Add ARMv8/CE acceleration for AES-XTS

* cipher/rijndael-armv8-aarch32-ce.S (_gcry_aes_xts_enc_armv8_ce)
(_gcry_aes_xts_dec_armv8_ce): New.
* cipher/rijndael-armv8-aarch64-ce.S (_gcry_aes_xts_enc_armv8_ce)
(_gcry_aes_xts_dec_armv8_ce): New.
* cipher/rijndael-armv8-ce.c (_gcry_aes_xts_enc_armv8_ce)
(_gcry_aes_xts_dec_armv8_ce, xts_crypt_fn_t)
(_gcry_aes_armv8_ce_xts_crypt): New.
* cipher/rijndael.c (_gcry_aes_armv8_ce_xts_crypt): New.
(_gcry_aes_xts_crypt) [USE_ARM_CE]: New.
--

Benchmark on Cortex-A53 (AArch64, 1152 Mhz):

Before:
 AES            |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      4.88 ns/B     195.5 MiB/s      5.62 c/B
        XTS dec |      4.94 ns/B     192.9 MiB/s      5.70 c/B
                =
 AES192         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      5.55 ns/B     171.8 MiB/s      6.39 c/B
        XTS dec |      5.61 ns/B     169.9 MiB/s      6.47 c/B
                =
 AES256         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      6.22 ns/B     153.3 MiB/s      7.17 c/B
        XTS dec |      6.29 ns/B     151.7 MiB/s      7.24 c/B
                =

After (~2.6x faster):
 AES            |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      1.83 ns/B     520.9 MiB/s      2.11 c/B
        XTS dec |      1.82 ns/B     524.9 MiB/s      2.09 c/B
                =
 AES192         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      1.97 ns/B     483.3 MiB/s      2.27 c/B
        XTS dec |      1.96 ns/B     486.9 MiB/s      2.26 c/B
                =
 AES256         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      2.11 ns/B     450.9 MiB/s      2.44 c/B
        XTS dec |      2.10 ns/B     453.8 MiB/s      2.42 c/B
                =

Benchmark on Cortex-A53 (AArch32, 1152 Mhz):

Before:
 AES            |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      6.52 ns/B     146.2 MiB/s      7.51 c/B
        XTS dec |      6.57 ns/B     145.2 MiB/s      7.57 c/B
                =
 AES192         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      7.10 ns/B     134.3 MiB/s      8.18 c/B
        XTS dec |      7.11 ns/B     134.2 MiB/s      8.19 c/B
                =
 AES256         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      7.30 ns/B     130.7 MiB/s      8.41 c/B
        XTS dec |      7.38 ns/B     129.3 MiB/s      8.50 c/B
                =

After (~2.7x faster):
Cipher:
 AES            |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      2.33 ns/B     409.6 MiB/s      2.68 c/B
        XTS dec |      2.35 ns/B     405.3 MiB/s      2.71 c/B
                =
 AES192         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      2.53 ns/B     377.6 MiB/s      2.91 c/B
        XTS dec |      2.54 ns/B     375.5 MiB/s      2.93 c/B
                =
 AES256         |  nanosecs/byte   mebibytes/sec   cycles/byte
        XTS enc |      2.75 ns/B     346.8 MiB/s      3.17 c/B
        XTS dec |      2.76 ns/B     345.2 MiB/s      3.18 c/B
                =

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

1 files changed, 28 insertions, 0 deletions

diff --git a/cipher/rijndael-armv8-ce.c b/cipher/rijndael-armv8-ce.c
index 334cf684..6af7108f 100644
--- a/cipher/rijndael-armv8-ce.c
+++ b/cipher/rijndael-armv8-ce.c
@@ -101,6 +101,16 @@ extern void _gcry_aes_ocb_auth_armv8_ce (const void *keysched,
                                          size_t nblocks,
                                          unsigned int nrounds,
                                          unsigned int blkn);
+extern void _gcry_aes_xts_enc_armv8_ce (const void *keysched,
+                                        unsigned char *outbuf,
+                                        const unsigned char *inbuf,
+                                        unsigned char *tweak,
+                                        size_t nblocks, unsigned int nrounds);
+extern void _gcry_aes_xts_dec_armv8_ce (const void *keysched,
+                                        unsigned char *outbuf,
+                                        const unsigned char *inbuf,
+                                        unsigned char *tweak,
+                                        size_t nblocks, unsigned int nrounds);
 
 typedef void (*ocb_crypt_fn_t) (const void *keysched, unsigned char *outbuf,
                                 const unsigned char *inbuf,
@@ -108,6 +118,11 @@ typedef void (*ocb_crypt_fn_t) (const void *keysched, unsigned char *outbuf,
                                 unsigned char *L_table, size_t nblocks,
                                 unsigned int nrounds, unsigned int blkn);
 
+typedef void (*xts_crypt_fn_t) (const void *keysched, unsigned char *outbuf,
+                                const unsigned char *inbuf,
+                                unsigned char *tweak, size_t nblocks,
+                                unsigned int nrounds);
+
 void
 _gcry_aes_armv8_ce_setkey (RIJNDAEL_context *ctx, const byte *key)
 {
@@ -361,4 +376,17 @@ _gcry_aes_armv8_ce_ocb_auth (gcry_cipher_hd_t c, void *abuf_arg,
 			      nblocks, nrounds, (unsigned int)blkn);
 }
 
+void
+_gcry_aes_armv8_ce_xts_crypt (RIJNDAEL_context *ctx, unsigned char *tweak,
+			      unsigned char *outbuf, const unsigned char *inbuf,
+			      size_t nblocks, int encrypt)
+{
+  const void *keysched = encrypt ? ctx->keyschenc32 : ctx->keyschdec32;
+  xts_crypt_fn_t crypt_fn = encrypt ? _gcry_aes_xts_enc_armv8_ce
+                                    : _gcry_aes_xts_dec_armv8_ce;
+  unsigned int nrounds = ctx->rounds;
+
+  crypt_fn(keysched, outbuf, inbuf, tweak, nblocks, nrounds);
+}
+
 #endif /* USE_ARM_CE */