changeset 11991:a6d24fc1deb7 libavcodec

Add MMX, SSE2, SSSE3 asm for VP8 bilinear MC
author darkshikari
date Mon, 28 Jun 2010 18:56:24 +0000
parents 3c51d7ac41c9
children da388061b227
files x86/vp8dsp-init.c x86/vp8dsp.asm
diffstat 2 files changed, 314 insertions(+), 110 deletions(-) [+]
line wrap: on
line diff
--- a/x86/vp8dsp-init.c	Mon Jun 28 16:04:14 2010 +0000
+++ b/x86/vp8dsp-init.c	Mon Jun 28 18:56:24 2010 +0000
@@ -23,6 +23,8 @@
 #include "libavutil/x86_cpu.h"
 #include "libavcodec/vp8dsp.h"
 
+#if HAVE_YASM
+
 /*
  * MC functions
  */
@@ -65,96 +67,149 @@
                                        uint8_t *src, int srcstride,
                                        int height, int mx, int my);
 
-#define TAP_W16(OPT, TAPTYPE) \
-static void ff_put_vp8_epel16_ ## TAPTYPE ## _ ## OPT(uint8_t *dst, \
-                                                      int dststride, \
-                                                      uint8_t *src, \
-                                                      int srcstride, \
-                                                      int height, \
-                                                      int mx, int my) \
+extern void ff_put_vp8_bilinear4_h_mmxext(uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+extern void ff_put_vp8_bilinear8_h_sse2  (uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+
+extern void ff_put_vp8_bilinear4_v_mmxext(uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+extern void ff_put_vp8_bilinear8_v_sse2  (uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+extern void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+extern void ff_put_vp8_bilinear8_h_ssse3 (uint8_t *dst, int dststride,
+                                          uint8_t *src, int srcstride,
+                                          int height, int mx, int my);
+
+#define TAP_W16(OPT, FILTERTYPE, TAPTYPE) \
+static void ff_put_vp8_ ## FILTERTYPE ## 16_ ## TAPTYPE ## _ ## OPT( \
+    uint8_t *dst,  int dststride, uint8_t *src, \
+    int srcstride, int height, int mx, int my) \
 { \
-    ff_put_vp8_epel8_ ## TAPTYPE ## _ ## OPT(dst,     dststride, \
-                                             src,     srcstride, \
-                                             height, mx, my); \
-    ff_put_vp8_epel8_ ## TAPTYPE ## _ ## OPT(dst + 8, dststride, \
-                                             src + 8, srcstride, \
-                                             height, mx, my); \
+    ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
+        dst,     dststride, src,     srcstride, height, mx, my); \
+    ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
+        dst + 8, dststride, src + 8, srcstride, height, mx, my); \
 }
-#define TAP_W8(OPT, TAPTYPE) \
-static void ff_put_vp8_epel8_ ## TAPTYPE ## _ ## OPT(uint8_t *dst, \
-                                                     int dststride, \
-                                                     uint8_t *src, \
-                                                     int srcstride, \
-                                                     int height, \
-                                                     int mx, int my) \
+#define TAP_W8(OPT, FILTERTYPE, TAPTYPE) \
+static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
+    uint8_t *dst,  int dststride, uint8_t *src, \
+    int srcstride, int height, int mx, int my) \
 { \
-    ff_put_vp8_epel4_ ## TAPTYPE ## _ ## OPT(dst,     dststride, \
-                                             src,     srcstride, \
-                                             height, mx, my); \
-    ff_put_vp8_epel4_ ## TAPTYPE ## _ ## OPT(dst + 4, dststride, \
-                                             src + 4, srcstride, \
-                                             height, mx, my); \
+    ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
+        dst,     dststride, src,     srcstride, height, mx, my); \
+    ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
+        dst + 4, dststride, src + 4, srcstride, height, mx, my); \
 }
 
-#if HAVE_YASM
-TAP_W8 (mmxext, h4)
-TAP_W8 (mmxext, h6)
-TAP_W16(mmxext, h6)
-TAP_W8 (mmxext, v4)
-TAP_W8 (mmxext, v6)
-TAP_W16(mmxext, v6)
+TAP_W8 (mmxext, epel, h4)
+TAP_W8 (mmxext, epel, h6)
+TAP_W16(mmxext, epel, h6)
+TAP_W8 (mmxext, epel, v4)
+TAP_W8 (mmxext, epel, v6)
+TAP_W16(mmxext, epel, v6)
+TAP_W8 (mmxext, bilinear, h)
+TAP_W16(mmxext, bilinear, h)
+TAP_W8 (mmxext, bilinear, v)
+TAP_W16(mmxext, bilinear, v)
 
