[FFmpeg-devel] [PATCH 3/3] xtea: precalculate part of the right side of the core equations

[Michael Niedermayer]

This improves speed a bit (5% or so)

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
---
 libavutil/xtea.c |   27 +++++++++++++--------------
 libavutil/xtea.h |    2 +-
 2 files changed, 14 insertions(+), 15 deletions(-)

diff --git a/libavutil/xtea.c b/libavutil/xtea.c
index beaa687..cdebf60 100644
--- a/libavutil/xtea.c
+++ b/libavutil/xtea.c
@@ -31,36 +31,35 @@ const int av_xtea_size = sizeof(AVXTEA);
 
 void av_xtea_init(AVXTEA *ctx, const uint8_t key[16])
 {
+    uint32_t sum = 0, delta = 0x9E3779B9;
     int i;
 
-    for (i = 0; i < 4; i++)
-        ctx->key[i] = AV_RB32(key + (i << 2));
+    for (i = 0; i < 32; i++){
+        ctx->key[2*i+0] = sum + AV_RB32(key + ((sum&   3) << 2));
+        sum += delta;
+        ctx->key[2*i+1] = sum + AV_RB32(key + ((sum&6144) >> 9));
+    }
 }
 
 static void xtea_crypt_ecb(AVXTEA *ctx, uint8_t *dst, const uint8_t *src,
                            int decrypt)
 {
     uint32_t v0, v1;
+    uint32_t *key = ctx->key;
     int i;
 
     v0 = AV_RB32(src);
     v1 = AV_RB32(src + 4);
 
     if (decrypt) {
-        uint32_t delta = 0x9E3779B9, sum = delta * 32;
-
-        for (i = 0; i < 32; i++) {
-            v1 -= (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (sum + ctx->key[(sum >> 11) & 3]);
-            sum -= delta;
-            v0 -= (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (sum + ctx->key[sum & 3]);
+        for (i=62; i>=0; i-=2) {
+            v1 -= (((v0 << 4) ^ (v0 >> 5)) + v0) ^ key[i+1];
+            v0 -= (((v1 << 4) ^ (v1 >> 5)) + v1) ^ key[i+0];
         }
     } else {
-        uint32_t sum = 0, delta = 0x9E3779B9;
-
-        for (i = 0; i < 32; i++) {
-            v0 += (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (sum + ctx->key[sum & 3]);
-            sum += delta;
-            v1 += (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (sum + ctx->key[(sum >> 11) & 3]);
+        for (i=0; i<64; i+=2) {
+            v0 += (((v1 << 4) ^ (v1 >> 5)) + v1) ^ key[i+0];
+            v1 += (((v0 << 4) ^ (v0 >> 5)) + v0) ^ key[i+1];
         }
     }
 
diff --git a/libavutil/xtea.h b/libavutil/xtea.h
index 7313fb7..e68ee2a 100644
--- a/libavutil/xtea.h
+++ b/libavutil/xtea.h
@@ -39,7 +39,7 @@ extern const int av_xtea_size;
  * libavutil.
  */
 typedef struct AVXTEA {
-    uint32_t key[16];
+    uint32_t key[64];
 } AVXTEA;
 
 /**
-- 
1.7.9.5