Mailing List Archive

[PATCH 3/3] Optimizations for AES aarch64-ce assembly implementation
* cipher/rijndael-armv8-aarch64-ce.S (vk14): Remove.
(vklast, __, _): New.
(aes_preload_keys): Setup vklast.
(do_aes_one128/192/256): Split to ...
(do_aes_one_part1, do_aes_part2_128/192/256): ... these and add
interleave ops.
(do_aes_one128/192/256): New using above part1 and part2 macros.
(aes_round_4): Rename to ...
(aes_round_4_multikey): ... this and allow different key used for
parallel blocks.
(aes_round_4): New using above multikey macro.
(aes_lastround_4): Reorder AES round and xor instructions, allow
different last key for parallel blocks.
(do_aes_4_128/192/256): Split to ...
(do_aes_4_part1_multikey, do_aes_4_part1)
(do_aes_4_part2_128/192/256): ... these.
(do_aes_4_128/192/256): New using above part1 and part2 macros.
(CLEAR_REG): Use movi for clearing registers.
(aes_clear_keys): Remove branching and clear all key registers.
(_gcry_aes_enc_armv8_ce, _gcry_aes_dec_armv8_ce): Adjust to macro
changes.
(_gcry_aes_cbc_enc_armv8_ce, _gcry_aes_cbc_dec_armv8_ce)
(_gcry_aes_cfb_enc_armv8_ce, _gcry_aes_cfb_enc_armv8_ce)
(_gcry_aes_ctr32le_enc_armv8_ce): Apply entry/loop-body/exit
optimization for better interleaving of input/output processing;
First/last round key and input/output xoring optimization to reduce
critical path length.
(_gcry_aes_ctr_enc_armv8_ce): Add fast path for counter incrementing
without byte-swaps when counter does not overflow 8-bit; Apply
entry/loop-body/exit optimization for better interleaving of
input/output processing; First/last round key and input/output
xoring optimization to reduce critical path length.
(_gcry_aes_ocb_enc_armv8_ce, _gcry_aes_ocb_dec_armv8_ce): Add aligned
processing for nblk and OCB offsets; Apply entry/loop-body/exit
optimization for better interleaving of input/output processing;
First/last round key and input/output xoring optimization to reduce
critical path length; Change to use same function body macro for
both encryption and decryption.
(_gcry_aes_xts_enc_armv8_ce, _gcry_aes_xts_dec_armv8_ce): Apply
entry/loop-body/exit optimization for better interleaving of
input/output processing; First/last round key and input/output
xoring optimization to reduce critical path length; Change to use
same function body macro for both encryption and decryption.
--

Benchmark on AWS Graviton2 (2500Mhz):

Before:
AES | nanosecs/byte mebibytes/sec cycles/byte
CBC enc | 0.663 ns/B 1439 MiB/s 1.66 c/B
CBC dec | 0.288 ns/B 3310 MiB/s 0.720 c/B
CFB enc | 0.657 ns/B 1453 MiB/s 1.64 c/B
CFB dec | 0.288 ns/B 3313 MiB/s 0.720 c/B
CTR dec | 0.314 ns/B 3039 MiB/s 0.785 c/B
XTS enc | 0.357 ns/B 2674 MiB/s 0.891 c/B
XTS dec | 0.358 ns/B 2666 MiB/s 0.894 c/B
OCB enc | 0.343 ns/B 2784 MiB/s 0.856 c/B
OCB dec | 0.341 ns/B 2795 MiB/s 0.853 c/B
GCM-SIV enc | 0.526 ns/B 1813 MiB/s 1.31 c/B

After:
AES | nanosecs/byte mebibytes/sec cycles/byte perf increase
CBC enc | 0.500 ns/B 1906 MiB/s 1.25 c/B +33%
CBC dec | 0.263 ns/B 3622 MiB/s 0.658 c/B +9%
CFB enc | 0.500 ns/B 1906 MiB/s 1.25 c/B +31%
CFB dec | 0.263 ns/B 3620 MiB/s 0.658 c/B +9%
CTR enc | 0.264 ns/B 3618 MiB/s 0.659 c/B +19%
XTS enc | 0.350 ns/B 2722 MiB/s 0.876 c/B +2%
OCB enc | 0.275 ns/B 3468 MiB/s 0.687 c/B +25%
OCB dec | 0.276 ns/B 3459 MiB/s 0.689 c/B +24%
GCM-SIV enc | 0.494 ns/B 1929 MiB/s 1.24 c/B +6%

Benchmark on Cortex-A53 (1152Mhz):

Before:
AES | nanosecs/byte mebibytes/sec cycles/byte
CBC enc | 1.41 ns/B 675.9 MiB/s 1.63 c/B
CBC dec | 0.910 ns/B 1048 MiB/s 1.05 c/B
CFB enc | 1.30 ns/B 732.2 MiB/s 1.50 c/B
CFB dec | 0.910 ns/B 1048 MiB/s 1.05 c/B
CTR enc | 1.03 ns/B 924.4 MiB/s 1.19 c/B
XTS enc | 1.25 ns/B 763.0 MiB/s 1.44 c/B
OCB enc | 1.21 ns/B 789.5 MiB/s 1.39 c/B
OCB dec | 1.21 ns/B 788.9 MiB/s 1.39 c/B
GCM-SIV enc | 1.92 ns/B 496.6 MiB/s 2.21 c/B

After:
AES | nanosecs/byte mebibytes/sec cycles/byte perf increase
CBC enc | 1.14 ns/B 836.6 MiB/s 1.31 c/B +24%
CBC dec | 0.843 ns/B 1132 MiB/s 0.971 c/B +8%
CFB enc | 1.19 ns/B 798.8 MiB/s 1.38 c/B +9%
CFB dec | 0.842 ns/B 1132 MiB/s 0.970 c/B +8%
CTR enc | 0.898 ns/B 1062 MiB/s 1.03 c/B +16%
XTS enc | 1.22 ns/B 779.9 MiB/s 1.41 c/B +2%
OCB enc | 0.992 ns/B 961.0 MiB/s 1.14 c/B +22%
OCB dec | 0.993 ns/B 960.5 MiB/s 1.14 c/B +22%
GCM-SIV enc | 1.88 ns/B 507.3 MiB/s 2.17 c/B +2%

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
---
cipher/rijndael-armv8-aarch64-ce.S | 1227 ++++++++++++++++------------
1 file changed, 713 insertions(+), 514 deletions(-)

diff --git a/cipher/rijndael-armv8-aarch64-ce.S b/cipher/rijndael-armv8-aarch64-ce.S
index a87d2ca5..9f8d9d49 100644
--- a/cipher/rijndael-armv8-aarch64-ce.S
+++ b/cipher/rijndael-armv8-aarch64-ce.S
@@ -44,7 +44,13 @@
#define vk11 v28
#define vk12 v29
#define vk13 v30
-#define vk14 v31
+#define vklast v31
+
+
+/* Helper macros */
+
+#define __ /*_*/
+#define _(...) __VA_ARGS__


/* AES macros */
@@ -54,39 +60,40 @@
ld1 {vk0.16b-vk3.16b}, [keysched], #64; \
ld1 {vk4.16b-vk7.16b}, [keysched], #64; \
ld1 {vk8.16b-vk10.16b}, [keysched], #48; \
+ mov vklast.16b, vk10.16b; \
b.lo 1f; \
ld1 {vk11.16b-vk12.16b}, [keysched], #32; \
+ mov vklast.16b, vk12.16b; \
b.eq 1f; \
- ld1 {vk13.16b-vk14.16b}, [keysched]; \
+ ld1 {vk13.16b-vklast.16b}, [keysched]; \
1: ;

-#define do_aes_one128(ed, mcimc, vo, vb) \
- aes##ed vb.16b, vk0.16b; \
+#define do_aes_one_part1(ed, mcimc, vb, vkfirst) \
+ aes##ed vb.16b, vkfirst.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk1.16b; \
- aes##mcimc vb.16b, vb.16b; \
+ aes##mcimc vb.16b, vb.16b;
+
+#define do_aes_one_part2_128(ed, mcimc, vb, iop1, iop2) \
aes##ed vb.16b, vk2.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk3.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop1; \
aes##ed vb.16b, vk4.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk5.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk6.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop2; \
aes##ed vb.16b, vk7.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk8.16b; \
aes##mcimc vb.16b, vb.16b; \
- aes##ed vb.16b, vk9.16b; \
- eor vo.16b, vb.16b, vk10.16b;
+ aes##ed vb.16b, vk9.16b;

-#define do_aes_one192(ed, mcimc, vo, vb) \
- aes##ed vb.16b, vk0.16b; \
- aes##mcimc vb.16b, vb.16b; \
- aes##ed vb.16b, vk1.16b; \
- aes##mcimc vb.16b, vb.16b; \
+#define do_aes_one_part2_192(ed, mcimc, vb, iop1, iop2) \
aes##ed vb.16b, vk2.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk3.16b; \
@@ -95,24 +102,21 @@
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk5.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop1; \
aes##ed vb.16b, vk6.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk7.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk8.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop2; \
aes##ed vb.16b, vk9.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk10.16b; \
aes##mcimc vb.16b, vb.16b; \
- aes##ed vb.16b, vk11.16b; \
- eor vo.16b, vb.16b, vk12.16b;
+ aes##ed vb.16b, vk11.16b;

-#define do_aes_one256(ed, mcimc, vo, vb) \
- aes##ed vb.16b, vk0.16b; \
- aes##mcimc vb.16b, vb.16b; \
- aes##ed vb.16b, vk1.16b; \
- aes##mcimc vb.16b, vb.16b; \
+#define do_aes_one_part2_256(ed, mcimc, vb, iop1, iop2) \
aes##ed vb.16b, vk2.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk3.16b; \
@@ -125,56 +129,78 @@
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk7.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop1; \
aes##ed vb.16b, vk8.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk9.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk10.16b; \
aes##mcimc vb.16b, vb.16b; \
+ iop2; \
aes##ed vb.16b, vk11.16b; \
aes##mcimc vb.16b, vb.16b; \
aes##ed vb.16b, vk12.16b; \
aes##mcimc vb.16b, vb.16b; \
- aes##ed vb.16b, vk13.16b; \
- eor vo.16b, vb.16b, vk14.16b;
+ aes##ed vb.16b, vk13.16b;

