void OpenGLRenderer::renderPoints(const glm::vec3* positions, const Color256* colors, const int num, const Camera &camera) {
//always use the point shaders to render points
glUseProgram(points_program_);
GLuint mvp_location = glGetUniformLocation(points_program_, "u_mvpMatrix");
//Declare CUDA device pointers for it to use
float3* dptr_pos;
float3* dptr_col;
//Setup position buffer
glBindBuffer(GL_ARRAY_BUFFER, buffers_[0]);
glBufferData(GL_ARRAY_BUFFER, 3 * num*sizeof(float), NULL, GL_DYNAMIC_DRAW);
glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 0, NULL);
glEnableVertexAttribArray(0);
//Setup color buffer
glBindBuffer(GL_ARRAY_BUFFER, buffers_[1]);
glBufferData(GL_ARRAY_BUFFER, 3 * num*sizeof(float), NULL, GL_DYNAMIC_DRAW);
glVertexAttribPointer(1, 3, GL_FLOAT, GL_FALSE, 0, NULL);
glEnableVertexAttribArray(1);
//Register position and normal buffers with CUDA
cudaGLRegisterBufferObject(buffers_[0]);
cudaGLRegisterBufferObject(buffers_[1]);
//Map buffers to CUDA
cudaGLMapBufferObject((void**)&dptr_pos, buffers_[0]);
cudaGLMapBufferObject((void**)&dptr_col, buffers_[1]);
//Copy data to buffer with CUDA
copyPointsToGL(positions, colors, dptr_pos, dptr_col, num);
//Unmap buffers from CUDA
cudaGLUnmapBufferObject(buffers_[0]);
cudaGLUnmapBufferObject(buffers_[1]);
//Unregister position and normal buffers with CUDA
cudaGLUnregisterBufferObject(buffers_[0]);
cudaGLUnregisterBufferObject(buffers_[1]);
//Send the MVP Matrix
glUniformMatrix4fv(mvp_location, 1, GL_FALSE, glm::value_ptr(camera.mvp));
//Draw
glPointSize(1.0f);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glDrawArrays(GL_POINTS, 0, 3 * num);
}
void raytrace()
{
uint* imagedata;
cutilSafeCall(cudaGLMapBufferObject((void**)&imagedata, pbo));
float3 A, B, C;
camera.getImagePlane(A, B, C);
dev_camera d_cam(camera.getPosition(), A, B, C, aperture, focal);
dev_light d_light(light.getPosition(), light.getColor(), 4096);
//need to change here.
float3 minAABB, maxAABB;
world.getAABB(minAABB, maxAABB);
sceneInfo scene(world.getNumTriangles(), world.getNumSpheres(), world.getNumBoxes(), minAABB, maxAABB);
//TODO: add control for clear buffer here.
//change here for the many object case
raytraceImage(imagedata, dev_lastframe_ptr, dev_num_layers, r_width, r_height, moved, d_cam, d_light, scene);
//for showing the real frame rate
cudaMemcpy(&frame_num, dev_num_layers, sizeof(float), cudaMemcpyDeviceToHost);
frame_num++;
cudaMemcpy(dev_num_layers, &frame_num, sizeof(int), cudaMemcpyHostToDevice);
cutilSafeCall(cudaGLUnmapBufferObject(pbo));
//download texture from pbo
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
glBindTexture(GL_TEXTURE_2D, framebuffer);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, r_width, r_height, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
CUT_CHECK_ERROR_GL();
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
dptr=NULL;
vbo = mesh->getVBO();
vbosize = mesh->getVBOsize();
float newcbo[] = {0.0, 1.0, 0.0,
0.0, 0.0, 1.0,
1.0, 0.0, 0.0};
cbo = newcbo;
cbosize = 9;
ibo = mesh->getIBO();
ibosize = mesh->getIBOsize();
cudaGLMapBufferObject((void**)&dptr, pbo);
cudaRasterizeCore(dptr, glm::vec2(width, height), frame, vbo, vbosize, cbo, cbosize, ibo, ibosize);
cudaGLUnmapBufferObject(pbo);
vbo = NULL;
cbo = NULL;
ibo = NULL;
frame++;
fpstracker++;
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
dptr=NULL;
vbo = mesh->getVBO();
vbosize = mesh->getVBOsize();
float newcbo[] = {0.0, 1.0, 0.0,
0.0, 0.0, 1.0,
1.0, 0.0, 0.0};
cbo = newcbo;
cbosize = 9;
ibo = mesh->getIBO();
ibosize = mesh->getIBOsize();
nbo = mesh->getNBO();
nbosize = mesh->getNBOsize();
cudaGLMapBufferObject((void**)&dptr, pbo);
// Invert camera to convert to view matrix
cudaRasterizeCore(glm::inverse(cam), projection, light, draw_mode, dptr, glm::vec2(width, height), frame, vbo, vbosize, nbo, nbosize, cbo, cbosize, ibo, ibosize);
cudaGLUnmapBufferObject(pbo);
vbo = NULL;
cbo = NULL;
ibo = NULL;
frame++;
fpstracker++;
}
void runCuda()
{
//////////////////////
// Timing cuda call //
//////////////////////
float time;
