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 VP9_FILTER_WEIGHT 128
michael@0: %define VP9_FILTER_SHIFT  7
michael@0: 
michael@0: ;void vp9_post_proc_down_and_across_mmx
michael@0: ;(
michael@0: ;    unsigned char *src_ptr,
michael@0: ;    unsigned char *dst_ptr,
michael@0: ;    int src_pixels_per_line,
michael@0: ;    int dst_pixels_per_line,
michael@0: ;    int rows,
michael@0: ;    int cols,
michael@0: ;    int flimit
michael@0: ;)
michael@0: global sym(vp9_post_proc_down_and_across_mmx) PRIVATE
michael@0: sym(vp9_post_proc_down_and_across_mmx):
michael@0:     push        rbp
michael@0:     mov         rbp, rsp
michael@0:     SHADOW_ARGS_TO_STACK 7
michael@0:     GET_GOT     rbx
michael@0:     push        rsi
michael@0:     push        rdi
michael@0:     ; end prolog
michael@0: 
michael@0: %if ABI_IS_32BIT=1 && CONFIG_PIC=1
michael@0:     ; move the global rd onto the stack, since we don't have enough registers
michael@0:     ; to do PIC addressing
michael@0:     movq        mm0, [GLOBAL(rd)]
michael@0:     sub         rsp, 8
michael@0:     movq        [rsp], mm0
michael@0: %define RD [rsp]
michael@0: %else
michael@0: %define RD [GLOBAL(rd)]
michael@0: %endif
michael@0: 
michael@0:         push        rbx
michael@0:         lea         rbx, [GLOBAL(Blur)]
michael@0:         movd        mm2, dword ptr arg(6) ;flimit
michael@0:         punpcklwd   mm2, mm2
michael@0:         punpckldq   mm2, mm2
michael@0: 
michael@0:         mov         rsi,        arg(0) ;src_ptr
michael@0:         mov         rdi,        arg(1) ;dst_ptr
michael@0: 
michael@0:         movsxd      rcx, DWORD PTR arg(4) ;rows
michael@0:         movsxd      rax, DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
michael@0:         pxor        mm0, mm0              ; mm0 = 00000000
michael@0: 
michael@0: .nextrow:
michael@0: 
michael@0:         xor         rdx,        rdx       ; clear out rdx for use as loop counter
michael@0: .nextcol:
michael@0: 
michael@0:         pxor        mm7, mm7              ; mm7 = 00000000
michael@0:         movq        mm6, [rbx + 32 ]      ; mm6 = kernel 2 taps
michael@0:         movq        mm3, [rsi]            ; mm4 = r0 p0..p7
michael@0:         punpcklbw   mm3, mm0              ; mm3 = p0..p3
michael@0:         movq        mm1, mm3              ; mm1 = p0..p3
michael@0:         pmullw      mm3, mm6              ; mm3 *= kernel 2 modifiers
michael@0: 
michael@0:         movq        mm6, [rbx + 48]       ; mm6 = kernel 3 taps
michael@0:         movq        mm5, [rsi + rax]      ; mm4 = r1 p0..p7
michael@0:         punpcklbw   mm5, mm0              ; mm5 = r1 p0..p3
michael@0:         pmullw      mm6, mm5              ; mm6 *= p0..p3 * kernel 3 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm6
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm7, mm1              ; mm7 = r0 p0..p3
michael@0:         psubusw     mm7, mm5              ; mm7 = r0 p0..p3 - r1 p0..p3
michael@0:         psubusw     mm5, mm1              ; mm5 = r1 p0..p3 - r0 p0..p3
michael@0:         paddusw     mm7, mm5              ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
michael@0:         pcmpgtw     mm7, mm2
michael@0: 
michael@0:         movq        mm6, [rbx + 64 ]      ; mm6 = kernel 4 modifiers
michael@0:         movq        mm5, [rsi + 2*rax]    ; mm4 = r2 p0..p7
michael@0:         punpcklbw   mm5, mm0              ; mm5 = r2 p0..p3
michael@0:         pmullw      mm6, mm5              ; mm5 *= kernel 4 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = r0 p0..p3
michael@0:         psubusw     mm6, mm5              ; mm6 = r0 p0..p3 - r2 p0..p3
michael@0:         psubusw     mm5, mm1              ; mm5 = r2 p0..p3 - r2 p0..p3
michael@0:         paddusw     mm6, mm5              ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0: 
michael@0:         neg         rax
michael@0:         movq        mm6, [rbx ]           ; kernel 0 taps
michael@0:         movq        mm5, [rsi+2*rax]      ; mm4 = r-2 p0..p7
michael@0:         punpcklbw   mm5, mm0              ; mm5 = r-2 p0..p3
michael@0:         pmullw      mm6, mm5              ; mm5 *= kernel 0 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = r0 p0..p3
michael@0:         psubusw     mm6, mm5              ; mm6 = p0..p3 - r-2 p0..p3
michael@0:         psubusw     mm5, mm1              ; mm5 = r-2 p0..p3 - p0..p3
michael@0:         paddusw     mm6, mm5              ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0:         movq        mm6, [rbx + 16]       ; kernel 1 taps
michael@0:         movq        mm4, [rsi+rax]        ; mm4 = r-1 p0..p7
michael@0:         punpcklbw   mm4, mm0              ; mm4 = r-1 p0..p3
michael@0:         pmullw      mm6, mm4              ; mm4 *= kernel 1 modifiers.