-TAP_W16(sse2,   h6)
-TAP_W16(sse2,   v6)
+TAP_W16(sse2,   epel, h6)
+TAP_W16(sse2,   epel, v6)
+TAP_W16(sse2,   bilinear, h)
+TAP_W16(sse2,   bilinear, v)
 
-TAP_W16(ssse3,  h6)
-TAP_W16(ssse3,  v6)
-#endif
+TAP_W16(ssse3,  epel, h6)
+TAP_W16(ssse3,  epel, v6)
+TAP_W16(ssse3,  bilinear, h)
+TAP_W16(ssse3,  bilinear, v)
 
 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \
-static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT \
-                                                (uint8_t *dst, int dststride, \
-                                                 uint8_t *src, int srcstride, \
-                                                 int height, int mx, int my) \
+static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT( \
+    uint8_t *dst, int dststride, uint8_t *src, \
+    int srcstride, int height, int mx, int my) \
 { \
     DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \
     uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \
     src -= srcstride * (TAPNUMY / 2 - 1); \
-    ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT(tmp, SIZE, \
-                                                         src, srcstride, \
-                                                         height + TAPNUMY - 1, \
-                                                         mx, my); \
-    ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT(dst, dststride, \
-                                                         tmpptr, SIZE, \
-                                                         height, mx, my); \
+    ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT( \
+        tmp, SIZE,      src,    srcstride, height + TAPNUMY - 1, mx, my); \
+    ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT( \
+        dst, dststride, tmpptr, SIZE,      height,               mx, my); \
 }
 
 #define HVTAPMMX(x, y) \
 HVTAP(mmxext, 8, x, y,  4,  8) \
 HVTAP(mmxext, 8, x, y,  8, 16)
 
-#if HAVE_YASM
 HVTAPMMX(4, 4)
 HVTAPMMX(4, 6)
 HVTAPMMX(6, 4)
 HVTAPMMX(6, 6)
 HVTAP(mmxext, 8, 6, 6, 16, 16)
-#endif
 
 #define HVTAPSSE2(x, y, w) \
 HVTAP(sse2,  16, x, y, w, 16) \
 HVTAP(ssse3, 16, x, y, w, 16)
 
-#if HAVE_YASM
 HVTAPSSE2(4, 4, 8)
 HVTAPSSE2(4, 6, 8)
 HVTAPSSE2(6, 4, 8)
 HVTAPSSE2(6, 6, 8)
 HVTAPSSE2(6, 6, 16)
-#endif
+
+#define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \
+static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \
+    uint8_t *dst, int dststride, uint8_t *src, \
+    int srcstride, int height, int mx, int my) \
+{ \
+    DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + 2)]; \
+    ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT( \
+        tmp, SIZE,      src, srcstride, height + 1, mx, my); \
+    ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT( \
+        dst, dststride, tmp, SIZE,      height,     mx, my); \
+}
+
+HVBILIN(mmxext, 8,  4,  8)
+HVBILIN(mmxext, 8,  8, 16)
+HVBILIN(mmxext, 8, 16, 16)
+HVBILIN(sse2,   8,  8, 16)
+HVBILIN(sse2,   8, 16, 16)
+HVBILIN(ssse3,  8,  8, 16)
+HVBILIN(ssse3,  8, 16, 16)
 
 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);
+#endif
+
+#define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
+    c->put_vp8_epel_pixels_tab[IDX][0][2] = ff_put_vp8_epel ## SIZE ## _h6_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][2][0] = ff_put_vp8_epel ## SIZE ## _v6_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][2][2] = ff_put_vp8_epel ## SIZE ## _h6v6_ ## OPT
+
+#define VP8_MC_FUNC(IDX, SIZE, OPT) \
+    c->put_vp8_epel_pixels_tab[IDX][0][1] = ff_put_vp8_epel ## SIZE ## _h4_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][1][0] = ff_put_vp8_epel ## SIZE ## _v4_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][1][1] = ff_put_vp8_epel ## SIZE ## _h4v4_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][1][2] = ff_put_vp8_epel ## SIZE ## _h6v4_ ## OPT; \
+    c->put_vp8_epel_pixels_tab[IDX][2][1] = ff_put_vp8_epel ## SIZE ## _h4v6_ ## OPT; \
+    VP8_LUMA_MC_FUNC(IDX, SIZE, OPT)
+
+#define VP8_BILINEAR_MC_FUNC(IDX, SIZE, OPT) \
+    c->put_vp8_bilinear_pixels_tab[IDX][0][1] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][0][2] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][1][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][1][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][1][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][2][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][2][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
+    c->put_vp8_bilinear_pixels_tab[IDX][2][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT
+
 
