denozord/strongswan-ext
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

aesni: Calculate GHASH for 4 blocks of associated data in parallel

Browse Source 
While associated data is usually not that large, in some specific cases
this can bring a significant performance boost.
This commit is contained in:
Martin Willi
2015-04-15 11:35:28 +02:00
parent 0eb593b0bb
commit edab6c658c
1 changed files with 18 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/libstrongswan/plugins/aesni/aesni_gcm.c
+18 -2
View File
@@ -269,14 +269,30 @@ static __m128i ghash(__m128i h, __m128i y, __m128i x)
*/ */
static __m128i icv_header(private_aesni_gcm_t *this, void *assoc, size_t alen) static __m128i icv_header(private_aesni_gcm_t *this, void *assoc, size_t alen)
{ {
u_int blocks, rem, i; u_int blocks, pblocks, rem, i;
__m128i h1, h2, h3, h4, d1, d2, d3, d4;
__m128i y, last, *ab; __m128i y, last, *ab;
h1 = this->hhhh;
h2 = this->hhh;
h3 = this->hh;
h4 = this->h;
y = _mm_setzero_si128(); y = _mm_setzero_si128();
ab = assoc; ab = assoc;
blocks = alen / AES_BLOCK_SIZE; blocks = alen / AES_BLOCK_SIZE;
pblocks = blocks - (blocks % GCM_CRYPT_PARALLELISM);
rem = alen % AES_BLOCK_SIZE; rem = alen % AES_BLOCK_SIZE;
for (i = 0; i < blocks; i++) for (i = 0; i < pblocks; i += GCM_CRYPT_PARALLELISM)
{
d1 = _mm_loadu_si128(ab + i + 0);
d2 = _mm_loadu_si128(ab + i + 1);
d3 = _mm_loadu_si128(ab + i + 2);
d4 = _mm_loadu_si128(ab + i + 3);
y = _mm_xor_si128(y, d1);
y = mult4xor(h1, h2, h3, h4, y, d2, d3, d4);
}
for (i = pblocks; i < blocks; i++)
{ {
y = ghash(this->h, y, _mm_loadu_si128(ab + i)); y = ghash(this->h, y, _mm_loadu_si128(ab + i));
} }