1/*
2 * Copyright (C) 2002-2012 Michael Niedermayer
3 * Copyright (C) 2012 Ronald S. Bultje
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22#include "config.h"
23#include "libavutil/attributes.h"
24#include "libavutil/avassert.h"
25#include "libavutil/common.h"
26#include "libavutil/cpu.h"
27#include "libavutil/mem.h"
28#include "libavutil/x86/asm.h"
29#include "libavutil/x86/cpu.h"
30#include "libavcodec/videodsp.h"
31
32#if HAVE_YASM
33typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
34                                const uint8_t *src, x86_reg src_stride,
35                                x86_reg start_y, x86_reg end_y, x86_reg bh);
36typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
37                                const uint8_t *src, x86_reg src_stride,
38                                x86_reg start_y, x86_reg end_y, x86_reg bh,
39                                x86_reg w);
40
41extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx;
42extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx;
43extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx;
44extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx;
45extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx;
46extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx;
47extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx;
48extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx;
49extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx;
50extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx;
51extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx;
52extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx;
53extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx;
54extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx;
55extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx;
56extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx;
57extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx;
58extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx;
59extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx;
60extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx;
61extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx;
62extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx;
63#if ARCH_X86_32
64static emu_edge_vfix_func *vfixtbl_mmx[22] = {
65    &ff_emu_edge_vfix1_mmx,  &ff_emu_edge_vfix2_mmx,  &ff_emu_edge_vfix3_mmx,
66    &ff_emu_edge_vfix4_mmx,  &ff_emu_edge_vfix5_mmx,  &ff_emu_edge_vfix6_mmx,
67    &ff_emu_edge_vfix7_mmx,  &ff_emu_edge_vfix8_mmx,  &ff_emu_edge_vfix9_mmx,
68    &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx,
69    &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx,
70    &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx,
71    &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx,
72    &ff_emu_edge_vfix22_mmx
73};
74#endif
75extern emu_edge_vvar_func ff_emu_edge_vvar_mmx;
76extern emu_edge_vfix_func ff_emu_edge_vfix16_sse;
77extern emu_edge_vfix_func ff_emu_edge_vfix17_sse;
78extern emu_edge_vfix_func ff_emu_edge_vfix18_sse;
79extern emu_edge_vfix_func ff_emu_edge_vfix19_sse;
80extern emu_edge_vfix_func ff_emu_edge_vfix20_sse;
81extern emu_edge_vfix_func ff_emu_edge_vfix21_sse;
82extern emu_edge_vfix_func ff_emu_edge_vfix22_sse;
83static emu_edge_vfix_func *vfixtbl_sse[22] = {
84    ff_emu_edge_vfix1_mmx,  ff_emu_edge_vfix2_mmx,  ff_emu_edge_vfix3_mmx,
85    ff_emu_edge_vfix4_mmx,  ff_emu_edge_vfix5_mmx,  ff_emu_edge_vfix6_mmx,
86    ff_emu_edge_vfix7_mmx,  ff_emu_edge_vfix8_mmx,  ff_emu_edge_vfix9_mmx,
87    ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx,
88    ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx,
89    ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse,
90    ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse,
91    ff_emu_edge_vfix22_sse
92};
93extern emu_edge_vvar_func ff_emu_edge_vvar_sse;
94
95typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
96                                x86_reg start_x, x86_reg bh);
97typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
98                                x86_reg start_x, x86_reg n_words, x86_reg bh);
99
100extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx;
101extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx;
102extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx;
103extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx;
104extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx;
105extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx;
106extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx;
107extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx;
108extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx;
109extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx;
110extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx;
111#if ARCH_X86_32
112static emu_edge_hfix_func *hfixtbl_mmx[11] = {
113    ff_emu_edge_hfix2_mmx,  ff_emu_edge_hfix4_mmx,  ff_emu_edge_hfix6_mmx,
114    ff_emu_edge_hfix8_mmx,  ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
115    ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx,
116    ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx
117};
118#endif
119extern emu_edge_hvar_func ff_emu_edge_hvar_mmx;
120extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2;
121extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2;
122extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2;
123extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2;
124static emu_edge_hfix_func *hfixtbl_sse2[11] = {
125    ff_emu_edge_hfix2_mmx,  ff_emu_edge_hfix4_mmx,  ff_emu_edge_hfix6_mmx,
126    ff_emu_edge_hfix8_mmx,  ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