cudaEvent_t start, stop;
cudaEventCreate(&start);
cudaEventCreate(&stop);
cudaEventRecord(start, 0);
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
dptr=NULL;
vbo = mesh->getVBO();
vbosize = mesh->getVBOsize();
nbo = mesh->getNBO();
nbosize = mesh->getNBOsize();
#if RGBONLY == 1
float newcbo[] = {0.0, 1.0, 0.0,
0.0, 0.0, 1.0,
1.0, 0.0, 0.0};
cbo = newcbo;
cbosize = 9;
#elif RGBONLY == 0
vec3 defaultColor(0.5f, 0.5f, 0.5f);
mesh->changeColor(defaultColor);
cbo = mesh->getCBO();
cbosize = mesh->getCBOsize();
#endif
ibo = mesh->getIBO();
ibosize = mesh->getIBOsize();
cudaGLMapBufferObject((void**)&dptr, pbo);
updateCamera();
cudaRasterizeCore(cam, dptr, glm::vec2(width, height), frame, vbo, vbosize, cbo, cbosize, ibo, ibosize, nbo, nbosize, lights, lightsize, alpha, beta, displayMode);
cudaGLUnmapBufferObject(pbo);
vbo = NULL;
cbo = NULL;
ibo = NULL;
frame++;
fpstracker++;
//////////////////////
// Timing cuda call //
//////////////////////
cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&time, start, stop);
printf("runCuda runtime: %3.1f ms \n", time);
}
void runCuda() {
if (camchanged) {
iteration = 0;
Camera &cam = renderState->camera;
glm::vec3 v = cam.view;
glm::vec3 u = cam.up;
glm::vec3 r = glm::cross(v, u);
glm::mat4 rotmat = glm::rotate(theta, r) * glm::rotate(phi, u);
cam.view = glm::vec3(rotmat * glm::vec4(v, 0.f));
cam.up = glm::vec3(rotmat * glm::vec4(u, 0.f));
cam.position += cammove.x * r + cammove.y * u + cammove.z * v;
// Camera to grid center
float distance = cam.resolution.x / 2 / tan(cam.fov.x / 2);
cam.toGrid = glm::vec3(cam.view.x*distance, cam.view.y*distance, cam.view.z*distance);
// Find camera right vector
float rAngle = -PI / 2;
float qx = cam.view.x * sin(rAngle / 2);
float qy = cam.view.y * sin(rAngle / 2);
float qz = cam.view.z * sin(rAngle / 2);
float qw = cos(rAngle / 2);
glm::quat q = glm::quat(qw, qx, qy, qz);
cam.right = q * cam.up;
theta = phi = 0;
cammove = glm::vec3();
camchanged = false;
}
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
if (iteration == 0) {
pathtraceFree();
pathtraceInit(scene);
}
if (iteration < renderState->iterations) {
uchar4 *pbo_dptr = NULL;
iteration++;
cudaGLMapBufferObject((void**)&pbo_dptr, pbo);
// execute the kernel
int frame = 0;
pathtrace(pbo_dptr, frame, iteration);
// unmap buffer object
cudaGLUnmapBufferObject(pbo);
} else {
saveImage();
pathtraceFree();
cudaDeviceReset();
exit(EXIT_SUCCESS);
}
}
void runCuda()
{
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
float *dptrvert=NULL;
float *velptr=NULL;
cudaGLMapBufferObject((void**)&dptrvert, planetVBO);
cudaGLMapBufferObject((void**)&velptr, velocityVBO);
// execute the kernel
cudaFlockingUpdateWrapper(DT, seekTarget);
#if VISUALIZE == 1
cudaUpdateVBO(dptrvert, velptr);
#endif
// unmap buffer object
cudaGLUnmapBufferObject(planetVBO);
cudaGLUnmapBufferObject(velocityVBO);
}
void Mandelbrot::WriteBuffer() {
checkCudaErrors( cudaGLMapBufferObject( ( void** ) &this->devArray, this->buffer ), __LINE__, false );
cudaMemcpy( this->devArray, this->devCalcArray, this->iSize, cudaMemcpyDeviceToDevice );
checkCudaErrors( cudaGLUnmapBufferObject( this->buffer ), __LINE__, false );
this->bIsFlushed = true;
}
void Renderer::render(const Camera& camera, float time) {
// calc cam vars
glm::vec3 A,B,C;
{
// camera ray
C = glm::normalize(camera.getLookAt()-camera.getPosition());
// calc A (screen x)
// calc B (screen y) then scale down relative to aspect
// fov is for screen x axis
A = glm::normalize(glm::cross(C,camera.getUp()));
B = 1.0f/camera.getAspect()*glm::normalize(glm::cross(A,C));
// scale by FOV
float tanFOV = tan(glm::radians(camera.getFOV()));
A *= tanFOV;
B *= tanFOV;
}
// cuda call
unsigned int* out_data;
checkCudaErrors(cudaGLMapBufferObject((void**)&out_data, pbo));
if (mode == RAYTRACE) {
raytrace1(out_data, image_width, image_height, time,
camera.getPosition(), A, B, C,
scene_d, sceneSize);
}
else if (mode == PATHTRACE) {
++filmIters;
pathtrace(out_data, image_width, image_height, time,
camera.getPosition(), A, B, C,
camera.m_lensRadius, camera.m_focalDist,
