avcodec/takdec: add x86 SIMD for rest of decorrelation modes

Signed-off-by: Paul B Mahol <onemda@gmail.com>

1 files changed, 19 insertions, 25 deletions

diff --git a/libavcodec/takdec.c b/libavcodec/takdec.c
index 5395596cfc..e5c0723940 100644
--- a/libavcodec/takdec.c
+++ b/libavcodec/takdec.c
@@ -28,6 +28,7 @@
 #include "libavutil/internal.h"
 #include "libavutil/samplefmt.h"
 #include "tak.h"
+#include "takdsp.h"
 #include "audiodsp.h"
 #include "thread.h"
 #include "avcodec.h"
@@ -47,6 +48,7 @@ typedef struct MCDParam {
 typedef struct TAKDecContext {
     AVCodecContext *avctx;                          ///< parent AVCodecContext
     AudioDSPContext adsp;
+    TAKDSPContext   tdsp;
     TAKStreamInfo   ti;
     GetBitContext   gb;                             ///< bitstream reader initialized to start at the current frame
 
@@ -172,6 +174,7 @@ static av_cold int tak_decode_init(AVCodecContext *avctx)
     TAKDecContext *s = avctx->priv_data;
 
     ff_audiodsp_init(&s->adsp);
+    ff_takdsp_init(&s->tdsp);
 
     s->avctx = avctx;
     avctx->bits_per_raw_sample = avctx->bits_per_coded_sample;
@@ -541,46 +544,32 @@ static int decode_channel(TAKDecContext *s, int chan)
 static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
 {
     GetBitContext *gb = &s->gb;
-    int32_t *p1       = s->decoded[c1] + 1;
-    int32_t *p2       = s->decoded[c2] + 1;
+    int32_t *p1       = s->decoded[c1] + (s->dmode > 5);
+    int32_t *p2       = s->decoded[c2] + (s->dmode > 5);
+    int32_t bp1       = p1[0];
+    int32_t bp2       = p2[0];
     int i;
     int dshift, dfactor;
 
+    length += s->dmode < 6;
+
     switch (s->dmode) {
     case 1: /* left/side */
-        for (i = 0; i < length; i++) {
-            int32_t a = p1[i];
-            int32_t b = p2[i];
-            p2[i]     = a + b;
-        }
+        s->tdsp.decorrelate_ls(p1, p2, length);
         break;
     case 2: /* side/right */
-        for (i = 0; i < length; i++) {
-            int32_t a = p1[i];
-            int32_t b = p2[i];
-            p1[i]     = b - a;
-        }
+        s->tdsp.decorrelate_sr(p1, p2, length);
         break;
     case 3: /* side/mid */
-        for (i = 0; i < length; i++) {
-            int32_t a = p1[i];
-            int32_t b = p2[i];
-            a        -= b >> 1;
-            p1[i]     = a;
-            p2[i]     = a + b;
-        }
+        s->tdsp.decorrelate_sm(p1, p2, length);
         break;
     case 4: /* side/left with scale factor */
         FFSWAP(int32_t*, p1, p2);
+        FFSWAP(int32_t, bp1, bp2);
     case 5: /* side/right with scale factor */
         dshift  = get_bits_esc4(gb);
         dfactor = get_sbits(gb, 10);
-        for (i = 0; i < length; i++) {
-            int32_t a = p1[i];
-            int32_t b = p2[i];
-            b         = dfactor * (b >> dshift) + 128 >> 8 << dshift;
-            p1[i]     = b - a;
-        }
+        s->tdsp.decorrelate_sf(p1, p2, length, dshift, dfactor);
         break;
     case 6:
         FFSWAP(int32_t*, p1, p2);
@@ -664,6 +653,11 @@ static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
     }
     }
 
+    if (s->dmode > 0 && s->dmode < 6) {
+        p1[0] = bp1;
+        p2[0] = bp2;
+    }
+
     return 0;
 }