Merge commit '9d218d573f8088c606d873e80df572582e6773ef'
[ffmpeg.git] / libavcodec / dcadsp.c
index abeba24..97e46fd 100644 (file)
  */
 
 #include "config.h"
+
 #include "libavutil/attributes.h"
 #include "libavutil/intreadwrite.h"
+
 #include "dcadsp.h"
 
-static void dca_lfe_fir_c(float *out, const float *in, const float *coefs,
-                          int decifactor, float scale)
+static void decode_hf_c(float dst[DCA_SUBBANDS][8],
+                        const int32_t vq_num[DCA_SUBBANDS],
+                        const int8_t hf_vq[1024][32], intptr_t vq_offset,
+                        int32_t scale[DCA_SUBBANDS][2],
+                        intptr_t start, intptr_t end)
+{
+    int i, l;
+
+    for (l = start; l < end; l++) {
+        /* 1 vector -> 32 samples but we only need the 8 samples
+         * for this subsubframe. */
+        const int8_t *ptr = &hf_vq[vq_num[l]][vq_offset];
+        float fscale = scale[l][0] * (1 / 16.0);
+        for (i = 0; i < 8; i++)
+            dst[l][i] = ptr[i] * fscale;
+    }
+}
+
+static inline void dca_lfe_fir(float *out, const float *in, const float *coefs,
+                               int decifactor)
 {
-    float *out2 = out + decifactor;
-    const float *cf0 = coefs;
-    const float *cf1 = coefs + 256;
+    float *out2    = out + 2 * decifactor - 1;
+    int num_coeffs = 256 / decifactor;
     int j, k;
 
     /* One decimated sample generates 2*decifactor interpolated ones */
     for (k = 0; k < decifactor; k++) {
         float v0 = 0.0;
         float v1 = 0.0;
-        for (j = 0; j < 256 / decifactor; j++) {
-            float s = in[-j];
-            v0 += s * *cf0++;
-            v1 += s * *--cf1;
+        for (j = 0; j < num_coeffs; j++, coefs++) {
+            v0 += in[-j]                 * *coefs;
+            v1 += in[j + 1 - num_coeffs] * *coefs;
         }
-        *out++  = v0 * scale;
-        *out2++ = v1 * scale;
+        *out++  = v0;
+        *out2-- = v1;
     }
 }
 
@@ -69,14 +87,33 @@ static void dca_qmf_32_subbands(float samples_in[32][8], int sb_act,
         }
 
         synth->synth_filter_float(imdct, synth_buf_ptr, synth_buf_offset,
-                                  synth_buf2, window, samples_out, raXin, scale);
+                                  synth_buf2, window, samples_out, raXin,
+                                  scale);
         samples_out += 32;
     }
 }
 
+static void dca_lfe_fir0_c(float *out, const float *in, const float *coefs)
+{
+    dca_lfe_fir(out, in, coefs, 32);
+}
+
+static void dca_lfe_fir1_c(float *out, const float *in, const float *coefs)
+{
+    dca_lfe_fir(out, in, coefs, 64);
+}
+
 av_cold void ff_dcadsp_init(DCADSPContext *s)
 {
-    s->lfe_fir = dca_lfe_fir_c;
+    s->lfe_fir[0]      = dca_lfe_fir0_c;
+    s->lfe_fir[1]      = dca_lfe_fir1_c;
     s->qmf_32_subbands = dca_qmf_32_subbands;
-    if (ARCH_ARM) ff_dcadsp_init_arm(s);
+    s->decode_hf       = decode_hf_c;
+
+    if (ARCH_AARCH64)
+        ff_dcadsp_init_aarch64(s);
+    if (ARCH_ARM)
+        ff_dcadsp_init_arm(s);
+    if (ARCH_X86)
+        ff_dcadsp_init_x86(s);
 }