Port pred8x8l_down_left_mmxext (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 26162 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 111 insertions, 0 deletions

diff --git a/libavcodec/x86/h264_intrapred.asm b/libavcodec/x86/h264_intrapred.asm
index 3810853362..dd2c499b9d 100644
--- a/libavcodec/x86/h264_intrapred.asm
+++ b/libavcodec/x86/h264_intrapred.asm
@@ -1374,6 +1374,115 @@ PRED8x8L_VERTICAL ssse3
 ;void pred8x8l_down_left(uint8_t *src, int has_topleft, int has_topright, int stride)
 ;-----------------------------------------------------------------------------
 
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+cglobal pred8x8l_down_left_mmxext, 4,5
+    sub          r0, r3
+    movq        mm0, [r0-8]
+    movq        mm3, [r0]
+    movq        mm1, [r0+8]
+    movq        mm2, mm3
+    movq        mm4, mm3
+    PALIGNR     mm2, mm0, 7, mm0
+    PALIGNR     mm1, mm4, 1, mm4
+    test         r1, r1
+    jz .fix_lt_2
+    test         r2, r2
+    jz .fix_tr_1
+    jmp .do_top
+.fix_lt_2:
+    movq        mm5, mm3
+    pxor        mm5, mm2
+    psllq       mm5, 56
+    psrlq       mm5, 56
+    pxor        mm2, mm5
+    test         r2, r2
+    jnz .do_top
+.fix_tr_1:
+    movq        mm5, mm3
+    pxor        mm5, mm1
+    psrlq       mm5, 56
+    psllq       mm5, 56
+    pxor        mm1, mm5
+    jmp .do_top
+.fix_tr_2:
+    punpckhbw   mm3, mm3
+    pshufw      mm1, mm3, 0xFF
+    jmp .do_topright
+.do_top:
+    PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+    movq        mm7, mm4
+    test         r2, r2
+    jz .fix_tr_2
+    movq        mm0, [r0+8]
+    movq        mm5, mm0
+    movq        mm2, mm0
+    movq        mm4, mm0
+    psrlq       mm5, 56
+    PALIGNR     mm2, mm3, 7, mm3
+    PALIGNR     mm5, mm4, 1, mm4
+    PRED4x4_LOWPASS mm1, mm2, mm5, mm0, mm4
+.do_topright:
+    lea          r1, [r0+r3*2]
+    movq        mm6, mm1
+    psrlq       mm1, 56
+    movq        mm4, mm1
+    lea          r2, [r1+r3*2]
+    movq        mm2, mm6
+    PALIGNR     mm2, mm7, 1, mm0
+    movq        mm3, mm6
+    PALIGNR     mm3, mm7, 7, mm0
+    PALIGNR     mm4, mm6, 1, mm0
+    movq        mm5, mm7
+    movq        mm1, mm7
+    movq        mm7, mm6
+    lea          r4, [r2+r3*2]
+    psllq       mm1, 8
+    PRED4x4_LOWPASS mm0, mm1, mm2, mm5, mm6
+    PRED4x4_LOWPASS mm1, mm3, mm4, mm7, mm6
+    movq  [r4+r3*2], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r4+r3*1], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r2+r3*2], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r2+r3*1], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r1+r3*2], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r1+r3*1], mm1
+    movq        mm2, mm0
+    psllq       mm1, 8
+    psrlq       mm2, 56
+    psllq       mm0, 8
+    por         mm1, mm2
+    movq  [r0+r3*2], mm1
+    psllq       mm1, 8
+    psrlq       mm0, 56
+    por         mm1, mm0
+    movq  [r0+r3*1], mm1
+    RET
+
 %macro PRED8x8L_DOWN_LEFT 1
 cglobal pred8x8l_down_left_%1, 4,4
     sub          r0, r3
diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c
index 0a45d4f3ff..e93fb917cc 100644
--- a/libavcodec/x86/h264_intrapred_init.c
+++ b/libavcodec/x86/h264_intrapred_init.c
@@ -67,6 +67,7 @@ void ff_pred8x8l_horizontal_mmxext (uint8_t *src, int has_topleft, int has_topri
 void ff_pred8x8l_horizontal_ssse3  (uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_vertical_mmxext   (uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_vertical_ssse3    (uint8_t *src, int has_topleft, int has_topright, int stride);
+void ff_pred8x8l_down_left_mmxext  (uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_down_left_sse2    (uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_down_left_ssse3   (uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred8x8l_down_right_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
@@ -132,6 +133,7 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
         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;
+        h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_mmxext;
         h->pred8x8l [HOR_DOWN_PRED       ] = ff_pred8x8l_horizontal_down_mmxext;
         h->pred4x4  [DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_mmxext;
         h->pred4x4  [VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_mmxext;