hevc: store the 'tc' chroma deblocking values as int16_t

smarter · Jan 10, 2014 · 5ffd7fc · 5ffd7fc
1 parent 7c6acc6
commit 5ffd7fc
Show file tree

Hide file tree

Showing 4 changed files with 12 additions and 14 deletions.
diff --git a/libavcodec/hevc_filter.c b/libavcodec/hevc_filter.c
@@ -45,7 +45,7 @@ static const uint8_t betatable[52] = {
     38, 40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60, 62, 64                      // QP 38...51
 };
 
-static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
+static int16_t chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
 {
     static const int qp_c[] = {
         29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
@@ -335,7 +335,8 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
 {
     uint8_t *src;
     int x, y, x_end, y_end, chroma;
-    int c_tc[2], beta[2], tc[2];
+    int beta[2], tc[2];
+    int16_t c_tc[2];
     uint8_t no_p[2] = { 0 };
     uint8_t no_q[2] = { 0 };
 

diff --git a/libavcodec/hevcdsp.h b/libavcodec/hevcdsp.h
@@ -85,20 +85,20 @@ typedef struct HEVCDSPContext {
                                     int *beta, int *tc,
                                     uint8_t *no_p, uint8_t *no_q);
     void (*hevc_h_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
-                                      int *tc, uint8_t *no_p, uint8_t *no_q);
+                                      int16_t *tc, uint8_t *no_p, uint8_t *no_q);
     void (*hevc_v_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
-                                      int *tc, uint8_t *no_p, uint8_t *no_q);
+                                      int16_t *tc, uint8_t *no_p, uint8_t *no_q);
     void (*hevc_h_loop_filter_luma_c)(uint8_t *pix, ptrdiff_t stride,
                                       int *beta, int *tc,
                                       uint8_t *no_p, uint8_t *no_q);
     void (*hevc_v_loop_filter_luma_c)(uint8_t *pix, ptrdiff_t stride,
                                       int *beta, int *tc,
                                       uint8_t *no_p, uint8_t *no_q);
     void (*hevc_h_loop_filter_chroma_c)(uint8_t *pix, ptrdiff_t stride,
-                                        int *tc, uint8_t *no_p,
+                                        int16_t *tc, uint8_t *no_p,
                                         uint8_t *no_q);
     void (*hevc_v_loop_filter_chroma_c)(uint8_t *pix, ptrdiff_t stride,
-                                        int *tc, uint8_t *no_p,
+                                        int16_t *tc, uint8_t *no_p,
                                         uint8_t *no_q);
 } HEVCDSPContext;
 

diff --git a/libavcodec/hevcdsp_template.c b/libavcodec/hevcdsp_template.c
@@ -1258,7 +1258,7 @@ static void FUNC(hevc_loop_filter_luma)(uint8_t *_pix,
 }
 
 static void FUNC(hevc_loop_filter_chroma)(uint8_t *_pix, ptrdiff_t _xstride,
-                                          ptrdiff_t _ystride, int *_tc,
+                                          ptrdiff_t _ystride, int16_t *_tc,
                                           uint8_t *_no_p, uint8_t *_no_q)
 {
     int d, j, no_p, no_q;
@@ -1292,14 +1292,14 @@ static void FUNC(hevc_loop_filter_chroma)(uint8_t *_pix, ptrdiff_t _xstride,
 }
 
 static void FUNC(hevc_h_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
-                                            int *tc, uint8_t *no_p,
+                                            int16_t *tc, uint8_t *no_p,
                                             uint8_t *no_q)
 {
     FUNC(hevc_loop_filter_chroma)(pix, stride, sizeof(pixel), tc, no_p, no_q);
 }
 
 static void FUNC(hevc_v_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
-                                            int *tc, uint8_t *no_p,
+                                            int16_t *tc, uint8_t *no_p,
                                             uint8_t *no_q)
 {
     FUNC(hevc_loop_filter_chroma)(pix, sizeof(pixel), stride, tc, no_p, no_q);

diff --git a/libavcodec/x86/hevc_deblock.asm b/libavcodec/x86/hevc_deblock.asm
@@ -256,12 +256,9 @@ ff_hevc_chroma_deblock_body:
     paddw         m5, m4
 
     ; tc calculations
-    movd          m6, [r2]       ; tc0
-    add           r2, 4
+    movd          m6, [r2]
     punpcklwd     m6, m6
-    movd          m7, [r2]       ; tc1
-    punpcklwd     m7, m7
-    shufps        m6, m7, 0      ; tc0, tc1
+    punpckldq     m6, m6         ; tc0, tc1
     movdqa        m4, m6
     pcmpeqw       m7, m7         ; set all bits to 1
     pxor          m4, m7         ; flip all bits of first reg