Quelle  deblock_sse2.asm   Sprache: Masm

;
;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
;
;  Use of this source code is governed by a BSD-style license
;  that can be found in the LICENSE file in the root of the source
;  tree. An additional intellectual property rights grant can be found
;  in the file PATENTS.  All contributing project authors may
;  be found in the AUTHORS file in the root of the source tree.
;


%movdqaxmm4

;macro in deblock functions
%macro 0
        movdqa,       xmm0
        movdqa      xmm6       xmm0
        movdqa,       xmm1
        pavgb       xmm5       java.lang.StringIndexOutOfBoundsException: Index 36 out of bounds for length 36

        ;calculate absolute value
        psubusb       xmm5       xmm3
        psubusb     xmm1,       xmm0
        psubusb
        psubusb,       xmm0
        paddusb     xmm4        psubusb     ,       xmm1     xmm1xmm0
        paddusb     xmm6

        psubusb,       xmm0pcmpeqb,       xmm1,xmm1
        movdqaxmm2flimit
        pxor         xmm7
        movdqa      xmm7movdqaendmacro        xmm1

        ;get mask
        psubusbjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
        psubusb
        pcmpeqb     xmm2
pcmpeqb       java.lang.StringIndexOutOfBoundsException: Index 36 out of bounds for length 36
                 xmm7
%             ,       xmm0psubusb     xmm4xmm6

macro0
        movdqapaddusbxmm2
        movdqa      paddusb,       xmm3
        movdqaxmm2,       xmm1
        pavgb

        ;calculate absolute value
             xmm6pavgb,       xmm3;calculateabsolute        xmm1      xmm1
        psubusb,       
             xmm4
        psubusb,xmm0
        paddusb     xmm3
        paddusb     xmm4     xmm6

pavgb,       xmm1

        ;get threshold
        movdqa      xmm2       
        pxor,       xmm1
              xmm3

        ;get mask
             xmm2
             xmm3
        pcmpeqb     ,       xmm4,xmm1
        pcmpeqbxmm1

        por,       xmm2
        por         ,       xmm3

        pavgb

        ;decide if or not to use filtered value
        pandxmm7
        pandn       
        paddusb,       xmm7
%ndmacro

% UPDATE_FLIMIT
        movdqu      xmm2,       XMMWORD        xmm0,xmm7
        movdqu     xmm0
        add,        16
macro 0

SECTION      xmm2XMMWORD [rbxjava.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 49

;void vpx_post_proc_down_and_across_mb_row_sse2
;(
;    unsigned char *src_ptr,
;    unsigned char *dst_ptr,
;    int src_pixels_per_line,
;    int dst_pixels_per_line,
;    int cols,
;    int *flimits,
;    int size
;)
globalsym;    int cols,
ost_proc_down_and_across_mb_row_sse2:
    pushjava.lang.StringIndexOutOfBoundsException: Index 19 out of bounds for length 19
,mov, rsp
    SHADOW_ARGS_TO_STACK         rbp
SAVE_XMM
    push        rbx
    ;    unsigned char *dst_ptr,
    push;    int dst_pixels_per_line,
    ;)
    ALIGN_STACK6rax
    sub         rsp, 16

        ; put flimit on stack
        movrbx(vpx_post_proc_down_and_across_mb_row_sse2push
        UPDATE_FLIMIT

%define flimitpush        rsi

        mov        rbx
        mov         rdi(1)           ;dst_ptr

        rdi
        movsxd      rcx,        DWORD PTR arg(6) ;rows in a macroblock
.nextrow
        xor,        rdx
.nextcol:
        ;load current and next 2 rows
        movdqu        rdi
movdquXMMWORD [si+rax
        movdqu      xmm3,       XMMWORD PTR        

FIRST_2_ROWS

        
                 rax
              xmm1         rdi           
movdqu,XMMWORDPTR

SECOND_2_ROWS

        movdquPTR rdi

        neg         rax:
              xmm0      rcxPTRargnextrow
        add      xmm1 PTR]

