SSE optimized 32-point DCT

Originally committed as revision 24077 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/dct.c

@@ -30,6 +30,7 @@
 #include <math.h>
 #include "libavutil/mathematics.h"
 #include "fft.h"
+#include "x86/fft.h"
 
 #define DCT32_FLOAT
 #include "dct32.c"
@@ -213,6 +214,7 @@ av_cold int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType inverse)
         s->dct_calc = dct32_func;
 
     s->dct32 = dct32;
+    if (HAVE_MMX)     ff_dct_init_mmx(s);
 
     return 0;
 }

libavcodec/fft.h

@@ -112,6 +112,7 @@ void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 void ff_fft_init_altivec(FFTContext *s);
 void ff_fft_init_mmx(FFTContext *s);
 void ff_fft_init_arm(FFTContext *s);
+void ff_dct_init_mmx(DCTContext *s);
 
 /**
  * Do the permutation needed BEFORE calling ff_fft_calc().

libavcodec/x86/fft.c

@@ -42,3 +42,11 @@ av_cold void ff_fft_init_mmx(FFTContext *s)
     }
 #endif
 }
+
+av_cold void ff_dct_init_mmx(DCTContext *s)
+{
+    int has_vectors = mm_support();
+    if (has_vectors & FF_MM_SSE && HAVE_SSE)
+        s->dct32 = ff_dct32_float_sse;
+}
+

libavcodec/x86/fft.h

@@ -32,5 +32,6 @@ void ff_imdct_calc_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input
 void ff_imdct_half_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input);
 void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
 void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
+void ff_dct32_float_sse(FFTSample *out, const FFTSample *in);
 
 #endif