@@ -272,7 +272,7 @@ 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, buf_mdct)[1024];

     /** @} */

     /**

@@ -1763,8 +1763,8 @@ static void apply_ltp(AACContext *ac, SingleChannelElement *sce)

     int i, sfb;

     if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {

-        float *predTime = ac->buf_mdct;

-        float *predFreq = sce->ret;

+        float *predTime = sce->ret;

+        float *predFreq = ac->buf_mdct;

         int16_t num_samples = 2048;

         if (ltp->lag < 1024)

@@ -1797,19 +1797,22 @@ static void update_ltp(AACContext *ac, SingleChannelElement *sce)

     const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;

     int i;

-    for (i = 0; i < 512; i++)

-        ac->buf_mdct[1535 - i] = ac->buf_mdct[512 + i];

-

     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {

         memcpy(saved_ltp,       saved, 512 * sizeof(float));

         memset(saved_ltp + 576, 0,     448 * sizeof(float));

-        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     swindow,     128);

+        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);

+        for (i = 0; i < 64; i++)

+            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];

     } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {

         memcpy(saved_ltp,       ac->buf_mdct + 512, 448 * sizeof(float));

         memset(saved_ltp + 576, 0,                  448 * sizeof(float));

-        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     swindow,     128);

+        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);

+        for (i = 0; i < 64; i++)

+            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];

     } else { // LONG_STOP or ONLY_LONG

-        ac->dsp.vector_fmul_reverse(saved_ltp,       ac->buf_mdct + 512,     lwindow,     1024);

+        ac->dsp.vector_fmul_reverse(saved_ltp,       ac->buf_mdct + 512,     &lwindow[512],     512);

+        for (i = 0; i < 512; i++)

+            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];

     }

     memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));