scene_d, sceneSize,
rand_d, rays_d, col_d, idx_d,
film_d, filmIters);
}
checkCudaErrors(cudaGLUnmapBufferObject(pbo));
// download texture from destination PBO
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo);
glActiveTexture(GL_TEXTURE0 + RENDER_TEXTURE);
glBindTexture(GL_TEXTURE_2D, result_texture);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_BGRA, GL_UNSIGNED_BYTE, NULL);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glActiveTexture(GL_TEXTURE0 + UNUSED_TEXTURE);
SDK_CHECK_ERROR_GL();
fullScreenQuad.display();
}
void MappedVboVoid::
unmap(DataStorageVoid* datap)
{
if (_is_mapped)
{
TIME_MAPPEDVBOVOID TaskInfo("Unmapping vbo %u of size %s", (unsigned)*_vbo, DataStorageVoid::getMemorySizeText(datap->numberOfBytes()).c_str());
#ifdef USE_CUDA
// make sure data is located in cuda
datap->AccessStorage<CudaGlobalStorage>( true, false );
#ifdef CUDA_MEMCHECK_TEST
// copy data back over the mapped memory
*mapped_gl_mem = *datap;
#endif
// sync from cuda to vbo
cudaGLUnmapBufferObject(*_vbo);
// release resources
mapped_gl_mem.reset();
datap->DiscardAllData();
_is_mapped = false;
// The memory bound with Cuda-OpenGL-interop can be relied on. So
// call cudaGetLastError to clear the cuda error state just in case.
// (I'm not sure why but it might be related to cuda out-of-memory
// errors elsewhere)
cudaGetLastError();
#else
// make sure data is located in cpu
datap->AccessStorage<CpuMemoryStorage>( true, false );
// sync from mem to vbo
glBindBuffer(_vbo->vbo_type(), *_vbo);
glUnmapBuffer(_vbo->vbo_type());
glBindBuffer(_vbo->vbo_type(), 0);
// release resources
mapped_gl_mem.reset();
datap->DiscardAllData();
_is_mapped = false;
#endif
TIME_MAPPEDVBOVOID ComputationSynchronize();
if (_tt)
{
TaskInfo("Unmapped vbo %u of size %s", (unsigned)*_vbo, DataStorageVoid::getMemorySizeText(datap->numberOfBytes()).c_str());
delete _tt;
_tt = 0;
}
}
}
void runCuda()
{
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
float4 *dptr=NULL;
float *dptrvert=NULL;
cudaGLMapBufferObject((void**)&dptr, pbo);
cudaGLMapBufferObject((void**)&dptrvert, planetVBO);
// execute the kernel
cudaNBodyUpdateWrapper(DT);
#if VISUALIZE == 1
cudaUpdatePBO(dptr, field_width, field_height);
cudaUpdateVBO(dptrvert, field_width, field_height);
#endif
// unmap buffer object
cudaGLUnmapBufferObject(planetVBO);
cudaGLUnmapBufferObject(pbo);
}
static void prepare_image_for_mapping(Data_Obj *dp)
{
#ifdef HAVE_OPENGL
int t;
cudaError_t e;
// unmap buffer before using w/ GL
if( BUF_IS_MAPPED(dp) ){
e = cudaGLUnmapBufferObject( OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
describe_cuda_driver_error2("update_cuda_viewer",
"cudaGLUnmapBufferObject",e);
NERROR1("failed to unmap buffer object");
}
CLEAR_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
}
//
//bind_texture(OBJ_DATA_PTR(dp));
glClear(GL_COLOR_BUFFER_BIT);
/*
sprintf(ERROR_STRING,"update_cuda_viewer: tex_id = %d, buf_id = %d",
OBJ_TEX_ID(dp),OBJ_BUF_ID(dp));
advise(ERROR_STRING);
*/
glBindTexture(GL_TEXTURE_2D, OBJ_TEX_ID(dp));
#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, OBJ_BUF_ID(dp));
#endif // HAVE_LIBGLEW
#ifdef FOOBAR
switch(OBJ_COMPS(dp)){
/* what used to be here??? */
}
#endif /* FOOBAR */
t=gl_pixel_type(dp);
glTexSubImage2D(GL_TEXTURE_2D, 0, // target, level
0, 0, // x0, y0
OBJ_COLS(dp), OBJ_ROWS(dp), // dx, dy
t,
GL_UNSIGNED_BYTE, // type
OFFSET(0)); // offset into PIXEL_UNPACK_BUFFER
#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
#endif // HAVE_LIBGLEW
}
//====================================
// Main loop
//====================================
void RunCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
uchar4 *dptr = NULL;
cudaGLMapBufferObject((void**)&dptr, m_pbo);
// Execute the kernel
CudaKernel(dptr, m_width, m_height, m_major, m_minor);
// Unmap buffer object
cudaGLUnmapBufferObject(m_pbo);
}
//-------------------------------
//---------RUNTIME STUFF---------
//-------------------------------
void runCuda() {
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
dptr = NULL;
cudaGLMapBufferObject((void **)&dptr, pbo);
