shithub: openh264

ref: 6239fbe131bbcaa71269f02bd28f228a28f7395a
dir: /codec/encoder/core/x86/intra_pred.asm/

View raw version
;*!
;* \copy
;*     Copyright (c)  2009-2013, Cisco Systems
;*     All rights reserved.
;*
;*     Redistribution and use in source and binary forms, with or without
;*     modification, are permitted provided that the following conditions
;*     are met:
;*
;*        * Redistributions of source code must retain the above copyright
;*          notice, this list of conditions and the following disclaimer.
;*
;*        * Redistributions in binary form must reproduce the above copyright
;*          notice, this list of conditions and the following disclaimer in
;*          the documentation and/or other materials provided with the
;*          distribution.
;*
;*     THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
;*     "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
;*     LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
;*     FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
;*     COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
;*     INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
;*     BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
;*     LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
;*     CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
;*     LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
;*     ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
;*     POSSIBILITY OF SUCH DAMAGE.
;*
;*
;*  intra_pred.asm
;*
;*  Abstract
;*      sse2 function for intra predict operations
;*
;*  History
;*      18/09/2009 Created
;*
;*
;*************************************************************************/
%include "asm_inc.asm"

;***********************************************************************
; Local Data (Read Only)
;***********************************************************************

SECTION .rodata align=16

align 16
sse2_plane_inc_minus dw -7, -6, -5, -4, -3, -2, -1, 0
align 16
sse2_plane_inc dw 1, 2, 3, 4, 5, 6, 7, 8
align 16
sse2_plane_dec dw 8, 7, 6, 5, 4, 3, 2, 1

; for chroma plane mode
sse2_plane_inc_c dw 1, 2, 3, 4
sse2_plane_dec_c dw 4, 3, 2, 1
align 16
sse2_plane_mul_b_c dw -3, -2, -1, 0, 1, 2, 3, 4

align 16
mmx_01bytes:        times 16    db 1

align 16
mmx_0x02: dw 0x02, 0x00, 0x00, 0x00


;***********************************************************************
; macros
;***********************************************************************
;dB 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
;%1 will keep the last result
%macro SSE_DB_1_2REG 2
    pxor %1, %1
    pcmpeqw %2, %2
    psubb %1, %2
%endmacro

;xmm0, xmm1, xmm2, eax, ecx
;lower 64 bits of xmm0 save the result
%macro SSE2_PRED_H_4X4_TWO_LINE 5
    movd        %1, [%4-1]
    movdqa      %3, %1
    punpcklbw   %1, %3
    movdqa      %3, %1
    punpcklbw   %1, %3

    ;add            %4, %5
    movd        %2, [%4+%5-1]
    movdqa      %3, %2
    punpcklbw   %2, %3
    movdqa      %3, %2
    punpcklbw   %2, %3
    punpckldq   %1, %2
%endmacro

%macro SUMW_HORIZON1 2
    movdqa      %2, %1
    psrldq      %2, 8
    paddusw     %1, %2
    movdqa      %2, %1
    psrldq      %2, 4
    paddusw     %1, %2
    movdqa      %2, %1
    psrldq      %2, 2
    paddusw     %1, %2
%endmacro

%macro LOAD_COLUMN 6
    movd    %1, [%5]
    movd    %2, [%5+%6]
    punpcklbw %1,   %2
    lea     %5, [%5+2*%6]
    movd    %3, [%5]
    movd    %2, [%5+%6]
    punpcklbw %3,   %2
    punpcklwd %1,   %3
    lea     %5, [%5+2*%6]
    movd    %4, [%5]
    movd    %2, [%5+%6]
    punpcklbw %4,   %2
    lea     %5, [%5+2*%6]
    movd    %3, [%5]
    movd    %2, [%5+%6]
    lea     %5, [%5+2*%6]
    punpcklbw %3,   %2
    punpcklwd %4,   %3
    punpckhdq %1,   %4
%endmacro

%macro SUMW_HORIZON 3
    movhlps     %2, %1          ; x2 = xx xx xx xx d7 d6 d5 d4
    paddw       %1, %2          ; x1 = xx xx xx xx d37 d26 d15 d04
    punpcklwd   %1, %3          ; x1 =  d37  d26 d15 d04
    movhlps     %2, %1          ; x2 = xxxx xxxx d37 d26
    paddd       %1, %2          ; x1 = xxxx xxxx d1357 d0246
    pshuflw     %2, %1, 0x4e    ; x2 = xxxx xxxx d0246 d1357
    paddd       %1, %2          ; x1 = xxxx xxxx xxxx  d01234567
%endmacro


%macro COPY_16_TIMES 2
    movdqa      %2, [%1-16]
    psrldq      %2, 15
    pmuludq     %2, [mmx_01bytes]
    pshufd      %2, %2, 0
%endmacro

%macro COPY_16_TIMESS 3
    movdqa      %2, [%1+%3-16]
    psrldq      %2, 15
    pmuludq     %2, [mmx_01bytes]
    pshufd      %2, %2, 0
%endmacro