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = r0 p0..p3
michael@0:         psubusw     mm6, mm4              ; mm6 = p0..p3 - r-2 p0..p3
michael@0:         psubusw     mm4, mm1              ; mm5 = r-1 p0..p3 - p0..p3
michael@0:         paddusw     mm6, mm4              ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0: 
michael@0:         paddusw     mm3, RD               ; mm3 += round value
michael@0:         psraw       mm3, VP9_FILTER_SHIFT     ; mm3 /= 128
michael@0: 
michael@0:         pand        mm1, mm7              ; mm1 select vals > thresh from source
michael@0:         pandn       mm7, mm3              ; mm7 select vals < thresh from blurred result
michael@0:         paddusw     mm1, mm7              ; combination
michael@0: 
michael@0:         packuswb    mm1, mm0              ; pack to bytes
michael@0: 
michael@0:         movd        [rdi], mm1            ;
michael@0:         neg         rax                   ; pitch is positive
michael@0: 
michael@0: 
michael@0:         add         rsi, 4
michael@0:         add         rdi, 4
michael@0:         add         rdx, 4
michael@0: 
michael@0:         cmp         edx, dword ptr arg(5) ;cols
michael@0:         jl          .nextcol
michael@0:         ; done with the all cols, start the across filtering in place
michael@0:         sub         rsi, rdx
michael@0:         sub         rdi, rdx
michael@0: 
michael@0: 
michael@0:         push        rax
michael@0:         xor         rdx,    rdx
michael@0:         mov         rax,    [rdi-4];
michael@0: 
michael@0: .acrossnextcol:
michael@0:         pxor        mm7, mm7              ; mm7 = 00000000
michael@0:         movq        mm6, [rbx + 32 ]      ;
michael@0:         movq        mm4, [rdi+rdx]        ; mm4 = p0..p7
michael@0:         movq        mm3, mm4              ; mm3 = p0..p7
michael@0:         punpcklbw   mm3, mm0              ; mm3 = p0..p3
michael@0:         movq        mm1, mm3              ; mm1 = p0..p3
michael@0:         pmullw      mm3, mm6              ; mm3 *= kernel 2 modifiers
michael@0: 
michael@0:         movq        mm6, [rbx + 48]
michael@0:         psrlq       mm4, 8                ; mm4 = p1..p7
michael@0:         movq        mm5, mm4              ; mm5 = p1..p7
michael@0:         punpcklbw   mm5, mm0              ; mm5 = p1..p4
michael@0:         pmullw      mm6, mm5              ; mm6 *= p1..p4 * kernel 3 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm6
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm7, mm1              ; mm7 = p0..p3
michael@0:         psubusw     mm7, mm5              ; mm7 = p0..p3 - p1..p4
michael@0:         psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
michael@0:         paddusw     mm7, mm5              ; mm7 = abs(p0..p3 - p1..p4)
michael@0:         pcmpgtw     mm7, mm2
michael@0: 
michael@0:         movq        mm6, [rbx + 64 ]
michael@0:         psrlq       mm4, 8                ; mm4 = p2..p7
michael@0:         movq        mm5, mm4              ; mm5 = p2..p7
michael@0:         punpcklbw   mm5, mm0              ; mm5 = p2..p5
michael@0:         pmullw      mm6, mm5              ; mm5 *= kernel 4 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = p0..p3
michael@0:         psubusw     mm6, mm5              ; mm6 = p0..p3 - p1..p4
michael@0:         psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
michael@0:         paddusw     mm6, mm5              ; mm6 = abs(p0..p3 - p1..p4)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0: 
michael@0:         movq        mm6, [rbx ]
michael@0:         movq        mm4, [rdi+rdx-2]      ; mm4 = p-2..p5
michael@0:         movq        mm5, mm4              ; mm5 = p-2..p5
michael@0:         punpcklbw   mm5, mm0              ; mm5 = p-2..p1
michael@0:         pmullw      mm6, mm5              ; mm5 *= kernel 0 modifiers
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = p0..p3
michael@0:         psubusw     mm6, mm5              ; mm6 = p0..p3 - p1..p4
michael@0:         psubusw     mm5, mm1              ; mm5 = p1..p4 - p0..p3
michael@0:         paddusw     mm6, mm5              ; mm6 = abs(p0..p3 - p1..p4)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0:         movq        mm6, [rbx + 16]
michael@0:         psrlq       mm4, 8                ; mm4 = p-1..p5
michael@0:         punpcklbw   mm4, mm0              ; mm4 = p-1..p2
michael@0:         pmullw      mm6, mm4              ; mm4 *= kernel 1 modifiers.
