Port pred8x8l_horizontal_up_mmxext/ssse3 (H.264 intra prediction) from x264

(authors: Jason, Loren, Holger) to FFmpeg. Patch by Daniel Kang <daniel dot
d dot kang at gmail com>, as part of Google's GCI 2010.

Originally committed as revision 26149 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/h264_intrapred.asm

@@ -2045,6 +2045,101 @@ INIT_MMX
 PRED8x8L_VERTICAL_LEFT ssse3
 %endif
 
+;-----------------------------------------------------------------------------
+; void pred8x8l_horizontal_up(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%ifdef CONFIG_GPL
+%macro PRED8x8L_HORIZONTAL_UP 1
+cglobal pred8x8l_horizontal_up_%1, 4,4
+    sub          r0, r3
+    lea          r2, [r0+r3*2]
+    movq        mm0, [r0+r3*1-8]
+    punpckhbw   mm0, [r0+r3*0-8]
+    movq        mm1, [r2+r3*1-8]
+    punpckhbw   mm1, [r0+r3*2-8]
+    mov          r2, r0
+    punpckhwd   mm1, mm0
+    lea          r0, [r0+r3*4]
+    movq        mm2, [r0+r3*1-8]
+    punpckhbw   mm2, [r0+r3*0-8]
+    lea          r0, [r0+r3*2]
+    movq        mm3, [r0+r3*1-8]
+    punpckhbw   mm3, [r0+r3*0-8]
+    punpckhwd   mm3, mm2
+    punpckhdq   mm3, mm1
+    lea          r0, [r0+r3*2]
+    movq        mm0, [r0+r3*0-8]
+    movq        mm1, [r2]
+    mov          r0, r2
+    movq        mm4, mm3
+    movq        mm2, mm3
+    PALIGNR     mm4, mm0, 7, mm0
+    PALIGNR     mm1, mm2, 1, mm2
+    test        r1, r1
+    jnz .do_left
+.fix_lt_1:
+    movq        mm5, mm3
+    pxor        mm5, mm4
+    psrlq       mm5, 56
+    psllq       mm5, 48
+    pxor        mm1, mm5
+.do_left:
+    movq       mm0, mm4
+    PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+    movq       mm4, mm0
+    movq       mm7, mm2
+    PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+    psllq      mm1, 56
+    PALIGNR    mm7, mm1, 7, mm3
+    lea         r1, [r0+r3*2]
+    pshufw     mm0, mm7, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
+    psllq      mm7, 56             ; l7 .. .. .. .. .. .. ..
+    movq       mm2, mm0
+    psllw      mm0, 8
+    psrlw      mm2, 8
+    por        mm2, mm0            ; l7 l6 l5 l4 l3 l2 l1 l0
+    movq       mm3, mm2
+    movq       mm4, mm2
+    movq       mm5, mm2
+    psrlq      mm2, 8
+    psrlq      mm3, 16
+    lea         r2, [r1+r3*2]
+    por        mm2, mm7            ; l7 l7 l6 l5 l4 l3 l2 l1
+    punpckhbw  mm7, mm7
+    por        mm3, mm7            ; l7 l7 l7 l6 l5 l4 l3 l2
+    pavgb      mm4, mm2
+    PRED4x4_LOWPASS mm1, mm3, mm5, mm2, mm6
+    movq       mm5, mm4
+    punpcklbw  mm4, mm1            ; p4 p3 p2 p1
+    punpckhbw  mm5, mm1            ; p8 p7 p6 p5
+    movq       mm6, mm5
+    movq       mm7, mm5
+    movq       mm0, mm5
+    PALIGNR    mm5, mm4, 2, mm1
+    pshufw     mm1, mm6, 11111001b
+    PALIGNR    mm6, mm4, 4, mm2
+    pshufw     mm2, mm7, 11111110b
+    PALIGNR    mm7, mm4, 6, mm3
+    pshufw     mm3, mm0, 11111111b
+    movq [r0+r3*1], mm4
+    movq [r0+r3*2], mm5
+    lea         r0, [r2+r3*2]
+    movq [r1+r3*1], mm6
+    movq [r1+r3*2], mm7
+    movq [r2+r3*1], mm0
+    movq [r2+r3*2], mm1
+    movq [r0+r3*1], mm2
+    movq [r0+r3*2], mm3
+    RET
+%endmacro
+
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+PRED8x8L_HORIZONTAL_UP mmxext
+%define PALIGNR PALIGNR_SSSE3
+PRED8x8L_HORIZONTAL_UP ssse3
+%endif
+
 ;-----------------------------------------------------------------------------
 ; void pred4x4_dc_mmxext(uint8_t *src, const uint8_t *topright, int stride)
 ;-----------------------------------------------------------------------------

libavcodec/x86/h264_intrapred_init.c

@@ -77,6 +77,8 @@ void ff_pred8x8l_vertical_right_sse2(uint8_t *src, int has_topleft, int has_topr
 void ff_pred8x8l_vertical_right_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_vertical_left_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_vertical_left_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
+void ff_pred8x8l_horizontal_up_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
+void ff_pred8x8l_horizontal_up_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred4x4_dc_mmxext          (uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_down_left_mmxext   (uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_tm_vp8_mmx         (uint8_t *src, const uint8_t *topright, int stride);
@@ -121,6 +123,7 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
         h->pred8x8l [VERT_PRED   ] = ff_pred8x8l_vertical_mmxext;
         h->pred8x8l [DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_mmxext;
         h->pred8x8l [VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_mmxext;
+        h->pred8x8l [HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_mmxext;
 #endif
         h->pred4x4  [DC_PRED     ] = ff_pred4x4_dc_mmxext;
 #if CONFIG_GPL
@@ -189,6 +192,7 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
         h->pred8x8l [DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_ssse3;
         h->pred8x8l [VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_ssse3;
         h->pred8x8l [VERT_LEFT_PRED      ] = ff_pred8x8l_vertical_left_ssse3;
+        h->pred8x8l [HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_ssse3;
 #endif
         if (codec_id == CODEC_ID_VP8) {
             h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_ssse3;