CODEWITHSUNDEEP
c++openglgpgpu
CivDav
CivDav

Reputation: 102

Passing data between CPU and GPU in textures

So I have a very complicated R^4 -> R^4 function, which I need to calculate for a lot of input glm::vec4s, in real time, so I want to do it on the GPU, for all vec4s parallel.

What I figured is that I would create a GL_RGBA32F texture, 1920x1 resolution (1920 is enough for my purposes), copy my input data onto the texture, then call a drawing of a line, so the rasterizer calls a fragment for each of my vec4s. Then either write the results back to the texture using imageload/store or render it to a 1920x1 framebuffer, and read it from there.

Problem is that for some reason opengl can't read my GL_RGBA32F texture.

Here is my code:

Setting up the texture (currently loaded with dummy data):

glm::vec4 texturedata[1920];
for (unsigned int i = 0; i < 1920; i++)
{
    texturedata[i] = glm::vec4(1.0f, 1.0f, 1.0f, 1.0f);
}
glGenTextures(1, &datatexture);
glBindTexture(GL_TEXTURE_2D, datatexture);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA32F, 1920, 1, 0, GL_RGBA, GL_FLOAT, texturedata);

Before each rendering:

glUseProgram(mprogram);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, datatexture);
glBindVertexArray(rasterizertriggervao);
glUniform1i(glGetUniformLocation(mprogram, "datatexture"), 0);
glDrawArrays(GL_LINES, 0, 2);

The rasterizertriggervao is 2 floats: -1, 1, and the vertex shader draws a nice line through the middle of my screen from that.

Fragment shader:

layout(binding = 0) uniform sampler2D datatexture;
out vec4 x;
void main()
{
    x = vec4( (texture(datatexture, vec2(gl_FragCoord.x/1920.0, 0.0))).x, 0.0f, 0.0f, 1.0f );
}

So this should draw a nice red line in the middle of my screen for me. It draws a black one. The rasterizer called all 1920x1 fragments, and the texture is correctly copied to the GPU (I have Nvidia Nsight installed, which allows me to debug the GPU, check the contents of textures and whatnot on the GPU directly, and I checked, the texture is full of 1.0f). However for some reason the sampling doesn't work.

I know that there are better ways to do GPGPU but this thing has to fit into a much bigger program nicely, and this is the way I need it to work, through textures :)

Upvotes: 1

Views: 1181

Answers (1)

derhass
derhass

Reputation: 45332

Your seem not to set the texture filter modes for your texture. Now, GL's defaults are (unfortunately) to use mip-mapping. But your texture is not mipmap-complete, so sampling from it will not work. You should add glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MIN_FILTER, GL_NEAREAST) and probably also glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MAG_FILTER, GL_NEAREAST).

As Andon M. Coleman already pointed out in the comments, you are not sampling the texture at the correct location. You should use vec2(gl_FragCoord.x/1920.0 + 0.5/1920.0, 0.5) in your case. I also agree with Andon M. Coleman's suggestion to directly use texelFetch(), since you can directly use the integer value `gl_FragCoord.x'.

Upvotes: 2

Related Questions