I am trying to draw a bunch of points on the screen. I'm using CUDA to generate the data (position and color), and OpenGL to draw it. I am trying to get CUDA to update a VBO and then OpenGL to draw it, but I get a blank screen. I am not sure if CUDA is not able to update the buffer, or that the buffer is not drawing properly. My GPU is a GTX 1080, and I'm trying to use OpenGL 4.0. Colors are specified by CUDA as well. If my problem is that I need a shader, how do I add that, but also still specify the color through CUDA?
UPDATE: problem seems to be openGL. Updated code to use triangle So new question to add. Why is my VBO not being rendered?
Here is the code:
GPUmain.cuh:
#include <cuda_runtime.h>
#include "device_launch_parameters.h"
#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
#include <thrust/remove.h>
#include <curand.h>
#include <GL/glew.h>
#include <SDL_opengl.h>
#include <cuda_gl_interop.h>
#define BUFFER_OFFSET(i) ((char *)NULL + (i))
//ver: x, y, z, r, g, b, a
struct ver {
// x, y, z pos
GLuint x, y, z;
// r, g, b, a color
GLubyte r, g, b, a;
};
class GPU {
public:
static int nParticles;
static GLuint vboid;
static cudaGraphicsResource *CGR;
//collection of vertices to be simulated and rendered
static thrust::device_vector<ver> rverts;
static void init(int w, int h);
static void compute();
static void render();
static void GPUmain();
static void free();
};
GPUmain.cu:
#include "GPUmain.cuh"
__global__ void uploadVerts(ver *vv, ver *vb) {
int id = threadIdx.x + (blockDim.x * blockIdx.x);
vb[id] = vv[id];
vb[id].x = vv[id].x;
vb[id].y = vv[id].y;
vb[id].z = vv[id].z;
vb[id].r = vv[id].r;
vb[id].g = vv[id].g;
vb[id].b = vv[id].b;
vb[id].a = vv[id].a;
}
__global__ void genGrid(ver *v) {
int i = threadIdx.x + (blockDim.x * blockIdx.x);
float x = (float)(i % ((int)1080));
float y = (float)(i / ((int)1920));
v[i].x = x;
v[i].y = y;
v[i].z = 1;
v[i].r = 255;
v[i].g = 0;
v[i].b = 0;
v[i].a = 0;
}
int GPU::nParticles;
GLuint GPU::vboid;
cudaGraphicsResource *GPU::CGR;
//collection of vertices to be simulated and rendered
thrust::device_vector<ver> GPU::rverts;
void GPU::init(int w, int h)
{
nParticles = w * h;
/*rverts.resize(nParticles, ver{0,0,0,0,0,0,0});
genGrid<<<nParticles/1024,1024>>>(thrust::raw_pointer_cast(&rverts[0]));*/
ver e[3] = {
ver{1024,200,2,255,0,0,255},
ver{499,288,173,0,255,0,255},
ver{462,1674,8,0,0,255,255}
};
glGenBuffers(1,&vboid);
glBindBuffer(GL_ARRAY_BUFFER,vboid);
glBufferData(GL_ARRAY_BUFFER,3*sizeof(ver),e,GL_DYNAMIC_DRAW);
glBindBuffer(GL_ARRAY_BUFFER, 0);
/*cudaGraphicsGLRegisterBuffer(&CGR,vboid,cudaGraphicsMapFlagsWriteDiscard);*/
}
void GPU::compute()
{
}
void GPU::render()
{
/*ver *verts;
size_t size;
cudaGraphicsMapResources(1, &CGR, 0);
cudaGraphicsResourceGetMappedPointer((void**)&verts, &size, CGR);
uploadVerts<<<nParticles/1024, 1024>>>(thrust::raw_pointer_cast(&rverts[0]), verts);
cudaGraphicsUnmapResources(1, &CGR, 0);
cudaDeviceSynchronize();*/
glClearColor(0, 0, 0, 0); // we clear the screen with black (else, frames would overlay...)