 av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
 {
@@ -168,53 +223,26 @@
     /* note that 4-tap width=16 functions are missing because w=16
      * is only used for luma, and luma is always a copy or sixtap. */
     if (mm_flags & FF_MM_MMXEXT) {
-        c->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_mmxext;
-        c->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_mmxext;
-        c->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_mmxext;
-        c->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_mmxext;
-        c->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_mmxext;
-        c->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_mmxext;
-        c->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_mmxext;
-        c->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_mmxext;
-        c->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_mmxext;
-        c->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_mmxext;
-        c->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_mmxext;
-        c->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_mmxext;
-        c->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_mmxext;
-        c->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_mmxext;
-        c->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_mmxext;
-        c->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_mmxext;
-        c->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_mmxext;
-        c->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_mmxext;
-        c->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_mmxext;
+        VP8_LUMA_MC_FUNC(0, 16, mmxext);
+        VP8_MC_FUNC(1, 8, mmxext);
+        VP8_MC_FUNC(1, 4, mmxext);
+        VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
+        VP8_BILINEAR_MC_FUNC(1, 8, mmxext);
+        VP8_BILINEAR_MC_FUNC(1, 4, mmxext);
     }
 
     if (mm_flags & FF_MM_SSE2) {
-        c->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_sse2;
-        c->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_sse2;
-        c->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_sse2;
-        c->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_sse2;
-        c->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_sse2;
-        c->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_sse2;
-        c->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_sse2;
-        c->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_sse2;
-        c->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_sse2;
-        c->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_sse2;
-        c->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_sse2;
+        VP8_LUMA_MC_FUNC(0, 16, sse2);
+        VP8_MC_FUNC(1, 8, sse2);
+        VP8_BILINEAR_MC_FUNC(0, 16, sse2);
+        VP8_BILINEAR_MC_FUNC(1, 8, sse2);
     }
 
     if (mm_flags & FF_MM_SSSE3) {
-        c->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_ssse3;
-        c->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_ssse3;
-        c->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_ssse3;
-        c->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_ssse3;
-        c->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_ssse3;
-        c->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_ssse3;
-        c->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_ssse3;
-        c->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_ssse3;
-        c->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_ssse3;
-        c->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_ssse3;
-        c->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_ssse3;
+        VP8_LUMA_MC_FUNC(0, 16, ssse3);
+        VP8_MC_FUNC(1, 8, ssse3);
+        VP8_BILINEAR_MC_FUNC(0, 16, ssse3);
+        VP8_BILINEAR_MC_FUNC(1, 8, ssse3);
     }
 
     if (mm_flags & FF_MM_SSE4) {
--- a/x86/vp8dsp.asm	Mon Jun 28 16:04:14 2010 +0000
+++ b/x86/vp8dsp.asm	Mon Jun 28 18:56:24 2010 +0000
@@ -98,13 +98,31 @@
                      times 8 dw -11
                      times 8 dw   2
 
+bilinear_filter_vw_m: times 8 dw 1
+                      times 8 dw 2
+                      times 8 dw 3
+                      times 8 dw 4
+                      times 8 dw 5
+                      times 8 dw 6
+                      times 8 dw 7
+
+bilinear_filter_vb_m: times 8 db 7, 1
+                      times 8 db 6, 2
+                      times 8 db 5, 3
+                      times 8 db 4, 4
+                      times 8 db 3, 5
+                      times 8 db 2, 6
+                      times 8 db 1, 7
+
 %ifdef PIC
-%define fourtap_filter_hw r11
-%define sixtap_filter_hw  r11
-%define fourtap_filter_hb r11
-%define sixtap_filter_hb  r11
-%define fourtap_filter_v  r11
-%define sixtap_filter_v   r11
+%define fourtap_filter_hw    r11
+%define sixtap_filter_hw     r11
+%define fourtap_filter_hb    r11
+%define sixtap_filter_hb     r11
+%define fourtap_filter_v     r11
+%define sixtap_filter_v      r11
+%define bilinear_filter_vw   r11
+%define bilinear_filter_vb   r11
 %else
 %define fourtap_filter_hw fourtap_filter_hw_m
 %define sixtap_filter_hw  sixtap_filter_hw_m
@@ -112,14 +130,16 @@
 %define sixtap_filter_hb  sixtap_filter_hb_m
 %define fourtap_filter_v  fourtap_filter_v_m
 %define sixtap_filter_v   sixtap_filter_v_m
+%define bilinear_filter_vw bilinear_filter_vw_m
+%define bilinear_filter_vb bilinear_filter_vb_m
 %endif
 
