texture cudaFilterModeLinear for float4?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi!

In the programming guide it reads: "cudaFilterModeLinear is only valid for returned

values of floating-point type."

Does this mean textures of type

texture<float4, 2, cudaReadModeElementType>

can not be interpolated, only

texture<float, 2, cudaReadModeElementType>

can?

bye

Sebastian

2

You can interpolate float2s and float4s. And also 16 bit floating point formats if you use the driver API.

3

Be aware that the hardware interpolator is 9bit precision only (it is built for RGBA8 image data). Use cudaReadModeElementType and do the interpolation in CUDA to get better precision.

Peter

4

This will also be much slower. It depends if you’re more concerned about speed or accuracy.

5

Well, lerp is not that expensive. In my experience, it is usually well hidden in the read latency as code pieces that do interpolation are usually memory bound anyway.

Peter

6

Thank you all! Got it working.
I am using the float4 interpolation now since accuracy is not very important to me.
Thanks for your help!
Sebastian

7

This is a bit misleading - to be exact the precision of the fractional part of the interpolant is 9-bit. This means there are only 2^9 possible interpolated positions between each texel, which can cause steps at large magnifications.

The resulting values will still be floating point.