127    ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2,
128    ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2
129};
130extern emu_edge_hvar_func ff_emu_edge_hvar_sse2;
131
132static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
133                                              ptrdiff_t dst_stride,
134                                              ptrdiff_t src_stride,
135                                              x86_reg block_w, x86_reg block_h,
136                                              x86_reg src_x, x86_reg src_y,
137                                              x86_reg w, x86_reg h,
138                                              emu_edge_vfix_func **vfix_tbl,
139                                              emu_edge_vvar_func *v_extend_var,
140                                              emu_edge_hfix_func **hfix_tbl,
141                                              emu_edge_hvar_func *h_extend_var)
142{
143    x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
144
145    if (!w || !h)
146        return;
147
148    if (src_y >= h) {
149        src -= src_y*src_stride;
150        src_y_add = h - 1;
151        src_y     = h - 1;
152    } else if (src_y <= -block_h) {
153        src -= src_y*src_stride;
154        src_y_add = 1 - block_h;
155        src_y     = 1 - block_h;
156    }
157    if (src_x >= w) {
158        src   += w - 1 - src_x;
159        src_x  = w - 1;
160    } else if (src_x <= -block_w) {
161        src   += 1 - block_w - src_x;
162        src_x  = 1 - block_w;
163    }
164
165    start_y = FFMAX(0, -src_y);
166    start_x = FFMAX(0, -src_x);
167    end_y   = FFMIN(block_h, h-src_y);
168    end_x   = FFMIN(block_w, w-src_x);
169    av_assert2(start_x < end_x && block_w > 0);
170    av_assert2(start_y < end_y && block_h > 0);
171
172    // fill in the to-be-copied part plus all above/below
173    src += (src_y_add + start_y) * src_stride + start_x;
174    w = end_x - start_x;
175    if (w <= 22) {
176        vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
177                        start_y, end_y, block_h);
178    } else {
179        v_extend_var(dst + start_x, dst_stride, src, src_stride,
180                     start_y, end_y, block_h, w);
181    }
182
183    // fill left
184    if (start_x) {
185        if (start_x <= 22) {
186            hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
187        } else {
188            h_extend_var(dst, dst_stride,
189                         start_x, (start_x + 1) >> 1, block_h);
190        }
191    }
192
193    // fill right
194    p = block_w - end_x;
195    if (p) {
196        if (p <= 22) {
197            hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
198                                   -!(p & 1), block_h);
199        } else {
200            h_extend_var(dst + end_x - (p & 1), dst_stride,
201                         -!(p & 1), (p + 1) >> 1, block_h);
202        }
203    }
204}
205
206#if ARCH_X86_32
207static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
208                                             ptrdiff_t buf_stride,
209                                             ptrdiff_t src_stride,
210                                             int block_w, int block_h,
211                                             int src_x, int src_y, int w, int h)
212{
213    emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
214                     src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx,
215                     hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
216}
217
218static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
219                                             ptrdiff_t buf_stride,
220                                             ptrdiff_t src_stride,
221                                             int block_w, int block_h,
222                                             int src_x, int src_y, int w, int h)
223{
224    emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
225                     src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
226                     hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
227}
228#endif
229
230static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
231                                              ptrdiff_t buf_stride,
232                                              ptrdiff_t src_stride,
233                                              int block_w, int block_h,
234                                              int src_x, int src_y, int w,
235                                              int h)
236{
237    emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
238                     src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
239                     hfixtbl_sse2, &ff_emu_edge_hvar_sse2);
240}
241#endif /* HAVE_YASM */
242
243void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h);
244void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h);
245
246av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
247{
248#if HAVE_YASM
249    int cpu_flags = av_get_cpu_flags();
250
251#if ARCH_X86_32
252    if (EXTERNAL_MMX(cpu_flags) && bpc <= 8) {
253        ctx->emulated_edge_mc = emulated_edge_mc_mmx;
254    }
255    if (EXTERNAL_AMD3DNOW(cpu_flags)) {
256        ctx->prefetch = ff_prefetch_3dnow;
257    }
258#endif /* ARCH_X86_32 */
259    if (EXTERNAL_MMXEXT(cpu_flags)) {
260        ctx->prefetch = ff_prefetch_mmxext;
261    }
262#if ARCH_X86_32
263    if (EXTERNAL_SSE(cpu_flags) && bpc <= 8) {
264        ctx->emulated_edge_mc = emulated_edge_mc_sse;
265    }
266#endif /* ARCH_X86_32 */
267    if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
268        ctx->emulated_edge_mc = emulated_edge_mc_sse2;
269    }
270#endif /* HAVE_YASM */
271}
272