Preferred endianness

[hackaugusto]

Some simple questions about endianness on the VM.

Bytes enumeration

Given the word W with bytes 0x04 0x03 0x02 0x01, in big-endian format (most significant byte is 0x04). How should the bytes be named?

byte	0x04	0x03	0x02	0x01	description
1	w[0]	w[1]	w[0]	w[3]	index is byte position from right-to-left (MSB is 0)
2	w[3]	w[2]	w[1]	w[0]	index is byte position from left-to-right (LSB is 0) (*)

Endianness on the stack

Would it be little endian or big endian?

stack	s[0]	s[1]	s[2]	s[3]	description
LE	0x01	0x02	0x03	0x04	LSB first
BE	0x04	0x03	0x02	0x01	MSB first (*)

* marks the preferred denotation

[bobbinth]

Great topic!

One note on terminology: we use "word" to denote an array of 4 field elements. So, words consist of field elements and not bytes. Field elements can be split up into bytes (8 bytes per element).

Regarding endianness: we typically use little-ending order. So, a word is (a0, a1, a2, a3) where a0 is the least significant element of the word.

When a word is placed on the stack, it is arranged in "stack order" which is the reverse of the normal order. So, it appears on the stack as [a3, a2, a1, a0, ...] where a3 is at the top of the stack. The reasoning for this is that if we were to read the word element-by-element and push each element onto the stack we'd first push a0 onto the stack, then a1 onto the stack etc.

Bytes within an element are assumed to be arranged in the little endian order as well. So, for example, when we use u32split operation ([a, ...] -> [a1, a0, ...]), the top 32 bits are put at the top of the stack, and lower 32 bits are put at the bottom of the stack.

We treat u64 values in a similar manner. A u64 value consisting of two 32-bit limbs (a0, a1) where a0 is the least significant limb, would be placed on the stack as [a1, a0, ...].