%macro LOAD_COLUMN_C 6
    movd    %1, [%5]
    movd    %2, [%5+%6]
    punpcklbw %1,%2
    lea     %5, [%5+2*%6]
    movd    %3, [%5]
    movd    %2, [%5+%6]
    punpcklbw %3,   %2
    punpckhwd %1,   %3
    lea     %5, [%5+2*%6]
%endmacro

%macro LOAD_2_LEFT_AND_ADD 0
    lea         r1, [r1+2*r2]
    movzx       r4, byte [r1-0x01]
    add         r3, r4
    movzx       r4, byte [r1+r2-0x01]
    add         r3, r4
%endmacro

;***********************************************************************
; Code
;***********************************************************************

SECTION .text

;***********************************************************************
;   void WelsI4x4LumaPredH_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride)
;
;   pred must align to 16
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredH_sse2
    push r3
    %assign push_num 1
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    movzx       r3, byte [r1-1]
    movd        xmm0,   r3d
    pmuludq     xmm0,   [mmx_01bytes]

    movzx       r3, byte [r1+r2-1]
    movd        xmm1,   r3d
    pmuludq     xmm1,   [mmx_01bytes]

    unpcklps    xmm0,   xmm1

    lea         r1, [r1+r2*2]
    movzx       r3, byte [r1-1]
    movd        xmm2,   r3d
    pmuludq     xmm2,   [mmx_01bytes]

    movzx       r3, byte [r1+r2-1]
    movd        xmm3,   r3d
    pmuludq     xmm3,   [mmx_01bytes]

    unpcklps    xmm2,   xmm3
    unpcklpd    xmm0,   xmm2

    movdqa      [r0],   xmm0
    pop r3
    ret

;***********************************************************************
; void WelsI16x16LumaPredPlane_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride);
;***********************************************************************
WELS_EXTERN WelsI16x16LumaPredPlane_sse2
    push r3
    push r4
    %assign push_num 2
    LOAD_3_PARA
    PUSH_XMM 8
    SIGN_EXTENSION r2, r2d
    sub     r1, 1
    sub     r1, r2

    ;for H
    pxor    xmm7,   xmm7
    movq    xmm0,   [r1]
    movdqa  xmm5,   [sse2_plane_dec]
    punpcklbw xmm0, xmm7
    pmullw  xmm0,   xmm5
    movq    xmm1,   [r1 + 9]
    movdqa  xmm6,   [sse2_plane_inc]
    punpcklbw xmm1, xmm7
    pmullw  xmm1,   xmm6
    psubw   xmm1,   xmm0

    SUMW_HORIZON    xmm1,xmm0,xmm2
    movd    r3d,    xmm1        ; H += (i + 1) * (top[8 + i] - top[6 - i]);
    movsx   r3, r3w
    imul    r3, 5
    add     r3, 32
    sar     r3, 6           ; b = (5 * H + 32) >> 6;
    SSE2_Copy8Times xmm1, r3d   ; xmm1 = b,b,b,b,b,b,b,b

    movzx   r4, BYTE [r1+16]
    sub r1, 3
    LOAD_COLUMN     xmm0, xmm2, xmm3, xmm4, r1, r2

    add     r1, 3
    movzx   r3, BYTE [r1+8*r2]
    add     r4, r3
    shl     r4, 4           ;   a = (left[15*stride] + top[15]) << 4;

    sub r1, 3
    add     r1, r2
    LOAD_COLUMN     xmm7, xmm2, xmm3, xmm4, r1, r2
    pxor    xmm4,   xmm4
    punpckhbw xmm0, xmm4
    pmullw  xmm0,   xmm5
    punpckhbw xmm7, xmm4
    pmullw  xmm7,   xmm6
    psubw   xmm7,   xmm0

    SUMW_HORIZON   xmm7,xmm0,xmm2
    movd    r3d,   xmm7         ; V
    movsx   r3, r3w
    imul    r3, 5
    add     r3, 32
    sar     r3, 6               ; c = (5 * V + 32) >> 6;
    SSE2_Copy8Times xmm4, r3d       ; xmm4 = c,c,c,c,c,c,c,c

    add     r4, 16
    imul    r3, -7
    add     r3, r4              ; s = a + 16 + (-7)*c
    SSE2_Copy8Times xmm0, r3d       ; xmm0 = s,s,s,s,s,s,s,s

    xor     r3, r3
    movdqa  xmm5,   [sse2_plane_inc_minus]

get_i16x16_luma_pred_plane_sse2_1:
    movdqa  xmm2,   xmm1
    pmullw  xmm2,   xmm5
    paddw   xmm2,   xmm0
    psraw   xmm2,   5
    movdqa  xmm3,   xmm1
    pmullw  xmm3,   xmm6
    paddw   xmm3,   xmm0
    psraw   xmm3,   5
    packuswb xmm2,  xmm3
    movdqa  [r0],   xmm2
    paddw   xmm0,   xmm4
    add     r0, 16
    inc     r3
    cmp     r3, 16
    jnz get_i16x16_luma_pred_plane_sse2_1
    POP_XMM
    pop r4
    pop r3
    ret

