michael@0: ;
michael@0: ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0: ;
michael@0: ;  Use of this source code is governed by a BSD-style license
michael@0: ;  that can be found in the LICENSE file in the root of the source
michael@0: ;  tree. An additional intellectual property rights grant can be found
michael@0: ;  in the file PATENTS.  All contributing project authors may
michael@0: ;  be found in the AUTHORS file in the root of the source tree.
michael@0: ;
michael@0: 
michael@0: 
michael@0: %include "vpx_ports/x86_abi_support.asm"
michael@0: 
michael@0: %define VP8_FILTER_WEIGHT 128
michael@0: %define VP8_FILTER_SHIFT  7
michael@0: 
michael@0: ;void vp8_mbpost_proc_down_mmx(unsigned char *dst,
michael@0: ;                             int pitch, int rows, int cols,int flimit)
michael@0: extern sym(vp8_rv)
michael@0: global sym(vp8_mbpost_proc_down_mmx) PRIVATE
michael@0: sym(vp8_mbpost_proc_down_mmx):
michael@0:     push        rbp
michael@0:     mov         rbp, rsp
michael@0:     SHADOW_ARGS_TO_STACK 5
michael@0:     GET_GOT     rbx
michael@0:     push        rsi
michael@0:     push        rdi
michael@0:     ; end prolog
michael@0: 
michael@0:     ALIGN_STACK 16, rax
michael@0:     sub         rsp, 136
michael@0: 
michael@0:     ; unsigned char d[16][8] at [rsp]
michael@0:     ; create flimit2 at [rsp+128]
michael@0:     mov         eax, dword ptr arg(4) ;flimit
michael@0:     mov         [rsp+128], eax
michael@0:     mov         [rsp+128+4], eax
michael@0: %define flimit2 [rsp+128]
michael@0: 
michael@0: %if ABI_IS_32BIT=0
michael@0:     lea         r8,       [GLOBAL(sym(vp8_rv))]
michael@0: %endif
michael@0: 
michael@0:     ;rows +=8;
michael@0:     add         dword ptr arg(2), 8
michael@0: 
michael@0:     ;for(c=0; c<cols; c+=4)
michael@0: .loop_col:
michael@0:             mov         rsi,        arg(0)  ;s
michael@0:             pxor        mm0,        mm0     ;
michael@0: 
michael@0:             movsxd      rax,        dword ptr arg(1) ;pitch       ;
michael@0: 
michael@0:             ; this copies the last row down into the border 8 rows
michael@0:             mov         rdi,        rsi
michael@0:             mov         rdx,        arg(2)
michael@0:             sub         rdx,        9
michael@0:             imul        rdx,        rax
michael@0:             lea         rdi,        [rdi+rdx]
michael@0:             movq        mm1,        QWORD ptr[rdi]              ; first row
michael@0:             mov         rcx,        8
michael@0: .init_borderd                                                    ; initialize borders
michael@0:             lea         rdi,        [rdi + rax]
michael@0:             movq        [rdi],      mm1
michael@0: 
michael@0:             dec         rcx
michael@0:             jne         .init_borderd
michael@0: 
michael@0:             neg         rax                                     ; rax = -pitch
michael@0: 
michael@0:             ; this copies the first row up into the border 8 rows
michael@0:             mov         rdi,        rsi
michael@0:             movq        mm1,        QWORD ptr[rdi]              ; first row
michael@0:             mov         rcx,        8
michael@0: .init_border                                                    ; initialize borders
michael@0:             lea         rdi,        [rdi + rax]
michael@0:             movq        [rdi],      mm1
michael@0: 
michael@0:             dec         rcx
michael@0:             jne         .init_border
michael@0: 
michael@0: 
michael@0:             lea         rsi,        [rsi + rax*8];              ; rdi = s[-pitch*8]
michael@0:             neg         rax
michael@0: 
michael@0: 
michael@0:             pxor        mm5,        mm5
michael@0:             pxor        mm6,        mm6     ;
michael@0: 
michael@0:             pxor        mm7,        mm7     ;
michael@0:             mov         rdi,        rsi
michael@0: 
michael@0:             mov         rcx,        15          ;
michael@0: 
michael@0: .loop_initvar:
michael@0:             movd        mm1,        DWORD PTR [rdi];
michael@0:             punpcklbw   mm1,        mm0     ;
michael@0: 
michael@0:             paddw       mm5,        mm1     ;
