1;******************************************************************************
2;* Copyright (c) 2010 David Conrad
3;*
4;* This file is part of FFmpeg.
5;*
6;* FFmpeg is free software; you can redistribute it and/or
7;* modify it under the terms of the GNU Lesser General Public
8;* License as published by the Free Software Foundation; either
9;* version 2.1 of the License, or (at your option) any later version.
10;*
11;* FFmpeg is distributed in the hope that it will be useful,
12;* but WITHOUT ANY WARRANTY; without even the implied warranty of
13;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14;* Lesser General Public License for more details.
15;*
16;* You should have received a copy of the GNU Lesser General Public
17;* License along with FFmpeg; if not, write to the Free Software
18;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19;******************************************************************************
20
21%include "libavutil/x86/x86util.asm"
22
23SECTION_RODATA
24pw_3: times 8 dw 3
25pw_7: times 8 dw 7
26pw_16: times 8 dw 16
27pw_32: times 8 dw 32
28pb_128: times 16 db 128
29
30section .text
31
32%macro UNPACK_ADD 6
33    mov%5   %1, %3
34    mov%6   m5, %4
35    mova    m4, %1
36    mova    %2, m5
37    punpcklbw %1, m7
38    punpcklbw m5, m7
39    punpckhbw m4, m7
40    punpckhbw %2, m7
41    paddw   %1, m5
42    paddw   %2, m4
43%endmacro
44
45%macro HPEL_FILTER 1
46; dirac_hpel_filter_v_sse2(uint8_t *dst, uint8_t *src, int stride, int width);
47cglobal dirac_hpel_filter_v_%1, 4,6,8, dst, src, stride, width, src0, stridex3
48    mov     src0q, srcq
49    lea     stridex3q, [3*strideq]
50    sub     src0q, stridex3q
51    pxor    m7, m7
52.loop:
53    ; 7*(src[0] + src[1])
54    UNPACK_ADD m0, m1, [srcq], [srcq + strideq], a,a
55    pmullw  m0, [pw_7]
56    pmullw  m1, [pw_7]
57
58    ; 3*( ... + src[-2] + src[3])
59    UNPACK_ADD m2, m3, [src0q + strideq], [srcq + stridex3q], a,a
60    paddw   m0, m2
61    paddw   m1, m3
62    pmullw  m0, [pw_3]
63    pmullw  m1, [pw_3]
64
65    ; ... - 7*(src[-1] + src[2])
66    UNPACK_ADD m2, m3, [src0q + strideq*2], [srcq + strideq*2], a,a
67    pmullw  m2, [pw_7]
68    pmullw  m3, [pw_7]
69    psubw   m0, m2
70    psubw   m1, m3
71
72    ; ... - (src[-3] + src[4])
73    UNPACK_ADD m2, m3, [src0q], [srcq + strideq*4], a,a
74    psubw   m0, m2
75    psubw   m1, m3
76
77    paddw   m0, [pw_16]
78    paddw   m1, [pw_16]
79    psraw   m0, 5
80    psraw   m1, 5
81    packuswb m0, m1
82    mova    [dstq], m0
83    add     dstq, mmsize
84    add     srcq, mmsize
85    add     src0q, mmsize
86    sub     widthd, mmsize
87    jg      .loop
88    RET
89
90; dirac_hpel_filter_h_sse2(uint8_t *dst, uint8_t *src, int width);
91cglobal dirac_hpel_filter_h_%1, 3,3,8, dst, src, width
92    dec     widthd
93    pxor    m7, m7
94    and     widthd, ~(mmsize-1)
95.loop:
96    ; 7*(src[0] + src[1])
97    UNPACK_ADD m0, m1, [srcq + widthq], [srcq + widthq + 1], u,u
98    pmullw  m0, [pw_7]
99    pmullw  m1, [pw_7]
100
101    ; 3*( ... + src[-2] + src[3])
102    UNPACK_ADD m2, m3, [srcq + widthq - 2], [srcq + widthq + 3], u,u
103    paddw   m0, m2
104    paddw   m1, m3
105    pmullw  m0, [pw_3]
106    pmullw  m1, [pw_3]
107
108    ; ... - 7*(src[-1] + src[2])
109    UNPACK_ADD m2, m3, [srcq + widthq - 1], [srcq + widthq + 2], u,u
110    pmullw  m2, [pw_7]
111    pmullw  m3, [pw_7]
112    psubw   m0, m2
113    psubw   m1, m3
114
115    ; ... - (src[-3] + src[4])
116    UNPACK_ADD m2, m3, [srcq + widthq - 3], [srcq + widthq + 4], u,u
117    psubw   m0, m2
118    psubw   m1, m3
119
120    paddw   m0, [pw_16]
121    paddw   m1, [pw_16]
122    psraw   m0, 5
123    psraw   m1, 5