;***********************************************************************
; void WelsIChromaPredPlane_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride);
;***********************************************************************
WELS_EXTERN WelsIChromaPredPlane_sse2
    push r3
    push r4
    %assign push_num 2
    LOAD_3_PARA
    PUSH_XMM 8
    SIGN_EXTENSION r2, r2d
    sub     r1, 1
    sub     r1, r2

    pxor    mm7,    mm7
    movq    mm0,    [r1]
    movq    mm5,    [sse2_plane_dec_c]
    punpcklbw mm0,  mm7
    pmullw  mm0,    mm5
    movq    mm1,    [r1 + 5]
    movq    mm6,    [sse2_plane_inc_c]
    punpcklbw mm1,  mm7
    pmullw  mm1,    mm6
    psubw   mm1,    mm0

    movq2dq xmm1,   mm1
    pxor    xmm2,   xmm2
    SUMW_HORIZON    xmm1,xmm0,xmm2
    movd    r3d,    xmm1
    movsx   r3, r3w
    imul    r3, 17
    add     r3, 16
    sar     r3, 5           ; b = (17 * H + 16) >> 5;
    SSE2_Copy8Times xmm1, r3d   ; mm1 = b,b,b,b,b,b,b,b

    movzx   r3, BYTE [r1+8]
    sub r1, 3
    LOAD_COLUMN_C   mm0, mm2, mm3, mm4, r1, r2

    add     r1, 3
    movzx   r4, BYTE [r1+4*r2]
    add     r4, r3
    shl     r4, 4           ; a = (left[7*stride] + top[7]) << 4;

    sub r1, 3
    add     r1, r2
    LOAD_COLUMN_C   mm7, mm2, mm3, mm4, r1, r2
    pxor    mm4,    mm4
    punpckhbw mm0,  mm4
    pmullw  mm0,    mm5
    punpckhbw mm7,  mm4
    pmullw  mm7,    mm6
    psubw   mm7,    mm0

    movq2dq xmm7,   mm7
    pxor    xmm2,   xmm2
    SUMW_HORIZON    xmm7,xmm0,xmm2
    movd    r3d,    xmm7            ; V
    movsx   r3, r3w
    imul    r3, 17
    add     r3, 16
    sar     r3, 5               ; c = (17 * V + 16) >> 5;
    SSE2_Copy8Times xmm4, r3d   ; mm4 = c,c,c,c,c,c,c,c

    add     r4, 16
    imul    r3, -3
    add     r3, r4      ; s = a + 16 + (-3)*c
    SSE2_Copy8Times xmm0, r3d   ; xmm0 = s,s,s,s,s,s,s,s

    xor     r3, r3
    movdqa  xmm5,   [sse2_plane_mul_b_c]

get_i_chroma_pred_plane_sse2_1:
    movdqa  xmm2,   xmm1
    pmullw  xmm2,   xmm5
    paddw   xmm2,   xmm0
    psraw   xmm2,   5
    packuswb xmm2,  xmm2
    movq    [r0],   xmm2
    paddw   xmm0,   xmm4
    add     r0, 8
    inc     r3
    cmp     r3, 8
    jnz get_i_chroma_pred_plane_sse2_1
    POP_XMM
    pop r4
    pop r3
    WELSEMMS
    ret

;***********************************************************************
;   0 |1 |2 |3 |4 |
;   6 |7 |8 |9 |10|
;   11|12|13|14|15|
;   16|17|18|19|20|
;   21|22|23|24|25|
;   7 is the start pixel of current 4x4 block
;   pred[7] = ([6]+[0]*2+[1]+2)/4
;
;   void WelsI4x4LumaPredDDR_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredDDR_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    movq        mm1,[r1+r2-8]       ;get value of 11,decreasing 8 is trying to improve the performance of movq mm1[8] = 11
    movq        mm2,[r1-8]          ;get value of 6 mm2[8] = 6
    sub     r1, r2          ;mov eax to above line of current block(postion of 1)
    punpckhbw   mm2,[r1-8]          ;mm2[8](high 8th byte of mm2) = [0](value of 0), mm2[7]= [6]
    movd        mm3,[r1]            ;get value 1, mm3[1] = [1],mm3[2]=[2],mm3[3]=[3]
    punpckhwd   mm1,mm2             ;mm1[8]=[0],mm1[7]=[6],mm1[6]=[11]
    psllq       mm3,18h             ;mm3[5]=[1]
    psrlq       mm1,28h             ;mm1[3]=[0],mm1[2]=[6],mm1[1]=[11]
    por         mm3,mm1             ;mm3[6]=[3],mm3[5]=[2],mm3[4]=[1],mm3[3]=[0],mm3[2]=[6],mm3[1]=[11]
    movq        mm1,mm3             ;mm1[6]=[3],mm1[5]=[2],mm1[4]=[1],mm1[3]=[0],mm1[2]=[6],mm1[1]=[11]
    lea         r1,[r1+r2*2-8h]     ;set eax point to 12
    movq        mm4,[r1+r2]     ;get value of 16, mm4[8]=[16]
    psllq       mm3,8               ;mm3[7]=[3],mm3[6]=[2],mm3[5]=[1],mm3[4]=[0],mm3[3]=[6],mm3[2]=[11],mm3[1]=0
    psrlq       mm4,38h             ;mm4[1]=[16]
    por         mm3,mm4             ;mm3[7]=[3],mm3[6]=[2],mm3[5]=[1],mm3[4]=[0],mm3[3]=[6],mm3[2]=[11],mm3[1]=[16]
    movq        mm2,mm3             ;mm2[7]=[3],mm2[6]=[2],mm2[5]=[1],mm2[4]=[0],mm2[3]=[6],mm2[2]=[11],mm2[1]=[16]
    movq        mm4,[r1+r2*2]       ;mm4[8]=[21]
    psllq       mm3,8               ;mm3[8]=[3],mm3[7]=[2],mm3[6]=[1],mm3[5]=[0],mm3[4]=[6],mm3[3]=[11],mm3[2]=[16],mm3[1]=0
    psrlq       mm4,38h             ;mm4[1]=[21]
    por         mm3,mm4             ;mm3[8]=[3],mm3[7]=[2],mm3[6]=[1],mm3[5]=[0],mm3[4]=[6],mm3[3]=[11],mm3[2]=[16],mm3[1]=[21]
    movq        mm4,mm3             ;mm4[8]=[3],mm4[7]=[2],mm4[6]=[1],mm4[5]=[0],mm4[4]=[6],mm4[3]=[11],mm4[2]=[16],mm4[1]=[21]
    pavgb       mm3,mm1             ;mm3=([11]+[21]+1)/2
    pxor        mm1,mm4             ;find odd value in the lowest bit of each byte
    pand        mm1,[mmx_01bytes]   ;set the odd bit
    psubusb     mm3,mm1             ;decrease 1 from odd bytes
    pavgb       mm2,mm3             ;mm2=(([11]+[21]+1)/2+1+[16])/2

