1D texture and linear interpolation

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I got a piece of code that works when I do not use interpolation:

g_table_texture.addressMode[0] = cudaAddressModeClamp;

	g_table_texture.addressMode[1] = cudaAddressModeClamp;

	g_table_texture.filterMode = cudaFilterModePoint;

	g_table_texture.normalized = false;

And I fetch a texture element using:

return tex1Dfetch(g_table_texture, int_texture_coordinate);

with 0 <= int_texture_coordinate < N

Now I change it to:

g_table_texture.addressMode[0] = cudaAddressModeClamp;

	g_table_texture.addressMode[1] = cudaAddressModeClamp;

	g_table_texture.filterMode = cudaFilterModeLinear;

	g_table_texture.normalized = true;

And I fetch a texture element using:

return tex1D(g_table_texture, float_texture_coordinate);

with float float_texture_coordinate = int_texture_coordinate;

float_texture_coordinate /= N;

so that 0 <= float_texture_coordinate <= 1

Now the return value is always 0.

Any idea what I am doing wrong?

2

Failing to check for error codes. If you had, it would have said invalid texture reference.

tex1D can only read textures bound to cudaArrays.

3

Failing to check for error codes. If you had, it would have said invalid texture reference.

tex1D can only read textures bound to cudaArrays.

4

I check for errors after each run of the kernel.

There is no error.

5

I check for errors after each run of the kernel.

There is no error.

6

It wasn’t a cudaArray. No runtime error, tho.

Moving from 1D to 2D worked with liner memory…

7

It wasn’t a cudaArray. No runtime error, tho.

Moving from 1D to 2D worked with liner memory…