Update to opus-HEAD-66611f1.

silk/dec_API.c

 /***********************************************************************
 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions
 are met:
 - Redistributions of source code must retain the above copyright notice,
 this list of conditions and the following disclaimer.
 - Redistributions in binary form must reproduce the above copyright
 notice, this list of conditions and the following disclaimer in the
 documentation and/or other materials provided with the distribution.
@@ skipping 13 matching lines @@
 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 POSSIBILITY OF SUCH DAMAGE.
 ***********************************************************************/
 
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 #include "API.h"
 #include "main.h"
 #include "stack_alloc.h"
+#include "os_support.h"
 
 /************************/
 /* Decoder Super Struct */
 /************************/
 typedef struct {
     silk_decoder_state          channel_state[ DECODER_NUM_CHANNELS ];
     stereo_dec_state                sStereo;
     opus_int                         nChannelsAPI;
     opus_int                         nChannelsInternal;
     opus_int                         prev_decode_only_middle;
@@ skipping 33 matching lines @@
 }
 
 /* Decode a frame */
 opus_int silk_Decode(                                   /* O    Returns error code                              */
     void*                           decState,           /* I/O  State                                           */
     silk_DecControlStruct*          decControl,         /* I/O  Control Structure                               */
     opus_int                        lostFlag,           /* I    0: no loss, 1 loss, 2 decode fec                */
     opus_int                        newPacketFlag,      /* I    Indicates first decoder call for this packet    */
     ec_dec                          *psRangeDec,        /* I/O  Compressor data structure                       */
     opus_int16                      *samplesOut,        /* O    Decoded output speech vector                    */
-    opus_int32                      *nSamplesOut        /* O    Number of samples decoded                       */
+    opus_int32                      *nSamplesOut,       /* O    Number of samples decoded                       */
+    int                             arch                /* I    Run-time architecture                           */
 )
 {
     opus_int   i, n, decode_only_middle = 0, ret = SILK_NO_ERROR;
     opus_int32 nSamplesOutDec, LBRR_symbol;
     opus_int16 *samplesOut1_tmp[ 2 ];
-    VARDECL( opus_int16, samplesOut1_tmp_storage );
+    VARDECL( opus_int16, samplesOut1_tmp_storage1 );
+    VARDECL( opus_int16, samplesOut1_tmp_storage2 );
     VARDECL( opus_int16, samplesOut2_tmp );
     opus_int32 MS_pred_Q13[ 2 ] = { 0 };
     opus_int16 *resample_out_ptr;
     silk_decoder *psDec = ( silk_decoder * )decState;
     silk_decoder_state *channel_state = psDec->channel_state;
     opus_int has_side;
     opus_int stereo_to_mono;
+    int delay_stack_alloc;
     SAVE_STACK;
 
     silk_assert( decControl->nChannelsInternal == 1 || decControl->nChannelsInternal == 2 );
 
     /**********************************/
     /* Test if first frame in payload */
     /**********************************/
     if( newPacketFlag ) {
         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
             channel_state[ n ].nFramesDecoded = 0;  /* Used to count frames in packet */
@@ skipping 78 matching lines @@
                     }
                 }
             }
         }
 
         if( lostFlag == FLAG_DECODE_NORMAL ) {
             /* Regular decoding: skip all LBRR data */
             for( i = 0; i < channel_state[ 0 ].nFramesPerPacket; i++ ) {
                 for( n = 0; n < decControl->nChannelsInternal; n++ ) {
                     if( channel_state[ n ].LBRR_flags[ i ] ) {
-                        opus_int pulses[ MAX_FRAME_LENGTH ];
+                        opus_int16 pulses[ MAX_FRAME_LENGTH ];
                         opus_int condCoding;
 
                         if( decControl->nChannelsInternal == 2 && n == 0 ) {
                             silk_stereo_decode_pred( psRangeDec, MS_pred_Q13 );
                             if( channel_state[ 1 ].LBRR_flags[ i ] == 0 ) {
                                 silk_stereo_decode_mid_only( psRangeDec, &decode_only_middle );
                             }
                         }
                         /* Use conditional coding if previous frame available */
                         if( i > 0 && channel_state[ n ].LBRR_flags[ i - 1 ] ) {
@@ skipping 34 matching lines @@
     /* Reset side channel decoder prediction memory for first frame with side coding */
     if( decControl->nChannelsInternal == 2 && decode_only_middle == 0 && psDec->prev_decode_only_middle == 1 ) {
         silk_memset( psDec->channel_state[ 1 ].outBuf, 0, sizeof(psDec->channel_state[ 1 ].outBuf) );
         silk_memset( psDec->channel_state[ 1 ].sLPC_Q14_buf, 0, sizeof(psDec->channel_state[ 1 ].sLPC_Q14_buf) );
         psDec->channel_state[ 1 ].lagPrev        = 100;
         psDec->channel_state[ 1 ].LastGainIndex  = 10;
         psDec->channel_state[ 1 ].prevSignalType = TYPE_NO_VOICE_ACTIVITY;
         psDec->channel_state[ 1 ].first_frame_after_reset = 1;
     }
 
