VP8 H/V inner loopfilter MMX/MMXEXT/SSE2 optimizations.

Originally committed as revision 24250 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/dsputil_mmx.c

@@ -61,7 +61,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
 
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_1  ) = {0x0101010101010101ULL, 0x0101010101010101ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_4  ) = {0x0404040404040404ULL, 0x0404040404040404ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;

libavcodec/x86/dsputil_mmx.h

@@ -47,7 +47,7 @@ extern const uint64_t ff_pw_96;
 extern const uint64_t ff_pw_128;
 extern const uint64_t ff_pw_255;
 
-extern const uint64_t ff_pb_1;
+extern const xmm_reg  ff_pb_1;
 extern const xmm_reg  ff_pb_3;
 extern const uint64_t ff_pb_7;
 extern const uint64_t ff_pb_1F;

libavcodec/x86/vp8dsp-init.c

@@ -229,6 +229,19 @@ extern void ff_vp8_v_loop_filter_simple_sse2  (uint8_t *dst, int stride, int fli
 extern void ff_vp8_h_loop_filter_simple_mmx   (uint8_t *dst, int stride, int flim);
 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);
 #endif
 
 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
@@ -270,6 +283,9 @@ 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;
     }
 
     /* note that 4-tap width=16 functions are missing because w=16
@@ -285,6 +301,9 @@ 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;
     }
 
     if (mm_flags & FF_MM_SSE) {
@@ -300,6 +319,9 @@ 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;
     }
 
     if (mm_flags & FF_MM_SSSE3) {