[FFmpeg-cvslog] xtea: optimize core, about 4-5% faster.

Michael Niedermayer git at videolan.org
Tue Sep 11 05:04:03 CEST 2012


ffmpeg | branch: master | Michael Niedermayer <michaelni at gmx.at> | Tue Sep 11 04:55:01 2012 +0200| [bb4ddeea5f6951d13833e6a1235763257ff0a658] | committer: Michael Niedermayer

xtea: optimize core, about 4-5% faster.

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=bb4ddeea5f6951d13833e6a1235763257ff0a658
---

 libavutil/xtea.c |   84 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 83 insertions(+), 1 deletion(-)

diff --git a/libavutil/xtea.c b/libavutil/xtea.c
index 0a5df1c..d9c2858 100644
--- a/libavutil/xtea.c
+++ b/libavutil/xtea.c
@@ -39,12 +39,17 @@ static void xtea_crypt_ecb(AVXTEA *ctx, uint8_t *dst, const uint8_t *src,
                            int decrypt)
 {
     uint32_t v0, v1;
-    int i;
+    uint32_t k0 = ctx->key[0];
+    uint32_t k1 = ctx->key[1];
+    uint32_t k2 = ctx->key[2];
+    uint32_t k3 = ctx->key[3];
 
     v0 = AV_RB32(src);
     v1 = AV_RB32(src + 4);
 
     if (decrypt) {
+#if CONFIG_SMALL
+        int i;
         uint32_t delta = 0x9E3779B9, sum = delta * 32;
 
         for (i = 0; i < 32; i++) {
@@ -52,7 +57,47 @@ static void xtea_crypt_ecb(AVXTEA *ctx, uint8_t *dst, const uint8_t *src,
             sum -= delta;
             v0 -= (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (sum + ctx->key[sum & 3]);
         }
+#else
+#define DSTEP(SUM, K0, K1) \
+            v1 -= (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (SUM + K0); \
+            v0 -= (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (SUM - 0x9E3779B9 + K1)
+
+        DSTEP(0xC6EF3720U, k2, k3);
+        DSTEP(0x28B7BD67U, k3, k2);
+        DSTEP(0x8A8043AEU, k0, k1);
+        DSTEP(0xEC48C9F5U, k1, k0);
+        DSTEP(0x4E11503CU, k2, k3);
+        DSTEP(0xAFD9D683U, k2, k2);
+        DSTEP(0x11A25CCAU, k3, k1);
+        DSTEP(0x736AE311U, k0, k0);
+        DSTEP(0xD5336958U, k1, k3);
+        DSTEP(0x36FBEF9FU, k1, k2);
+        DSTEP(0x98C475E6U, k2, k1);
+        DSTEP(0xFA8CFC2DU, k3, k0);
+        DSTEP(0x5C558274U, k0, k3);
+        DSTEP(0xBE1E08BBU, k1, k2);
+        DSTEP(0x1FE68F02U, k1, k1);
+        DSTEP(0x81AF1549U, k2, k0);
+        DSTEP(0xE3779B90U, k3, k3);
+        DSTEP(0x454021D7U, k0, k2);
+        DSTEP(0xA708A81EU, k1, k1);
+        DSTEP(0x08D12E65U, k1, k0);
+        DSTEP(0x6A99B4ACU, k2, k3);
+        DSTEP(0xCC623AF3U, k3, k2);
+        DSTEP(0x2E2AC13AU, k0, k1);
+        DSTEP(0x8FF34781U, k0, k0);
+        DSTEP(0xF1BBCDC8U, k1, k3);
+        DSTEP(0x5384540FU, k2, k2);
+        DSTEP(0xB54CDA56U, k3, k1);
+        DSTEP(0x1715609DU, k0, k0);
+        DSTEP(0x78DDE6E4U, k0, k3);
+        DSTEP(0xDAA66D2BU, k1, k2);
+        DSTEP(0x3C6EF372U, k2, k1);
+        DSTEP(0x9E3779B9U, k3, k0);
+#endif
     } else {
+#if CONFIG_SMALL
+        int i;
         uint32_t sum = 0, delta = 0x9E3779B9;
 
         for (i = 0; i < 32; i++) {
@@ -60,6 +105,43 @@ static void xtea_crypt_ecb(AVXTEA *ctx, uint8_t *dst, const uint8_t *src,
             sum += delta;
             v1 += (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (sum + ctx->key[(sum >> 11) & 3]);
         }
+#else
+#define ESTEP(SUM, K0, K1) \
+            v0 += (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (SUM + K0);\
+            v1 += (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (SUM + 0x9E3779B9 + K1)
+        ESTEP(0x00000000U, k0, k3);
+        ESTEP(0x9E3779B9U, k1, k2);
+        ESTEP(0x3C6EF372U, k2, k1);
+        ESTEP(0xDAA66D2BU, k3, k0);
+        ESTEP(0x78DDE6E4U, k0, k0);
+        ESTEP(0x1715609DU, k1, k3);
+        ESTEP(0xB54CDA56U, k2, k2);
+        ESTEP(0x5384540FU, k3, k1);
+        ESTEP(0xF1BBCDC8U, k0, k0);
+        ESTEP(0x8FF34781U, k1, k0);
+        ESTEP(0x2E2AC13AU, k2, k3);
+        ESTEP(0xCC623AF3U, k3, k2);
+        ESTEP(0x6A99B4ACU, k0, k1);
+        ESTEP(0x08D12E65U, k1, k1);
+        ESTEP(0xA708A81EU, k2, k0);
+        ESTEP(0x454021D7U, k3, k3);
+        ESTEP(0xE3779B90U, k0, k2);
+        ESTEP(0x81AF1549U, k1, k1);
+        ESTEP(0x1FE68F02U, k2, k1);
+        ESTEP(0xBE1E08BBU, k3, k0);
+        ESTEP(0x5C558274U, k0, k3);
+        ESTEP(0xFA8CFC2DU, k1, k2);
+        ESTEP(0x98C475E6U, k2, k1);
+        ESTEP(0x36FBEF9FU, k3, k1);
+        ESTEP(0xD5336958U, k0, k0);
+        ESTEP(0x736AE311U, k1, k3);
+        ESTEP(0x11A25CCAU, k2, k2);
+        ESTEP(0xAFD9D683U, k3, k2);
+        ESTEP(0x4E11503CU, k0, k1);
+        ESTEP(0xEC48C9F5U, k1, k0);
+        ESTEP(0x8A8043AEU, k2, k3);
+        ESTEP(0x28B7BD67U, k3, k2);
+#endif
     }
 
     AV_WB32(dst, v0);



More information about the ffmpeg-cvslog mailing list