lavc: VP9 decoder

[ffmpeg] / libavcodec / x86 / vp3dsp.asm

diff --git a/libavcodec/x86/vp3dsp.asm b/libavcodec/x86/vp3dsp.asm
index fc1e776a133ab7859ece8dd425336fd1a60c8d04..fc8a047224fa4eb7a87f6d0a21dae28363f8f7df 100644 (file)
--- a/libavcodec/x86/vp3dsp.asm
+++ b/libavcodec/x86/vp3dsp.asm
@@ -33,12 +33,13 @@ vp3_idct_data: times 8 dw 64277
                times 8 dw 25080
                times 8 dw 12785
 
+pb_7:  times 8 db 0x07
+pb_1F: times 8 db 0x1f
+pb_81: times 8 db 0x81
+
 cextern pb_1
 cextern pb_3
-cextern pb_7
-cextern pb_1F
 cextern pb_80
-cextern pb_81
 
 cextern pw_8
 
@@ -500,22 +501,22 @@ cglobal vp3_h_loop_filter, 3, 4
 
     ; at this point, function has completed dequantization + dezigzag +
     ; partial transposition; now do the idct itself
-%define I(x) [%1+16* x     ]
-%define J(x) [%1+16*(x-4)+8]
+%define I(x) [%1+16*x]
+%define J(x) [%1+16*x]
     RowIDCT
     Transpose
 
-%define I(x) [%1+16* x   +64]
-%define J(x) [%1+16*(x-4)+72]
+%define I(x) [%1+16*x+8]
+%define J(x) [%1+16*x+8]
     RowIDCT
     Transpose
 
-%define I(x) [%1+16*x]
-%define J(x) [%1+16*x]
+%define I(x) [%1+16* x]
+%define J(x) [%1+16*(x-4)+8]
     ColumnIDCT
 
-%define I(x) [%1+16*x+8]
-%define J(x) [%1+16*x+8]
+%define I(x) [%1+16* x   +64]
+%define J(x) [%1+16*(x-4)+72]
     ColumnIDCT
 %endif ; mmsize == 16/8
 %endmacro
@@ -533,10 +534,17 @@ cglobal vp3_idct_put, 3, 4, 9
     mova          m1, [r2+mmsize*2+%%i]
     mova          m2, [r2+mmsize*4+%%i]
     mova          m3, [r2+mmsize*6+%%i]
+%if mmsize == 8
+    packsswb      m0, [r2+mmsize*8+%%i]
+    packsswb      m1, [r2+mmsize*10+%%i]
+    packsswb      m2, [r2+mmsize*12+%%i]
+    packsswb      m3, [r2+mmsize*14+%%i]
+%else
     packsswb      m0, [r2+mmsize*1+%%i]
     packsswb      m1, [r2+mmsize*3+%%i]
     packsswb      m2, [r2+mmsize*5+%%i]
     packsswb      m3, [r2+mmsize*7+%%i]
+%endif
     paddb         m0, m4
     paddb         m1, m4
     paddb         m2, m4
@@ -560,49 +568,95 @@ cglobal vp3_idct_put, 3, 4, 9
     movq   [r0+r1*2], m3
     movhps [r0+r3  ], m3
 %endif
-%assign %%i %%i+64
+%assign %%i %%i+8
+%endrep
+
+    pxor          m0, m0
+%assign %%offset 0
+%rep 128/mmsize
+    mova [r2+%%offset], m0
+%assign %%offset %%offset+mmsize
 %endrep
     RET
 
