Optimisation for amd64

Increased major number as the ABI was broken
by insertion of padding into the BLAKE2
parameter structures (except for BLAKE2Xs)

Signed-off-by: Mattias Andrée <maandree@kth.se>

1 files changed, 23 insertions, 1 deletions

diff --git a/libblake_blake2b_force_update.c b/libblake_blake2b_force_update.c
index 60b8fab..2446e16 100644
--- a/libblake_blake2b_force_update.c
+++ b/libblake_blake2b_force_update.c
@@ -8,8 +8,30 @@ libblake_blake2b_force_update(struct libblake_blake2b_state *state, const void *
 	size_t off = 0;
 
 	for (; len - off >= 128; off += 128) {
+		/* The following optimisations have been tested:
+		 * 
+		 * 1)
+		 *     `*(__uint128_t *)state->t += 128;`
+		 *     result: slower
+		 * 
+		 * 2)
+		 *     addq, adcq using `__asm__ __volatile__`
+		 *     result: slower (as 1)
+		 * 
+		 * 3)
+		 *     using `__builtin_add_overflow`
+		 *     result: no difference
+		 * 
+		 * These testes where preformed on amd64 with a compile-time
+		 * assumption that `UINT_LEAST64_C(0xFFFFffffFFFFffff) + 1 == 0`,
+		 * which the compiler accepted and those included the attempted
+		 * optimisations.
+		 * 
+		 * UNLIKELY does not seem to make any difference, but it
+		 * does change the output, theoretically of the better.
+		 */
 		state->t[0] = (state->t[0] + 128) & UINT_LEAST64_C(0xFFFFffffFFFFffff);
-		if (state->t[0] < 128)
+		if (UNLIKELY(state->t[0] < 128))
 			state->t[1] = (state->t[1] + 1) & UINT_LEAST64_C(0xFFFFffffFFFFffff);
 
 		libblake_internal_blake2b_compress(state, &data[off]);