Change function prototypes for width=8 inner and mbedge loopfilter functions

so that it does both U and V planes at the same time. This will have speed
advantages when using SSE2 (or higher) optimizations, since we can do both
the U and V rows together in a single xmm register.

This also renames filter16 to filter16y and filter8 to filter8uv so that it's
more obvious what each function is used for.

Originally committed as revision 24337 to svn://svn.ffmpeg.org/ffmpeg/trunk

5 files changed, 96 insertions, 58 deletions

diff --git a/libavcodec/vp8.c b/libavcodec/vp8.c
index 0b4434b32f..031c845800 100644
--- a/libavcodec/vp8.c
+++ b/libavcodec/vp8.c
@@ -1245,31 +1245,45 @@ static void filter_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb, int mb_
      bedge_lim = 2* filter_level    + inner_limit;
 
     if (mb_x) {
-        s->vp8dsp.vp8_h_loop_filter16(dst[0], s->linesize,   mbedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter8 (dst[1], s->uvlinesize, mbedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter8 (dst[2], s->uvlinesize, mbedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter16y(dst[0],     s->linesize,
+                                       mbedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter8uv(dst[1],     dst[2],      s->uvlinesize,
+                                       mbedge_lim, inner_limit, hev_thresh);
     }
 
     if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
-        s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+ 4, s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+ 8, s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+12, s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter8_inner (dst[1]+ 4, s->uvlinesize, bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_h_loop_filter8_inner (dst[2]+ 4, s->uvlinesize, bedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 4, s->linesize, bedge_lim,
+                                             inner_limit,   hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 8, s->linesize, bedge_lim,
+                                             inner_limit,   hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+12, s->linesize, bedge_lim,
+                                             inner_limit,   hev_thresh);
+        s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4,    dst[2] + 4,
+                                             s->uvlinesize, bedge_lim,
+                                             inner_limit,   hev_thresh);
     }
 
     if (mb_y) {
-        s->vp8dsp.vp8_v_loop_filter16(dst[0], s->linesize,   mbedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter8 (dst[1], s->uvlinesize, mbedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter8 (dst[2], s->uvlinesize, mbedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter16y(dst[0],     s->linesize,
+                                       mbedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter8uv(dst[1],     dst[2],      s->uvlinesize,
+                                       mbedge_lim, inner_limit, hev_thresh);
     }
 
     if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
-        s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+ 4*s->linesize,   s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+ 8*s->linesize,   s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+12*s->linesize,   s->linesize,   bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter8_inner (dst[1]+ 4*s->uvlinesize, s->uvlinesize, bedge_lim, inner_limit, hev_thresh);
-        s->vp8dsp.vp8_v_loop_filter8_inner (dst[2]+ 4*s->uvlinesize, s->uvlinesize, bedge_lim, inner_limit, hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 4*s->linesize,
+                                             s->linesize,   bedge_lim,
+                                             inner_limit,   hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 8*s->linesize,
+                                             s->linesize,   bedge_lim,
+                                             inner_limit,   hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+12*s->linesize,
+                                             s->linesize,   bedge_lim,
+                                             inner_limit, hev_thresh);
+        s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * s->uvlinesize,
+                                             dst[2] + 4 * s->uvlinesize,
+                                             s->uvlinesize, bedge_lim,
+                                             inner_limit,   hev_thresh);
     }
 }
 
diff --git a/libavcodec/vp8dsp.c b/libavcodec/vp8dsp.c
index 0d2dc01e70..482d62af1b 100644
--- a/libavcodec/vp8dsp.c
+++ b/libavcodec/vp8dsp.c
@@ -196,8 +196,8 @@ static av_always_inline void filter_mbedge(uint8_t *p, int stride)
     p[ 2*stride] = cm[q2 - a2];
 }
 
-#define LOOP_FILTER(dir, size, stridea, strideb) \
-static void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
+#define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
+static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
                                      int flim_E, int flim_I, int hev_thresh)\
 {\
     int i;\
@@ -211,7 +211,7 @@ static void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
         }\
 }\
 \
-static void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
+static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
                                       int flim_E, int flim_I, int hev_thresh)\
 {\
     int i;\
@@ -226,10 +226,26 @@ static void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int st
         }\
 }
 
-LOOP_FILTER(v, 16, 1, stride)
-LOOP_FILTER(h, 16, stride, 1)
-LOOP_FILTER(v,  8, 1, stride)
-LOOP_FILTER(h,  8, stride, 1)
+LOOP_FILTER(v, 16, 1, stride,)
+LOOP_FILTER(h, 16, stride, 1,)
+
+#define UV_LOOP_FILTER(dir, stridea, strideb) \
+LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
+static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, int stride,\
+                                      int fE, int fI, int hev_thresh)\
+{\
+  vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
+  vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
+}\
+static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, int stride,\
+                                      int fE, int fI, int hev_thresh)\
+{\
+  vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
+  vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
+}
+
+UV_LOOP_FILTER(v, 1, stride)
+UV_LOOP_FILTER(h, stride, 1)
 
 static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
 {
@@ -443,15 +459,15 @@ av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
     dsp->vp8_idct_add    = vp8_idct_add_c;
     dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
 
-    dsp->vp8_v_loop_filter16 = vp8_v_loop_filter16_c;
-    dsp->vp8_h_loop_filter16 = vp8_h_loop_filter16_c;
-    dsp->vp8_v_loop_filter8  = vp8_v_loop_filter8_c;
-    dsp->vp8_h_loop_filter8  = vp8_h_loop_filter8_c;
+    dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
+    dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
+    dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
+    dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
 