-#define aes_round_4(ed, mcimc, b0, b1, b2, b3, key) \
- aes##ed b0.16b, key.16b; \
+#define do_aes_one128(ed, mcimc, vo, vb, vkfirst) \
+ do_aes_one_part1(ed, mcimc, vb, vkfirst); \
+ do_aes_one_part2_128(ed, mcimc, vb, __, __); \
+ eor vo.16b, vb.16b, vklast.16b;
+
+#define do_aes_one192(ed, mcimc, vo, vb, vkfirst) \
+ do_aes_one_part1(ed, mcimc, vb, vkfirst); \
+ do_aes_one_part2_192(ed, mcimc, vb, __, __); \
+ eor vo.16b, vb.16b, vklast.16b;
+
+#define do_aes_one256(ed, mcimc, vo, vb, vkfirst) \
+ do_aes_one_part1(ed, mcimc, vb, vkfirst); \
+ do_aes_one_part2_256(ed, mcimc, vb, __, __); \
+ eor vo.16b, vb.16b, vklast.16b;
+
+#define aes_round_4_multikey(ed, mcimc, b0, b1, b2, b3, key0, key1, key2, key3) \
+ aes##ed b0.16b, key0.16b; \
aes##mcimc b0.16b, b0.16b; \
- aes##ed b1.16b, key.16b; \
+ aes##ed b1.16b, key1.16b; \
aes##mcimc b1.16b, b1.16b; \
- aes##ed b2.16b, key.16b; \
+ aes##ed b2.16b, key2.16b; \
aes##mcimc b2.16b, b2.16b; \
- aes##ed b3.16b, key.16b; \
+ aes##ed b3.16b, key3.16b; \
aes##mcimc b3.16b, b3.16b;

-#define aes_lastround_4(ed, b0, b1, b2, b3, key1, key2) \
+#define aes_round_4(ed, mcimc, b0, b1, b2, b3, key) \
+ aes_round_4_multikey(ed, mcimc, b0, b1, b2, b3, key, key, key, key);
+
+#define aes_lastround_4(ed, o0, o1, o2, o3, b0, b1, b2, b3, key1, b0_key2, b1_key2, b2_key2, b3_key2) \
aes##ed b0.16b, key1.16b; \
- eor b0.16b, b0.16b, key2.16b; \
aes##ed b1.16b, key1.16b; \
- eor b1.16b, b1.16b, key2.16b; \
aes##ed b2.16b, key1.16b; \
- eor b2.16b, b2.16b, key2.16b; \
aes##ed b3.16b, key1.16b; \
- eor b3.16b, b3.16b, key2.16b;
+ eor o0.16b, b0.16b, b0_key2.16b; \
+ eor o1.16b, b1.16b, b1_key2.16b; \
+ eor o2.16b, b2.16b, b2_key2.16b; \
+ eor o3.16b, b3.16b, b3_key2.16b;

-#define do_aes_4_128(ed, mcimc, b0, b1, b2, b3) \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk0); \
+#define do_aes_4_part1_multikey(ed, mcimc, b0, b1, b2, b3, key0, key1, key2, key3) \
+ aes_round_4_multikey(ed, mcimc, b0, b1, b2, b3, key0, key1, key2, key3); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk1); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk2); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk3); \
+ aes_round_4(ed, mcimc, b0, b1, b2, b3, vk3);
+
+#define do_aes_4_part1(ed, mcimc, b0, b1, b2, b3, vkfirst) \
+ do_aes_4_part1_multikey(ed, mcimc, b0, b1, b2, b3, vkfirst, vkfirst, vkfirst, vkfirst);
+
+#define do_aes_4_part2_128(ed, mcimc, o0, o1, o2, o3, b0, b1, b2, b3, \
+ b0_key, b1_key, b2_key, b3_key) \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk4); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk5); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk6); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk7); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk8); \
- aes_lastround_4(ed, b0, b1, b2, b3, vk9, vk10);
+ aes_lastround_4(ed, o0, o1, o2, o3, b0, b1, b2, b3, vk9, b0_key, b1_key, b2_key, b3_key);

-#define do_aes_4_192(ed, mcimc, b0, b1, b2, b3) \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk0); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk1); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk2); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk3); \
+#define do_aes_4_part2_192(ed, mcimc, o0, o1, o2, o3, b0, b1, b2, b3, \
+ b0_key, b1_key, b2_key, b3_key) \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk4); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk5); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk6); \
@@ -182,13 +208,10 @@
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk8); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk9); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk10); \
- aes_lastround_4(ed, b0, b1, b2, b3, vk11, vk12);
+ aes_lastround_4(ed, o0, o1, o2, o3, b0, b1, b2, b3, vk11, b0_key, b1_key, b2_key, b3_key);

-#define do_aes_4_256(ed, mcimc, b0, b1, b2, b3) \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk0); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk1); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk2); \
- aes_round_4(ed, mcimc, b0, b1, b2, b3, vk3); \
+#define do_aes_4_part2_256(ed, mcimc, o0, o1, o2, o3, b0, b1, b2, b3, \
+ b0_key, b1_key, b2_key, b3_key) \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk4); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk5); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk6); \
@@ -198,15 +221,25 @@
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk10); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk11); \
aes_round_4(ed, mcimc, b0, b1, b2, b3, vk12); \
- aes_lastround_4(ed, b0, b1, b2, b3, vk13, vk14);
+ aes_lastround_4(ed, o0, o1, o2, o3, b0, b1, b2, b3, vk13, b0_key, b1_key, b2_key, b3_key);

+#define do_aes_4_128(ed, mcimc, b0, b1, b2, b3) \
+ do_aes_4_part1(ed, mcimc, b0, b1, b2, b3, vk0); \
+ do_aes_4_part2_128(ed, mcimc, b0, b1, b2, b3, b0, b1, b2, b3, vklast, vklast, vklast, vklast);
+
+#define do_aes_4_192(ed, mcimc, b0, b1, b2, b3) \
+ do_aes_4_part1(ed, mcimc, b0, b1, b2, b3, vk0); \
+ do_aes_4_part2_192(ed, mcimc, b0, b1, b2, b3, b0, b1, b2, b3, vklast, vklast, vklast, vklast);
+
+#define do_aes_4_256(ed, mcimc, b0, b1, b2, b3) \
+ do_aes_4_part1(ed, mcimc, b0, b1, b2, b3, vk0); \
+ do_aes_4_part2_256(ed, mcimc, b0, b1, b2, b3, b0, b1, b2, b3, vklast, vklast, vklast, vklast);

/* Other functional macros */

-#define CLEAR_REG(reg) eor reg.16b, reg.16b, reg.16b;
+#define CLEAR_REG(reg) movi reg.16b, #0;

#define aes_clear_keys(nrounds) \
- cmp nrounds, #12; \
CLEAR_REG(vk0); \
CLEAR_REG(vk1); \
CLEAR_REG(vk2); \
@@ -218,13 +251,10 @@
CLEAR_REG(vk9); \
CLEAR_REG(vk8); \
CLEAR_REG(vk10); \
- b.lo 1f; \
CLEAR_REG(vk11); \
CLEAR_REG(vk12); \
- b.eq 1f; \
CLEAR_REG(vk13); \
- CLEAR_REG(vk14); \
-1: ;
+ CLEAR_REG(vklast);


/*
@@ -252,7 +282,7 @@ _gcry_aes_enc_armv8_ce:
b.eq .Lenc1_192

.Lenc1_128:
- do_aes_one128(e, mc, v0, v0);
+ do_aes_one128(e, mc, v0, v0, vk0);

.Lenc1_tail:
CLEAR_REG(vk0)
@@ -266,6 +296,7 @@ _gcry_aes_enc_armv8_ce:
CLEAR_REG(vk8)
CLEAR_REG(vk9)
CLEAR_REG(vk10)
+ CLEAR_REG(vklast)
st1 {v0.16b}, [x1]
CLEAR_REG(v0)

@@ -273,19 +304,18 @@ _gcry_aes_enc_armv8_ce:
ret

.Lenc1_192:
- do_aes_one192(e, mc, v0, v0);
+ do_aes_one192(e, mc, v0, v0, vk0);

CLEAR_REG(vk11)
CLEAR_REG(vk12)
b .Lenc1_tail

.Lenc1_256:
- do_aes_one256(e, mc, v0, v0);
+ do_aes_one256(e, mc, v0, v0, vk0);

CLEAR_REG(vk11)
CLEAR_REG(vk12)
CLEAR_REG(vk13)
- CLEAR_REG(vk14)
b .Lenc1_tail
CFI_ENDPROC();
ELF(.size _gcry_aes_enc_armv8_ce,.-_gcry_aes_enc_armv8_ce;)
@@ -316,7 +346,7 @@ _gcry_aes_dec_armv8_ce:
b.eq .Ldec1_192

.Ldec1_128:
- do_aes_one128(d, imc, v0, v0);
+ do_aes_one128(d, imc, v0, v0, vk0);

.Ldec1_tail:
CLEAR_REG(vk0)
@@ -330,6 +360,7 @@ _gcry_aes_dec_armv8_ce:
CLEAR_REG(vk8)
CLEAR_REG(vk9)
CLEAR_REG(vk10)
+ CLEAR_REG(vklast)
st1 {v0.16b}, [x1]
CLEAR_REG(v0)

@@ -337,19 +368,18 @@ _gcry_aes_dec_armv8_ce:
ret

.Ldec1_192:
- do_aes_one192(d, imc, v0, v0);
+ do_aes_one192(d, imc, v0, v0, vk0);

CLEAR_REG(vk11)
CLEAR_REG(vk12)
b .Ldec1_tail

.Ldec1_256:
- do_aes_one256(d, imc, v0, v0);
+ do_aes_one256(d, imc, v0, v0, vk0);

CLEAR_REG(vk11)
CLEAR_REG(vk12)
CLEAR_REG(vk13)
- CLEAR_REG(vk14)
b .Ldec1_tail
CFI_ENDPROC();
ELF(.size _gcry_aes_dec_armv8_ce,.-_gcry_aes_dec_armv8_ce;)
@@ -381,26 +411,38 @@ _gcry_aes_cbc_enc_armv8_ce:
cbz x4, .Lcbc_enc_skip

cmp w5, #0
- ld1 {v1.16b}, [x3] /* load IV */
- cset x5, eq
+ ld1 {v4.16b}, [x3] /* load IV */
+ csetm x5, eq

aes_preload_keys(x0, w6);
- lsl x5, x5, #4
+ and x5, x5, #16
+
+ ld1 {v3.16b}, [x2], #16; /* load plaintext */
+ mov v0.16b, vk0.16b;
+ sub x4, x4, #1;
+ eor v16.16b, vk0.16b, vklast.16b;
+ eor v4.16b, v4.16b, v3.16b;
+ do_aes_one_part1(e, mc, v4, v0);

- b.eq .Lcbc_enc_loop192
- b.hi .Lcbc_enc_loop256
+ b.eq .Lcbc_enc_entry_192
+ b.hi .Lcbc_enc_entry_256

