Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).

Originally committed as revision 24029 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/dsputil_mmx.c

@@ -63,12 +63,16 @@ 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_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_4  ) = {0x0404040404040404ULL, 0x0404040404040404ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_80 ) = {0x8080808080808080ULL, 0x8080808080808080ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_F8 ) = {0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_FE ) = {0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL};
 
 DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
 DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };

libavcodec/x86/vp8dsp-init.c

@@ -222,6 +222,13 @@ extern void ff_vp8_idct_dc_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
 extern void ff_vp8_idct_dc_add_sse4(uint8_t *dst, DCTELEM block[16], int stride);
 extern void ff_vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16]);
 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
+
+extern void ff_vp8_v_loop_filter_simple_mmx   (uint8_t *dst, int stride, int flim);
+extern void ff_vp8_v_loop_filter_simple_mmxext(uint8_t *dst, int stride, int flim);
+extern void ff_vp8_v_loop_filter_simple_sse2  (uint8_t *dst, int stride, int flim);
+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);
 #endif
 
 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
@@ -260,6 +267,9 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
         c->put_vp8_epel_pixels_tab[1][0][0]     =
         c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
+
+        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;
     }
 
     /* note that 4-tap width=16 functions are missing because w=16
@@ -272,6 +282,9 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
         VP8_BILINEAR_MC_FUNC(1, 8, mmxext);
         VP8_BILINEAR_MC_FUNC(2, 4, mmxext);
+
+        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;
     }
 
     if (mm_flags & FF_MM_SSE) {
@@ -284,6 +297,9 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         VP8_MC_FUNC(1, 8, sse2);
         VP8_BILINEAR_MC_FUNC(0, 16, sse2);
         VP8_BILINEAR_MC_FUNC(1, 8, sse2);
+
+        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;
     }
 
     if (mm_flags & FF_MM_SSSE3) {

libavcodec/x86/x86util.asm

@@ -37,6 +37,14 @@
     SWAP %2, %4, %3
 %endmacro
 
+%macro TRANSPOSE4x4B 5
+    SBUTTERFLY bw, %1, %2, %5
+    SBUTTERFLY bw, %3, %4, %5
+    SBUTTERFLY wd, %1, %3, %5
+    SBUTTERFLY wd, %2, %4, %5
+    SWAP %2, %3
+%endmacro
+
 %macro TRANSPOSE4x4W 5
     SBUTTERFLY wd, %1, %2, %5
     SBUTTERFLY wd, %3, %4, %5