x86: hpeldsp: avg_pixels_xy2 for mmx2&3dnow

This is a port of the inline assembly of the mmx version to use the
pavg(us|)b instruction.

        8    16
mmx   1498  4355
mmx2  1242  3509

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

2 files changed, 86 insertions, 0 deletions

diff --git a/libavcodec/x86/hpeldsp.asm b/libavcodec/x86/hpeldsp.asm
index 937b59bf4f..2adead218c 100644
--- a/libavcodec/x86/hpeldsp.asm
+++ b/libavcodec/x86/hpeldsp.asm
@@ -29,6 +29,7 @@
 
 SECTION_RODATA
 cextern pb_1
+cextern pw_2
 
 SECTION_TEXT
 
@@ -494,3 +495,79 @@ INIT_MMX mmxext
 AVG_APPROX_PIXELS8_XY2
 INIT_MMX 3dnow
 AVG_APPROX_PIXELS8_XY2
+
+
+; void ff_avg_pixels16_xy2(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
+%macro AVG_PIXELS_XY2 0
+%if cpuflag(sse2)
+cglobal avg_pixels16_xy2, 4,5,8
+%else
+cglobal avg_pixels8_xy2, 4,5
+%endif
+    pxor        m7, m7
+    mova        m6, [pw_2]
+    movu        m0, [r1]
+    movu        m4, [r1+1]
+    mova        m1, m0
+    mova        m5, m4
+    punpcklbw   m0, m7
+    punpcklbw   m4, m7
+    punpckhbw   m1, m7
+    punpckhbw   m5, m7
+    paddusw     m4, m0
+    paddusw     m5, m1
+    xor         r4, r4
+    add         r1, r2
+.loop:
+    movu        m0, [r1+r4]
+    movu        m2, [r1+r4+1]
+    mova        m1, m0
+    mova        m3, m2
+    punpcklbw   m0, m7
+    punpcklbw   m2, m7
+    punpckhbw   m1, m7
+    punpckhbw   m3, m7
+    paddusw     m0, m2
+    paddusw     m1, m3
+    paddusw     m4, m6
+    paddusw     m5, m6
+    paddusw     m4, m0
+    paddusw     m5, m1
+    psrlw       m4, 2
+    psrlw       m5, 2
+    mova        m3, [r0+r4]
+    packuswb    m4, m5
+    PAVGB       m4, m3
+    mova   [r0+r4], m4
+    add         r4, r2
+
+    movu        m2, [r1+r4]
+    movu        m4, [r1+r4+1]
+    mova        m3, m2
+    mova        m5, m4
+    punpcklbw   m2, m7
+    punpcklbw   m4, m7
+    punpckhbw   m3, m7
+    punpckhbw   m5, m7
+    paddusw     m4, m2
+    paddusw     m5, m3
+    paddusw     m0, m6
+    paddusw     m1, m6
+    paddusw     m0, m4
+    paddusw     m1, m5
+    psrlw       m0, 2
+    psrlw       m1, 2
+    mova        m3, [r0+r4]
+    packuswb    m0, m1
+    PAVGB       m0, m3
+    mova   [r0+r4], m0
+    add         r4, r2
+    sub        r3d, 2
+    jnz .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmxext
+AVG_PIXELS_XY2
+INIT_MMX 3dnow
+AVG_PIXELS_XY2
diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c
index 6e9f66340c..5e2ecb53a8 100644
--- a/libavcodec/x86/hpeldsp_init.c
+++ b/libavcodec/x86/hpeldsp_init.c
@@ -74,6 +74,10 @@ void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
                               ptrdiff_t line_size, int h);
 void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
+void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
+                               ptrdiff_t line_size, int h);
+void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
 void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
                                       ptrdiff_t line_size, int h);
 void ff_avg_approx_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
@@ -156,6 +160,7 @@ CALL_2X_PIXELS_EXPORT(ff_put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)
     CALL_2X_PIXELS(avg_pixels16           ## CPUEXT, ff_avg_pixels8           ## CPUEXT, 8) \
     CALL_2X_PIXELS(avg_pixels16_x2        ## CPUEXT, ff_avg_pixels8_x2        ## CPUEXT, 8) \
     CALL_2X_PIXELS(avg_pixels16_y2        ## CPUEXT, ff_avg_pixels8_y2        ## CPUEXT, 8) \
+    CALL_2X_PIXELS(avg_pixels16_xy2       ## CPUEXT, ff_avg_pixels8_xy2       ## CPUEXT, 8) \
     CALL_2X_PIXELS(avg_approx_pixels16_xy2## CPUEXT, ff_avg_approx_pixels8_xy2## CPUEXT, 8)
 
 HPELDSP_AVG_PIXELS16(_3dnow)
@@ -209,6 +214,7 @@ static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)
     c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
     c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
     c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
+    c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;
 
     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
@@ -216,6 +222,7 @@ static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)
     c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
+    c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
 
     if (!(flags & CODEC_FLAG_BITEXACT)) {
         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
@@ -243,6 +250,7 @@ static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
     c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
     c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
     c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
+    c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
 
     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;
     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;
@@ -250,6 +258,7 @@ static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
     c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow;
     c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow;
     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;
+    c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;
 
     if (!(flags & CODEC_FLAG_BITEXACT)){
         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;