]> git.sesse.net Git - x264/blobdiff - common/quant.c
MBAFF: First edge deblocking
[x264] / common / quant.c
index 5d319d1399ad145a353de4682e8c74dd281e4641..cc4ea8649da84894b4973fb4901822d41843d3d8 100644 (file)
@@ -1,7 +1,7 @@
 /*****************************************************************************
  * quant.c: quantization and level-run
  *****************************************************************************
- * Copyright (C) 2005-2010 x264 project
+ * Copyright (C) 2005-2011 x264 project
  *
  * Authors: Loren Merritt <lorenm@u.washington.edu>
  *          Fiona Glaser <fiona@x264.com>
@@ -141,9 +141,69 @@ static void dequant_4x4_dc( dctcoef dct[16], int dequant_mf[6][16], int i_qp )
     }
 }
 
+static ALWAYS_INLINE void idct_dequant_2x2_dconly( dctcoef out[4], dctcoef dct[4], int dequant_mf )
+{
+    int d0 = dct[0] + dct[1];
+    int d1 = dct[2] + dct[3];
+    int d2 = dct[0] - dct[1];
+    int d3 = dct[2] - dct[3];
+    out[0] = (d0 + d1) * dequant_mf >> 5;
+    out[1] = (d0 - d1) * dequant_mf >> 5;
+    out[2] = (d2 + d3) * dequant_mf >> 5;
+    out[3] = (d2 - d3) * dequant_mf >> 5;
+}
+
+static ALWAYS_INLINE int idct_dequant_round_2x2_dc( dctcoef ref[4], dctcoef dct[4], int dequant_mf )
+{
+    dctcoef out[4];
+    idct_dequant_2x2_dconly( out, dct, dequant_mf );
+    return ((ref[0] ^ (out[0]+32))
+          | (ref[1] ^ (out[1]+32))
+          | (ref[2] ^ (out[2]+32))
+          | (ref[3] ^ (out[3]+32))) >> 6;
+}
+
+static int optimize_chroma_dc( dctcoef dct[4], int dequant_mf )
+{
+    /* dequant_mf = h->dequant4_mf[CQM_4IC + b_inter][i_qp%6][0] << i_qp/6, max 32*64 */
+    dctcoef dct_orig[4];
+    int coeff, nz;
+
+    idct_dequant_2x2_dconly( dct_orig, dct, dequant_mf );
+    dct_orig[0] += 32;
+    dct_orig[1] += 32;
+    dct_orig[2] += 32;
+    dct_orig[3] += 32;
+
+    /* If the DC coefficients already round to zero, terminate early. */
+    if( !((dct_orig[0]|dct_orig[1]|dct_orig[2]|dct_orig[3])>>6) )
+        return 0;
+
+    /* Start with the highest frequency coefficient... is this the best option? */
+    for( nz = 0, coeff = 3; coeff >= 0; coeff-- )
+    {
+        int level = dct[coeff];
+        int sign = level>>31 | 1; /* dct2x2[coeff] < 0 ? -1 : 1 */
+
+        while( level )
+        {
+            dct[coeff] = level - sign;
+            if( idct_dequant_round_2x2_dc( dct_orig, dct, dequant_mf ) )
+            {
+                nz = 1;
+                dct[coeff] = level;
+                break;
+            }
+            level -= sign;
+        }
+    }
+
+    return nz;
+}
+
 static void x264_denoise_dct( dctcoef *dct, uint32_t *sum, udctcoef *offset, int size )
 {
-    for( int i = 1; i < size; i++ )
+    for( int i = 0; i < size; i++ )
     {
         int level = dct[i];
         int sign = level>>31;
@@ -272,6 +332,8 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
     pf->dequant_4x4_dc = dequant_4x4_dc;
     pf->dequant_8x8 = dequant_8x8;
 
+    pf->optimize_chroma_dc = optimize_chroma_dc;
+
     pf->denoise_dct = x264_denoise_dct;
     pf->decimate_score15 = x264_decimate_score15;
     pf->decimate_score16 = x264_decimate_score16;
@@ -427,6 +489,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
             pf->dequant_4x4 = x264_dequant_4x4_flat16_sse2;
             pf->dequant_8x8 = x264_dequant_8x8_flat16_sse2;
         }
+        pf->optimize_chroma_dc = x264_optimize_chroma_dc_sse2;
         pf->denoise_dct = x264_denoise_dct_sse2;
         pf->decimate_score15 = x264_decimate_score15_sse2;
         pf->decimate_score16 = x264_decimate_score16_sse2;
@@ -457,6 +520,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
         pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3;
         pf->quant_4x4 = x264_quant_4x4_ssse3;
         pf->quant_8x8 = x264_quant_8x8_ssse3;
+        pf->optimize_chroma_dc = x264_optimize_chroma_dc_ssse3;
         pf->denoise_dct = x264_denoise_dct_ssse3;
         pf->decimate_score15 = x264_decimate_score15_ssse3;
         pf->decimate_score16 = x264_decimate_score16_ssse3;
@@ -473,6 +537,16 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
         pf->quant_4x4_dc = x264_quant_4x4_dc_sse4;
         pf->quant_4x4 = x264_quant_4x4_sse4;
         pf->quant_8x8 = x264_quant_8x8_sse4;
+        pf->optimize_chroma_dc = x264_optimize_chroma_dc_sse4;
+    }
+
+    if( cpu&X264_CPU_AVX )
+    {
+        pf->dequant_4x4 = x264_dequant_4x4_avx;
+        pf->dequant_8x8 = x264_dequant_8x8_avx;
+        pf->dequant_4x4_dc = x264_dequant_4x4dc_avx;
+        pf->optimize_chroma_dc = x264_optimize_chroma_dc_avx;
+        pf->denoise_dct = x264_denoise_dct_avx;
     }
 #endif // HAVE_MMX