    movd        [r0+12],mm2
    psrlq       mm2,8
    movd        [r0+8],mm2
    psrlq       mm2,8
    movd        [r0+4],mm2
    psrlq       mm2,8
    movd        [r0],mm2
    WELSEMMS
    ret

;***********************************************************************
;   0 |1 |2 |3 |4 |
;   5 |6 |7 |8 |9 |
;   10|11|12|13|14|
;   15|16|17|18|19|
;   20|21|22|23|24|
;   6 is the start pixel of current 4x4 block
;   pred[6] = ([1]+[2]+[3]+[4]+[5]+[10]+[15]+[20]+4)/8
;
;   void WelsI4x4LumaPredDc_sse2(uint8_t *pred,uint8_t *pRef,int32_t stride)
;
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredDc_sse2
    push r3
    push r4
    %assign push_num 2
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    movzx       r4, byte [r1-1h]
    sub         r1, r2
    movd        xmm0,   [r1]
    pxor        xmm1,   xmm1
    psadbw      xmm0,   xmm1
    xor r3, r3
    movd        r3d,    xmm0
    add         r3, r4
    movzx       r4, byte [r1+r2*2-1h]
    add         r3, r4

    lea         r1, [r1+r2*2-1]
    movzx       r4, byte [r1+r2]
    add         r3, r4

    movzx       r4, byte [r1+r2*2]
    add         r3, r4
    add         r3, 4
    sar         r3, 3
    imul        r3, 0x01010101

    movd        xmm0,   r3d
    pshufd      xmm0,   xmm0,   0
    movdqa      [r0],   xmm0
    pop r4
    pop r3
    ret

;***********************************************************************
;   void WelsIChromaPredH_mmx(uint8_t *pred, uint8_t *pRef, int32_t stride)
;   copy 8 pixel of 8 line from left
;***********************************************************************
%macro MMX_PRED_H_8X8_ONE_LINE 4
    movq        %1,     [%3-8]
    psrlq       %1,     38h

    ;pmuludq        %1,     [mmx_01bytes]       ;extend to 4 bytes
    pmullw      %1,     [mmx_01bytes]
    pshufw      %1,     %1, 0
    movq        [%4],   %1
%endmacro

%macro MMX_PRED_H_8X8_ONE_LINEE 4
    movq        %1,     [%3+r2-8]
    psrlq       %1,     38h

    ;pmuludq        %1,     [mmx_01bytes]       ;extend to 4 bytes
    pmullw      %1,     [mmx_01bytes]
    pshufw      %1,     %1, 0
    movq        [%4],   %1
%endmacro

WELS_EXTERN WelsIChromaPredH_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    movq        mm0,    [r1-8]
    psrlq       mm0,    38h

    ;pmuludq        mm0,    [mmx_01bytes]       ;extend to 4 bytes
    pmullw      mm0,        [mmx_01bytes]
    pshufw      mm0,    mm0,    0
    movq        [r0],   mm0

    MMX_PRED_H_8X8_ONE_LINEE    mm0, mm1, r1,r0+8

    lea         r1,[r1+r2*2]
    MMX_PRED_H_8X8_ONE_LINE mm0, mm1, r1,r0+16

    MMX_PRED_H_8X8_ONE_LINEE    mm0, mm1, r1,r0+24

    lea         r1,[r1+r2*2]
    MMX_PRED_H_8X8_ONE_LINE mm0, mm1, r1,r0+32

    MMX_PRED_H_8X8_ONE_LINEE    mm0, mm1, r1,r0+40

    lea         r1,[r1+r2*2]
    MMX_PRED_H_8X8_ONE_LINE mm0, mm1, r1,r0+48

    MMX_PRED_H_8X8_ONE_LINEE    mm0, mm1, r1,r0+56
    WELSEMMS
    ret

;***********************************************************************
;   void WelsI4x4LumaPredV_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride)
;   copy pixels from top 4 pixels
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredV_sse2
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movd        xmm0,   [r1]
    pshufd      xmm0,   xmm0,   0
    movdqa      [r0],   xmm0
    ret

