Replaces inline copies and initialization with OPUS_*() macros.

This is a bit faster at -O2 because memcpy()/memmove()/memset() are
vectorized. The code is also cleaner.

3 files changed, 30 insertions, 48 deletions

diff --git a/celt/bands.c b/celt/bands.c
index cce56e2f..cd7d88d1 100644
--- a/celt/bands.c
+++ b/celt/bands.c
@@ -249,8 +249,7 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
          } while (++j<band_end);
       }
       celt_assert(start <= end);
-      for (i=M*eBands[end];i<N;i++)
-         *f++ = 0;
+      OPUS_CLEAR(&freq[c*N+M*eBands[end]], N-M*eBands[end]);
    } while (++c<C);
 }
 
@@ -409,8 +408,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N)
    Er = MULT16_16(mid2, mid2) + side + 2*xp;
    if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28))
    {
-      for (j=0;j<N;j++)
-         Y[j] = X[j];
+      OPUS_COPY(Y, X, N);
       return;
    }
 
@@ -567,8 +565,7 @@ static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard
          for (j=0;j<N0;j++)
             tmp[i*N0+j] = X[j*stride+i];
    }
-   for (j=0;j<N;j++)
-      X[j] = tmp[j];
+   OPUS_COPY(X, tmp, N);
    RESTORE_STACK;
 }
 
@@ -591,8 +588,7 @@ static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard)
          for (j=0;j<N0;j++)
             tmp[j*stride+i] = X[i*N0+j];
    }
-   for (j=0;j<N;j++)
-      X[j] = tmp[j];
+   OPUS_COPY(X, tmp, N);
    RESTORE_STACK;
 }
 
@@ -1021,8 +1017,7 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
             fill &= cm_mask;
             if (!fill)
             {
-               for (j=0;j<N;j++)
-                  X[j] = 0;
+               OPUS_CLEAR(X, N);
             } else {
                if (lowband == NULL)
                {
@@ -1098,9 +1093,7 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X,
 
    if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1))
    {
-      int j;
-      for (j=0;j<N;j++)
-         lowband_scratch[j] = lowband[j];
+      OPUS_COPY(lowband_scratch, lowband, N);
       lowband = lowband_scratch;
    }
 
diff --git a/celt/celt_encoder.c b/celt/celt_encoder.c
index 917f9195..db183430 100644
--- a/celt/celt_encoder.c
+++ b/celt/celt_encoder.c
@@ -276,8 +276,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
       }
       /*printf("\n");*/
       /* First few samples are bad because we don't propagate the memory */
-      for (i=0;i<12;i++)
-         tmp[i] = 0;
+      OPUS_CLEAR(tmp, 12);
 
 #ifdef FIXED_POINT
       /* Normalize tmp to max range */
@@ -453,8 +452,7 @@ static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS
          int bound = B*N/upsample;
          for (i=0;i<bound;i++)
             out[c*B*N+i] *= upsample;
-         for (;i<B*N;i++)
-            out[c*B*N+i] = 0;
+         OPUS_CLEAR(&out[c*B*N+bound], B*N-bound);
       } while (++c<C);
    }
 }
@@ -489,8 +487,7 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
    Nu = N/upsample;
    if (upsample!=1)
    {
-      for (i=0;i<N;i++)
-         inp[i] = 0;
+      OPUS_CLEAR(inp, N);
    }
    for (i=0;i<Nu;i++)
       inp[i*upsample] = SCALEIN(pcmp[CC*i]);
@@ -586,8 +583,7 @@ static int tf_analysis(const CELTMode *m, int len, int isTransient,
       N = (m->eBands[i+1]-m->eBands[i])<<LM;
       /* band is too narrow to be split down to LM=-1 */
       narrow = (m->eBands[i+1]-m->eBands[i])==1;
-      for (j=0;j<N;j++)
-         tmp[j] = X[tf_chan*N0 + j+(m->eBands[i]<<LM)];
+      OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<<LM)], N);
       /* Just add the right channel if we're in stereo */
       /*if (C==2)
          for (j=0;j<N;j++)
@@ -597,8 +593,7 @@ static int tf_analysis(const CELTMode *m, int len, int isTransient,
       /* Check the -1 case for transients */
       if (isTransient && !narrow)
       {
-         for (j=0;j<N;j++)
-            tmp_1[j] = tmp[j];
+         OPUS_COPY(tmp_1, tmp, N);
          haar1(tmp_1, N>>LM, 1<<LM);
          L1 = l1_metric(tmp_1, N, LM+1, bias);
          if (L1<best_L1)
@@ -903,8 +898,7 @@ static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16
    SAVE_STACK;
    ALLOC(follower, C*nbEBands, opus_val16);
    ALLOC(noise_floor, C*nbEBands, opus_val16);
-   for (i=0;i<nbEBands;i++)
-      offsets[i] = 0;
+   OPUS_CLEAR(offsets, nbEBands);
    /* Dynamic allocation code */
    maxDepth=-QCONST16(31.9f, DB_SHIFT);
    for (i=0;i<end;i++)
@@ -1566,8 +1560,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
    amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C);
 
