraphicsMapResources(1, &cuda_pbo_resource, 0);
cudaGraphicsResourceGetMappedPointer((void **)&d_out, NULL,
cuda_pbo_resource);
kernelLauncher(d_out, d_vol, W, H, volumeSize, method, zs, theta,
threshold, dist);
cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0);
char title[128];
sprintf(title, "Volume Visualizer : objId =%d, method = %d,"
" dist = %.1f, theta = %.1f", id, method, dist,
theta);
glutSetWindowTitle(title);
}
// This is the normal display path
void display(void)
{
sdkStartTimer(&timer);
// Sobel operation
Pixel *data = NULL;
// map PBO to get CUDA device pointer
checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
size_t num_bytes;
checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&data, &num_bytes,
cuda_pbo_resource));
//printf("CUDA mapped PBO: May access %ld bytes\n", num_bytes);
sobelFilter(data, imWidth, imHeight, g_SobelDisplayMode, imageScale);
checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_pbo_resource, 0));
glClear(GL_COLOR_BUFFER_BIT);
glBindTexture(GL_TEXTURE_2D, texid);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo_buffer);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, imWidth, imHeight,
GL_LUMINANCE, GL_UNSIGNED_BYTE, OFFSET(0));
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glDisable(GL_DEPTH_TEST);
glEnable(GL_TEXTURE_2D);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
glBegin(GL_QUADS);
glVertex2f(0, 0);
glTexCoord2f(0, 0);
glVertex2f(0, 1);
glTexCoord2f(1, 0);
glVertex2f(1, 1);
glTexCoord2f(1, 1);
glVertex2f(1, 0);
glTexCoord2f(0, 1);
glEnd();
glBindTexture(GL_TEXTURE_2D, 0);
glutSwapBuffers();
sdkStopTimer(&timer);
computeFPS();
}
void
OsdCudaGLVertexBuffer::map() {
if (_devicePtr) return;
size_t num_bytes;
void *ptr;
cudaError_t err = cudaGraphicsMapResources(1, &_cudaResource, 0);
if (err != cudaSuccess)
OsdError(OSD_CUDA_GL_ERROR, "OsdCudaGLVertexBuffer::map failed.\n%s\n", cudaGetErrorString(err));
err = cudaGraphicsResourceGetMappedPointer(&ptr, &num_bytes, _cudaResource);
if (err != cudaSuccess)
OsdError(OSD_CUDA_GL_ERROR, "OsdCudaGLVertexBuffer::map failed.\n%s\n", cudaGetErrorString(err));
_devicePtr = ptr;
}
// Run the Cuda part of the computation
void runCuda(glm::mat3 rot, glm::vec3 campos, float focalLength, LOD l) {
uchar4 *dptr = NULL;
// Map the buffer object
size_t num_bytes;
cudaGraphicsMapResources(1, &cuda_pbo, 0);
// Get Address for kernel
cudaGraphicsResourceGetMappedPointer((void**)&dptr, &num_bytes, cuda_pbo);
launchKernel(dptr, window_width, window_height, focalLength, rot, campos, l);
// Unmap the buffer object
cudaGraphicsUnmapResources(1, &cuda_pbo, 0);
}
void CGLUtil::gpuMapRGBResources(const cv::cuda::GpuMat& cvgmRGBs_){
int nPyrLevel_ = getLevel( cvgmRGBs_.cols );
// map OpenGL buffer object for writing from CUDA
void *pDev;
cudaGraphicsMapResources(1, &_apResourceRGBVBO[nPyrLevel_], 0);
size_t nSize;
cudaGraphicsResourceGetMappedPointer((void **)&pDev, &nSize, _apResourceRGBVBO[nPyrLevel_] );
cv::cuda::GpuMat cvgmRGBs(cvgmRGBs_.size(),CV_8UC3,pDev);
cvgmRGBs_.copyTo(cvgmRGBs);