;***********************************************************************
;   void WelsIChromaPredV_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride)
;   copy 8 pixels from top 8 pixels
;***********************************************************************
WELS_EXTERN WelsIChromaPredV_sse2
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub     r1,     r2
    movq        xmm0,       [r1]
    movdqa      xmm1,       xmm0
    punpcklqdq  xmm0,       xmm1
    movdqa      [r0],       xmm0
    movdqa      [r0+16],    xmm0
    movdqa      [r0+32],    xmm0
    movdqa      [r0+48],    xmm0
    ret

;***********************************************************************
;   lt|t0|t1|t2|t3|
;   l0|
;   l1|
;   l2|
;   l3|
;   t3 will never been used
;   destination:
;   |a |b |c |d |
;   |e |f |a |b |
;   |g |h |e |f |
;   |i |j |g |h |

;   a = (1 + lt + l0)>>1
;   e = (1 + l0 + l1)>>1
;   g = (1 + l1 + l2)>>1
;   i = (1 + l2 + l3)>>1

;   d = (2 + t0 + (t1<<1) + t2)>>2
;   c = (2 + lt + (t0<<1) + t1)>>2
;   b = (2 + l0 + (lt<<1) + t0)>>2

;   f = (2 + l1 + (l0<<1) + lt)>>2
;   h = (2 + l2 + (l1<<1) + l0)>>2
;   j = (2 + l3 + (l2<<1) + l1)>>2
;   [b a f e h g j i] + [d c b a] --> mov to memory
;
;   void WelsI4x4LumaPredHD_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredHD_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movd        mm0, [r1-1]            ; mm0 = [xx xx xx xx t2 t1 t0 lt]
    psllq       mm0, 20h                ; mm0 = [t2 t1 t0 lt xx xx xx xx]

    movd        mm1, [r1+2*r2-4]
    punpcklbw   mm1, [r1+r2-4]        ; mm1[7] = l0, mm1[6] = l1
    lea         r1, [r1+2*r2]
    movd        mm2, [r1+2*r2-4]
    punpcklbw   mm2, [r1+r2-4]        ; mm2[7] = l2, mm2[6] = l3
    punpckhwd   mm2, mm1                ; mm2 = [l0 l1 l2 l3 xx xx xx xx]
    psrlq       mm2, 20h
    pxor        mm0, mm2                ; mm0 = [t2 t1 t0 lt l0 l1 l2 l3]

    movq        mm1, mm0
    psrlq       mm1, 10h                ; mm1 = [xx xx t2 t1 t0 lt l0 l1]
    movq        mm2, mm0
    psrlq       mm2, 8h                 ; mm2 = [xx t2 t1 t0 lt l0 l1 l2]
    movq        mm3, mm2
    movq        mm4, mm1
    pavgb       mm1, mm0

    pxor        mm4, mm0                ; find odd value in the lowest bit of each byte
    pand        mm4, [mmx_01bytes]      ; set the odd bit
    psubusb     mm1, mm4                ; decrease 1 from odd bytes

    pavgb       mm2, mm1                ; mm2 = [xx xx d  c  b  f  h  j]

    movq        mm4, mm0
    pavgb       mm3, mm4                ; mm3 = [xx xx xx xx a  e  g  i]
    punpcklbw   mm3, mm2                ; mm3 = [b  a  f  e  h  g  j  i]

    psrlq       mm2, 20h
    psllq       mm2, 30h                ; mm2 = [d  c  0  0  0  0  0  0]
    movq        mm4, mm3
    psrlq       mm4, 10h                ; mm4 = [0  0  b  a  f  e  h  j]
    pxor        mm2, mm4                ; mm2 = [d  c  b  a  xx xx xx xx]
    psrlq       mm2, 20h                ; mm2 = [xx xx xx xx  d  c  b  a]

    movd        [r0], mm2
    movd        [r0+12], mm3
    psrlq       mm3, 10h
    movd        [r0+8], mm3
    psrlq       mm3, 10h
    movd        [r0+4], mm3
    WELSEMMS
    ret

;***********************************************************************
;   lt|t0|t1|t2|t3|
;   l0|
;   l1|
;   l2|
;   l3|
;   t3 will never been used
;   destination:
;   |a |b |c |d |
;   |c |d |e |f |
;   |e |f |g |g |
;   |g |g |g |g |

;   a = (1 + l0 + l1)>>1
;   c = (1 + l1 + l2)>>1
;   e = (1 + l2 + l3)>>1
;   g = l3

;   b = (2 + l0 + (l1<<1) + l2)>>2
;   d = (2 + l1 + (l2<<1) + l3)>>2
;   f = (2 + l2 + (l3<<1) + l3)>>2