michael@0:             pmullw      mm1,        mm1     ;
michael@0: 
michael@0:             movq        mm2,        mm1     ;
michael@0:             punpcklwd   mm1,        mm0     ;
michael@0: 
michael@0:             punpckhwd   mm2,        mm0     ;
michael@0:             paddd       mm6,        mm1     ;
michael@0: 
michael@0:             paddd       mm7,        mm2     ;
michael@0:             lea         rdi,        [rdi+rax]   ;
michael@0: 
michael@0:             dec         rcx
michael@0:             jne         .loop_initvar
michael@0:             ;save the var and sum
michael@0:             xor         rdx,        rdx
michael@0: .loop_row:
michael@0:             movd        mm1,        DWORD PTR [rsi]     ; [s-pitch*8]
michael@0:             movd        mm2,        DWORD PTR [rdi]     ; [s+pitch*7]
michael@0: 
michael@0:             punpcklbw   mm1,        mm0
michael@0:             punpcklbw   mm2,        mm0
michael@0: 
michael@0:             paddw       mm5,        mm2
michael@0:             psubw       mm5,        mm1
michael@0: 
michael@0:             pmullw      mm2,        mm2
michael@0:             movq        mm4,        mm2
michael@0: 
michael@0:             punpcklwd   mm2,        mm0
michael@0:             punpckhwd   mm4,        mm0
michael@0: 
michael@0:             paddd       mm6,        mm2
michael@0:             paddd       mm7,        mm4
michael@0: 
michael@0:             pmullw      mm1,        mm1
michael@0:             movq        mm2,        mm1
michael@0: 
michael@0:             punpcklwd   mm1,        mm0
michael@0:             psubd       mm6,        mm1
michael@0: 
michael@0:             punpckhwd   mm2,        mm0
michael@0:             psubd       mm7,        mm2
michael@0: 
michael@0: 
michael@0:             movq        mm3,        mm6
michael@0:             pslld       mm3,        4
michael@0: 
michael@0:             psubd       mm3,        mm6
michael@0:             movq        mm1,        mm5
michael@0: 
michael@0:             movq        mm4,        mm5
michael@0:             pmullw      mm1,        mm1
michael@0: 
michael@0:             pmulhw      mm4,        mm4
michael@0:             movq        mm2,        mm1
michael@0: 
michael@0:             punpcklwd   mm1,        mm4
michael@0:             punpckhwd   mm2,        mm4
michael@0: 
michael@0:             movq        mm4,        mm7
michael@0:             pslld       mm4,        4
michael@0: 
michael@0:             psubd       mm4,        mm7
michael@0: 
michael@0:             psubd       mm3,        mm1
michael@0:             psubd       mm4,        mm2
michael@0: 
michael@0:             psubd       mm3,        flimit2
michael@0:             psubd       mm4,        flimit2
michael@0: 
michael@0:             psrad       mm3,        31
michael@0:             psrad       mm4,        31
michael@0: 
michael@0:             packssdw    mm3,        mm4
michael@0:             packsswb    mm3,        mm0
michael@0: 
michael@0:             movd        mm1,        DWORD PTR [rsi+rax*8]
michael@0: 
michael@0:             movq        mm2,        mm1
michael@0:             punpcklbw   mm1,        mm0
michael@0: 
michael@0:             paddw       mm1,        mm5
michael@0:             mov         rcx,        rdx
michael@0: 
michael@0:             and         rcx,        127
michael@0: %if ABI_IS_32BIT=1 && CONFIG_PIC=1
michael@0:             push        rax
michael@0:             lea         rax,        [GLOBAL(sym(vp8_rv))]
michael@0:             movq        mm4,        [rax + rcx*2] ;vp8_rv[rcx*2]
michael@0:             pop         rax
michael@0: %elif ABI_IS_32BIT=0
michael@0:             movq        mm4,        [r8 + rcx*2] ;vp8_rv[rcx*2]
michael@0: %else
michael@0:             movq        mm4,        [sym(vp8_rv) + rcx*2]
michael@0: %endif
michael@0:             paddw       mm1,        mm4
michael@0:             psraw       mm1,        4
michael@0: 
michael@0:             packuswb    mm1,        mm0
michael@0:             pand        mm1,        mm3
michael@0: 
michael@0:             pandn       mm3,        mm2
michael@0:             por         mm1,        mm3
michael@0: 