#define CBC_ENC(bits) \
- .Lcbc_enc_loop##bits: \
- ld1 {v0.16b}, [x2], #16; /* load plaintext */ \
- eor v1.16b, v0.16b, v1.16b; \
- sub x4, x4, #1; \
- \
- do_aes_one##bits(e, mc, v1, v1); \
+ .Lcbc_enc_entry_##bits: \
+ cbz x4, .Lcbc_enc_done_##bits; \
\
- st1 {v1.16b}, [x1], x5; /* store ciphertext */ \
+ .Lcbc_enc_loop_##bits: \
+ do_aes_one_part2_##bits(e, mc, v4, \
+ _(ld1 {v0.16b}, [x2], #16 /* load plaintext */), \
+ _(eor v0.16b, v0.16b, v16.16b)); \
+ sub x4, x4, #1; \
+ eor v3.16b, v4.16b, vklast.16b; \
+ do_aes_one_part1(e, mc, v4, v0); \
+ st1 {v3.16b}, [x1], x5; /* store ciphertext */ \
+ cbnz x4, .Lcbc_enc_loop_##bits; \
\
- cbnz x4, .Lcbc_enc_loop##bits; \
+ .Lcbc_enc_done_##bits: \
+ do_aes_one_part2_##bits(e, mc, v4, __, __); \
b .Lcbc_enc_done;

CBC_ENC(128)
@@ -410,11 +452,14 @@ _gcry_aes_cbc_enc_armv8_ce:
#undef CBC_ENC

.Lcbc_enc_done:
+ eor v3.16b, v4.16b, vklast.16b;
+ st1 {v3.16b}, [x1]; /* store ciphertext */
aes_clear_keys(w6)
+ st1 {v3.16b}, [x3] /* store IV */

- st1 {v1.16b}, [x3] /* store IV */
-
- CLEAR_REG(v1)
+ CLEAR_REG(v16)
+ CLEAR_REG(v4)
+ CLEAR_REG(v3)
CLEAR_REG(v0)

.Lcbc_enc_skip:
@@ -445,7 +490,10 @@ _gcry_aes_cbc_dec_armv8_ce:

cbz x4, .Lcbc_dec_skip

- ld1 {v0.16b}, [x3] /* load IV */
+ add sp, sp, #-64;
+ CFI_ADJUST_CFA_OFFSET(64);
+
+ ld1 {v16.16b}, [x3] /* load IV */

aes_preload_keys(x0, w5);

@@ -457,44 +505,61 @@ _gcry_aes_cbc_dec_armv8_ce:
cmp x4, #4; \
b.lo .Lcbc_dec_loop_##bits; \
\
- .Lcbc_dec_loop4_##bits: \
- \
- ld1 {v1.16b-v4.16b}, [x2], #64; /* load ciphertext */ \
+ ld1 {v0.16b-v3.16b}, [x2], #64; /* load ciphertext */ \
+ cmp x4, #8; \
sub x4, x4, #4; \
- mov v5.16b, v1.16b; \
- mov v6.16b, v2.16b; \
- mov v7.16b, v3.16b; \
- mov v16.16b, v4.16b; \
- cmp x4, #4; \
+ eor v4.16b, v16.16b, vklast.16b; \
+ eor v5.16b, v0.16b, vklast.16b; \
+ eor v6.16b, v1.16b, vklast.16b; \
+ eor v7.16b, v2.16b, vklast.16b; \
+ mov v16.16b, v3.16b; /* next IV */ \
\
- do_aes_4_##bits(d, imc, v1, v2, v3, v4); \
+ do_aes_4_part1(d, imc, v0, v1, v2, v3, vk0); \
+ b.lo .Lcbc_dec_done4_##bits; \
\
- eor v1.16b, v1.16b, v0.16b; \
- eor v2.16b, v2.16b, v5.16b; \
- st1 {v1.16b-v2.16b}, [x1], #32; /* store plaintext */ \
- eor v3.16b, v3.16b, v6.16b; \
- eor v4.16b, v4.16b, v7.16b; \
- mov v0.16b, v16.16b; /* next IV */ \
- st1 {v3.16b-v4.16b}, [x1], #32; /* store plaintext */ \
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */ \
+ \
+ .Lcbc_dec_loop4_##bits: \
+ do_aes_4_part2_##bits(d, imc, v8, v9, v10, v11, v0, v1, v2, v3, v4, v5, v6, v7); \
+ ld1 {v0.16b-v3.16b}, [x2], #64; /* load ciphertext */ \
+ cmp x4, #8; \
+ sub x4, x4, #4; \
+ eor v4.16b, v16.16b, vklast.16b; \
+ eor v5.16b, v0.16b, vklast.16b; \
+ eor v6.16b, v1.16b, vklast.16b; \
+ eor v7.16b, v2.16b, vklast.16b; \
+ mov v16.16b, v3.16b; /* next IV */ \
+ \
+ do_aes_4_part1(d, imc, v0, v1, v2, v3, vk0); \
+ st1 {v8.16b-v11.16b}, [x1], #64; /* store plaintext */ \
\
b.hs .Lcbc_dec_loop4_##bits; \
- CLEAR_REG(v3); \
+ \
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */ \
+ \
+ .Lcbc_dec_done4_##bits: \
+ do_aes_4_part2_##bits(d, imc, v0, v1, v2, v3, v0, v1, v2, v3, v4, v5, v6, v7); \
+ \
CLEAR_REG(v4); \
CLEAR_REG(v5); \
CLEAR_REG(v6); \
CLEAR_REG(v7); \
- CLEAR_REG(v16); \
+ st1 {v0.16b-v3.16b}, [x1], #64; /* store plaintext */ \
+ CLEAR_REG(v0); \
+ CLEAR_REG(v3); \
cbz x4, .Lcbc_dec_done; \
\
.Lcbc_dec_loop_##bits: \
ld1 {v1.16b}, [x2], #16; /* load ciphertext */ \
sub x4, x4, #1; \
+ eor v16.16b, v16.16b, vklast.16b; \
mov v2.16b, v1.16b; \
\
- do_aes_one##bits(d, imc, v1, v1); \
+ do_aes_one_part1(d, imc, v1, vk0); \
+ do_aes_one_part2_##bits(d, imc, v1, __, __); \
+ eor v1.16b, v1.16b, v16.16b; \
\
- eor v1.16b, v1.16b, v0.16b; \
- mov v0.16b, v2.16b; \
+ mov v16.16b, v2.16b; \
st1 {v1.16b}, [x1], #16; /* store plaintext */ \
\
cbnz x4, .Lcbc_dec_loop_##bits; \
@@ -509,12 +574,15 @@ _gcry_aes_cbc_dec_armv8_ce:
.Lcbc_dec_done:
aes_clear_keys(w5)

- st1 {v0.16b}, [x3] /* store IV */
+ st1 {v16.16b}, [x3] /* store IV */

- CLEAR_REG(v0)
+ CLEAR_REG(v16)
CLEAR_REG(v1)
CLEAR_REG(v2)

+ add sp, sp, #64;
+ CFI_ADJUST_CFA_OFFSET(-64);
+
.Lcbc_dec_skip:
ret
CFI_ENDPROC();
@@ -544,9 +612,13 @@ _gcry_aes_ctr_enc_armv8_ce:

cbz x4, .Lctr_enc_skip

- mov x6, #1
+ add x8, sp, #-64
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
+ mov w6, #(1 << 24)
movi v16.16b, #0
- mov v16.D[1], x6
+ mov v16.S[3], w6 /* 1 */

/* load IV */
ldp x9, x10, [x3]
@@ -554,6 +626,9 @@ _gcry_aes_ctr_enc_armv8_ce:
rev x9, x9
rev x10, x10

+ mov x12, #(4 << 56)
+ lsl x11, x10, #56
+
aes_preload_keys(x0, w5);