rasterize(dptr);
cudaGLUnmapBufferObject(pbo);
frame++;
fpstracker++;
}
// This is the normal display path
void display(void)
{
cutilCheckError(cutStartTimer(timer));
// Sobel operation
Pixel *data = NULL;
cutilSafeCall(cudaGLMapBufferObject((void**)&data, pbo_buffer));
sobelFilter(data, imWidth, imHeight, g_SobelDisplayMode, imageScale );
cutilSafeCall(cudaGLUnmapBufferObject(pbo_buffer));
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();
}
// display results using OpenGL
void display()
{
cutilCheckError(cutStartTimer(timer));
// execute filter, writing results to pbo
unsigned int *d_result;
cutilSafeCall(cudaGLMapBufferObject((void**)&d_result, pbo));
gaussianFilterRGBA(d_img, d_result, d_temp, width, height, sigma, order, nthreads);
cutilSafeCall(cudaGLUnmapBufferObject(pbo));
// load texture from pbo
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
glBindTexture(GL_TEXTURE_2D, texid);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, 0);
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
// display results
glClear(GL_COLOR_BUFFER_BIT);
glEnable(GL_TEXTURE_2D);
glDisable(GL_DEPTH_TEST);
glBegin(GL_QUADS);
glTexCoord2f(0, 1); glVertex2f(0, 0);
glTexCoord2f(1, 1); glVertex2f(1, 0);
glTexCoord2f(1, 0); glVertex2f(1, 1);
glTexCoord2f(0, 0); glVertex2f(0, 1);
glEnd();
glDisable(GL_TEXTURE_2D);
if (g_CheckRender && g_CheckRender->IsQAReadback() && g_Verify) {
// readback for QA testing
printf("> (Frame %d) Readback BackBuffer\n", frameCount);
g_CheckRender->readback( width, height );
g_CheckRender->savePPM(sOriginal[g_Index], true, NULL);
if (!g_CheckRender->PPMvsPPM(sOriginal[g_Index], sReference[g_Index], MAX_EPSILON_ERROR, 0.50f )) {
g_TotalErrors++;
}
g_Verify = false;
}
glutSwapBuffers();
cutilCheckError(cutStopTimer(timer));
computeFPS();
}
void initCuda(){
// Use device with highest Gflops/s
cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() );
initPBO(&pbo);
dptr=NULL;
cudaGLMapBufferObject((void**)&dptr, pbo);
clearPBOpos(dptr,width,height);
cudaGLUnmapBufferObject(pbo);
// Clean up on program exit
atexit(cleanupCuda);
SetScissorWindow(glm::vec4(300,300,500,500));
texture.mapptr = stbi_load("cow.jpeg",&texture.width, &texture.height,&texture.depth,0);
runCuda();
}
// Run the Cuda part of the computation
void runCuda()
{
uchar4 *dptr=NULL;
// map OpenGL buffer object for writing from CUDA on a single GPU
// no data is moved (Win & Linux). When mapped to CUDA, OpenGL
// should not use this buffer
cudaGLMapBufferObject((void**)&dptr, pbo);
// execute the kernel
launch_kernel(dptr, image_width, image_height, animTime);
// unmap buffer object
cudaGLUnmapBufferObject(pbo);
}
void runCuda() {
if (camchanged) {
iteration = 0;
Camera &cam = renderState->camera;
glm::vec3 v = cam.view;
glm::vec3 u = cam.up;
glm::vec3 r = glm::cross(v, u);
glm::mat4 rotmat = glm::rotate(theta, r) * glm::rotate(phi, u);
cam.view = glm::vec3(rotmat * glm::vec4(v, 0.f));
cam.up = glm::vec3(rotmat * glm::vec4(u, 0.f));
cam.position += cammove.x * r + cammove.y * u + cammove.z * v;
theta = phi = 0;
cammove = glm::vec3();
camchanged = false;
}
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
if (iteration == 0) {
pathtraceFree();
pathtraceInit(scene);
}
if (iteration < renderState->iterations) {
uchar4 *pbo_dptr = NULL;
iteration++;
cudaGLMapBufferObject((void**)&pbo_dptr, pbo);
// execute the kernel
if( scene->blur ) {
for( int frame = 0; frame<(scene->frames); frame++ ){
pathtrace(pbo_dptr, frame, scene->frames, iteration);
//printf("frame is : %d", frame);
}
} else {
pathtrace(pbo_dptr, 0, 0, iteration);
}
// unmap buffer object
cudaGLUnmapBufferObject(pbo);
} else {
saveImage();
pathtraceFree();
cudaDeviceReset();
exit(EXIT_SUCCESS);
}
}
////////////////////////////////////////////////////////////////////////////////
//! Run the Cuda part of the computation
////////////////////////////////////////////////////////////////////////////////
void runCuda(GLuint vbo)
{
// map OpenGL buffer object for writing from CUDA
float4 *dptr;
cutilSafeCall(cudaGLMapBufferObject((void**)&dptr, vbo));