;   [g g f e d c b a] + [g g g g] --> mov to memory
;
;   void WelsI4x4LumaPredHU_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredHU_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    movd        mm0, [r1-4]            ; mm0[3] = l0
    punpcklbw   mm0, [r1+r2-4]        ; mm0[7] = l1, mm0[6] = l0
    lea         r1, [r1+2*r2]
    movd        mm2, [r1-4]            ; mm2[3] = l2
    movd        mm4, [r1+r2-4]        ; mm4[3] = l3
    punpcklbw   mm2, mm4
    punpckhwd   mm0, mm2                ; mm0 = [l3 l2 l1 l0 xx xx xx xx]

    psrlq       mm4, 18h
    psllq       mm4, 38h                ; mm4 = [l3 xx xx xx xx xx xx xx]
    psrlq       mm0, 8h
    pxor        mm0, mm4                ; mm0 = [l3 l3 l2 l1 l0 xx xx xx]

    movq        mm1, mm0
    psllq       mm1, 8h                 ; mm1 = [l3 l2 l1 l0 xx xx xx xx]
    movq        mm3, mm1                ; mm3 = [l3 l2 l1 l0 xx xx xx xx]
    pavgb       mm1, mm0                ; mm1 = [g  e  c  a  xx xx xx xx]

    movq        mm2, mm0
    psllq       mm2, 10h                ; mm2 = [l2 l1 l0 xx xx xx xx xx]
    movq        mm5, mm2
    pavgb       mm2, mm0

    pxor        mm5, mm0                ; find odd value in the lowest bit of each byte
    pand        mm5, [mmx_01bytes]      ; set the odd bit
    psubusb     mm2, mm5                ; decrease 1 from odd bytes

    pavgb       mm2, mm3                ; mm2 = [f  d  b  xx xx xx xx xx]

    psrlq       mm2, 8h
    pxor        mm2, mm4                ; mm2 = [g  f  d  b  xx xx xx xx]

    punpckhbw   mm1, mm2                ; mm1 = [g  g  f  e  d  c  b  a]
    punpckhbw   mm4, mm4                ; mm4 = [g  g  xx xx xx xx xx xx]
    punpckhbw   mm4, mm4                ; mm4 = [g  g  g  g  xx xx xx xx]

    psrlq       mm4, 20h
    movd        [r0+12], mm4

    movd        [r0], mm1
    psrlq       mm1, 10h
    movd        [r0+4], mm1
    psrlq       mm1, 10h
    movd        [r0+8], mm1
    WELSEMMS
    ret



;***********************************************************************
;   lt|t0|t1|t2|t3|
;   l0|
;   l1|
;   l2|
;   l3|
;   l3 will never been used
;   destination:
;   |a |b |c |d |
;   |e |f |g |h |
;   |i |a |b |c |
;   |j |e |f |g |

;   a = (1 + lt + t0)>>1
;   b = (1 + t0 + t1)>>1
;   c = (1 + t1 + t2)>>1
;   d = (1 + t2 + t3)>>1

;   e = (2 + l0 + (lt<<1) + t0)>>2
;   f = (2 + lt + (t0<<1) + t1)>>2
;   g = (2 + t0 + (t1<<1) + t2)>>2

;   h = (2 + t1 + (t2<<1) + t3)>>2
;   i = (2 + lt + (l0<<1) + l1)>>2
;   j = (2 + l0 + (l1<<1) + l2)>>2
;
;   void WelsI4x4LumaPredVR_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredVR_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movq        mm0, [r1-1]            ; mm0 = [xx xx xx t3 t2 t1 t0 lt]
    psllq       mm0, 18h                ; mm0 = [t3 t2 t1 t0 lt xx xx xx]

    movd        mm1, [r1+2*r2-4]
    punpcklbw   mm1, [r1+r2-4]        ; mm1[7] = l0, mm1[6] = l1
    lea         r1, [r1+2*r2]
    movq        mm2, [r1+r2-8]        ; mm2[7] = l2
    punpckhwd   mm2, mm1                ; mm2 = [l0 l1 l2 xx xx xx xx xx]
    psrlq       mm2, 28h
    pxor        mm0, mm2                ; mm0 = [t3 t2 t1 t0 lt l0 l1 l2]

    movq        mm1, mm0
    psllq       mm1, 8h                 ; mm1 = [t2 t1 t0 lt l0 l1 l2 xx]
    pavgb       mm1, mm0                ; mm1 = [d  c  b  a  xx xx xx xx]

    movq        mm2, mm0
    psllq       mm2, 10h                ; mm2 = [t1 t0 lt l0 l1 l2 xx xx]
    movq        mm3, mm2
    pavgb       mm2, mm0

    pxor        mm3, mm0                ; find odd value in the lowest bit of each byte
    pand        mm3, [mmx_01bytes]      ; set the odd bit
    psubusb     mm2, mm3                ; decrease 1 from odd bytes

    movq        mm3, mm0
    psllq       mm3, 8h                 ; mm3 = [t2 t1 t0 lt l0 l1 l2 xx]
    pavgb       mm3, mm2                ; mm3 = [h  g  f  e  i  j  xx xx]
    movq        mm2, mm3