cudaGraphicsUnmapResources(1, &_apResourceRGBVBO[nPyrLevel_], 0);
// render from the vbo
glBindBuffer(GL_ARRAY_BUFFER, _auRGBVBO[nPyrLevel_]);
glColorPointer(3, GL_UNSIGNED_BYTE, 0, 0);
glEnableClientState(GL_COLOR_ARRAY);//you cant move glEnableClientState infront of cuda GraphicsMapResources, otherwise, you will have weird problem
//glDrawArrays(GL_POINTS, 0, btl::kinect::__aKinectWxH[usPyrLevel_] );
//glDisableClientState(GL_COLOR_ARRAY);
//glBindBuffer( GL_ARRAY_BUFFER, 0 );
}
void CGLUtil::gpuMapNlResources(const cv::cuda::GpuMat& cvgmNls_){
int nPyrLevel_ = getLevel( cvgmNls_.cols );
// map OpenGL buffer object for writing from CUDA
void *pDev;
cudaGraphicsMapResources(1, &_apResourceNlVBO[nPyrLevel_], 0);
size_t nSize;
cudaGraphicsResourceGetMappedPointer((void **)&pDev, &nSize, _apResourceNlVBO[nPyrLevel_] );
cv::cuda::GpuMat cvgmNls(cvgmNls_.size(),CV_32FC3,pDev);
cvgmNls_.copyTo(cvgmNls);
cudaGraphicsUnmapResources(1, &_apResourceNlVBO[nPyrLevel_], 0);
// render from the vbo
glBindBuffer(GL_ARRAY_BUFFER, _auNlVBO[nPyrLevel_]);
glNormalPointer(GL_FLOAT, 12, 0); //12 is the stride = the number of bytes occupied by each normal
glEnableClientState(GL_NORMAL_ARRAY);//you cant move glEnableClientState infront of cuda GraphicsMapResources, otherwise, you will have weird problem
//glColor3f(1.0, 0.0, 0.0);
//glDrawArrays(GL_POINTS, 0, btl::kinect::__aKinectWxH[usPyrLevel_] );
//glDisableClientState(GL_NORMAL_ARRAY);
//glBindBuffer( GL_ARRAY_BUFFER, 0 );
}
void pbf_dam_sim::simulateOneStep()
{
//simulatee.external.body_force
cudaGraphicsMapResources(1, &cu_res);
#ifndef NDEBUG
gpuErrchk(cudaPeekAtLastError());
gpuErrchk(cudaDeviceSynchronize());
#endif
size_t pos_size;
cudaGraphicsResourceGetMappedPointer((void**)&simulatee.phase.x, &pos_size, cu_res);
#ifndef NDEBUG
gpuErrchk(cudaPeekAtLastError());
gpuErrchk(cudaDeviceSynchronize());
#endif
one_step(simulatee, buffer, domain, 3);
cudaGraphicsUnmapResources(1, &cu_res);
simulatee.phase.x = NULL;
}
void CGLUtil::gpuMapPtResources(const cv::cuda::GpuMat& cvgmPts_){
int nPyrLevel_ = getLevel( cvgmPts_.cols );
// map OpenGL buffer object for writing from CUDA
void *pDev;
cudaGraphicsMapResources(1, &_apResourcePtVBO[nPyrLevel_], 0);
size_t nSize;
cudaGraphicsResourceGetMappedPointer((void **)&pDev, &nSize, _apResourcePtVBO[nPyrLevel_] );
cv::cuda::GpuMat cvgmPts( cvgmPts_.size(),CV_32FC3,pDev );
cvgmPts_.copyTo(cvgmPts); // the operation of the Buffer must be done before cudaGraphicsUnmapResources(), otherwise, the buffer will affects each other
cudaGraphicsUnmapResources(1, &_apResourcePtVBO[nPyrLevel_], 0);
// render from the vbo
glBindBuffer(GL_ARRAY_BUFFER, _auPtVBO[nPyrLevel_]);
glVertexPointer(3, GL_FLOAT, 0, 0);
glEnableClientState(GL_VERTEX_ARRAY);//you cant move glEnableClientState in front of cuda GraphicsMapResources, otherwise, you will have weird problem
//glColor3f(1.0, 0.0, 0.0);
//glDrawArrays(GL_POINTS, 0, btl::kinect::__aKinectWxH[usPyrLevel_] );
//glDisableClientState(GL_VERTEX_ARRAY);
//glBindBuffer( GL_ARRAY_BUFFER, 0 );
}