// execute the kernel
// dim3 block(8, 8, 1);
// dim3 grid(mesh_width / block.x, mesh_height / block.y, 1);
// kernel<<< grid, block>>>(dptr, mesh_width, mesh_height, anim);
launch_kernel(dptr, mesh_width, mesh_height, anim);
// unmap buffer object
cutilSafeCall(cudaGLUnmapBufferObject(vbo));
}
//====================================
// Main loop
//====================================
void runCUDA() {
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not
// use this buffer
float4 *dptr = NULL;
float *dptrvert = NULL;
cudaGLMapBufferObject((void**)&dptrvert, planetVBO);
// execute the kernel
Nbody::stepSimulation(DT);
#if VISUALIZE
Nbody::copyPlanetsToVBO(dptrvert);
#endif
// unmap buffer object
cudaGLUnmapBufferObject(planetVBO);
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
dptr=NULL;
vbo = mesh->getVBO();
vbosize = mesh->getVBOsize();
float newcbo[] = {0.0, 1.0, 0.0,
0.0, 0.0, 1.0,
1.0, 0.0, 0.0};
cbo = newcbo;
cbosize = 9;
ibo = mesh->getIBO();
ibosize = mesh->getIBOsize();
nbo = mesh->getNBO();
nbosize = mesh->getNBOsize();
// Update view and model to projection transform matrices in each step when interacting with keyboard or mouse
*view = glm::lookAt(cam.position, glm::vec3(0.0f), cam.up);
*transformModel2Projection = utilityCore::glmMat4ToCudaMat4(*projection * *view * *model);
viewPort = glm::normalize(utilityCore::multiplyMat(utilityCore::glmMat4ToCudaMat4(*projection * *view), glm::vec4(cam.view, 1.0f)));
// Transformation Feedback
std::cout << "\n The model-view-projection transformation is:" << std::endl;
utilityCore::printMat4(*projection * *view * *model);
std::cout << "\n The view port in the clip space is:" << std::endl;
utilityCore::printVec3(viewPort);
cudaGLMapBufferObject((void**)&dptr, pbo);
cudaRasterizeCore(dptr, glm::vec2(width, height), frame, vbo, vbosize, cbo, cbosize, ibo, ibosize, nbo, nbosize, transformModel2Projection, viewPort, antialiasing, depthFlag, flatcolorFlag, color, multicolorFlag);
cudaGLUnmapBufferObject(pbo);
vbo = NULL;
cbo = NULL;
ibo = NULL;
nbo = NULL;
frame++;
fpstracker++;
}
static int cu2_unmap_buf(QSP_ARG_DECL Data_Obj *dp)
{
#ifdef HAVE_OPENGL
cudaError_t e;
e = cudaGLUnmapBufferObject( OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
describe_cuda_driver_error2("cu2_unmap_buf",
"cudaGLUnmapBufferObject",e);
ERROR1("failed to unmap buffer object");
return -1;
}
return 0;
#else // ! HAVE_OPENGL
WARN("cu2_unmap_buf: Sorry, no OpenGL support in this build!?");
return -1;
#endif // ! HAVE_OPENGL
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
vbo = mesh->getVBO();
vbosize = mesh->getVBOsize();
nbo = mesh->getNBO();
nbosize = mesh->getNBOsize();
float newcbo[] = {0.0, 1.0, 0.0,
0.0, 0.0, 1.0,
1.0, 0.0, 0.0};
cbo = newcbo;
cbosize = 9;
ibo = mesh->getIBO();
ibosize = mesh->getIBOsize();
calcuatetransformationMatrix( eye,glm::vec2(width, height), front, back);
dptr=NULL;
cudaGLMapBufferObject((void**)&dptr, pbo);
if(ReadBlendType() == ADD)
{
drawTexture(dptr,width, height,texture);
}
//clearPBOpos(dptr,width,height);
cudaRasterizeCore(dptr, glm::vec2(width, height), frame, vbo, vbosize, nbo, nbosize, cbo, cbosize, ibo, ibosize);
cudaGLUnmapBufferObject(pbo);
vbo = NULL;
cbo = NULL;
ibo = NULL;
frame++;
fpstracker++;
}
// display results using OpenGL
void display()
{
sdkStartTimer(&timer);
// execute filter, writing results to pbo
unsigned int *d_result;
checkCudaErrors(cudaGLMapBufferObject((void **)&d_result, pbo));
gaussianFilterRGBA(d_img, d_result, d_temp, width, height, sigma, order, nthreads);
checkCudaErrors(cudaGLUnmapBufferObject(pbo));
// load texture from pbo
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
glBindTexture(GL_TEXTURE_2D, texid);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, 0);
glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
// display results
glClear(GL_COLOR_BUFFER_BIT);
glEnable(GL_TEXTURE_2D);
glDisable(GL_DEPTH_TEST);
glBegin(GL_QUADS);
glTexCoord2f(0, 1);
glVertex2f(0, 0);
glTexCoord2f(1, 1);
glVertex2f(1, 0);
glTexCoord2f(1, 0);
glVertex2f(1, 1);
glTexCoord2f(0, 0);
