Simplify the methods for pixel reading and writing

[maarzt]

When writing a OpenCL kernel in clEsperanto, the code for reading a pixel usually looks like this:

const POS_src_TYPE spos = POS_src_INSTANCE(sx, sy, 0, 0);
const float a = READ_src_IMAGE(src,sampler,spos).x;

I understand the benefits that clEsperanto gives us here. But I would suggest to go further and simplify this by adding a C #define such that the example above could be rewritten as:

const float a = READ_PIXEL_2D(src, sx, sy);

The define could look like this:

#define READ_PIXEL_2D(image, x, y) (READ_ ## image ## _IMAGE((image), sampler, POS_ ## image ## _INSTANCE((x), (y), 0, 0)).x)

The same should be done for image writing, and 3d images.

[haesleinhuepf]

Hey @maarzt ,

great to see you here!

The same should be done for image writing, and 3d images.

I think we could even formulate that in a 2D-3D-agnostic way. E.g.

#define READ_PIXEL(image, x, y, z) (READ_ ## image ## _IMAGE((image), sampler, POS_ ## image ## _INSTANCE((x), (y), (z), 0)).x)

And the change appears backwards-compatible. Thus, we can build it in the preamble any time and update kernels step by step.

Thanks for the suggestion!

Cheers,
Robert