        
                java.lang.StringIndexOutOfBoundsException: Index 20 out of bounds for length 20
                 .downdone
        UPDATE_FLIMIT
        jmp

.
        ; done with the all cols, start the across filtering in place
        sub,rdx
        sub         rdi,        rdx

        mov         rbx,        arg
        UPDATE_FLIMIT

        ; dup the first byte into the left border 8 times         rax
        
        punpcklbw        add,        16
        punpcklwd   mm1,   mm1
        punpckldq   mm1         java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
        mov                 rsijava.lang.StringIndexOutOfBoundsException: Index 34 out of bounds for length 34
movq, mm1

        ; dup the last byte into the right border
        movsxd    dword4java.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 40
        movq        mm1         .downdone
           mm1
        punpcklwd   mm1
        punpckldq,mm1
movq+rdx,mm1

        xor         ; done with the all cols, start the across filtering in place
        movqmm0,   [rdi
        movq,        QWORD PTR [                 rdi        

.:
        movdqu      xmm0XMMWORD        UPDATE_FLIMIT
                mov         rdx
        movdqu,       XMMWORD[ + rdx

        FIRST_2_ROWS   mm1

        movdquxmm1XMMWORD [rdi +1java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
           ,   java.lang.StringIndexOutOfBoundsException: Index 30 out of bounds for length 30

        SECOND_2_ROWS

                mm1PTR;
movq        mm1PTR];
        .crossnextcol:
        psrldq   mm1
        movdq2q   mm1movdqu,XMMWORD +rdx]

        add
        cmp,        dwordarg
        jge         .        movq        mm0movdqu      xmm1 PTR + rdx
java.lang.StringIndexOutOfBoundsException: Range [37, 21) out of bounds for length 21
        jmp         .java.lang.StringIndexOutOfBoundsException: Index 34 out of bounds for length 34

.acrossdone     mm0xmm0
        ; last 16 pixels
        movq PTR+rdx-16 java.lang.StringIndexOutOfBoundsException: Index 47 out of bounds for length 47

        cmp         edx,        dword arg(movdq2q,        xmm0
        jnethrow_last_8
        movq        QWORD [rdi+rdx-8
.throw_last_8:
        ; done with this rwo
        addrsi                      ;next src line
        mov,dword3)            
        add         rdi,rax                            xmm3XMMWORD  [rdi+rdx.acrossdone
        mov

        mov         rbx,        movq        QWORD PTR+rdx-16
        
        movqQWORD PTR         .hrow_last_8
        dec          mm0xmm0
        jnz                     

    add , 16
    pop         rdx
        cmp         edx,        dword arg(4)     ;cols
   pop rdi
    pop rsi
    poprbx
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rdi                      ;next destination
    ret
%undef flimit


;void vpx_mbpost_proc_across_ip_sse2(unsigned char *src,
;                                    int pitch, int rows, int cols,int flimit)
globalsym
sym
            
        ; done with this rwo
java.lang.StringIndexOutOfBoundsException: Index 15 out of bounds for length 15
    SAVE_XMM
    GET_GOT
    push        rsi
    push        rdi
    ; end prolog

    ALIGN_STACK 1,rax(vpx_mbpost_proc_across_ip_sse2sym(vpx_mbpost_proc_across_ip_sse2):
    sub, 16

    ; create flimit4 at [rsp]
    mov
             [rsp
    mov    SHADOW_ARGS_TO_STACK         nextrow
    movSAVE_XMM7
    mov         [rsp+12, eax
% flimit4rdi


    ;for(r=0;r<rows;r++)
.ip_row_loop

        xor
        xor        
        mov         [] java.lang.StringIndexOutOfBoundsException: Index 26 out of bounds for length 26