glVertex2f(0, 1);
glEnd();
glDisable(GL_TEXTURE_2D);
glutSwapBuffers();
sdkStopTimer(&timer);
computeFPS();
}
static void select_reference_color(int x, int y) {
static float offset = SCREEN_WIDTH/1280.0f;
printf("(After scaling) Handling click on x = %d, y = %d\n", x, y);
GLuint left_buffer, right_buffer, buffer;
kb_images_current_buffers(&left_buffer, &right_buffer);
printf("left texture id = %d, right = %d\n", left_buffer, right_buffer);
uchar4 *gpu_buffer = NULL;
if (x < 640*offset) {
/* The clicked pixel is in the left image */
buffer = left_buffer;
} else {
/* The clicked pixel is in the right image */
buffer = right_buffer;
x -= 640*offset;
}
cutilSafeCall(cudaGLMapBufferObject((void**)&gpu_buffer, buffer));
uchar4 pixel;
cudaMemcpy(&pixel, gpu_buffer + (y * 640) + x, sizeof(uchar4), cudaMemcpyDeviceToHost);
printf("pixel-value: %d, %d, %d (%d)\n", pixel.x, pixel.y, pixel.z, pixel.w);
static char rgbbuffer[4096];
snprintf(rgbbuffer, sizeof(rgbbuffer), "%d,%d,%d", pixel.x, pixel.y, pixel.z);
kb_ui_call_javascript("SetRGB", rgbbuffer);
double r = pixel.z;
double g = pixel.y;
double b = pixel.x;
double nominator = sqrt((r * r) + (g * g) + (b * b));
printf("nominator = %f\n", nominator);
reference_color.x = r / nominator;
reference_color.y = g / nominator;
reference_color.z = b / nominator;
printf("reference_color final: %f %f %f\n",
reference_color.x, reference_color.y, reference_color.z);
cutilSafeCall(cudaGLUnmapBufferObject(buffer));
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
if(iterations<renderCam->iterations){
uchar4 *dptr=NULL;
iterations++;
cudaGLMapBufferObject((void**)&dptr, pbo);
//pack geom and material arrays
geom* geoms = new geom[renderScene->objects.size()];
material* materials = new material[renderScene->materials.size()];
map* maps = new map[renderScene->maps.size()];
for(int i=0; i<renderScene->objects.size(); i++){
geoms[i] = renderScene->objects[i];
}
for(int i=0; i<renderScene->materials.size(); i++){
materials[i] = renderScene->materials[i];
}
for(int i=0; i<renderScene->maps.size(); i++){
maps[i] = renderScene->maps[i];
}
// execute the kernel
if(!textureMode)
cudaRaytraceCore(dptr, renderCam, targetFrame, iterations, materials, renderScene->materials.size(),maps,renderScene->maps.size(), geoms, renderScene->objects.size(), mblur,dof);
else
cudaRaytraceCoreT(dptr, renderCam, targetFrame, iterations, materials, renderScene->materials.size(),maps,renderScene->maps.size(), geoms, renderScene->objects.size(), mblur,dof);
// unmap buffer object
cudaGLUnmapBufferObject(pbo);
}else{
if(!finishedRender){
//output image file
image outputImage(renderCam->resolution.x, renderCam->resolution.y);
image depthImage(renderCam->resolution.x, renderCam->resolution.y);
for(int x=0; x<renderCam->resolution.x; x++){
for(int y=0; y<renderCam->resolution.y; y++){
int index = x + (y * renderCam->resolution.x);
glm::vec3 justRGB(renderCam->image[index].x,renderCam->image[index].y,renderCam->image[index].z);
outputImage.writePixelRGB(renderCam->resolution.x-1-x,y,justRGB);
float d = abs(renderCam->image[index].w-renderCam->positions[targetFrame].z)/40.0f;
depthImage.writePixelRGB(renderCam->resolution.x-1-x,y, glm::vec3(d,d,d));
}
}
gammaSettings gamma;
gamma.applyGamma = true;
gamma.gamma = 1.0/2.2;
gamma.divisor = renderCam->iterations;
outputImage.setGammaSettings(gamma);
string filename = renderCam->imageName;
string s;
stringstream out;
out << targetFrame;
s = out.str();
utilityCore::replaceString(filename, ".bmp", "."+s+".bmp");
utilityCore::replaceString(filename, ".png", "."+s+".png");
outputImage.saveImageRGB(filename);
depthImage.saveImageRGB("depth."+s+".bmp");
cout << "Saved frame " << s << " to " << filename << endl;
finishedRender = true;
if(singleFrameMode==true){
//cudaDeviceReset();
exit(0);
}
}
if(targetFrame<renderCam->frames-1){
//clear image buffer and move onto next frame
targetFrame++;
iterations = 0;
for(int i=0; i<renderCam->resolution.x*renderCam->resolution.y; i++){
renderCam->image[i] = glm::vec4(0,0,0,-1);
}
//cudaDeviceReset();
finishedRender = false;