michael@0:         paddusw     mm3, mm6              ; mm3 += mm5
michael@0: 
michael@0:         ; thresholding
michael@0:         movq        mm6, mm1              ; mm6 = p0..p3
michael@0:         psubusw     mm6, mm4              ; mm6 = p0..p3 - p1..p4
michael@0:         psubusw     mm4, mm1              ; mm5 = p1..p4 - p0..p3
michael@0:         paddusw     mm6, mm4              ; mm6 = abs(p0..p3 - p1..p4)
michael@0:         pcmpgtw     mm6, mm2
michael@0:         por         mm7, mm6              ; accumulate thresholds
michael@0: 
michael@0:         paddusw     mm3, RD               ; mm3 += round value
michael@0:         psraw       mm3, VP9_FILTER_SHIFT     ; mm3 /= 128
michael@0: 
michael@0:         pand        mm1, mm7              ; mm1 select vals > thresh from source
michael@0:         pandn       mm7, mm3              ; mm7 select vals < thresh from blurred result
michael@0:         paddusw     mm1, mm7              ; combination
michael@0: 
michael@0:         packuswb    mm1, mm0              ; pack to bytes
michael@0:         mov         DWORD PTR [rdi+rdx-4],  eax   ; store previous four bytes
michael@0:         movd        eax,    mm1
michael@0: 
michael@0:         add         rdx, 4
michael@0:         cmp         edx, dword ptr arg(5) ;cols
michael@0:         jl          .acrossnextcol;
michael@0: 
michael@0:         mov         DWORD PTR [rdi+rdx-4],  eax
michael@0:         pop         rax
michael@0: 
michael@0:         ; done with this rwo
michael@0:         add         rsi,rax               ; next line
michael@0:         movsxd      rax, dword ptr arg(3) ;dst_pixels_per_line ; destination pitch?
michael@0:         add         rdi,rax               ; next destination
michael@0:         movsxd      rax, dword ptr arg(2) ;src_pixels_per_line ; destination pitch?
michael@0: 
michael@0:         dec         rcx                   ; decrement count
michael@0:         jnz         .nextrow               ; next row
michael@0:         pop         rbx
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 RD
michael@0: 
michael@0: 
michael@0: ;void vp9_mbpost_proc_down_mmx(unsigned char *dst,
michael@0: ;                             int pitch, int rows, int cols,int flimit)
michael@0: extern sym(vp9_rv)
michael@0: global sym(vp9_mbpost_proc_down_mmx) PRIVATE
michael@0: sym(vp9_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(vp9_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:             neg         rax                                     ; rax = -pitch
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(vp9_rv))]
michael@0:             movq        mm4,        [rax + rcx*2] ;vp9_rv[rcx*2]
michael@0:             pop         rax
michael@0: %elif ABI_IS_32BIT=0
michael@0:             movq        mm4,        [r8 + rcx*2] ;vp9_rv[rcx*2]
michael@0: %else
michael@0:             movq        mm4,        [sym(vp9_rv) + rcx*2]
michael@0: %endif
michael@0:             paddw       mm1,        mm4
michael@0:             ;paddw     xmm1,       eight8s
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 vp9_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(vp9_plane_add_noise_mmx) PRIVATE
michael@0: sym(vp9_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