b.eq .Lctr_enc_entry_192
@@ -564,73 +639,71 @@ _gcry_aes_ctr_enc_armv8_ce:
cmp x4, #4; \
b.lo .Lctr_enc_loop_##bits; \
\
- .Lctr_enc_loop4_##bits: \
- cmp x10, #0xfffffffffffffffc; \
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */ \
+ \
+ adds x11, x11, x12; \
+ add v9.4s, v16.4s, v16.4s; /* 2 */ \
+ add v10.4s, v16.4s, v9.4s; /* 3 */ \
+ add v11.4s, v9.4s, v9.4s; /* 4 */ \
+ mov x7, #1; \
sub x4, x4, #4; \
- b.lo .Lctr_enc_loop4_##bits##_nocarry; \
+ ld1 {v5.16b-v8.16b}, [x2], #64; /* preload ciphertext */ \
+ b.cs .Lctr_enc_carry4_##bits; \
\
- adds x10, x10, #1; \
mov v1.16b, v0.16b; \
- adc x9, x9, xzr; \
- mov v2.D[1], x10; \
- mov v2.D[0], x9; \
- \
- adds x10, x10, #1; \
- rev64 v2.16b, v2.16b; \
- adc x9, x9, xzr; \
- mov v3.D[1], x10; \
- mov v3.D[0], x9; \
- \
- adds x10, x10, #1; \
- rev64 v3.16b, v3.16b; \
- adc x9, x9, xzr; \
- mov v4.D[1], x10; \
- mov v4.D[0], x9; \
+ add x10, x10, #4; \
+ add v2.16b, v0.16b, v16.16b; \
+ add v3.8h, v0.8h, v9.8h; \
+ add v4.4s, v0.4s, v10.4s; \
+ add v0.2d, v0.2d, v11.2d; \
\
- adds x10, x10, #1; \
- rev64 v4.16b, v4.16b; \
- adc x9, x9, xzr; \
- mov v0.D[1], x10; \
- mov v0.D[0], x9; \
- rev64 v0.16b, v0.16b; \
+ .Lctr_enc_entry4_##bits##_carry_done: \
+ mov x7, #0; \
+ cmp x4, #4; \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ b.lo .Lctr_enc_done4_##bits; \
\
- b .Lctr_enc_loop4_##bits##_store_ctr; \
+ st1 {v12.16b-v15.16b}, [x8]; /* store callee saved registers */ \
\
- .Lctr_enc_loop4_##bits##_nocarry: \
+ .Lctr_enc_loop4_##bits: \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ eor v8.16b, v8.16b, vklast.16b; \
+ do_aes_4_part2_##bits(e, mc, v12, v13, v14, v15, v1, v2, v3, v4, v5, v6, v7, v8); \
+ ld1 {v5.16b-v8.16b}, [x2], #64; /* preload ciphertext */ \
+ adds x11, x11, x12; \
+ sub x4, x4, #4; \
+ b.cs .Lctr_enc_carry4_##bits; \
\
- add v3.2d, v16.2d, v16.2d; /* 2 */ \
- rev64 v6.16b, v0.16b; \
+ mov v1.16b, v0.16b; \
add x10, x10, #4; \
- add v4.2d, v3.2d, v16.2d; /* 3 */ \
- add v0.2d, v3.2d, v3.2d; /* 4 */ \
- rev64 v1.16b, v6.16b; \
- add v2.2d, v6.2d, v16.2d; \
- add v3.2d, v6.2d, v3.2d; \
- add v4.2d, v6.2d, v4.2d; \
- add v0.2d, v6.2d, v0.2d; \
- rev64 v2.16b, v2.16b; \
- rev64 v3.16b, v3.16b; \
- rev64 v0.16b, v0.16b; \
- rev64 v4.16b, v4.16b; \
+ add v2.16b, v0.16b, v16.16b; \
+ add v3.8h, v0.8h, v9.8h; \
+ add v4.4s, v0.4s, v10.4s; \
+ add v0.2d, v0.2d, v11.2d; \
\
- .Lctr_enc_loop4_##bits##_store_ctr: \
- \
- st1 {v0.16b}, [x3]; \
+ .Lctr_enc_loop4_##bits##_carry_done: \
cmp x4, #4; \
- ld1 {v5.16b-v7.16b}, [x2], #48; /* preload ciphertext */ \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ st1 {v12.16b-v15.16b}, [x1], #64; /* store plaintext */ \
\
- do_aes_4_##bits(e, mc, v1, v2, v3, v4); \
+ b.hs .Lctr_enc_loop4_##bits; \
\
- eor v1.16b, v1.16b, v5.16b; \
- ld1 {v5.16b}, [x2], #16; /* load ciphertext */ \
- eor v2.16b, v2.16b, v6.16b; \
- eor v3.16b, v3.16b, v7.16b; \
- eor v4.16b, v4.16b, v5.16b; \
- st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
+ ld1 {v12.16b-v15.16b}, [x8]; /* restore callee saved registers */ \
+ \
+ .Lctr_enc_done4_##bits: \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ eor v8.16b, v8.16b, vklast.16b; \
+ do_aes_4_part2_##bits(e, mc, v5, v6, v7, v8, v1, v2, v3, v4, v5, v6, v7, v8); \
+ \
+ st1 {v5.16b-v8.16b}, [x1], #64; /* store plaintext */ \
\
- b.hs .Lctr_enc_loop4_##bits; \
CLEAR_REG(v3); \
CLEAR_REG(v4); \
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */ \
CLEAR_REG(v5); \
CLEAR_REG(v6); \
CLEAR_REG(v7); \
@@ -641,19 +714,48 @@ _gcry_aes_ctr_enc_armv8_ce:
adds x10, x10, #1; \
mov v1.16b, v0.16b; \
adc x9, x9, xzr; \
- mov v0.D[1], x10; \
- mov v0.D[0], x9; \
+ dup v0.2d, x10; \
sub x4, x4, #1; \
+ ins v0.D[0], x9; \
ld1 {v2.16b}, [x2], #16; /* load ciphertext */ \
rev64 v0.16b, v0.16b; \
\
- do_aes_one##bits(e, mc, v1, v1); \
+ do_aes_one_part1(e, mc, v1, vk0); \
+ eor v2.16b, v2.16b, vklast.16b; \
+ do_aes_one_part2_##bits(e, mc, v1, __, __); \
\
- eor v1.16b, v2.16b, v1.16b; \
+ eor v1.16b, v1.16b, v2.16b; \
st1 {v1.16b}, [x1], #16; /* store plaintext */ \
\
cbnz x4, .Lctr_enc_loop_##bits; \
- b .Lctr_enc_done;
+ b .Lctr_enc_done; \
+ \
+ .Lctr_enc_carry4_##bits: \
+ \
+ adds x13, x10, #1; \
+ mov v1.16b, v0.16b; \
+ adc x14, x9, xzr; \
+ dup v2.2d, x13; \
+ adds x13, x10, #2; \
+ ins v2.D[0], x14; \
+ adc x14, x9, xzr; \
+ rev64 v2.16b, v2.16b; \
+ dup v3.2d, x13; \
+ adds x13, x10, #3; \
+ ins v3.D[0], x14; \
+ adc x14, x9, xzr; \
+ rev64 v3.16b, v3.16b; \
+ dup v4.2d, x13; \
+ adds x10, x10, #4; \
+ ins v4.D[0], x14; \
+ adc x9, x9, xzr; \
+ rev64 v4.16b, v4.16b; \
+ dup v0.2d, x10; \
+ ins v0.D[0], x9; \
+ rev64 v0.16b, v0.16b; \
+ \
+ cbz x7, .Lctr_enc_loop4_##bits##_carry_done; \
+ b .Lctr_enc_entry4_##bits##_carry_done;

CTR_ENC(128)
CTR_ENC(192)
@@ -669,6 +771,10 @@ _gcry_aes_ctr_enc_armv8_ce:
CLEAR_REG(v0)
CLEAR_REG(v1)
CLEAR_REG(v2)
+ CLEAR_REG(v16)
+
+ add sp, sp, #128;
+ CFI_ADJUST_CFA_OFFSET(-128);

.Lctr_enc_skip:
ret
@@ -700,6 +806,10 @@ _gcry_aes_ctr32le_enc_armv8_ce:

cbz x4, .Lctr32le_enc_skip

+ add x8, sp, #-64
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
mov w6, #1
movi v16.16b, #0
mov v16.S[0], w6
@@ -712,38 +822,66 @@ _gcry_aes_ctr32le_enc_armv8_ce:
b.eq .Lctr32le_enc_entry_192
b.hi .Lctr32le_enc_entry_256

-#define CTR_ENC(bits) \
+#define CTR32LE_ENC(bits) \
.Lctr32le_enc_entry_##bits: \
cmp x4, #4; \
b.lo .Lctr32le_enc_loop_##bits; \
\
- .Lctr32le_enc_loop4_##bits: \
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */ \
+ add v9.4s, v16.4s, v16.4s; /* 2 */ \
+ cmp x4, #8; \
+ add v10.4s, v9.4s, v16.4s; /* 3 */ \
sub x4, x4, #4; \
+ add v11.4s, v9.4s, v9.4s; /* 4 */ \
+ \
+ ld1 {v5.16b-v8.16b}, [x2], #64; /* preload ciphertext */ \
\
- add v3.4s, v16.4s, v16.4s; /* 2 */ \
mov v1.16b, v0.16b; \
add v2.4s, v0.4s, v16.4s; \
- add v4.4s, v3.4s, v16.4s; /* 3 */ \
- add v6.4s, v3.4s, v3.4s; /* 4 */ \
- add v3.4s, v0.4s, v3.4s; \
- add v4.4s, v0.4s, v4.4s; \
- add v0.4s, v0.4s, v6.4s; \
+ add v3.4s, v0.4s, v9.4s; \
+ add v4.4s, v0.4s, v10.4s; \
+ add v0.4s, v0.4s, v11.4s; \
\
- cmp x4, #4; \
- ld1 {v5.16b-v7.16b}, [x2], #48; /* preload ciphertext */ \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ b.lo .Lctr32le_enc_done4_##bits; \
\
- do_aes_4_##bits(e, mc, v1, v2, v3, v4); \
+ st1 {v12.16b-v15.16b}, [x8]; /* store callee saved registers */ \
\
- eor v1.16b, v1.16b, v5.16b; \
- ld1 {v5.16b}, [x2], #16; /* load ciphertext */ \
- eor v2.16b, v2.16b, v6.16b; \
- eor v3.16b, v3.16b, v7.16b; \
- eor v4.16b, v4.16b, v5.16b; \
- st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
+ .Lctr32le_enc_loop4_##bits: \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ eor v8.16b, v8.16b, vklast.16b; \
+ do_aes_4_part2_##bits(e, mc, v12, v13, v14, v15, v1, v2, v3, v4, v5, v6, v7, v8); \
+ ld1 {v5.16b-v8.16b}, [x2], #64; /* preload ciphertext */ \
+ \
+ cmp x4, #8; \
+ sub x4, x4, #4; \
+ \
+ mov v1.16b, v0.16b; \
+ add v2.4s, v0.4s, v16.4s; \
+ add v3.4s, v0.4s, v9.4s; \
+ add v4.4s, v0.4s, v10.4s; \
+ add v0.4s, v0.4s, v11.4s; \
+ \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ st1 {v12.16b-v15.16b}, [x1], #64; /* store plaintext */ \
\
b.hs .Lctr32le_enc_loop4_##bits; \
+ \
+ ld1 {v12.16b-v15.16b}, [x8]; /* restore callee saved registers */ \
+ \
+ .Lctr32le_enc_done4_##bits: \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ eor v8.16b, v8.16b, vklast.16b; \
+ do_aes_4_part2_##bits(e, mc, v5, v6, v7, v8, v1, v2, v3, v4, v5, v6, v7, v8); \
+ \
+ st1 {v5.16b-v8.16b}, [x1], #64; /* store plaintext */ \
CLEAR_REG(v3); \
CLEAR_REG(v4); \
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */ \
CLEAR_REG(v5); \
CLEAR_REG(v6); \
CLEAR_REG(v7); \
@@ -756,19 +894,21 @@ _gcry_aes_ctr32le_enc_armv8_ce:
sub x4, x4, #1; \
add v0.4s, v0.4s, v16.4s; \
\
- do_aes_one##bits(e, mc, v1, v1); \
+ do_aes_one_part1(e, mc, v1, vk0); \
+ eor v2.16b, v2.16b, vklast.16b; \
+ do_aes_one_part2_##bits(e, mc, v1, __, __); \
\
- eor v1.16b, v2.16b, v1.16b; \
+ eor v1.16b, v1.16b, v2.16b; \
st1 {v1.16b}, [x1], #16; /* store plaintext */ \
\
cbnz x4, .Lctr32le_enc_loop_##bits; \
b .Lctr32le_enc_done;

- CTR_ENC(128)
- CTR_ENC(192)
- CTR_ENC(256)
+ CTR32LE_ENC(128)
+ CTR32LE_ENC(192)
+ CTR32LE_ENC(256)

-#undef CTR_ENC
+#undef CTR32LE_ENC

.Lctr32le_enc_done:
aes_clear_keys(w5)
@@ -778,6 +918,10 @@ _gcry_aes_ctr32le_enc_armv8_ce:
CLEAR_REG(v0)
CLEAR_REG(v1)
CLEAR_REG(v2)
+ CLEAR_REG(v16)
+
+ add sp, sp, #128;
+ CFI_ADJUST_CFA_OFFSET(-128);

.Lctr32le_enc_skip:
ret
@@ -813,21 +957,34 @@ _gcry_aes_cfb_enc_armv8_ce:

aes_preload_keys(x0, w5);

+ ld1 {v1.16b}, [x2], #16; /* load plaintext */
+ eor v3.16b, vk0.16b, vklast.16b;
+ eor v0.16b, v0.16b, vklast.16b;
+ sub x4, x4, #1;
+ mov v4.16b, v3.16b;
+ do_aes_one_part1(e, mc, v0, v4);
+
b.eq .Lcfb_enc_entry_192
b.hi .Lcfb_enc_entry_256