}
}
}
void runCuda(){
// Map OpenGL buffer object for writing from CUDA on a single GPU
// No data is moved (Win & Linux). When mapped to CUDA, OpenGL should not use this buffer
if((unsigned int)iterations < renderCam->iterations){
uchar4 *dptr=NULL;
++iterations;
cudaGLMapBufferObject((void**)&dptr, pbo);
//pack geom and material arrays
unsigned int objectsSize = renderScene->objects.size(), materialsSize = renderScene->materials.size(), lightsSize = renderScene->lights.size();
geom* geoms = new geom[objectsSize];
material* materials = new material[materialsSize];
light* lights = new light[lightsSize];
for(unsigned int i=0; i< objectsSize; ++i){
geoms[i] = renderScene->objects[i];
}
for(unsigned int i=0; i< materialsSize; ++i){
materials[i] = renderScene->materials[i];
}
for(unsigned int i=0; i< lightsSize; ++i){
lights[i] = renderScene->lights[i];
}
// execute the kernel
cudaRaytraceCore(dptr, renderCam, targetFrame, iterations, materials, materialsSize, geoms, objectsSize, lights, lightsSize);
// unmap buffer object
cudaGLUnmapBufferObject(pbo);
}else{
if(!finishedRender){
//output image file
image outputImage(renderCam->resolution.x, renderCam->resolution.y);
for(int x=0; x<renderCam->resolution.x; ++x){
for(int y=0; y<renderCam->resolution.y; ++y){
int index = x + (y * renderCam->resolution.x);
outputImage.writePixelRGB(x,y,renderCam->image[index]);
}
}
gammaSettings gamma;
gamma.applyGamma = true;
gamma.gamma = 1.0/2.2;
gamma.divisor = renderCam->iterations;
outputImage.setGammaSettings(gamma);
string filename = renderCam->imageName;
string s;
stringstream out;
out << targetFrame;
s = out.str();
utilityCore::replaceString(filename, ".bmp", "."+s+".bmp");
utilityCore::replaceString(filename, ".png", "."+s+".png");
outputImage.saveImageRGB(filename);
cout << "Saved frame " << s << " to " << filename << endl;
finishedRender = true;
if(singleFrameMode==true){
cudaDeviceReset();
exit(0);
}
}
if(targetFrame < renderCam->frames - 1){
//clear image buffer and move onto next frame
++targetFrame;
iterations = 0;
for(int i=0; i<renderCam->resolution.x*renderCam->resolution.y; ++i){
renderCam->image[i] = glm::vec3(0,0,0);
}
cudaDeviceReset();
finishedRender = false;
}
}
}
// This is the normal display path
static void update_pf_viewer(QSP_ARG_DECL Platform_Viewer *pvp, Data_Obj *dp)
{
#ifdef HAVE_OPENGL
int t;
//cudaError_t e;
// unmap buffer before using w/ GL
if( BUF_IS_MAPPED(dp) ){
if( (*PF_UNMAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))
(QSP_ARG dp) < 0 ) {
WARN("update_pf_viewer: buffer unmap error!?");
}
#ifdef FOOBAR
e = cudaGLUnmapBufferObject( OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
describe_cuda_driver_error2("update_pf_viewer",
"cudaGLUnmapBufferObject",e);
NERROR1("failed to unmap buffer object");
}
#endif // FOOBAR
CLEAR_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
}
//
//bind_texture(OBJ_DATA_PTR(dp));
glClear(GL_COLOR_BUFFER_BIT);
/*
sprintf(ERROR_STRING,"update_pf_viewer: tex_id = %d, buf_id = %d",
OBJ_TEX_ID(dp),OBJ_BUF_ID(dp));
advise(ERROR_STRING);
*/
glBindTexture(GL_TEXTURE_2D, OBJ_TEX_ID(dp));
// is glBindBuffer REALLY part of libGLEW???
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, OBJ_BUF_ID(dp));
//#endif // HAVE_LIBGLEW
#ifdef FOOBAR
switch(OBJ_COMPS(dp)){
/* what used to be here??? */
}
#endif /* FOOBAR */
t=gl_pixel_type(dp);
glTexSubImage2D(GL_TEXTURE_2D, 0, // target, level
0, 0, // x0, y0
OBJ_COLS(dp), OBJ_ROWS(dp), // dx, dy
t,
GL_UNSIGNED_BYTE, // type
OFFSET(0)); // offset into PIXEL_UNPACK_BUFFER
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
//#endif // HAVE_LIBGLEW
glBegin(GL_QUADS);
glTexCoord2f(0, 1); glVertex2f(-1.0, -1.0);
glTexCoord2f(0, 0); glVertex2f(-1.0, 1.0);
glTexCoord2f(1, 0); glVertex2f(1.0, 1.0);
glTexCoord2f(1, 1); glVertex2f(1.0, -1.0);
glEnd();
glBindTexture(GL_TEXTURE_2D, 0);
#ifdef FOOBAR
e = cudaGLMapBufferObject( &OBJ_DATA_PTR(dp), OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
WARN("Error mapping buffer object!?");
// should we return now, with possibly other cleanup???
