[FFmpeg-devel] [PATCH] Allow arch-specific mdct code to request interleaving of cos/sin tables

[Mans Rullgard]

---
 libavcodec/dsputil.h |    3 +++
 libavcodec/mdct.c    |   27 +++++++++++++++++++--------
 2 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 6c81e3c..c845046 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -727,6 +727,9 @@ typedef struct MDCTContext {
     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
     void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
     void (*mdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
+    int permutation;
+#define FF_MDCT_PERM_NONE       0
+#define FF_MDCT_PERM_INTERLEAVE 1
 } MDCTContext;
 
 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
diff --git a/libavcodec/mdct.c b/libavcodec/mdct.c
index dad966d..6a893f9 100644
--- a/libavcodec/mdct.c
+++ b/libavcodec/mdct.c
@@ -76,39 +76,51 @@ av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
 {
     int n, n4, i;
     double alpha, theta;
+    int tstep;
 
     memset(s, 0, sizeof(*s));
     n = 1 << nbits;
     s->nbits = nbits;
     s->n = n;
     n4 = n >> 2;
-    s->tcos = av_malloc(n4 * sizeof(FFTSample));
+    s->tcos = av_malloc(n/2 * sizeof(FFTSample));
     if (!s->tcos)
         goto fail;
-    s->tsin = av_malloc(n4 * sizeof(FFTSample));
-    if (!s->tsin)
-        goto fail;
 
     s->imdct_calc = ff_imdct_calc_c;
     s->imdct_half = ff_imdct_half_c;
     s->mdct_calc  = ff_mdct_calc_c;
 
+    s->permutation = FF_MDCT_PERM_NONE;
+
     if (ARCH_ARM) ff_mdct_init_arm(s);
     if (HAVE_MMX) ff_mdct_init_mmx(s);
 
+    switch (s->permutation) {
+    case FF_MDCT_PERM_NONE:
+        s->tsin = s->tcos + n4;
+        tstep = 1;
+        break;
+    case FF_MDCT_PERM_INTERLEAVE:
+        s->tsin = s->tcos + 1;
+        tstep = 2;
+        break;
+    default:
+        goto fail;
+    }
+
     theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0);
     scale = sqrt(fabs(scale));
     for(i=0;i<n4;i++) {
         alpha = 2 * M_PI * (i + theta) / n;
-        s->tcos[i] = -cos(alpha) * scale;
-        s->tsin[i] = -sin(alpha) * scale;
+        s->tcos[i*tstep] = -cos(alpha) * scale;
+        s->tsin[i*tstep] = -sin(alpha) * scale;
     }
     if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
         goto fail;
     return 0;
  fail:
     av_freep(&s->tcos);
-    av_freep(&s->tsin);
     return -1;
 }
 
@@ -236,6 +248,5 @@ void ff_mdct_calc_c(MDCTContext *s, FFTSample *out, const FFTSample *input)
 av_cold void ff_mdct_end(MDCTContext *s)
 {
     av_freep(&s->tcos);
-    av_freep(&s->tsin);
     ff_fft_end(&s->fft);
 }
-- 
1.6.4.2