124    packuswb m0, m1
125    mova    [dstq + widthq], m0
126    sub     widthd, mmsize
127    jge     .loop
128    RET
129%endmacro
130
131%macro PUT_RECT 1
132; void put_rect_clamped(uint8_t *dst, int dst_stride, int16_t *src, int src_stride, int width, int height)
133cglobal put_signed_rect_clamped_%1, 5,9,3, dst, dst_stride, src, src_stride, w, dst2, src2
134    mova    m0, [pb_128]
135    add     wd, (mmsize-1)
136    and     wd, ~(mmsize-1)
137
138%if ARCH_X86_64
139    movsxd   dst_strideq, dst_strided
140    movsxd   src_strideq, src_strided
141    mov   r7d, r5m
142    mov   r8d, wd
143    %define wspill r8d
144    %define hd r7d
145%else
146    mov    r4m, wd
147    %define wspill r4m
148    %define hd r5mp
149%endif
150
151.loopy
152    lea     src2q, [srcq+src_strideq*2]
153    lea     dst2q, [dstq+dst_strideq]
154.loopx:
155    sub      wd, mmsize
156    mova     m1, [srcq +2*wq]
157    mova     m2, [src2q+2*wq]
158    packsswb m1, [srcq +2*wq+mmsize]
159    packsswb m2, [src2q+2*wq+mmsize]
160    paddb    m1, m0
161    paddb    m2, m0
162    mova    [dstq +wq], m1
163    mova    [dst2q+wq], m2
164    jg      .loopx
165
166    lea   srcq, [srcq+src_strideq*4]
167    lea   dstq, [dstq+dst_strideq*2]
168    sub     hd, 2
169    mov     wd, wspill
170    jg      .loopy
171    RET
172%endm
173
174%macro ADD_RECT 1
175; void add_rect_clamped(uint8_t *dst, uint16_t *src, int stride, int16_t *idwt, int idwt_stride, int width, int height)
176cglobal add_rect_clamped_%1, 7,9,3, dst, src, stride, idwt, idwt_stride, w, h
177    mova    m0, [pw_32]
178    add     wd, (mmsize-1)
179    and     wd, ~(mmsize-1)
180
181%if ARCH_X86_64
182    movsxd   strideq, strided
183    movsxd   idwt_strideq, idwt_strided
184    mov   r8d, wd
185    %define wspill r8d
186%else
187    mov    r5m, wd
188    %define wspill r5m
189%endif
190
191.loop:
192    sub     wd, mmsize
193    movu    m1, [srcq +2*wq] ; FIXME: ensure alignment
194    paddw   m1, m0
195    psraw   m1, 6
196    movu    m2, [srcq +2*wq+mmsize] ; FIXME: ensure alignment
197    paddw   m2, m0
198    psraw   m2, 6
199    paddw   m1, [idwtq+2*wq]
200    paddw   m2, [idwtq+2*wq+mmsize]
201    packuswb m1, m2
202    mova    [dstq +wq], m1
203    jg      .loop
204
205    lea   srcq, [srcq + 2*strideq]
206    add   dstq, strideq
207    lea  idwtq, [idwtq+ 2*idwt_strideq]
208    sub     hd, 1
209    mov     wd, wspill
210    jg      .loop
211    RET
212%endm
213
214%macro ADD_OBMC 2
215; void add_obmc(uint16_t *dst, uint8_t *src, int stride, uint8_t *obmc_weight, int yblen)
216cglobal add_dirac_obmc%1_%2, 6,6,5, dst, src, stride, obmc, yblen
217    pxor        m4, m4
218.loop:
219%assign i 0
220%rep %1 / mmsize
221    mova        m0, [srcq+i]
222    mova        m1, m0
223    punpcklbw   m0, m4
224    punpckhbw   m1, m4
225    mova        m2, [obmcq+i]
226    mova        m3, m2
227   punpcklbw   m2, m4
228    punpckhbw   m3, m4
229    pmullw      m0, m2
230    pmullw      m1, m3
231    movu        m2, [dstq+2*i]
232    movu        m3, [dstq+2*i+mmsize]
233    paddw       m0, m2
234    paddw       m1, m3
235    movu        [dstq+2*i], m0
236    movu        [dstq+2*i+mmsize], m1
237%assign i i+mmsize
238%endrep
239    lea         srcq, [srcq+strideq]
240    lea         dstq, [dstq+2*strideq]
241    add         obmcq, 32
242    sub         yblend, 1
243    jg          .loop
244    RET
245%endm
246
247INIT_MMX
248%if ARCH_X86_64 == 0
249PUT_RECT mmx
250ADD_RECT mmx
251
252HPEL_FILTER mmx
253ADD_OBMC 32, mmx
254ADD_OBMC 16, mmx
255%endif
256ADD_OBMC 8, mmx
257
258INIT_XMM
259PUT_RECT sse2
260ADD_RECT sse2
261
262HPEL_FILTER sse2
263ADD_OBMC 32, sse2
264ADD_OBMC 16, sse2
265