That looks like a neat routine.  Nice job calc84!

Wow, could you explain in details how you managed to achieve such speed? I totally don't understand any of your code

This is a little weird to explain. This zooms by shifting the screen in parts to each edge. For vertical zooming, it's as simple as copying data from L₆+12 to L₆ and copying from L₆+755 to L₆+767 (reversed) multiple times, each time with a different copy size. Horizontal zooming was a bit tougher and takes up most of the variables and calculations. I multiply a certain number of bytes on the left side of the screen by 2 and divide bytes on the right side by 2, making sure to carry pixels that would be bumped off from byte to byte.

What I mean is that everything is moved towards the edges of the screen. The distance from the middle in either the vertical or horizontal direction determines how far it will be moved towards either the vertical or horizontal edge closest to it. The middle pixels will move towards the edges by only one pixel, whereas pixels halfway between an edge and the middle will be moved all the way to the edge.

And we don't want to leave grayscale programs stranded, though they'll take a decently sized speed hit: 4 seconds for black and white vs 12 seconds for grayscale. It takes an extra 4 seconds to shift the back buffer and another 4 to update the screen using DispGraph^rr 128 times throughout the process. On 15MHz calculators, this routine should take about 8 seconds.

EDIT: Uploaded a new GIF and source, as I realized I was copying the pictures to the wrong buffers so the grays were reversed.

Yeah, the hardest parts about this routine were the horizontal zooming and indicating which pixel rows/columns to shift outwards from, and in what order. The horizontal zooming is quite math-based, and took a lot of pacing around my room and thinking to produce and finally get right.

Well the thing about shifting pixels outwards horizontally is that, whereas a vertictical shift of one pixel is 12 bytes, a horizontal shift of one pixel is only one bit. Shifting data by one bit cannot be done using conj().