R32V2020 Big vs Little Endian

Historical Context

• There's a long history of endian battles.

Endianness is of interest in computer science because two conflicting and incompatible formats are in common use: words may be represented in big-endian or little-endian format, depending on whether bits or bytes or other components are ordered from the big end (most significant bit) or the little end (least significant bit). - Endianness

• Both big and little endian have advantages and disadvantages.
  • Readability.
    • Reading memory and swapping around the bytes when displaying them is unnatural.
  • Connection to peripherals seems unnatural with bytes on d24-d31, shorts on d16-d31.
    • Some are big and some are little endian.
• Software solutions are really big and painful.
  • Shifts (by 8), masks, ORs, etc.
  • Doing this eliminates much of the need for slow shifting

Solution

• endswap - opcode that swaps the endian of a long
  • d0-d7 get moved to d24-d32, d8-d15 get moved to d16-d23, d16-d23 get moved to d8-d15, and d24-d31 get moved to d0-d7
  • Relatively easy to implement in the FPGA ALU

Example

• "Natural" data bus connections
  • Byte data length peripherals always connect to d0-d7
  • Short data length peripherals always connect to d0-d15
  • Long data length peripherals connect to d0-d32
• Data memory should be stored in natural sense as well
  • 32-bit memory packed with byte data
    • [d32..d24|d23..d16|d15..d8|d7..d0]
    • Reads in memory dump with most significant byte first
    • If you want to send that 32-bit data to an 8-bit peripheral you end up doing shifts/masks with multiple registers
  • Much easier to do endswap which flips the order into a register and put out the register one at a time
  • Still a use for shl8 and shr8 which shift the 32-bit register by 8 bits
  • Less of a need for arbitrary shifts
    • Could get by with 8-bit shifts and 1-bit shifts