#define CFB_ENC(bits) \
.Lcfb_enc_entry_##bits: \
+ cbz x4, .Lcfb_enc_done_##bits; \
+ \
.Lcfb_enc_loop_##bits: \
- ld1 {v1.16b}, [x2], #16; /* load plaintext */ \
+ eor v2.16b, v1.16b, vklast.16b; \
+ do_aes_one_part2_##bits(e, mc, v0, \
+ _(eor v4.16b, v3.16b, v1.16b), \
+ _(ld1 {v1.16b}, [x2], #16 /* load plaintext */)); \
sub x4, x4, #1; \
- \
- do_aes_one##bits(e, mc, v0, v0); \
- \
- eor v0.16b, v1.16b, v0.16b; \
- st1 {v0.16b}, [x1], #16; /* store ciphertext */ \
- \
+ eor v2.16b, v2.16b, v0.16b; \
+ do_aes_one_part1(e, mc, v0, v4); \
+ st1 {v2.16b}, [x1], #16; /* store ciphertext */ \
cbnz x4, .Lcfb_enc_loop_##bits; \
+ \
+ .Lcfb_enc_done_##bits: \
+ eor v2.16b, v1.16b, vklast.16b; \
+ do_aes_one_part2_##bits(e, mc, v0, __, __); \
b .Lcfb_enc_done;

CFB_ENC(128)
@@ -837,12 +994,16 @@ _gcry_aes_cfb_enc_armv8_ce:
#undef CFB_ENC

.Lcfb_enc_done:
+ eor v2.16b, v2.16b, v0.16b;
+ st1 {v2.16b}, [x1]; /* store ciphertext */
aes_clear_keys(w5)
-
- st1 {v0.16b}, [x3] /* store IV */
+ st1 {v2.16b}, [x3] /* store IV */

CLEAR_REG(v0)
CLEAR_REG(v1)
+ CLEAR_REG(v2)
+ CLEAR_REG(v3)
+ CLEAR_REG(v4)

.Lcfb_enc_skip:
ret
@@ -873,6 +1034,9 @@ _gcry_aes_cfb_dec_armv8_ce:

cbz x4, .Lcfb_dec_skip

+ add sp, sp, #-64;
+ CFI_ADJUST_CFA_OFFSET(64);
+
/* load IV */
ld1 {v0.16b}, [x3]

@@ -886,42 +1050,60 @@ _gcry_aes_cfb_dec_armv8_ce:
cmp x4, #4; \
b.lo .Lcfb_dec_loop_##bits; \
\
- .Lcfb_dec_loop4_##bits: \
- \
- ld1 {v2.16b-v4.16b}, [x2], #48; /* load ciphertext */ \
+ ld1 {v2.16b-v5.16b}, [x2], #64; /* load ciphertext */ \
+ cmp x4, #8; \
mov v1.16b, v0.16b; \
sub x4, x4, #4; \
- cmp x4, #4; \
- mov v5.16b, v2.16b; \
- mov v6.16b, v3.16b; \
- mov v7.16b, v4.16b; \
- ld1 {v0.16b}, [x2], #16; /* load next IV / ciphertext */ \
+ eor v6.16b, v2.16b, vklast.16b; \
+ eor v7.16b, v3.16b, vklast.16b; \
+ eor v16.16b, v4.16b, vklast.16b; \
+ mov v0.16b, v5.16b; /* next IV */ \
+ eor v5.16b, v5.16b, vklast.16b; \
\
- do_aes_4_##bits(e, mc, v1, v2, v3, v4); \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ b.lo .Lcfb_dec_done4_##bits; \
\
- eor v1.16b, v1.16b, v5.16b; \
- eor v2.16b, v2.16b, v6.16b; \
- eor v3.16b, v3.16b, v7.16b; \
- eor v4.16b, v4.16b, v0.16b; \
- st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */ \
+ \
+ .Lcfb_dec_loop4_##bits: \
+ do_aes_4_part2_##bits(e, mc, v8, v9, v10, v11, v1, v2, v3, v4, v6, v7, v16, v5); \
+ ld1 {v2.16b-v5.16b}, [x2], #64; /* load ciphertext */ \
+ cmp x4, #8; \
+ mov v1.16b, v0.16b; \
+ sub x4, x4, #4; \
+ eor v6.16b, v2.16b, vklast.16b; \
+ eor v7.16b, v3.16b, vklast.16b; \
+ eor v16.16b, v4.16b, vklast.16b; \
+ mov v0.16b, v5.16b; /* next IV */ \
+ eor v5.16b, v5.16b, vklast.16b; \
+ \
+ do_aes_4_part1(e, mc, v1, v2, v3, v4, vk0); \
+ st1 {v8.16b-v11.16b}, [x1], #64; /* store plaintext */ \
\
b.hs .Lcfb_dec_loop4_##bits; \
- CLEAR_REG(v3); \
- CLEAR_REG(v4); \
+ \
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */ \
+ \
+ .Lcfb_dec_done4_##bits: \
+ do_aes_4_part2_##bits(e, mc, v1, v2, v3, v4, v1, v2, v3, v4, v6, v7, v16, v5); \
+ \
CLEAR_REG(v5); \
CLEAR_REG(v6); \
CLEAR_REG(v7); \
+ st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
+ CLEAR_REG(v3); \
+ CLEAR_REG(v4); \
cbz x4, .Lcfb_dec_done; \
\
.Lcfb_dec_loop_##bits: \
- \
ld1 {v1.16b}, [x2], #16; /* load ciphertext */ \
- \
sub x4, x4, #1; \
\
- do_aes_one##bits(e, mc, v0, v0); \
+ do_aes_one_part1(e, mc, v0, vk0); \
+ eor v2.16b, v1.16b, vklast.16b; \
+ do_aes_one_part2_##bits(e, mc, v0, __, __); \
+ eor v2.16b, v2.16b, v0.16b; \
\
- eor v2.16b, v1.16b, v0.16b; \
mov v0.16b, v1.16b; \
st1 {v2.16b}, [x1], #16; /* store plaintext */ \
\
@@ -942,6 +1124,10 @@ _gcry_aes_cfb_dec_armv8_ce:
CLEAR_REG(v0)
CLEAR_REG(v1)
CLEAR_REG(v2)
+ CLEAR_REG(v16)
+
+ add sp, sp, #64;
+ CFI_ADJUST_CFA_OFFSET(-64);

.Lcfb_dec_skip:
ret
@@ -972,7 +1158,7 @@ _gcry_aes_ocb_enc_armv8_ce:
* x3: offset
* x4: checksum
* x5: Ltable
- * x6: nblocks (0 < nblocks <= 32)
+ * x6: nblocks (0 < nblocks)
* w7: nrounds
* %st+0: blkn => w12
*/
@@ -982,110 +1168,203 @@ _gcry_aes_ocb_enc_armv8_ce:
ld1 {v0.16b}, [x3] /* load offset */
ld1 {v16.16b}, [x4] /* load checksum */

+ add x16, sp, #-64;
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
aes_preload_keys(x0, w7);

- b.eq .Locb_enc_entry_192
- b.hi .Locb_enc_entry_256
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */

-#define OCB_ENC(bits, ...) \
- .Locb_enc_entry_##bits: \
- cmp x6, #4; \
- add x12, x12, #1; \
- b.lo .Locb_enc_loop_##bits; \
+ eor v0.16b, v0.16b, vk0.16b; /* offset ^ first key */
+ eor v9.16b, vk0.16b, vklast.16b; /* first key ^ last key */
+
+ b.eq .Locb_ecry_entry_192
+ b.hi .Locb_ecry_entry_256
+
+#define OCB_CRYPT(bits, ed, mcimc) \
+ .Locb_##ed##cry_entry_##bits: \
+ /* Get number of blocks to align nblk to 4. */ \
+ neg x13, x12; \
+ add x12, x12, #1; /* Pre-increment nblk for ntz calculation */ \
+ and x13, x13, #(4-1); \
+ cmp x13, x6; \
+ csel x13, x6, x13, hi; \
+ cbz x13, .Locb_##ed##cry_alignment_ok_##bits; \
+ \
+ /* Number of blocks after alignment. */ \
+ sub x14, x6, x13; \
\
- .Locb_enc_loop4_##bits: \
+ /* If number after alignment is less than 4, skip aligned handling \
+ * completely. */ \
+ cmp x14, #4; \
+ csel x13, x6, x13, lo; \
+ \
+ .Locb_##ed##cry_unaligned_entry_##bits: \
+ cmp x13, #4; \
+ \
+ .Locb_##ed##cry_loop1_##bits: \
\
/* Offset_i = Offset_{i-1} xor L_{ntz(i)} */ \
/* Checksum_i = Checksum_{i-1} xor P_i */ \
/* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i) */ \
\
- add w9, w12, #1; \
- add w10, w12, #2; \
- add w11, w12, #3; \
- rbit w8, w12; \
- add w12, w12, #4; \
- rbit w9, w9; \
- rbit w10, w10; \
- rbit w11, w11; \
- clz w8, w8; /* ntz(i+0) */ \
- clz w9, w9; /* ntz(i+1) */ \
- clz w10, w10; /* ntz(i+2) */ \
- clz w11, w11; /* ntz(i+3) */ \
+ rbit x8, x12; \
+ add x12, x12, #1; \
+ clz x8, x8; /* ntz(i) */ \
add x8, x5, x8, lsl #4; \
- ld1 {v1.16b-v4.16b}, [x2], #64; /* load P_i+<0-3> */ \
- add x9, x5, x9, lsl #4; \
- add x10, x5, x10, lsl #4; \
- add x11, x5, x11, lsl #4; \
\
- sub x6, x6, #4; \
+ ld1 {v1.16b}, [x2], #16; /* load plaintext */ \
+ ld1 {v2.16b}, [x8]; /* load L_{ntz(i)} */ \
+ eor v0.16b, v0.16b, v2.16b; \
+ sub x13, x13, #1; \
+ ENC(eor v16.16b, v16.16b, v1.16b); \
+ sub x6, x6, #1; \
\
- ld1 {v5.16b}, [x8]; /* load L_{ntz(i+0)} */ \
- eor v16.16b, v16.16b, v1.16b; /* Checksum_i+0 */ \
- ld1 {v6.16b}, [x9]; /* load L_{ntz(i+1)} */ \
- eor v16.16b, v16.16b, v2.16b; /* Checksum_i+1 */ \
- ld1 {v7.16b}, [x10]; /* load L_{ntz(i+2)} */ \
- eor v16.16b, v16.16b, v3.16b; /* Checksum_i+2 */ \
- eor v5.16b, v5.16b, v0.16b; /* Offset_i+0 */ \
- ld1 {v0.16b}, [x11]; /* load L_{ntz(i+3)} */ \
- eor v16.16b, v16.16b, v4.16b; /* Checksum_i+3 */ \
- eor v6.16b, v6.16b, v5.16b; /* Offset_i+1 */ \
- eor v1.16b, v1.16b, v5.16b; /* P_i+0 xor Offset_i+0 */ \
- eor v7.16b, v7.16b, v6.16b; /* Offset_i+2 */ \
- eor v2.16b, v2.16b, v6.16b; /* P_i+1 xor Offset_i+1 */ \
- eor v0.16b, v0.16b, v7.16b; /* Offset_i+3 */ \
- cmp x6, #4; \
- eor v3.16b, v3.16b, v7.16b; /* P_i+2 xor Offset_i+2 */ \
- eor v4.16b, v4.16b, v0.16b; /* P_i+3 xor Offset_i+3 */ \
+ do_aes_one_part1(ed, mcimc, v1, v0); \
+ eor v2.16b, v0.16b, v9.16b; \
+ do_aes_one_part2_##bits(ed, mcimc, v1, __, __); \
+ eor v1.16b, v1.16b, v2.16b; \
+ st1 {v1.16b}, [x1], #16; /* store ciphertext */ \
+ DEC(eor v16.16b, v16.16b, v1.16b); \
\
- do_aes_4_##bits(e, mc, v1, v2, v3, v4); \
+ cbnz x13, .Locb_##ed##cry_loop1_##bits; \
\
- eor v1.16b, v1.16b, v5.16b; /* xor Offset_i+0 */ \
- eor v2.16b, v2.16b, v6.16b; /* xor Offset_i+1 */ \
- eor v3.16b, v3.16b, v7.16b; /* xor Offset_i+2 */ \
- eor v4.16b, v4.16b, v0.16b; /* xor Offset_i+3 */ \
- st1 {v1.16b-v4.16b}, [x1], #64; \
+ cbz x6, .Locb_##ed##cry_done; \
\
- b.hs .Locb_enc_loop4_##bits; \
- CLEAR_REG(v3); \
- CLEAR_REG(v4); \
- CLEAR_REG(v5); \
- CLEAR_REG(v6); \
- CLEAR_REG(v7); \
- cbz x6, .Locb_enc_done; \
+ /* nblk is now aligned and we have 4 or more blocks. So jump directly to \
+ * aligned processing. */ \
+ b .Locb_##ed##cry_aligned_entry_##bits; \
\
- .Locb_enc_loop_##bits: \
+ .Locb_##ed##cry_alignment_ok_##bits: \
+ cbz x6, .Locb_##ed##cry_done; \
+ \
+ /* Short buffers do not benefit from L-array optimization. */ \
+ cmp x6, #4; \
+ mov x13, x6; \
+ b.lo .Locb_##ed##cry_unaligned_entry_##bits; \
+ \
+ .Locb_##ed##cry_aligned_entry_##bits: \
+ /* Prepare L-array optimization. \
+ * Since nblk is aligned to 4, offsets will have following construction: \
+ * - block1 = ntz{0} = offset ^ L[0] \
+ * - block2 = ntz{1} = offset ^ L[0] ^ L[1] \
+ * - block3 = ntz{0} = offset ^ L[1] \
+ * - block4 = ntz{x} = offset ^ L[1] ^ L[ntz{x}] \
+ */ \
+ ld1 {v10.16b-v11.16b}, [x5]; /* preload L[0] && L[1] */ \
+ mov x15, #4; \
+ \
+ st1 {v12.16b-v15.16b}, [x16]; /* store callee saved registers */ \
\
/* Offset_i = Offset_{i-1} xor L_{ntz(i)} */ \
/* Checksum_i = Checksum_{i-1} xor P_i */ \
/* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i) */ \
\
- rbit x8, x12; \
- add x12, x12, #1; \
- clz x8, x8; /* ntz(i) */ \
- add x8, x5, x8, lsl #4; \
+ add x11, x12, #3; \
+ ld1 {v1.16b-v4.16b}, [x2], #64; /* load P_i+<0-3> */ \
+ rbit x11, x11; \
+ eor v6.16b, v10.16b, v11.16b; /* L[0] ^ L[1] */ \
+ ENC(eor v16.16b, v16.16b, v1.16b); /* Checksum_i+0 */ \
+ add x12, x12, #4; \
+ clz x11, x11; /* ntz(i+3) */ \
+ add x15, x15, #4; \
+ add x11, x5, x11, lsl #4; \
\
- ld1 {v1.16b}, [x2], #16; /* load plaintext */ \
- ld1 {v2.16b}, [x8]; /* load L_{ntz(i)} */ \
- sub x6, x6, #1; \
- eor v0.16b, v0.16b, v2.16b; \
- eor v16.16b, v16.16b, v1.16b; \
- eor v1.16b, v1.16b, v0.16b; \
+ eor v5.16b, v0.16b, v10.16b; /* Offset_i+0 */ \
+ ENC(eor v16.16b, v16.16b, v2.16b); /* Checksum_i+1 */ \
+ ld1 {v8.16b}, [x11]; /* load L_{ntz(i+3)} */ \
+ ENC(eor v16.16b, v16.16b, v3.16b); /* Checksum_i+2 */ \
+ eor v6.16b, v0.16b, v6.16b; /* Offset_i+1 */ \
+ ENC(eor v16.16b, v16.16b, v4.16b); /* Checksum_i+3 */ \
+ eor v7.16b, v0.16b, v11.16b; /* Offset_i+2 */ \
+ eor v8.16b, v8.16b, v11.16b; /* L[1] ^ L[ntz{x}] */ \
+ cmp x15, x13; \
+ eor v0.16b, v0.16b, v8.16b; /* Offset_i+3 */ \
+ \
+ do_aes_4_part1_multikey(ed, mcimc, v1, v2, v3, v4, v5, v6, v7, v0); /* P_i+j xor Offset_i+j */ \
+ b.hi .Locb_##ed##cry_aligned_done4_##bits; \
+ \
+ .Locb_##ed##cry_aligned_loop4_##bits: \
+ add x11, x12, #3; \
+ eor v5.16b, v5.16b, v9.16b; \
+ eor v6.16b, v6.16b, v9.16b; \
+ rbit x11, x11; \
+ eor v7.16b, v7.16b, v9.16b; \
+ eor v8.16b, v0.16b, v9.16b; \
+ clz x11, x11; /* ntz(i+3) */ \
+ do_aes_4_part2_##bits(ed, mcimc, v12, v13, v14, v15, v1, v2, v3, v4, v5, v6, v7, v8); /* xor Offset_i+j */ \
\
- do_aes_one##bits(e, mc, v1, v1); \
+ /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */ \
+ /* Checksum_i = Checksum_{i-1} xor P_i */ \
+ /* C_i = Offset_i xor ENCIPHER(K, P_i xor Offset_i) */ \
\
- eor v1.16b, v1.16b, v0.16b; \
- st1 {v1.16b}, [x1], #16; /* store ciphertext */ \
+ add x12, x12, #4; \
+ ld1 {v1.16b-v4.16b}, [x2], #64; /* load P_i+<0-3> */ \
+ eor v6.16b, v10.16b, v11.16b; /* L[0] ^ L[1] */ \
+ add x15, x15, #4; \
+ DEC(eor v16.16b, v16.16b, v12.16b); /* Checksum_i+0 */ \
+ ENC(eor v16.16b, v16.16b, v1.16b); /* Checksum_i+0 */ \
+ add x11, x5, x11, lsl #4; \
+ \
+ eor v5.16b, v0.16b, v10.16b; /* Offset_i+0 */ \
+ ENC(eor v16.16b, v16.16b, v2.16b); /* Checksum_i+1 */ \
+ DEC(eor v16.16b, v16.16b, v13.16b); /* Checksum_1+2 */ \
+ ld1 {v8.16b}, [x11]; /* load L_{ntz(i+3)} */ \
+ ENC(eor v16.16b, v16.16b, v3.16b); /* Checksum_i+2 */ \
+ DEC(eor v16.16b, v16.16b, v14.16b); /* Checksum_i+0+3 */ \
+ eor v6.16b, v0.16b, v6.16b; /* Offset_i+1 */ \
+ ENC(eor v16.16b, v16.16b, v4.16b); /* Checksum_i+3 */ \
+ DEC(eor v16.16b, v16.16b, v15.16b); /* Checksum_i+0+1+2 */ \
+ eor v7.16b, v0.16b, v11.16b; /* Offset_i+2 */ \
+ eor v8.16b, v8.16b, v11.16b; /* L[1] ^ L[ntz{x}] */ \
+ cmp x15, x13; \
+ eor v0.16b, v0.16b, v8.16b; /* Offset_i+3 */ \
+ \
+ do_aes_4_part1_multikey(ed, mcimc, v1, v2, v3, v4, v5, v6, v7, v0); /* P_i+j xor Offset_i+j */ \
+ st1 {v12.16b-v15.16b}, [x1], #64; \
+ \
+ b.ls .Locb_##ed##cry_aligned_loop4_##bits; \
+ \
+ .Locb_##ed##cry_aligned_done4_##bits: \
+ eor v5.16b, v5.16b, v9.16b; \
+ eor v6.16b, v6.16b, v9.16b; \
+ eor v7.16b, v7.16b, v9.16b; \
+ eor v8.16b, v0.16b, v9.16b; \
+ do_aes_4_part2_##bits(ed, mcimc, v1, v2, v3, v4, v1, v2, v3, v4, v5, v6, v7, v8); /* xor Offset_i+j */ \
+ DEC(eor v16.16b, v16.16b, v1.16b); /* Checksum_i+0 */ \
+ DEC(eor v5.16b, v2.16b, v3.16b); /* Checksum_1+2 */ \
+ DEC(eor v16.16b, v16.16b, v4.16b); /* Checksum_i+0+3 */ \
+ st1 {v1.16b-v4.16b}, [x1], #64; \
+ DEC(eor v16.16b, v16.16b, v5.16b); /* Checksum_i+0+1+2 */ \
\
- cbnz x6, .Locb_enc_loop_##bits; \
- b .Locb_enc_done;
+ sub x15, x15, #4; \
+ CLEAR_REG(v3); \
+ CLEAR_REG(v4); \
+ ld1 {v12.16b-v15.16b}, [x16]; /* restore callee saved registers */ \
+ sub x13, x13, x15; \
+ sub x6, x6, x15; \
+ CLEAR_REG(v5); \
+ CLEAR_REG(v6); \
+ \
+ /* Handle tailing 1…3 blocks in unaligned loop. */ \
+ mov x13, x6; \
+ cbnz x6, .Locb_##ed##cry_unaligned_entry_##bits; \
+ \
+ b .Locb_##ed##cry_done;

- OCB_ENC(128)
- OCB_ENC(192)
- OCB_ENC(256)
+#define ENC(...) __VA_ARGS__
+#define DEC(...) /*_*/
+ OCB_CRYPT(128, e, mc)
+ OCB_CRYPT(192, e, mc)
+ OCB_CRYPT(256, e, mc)
+#undef ENC
+#undef DEC

-#undef OCB_ENC
+.Locb_ecry_done:
+ eor v0.16b, v0.16b, vk0.16b; /* restore offset */

-.Locb_enc_done:
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */
aes_clear_keys(w7)

st1 {v16.16b}, [x4] /* store checksum */
@@ -1094,8 +1373,12 @@ _gcry_aes_ocb_enc_armv8_ce:
CLEAR_REG(v0)
CLEAR_REG(v1)
CLEAR_REG(v2)
+ CLEAR_REG(v7)
CLEAR_REG(v16)

