1;
2;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3;
4;  Use of this source code is governed by a BSD-style license
5;  that can be found in the LICENSE file in the root of the source
6;  tree. An additional intellectual property rights grant can be found
7;  in the file PATENTS.  All contributing project authors may
8;  be found in the AUTHORS file in the root of the source tree.
9;
10
11
12%include "vpx_ports/x86_abi_support.asm"
13
14;void vpx_plane_add_noise_sse2(uint8_t *start, const int8_t *noise,
15;                              int blackclamp, int whiteclamp,
16;                              int width, int height, int pitch)
17global sym(vpx_plane_add_noise_sse2) PRIVATE
18sym(vpx_plane_add_noise_sse2):
19    push        rbp
20    mov         rbp, rsp
21    SHADOW_ARGS_TO_STACK 7
22    GET_GOT     rbx
23    push        rsi
24    push        rdi
25
26    mov         rdx, 0x01010101
27    mov         rax, arg(2)
28    mul         rdx
29    movd        xmm3, rax
30    pshufd      xmm3, xmm3, 0  ; xmm3 is 16 copies of char in blackclamp
31
32    mov         rdx, 0x01010101
33    mov         rax, arg(3)
34    mul         rdx
35    movd        xmm4, rax
36    pshufd      xmm4, xmm4, 0  ; xmm4 is 16 copies of char in whiteclamp
37
38    movdqu      xmm5, xmm3     ; both clamp = black clamp + white clamp
39    paddusb     xmm5, xmm4
40
41.addnoise_loop:
42    call sym(LIBVPX_RAND) WRT_PLT
43    mov     rcx, arg(1) ;noise
44    and     rax, 0xff
45    add     rcx, rax
46
47    mov     rdi, rcx
48    movsxd  rcx, dword arg(4) ;[Width]
49    mov     rsi, arg(0) ;Pos
50    xor     rax, rax
51
52.addnoise_nextset:
53      movdqu      xmm1,[rsi+rax]         ; get the source
54
55      psubusb     xmm1, xmm3 ; subtract black clamp
56      paddusb     xmm1, xmm5 ; add both clamp
57      psubusb     xmm1, xmm4 ; subtract whiteclamp
58
59      movdqu      xmm2,[rdi+rax]         ; get the noise for this line
60      paddb       xmm1,xmm2              ; add it in
61      movdqu      [rsi+rax],xmm1         ; store the result
62
63      add         rax,16                 ; move to the next line
64
65      cmp         rax, rcx
66      jl          .addnoise_nextset
67
68    movsxd  rax, dword arg(6) ; Pitch
69    add     arg(0), rax ; Start += Pitch
70    sub     dword arg(5), 1   ; Height -= 1
71    jg      .addnoise_loop
72
73    ; begin epilog
74    pop rdi
75    pop rsi
76    RESTORE_GOT
77    UNSHADOW_ARGS
78    pop         rbp
79    ret
80
81SECTION_RODATA
82align 16
83rd42:
84    times 8 dw 0x04
85four8s:
86    times 4 dd 8
87