-    ALLOC( samplesOut1_tmp_storage,
-           decControl->nChannelsInternal*(
-               channel_state[ 0 ].frame_length + 2 ),
+    /* Check if the temp buffer fits into the output PCM buffer. If it fits,
+       we can delay allocating the temp buffer until after the SILK peak stack
+       usage. We need to use a < and not a <= because of the two extra samples. */
+    delay_stack_alloc = decControl->internalSampleRate*decControl->nChannelsInternal
+          < decControl->API_sampleRate*decControl->nChannelsAPI;
+    ALLOC( samplesOut1_tmp_storage1, delay_stack_alloc ? ALLOC_NONE
+           : decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2 ),
            opus_int16 );
-    samplesOut1_tmp[ 0 ] = samplesOut1_tmp_storage;
-    samplesOut1_tmp[ 1 ] = samplesOut1_tmp_storage
-                           + channel_state[ 0 ].frame_length + 2;
+    if ( delay_stack_alloc )
+    {
+       samplesOut1_tmp[ 0 ] = samplesOut;
+       samplesOut1_tmp[ 1 ] = samplesOut + channel_state[ 0 ].frame_length + 2;
+    } else {
+       samplesOut1_tmp[ 0 ] = samplesOut1_tmp_storage1;
+       samplesOut1_tmp[ 1 ] = samplesOut1_tmp_storage1 + channel_state[ 0 ].frame_length + 2;
+    }
 
     if( lostFlag == FLAG_DECODE_NORMAL ) {
         has_side = !decode_only_middle;
     } else {
         has_side = !psDec->prev_decode_only_middle
               || (decControl->nChannelsInternal == 2 && lostFlag == FLAG_DECODE_LBRR && channel_state[1].LBRR_flags[ channel_state[1].nFramesDecoded ] == 1 );
     }
     /* Call decoder for one frame */
     for( n = 0; n < decControl->nChannelsInternal; n++ ) {
         if( n == 0 || has_side ) {
             opus_int FrameIndex;
             opus_int condCoding;
 
             FrameIndex = channel_state[ 0 ].nFramesDecoded - n;
             /* Use independent coding if no previous frame available */
             if( FrameIndex <= 0 ) {
                 condCoding = CODE_INDEPENDENTLY;
             } else if( lostFlag == FLAG_DECODE_LBRR ) {
                 condCoding = channel_state[ n ].LBRR_flags[ FrameIndex - 1 ] ? CODE_CONDITIONALLY : CODE_INDEPENDENTLY;
             } else if( n > 0 && psDec->prev_decode_only_middle ) {
                 /* If we skipped a side frame in this packet, we don't
                    need LTP scaling; the LTP state is well-defined. */
                 condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
             } else {
                 condCoding = CODE_CONDITIONALLY;
             }
-            ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding);
+            ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding, arch);
         } else {
             silk_memset( &samplesOut1_tmp[ n ][ 2 ], 0, nSamplesOutDec * sizeof( opus_int16 ) );
         }
         channel_state[ n ].nFramesDecoded++;
     }
 
     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) {
         /* Convert Mid/Side to Left/Right */
         silk_stereo_MS_to_LR( &psDec->sStereo, samplesOut1_tmp[ 0 ], samplesOut1_tmp[ 1 ], MS_pred_Q13, channel_state[ 0 ].fs_kHz, nSamplesOutDec );
     } else {
         /* Buffering */
         silk_memcpy( samplesOut1_tmp[ 0 ], psDec->sStereo.sMid, 2 * sizeof( opus_int16 ) );
         silk_memcpy( psDec->sStereo.sMid, &samplesOut1_tmp[ 0 ][ nSamplesOutDec ], 2 * sizeof( opus_int16 ) );
     }
 
     /* Number of output samples */
     *nSamplesOut = silk_DIV32( nSamplesOutDec * decControl->API_sampleRate, silk_SMULBB( channel_state[ 0 ].fs_kHz, 1000 ) );
 
     /* Set up pointers to temp buffers */
     ALLOC( samplesOut2_tmp,
            decControl->nChannelsAPI == 2 ? *nSamplesOut : ALLOC_NONE, opus_int16 );
     if( decControl->nChannelsAPI == 2 ) {
         resample_out_ptr = samplesOut2_tmp;
     } else {
         resample_out_ptr = samplesOut;
     }
 
+    ALLOC( samplesOut1_tmp_storage2, delay_stack_alloc
+           ? decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2 )
+           : ALLOC_NONE,
+           opus_int16 );
+    if ( delay_stack_alloc ) {
+       OPUS_COPY(samplesOut1_tmp_storage2, samplesOut, decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2));
+       samplesOut1_tmp[ 0 ] = samplesOut1_tmp_storage2;
+       samplesOut1_tmp[ 1 ] = samplesOut1_tmp_storage2 + channel_state[ 0 ].frame_length + 2;
+    }
     for( n = 0; n < silk_min( decControl->nChannelsAPI, decControl->nChannelsInternal ); n++ ) {
 
         /* Resample decoded signal to API_sampleRate */
         ret += silk_resampler( &channel_state[ n ].resampler_state, resample_out_ptr, &samplesOut1_tmp[ n ][ 1 ], nSamplesOutDec );
 
         /* Interleave if stereo output and stereo stream */
         if( decControl->nChannelsAPI == 2 ) {
             for( i = 0; i < *nSamplesOut; i++ ) {
                 samplesOut[ n + 2 * i ] = resample_out_ptr[ i ];
             }
@@ skipping 63 matching lines @@
     Silk_TOC->inbandFECFlag = flags & 1;
     for( i = nFramesPerPayload - 1; i >= 0 ; i-- ) {
         flags = silk_RSHIFT( flags, 1 );
         Silk_TOC->VADFlags[ i ] = flags & 1;
         Silk_TOC->VADFlag |= flags & 1;
     }
 
     return ret;
 }
 #endif