Most information about memory mapped displays on the net are about those in which there is essentially a location in main memory for each pixel on the display. A hypothetical 1024 x 512 display would therefore have 524,288 locations each mapped to a unique pixel. To set a pixel to a particular color, all you basically need to do is write an RGB value into the corresponding memory location for that pixel, eg: MOV 0xFF3C0A, 0x00BB75. The first argument in that instruction is the address holding the RGB value and the second is the address of the pixel you are copying it to.
However, this can use up a quite a lot of main memory, which might not be ideal for certain systems. In a machine with a 32 bit or higher word size, it could instead be possible to have a single "display output register" into which you would write a single word of data containing the X and Y coordinates as well as the RGB value for the pixel you want to set. For a 1024 x 512 display with a 12-bit color depth, 31 bits would suffice to specify all this. The data 'written' into that register would either be sent to a display adapter or sent directly to the display and decoded there. This would save over half a megabyte of memory while still retaining the memory-mapped model.
So is this actually a thing or not?