[FFmpeg-cvslog] hevc: use intreadwrite

[Christophe Gisquet]

ffmpeg | branch: master | Christophe Gisquet <christophe.gisquet at gmail.com> | Fri Aug  8 18:21:02 2014 +0000| [cf6090dc6252f2b276aa4133e3d73a89f4c6046c] | committer: Anton Khirnov

hevc: use intreadwrite

When dealing with MVs, both components may be processed at a time.

Signed-off-by: Anton Khirnov <anton at khirnov.net>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=cf6090dc6252f2b276aa4133e3d73a89f4c6046c
---

 libavcodec/hevc.c     |    3 +--
 libavcodec/hevc.h     |    2 +-
 libavcodec/hevc_mvs.c |   26 +++++++++++---------------
 3 files changed, 13 insertions(+), 18 deletions(-)

diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 4794e4b..ef41ddf 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1714,8 +1714,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                 }
 
                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
-                    lc->pu.mvd.x = 0;
-                    lc->pu.mvd.y = 0;
+                    AV_ZERO32(&lc->pu.mvd);
                 } else {
                     hls_mvd_coding(s, x0, y0, 1);
                 }
diff --git a/libavcodec/hevc.h b/libavcodec/hevc.h
index 03980b7..93c5125 100644
--- a/libavcodec/hevc.h
+++ b/libavcodec/hevc.h
@@ -621,7 +621,7 @@ typedef struct Mv {
 } Mv;
 
 typedef struct MvField {
-    Mv mv[2];
+    DECLARE_ALIGNED(4, Mv, mv)[2];
     int8_t ref_idx[2];
     int8_t pred_flag[2];
     uint8_t is_intra;
diff --git a/libavcodec/hevc_mvs.c b/libavcodec/hevc_mvs.c
index 8d86b59..89b514f 100644
--- a/libavcodec/hevc_mvs.c
+++ b/libavcodec/hevc_mvs.c
@@ -118,20 +118,21 @@ static int isDiffMER(HEVCContext *s, int xN, int yN, int xP, int yP)
            yN >> plevel == yP >> plevel;
 }
 
+#define MATCH_MV(x) (AV_RN32A(&A.x) == AV_RN32A(&B.x))
 #define MATCH(x) (A.x == B.x)
 
 // check if the mv's and refidx are the same between A and B
 static int compareMVrefidx(struct MvField A, struct MvField B)
 {
     if (A.pred_flag[0] && A.pred_flag[1] && B.pred_flag[0] && B.pred_flag[1])
-        return MATCH(ref_idx[0]) && MATCH(mv[0].x) && MATCH(mv[0].y) &&
-               MATCH(ref_idx[1]) && MATCH(mv[1].x) && MATCH(mv[1].y);
+        return MATCH(ref_idx[0]) && MATCH_MV(mv[0]) &&
+               MATCH(ref_idx[1]) && MATCH_MV(mv[1]);
 
     if (A.pred_flag[0] && !A.pred_flag[1] && B.pred_flag[0] && !B.pred_flag[1])
-        return MATCH(ref_idx[0]) && MATCH(mv[0].x) && MATCH(mv[0].y);
+        return MATCH(ref_idx[0]) && MATCH_MV(mv[0]);
 
     if (!A.pred_flag[0] && A.pred_flag[1] && !B.pred_flag[0] && B.pred_flag[1])
-        return MATCH(ref_idx[1]) && MATCH(mv[1].x) && MATCH(mv[1].y);
+        return MATCH(ref_idx[1]) && MATCH_MV(mv[1]);
 
     return 0;
 }
@@ -508,16 +509,13 @@ static void derive_spatial_merge_candidates(HEVCContext *s, int x0, int y0,
             if (l0_cand.pred_flag[0] && l1_cand.pred_flag[1] &&
                 (refPicList[0].list[l0_cand.ref_idx[0]] !=
                  refPicList[1].list[l1_cand.ref_idx[1]] ||
-                 l0_cand.mv[0].x != l1_cand.mv[1].x ||
-                 l0_cand.mv[0].y != l1_cand.mv[1].y)) {
+                 AV_RN32A(&l0_cand.mv[0]) != AV_RN32A(&l1_cand.mv[1]))) {
                 mergecandlist[nb_merge_cand].ref_idx[0]   = l0_cand.ref_idx[0];
                 mergecandlist[nb_merge_cand].ref_idx[1]   = l1_cand.ref_idx[1];
                 mergecandlist[nb_merge_cand].pred_flag[0] = 1;
                 mergecandlist[nb_merge_cand].pred_flag[1] = 1;
-                mergecandlist[nb_merge_cand].mv[0].x      = l0_cand.mv[0].x;
-                mergecandlist[nb_merge_cand].mv[0].y      = l0_cand.mv[0].y;
-                mergecandlist[nb_merge_cand].mv[1].x      = l1_cand.mv[1].x;
-                mergecandlist[nb_merge_cand].mv[1].y      = l1_cand.mv[1].y;
+                AV_COPY32(&mergecandlist[nb_merge_cand].mv[0], &l0_cand.mv[0]);
+                AV_COPY32(&mergecandlist[nb_merge_cand].mv[1], &l1_cand.mv[1]);
                 mergecandlist[nb_merge_cand].is_intra     = 0;
                 if (merge_idx == nb_merge_cand)
                     return;
@@ -530,10 +528,8 @@ static void derive_spatial_merge_candidates(HEVCContext *s, int x0, int y0,
     while (nb_merge_cand < s->sh.max_num_merge_cand) {
         mergecandlist[nb_merge_cand].pred_flag[0] = 1;
         mergecandlist[nb_merge_cand].pred_flag[1] = s->sh.slice_type == B_SLICE;
-        mergecandlist[nb_merge_cand].mv[0].x      = 0;
-        mergecandlist[nb_merge_cand].mv[0].y      = 0;
-        mergecandlist[nb_merge_cand].mv[1].x      = 0;
-        mergecandlist[nb_merge_cand].mv[1].y      = 0;
+        AV_ZERO32(mergecandlist[nb_merge_cand].mv + 0);
+        AV_ZERO32(mergecandlist[nb_merge_cand].mv + 1);
         mergecandlist[nb_merge_cand].is_intra     = 0;
         mergecandlist[nb_merge_cand].ref_idx[0]   = zero_idx < nb_refs ? zero_idx : 0;
         mergecandlist[nb_merge_cand].ref_idx[1]   = zero_idx < nb_refs ? zero_idx : 0;
@@ -554,7 +550,7 @@ void ff_hevc_luma_mv_merge_mode(HEVCContext *s, int x0, int y0, int nPbW,
 {
     int singleMCLFlag = 0;
     int nCS = 1 << log2_cb_size;
-    struct MvField mergecand_list[MRG_MAX_NUM_CANDS] = { { { { 0 } } } };
+    LOCAL_ALIGNED(4, MvField, mergecand_list, [MRG_MAX_NUM_CANDS]);
     int nPbW2 = nPbW;
     int nPbH2 = nPbH;
     HEVCLocalContext *lc = &s->HEVClc;