Add software prefetch to AVX2 base64 encoder hot loop

Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
MergeDate: Fri Feb 13 14:31:43 2026
(Merged from https://github.com/openssl/openssl/pull/29858)

diff --git a/crypto/evp/enc_b64_avx2.c b/crypto/evp/enc_b64_avx2.c
index b3f46e66aec4037bd3e0a69e3180bbf79e38fbed..141e7fcc87e5f761b27b8082089e7ad596bb345b 100644 (file)
--- a/crypto/evp/enc_b64_avx2.c
+++ b/crypto/evp/enc_b64_avx2.c
@@ -493,6 +493,7 @@ int encode_base64_avx2(EVP_ENCODE_CTX *ctx, unsigned char *dst,
 
     /* Process 96 bytes at a time */
     for (; i + 100 <= srclen; i += 96) {
+        _mm_prefetch((const char *)(input + i + 192), _MM_HINT_T0);
         /* We shave off 4 bytes from the beginning and the end */
         const __m128i lo0 = _mm_loadu_si128((const __m128i *)(input + i + 4 * 3 * 0));
         const __m128i hi0 = _mm_loadu_si128((const __m128i *)(input + i + 4 * 3 * 1));