+ add sp, sp, #128;
+ CFI_ADJUST_CFA_OFFSET(-128);
+
ret
CFI_ENDPROC();
ELF(.size _gcry_aes_ocb_enc_armv8_ce,.-_gcry_aes_ocb_enc_armv8_ce;)
@@ -1124,7 +1407,7 @@ _gcry_aes_ocb_dec_armv8_ce:
* x3: offset
* x4: checksum
* x5: Ltable
- * x6: nblocks (0 < nblocks <= 32)
+ * x6: nblocks (0 < nblocks)
* w7: nrounds
* %st+0: blkn => w12
*/
@@ -1134,110 +1417,34 @@ _gcry_aes_ocb_dec_armv8_ce:
ld1 {v0.16b}, [x3] /* load offset */
ld1 {v16.16b}, [x4] /* load checksum */

+ add x16, sp, #-64;
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
aes_preload_keys(x0, w7);

- b.eq .Locb_dec_entry_192
- b.hi .Locb_dec_entry_256
+ st1 {v8.16b-v11.16b}, [sp]; /* store callee saved registers */

-#define OCB_DEC(bits) \
- .Locb_dec_entry_##bits: \
- cmp x6, #4; \
- add w12, w12, #1; \
- b.lo .Locb_dec_loop_##bits; \
- \
- .Locb_dec_loop4_##bits: \
- \
- /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */ \
- /* P_i = Offset_i xor DECIPHER(K, C_i xor Offset_i) */ \
- /* Checksum_i = Checksum_{i-1} xor P_i */ \
- \
- add w9, w12, #1; \
- add w10, w12, #2; \
- add w11, w12, #3; \
- rbit w8, w12; \
- add w12, w12, #4; \
- rbit w9, w9; \
- rbit w10, w10; \
- rbit w11, w11; \
- clz w8, w8; /* ntz(i+0) */ \
- clz w9, w9; /* ntz(i+1) */ \
- clz w10, w10; /* ntz(i+2) */ \
- clz w11, w11; /* ntz(i+3) */ \
- add x8, x5, x8, lsl #4; \
- ld1 {v1.16b-v4.16b}, [x2], #64; /* load C_i+<0-3> */ \
- add x9, x5, x9, lsl #4; \
- add x10, x5, x10, lsl #4; \
- add x11, x5, x11, lsl #4; \
- \
- sub x6, x6, #4; \
- \
- ld1 {v5.16b}, [x8]; /* load L_{ntz(i+0)} */ \
- ld1 {v6.16b}, [x9]; /* load L_{ntz(i+1)} */ \
- ld1 {v7.16b}, [x10]; /* load L_{ntz(i+2)} */ \
- eor v5.16b, v5.16b, v0.16b; /* Offset_i+0 */ \
- ld1 {v0.16b}, [x11]; /* load L_{ntz(i+3)} */ \
- eor v6.16b, v6.16b, v5.16b; /* Offset_i+1 */ \
- eor v1.16b, v1.16b, v5.16b; /* C_i+0 xor Offset_i+0 */ \
- eor v7.16b, v7.16b, v6.16b; /* Offset_i+2 */ \
- eor v2.16b, v2.16b, v6.16b; /* C_i+1 xor Offset_i+1 */ \
- eor v0.16b, v0.16b, v7.16b; /* Offset_i+3 */ \
- cmp x6, #4; \
- eor v3.16b, v3.16b, v7.16b; /* C_i+2 xor Offset_i+2 */ \
- eor v4.16b, v4.16b, v0.16b; /* C_i+3 xor Offset_i+3 */ \
- \
- do_aes_4_##bits(d, imc, v1, v2, v3, v4); \
- \
- eor v1.16b, v1.16b, v5.16b; /* xor Offset_i+0 */ \
- eor v2.16b, v2.16b, v6.16b; /* xor Offset_i+1 */ \
- eor v16.16b, v16.16b, v1.16b; /* Checksum_i+0 */ \
- eor v3.16b, v3.16b, v7.16b; /* xor Offset_i+2 */ \
- eor v16.16b, v16.16b, v2.16b; /* Checksum_i+1 */ \
- eor v4.16b, v4.16b, v0.16b; /* xor Offset_i+3 */ \
- eor v16.16b, v16.16b, v3.16b; /* Checksum_i+2 */ \
- eor v16.16b, v16.16b, v4.16b; /* Checksum_i+3 */ \
- st1 {v1.16b-v4.16b}, [x1], #64; \
- \
- b.hs .Locb_dec_loop4_##bits; \
- CLEAR_REG(v3); \
- CLEAR_REG(v4); \
- CLEAR_REG(v5); \
- CLEAR_REG(v6); \
- CLEAR_REG(v7); \
- cbz x6, .Locb_dec_done; \
- \
- .Locb_dec_loop_##bits: \
- \
- /* Offset_i = Offset_{i-1} xor L_{ntz(i)} */ \
- /* P_i = Offset_i xor DECIPHER(K, C_i xor Offset_i) */ \
- /* Checksum_i = Checksum_{i-1} xor P_i */ \
- \
- rbit w8, w12; \
- add w12, w12, #1; \
- clz w8, w8; /* ntz(i) */ \
- add x8, x5, x8, lsl #4; \
- \
- ld1 {v1.16b}, [x2], #16; /* load ciphertext */ \
- ld1 {v2.16b}, [x8]; /* load L_{ntz(i)} */ \
- sub x6, x6, #1; \
- eor v0.16b, v0.16b, v2.16b; \
- eor v1.16b, v1.16b, v0.16b; \
- \
- do_aes_one##bits(d, imc, v1, v1) \
- \
- eor v1.16b, v1.16b, v0.16b; \
- st1 {v1.16b}, [x1], #16; /* store plaintext */ \
- eor v16.16b, v16.16b, v1.16b; \
- \
- cbnz x6, .Locb_dec_loop_##bits; \
- b .Locb_dec_done;
+ eor v0.16b, v0.16b, vk0.16b; /* offset ^ first key */
+ eor v9.16b, vk0.16b, vklast.16b; /* first key ^ last key */
+
+ b.eq .Locb_dcry_entry_192
+ b.hi .Locb_dcry_entry_256
+
+#define ENC(...) /*_*/
+#define DEC(...) __VA_ARGS__
+ OCB_CRYPT(128, d, imc)
+ OCB_CRYPT(192, d, imc)
+ OCB_CRYPT(256, d, imc)
+#undef ENC
+#undef DEC

- OCB_DEC(128)
- OCB_DEC(192)
- OCB_DEC(256)
+#undef OCB_CRYPT

-#undef OCB_DEC
+.Locb_dcry_done:
+ eor v0.16b, v0.16b, vk0.16b; /* restore offset */

-.Locb_dec_done:
+ ld1 {v8.16b-v11.16b}, [sp]; /* restore callee saved registers */
aes_clear_keys(w7)

st1 {v16.16b}, [x4] /* store checksum */
@@ -1248,6 +1455,9 @@ _gcry_aes_ocb_dec_armv8_ce:
CLEAR_REG(v2)
CLEAR_REG(v16)

+ add sp, sp, #128;
+ CFI_ADJUST_CFA_OFFSET(-128);
+
ret
CFI_ENDPROC();
ELF(.size _gcry_aes_ocb_dec_armv8_ce,.-_gcry_aes_ocb_dec_armv8_ce;)
@@ -1371,7 +1581,7 @@ _gcry_aes_ocb_auth_armv8_ce:
eor v0.16b, v0.16b, v2.16b; \
eor v1.16b, v1.16b, v0.16b; \
\
- do_aes_one##bits(e, mc, v1, v1) \
+ do_aes_one##bits(e, mc, v1, v1, vk0) \
\
eor v16.16b, v16.16b, v1.16b; \
\
@@ -1425,6 +1635,10 @@ _gcry_aes_xts_enc_armv8_ce:

cbz x4, .Lxts_enc_skip

+ add x16, sp, #-64;
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
/* load tweak */
ld1 {v0.16b}, [x3]

@@ -1435,18 +1649,66 @@ _gcry_aes_xts_enc_armv8_ce:
mov v16.D[1], x7

aes_preload_keys(x0, w5);
+ eor vklast.16b, vklast.16b, vk0.16b;

- b.eq .Lxts_enc_entry_192
- b.hi .Lxts_enc_entry_256
+ b.eq .Lxts_ecry_entry_192
+ b.hi .Lxts_ecry_entry_256

