ac3enc: move inner loop of compute_rematrixing_strategy to ac3dsp

5 files changed, 70 insertions, 12 deletions

diff --git a/libavcodec/ac3dsp.c b/libavcodec/ac3dsp.c
index 98c73573cb..69beab64ca 100644
--- a/libavcodec/ac3dsp.c
+++ b/libavcodec/ac3dsp.c
@@ -23,6 +23,7 @@
 #include "avcodec.h"
 #include "ac3.h"
 #include "ac3dsp.h"
+#include "mathops.h"
 
 static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
 {
@@ -169,6 +170,48 @@ static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
     }
 }
 
+static void ac3_sum_square_butterfly_int32_c(int64_t sum[4],
+                                             const int32_t *coef0,
+                                             const int32_t *coef1,
+                                             int len)
+{
+    int i;
+
+    sum[0] = sum[1] = sum[2] = sum[3] = 0;
+
+    for (i = 0; i < len; i++) {
+        int lt = coef0[i];
+        int rt = coef1[i];
+        int md = lt + rt;
+        int sd = lt - rt;
+        MAC64(sum[0], lt, lt);
+        MAC64(sum[1], rt, rt);
+        MAC64(sum[2], md, md);
+        MAC64(sum[3], sd, sd);
+    }
+}
+
+static void ac3_sum_square_butterfly_float_c(float sum[4],
+                                             const float *coef0,
+                                             const float *coef1,
+                                             int len)
+{
+    int i;
+
+    sum[0] = sum[1] = sum[2] = sum[3] = 0;
+
+    for (i = 0; i < len; i++) {
+        float lt = coef0[i];
+        float rt = coef1[i];
+        float md = lt + rt;
+        float sd = lt - rt;
+        sum[0] += lt * lt;
+        sum[1] += rt * rt;
+        sum[2] += md * md;
+        sum[3] += sd * sd;
+    }
+}
+
 av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
 {
     c->ac3_exponent_min = ac3_exponent_min_c;
@@ -180,6 +223,8 @@ av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
     c->update_bap_counts = ac3_update_bap_counts_c;
     c->compute_mantissa_size = ac3_compute_mantissa_size_c;
     c->extract_exponents = ac3_extract_exponents_c;
+    c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c;
+    c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c;
 
     if (ARCH_ARM)
         ff_ac3dsp_init_arm(c, bit_exact);
diff --git a/libavcodec/ac3dsp.h b/libavcodec/ac3dsp.h
index 8eeafd68ac..621841e384 100644
--- a/libavcodec/ac3dsp.h
+++ b/libavcodec/ac3dsp.h
@@ -125,6 +125,12 @@ typedef struct AC3DSPContext {
     int (*compute_mantissa_size)(uint16_t mant_cnt[6][16]);
 
     void (*extract_exponents)(uint8_t *exp, int32_t *coef, int nb_coefs);
+
+    void (*sum_square_butterfly_int32)(int64_t sum[4], const int32_t *coef0,
+                                       const int32_t *coef1, int len);
+
+    void (*sum_square_butterfly_float)(float sum[4], const float *coef0,
+                                       const float *coef1, int len);
 } AC3DSPContext;
 
 void ff_ac3dsp_init    (AC3DSPContext *c, int bit_exact);
diff --git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c
index ea3a46cdfa..b63f3d8df1 100644
--- a/libavcodec/ac3enc_fixed.c
+++ b/libavcodec/ac3enc_fixed.c
@@ -103,6 +103,12 @@ static void scale_coefficients(AC3EncodeContext *s)
     }
 }
 
+static void sum_square_butterfly(AC3EncodeContext *s, int64_t sum[4],
+                                 const int32_t *coef0, const int32_t *coef1,
+                                 int len)
+{
+    s->ac3dsp.sum_square_butterfly_int32(sum, coef0, coef1, len);
+}
 
 /**
  * Clip MDCT coefficients to allowable range.
diff --git a/libavcodec/ac3enc_float.c b/libavcodec/ac3enc_float.c
index 718cc1f2b2..99a0775be2 100644
--- a/libavcodec/ac3enc_float.c
+++ b/libavcodec/ac3enc_float.c
@@ -110,6 +110,12 @@ static void scale_coefficients(AC3EncodeContext *s)
                                chan_size * s->channels);
 }
 
+static void sum_square_butterfly(AC3EncodeContext *s, float sum[4],
+                                 const float *coef0, const float *coef1,
+                                 int len)
+{
+    s->ac3dsp.sum_square_butterfly_float(sum, coef0, coef1, len);
+}
 
 /**
  * Clip MDCT coefficients to allowable range.
diff --git a/libavcodec/ac3enc_template.c b/libavcodec/ac3enc_template.c
index c7243c7644..1ee1b617db 100644
--- a/libavcodec/ac3enc_template.c
+++ b/libavcodec/ac3enc_template.c
@@ -43,6 +43,9 @@ static int normalize_samples(AC3EncodeContext *s);
 
 static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
 
+static void sum_square_butterfly(AC3EncodeContext *s, CoefSumType sum[4],
+                                 const CoefType *coef0, const CoefType *coef1,
+                                 int len);
 
 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
 {
@@ -356,7 +359,7 @@ static void apply_channel_coupling(AC3EncodeContext *s)
 static void compute_rematrixing_strategy(AC3EncodeContext *s)
 {
     int nb_coefs;
-    int blk, bnd, i;
+    int blk, bnd;
     AC3Block *block, *av_uninit(block0);
 
     if (s->channel_mode != AC3_CHMODE_STEREO)
@@ -384,17 +387,9 @@ static void compute_rematrixing_strategy(AC3EncodeContext *s)
             /* calculate calculate sum of squared coeffs for one band in one block */
             int start = ff_ac3_rematrix_band_tab[bnd];
             int end   = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
-            CoefSumType sum[4] = {0,};
-            for (i = start; i < end; i++) {
-                CoefType lt = block->mdct_coef[1][i];
-                CoefType rt = block->mdct_coef[2][i];
-                CoefType md = lt + rt;
-                CoefType sd = lt - rt;
-                MAC_COEF(sum[0], lt, lt);
-                MAC_COEF(sum[1], rt, rt);
-                MAC_COEF(sum[2], md, md);
-                MAC_COEF(sum[3], sd, sd);
-            }
+            CoefSumType sum[4];
+            sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
+                                 block->mdct_coef[2] + start, end - start);
 
             /* compare sums to determine if rematrixing will be used for this band */
             if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))