glClear(GL_COLOR_BUFFER_BIT); // clear the buffer
glBindBuffer(GL_ARRAY_BUFFER, vboid);
glEnableClientState(GL_VERTEX_ARRAY);
glEnableClientState(GL_COLOR_ARRAY);
glVertexPointer(3, GL_INT, 4 * sizeof(GLubyte), 0);
glColorPointer(4, GL_BYTE, 3 * sizeof(GLuint), BUFFER_OFFSET(3 * sizeof(GLuint)));
glDrawArrays(GL_TRIANGLES, 0, 3);
glDisableClientState(GL_VERTEX_ARRAY);
glDisableClientState(GL_COLOR_ARRAY);
glBindBuffer(GL_ARRAY_BUFFER, 0);
}
void GPU::GPUmain()
{
compute();
render();
}
void GPU::free()
{
cudaGraphicsUnregisterResource(CGR);
glBindBuffer(GL_ARRAY_BUFFER, vboid);
glDeleteBuffers(1, &vboid);
glBindBuffer(GL_ARRAY_BUFFER, 0);
rverts.clear();
thrust::device_vector<ver>().swap(rverts);
}
The relevant (that contain OpenGL code) parts of window.cpp:
bool Window::init()
{
//initialize SDL
if (SDL_Init(SDL_INIT_EVERYTHING) != 0) {
log << "Failed to initialize SDL!\n";
return false;
}
//set window atributes
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 4);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 0);
SDL_GL_SetAttribute(SDL_GL_STENCIL_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_DOUBLEBUFFER, 1);
//creat window
window = SDL_CreateWindow(
name.c_str(),
SDL_WINDOWPOS_CENTERED,
SDL_WINDOWPOS_CENTERED,
width,
height,
SDL_WINDOW_OPENGL
);
//create opengl context in the window
glcontext = SDL_GL_CreateContext(window);
SDL_GL_SetSwapInterval(1);
//check if the window was created
if (window == nullptr) {
log << "Failed to create window!\n";
return false;
}
//turn on experimental features
glewExperimental = GL_TRUE;
//initiallize glew
if (glewInit() != GLEW_OK) {
log << "Failed to Init GLEW";
return false;
}
//set drawing parameters
glViewport(0, 0, width, height);
glMatrixMode(GL_PROJECTION);
glLoadIdentity();
glOrtho(0, width, 0, height, 0, 255);
glPointSize(1);
glEnable(GL_BLEND); // Allow Transparency
glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); // how transparency acts
std::cout << sizeof(ver);
GPU::init(width, height);
return true;
}
void Window::renderFrame()
{
GPU::render();
SDL_GL_SwapWindow(window); //swap buffers
}
If you use the fixed-function attributes and client side capabilities, then you've to use a compatibility profile context.
See Fixed Function Pipeline and Legacy OpenGL.
If you want to use a core profile, then you've to use Vertex Array Object and Shader:
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_COMPATIBILITY);
The following geometry
ver e[3] = { // x y z r g b a ver{1024, 200, 2, 255, 0, 0, 255}, ver{ 499, 288, 173, 0, 255, 0, 255}, ver{462, 1674, 8, 0, 0, 255, 255} };
is clipped by the near plane of the orthographic projection. Note, in view space the z-axis points out of the viewport.
Change the orthographic projection (or invert the z coordinates of the geometry):
glOrtho(0, width, 0, height, 0, 255);
glOrtho(0, width, 0, height, -255, 0);
The stride parameter of glVertexPointer respectively glColorPointer is the offset between consecutive attributes. So it has to be sizeof(ver).
The type of the color attributes is GL_UNSIGNED_BYTE rather than GL_BYTE:
glVertexPointer(3, GL_INT, 4 * sizeof(GLubyte), 0);
glColorPointer(4, GL_BYTE, 3 * sizeof(GLuint), BUFFER_OFFSET(3 * sizeof(GLuint)));
glVertexPointer(3, GL_INT, sizeof(ver), 0);
glColorPointer(4, GL_UNSIGNED_BYTE, sizeof(ver), BUFFER_OFFSET(3 * sizeof(GLuint)));