 cglobal vp3_idct_add, 3, 4, 9
     VP3_IDCT      r2
 
-    mov           r3, 4
-    pxor          m4, m4
     movsxdifnidn  r1, r1d
-.loop:
+    lea           r3, [r1*3]
+    pxor          m4, m4
+%if mmsize == 16
+%assign %%i 0
+%rep 2
     movq          m0, [r0]
     movq          m1, [r0+r1]
-%if mmsize == 8
-    mova          m2, m0
-    mova          m3, m1
-%endif
+    movq          m2, [r0+r1*2]
+    movq          m3, [r0+r3]
     punpcklbw     m0, m4
     punpcklbw     m1, m4
-%if mmsize == 8
-    punpckhbw     m2, m4
-    punpckhbw     m3, m4
-%endif
-    paddsw        m0, [r2+ 0]
-    paddsw        m1, [r2+16]
-%if mmsize == 8
-    paddsw        m2, [r2+ 8]
-    paddsw        m3, [r2+24]
-    packuswb      m0, m2
-    packuswb      m1, m3
-%else ; mmsize == 16
+    punpcklbw     m2, m4
+    punpcklbw     m3, m4
+    paddsw        m0, [r2+ 0+%%i]
+    paddsw        m1, [r2+16+%%i]
+    paddsw        m2, [r2+32+%%i]
+    paddsw        m3, [r2+48+%%i]
     packuswb      m0, m1
+    packuswb      m2, m3
+    movq   [r0     ], m0
+    movhps [r0+r1  ], m0
+    movq   [r0+r1*2], m2
+    movhps [r0+r3  ], m2
+%if %%i == 0
+    lea           r0, [r0+r1*4]
 %endif
-    movq     [r0   ], m0
-%if mmsize == 8
-    movq     [r0+r1], m1
-%else ; mmsize == 16
-    movhps   [r0+r1], m0
+%assign %%i %%i+64
+%endrep
+%else
+%assign %%i 0
+%rep 2
+    movq          m0, [r0]
+    movq          m1, [r0+r1]
+    movq          m2, [r0+r1*2]
+    movq          m3, [r0+r3]
+    movq          m5, m0
+    movq          m6, m1
+    movq          m7, m2
+    punpcklbw     m0, m4
+    punpcklbw     m1, m4
+    punpcklbw     m2, m4
+    punpckhbw     m5, m4
+    punpckhbw     m6, m4
+    punpckhbw     m7, m4
+    paddsw        m0, [r2+ 0+%%i]
+    paddsw        m1, [r2+16+%%i]
+    paddsw        m2, [r2+32+%%i]
+    paddsw        m5, [r2+64+%%i]
+    paddsw        m6, [r2+80+%%i]
+    paddsw        m7, [r2+96+%%i]
+    packuswb      m0, m5
+    movq          m5, m3
+    punpcklbw     m3, m4
+    punpckhbw     m5, m4
+    packuswb      m1, m6
+    paddsw        m3, [r2+48+%%i]
+    paddsw        m5, [r2+112+%%i]
+    packuswb      m2, m7
+    packuswb      m3, m5
+    movq   [r0     ], m0
+    movq   [r0+r1  ], m1
+    movq   [r0+r1*2], m2
+    movq   [r0+r3  ], m3
+%if %%i == 0
+    lea           r0, [r0+r1*4]
+%endif
+%assign %%i %%i+8
+%endrep
 %endif
-    lea           r0, [r0+r1*2]
-    add           r2, 32
-    dec           r3
-    jg .loop
+%assign %%i 0
+%rep 128/mmsize
+    mova    [r2+%%i], m4
+%assign %%i %%i+mmsize
+%endrep
     RET
 %endmacro
 
@@ -620,7 +674,7 @@ vp3_idct_funcs
     paddusb       m2, m0
     movq          m4, [r0+r1*2]
     paddusb       m3, m0
-    movq          m5, [r0+r3  ]
+    movq          m5, [r0+r2  ]
     paddusb       m4, m0
     paddusb       m5, m0
     psubusb       m2, m1
@@ -630,7 +684,7 @@ vp3_idct_funcs
     movq   [r0+r1  ], m3
     psubusb       m5, m1
     movq   [r0+r1*2], m4
-    movq   [r0+r3  ], m5
+    movq   [r0+r2  ], m5
 %endmacro
 
 INIT_MMX mmxext
@@ -638,11 +692,12 @@ cglobal vp3_idct_dc_add, 3, 4
 %if ARCH_X86_64
     movsxd        r1, r1d
 %endif
-    lea           r3, [r1*3]
-    movsx         r2, word [r2]
-    add           r2, 15
-    sar           r2, 5
-    movd          m0, r2d
+    movsx         r3, word [r2]
+    mov    word [r2], 0
+    lea           r2, [r1*3]
+    add           r3, 15
+    sar           r3, 5
+    movd          m0, r3d
     pshufw        m0, m0, 0x0
     pxor          m1, m1
     psubw         m1, m0