Add an x86 optimized version of extend().

[fjl] / idct_float.c

diff --git a/idct_float.c b/idct_float.c
index ede97cda9951a598f714e4d5b453dfe15b1115af..34ff5b690cabb463f1e156ba573b6cd3e865895c 100644 (file)
--- a/idct_float.c
+++ b/idct_float.c
@@ -50,9 +50,11 @@ static inline void idct1d_float(double y0, double y1, double y2, double y3, doub
        // constants
        static const double a1 = 0.7071067811865474;   // sqrt(2)
        static const double a2 = 0.5411961001461971;   // cos(3/8 pi) * sqrt(2)
-       static const double a3 = a1;
+       // static const double a3 = a1;
+       static const double a3 = 0.7071067811865474;
        static const double a4 = 1.3065629648763766;   // cos(pi/8) * sqrt(2)
-       static const double a5 = 0.5 * (a4 - a2);
+       // static const double a5 = 0.5 * (a4 - a2);
+       static const double a5 = 0.3826834323650897;
 
        // phase 1
        const double p1_0 = y0;