    psrlq       mm1, 20h                ; mm1 = [xx xx xx xx d  c  b  a]
    movd        [r0], mm1

    psrlq       mm2, 20h                ; mm2 = [xx xx xx xx h  g  f  e]
    movd        [r0+4], mm2

    movq        mm4, mm3
    psllq       mm4, 20h
    psrlq       mm4, 38h                ; mm4 = [xx xx xx xx xx xx xx i]

    movq        mm5, mm3
    psllq       mm5, 28h
    psrlq       mm5, 38h                ; mm5 = [xx xx xx xx xx xx xx j]

    psllq       mm1, 8h
    pxor        mm4, mm1                ; mm4 = [xx xx xx xx c  b  a  i]
    movd        [r0+8], mm4

    psllq       mm2, 8h
    pxor        mm5, mm2                ; mm5 = [xx xx xx xx g  f  e  j]
    movd        [r0+12], mm5
    WELSEMMS
    ret

;***********************************************************************
;   lt|t0|t1|t2|t3|t4|t5|t6|t7
;   l0|
;   l1|
;   l2|
;   l3|
;   lt,t0,t1,t2,t3 will never been used
;   destination:
;   |a |b |c |d |
;   |b |c |d |e |
;   |c |d |e |f |
;   |d |e |f |g |

;   a = (2 + t0 + t2 + (t1<<1))>>2
;   b = (2 + t1 + t3 + (t2<<1))>>2
;   c = (2 + t2 + t4 + (t3<<1))>>2
;   d = (2 + t3 + t5 + (t4<<1))>>2

;   e = (2 + t4 + t6 + (t5<<1))>>2
;   f = (2 + t5 + t7 + (t6<<1))>>2
;   g = (2 + t6 + t7 + (t7<<1))>>2

;   [g f e d c b a] --> mov to memory
;
;   void WelsI4x4LumaPredDDL_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredDDL_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movq        mm0, [r1]              ; mm0 = [t7 t6 t5 t4 t3 t2 t1 t0]
    movq        mm1, mm0
    movq        mm2, mm0

    movq        mm3, mm0
    psrlq       mm3, 38h
    psllq       mm3, 38h                ; mm3 = [t7 xx xx xx xx xx xx xx]

    psllq       mm1, 8h                 ; mm1 = [t6 t5 t4 t3 t2 t1 t0 xx]
    psrlq       mm2, 8h
    pxor        mm2, mm3                ; mm2 = [t7 t7 t6 t5 t4 t3 t2 t1]

    movq        mm3, mm1
    pavgb       mm1, mm2
    pxor        mm3, mm2                ; find odd value in the lowest bit of each byte
    pand        mm3, [mmx_01bytes]      ; set the odd bit
    psubusb     mm1, mm3                ; decrease 1 from odd bytes

    pavgb       mm0, mm1                ; mm0 = [g f e d c b a xx]

    psrlq       mm0, 8h
    movd        [r0], mm0
    psrlq       mm0, 8h
    movd        [r0+4], mm0
    psrlq       mm0, 8h
    movd        [r0+8], mm0
    psrlq       mm0, 8h
    movd        [r0+12], mm0
    WELSEMMS
    ret


;***********************************************************************
;   lt|t0|t1|t2|t3|t4|t5|t6|t7
;   l0|
;   l1|
;   l2|
;   l3|
;   lt,t0,t1,t2,t3 will never been used
;   destination:
;   |a |b |c |d |
;   |e |f |g |h |
;   |b |c |d |i |
;   |f |g |h |j |

;   a = (1 + t0 + t1)>>1
;   b = (1 + t1 + t2)>>1
;   c = (1 + t2 + t3)>>1
;   d = (1 + t3 + t4)>>1
;   i = (1 + t4 + t5)>>1

;   e = (2 + t0 + (t1<<1) + t2)>>2
;   f = (2 + t1 + (t2<<1) + t3)>>2
;   g = (2 + t2 + (t3<<1) + t4)>>2
;   h = (2 + t3 + (t4<<1) + t5)>>2
;   j = (2 + t4 + (t5<<1) + t6)>>2

;   [i d c b a] + [j h g f e] --> mov to memory
;
;   void WelsI4x4LumaPredVL_mmx(uint8_t *pred,uint8_t *pRef,int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI4x4LumaPredVL_mmx
    %assign push_num 0
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movq        mm0, [r1]              ; mm0 = [t7 t6 t5 t4 t3 t2 t1 t0]
    movq        mm1, mm0
    movq        mm2, mm0

    psrlq       mm1, 8h                 ; mm1 = [xx t7 t6 t5 t4 t3 t2 t1]
    psrlq       mm2, 10h                ; mm2 = [xx xx t7 t6 t5 t4 t3 t2]

    movq        mm3, mm1
    pavgb       mm3, mm0                ; mm3 = [xx xx xx i  d  c  b  a]

    movq        mm4, mm2
    pavgb       mm2, mm0
    pxor        mm4, mm0                ; find odd value in the lowest bit of each byte
    pand        mm4, [mmx_01bytes]      ; set the odd bit
    psubusb     mm2, mm4                ; decrease 1 from odd bytes

    pavgb       mm2, mm1                ; mm2 = [xx xx xx j  h  g  f  e]

    movd        [r0], mm3
    psrlq       mm3, 8h
    movd        [r0+8], mm3

    movd        [r0+4], mm2
    psrlq       mm2, 8h
    movd        [r0+12], mm2
    WELSEMMS
    ret