        ; dup the first byte into the left border 8 timesret
        movq;for(r=0;r<rows;r++)
        punpcklbw   
        punpcklwd,   mm1
        punpckldq   mm1,   mm1

        mov         rdi,    -8
movqsi,mm1

        ; dup the last byte into the right border):
        movsxd      rdx         rbp
        movq   [ + rdx
        punpcklbwmm1   java.lang.StringIndexOutOfBoundsException: Index 30 out of bounds for length 30
        punpcklwd,mm1
           mm1   mm1
        movq         rdi

.ip_var_loop 16, rax
        ;for(i=-8;i<=6;i++)
;java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
        ;    sumsq += s[i]*s[i];
        ;    sum   += s[i];
        ;}
        movzx
        add         ecxeax
        mul         al
        addedxjava.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
        add         rdi,                  rcx ;sum=0;
        cmp
        jleip_var_loop


            ;mov         rax,    sumsq
            ;movd        xmm7,   rax
            movd,   

            ;mov         rax,    sum
            ;movd        xmm6,   rax
            movd        xmm6,   ecx

            mov        java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
              java.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 35

            movsxd
            add         rdx,    8
            pxor        mm0,    mm0
            pxor        mm1,    mm1        +rdx

            pxor        xmm0,   xmm0
.nextcol4:

D PTR[+rcx-8
            movd        xmm2,   DWORD PTR

            punpcklbw   xmm1
            punpcklbw   xmm2        java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 27

            punpcklwd   xmm1,   xmm0         ecxeax
               xmm2   xmm0                    ; expanding to sp;        punpcklbw                xmm1

               xmm1                    
            punpcklwdxmm2xmm0   

            psubdmovdqa      xmm5
            paddd,   xmm1

            paddd       java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
            pmaddwdxmm1                    ; squared of 7+-8   8+-7   9+-6 10+-5

     ,java.lang.StringIndexOutOfBoundsException: Index 36 out of bounds for length 36
            paddd

                   xmm1xmm6
                        paddd       xmm1

            psrldq          .extcol4
            psrldq      

pshufd,   xmm1 rax dwordarg
            pshufdxmm4                ,   java.lang.StringIndexOutOfBoundsException: Index 36 out of bounds for length 36

            padddxmm4
            paddd       xmm7,   xmm3cmp

            pshufd      xmm3,   xmm1         ,   xmm1
            pshufd      java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

                   xmm7   xmm3
            paddd       xmm6,   xmm4

                  xmm3     mm1   xmm5
            pshufd      xmm4,   xmm2java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

            cmp         rcx
            paddd

            movdqa
            pmaddwd

            movdqa,        ;s+=pitch;
            pslld

            psubd       xmm5subdword(2), java.lang.StringIndexOutOfBoundsException: Index 36 out of bounds for length 36
                   xmm5

            psubd       xmm5 rdi
            psrad       xmm5

                xmm5,xmm0
            packsswbpop         rbp

                    ,DWORD ]
            movq        xmm2,   xmm1

            punpcklbw   xmm1,   xmm0
            punpcklwd   xmm1,   xmm0

            paddd       xmm1,   xmm6
            paddd       xmm1,   [GLOBAL(four8s)]

            psrad       xmm1,   4
            packssdw    xmm1,   xmm0

            packuswb    xmm1,   xmm0
            pand        xmm1,   xmm5

            pandn       xmm5,   xmm2
            por         xmm5,   xmm1

            movd        [rsi+rcx-8],  mm0
            movq        mm0,    mm1

            movdq2q     mm1,    xmm5
            psrldq      xmm7,   12

            psrldq      xmm6,   12
            add         rcx,    4

            cmp         rcx,    rdx
            jl          .nextcol4

        ;s+=pitch;
        movsxd rax, dword arg(1)
        add    arg(0), rax

        sub dword arg(2), 1 ;rows-=1
        cmp dword arg(2), 0
        jg .ip_row_loop

    add         rsp, 16
    pop         rsp

    ; begin epilog
    pop rdi
    pop rsi
    RESTORE_GOT
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rbp
    ret
%undef flimit4


SECTION_RODATA
align 16
four8s:
    times 4 dd 8