[FFmpeg-devel] [PATCH 4/5] wma lossless: reuse scalarproduct_and_madd_int16

[Christophe Gisquet]

Approximate relative speedup depending on instruction set:
plain C: -6%
mmxext:  51%
sse2:    54%
---
 libavcodec/wmalosslessdec.c | 47 +++++++++++++++++----------------------------
 1 file changed, 18 insertions(+), 29 deletions(-)

diff --git a/libavcodec/wmalosslessdec.c b/libavcodec/wmalosslessdec.c
index 90a0109..cff02a8 100644
--- a/libavcodec/wmalosslessdec.c
+++ b/libavcodec/wmalosslessdec.c
@@ -29,6 +29,7 @@
 #include "internal.h"
 #include "get_bits.h"
 #include "put_bits.h"
+#include "dsputil.h"
 #include "wma.h"
 #include "wma_common.h"
 
@@ -66,6 +67,7 @@ typedef struct {
 typedef struct WmallDecodeCtx {
     /* generic decoder variables */
     AVCodecContext  *avctx;
+    DSPContext      dsp;                           ///< accelerated DSP functions
     AVFrame         frame;
     uint8_t         frame_data[MAX_FRAMESIZE + FF_INPUT_BUFFER_PADDING_SIZE];  ///< compressed frame data
     PutBitContext   pb;                             ///< context for filling the frame_data buffer
@@ -141,9 +143,9 @@ typedef struct WmallDecodeCtx {
         int scaling;
         int coefsend;
         int bitsend;
-        int16_t coefs[MAX_ORDER];
-        int16_t lms_prevvalues[MAX_ORDER * 2];
-        int16_t lms_updates[MAX_ORDER * 2];
+        DECLARE_ALIGNED(16, int16_t, coefs)[MAX_ORDER];
+        DECLARE_ALIGNED(16, int16_t, lms_prevvalues)[MAX_ORDER * 2];
+        DECLARE_ALIGNED(16, int16_t, lms_updates)[MAX_ORDER * 2];
         int recent;
     } cdlms[WMALL_MAX_CHANNELS][9];
 
@@ -179,6 +181,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
     int i, log2_max_num_subframes;
 
     s->avctx = avctx;
+    ff_dsputil_init(&s->dsp, avctx);
     init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
 
     if (avctx->extradata_size >= 18) {
@@ -684,35 +687,12 @@ static void revert_mclms(WmallDecodeCtx *s, int tile_size)
     }
 }
 
-static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
-{
-    int pred = 0, icoef;
-    int recent = s->cdlms[ich][ilms].recent;
-
-    for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
-        pred += s->cdlms[ich][ilms].coefs[icoef] *
-                s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
-
-    return pred;
-}
-
-static void lms_update(WmallDecodeCtx *s, int ich, int ilms,
-                       int input, int residue)
+static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int input)
 {
     int icoef;
     int recent = s->cdlms[ich][ilms].recent;
     int range  = 1 << s->bits_per_sample - 1;
 
-    if (residue < 0) {
-        for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
-            s->cdlms[ich][ilms].coefs[icoef] -=
-                s->cdlms[ich][ilms].lms_updates[icoef + recent];
-    } else if (residue > 0) {
-        for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
-            s->cdlms[ich][ilms].coefs[icoef] +=
-                s->cdlms[ich][ilms].lms_updates[icoef + recent];
-    }
-
     if (recent)
         recent--;
     else {
@@ -773,6 +753,9 @@ static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
     s->update_speed[ich] = 8;
 }
 
+/** Get sign of integer (1 for positive, -1 for negative and 0 for zero) */
+#define WMASIGN(x) ((x > 0) - (x < 0))
+
 static void revert_cdlms(WmallDecodeCtx *s, int ch,
                          int coef_begin, int coef_end)
 {
@@ -783,9 +766,15 @@ static void revert_cdlms(WmallDecodeCtx *s, int ch,
         for (icoef = coef_begin; icoef < coef_end; icoef++) {
             pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
             residue = s->channel_residues[ch][icoef];
-            pred += lms_predict(s, ch, ilms);
+            pred += s->dsp.scalarproduct_and_madd_int16(s->cdlms[ch][ilms].coefs,
+                                                        s->cdlms[ch][ilms].lms_prevvalues
+                                                            + s->cdlms[ch][ilms].recent,
+                                                        s->cdlms[ch][ilms].lms_updates
+                                                            + s->cdlms[ch][ilms].recent,
+                                                        s->cdlms[ch][ilms].order,
+                                                        WMASIGN(residue));
             input = residue + (pred >> s->cdlms[ch][ilms].scaling);
-            lms_update(s, ch, ilms, input, residue);
+            lms_update(s, ch, ilms, input);
             s->channel_residues[ch][icoef] = input;
         }
     }
-- 
1.8.0.msysgit.0