even on a basic ARM9 without any fancy simd, we can find some fun doing "data-oriented" tricks. for my DM30 console, i'm using an 8x12 font, with the framebuffer stored as 1bpp 800x600 and the font stored as an array of 1bpp 8x12 glyphs. since our CPU is 32 bit, instead of doing byte loads/stores, we can load four characters out of the character map at a time, then load four rows of each character's glyph, then swizzle the pixels around in registers to splat out a 32x4 tile at a time