Use ff_imdct_half() and vector_fmul_window() for IMDCT and windowing. Reduce

buffer sizes accordingly. This produces a ~10% overall decoding perfomance
improvement.

Originally committed as revision 14908 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/aac.c

@@ -1166,11 +1166,11 @@ static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
     float * in = sce->coeffs;
     float * out = sce->ret;
     float * saved = sce->saved;
-    const float * lwindow      = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
     const float * swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
     const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
     const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
     float * buf = ac->buf_mdct;
+    DECLARE_ALIGNED(16, float, temp[128]);
     int i;
 
     // imdct
@@ -1179,12 +1179,10 @@ static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
             av_log(ac->avccontext, AV_LOG_WARNING,
                    "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
                    "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
-        for (i = 0; i < 2048; i += 256) {
-            ff_imdct_calc(&ac->mdct_small, buf + i, in + i/2);
-            ac->dsp.vector_fmul_reverse(ac->revers + i/2, buf + i + 128, swindow, 128);
-        }
+        for (i = 0; i < 1024; i += 128)
+            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
     } else
-        ff_imdct_calc(&ac->mdct, buf, in);
+        ff_imdct_half(&ac->mdct, buf, in);
 
     /* window overlapping
      * NOTE: To simplify the overlapping code, all 'meaningless' short to long
@@ -1194,38 +1192,38 @@ static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
      */
     if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
         (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
-        ac->dsp.vector_fmul_add_add(out,       buf,       lwindow_prev, saved,       ac->add_bias, 1024, 1);
+        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, ac->add_bias, 512);
     } else {
         for (i = 0; i < 448; i++)
             out[i] = saved[i] + ac->add_bias;
 
         if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
-            ac->dsp.vector_fmul_add_add(out + 448 + 0*128, buf + 0*128, swindow_prev, saved + 448 ,       ac->add_bias, 128, 1);
-            ac->dsp.vector_fmul_add_add(out + 448 + 1*128, buf + 2*128, swindow,      ac->revers + 0*128, ac->add_bias, 128, 1);
-            ac->dsp.vector_fmul_add_add(out + 448 + 2*128, buf + 4*128, swindow,      ac->revers + 1*128, ac->add_bias, 128, 1);
-            ac->dsp.vector_fmul_add_add(out + 448 + 3*128, buf + 6*128, swindow,      ac->revers + 2*128, ac->add_bias, 128, 1);
-            ac->dsp.vector_fmul_add_add(out + 448 + 4*128, buf + 8*128, swindow,      ac->revers + 3*128, ac->add_bias,  64, 1);
+            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, ac->add_bias, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      ac->add_bias, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      ac->add_bias, 64);
+            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      ac->add_bias, 64);
+            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      ac->add_bias, 64);
+            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
         } else {
-            ac->dsp.vector_fmul_add_add(out + 448,         buf + 448,   swindow_prev, saved + 448,        ac->add_bias, 128, 1);
+            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, ac->add_bias, 64);
             for (i = 576; i < 1024; i++)
-                out[i] = buf[i] + saved[i] + ac->add_bias;
+                out[i] = buf[i-512] + ac->add_bias;
         }
     }
 
     // buffer update
     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
-        ac->dsp.vector_fmul_add_add(saved,       buf + 1024 + 64,    swindow + 64, ac->revers + 3*128+64,  0, 64, 1);
-        ac->dsp.vector_fmul_add_add(saved + 64,  buf + 1024 + 2*128, swindow,      ac->revers + 4*128,     0, 128, 1);
-        ac->dsp.vector_fmul_add_add(saved + 192, buf + 1024 + 4*128, swindow,      ac->revers + 5*128,     0, 128, 1);
-        ac->dsp.vector_fmul_add_add(saved + 320, buf + 1024 + 6*128, swindow,      ac->revers + 6*128,     0, 128, 1);
-        memcpy(                     saved + 448, ac->revers + 7*128, 128 * sizeof(float));
-        memset(                     saved + 576, 0,                  448 * sizeof(float));
+        for (i = 0; i < 64; i++)
+            saved[i] = temp[64 + i] - ac->add_bias;
+        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
+        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
+        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
+        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
     } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
-        memcpy(saved, buf + 1024, 448 * sizeof(float));
-        ac->dsp.vector_fmul_reverse(saved + 448, buf + 1024 + 448, swindow, 128);
-        memset(saved + 576, 0, 448 * sizeof(float));
+        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
+        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
     } else { // LONG_STOP or ONLY_LONG
-        ac->dsp.vector_fmul_reverse(saved, buf + 1024, lwindow, 1024);
+        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
     }
 }
 

libavcodec/aac.h

@@ -206,7 +206,7 @@ typedef struct {
     int band_type_run_end[120];               ///< band type run end points
     float sf[120];                            ///< scalefactors
     DECLARE_ALIGNED_16(float, coeffs[1024]);  ///< coefficients for IMDCT
-    DECLARE_ALIGNED_16(float, saved[1024]);   ///< overlap
+    DECLARE_ALIGNED_16(float, saved[512]);    ///< overlap
     DECLARE_ALIGNED_16(float, ret[1024]);     ///< PCM output
 } SingleChannelElement;
 
@@ -247,8 +247,7 @@ typedef struct {
      * @defgroup temporary aligned temporary buffers (We do not want to have these on the stack.)
      * @{
      */
-    DECLARE_ALIGNED_16(float, buf_mdct[2048]);
-    DECLARE_ALIGNED_16(float, revers[1024]);
+    DECLARE_ALIGNED_16(float, buf_mdct[1024]);
     /** @} */
 
     /**