Further optimization and minor clean up of atrac codec: Unroll iqmf_dewindowing for non-ARM speeds up decoder by +10% on H300. Removed some non-used arrays. Codec is still not fully realtime on Coldfire targets.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@24648 a1c6a512-1295-4272-9138-f99709370657

3 files changed, 78 insertions, 16 deletions

diff --git a/apps/codecs/libatrac/atrac3.c b/apps/codecs/libatrac/atrac3.c
index ac63925ce7..467f42f161 100644
--- a/apps/codecs/libatrac/atrac3.c
+++ b/apps/codecs/libatrac/atrac3.c
@@ -98,6 +98,20 @@ static channel_unit     channel_units[2] IBSS_ATTR_LARGE_IRAM;
  * @param in        input buffer
  * @param win       windowing coefficients
  * @param nIn       size of spectrum buffer
+ * Reference implementation:
+ *
+ * for (j = nIn; j != 0; j--) {
+ *          s1 = fixmul32(in[0], win[0]);
+ *          s2 = fixmul32(in[1], win[1]);
+ *          for (i = 2; i < 48; i += 2) {
+ *              s1 += fixmul31(in[i  ], win[i  ]);
+ *              s2 += fixmul31(in[i+1], win[i+1]);
+ *          }
+ *          out[0] = s2;
+ *          out[1] = s1;
+ *          in += 2;
+ *          out += 2;
+ *      }
  */
  
 #if defined(CPU_ARM)
@@ -116,16 +130,62 @@ static channel_unit     channel_units[2] IBSS_ATTR_LARGE_IRAM;
         int32_t i, j, s1, s2;
         
         for (j = nIn; j != 0; j--) {
-            /* i=0 */
-            s1 = fixmul31(win[0], in[0]);
-            s2 = fixmul31(win[1], in[1]);
-    
-            /* i=2..46 */
-            for (i = 2; i < 48; i += 2) {
-                s1 += fixmul31(win[i  ], in[i  ]);
-                s2 += fixmul31(win[i+1], in[i+1]);
-            }
-    
+            i = 0;
+            /*  0.. 7 */
+            s1  = fixmul31(win[i], in[i]); i++;
+            s2  = fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            /*  8..15 */
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            /* 16..23 */
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            /* 24..31 */
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            /* 32..39 */
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            /* 40..47 */
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]); i++;
+            s1 += fixmul31(win[i], in[i]); i++;
+            s2 += fixmul31(win[i], in[i]);
+
             out[0] = s2;
             out[1] = s1;
     
diff --git a/apps/codecs/libatrac/atrac3data.h b/apps/codecs/libatrac/atrac3data.h
index 8e7fe97c32..0da9d9e758 100644
--- a/apps/codecs/libatrac/atrac3data.h
+++ b/apps/codecs/libatrac/atrac3data.h
@@ -117,18 +117,18 @@ static const int8_t decTable1[18] = {0,0, 0,1, 0,-1, 1,0, -1,0, 1,1, 1,-1, -1,1,
 
 
 /* tables for the scalefactor decoding */
-
+/* not needed anymore
 static const float iMaxQuant[8] = {
   0.0, 1.0/1.5, 1.0/2.5, 1.0/3.5, 1.0/4.5, 1.0/7.5, 1.0/15.5, 1.0/31.5
 };
-
+*/
 static const uint16_t subbandTab[33] = {
   0, 8, 16, 24, 32, 40, 48, 56, 64, 80, 96, 112, 128, 144, 160, 176, 192, 224,
   256, 288, 320, 352, 384, 416, 448, 480, 512, 576, 640, 704, 768, 896, 1024
 };
 
 /* transform data */
-
+/* not needed anymore
 static const float qmf_48tap_half[24] = {
    -0.00001461907, -0.00009205479, -0.000056157569, 0.00030117269,
     0.0002422519,-0.00085293897, -0.0005205574, 0.0020340169,
@@ -137,8 +137,9 @@ static const float qmf_48tap_half[24] = {
    -0.007801671, -0.034090221, 0.01880949, 0.054326009,
    -0.043596379, -0.099384367, 0.13207909, 0.46424159
 };
-
+*/
 /* joint stereo related tables */
+/* not needed anymore
 static const float matrixCoeffs[8] = {0.0, 2.0, 2.0, 2.0, 0.0, 0.0, 1.0, 1.0};
-
+*/
 #endif /* AVCODEC_ATRAC3DATA_H */
diff --git a/apps/codecs/libatrac/main.c b/apps/codecs/libatrac/main.c
index 826dacf819..e0a3f8507d 100644
--- a/apps/codecs/libatrac/main.c
+++ b/apps/codecs/libatrac/main.c
@@ -9,6 +9,8 @@
 #include "atrac3.h"
 #include "../librm/rm.h"
 
+ATRAC3Context q IBSS_ATTR;
+
 static unsigned char wav_header[44]={
     'R','I','F','F',//  0 - ChunkID
     0,0,0,0,        //  4 - ChunkSize (filesize-8)
@@ -101,7 +103,6 @@ int main(int argc, char *argv[])
     int16_t outbuf[2048];
     uint16_t fs,sps,h;
     uint32_t packet_count;
-    ATRAC3Context q;
     RMContext rmctx;
     RMPacket pkt;