michael@0:             and         rcx,        15
michael@0:             movd        DWORD PTR   [rsp+rcx*4], mm1 ;d[rcx*4]
michael@0: 
michael@0:             mov         rcx,        rdx
michael@0:             sub         rcx,        8
michael@0: 
michael@0:             and         rcx,        15
michael@0:             movd        mm1,        DWORD PTR [rsp+rcx*4] ;d[rcx*4]
michael@0: 
michael@0:             movd        [rsi],      mm1
michael@0:             lea         rsi,        [rsi+rax]
michael@0: 
michael@0:             lea         rdi,        [rdi+rax]
michael@0:             add         rdx,        1
michael@0: 
michael@0:             cmp         edx,        dword arg(2) ;rows
michael@0:             jl          .loop_row
michael@0: 
michael@0: 
michael@0:         add         dword arg(0), 4 ; s += 4
michael@0:         sub         dword arg(3), 4 ; cols -= 4
michael@0:         cmp         dword arg(3), 0
michael@0:         jg          .loop_col
michael@0: 
michael@0:     add         rsp, 136
michael@0:     pop         rsp
michael@0: 
michael@0:     ; begin epilog
michael@0:     pop rdi
michael@0:     pop rsi
michael@0:     RESTORE_GOT
michael@0:     UNSHADOW_ARGS
michael@0:     pop         rbp
michael@0:     ret
michael@0: %undef flimit2
michael@0: 
michael@0: 
michael@0: ;void vp8_plane_add_noise_mmx (unsigned char *Start, unsigned char *noise,
michael@0: ;                            unsigned char blackclamp[16],
michael@0: ;                            unsigned char whiteclamp[16],
michael@0: ;                            unsigned char bothclamp[16],
michael@0: ;                            unsigned int Width, unsigned int Height, int Pitch)
michael@0: extern sym(rand)
michael@0: global sym(vp8_plane_add_noise_mmx) PRIVATE
michael@0: sym(vp8_plane_add_noise_mmx):
michael@0:     push        rbp
michael@0:     mov         rbp, rsp
michael@0:     SHADOW_ARGS_TO_STACK 8
michael@0:     GET_GOT     rbx
michael@0:     push        rsi
michael@0:     push        rdi
michael@0:     ; end prolog
michael@0: 
michael@0: .addnoise_loop:
michael@0:     call sym(rand) WRT_PLT
michael@0:     mov     rcx, arg(1) ;noise
michael@0:     and     rax, 0xff
michael@0:     add     rcx, rax
michael@0: 
michael@0:     ; we rely on the fact that the clamping vectors are stored contiguously
michael@0:     ; in black/white/both order. Note that we have to reload this here because
michael@0:     ; rdx could be trashed by rand()
michael@0:     mov     rdx, arg(2) ; blackclamp
michael@0: 
michael@0: 
michael@0:             mov     rdi, rcx
michael@0:             movsxd  rcx, dword arg(5) ;[Width]
michael@0:             mov     rsi, arg(0) ;Pos
michael@0:             xor         rax,rax
michael@0: 
michael@0: .addnoise_nextset:
michael@0:             movq        mm1,[rsi+rax]         ; get the source
michael@0: 
michael@0:             psubusb     mm1, [rdx]    ;blackclamp        ; clamp both sides so we don't outrange adding noise
michael@0:             paddusb     mm1, [rdx+32] ;bothclamp
michael@0:             psubusb     mm1, [rdx+16] ;whiteclamp
michael@0: 
michael@0:             movq        mm2,[rdi+rax]         ; get the noise for this line
michael@0:             paddb       mm1,mm2              ; add it in
michael@0:             movq        [rsi+rax],mm1         ; store the result
michael@0: 
michael@0:             add         rax,8                 ; move to the next line
michael@0: 
michael@0:             cmp         rax, rcx
michael@0:             jl          .addnoise_nextset
michael@0: 
michael@0:     movsxd  rax, dword arg(7) ; Pitch
michael@0:     add     arg(0), rax ; Start += Pitch
michael@0:     sub     dword arg(6), 1   ; Height -= 1
michael@0:     jg      .addnoise_loop
michael@0: 
michael@0:     ; begin epilog
michael@0:     pop rdi
michael@0:     pop rsi
michael@0:     RESTORE_GOT
michael@0:     UNSHADOW_ARGS
michael@0:     pop         rbp
michael@0:     ret
michael@0: 
michael@0: 
michael@0: SECTION_RODATA
michael@0: align 16
michael@0: Blur:
michael@0:     times 16 dw 16
michael@0:     times  8 dw 64
michael@0:     times 16 dw 16
michael@0:     times  8 dw  0
michael@0: 
michael@0: rd:
michael@0:     times 4 dw 0x40