-filter_v4_shuf1: db 0, 3, 1, 4, 2, 5, 3, 6, 4, 7, 5,  8, 6,  9, 7, 10
-filter_v4_shuf2: db 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,  7, 7,  8, 8,  9
+filter_h2_shuf:  db 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5,  6, 6,  7,  7,  8
+filter_h4_shuf:  db 0, 3, 1, 4, 2, 5, 3, 6, 4, 7, 5,  8, 6,  9,  7, 10
 
-filter_v6_shuf1: db 0, 5, 1, 6, 2, 7, 3, 8, 4, 9, 5, 10, 6, 11,  7, 12
-filter_v6_shuf2: db 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,  7, 7,  8,  8,  9
-filter_v6_shuf3: db 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8,  9, 9, 10, 10, 11
+filter_h6_shuf1: db 0, 5, 1, 6, 2, 7, 3, 8, 4, 9, 5, 10, 6, 11,  7, 12
+filter_h6_shuf2: db 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,  7, 7,  8,  8,  9
+filter_h6_shuf3: db 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8,  9, 9, 10, 10, 11
 
 cextern pw_4
 cextern pw_64
@@ -361,8 +381,8 @@
 cglobal put_vp8_epel8_h4_ssse3, 6, 6, 7
     shl      r5d, 4
     mova      m2, [pw_64]
-    mova      m3, [filter_v4_shuf1]
-    mova      m4, [filter_v4_shuf2]
+    mova      m3, [filter_h4_shuf]
+    mova      m4, [filter_h6_shuf2]
 %ifdef PIC
     lea      r11, [fourtap_filter_hb_m]
 %endif
@@ -391,8 +411,8 @@
 
 cglobal put_vp8_epel8_h6_ssse3, 6, 6, 8
     lea      r5d, [r5*3]
-    mova      m3, [filter_v6_shuf1]
-    mova      m4, [filter_v6_shuf2]
+    mova      m3, [filter_h6_shuf1]
+    mova      m4, [filter_h6_shuf2]
 %ifdef PIC
     lea      r11, [sixtap_filter_hb_m]
 %endif
@@ -406,7 +426,7 @@
     mova      m2, m0
     pshufb    m0, m3
     pshufb    m1, m4
-    pshufb    m2, [filter_v6_shuf3]
+    pshufb    m2, [filter_h6_shuf3]
     pmaddubsw m0, m5
     pmaddubsw m1, m6
     pmaddubsw m2, m7
@@ -634,6 +654,162 @@
     jg .nextrow
     REP_RET
 