}
#endif // FOOBAR
if( (*PF_MAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))(QSP_ARG dp) < 0 ){
WARN("update_pf_viewer: Error mapping buffer!?");
}
SET_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
#else // ! HAVE_OPENGL
NO_OGL_MSG
#endif // ! HAVE_OPENGL
}
void CMarchingCubes::ComputeIsosurface(ElemType* _pFval, ElemType _isoValue, RenderData* _pRender)
{
int threads = 128;
dim3 grid(m_NumVoxels / threads, 1, 1);
// get around maximum grid size of 65535 in each dimension
if (grid.x > 65535) {
grid.y = grid.x / 32768;
grid.x = 32768;
}
uint totalVerts = 0;
int size = m_GridSize.x * m_GridSize.y * m_GridSize.z * sizeof(float);
//////////////////////////////////////////////////////////////////////////
int len = m_GridSize.x * m_GridSize.y * m_GridSize.z;
float *pFvalTemp = new float[len];
for (int i = 0; i < len; i++)
{
pFvalTemp[i] = _pFval[i];
}
//////////////////////////////////////////////////////////////////////////
float* pdVolumeFval; // ¶¥µãº¯ÊýÖµÎÆÀí(n¡¡Surface)
cutilSafeCall(cudaMalloc((void**) &pdVolumeFval, size));
cutilSafeCall(cudaMemcpy(pdVolumeFval, pFvalTemp, size, cudaMemcpyHostToDevice) );
bindVolumeValTexture(pdVolumeFval);
delete []pFvalTemp;
// calculate number of vertices need per voxel
launch_classifyVoxel(grid, threads,
m_pdVoxelVerts, m_pdVoxelOccupied, pdVolumeFval,
m_GridSize, m_NumVoxels, _isoValue);
#if DEBUG_BUFFERS
printf("voxelVerts:\n");
dumpBuffer(m_pdVoxelVerts, m_NumVoxels);
#endif
#if SKIP_EMPTY_VOXELS
// scan voxel occupied array
cudppScan(m_Scanplan, m_pdVoxelOccupiedScan, m_pdVoxelOccupied, m_NumVoxels);
#if DEBUG_BUFFERS
printf("voxelOccupiedScan:\n");
dumpBuffer(m_pdVoxelOccupiedScan, m_NumVoxels);
#endif
// read back values to calculate total number of non-empty voxels
// since we are using an exclusive scan, the total is the last value of
// the scan result plus the last value in the input array
{
uint lastElement, lastScanElement;
cutilSafeCall(cudaMemcpy((void *) &lastElement,
(void *) (m_pdVoxelOccupied + m_NumVoxels - 1),
sizeof(uint), cudaMemcpyDeviceToHost));
cutilSafeCall(cudaMemcpy((void *) &lastScanElement,
(void *) (m_pdVoxelOccupiedScan + m_NumVoxels - 1),
sizeof(uint), cudaMemcpyDeviceToHost));
m_ActiveVoxels = lastElement + lastScanElement;
}
if (0 == m_ActiveVoxels) {
// return if there are no full voxels
totalVerts = 0;
return;
}
// compact voxel index array
launch_compactVoxels(grid, threads, m_pdCompactedVoxelArray, m_pdVoxelOccupied, m_pdVoxelOccupiedScan, m_NumVoxels);
cutilCheckMsg("compactVoxels failed");
#endif // SKIP_EMPTY_VOXELS
// scan voxel vertex count array
cudppScan(m_Scanplan, m_pdVoxelVertsScan, m_pdVoxelVerts, m_NumVoxels);
#if DEBUG_BUFFERS
printf("voxelVertsScan:\n");
dumpBuffer(m_pdVoxelVertsScan, m_NumVoxels);
#endif
// readback total number of vertices
{
uint lastElement, lastScanElement;
cutilSafeCall(cudaMemcpy((void *) &lastElement,
(void *) (m_pdVoxelVerts + m_NumVoxels - 1),
sizeof(uint), cudaMemcpyDeviceToHost));
cutilSafeCall(cudaMemcpy((void *) &lastScanElement,
(void *) (m_pdVoxelVertsScan + m_NumVoxels - 1),
sizeof(uint), cudaMemcpyDeviceToHost));
totalVerts = lastElement + lastScanElement;
}
// create VBOs
GLuint posVbo, normalVbo;
createVBO(&posVbo, totalVerts * sizeof(float) * 4);
cutilSafeCall(cudaGLRegisterBufferObject(posVbo));
createVBO(&normalVbo, totalVerts * sizeof(float) * 4);
cutilSafeCall(cudaGLRegisterBufferObject(normalVbo));
// generate triangles, writing to vertex buffers
float4 *d_pos = 0, *d_normal = 0;
cutilSafeCall(cudaGLMapBufferObject((void**)&d_pos, posVbo));
cutilSafeCall(cudaGLMapBufferObject((void**)&d_normal, normalVbo));
#if SKIP_EMPTY_VOXELS
dim3 grid2((int) ceil(m_ActiveVoxels / (float) NTHREADS), 1, 1);
#else
dim3 grid2((int) ceil(m_NumVoxels / (float) NTHREADS), 1, 1);
#endif
while(grid2.x > 65535) {
grid2.x/=2;
grid2.y*=2;
}
launch_generateTriangles(grid2, NTHREADS, d_pos, d_normal,
m_pdCompactedVoxelArray, m_pdVoxelVertsScan,
m_pdVolume, pdVolumeFval,
m_GridSize, _isoValue,
m_ActiveVoxels, m_MaxVerts);
cutilSafeCall(cudaGLUnmapBufferObject(normalVbo));
cutilSafeCall(cudaGLUnmapBufferObject(posVbo));
_pRender->posVbo = posVbo;
_pRender->normalVbo = normalVbo;
_pRender->totalVerts = totalVerts;
cutilSafeCall(cudaFree(pdVolumeFval));
}