On x86_64, they can be done in 16-byte blocks
64-bit:
function old new delta
xorbuf_3 - 84 +84
xorbuf64_3_aligned64 - 58 +58
smix1 687 712 +25
xwrite_encrypted 520 534 +14
xorbuf16_aligned_long - 13 +13
tls_xread_record 733 742 +9
xorbuf 21 13 -8
xorbuf_aligned_AES_BLOCK_SIZE 15 - -15
blockmix 814 762 -52
blockmix_salsa8 317 198 -119
blockmix_xor_save 1620 1499 -121
blockmix_xor 1543 1322 -221
------------------------------------------------------------------------------
(add/remove: 4/1 grow/shrink: 3/5 up/down: 203/-536) Total: -333 bytes
32-bit:
function old new delta
xorbuf_3 - 76 +76
xorbuf64_3_aligned64 - 36 +36
xorbuf16_aligned_long - 23 +23
xwrite_encrypted 499 507 +8
tls_xread_record 646 650 +4
xorbuf 22 11 -11
xorbuf_aligned_AES_BLOCK_SIZE 23 - -23
blockmix 1083 938 -145
blockmix_salsa8 415 210 -205
blockmix_salsa8_xor 601 163 -438
blockmix_xor 2103 1533 -570
blockmix_xor_save 2614 1859 -755
------------------------------------------------------------------------------
(add/remove: 4/1 grow/shrink: 2/6 up/down: 147/-2147) Total: -2000 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>