+%macro FILTER_BILINEAR 3
+cglobal put_vp8_bilinear%2_v_%1, 7,7,%3
+    mov      r5d, 8*16
+    shl      r6d, 4
+    sub      r5d, r6d
+%ifdef PIC
+    lea      r11, [bilinear_filter_vw_m]
+%endif
+    pxor      m6, m6
+    mova      m4, [bilinear_filter_vw+r5d-16]
+    mova      m5, [bilinear_filter_vw+r6d-16]
+.nextrow
+    movh      m0, [r2+r3*0]
+    movh      m1, [r2+r3*1]
+    movh      m3, [r2+r3*2]
+    punpcklbw m0, m6
+    punpcklbw m1, m6
+    punpcklbw m3, m6
+    mova      m2, m1
+    pmullw    m0, m4
+    pmullw    m1, m5
+    pmullw    m2, m4
+    pmullw    m3, m5
+    paddsw    m0, m1
+    paddsw    m2, m3
+    psraw     m0, 2
+    psraw     m2, 2
+    pavgw     m0, m6
+    pavgw     m2, m6
+%ifidn %1, mmxext
+    packuswb  m0, m0
+    packuswb  m2, m2
+    movh [r0+r1*0], m0
+    movh [r0+r1*1], m2
+%else
+    packuswb  m0, m2
+    movh   [r0+r1*0], m0
+    movhps [r0+r1*1], m0
+%endif
+
+    lea       r0, [r0+r1*2]
+    lea       r2, [r2+r3*2]
+    sub       r4, 2
+    jg .nextrow
+    REP_RET
+
+cglobal put_vp8_bilinear%2_h_%1, 7,7,%3
+    mov      r6d, 8*16
+    shl      r5d, 4
+    sub      r6d, r5d
+%ifdef PIC
+    lea      r11, [bilinear_filter_vw_m]
+%endif
+    pxor      m6, m6
+    mova      m4, [bilinear_filter_vw+r6d-16]
+    mova      m5, [bilinear_filter_vw+r5d-16]
+.nextrow
+    movh      m0, [r2+r3*0+0]
+    movh      m1, [r2+r3*0+1]
+    movh      m2, [r2+r3*1+0]
+    movh      m3, [r2+r3*1+1]
+    punpcklbw m0, m6
+    punpcklbw m1, m6
+    punpcklbw m2, m6
+    punpcklbw m3, m6
+    pmullw    m0, m4
+    pmullw    m1, m5
+    pmullw    m2, m4
+    pmullw    m3, m5
+    paddsw    m0, m1
+    paddsw    m2, m3
+    psraw     m0, 2
+    psraw     m2, 2
+    pavgw     m0, m6
+    pavgw     m2, m6
+%ifidn %1, mmxext
+    packuswb  m0, m0
+    packuswb  m2, m2
+    movh [r0+r1*0], m0
+    movh [r0+r1*1], m2
+%else
+    packuswb  m0, m2
+    movh   [r0+r1*0], m0
+    movhps [r0+r1*1], m0
+%endif
+
+    lea       r0, [r0+r1*2]
+    lea       r2, [r2+r3*2]
+    sub       r4, 2
+    jg .nextrow
+    REP_RET
+%endmacro
+
+INIT_MMX
+FILTER_BILINEAR mmxext, 4, 0
+INIT_XMM
+FILTER_BILINEAR   sse2, 8, 7
+
+cglobal put_vp8_bilinear8_v_ssse3, 7,7,5
+    shl      r6d, 4
+%ifdef PIC
+    lea      r11, [bilinear_filter_vb_m]
+%endif
+    pxor      m4, m4
+    mova      m3, [bilinear_filter_vb+r6d-16]
+.nextrow
+    movh      m0, [r2+r3*0]
+    movh      m1, [r2+r3*1]
+    movh      m2, [r2+r3*2]
+    punpcklbw m0, m1
+    punpcklbw m1, m2
+    pmaddubsw m0, m3
+    pmaddubsw m1, m3
+    psraw     m0, 2
+    psraw     m1, 2
+    pavgw     m0, m4
+    pavgw     m1, m4
+    packuswb  m0, m1
+    movh   [r0+r1*0], m0
+    movhps [r0+r1*1], m0
+
+    lea       r0, [r0+r1*2]
+    lea       r2, [r2+r3*2]
+    sub       r4, 2
+    jg .nextrow
+    REP_RET
+
+cglobal put_vp8_bilinear8_h_ssse3, 7,7,5
+    shl      r5d, 4
+%ifdef PIC
+    lea      r11, [bilinear_filter_vb_m]
+%endif
+    pxor      m4, m4
+    mova      m2, [filter_h2_shuf]
+    mova      m3, [bilinear_filter_vb+r5d-16]
+.nextrow
+    movu      m0, [r2+r3*0]
+    movu      m1, [r2+r3*1]
+    pshufb    m0, m2
+    pshufb    m1, m2
+    pmaddubsw m0, m3
+    pmaddubsw m1, m3
+    psraw     m0, 2
+    psraw     m1, 2
+    pavgw     m0, m4
+    pavgw     m1, m4
+    packuswb  m0, m1
+    movh   [r0+r1*0], m0
+    movhps [r0+r1*1], m0
+
+    lea       r0, [r0+r1*2]
+    lea       r2, [r2+r3*2]
+    sub       r4, 2
+    jg .nextrow
+    REP_RET
+
 ;-----------------------------------------------------------------------------
 ; IDCT functions:
 ;