-#define XTS_ENC(bits) \
- .Lxts_enc_entry_##bits: \
+#define XTS_CRYPT(bits, ed, mcimc) \
+ .Lxts_##ed##cry_entry_##bits: \
cmp x4, #4; \
- b.lo .Lxts_enc_loop_##bits; \
+ b.lo .Lxts_##ed##cry_loop_##bits; \
\
- .Lxts_enc_loop4_##bits: \
+ st1 {v8.16b}, [sp]; /* store callee saved registers */ \
+ ext v4.16b, v0.16b, v0.16b, #8; \
+ mov v8.16b, v0.16b; \
+ \
+ sshr v2.2d, v4.2d, #63; \
+ add v5.2d, v0.2d, v0.2d; \
+ and v2.16b, v2.16b, v16.16b; \
+ add v4.2d, v4.2d, v4.2d; \
+ eor v5.16b, v5.16b, v2.16b; \
+ \
+ sshr v2.2d, v4.2d, #63; \
+ add v6.2d, v5.2d, v5.2d; \
+ and v2.16b, v2.16b, v16.16b; \
+ add v4.2d, v4.2d, v4.2d; \
+ eor v6.16b, v6.16b, v2.16b; \
+ \
+ sshr v2.2d, v4.2d, #63; \
+ add v7.2d, v6.2d, v6.2d; \
+ and v2.16b, v2.16b, v16.16b; \
+ add v4.2d, v4.2d, v4.2d; \
+ eor v7.16b, v7.16b, v2.16b; \
+ \
+ sshr v2.2d, v4.2d, #63; \
+ add v3.2d, v7.2d, v7.2d; \
+ and v2.16b, v2.16b, v16.16b; \
+ add v4.2d, v4.2d, v4.2d; \
+ eor v0.16b, v3.16b, v2.16b; \
+ ld1 {v1.16b-v4.16b}, [x2], #64; /* load plaintext */ \
+ cmp x4, #8; \
+ sub x4, x4, #4; \
+ \
+ eor v8.16b, v8.16b, vk0.16b; \
+ eor v5.16b, v5.16b, vk0.16b; \
+ eor v6.16b, v6.16b, vk0.16b; \
+ eor v7.16b, v7.16b, vk0.16b; \
+ \
+ do_aes_4_part1_multikey(ed, mcimc, v1, v2, v3, v4, v8, v5, v6, v7); \
+ b.lo .Lxts_##ed##cry_done4_##bits; \
+ \
+ st1 {v9.16b-v12.16b}, [x16]; /* store callee saved registers */ \
+ \
+ .Lxts_##ed##cry_loop4_##bits: \
+ eor v8.16b, v8.16b, vklast.16b; \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ do_aes_4_part2_##bits(ed, mcimc, v9, v10, v11, v12, v1, v2, v3, v4, v8, v5, v6, v7); \
\
ext v4.16b, v0.16b, v0.16b, #8; \
+ mov v8.16b, v0.16b; \
\
sshr v2.2d, v4.2d, #63; \
add v5.2d, v0.2d, v0.2d; \
@@ -1470,62 +1732,66 @@ _gcry_aes_xts_enc_armv8_ce:
add v3.2d, v7.2d, v7.2d; \
and v2.16b, v2.16b, v16.16b; \
add v4.2d, v4.2d, v4.2d; \
- eor v3.16b, v3.16b, v2.16b; \
- ld1 {v1.16b-v2.16b}, [x2], #32; /* load plaintext */ \
- st1 {v3.16b}, [x3]; \
+ eor v0.16b, v3.16b, v2.16b; \
+ ld1 {v1.16b-v4.16b}, [x2], #64; /* load plaintext */ \
+ cmp x4, #8; \
sub x4, x4, #4; \
- eor v1.16b, v1.16b, v0.16b; \
\
- ld1 {v3.16b-v4.16b}, [x2], #32; /* load plaintext */ \
- cmp x4, #4; \
- eor v2.16b, v2.16b, v5.16b; \
- eor v3.16b, v3.16b, v6.16b; \
- eor v4.16b, v4.16b, v7.16b; \
+ eor v8.16b, v8.16b, vk0.16b; \
+ eor v5.16b, v5.16b, vk0.16b; \
+ eor v6.16b, v6.16b, vk0.16b; \
+ eor v7.16b, v7.16b, vk0.16b; \
\
- do_aes_4_##bits(e, mc, v1, v2, v3, v4); \
+ do_aes_4_part1_multikey(ed, mcimc, v1, v2, v3, v4, v8, v5, v6, v7); \
+ \
+ st1 {v9.16b-v12.16b}, [x1], #64; /* store plaintext */ \
+ \
+ b.hs .Lxts_##ed##cry_loop4_##bits; \
+ \
+ ld1 {v9.16b-v12.16b}, [x16]; /* restore callee saved registers */ \
+ \
+ .Lxts_##ed##cry_done4_##bits: \
+ eor v8.16b, v8.16b, vklast.16b; \
+ eor v5.16b, v5.16b, vklast.16b; \
+ eor v6.16b, v6.16b, vklast.16b; \
+ eor v7.16b, v7.16b, vklast.16b; \
+ do_aes_4_part2_##bits(ed, mcimc, v1, v2, v3, v4, v1, v2, v3, v4, v8, v5, v6, v7); \
\
- eor v1.16b, v1.16b, v0.16b; \
- ld1 {v0.16b}, [x3]; \
- eor v2.16b, v2.16b, v5.16b; \
- eor v3.16b, v3.16b, v6.16b; \
- eor v4.16b, v4.16b, v7.16b; \
st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
\
- b.hs .Lxts_enc_loop4_##bits; \
- CLEAR_REG(v3); \
CLEAR_REG(v4); \
+ ld1 {v8.16b}, [sp]; /* restore callee saved registers */ \
CLEAR_REG(v5); \
CLEAR_REG(v6); \
CLEAR_REG(v7); \
- cbz x4, .Lxts_enc_done; \
+ cbz x4, .Lxts_##ed##cry_done; \
\
- .Lxts_enc_loop_##bits: \
+ .Lxts_##ed##cry_loop_##bits: \
\
ld1 {v1.16b}, [x2], #16; /* load plaintext */ \
ext v3.16b, v0.16b, v0.16b, #8; \
- mov v2.16b, v0.16b; \
+ eor v2.16b, v0.16b, vk0.16b; \
sshr v3.2d, v3.2d, #63; \
add v0.2d, v0.2d, v0.2d; \
and v3.16b, v3.16b, v16.16b; \
- eor v1.16b, v1.16b, v2.16b; \
- eor v0.16b, v0.16b, v3.16b; \
sub x4, x4, #1; \
+ eor v0.16b, v0.16b, v3.16b; \
\
- do_aes_one##bits(e, mc, v1, v1); \
- \
+ do_aes_one_part1(ed, mcimc, v1, v2); \
+ eor v2.16b, v2.16b, vklast.16b; \
+ do_aes_one_part2_##bits(ed, mcimc, v1, __, __); \
eor v1.16b, v1.16b, v2.16b; \
+ \
st1 {v1.16b}, [x1], #16; /* store ciphertext */ \
\
- cbnz x4, .Lxts_enc_loop_##bits; \
- b .Lxts_enc_done;
+ cbnz x4, .Lxts_##ed##cry_loop_##bits; \
+ b .Lxts_##ed##cry_done;

- XTS_ENC(128)
- XTS_ENC(192)
- XTS_ENC(256)
+ XTS_CRYPT(128, e, mc)
+ XTS_CRYPT(192, e, mc)
+ XTS_CRYPT(256, e, mc)

-#undef XTS_ENC
-
-.Lxts_enc_done:
+.Lxts_ecry_done:
aes_clear_keys(w5)

st1 {v0.16b}, [x3] /* store tweak */
@@ -1533,6 +1799,11 @@ _gcry_aes_xts_enc_armv8_ce:
CLEAR_REG(v0)
CLEAR_REG(v1)
CLEAR_REG(v2)
+ CLEAR_REG(v3)
+ CLEAR_REG(v16)
+
+ add sp, sp, 128;
+ CFI_ADJUST_CFA_OFFSET(-128);

.Lxts_enc_skip:
ret
@@ -1565,6 +1836,10 @@ _gcry_aes_xts_dec_armv8_ce:

cbz x4, .Lxts_dec_skip

+ add x16, sp, #-64;
+ add sp, sp, #-128;
+ CFI_ADJUST_CFA_OFFSET(128);
+
/* load tweak */
ld1 {v0.16b}, [x3]

@@ -1575,97 +1850,18 @@ _gcry_aes_xts_dec_armv8_ce:
mov v16.D[1], x7

aes_preload_keys(x0, w5);
+ eor vklast.16b, vklast.16b, vk0.16b;

- b.eq .Lxts_dec_entry_192
- b.hi .Lxts_dec_entry_256
+ b.eq .Lxts_dcry_entry_192
+ b.hi .Lxts_dcry_entry_256

-#define XTS_DEC(bits) \
- .Lxts_dec_entry_##bits: \
- cmp x4, #4; \
- b.lo .Lxts_dec_loop_##bits; \
- \
- .Lxts_dec_loop4_##bits: \
- \
- ext v4.16b, v0.16b, v0.16b, #8; \
- \
- sshr v2.2d, v4.2d, #63; \
- add v5.2d, v0.2d, v0.2d; \
- and v2.16b, v2.16b, v16.16b; \
- add v4.2d, v4.2d, v4.2d; \
- eor v5.16b, v5.16b, v2.16b; \
- \
- sshr v2.2d, v4.2d, #63; \
- add v6.2d, v5.2d, v5.2d; \
- and v2.16b, v2.16b, v16.16b; \
- add v4.2d, v4.2d, v4.2d; \
- eor v6.16b, v6.16b, v2.16b; \
- \
- sshr v2.2d, v4.2d, #63; \
- add v7.2d, v6.2d, v6.2d; \
- and v2.16b, v2.16b, v16.16b; \
- add v4.2d, v4.2d, v4.2d; \
- eor v7.16b, v7.16b, v2.16b; \
- \
- sshr v2.2d, v4.2d, #63; \
- add v3.2d, v7.2d, v7.2d; \
- and v2.16b, v2.16b, v16.16b; \
- add v4.2d, v4.2d, v4.2d; \
- eor v3.16b, v3.16b, v2.16b; \
- ld1 {v1.16b-v2.16b}, [x2], #32; /* load plaintext */ \
- st1 {v3.16b}, [x3]; \
- sub x4, x4, #4; \
- eor v1.16b, v1.16b, v0.16b; \
- \
- ld1 {v3.16b-v4.16b}, [x2], #32; /* load plaintext */ \
- cmp x4, #4; \
- eor v2.16b, v2.16b, v5.16b; \
- eor v3.16b, v3.16b, v6.16b; \
- eor v4.16b, v4.16b, v7.16b; \
- \
- do_aes_4_##bits(d, imc, v1, v2, v3, v4); \
- \
- eor v1.16b, v1.16b, v0.16b; \
- ld1 {v0.16b}, [x3]; \
- eor v2.16b, v2.16b, v5.16b; \
- eor v3.16b, v3.16b, v6.16b; \
- eor v4.16b, v4.16b, v7.16b; \
- st1 {v1.16b-v4.16b}, [x1], #64; /* store plaintext */ \
- \
- b.hs .Lxts_dec_loop4_##bits; \
- CLEAR_REG(v3); \
- CLEAR_REG(v4); \
- CLEAR_REG(v5); \
- CLEAR_REG(v6); \
- CLEAR_REG(v7); \
- cbz x4, .Lxts_dec_done; \
- \
- .Lxts_dec_loop_##bits: \
- \
- ld1 {v1.16b}, [x2], #16; /* load plaintext */ \
- ext v3.16b, v0.16b, v0.16b, #8; \
- mov v2.16b, v0.16b; \
- sshr v3.2d, v3.2d, #63; \
- add v0.2d, v0.2d, v0.2d; \
- and v3.16b, v3.16b, v16.16b; \
- eor v1.16b, v1.16b, v2.16b; \
- eor v0.16b, v0.16b, v3.16b; \
- sub x4, x4, #1; \
- \
- do_aes_one##bits(d, imc, v1, v1); \
- \
- eor v1.16b, v1.16b, v2.16b; \
- st1 {v1.16b}, [x1], #16; /* store ciphertext */ \
- \
- cbnz x4, .Lxts_dec_loop_##bits; \
- b .Lxts_dec_done;
-
- XTS_DEC(128)
- XTS_DEC(192)
- XTS_DEC(256)
+ XTS_CRYPT(128, d, imc)
+ XTS_CRYPT(192, d, imc)
+ XTS_CRYPT(256, d, imc)

-#undef XTS_DEC
+#undef XTS_CRYPT

-.Lxts_dec_done:
+.Lxts_dcry_done:
aes_clear_keys(w5)

st1 {v0.16b}, [x3] /* store tweak */
@@ -1674,6 +1870,9 @@ _gcry_aes_xts_dec_armv8_ce:
CLEAR_REG(v1)
CLEAR_REG(v2)

+ add sp, sp, 128;
+ CFI_ADJUST_CFA_OFFSET(-128);
+
.Lxts_dec_skip:
ret
CFI_ENDPROC();
--
2.32.0


_______________________________________________
Gcrypt-devel mailing list
Gcrypt-devel@gnupg.org
http://lists.gnupg.org/mailman/listinfo/gcrypt-devel