[PATCH 5/5] crypto: arm/blake2b - add NEON-optimized BLAKE2b implementation

[Eric Biggers]

On Tue, Dec 15, 2020 at 03:47:08PM -0800, Eric Biggers wrote:
> +// Execute one round of BLAKE2b by updating the state matrix v[0..15] in the
> +// NEON registers q0-q7.  The message block is in q8..q15.  The stack pointer
> +// points to a 32-byte aligned buffer containing a copy of q8 and q9, so that
> +// they can be reloaded if q8 and q9 are used as temporary registers.  The macro
> +// arguments s0-s15 give the order in which the message words are used in this
> +// round.  'final' is "true" if this is the final round, i.e. round 12 of 12.
> +.macro	_blake2b_round	s0, s1, s2, s3, s4, s5, s6, s7, \
> +			s8, s9, s10, s11, s12, s13, s14, s15, final="false"
[...]
> +	// Reloading q8-q9 can be skipped on the final round.
> +.if \final != "true"
> +	vld1.8		{q8-q9}, [sp, :256]
> +.endif
> +.endm
[...]
> +	_blake2b_round 14, 10,  4,  8,  9, 15, 13,  6, \
> +			1, 12,  0,  2, 11,  7, 5,   3,  "true"

Apparently using the strings "true" and "false" here sometimes causes a build
error where they get treated as symbols
(https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org/thread/2JPD4H3VFBSKWPUCPEPRAXBVMSR2UCQI/),
though somehow it worked fine for me.  I'll change this to use 1 and 0 instead.

- Eric