;***********************************************************************
;
;   void WelsIChromaPredDc_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride)
;***********************************************************************
WELS_EXTERN WelsIChromaPredDc_sse2
    push r3
    push r4
    %assign push_num 2
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movq        mm0, [r1]

    movzx       r3, byte [r1+r2-0x01] ; l1
    lea             r1, [r1+2*r2]
    movzx       r4, byte [r1-0x01]     ; l2
    add     r3, r4
    movzx       r4, byte [r1+r2-0x01] ; l3
    add     r3, r4
    lea             r1, [r1+2*r2]
    movzx       r4, byte [r1-0x01]     ; l4
    add     r3, r4
    movd            mm1, r3d                 ; mm1 = l1+l2+l3+l4

    movzx       r3, byte [r1+r2-0x01] ; l5
    lea             r1, [r1+2*r2]
    movzx       r4, byte [r1-0x01]     ; l6
    add     r3, r4
    movzx       r4, byte [r1+r2-0x01] ; l7
    add     r3, r4
    lea             r1, [r1+2*r2]
    movzx       r4, byte [r1-0x01]     ; l8
    add     r3, r4
    movd            mm2, r3d                 ; mm2 = l5+l6+l7+l8

    movq        mm3, mm0
    psrlq       mm0, 0x20
    psllq       mm3, 0x20
    psrlq       mm3, 0x20
    pxor        mm4, mm4
    psadbw      mm0, mm4
    psadbw      mm3, mm4                 ; sum1 = mm3+mm1, sum2 = mm0, sum3 = mm2

    paddq       mm3, mm1
    movq        mm1, mm2
    paddq       mm1, mm0;                ; sum1 = mm3, sum2 = mm0, sum3 = mm2, sum4 = mm1

    movq        mm4, [mmx_0x02]

    paddq       mm0, mm4
    psrlq       mm0, 0x02

    paddq       mm2, mm4
    psrlq       mm2, 0x02

    paddq       mm3, mm4
    paddq       mm3, mm4
    psrlq       mm3, 0x03

    paddq       mm1, mm4
    paddq       mm1, mm4
    psrlq       mm1, 0x03

    pmuludq     mm0, [mmx_01bytes]
    pmuludq     mm3, [mmx_01bytes]
    psllq       mm0, 0x20
    pxor        mm0, mm3                 ; mm0 = m_up

    pmuludq     mm2, [mmx_01bytes]
    pmuludq     mm1, [mmx_01bytes]
    psllq       mm1, 0x20
    pxor        mm1, mm2                 ; mm2 = m_down

    movq        [r0], mm0
    movq        [r0+0x08], mm0
    movq        [r0+0x10], mm0
    movq        [r0+0x18], mm0

    movq        [r0+0x20], mm1
    movq        [r0+0x28], mm1
    movq        [r0+0x30], mm1
    movq        [r0+0x38], mm1

    pop r4
    pop r3
    WELSEMMS
    ret



;***********************************************************************
;
;   void WelsI16x16LumaPredDc_sse2(uint8_t *pred, uint8_t *pRef, int32_t stride)
;***********************************************************************
WELS_EXTERN WelsI16x16LumaPredDc_sse2
    push r3
    push r4
    %assign push_num 2
    LOAD_3_PARA
    SIGN_EXTENSION r2, r2d
    sub         r1, r2
    movdqa      xmm0, [r1]             ; read one row
    pxor        xmm1, xmm1
    psadbw      xmm0, xmm1
    movdqa      xmm1, xmm0
    psrldq      xmm1, 0x08
    pslldq      xmm0, 0x08
    psrldq      xmm0, 0x08
    paddw       xmm0, xmm1

    movzx       r3, byte [r1+r2-0x01]
    movzx       r4, byte [r1+2*r2-0x01]
    add     r3, r4
    lea         r1, [r1+r2]
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    LOAD_2_LEFT_AND_ADD
    add         r3, 0x10
    movd        xmm1, r3d
    paddw       xmm0, xmm1
    psrld       xmm0, 0x05
    pmuludq     xmm0, [mmx_01bytes]
    pshufd      xmm0, xmm0, 0

    movdqa      [r0], xmm0
    movdqa      [r0+0x10], xmm0
    movdqa      [r0+0x20], xmm0
    movdqa      [r0+0x30], xmm0
    movdqa      [r0+0x40], xmm0
    movdqa      [r0+0x50], xmm0
    movdqa      [r0+0x60], xmm0
    movdqa      [r0+0x70], xmm0
    movdqa      [r0+0x80], xmm0
    movdqa      [r0+0x90], xmm0
    movdqa      [r0+0xa0], xmm0
    movdqa      [r0+0xb0], xmm0
    movdqa      [r0+0xc0], xmm0
    movdqa      [r0+0xd0], xmm0
    movdqa      [r0+0xe0], xmm0
    movdqa      [r0+0xf0], xmm0

    pop r4
    pop r3
    ret