Faster b-adapt + adaptive quantization

Factor out pow to be only called once per macroblock.  Speeds up b-adapt, especially b-adapt 2, considerably.
Speed boost is as high as 24% with b-adapt 2 + b-frames 16.

diff --git a/common/frame.c b/common/frame.c
index 26c564075f870faa9121ad2251ab29ff58116910..85bbcc0d1c97e37660c7c158745619cb4bdf1963 100644 (file)
--- a/common/frame.c
+++ b/common/frame.c
@@ -124,7 +124,11 @@ x264_frame_t *x264_frame_new( x264_t *h )
             CHECKED_MALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) );
 
     if( h->param.rc.i_aq_mode )
+    {
         CHECKED_MALLOC( frame->f_qp_offset, h->mb.i_mb_count * sizeof(float) );
+        if( h->frames.b_have_lowres )
+            CHECKED_MALLOC( frame->i_inv_qscale_factor, h->mb.i_mb_count * sizeof(uint16_t) );
+    }
 
     x264_pthread_mutex_init( &frame->mutex, NULL );
     x264_pthread_cond_init( &frame->cv, NULL );

diff --git a/common/frame.h b/common/frame.h
index 9c27382482ad2e692d3af76acd52a9c88b104c17..aad77f5e943c5e6dbd3d7fd711dc76e16ecc42af 100644 (file)
--- a/common/frame.h
+++ b/common/frame.h
@@ -83,6 +83,7 @@ typedef struct
     float   *f_qp_offset;
     int     b_intra_calculated;
     uint16_t *i_intra_cost;
+    uint16_t *i_inv_qscale_factor;
 
     /* threading */
     int     i_lines_completed; /* in pixels */

diff --git a/encoder/encoder.c b/encoder/encoder.c
index 4a9860fa0ee44b425c632a94371543aa24611fa7..c22f7b0cf5cf6a9e35a45c46518da8d8454e18a2 100644 (file)
--- a/encoder/encoder.c
+++ b/encoder/encoder.c
@@ -511,7 +511,8 @@ static int x264_validate_parameters( x264_t *h )
         h->mb.i_psy_trellis = 0;
     h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
     h->param.rc.i_aq_mode = x264_clip3( h->param.rc.i_aq_mode, 0, 1 );
-    if( h->param.rc.f_aq_strength <= 0 )
+    h->param.rc.f_aq_strength = x264_clip3f( h->param.rc.f_aq_strength, 0, 3 );
+    if( h->param.rc.f_aq_strength == 0 )
         h->param.rc.i_aq_mode = 0;
     h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
 

diff --git a/encoder/ratecontrol.c b/encoder/ratecontrol.c
index 002b79db0dc6223f187647b9ac0a2522c2d01612..b0ee384a5f16cf016460a7f10b9fb45155da1c27 100644 (file)
--- a/encoder/ratecontrol.c
+++ b/encoder/ratecontrol.c
@@ -201,6 +201,8 @@ void x264_adaptive_quant_frame( x264_t *h, x264_frame_t *frame )
             /* 10 constant chosen to result in approximately the same overall bitrate as without AQ. */
             float qp_adj = h->param.rc.f_aq_strength * 1.5 * (logf(energy) - 10.0);
             frame->f_qp_offset[mb_x + mb_y*h->mb.i_mb_stride] = qp_adj;
+            if( h->frames.b_have_lowres )
+                frame->i_inv_qscale_factor[mb_x+mb_y*h->mb.i_mb_stride] = FIX8(pow(2.0,-qp_adj/6.0));
         }
 }
 

diff --git a/encoder/slicetype.c b/encoder/slicetype.c
index 180448e94186913ada2fc8de343a4426aa34dad6..a8c028cfe4902c1c59885dcaf2f6cc87c13101be 100644 (file)
--- a/encoder/slicetype.c
+++ b/encoder/slicetype.c
@@ -306,10 +306,7 @@ static int x264_slicetype_frame_cost( x264_t *h, x264_mb_analysis_t *a,
                     int i_mb_cost = x264_slicetype_mb_cost( h, a, frames, p0, p1, b, dist_scale_factor, do_search );
                     int i_mb_cost_aq = i_mb_cost;
                     if( h->param.rc.i_aq_mode )
-                    {
-                        x264_emms();
-                        i_mb_cost_aq *= pow(2.0,-(frames[b]->f_qp_offset[h->mb.i_mb_x + h->mb.i_mb_y*h->mb.i_mb_stride])/6.0);
-                    }
+                        i_mb_cost_aq = (i_mb_cost_aq * frames[b]->i_inv_qscale_factor[h->mb.i_mb_x + h->mb.i_mb_y*h->mb.i_mb_stride] + 128) >> 8;
                     row_satd[ h->mb.i_mb_y ] += i_mb_cost_aq;
                     if( h->mb.i_mb_y > 0 && h->mb.i_mb_y < h->sps->i_mb_height - 1 &&
                         h->mb.i_mb_x > 0 && h->mb.i_mb_x < h->sps->i_mb_width - 1 )
@@ -329,10 +326,7 @@ static int x264_slicetype_frame_cost( x264_t *h, x264_mb_analysis_t *a,
                     int i_mb_cost = x264_slicetype_mb_cost( h, a, frames, p0, p1, b, dist_scale_factor, do_search );
                     int i_mb_cost_aq = i_mb_cost;
                     if( h->param.rc.i_aq_mode )
-                    {
-                        x264_emms();
-                        i_mb_cost_aq *= pow(2.0,-(frames[b]->f_qp_offset[h->mb.i_mb_x + h->mb.i_mb_y*h->mb.i_mb_stride])/6.0);
-                    }
+                        i_mb_cost_aq = (i_mb_cost_aq * frames[b]->i_inv_qscale_factor[h->mb.i_mb_x + h->mb.i_mb_y*h->mb.i_mb_stride] + 128) >> 8;
                     i_score += i_mb_cost;
                     i_score_aq += i_mb_cost_aq;
                 }