I've been talking to the person who made the giant DS, and while he won't share his code, he did go into how he did it. I was thinking of scaling up wonderswan and other consoles and was wondering is anyone have ideas or input on this.
Use internal fpga block RAM to store the scanline to be doubled, redraw it on the next h-retrace If scaling bigger in the X direction, hold data for 2 pixels instead of just one
The hardware part of this is really easy on handhelds since they have pixel buses for their LCDs. On full consoles it's much rarer to have the pixel bus exposed. (It does happen though; Neo Geo, 3DO and N64 all do. Probably PCE too.) Something you have to think about with handhelds though is that they don't have proper horizontal timing or blanking for a TV. Taking WS (224x144) as an example, you'd probably want to scale to 720p since 144 x 5 = 720. Assuming the WS has square pixels, 224 x 5 = 1120. On a 16:9, square pixels @ 720p = 1280, so you'll need to blank 80 pixels on either side of the line. You'll also have to reconstruct your own H-sync information since the WS' won't be correct for a TV. Most importantly, you probably can't get the desired output pixel frequency of 5x (input + added blanking pixels) from the pixel clock so you'll have to synthesize it yourself with your own oscillator. This means you'll need to buffer MULTIPLE lines since your frequency will not be in phase with the pixel clock. Even though the output pixel clock (and width) will be slightly off, you can resynchronize to the input video each frame and burn up any excess cycles. The output should lag behind the input only enough so that by the time the output catches up to the input, the frame is finished being drawn. (I think that'd only require buffering 2-3 lines if you synthesize a pretty close frequency.)