    ALLOC(surround_dynalloc, C*nbEBands, opus_val16);
-   for(i=0;i<st->end;i++)
-      surround_dynalloc[i] = 0;
+   OPUS_CLEAR(surround_dynalloc, st->end);
    /* This computes how much masking takes place between surround channels */
    if (st->start==0&&st->energy_mask&&!st->lfe)
    {
@@ -1629,8 +1622,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
                disabling masking. */
             mask_avg = 0;
             diff = 0;
-            for(i=0;i<mask_end;i++)
-               surround_dynalloc[i] = 0;
+            OPUS_CLEAR(surround_dynalloc, mask_end);
          } else {
             for(i=0;i<mask_end;i++)
                surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT));
@@ -1666,8 +1658,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 
    if (!secondMdct)
    {
-      for (i=0;i<C*nbEBands;i++)
-         bandLogE2[i] = bandLogE[i];
+      OPUS_COPY(bandLogE2, bandLogE, C*nbEBands);
    }
 
    /* Last chance to catch any transient we might have missed in the
@@ -2059,16 +2050,13 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
 #endif
 
    if (CC==2&&C==1) {
-      for (i=0;i<nbEBands;i++)
-         oldBandE[nbEBands+i]=oldBandE[i];
+      OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands);
    }
 
    if (!isTransient)
    {
-      for (i=0;i<CC*nbEBands;i++)
-         oldLogE2[i] = oldLogE[i];
-      for (i=0;i<CC*nbEBands;i++)
-         oldLogE[i] = oldBandE[i];
+      OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands);
+      OPUS_COPY(oldLogE, oldBandE, CC*nbEBands);
    } else {
       for (i=0;i<CC*nbEBands;i++)
          oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]);
diff --git a/src/opus_encoder.c b/src/opus_encoder.c
index dc73a46b..dd4999b3 100644
--- a/src/opus_encoder.c
+++ b/src/opus_encoder.c
@@ -1425,8 +1425,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_
     ec_enc_init(&enc, data, max_data_bytes-1);
 
     ALLOC(pcm_buf, (total_buffer+frame_size)*st->channels, opus_val16);
-    for (i=0;i<total_buffer*st->channels;i++)
-       pcm_buf[i] = st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels+i];
+    OPUS_COPY(pcm_buf, &st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels], total_buffer*st->channels);
 
     if (st->mode == MODE_CELT_ONLY)
        hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 );
@@ -1611,8 +1610,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_
             prefill_offset = st->channels*(st->encoder_buffer-st->delay_compensation-st->Fs/400);
             gain_fade(st->delay_buffer+prefill_offset, st->delay_buffer+prefill_offset,
                   0, Q15ONE, celt_mode->overlap, st->Fs/400, st->channels, celt_mode->window, st->Fs);
-            for(i=0;i<prefill_offset;i++)
-               st->delay_buffer[i]=0;
+            OPUS_CLEAR(st->delay_buffer, prefill_offset);
 #ifdef FIXED_POINT
             pcm_silk = st->delay_buffer;
 #else
@@ -1739,15 +1737,18 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_
     ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_val16);
     if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0)
     {
-       for (i=0;i<st->channels*st->Fs/400;i++)
-          tmp_prefill[i] = st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels + i];
+       OPUS_COPY(tmp_prefill, &st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels], st->channels*st->Fs/400);
     }
 
-    for (i=0;i<st->channels*(st->encoder_buffer-(frame_size+total_buffer));i++)
-        st->delay_buffer[i] = st->delay_buffer[i+st->channels*frame_size];
-    for (;i<st->encoder_buffer*st->channels;i++)
-        st->delay_buffer[i] = pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels+i];
-
+    if (st->channels*(st->encoder_buffer-(frame_size+total_buffer)) > 0)
+    {
+       OPUS_MOVE(st->delay_buffer, &st->delay_buffer[st->channels*frame_size], st->channels*(st->encoder_buffer-frame_size-total_buffer));
+       OPUS_COPY(&st->delay_buffer[st->channels*(st->encoder_buffer-frame_size-total_buffer)],
+             &pcm_buf[0],
+             (frame_size+total_buffer)*st->channels);
+    } else {
+       OPUS_COPY(st->delay_buffer, &pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels], st->encoder_buffer*st->channels);
+    }
     /* gain_fade() and stereo_fade() need to be after the buffer copying
        because we don't want any of this to affect the SILK part */
     if( st->prev_HB_gain < Q15ONE || HB_gain < Q15ONE ) {