Changeset 2616 for pjproject/trunk/third_party/g7221/encode/dct4_a.c

Timestamp:

Apr 18, 2009 2:29:28 PM (15 years ago)

Author:

bennylp

Message:

More ticket #774: optimization for siren7/siren14 codecs

File:

• pjproject/trunk/third_party/g7221/encode/dct4_a.c (modified) (7 diffs)

pjproject/trunk/third_party/g7221/encode/dct4_a.c

@@ -136 +136 @@
 
         /*    set_span      = 1 << (DCT_LENGTH_LOG - set_count_log); */
-        set_span = shr(dct_length,set_count_log);
-
-        set_count     = shl(1,set_count_log);
+        set_span = shr_nocheck(dct_length,set_count_log);
+
+        set_count     = shl_nocheck(1,set_count_log);
 
         in_ptr        = in_buffer;
@@ -169 +169 @@
                 in_val_low      = *in_ptr++;
                 in_val_high     = *in_ptr++;
-                acca            = L_add(in_val_low,in_val_high);
-                acca            = L_shr(acca,1);
+                // blp: addition of two 16bits vars, there's no way
+                //      they'll overflow a 32bit var
+                //acca            = L_add(in_val_low,in_val_high);
+                acca = (in_val_low + in_val_high);
+                acca            = L_shr_nocheck(acca,1);
                 out_val_low     = extract_l(acca);
 
                 acca            = L_sub(in_val_low,in_val_high);
-                acca            = L_shr(acca,1);
+                acca            = L_shr_nocheck(acca,1);
                 out_val_high    = extract_l(acca);
 
@@ -214 +217 @@
 
     temp = sub(dct_length_log,1);
-    temp = shl(1,temp);
+    temp = shl_nocheck(1,temp);
 
     for (pairs_left=temp; pairs_left > 0; pairs_left--)
@@ -220 +223 @@
         for ( k=0; k<CORE_SIZE; k++ )
         {
+#if PJ_HAS_INT64
+            /* blp: danger danger! not really compatible but faster */
+            pj_int64_t sum64=0;
+            move32();
+            
+            for ( i=0; i<CORE_SIZE; i++ )
+            {
+                sum64 += L_mult(pair_ptr[i], dct_core_a[i][k]);
+            }
+            sum = L_saturate(sum64);
+#else
             sum=0L;
             move32();
@@ -226 +240 @@
                 sum = L_mac(sum, pair_ptr[i],dct_core_a[i][k]);
             }
+#endif
             buffer_swap[k] = itu_round(sum);
         }
@@ -251 +266 @@
         /*===========================================================*/
         /*    set_span      = 1 << (DCT_LENGTH_LOG - set_count_log); */
-        set_span = shr(dct_length,set_count_log);
-
-        set_count     = shl(1,set_count_log);
+        set_span = shr_nocheck(dct_length,set_count_log);
+
+        set_count     = shl_nocheck(1,set_count_log);
         next_in_base  = in_buffer;
         move16();
@@ -278 +293 @@
             in_ptr_low     = next_in_base;
             move16();
-            temp           = shr(set_span,1);
+            temp           = shr_nocheck(set_span,1);
 
             /* address arithmetic */