-    dsp->vp8_v_loop_filter16_inner = vp8_v_loop_filter16_inner_c;
-    dsp->vp8_h_loop_filter16_inner = vp8_h_loop_filter16_inner_c;
-    dsp->vp8_v_loop_filter8_inner  = vp8_v_loop_filter8_inner_c;
-    dsp->vp8_h_loop_filter8_inner  = vp8_h_loop_filter8_inner_c;
+    dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
+    dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
+    dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
+    dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
 
     dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
     dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
diff --git a/libavcodec/vp8dsp.h b/libavcodec/vp8dsp.h
index b660048cc5..64a3bfbc57 100644
--- a/libavcodec/vp8dsp.h
+++ b/libavcodec/vp8dsp.h
@@ -35,16 +35,24 @@ typedef struct VP8DSPContext {
     void (*vp8_idct_dc_add)(uint8_t *dst, DCTELEM block[16], int stride);
 
     // loop filter applied to edges between macroblocks
-    void (*vp8_v_loop_filter16)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_h_loop_filter16)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_v_loop_filter8)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_h_loop_filter8)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter16y)(uint8_t *dst, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter16y)(uint8_t *dst, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter8uv)(uint8_t *dstU, uint8_t *dstV, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter8uv)(uint8_t *dstU, uint8_t *dstV, int stride,
+                                 int flim_E, int flim_I, int hev_thresh);
 
     // loop filter applied to inner macroblock edges
-    void (*vp8_v_loop_filter16_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_h_loop_filter16_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_v_loop_filter8_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
-    void (*vp8_h_loop_filter8_inner)(uint8_t *dst, int stride, int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter16y_inner)(uint8_t *dst, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter16y_inner)(uint8_t *dst, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_v_loop_filter8uv_inner)(uint8_t *dstU, uint8_t *dstV, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
+    void (*vp8_h_loop_filter8uv_inner)(uint8_t *dstU, uint8_t *dstV, int stride,
+                                       int flim_E, int flim_I, int hev_thresh);
 
     void (*vp8_v_loop_filter_simple)(uint8_t *dst, int stride, int flim);
     void (*vp8_h_loop_filter_simple)(uint8_t *dst, int stride, int flim);
diff --git a/libavcodec/x86/vp8dsp-init.c b/libavcodec/x86/vp8dsp-init.c
index edff2883f7..dbaa0dc0a5 100644
--- a/libavcodec/x86/vp8dsp-init.c
+++ b/libavcodec/x86/vp8dsp-init.c
@@ -230,18 +230,18 @@ extern void ff_vp8_h_loop_filter_simple_mmx   (uint8_t *dst, int stride, int fli
 extern void ff_vp8_h_loop_filter_simple_mmxext(uint8_t *dst, int stride, int flim);
 extern void ff_vp8_h_loop_filter_simple_sse2  (uint8_t *dst, int stride, int flim);
 
-extern void ff_vp8_v_loop_filter16_inner_mmx   (uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
-extern void ff_vp8_v_loop_filter16_inner_mmxext(uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
-extern void ff_vp8_v_loop_filter16_inner_sse2  (uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
-extern void ff_vp8_h_loop_filter16_inner_mmx   (uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
-extern void ff_vp8_h_loop_filter16_inner_mmxext(uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
-extern void ff_vp8_h_loop_filter16_inner_sse2  (uint8_t *dst, int stride,
-                                                int e, int i, int hvt);
+extern void ff_vp8_v_loop_filter16y_inner_mmx   (uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
+extern void ff_vp8_v_loop_filter16y_inner_mmxext(uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
+extern void ff_vp8_v_loop_filter16y_inner_sse2  (uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
+extern void ff_vp8_h_loop_filter16y_inner_mmx   (uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
+extern void ff_vp8_h_loop_filter16y_inner_mmxext(uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
+extern void ff_vp8_h_loop_filter16y_inner_sse2  (uint8_t *dst, int stride,
+                                                 int e, int i, int hvt);
 #endif
 
 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
@@ -284,8 +284,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx;
         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx;
 
-        c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_mmx;
-        c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_mmx;
+        c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx;
+        c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx;
     }
 
     /* note that 4-tap width=16 functions are missing because w=16
@@ -302,8 +302,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmxext;
         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmxext;
 
-        c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_mmxext;
-        c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_mmxext;
+        c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext;
+        c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext;
     }
 
     if (mm_flags & FF_MM_SSE) {
@@ -320,8 +320,8 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2;
         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2;
 
-        c->vp8_v_loop_filter16_inner = ff_vp8_v_loop_filter16_inner_sse2;
-        c->vp8_h_loop_filter16_inner = ff_vp8_h_loop_filter16_inner_sse2;
+        c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2;
+        c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2;
     }
 
     if (mm_flags & FF_MM_SSSE3) {
diff --git a/libavcodec/x86/vp8dsp.asm b/libavcodec/x86/vp8dsp.asm
index 70f0ae9286..116064c42f 100644
--- a/libavcodec/x86/vp8dsp.asm
+++ b/libavcodec/x86/vp8dsp.asm
@@ -1379,7 +1379,7 @@ SIMPLE_LOOPFILTER sse2,   h, 6
 ;-----------------------------------------------------------------------------
 
 %macro INNER_LOOPFILTER 4
-cglobal vp8_%2_loop_filter16_inner_%1, 5, %3, %4
+cglobal vp8_%2_loop_filter16y_inner_%1, 5, %3, %4
 %define dst_reg     r0
 %define mstride_reg r1
 %define E_reg       r2