How to sample 2x2 neighbor points of a given point, like textureGather() in openGL

[zhouwangpei123]

Given a point (x, y) in an image, let intx = floor(x), inty = floor(y), can halide gather its four corner points (intx, inty), (intx+1, inty), (intx, inty+1), (intx+1, inty+1) in one time.

[abadams]

In Halide that has to be expressed as four separate loads.

[zhouwangpei123]

thanks for the clarification.

[soufianekhiat]

If I generate 4 loads dors Halide would be able to have intrinsic for gather4 on GPU?