2 * Copyright (C) 2010 David Conrad
4 * This file is part of FFmpeg.
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.
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.
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 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/x86/cpu.h"
22 #include "libavcodec/diracdsp.h"
25 DECL_DIRAC_PIXOP(put, mmx);
26 DECL_DIRAC_PIXOP(avg, mmx);
27 DECL_DIRAC_PIXOP(avg, mmxext);
29 void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h);
30 void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h);
31 void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h);
32 void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h);
34 void ff_add_rect_clamped_mmx(uint8_t *, const uint16_t *, int, const int16_t *, int, int, int);
35 void ff_add_rect_clamped_sse2(uint8_t *, const uint16_t *, int, const int16_t *, int, int, int);
37 void ff_add_dirac_obmc8_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
38 void ff_add_dirac_obmc16_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
39 void ff_add_dirac_obmc32_mmx(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
41 void ff_add_dirac_obmc16_sse2(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
42 void ff_add_dirac_obmc32_sse2(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
44 void ff_put_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
45 void ff_put_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
46 void ff_put_signed_rect_clamped_mmx(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
47 void ff_put_signed_rect_clamped_sse2(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height);
48 void ff_put_signed_rect_clamped_10_sse4(uint8_t *dst, int dst_stride, const uint8_t *src, int src_stride, int width, int height);
50 void ff_dequant_subband_32_sse4(uint8_t *src, uint8_t *dst, ptrdiff_t stride, const int qf, const int qs, int tot_v, int tot_h);
54 #define HPEL_FILTER(MMSIZE, EXT) \
55 void ff_dirac_hpel_filter_v_ ## EXT(uint8_t *, const uint8_t *, int, int); \
56 void ff_dirac_hpel_filter_h_ ## EXT(uint8_t *, const uint8_t *, int); \
58 static void dirac_hpel_filter_ ## EXT(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, \
59 const uint8_t *src, int stride, int width, int height) \
63 ff_dirac_hpel_filter_v_ ## EXT(dstv-MMSIZE, src-MMSIZE, stride, width+MMSIZE+5); \
64 ff_dirac_hpel_filter_h_ ## EXT(dsth, src, width); \
65 ff_dirac_hpel_filter_h_ ## EXT(dstc, dstv, width); \
74 #define PIXFUNC(PFX, IDX, EXT) \
75 /*MMXDISABLEDc->PFX ## _dirac_pixels_tab[0][IDX] = ff_ ## PFX ## _dirac_pixels8_ ## EXT;*/ \
76 c->PFX ## _dirac_pixels_tab[1][IDX] = ff_ ## PFX ## _dirac_pixels16_ ## EXT; \
77 c->PFX ## _dirac_pixels_tab[2][IDX] = ff_ ## PFX ## _dirac_pixels32_ ## EXT
79 #define DIRAC_PIXOP(OPNAME2, OPNAME, EXT)\
80 void ff_ ## OPNAME2 ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
83 ff_ ## OPNAME2 ## _dirac_pixels16_c(dst, src, stride, h);\
85 OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
87 void ff_ ## OPNAME2 ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
90 ff_ ## OPNAME2 ## _dirac_pixels32_c(dst, src, stride, h);\
92 OPNAME ## _pixels16_ ## EXT(dst , src[0] , stride, h);\
93 OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
97 DIRAC_PIXOP(put, ff_put, mmx)
98 DIRAC_PIXOP(avg, ff_avg, mmx)
99 DIRAC_PIXOP(avg, ff_avg, mmxext)
101 void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
104 ff_put_dirac_pixels16_c(dst, src, stride, h);
106 ff_put_pixels16_sse2(dst, src[0], stride, h);
108 void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
111 ff_avg_dirac_pixels16_c(dst, src, stride, h);
113 ff_avg_pixels16_sse2(dst, src[0], stride, h);
115 void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
118 ff_put_dirac_pixels32_c(dst, src, stride, h);
120 ff_put_pixels16_sse2(dst , src[0] , stride, h);
121 ff_put_pixels16_sse2(dst+16, src[0]+16, stride, h);
124 void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
127 ff_avg_dirac_pixels32_c(dst, src, stride, h);
129 ff_avg_pixels16_sse2(dst , src[0] , stride, h);
130 ff_avg_pixels16_sse2(dst+16, src[0]+16, stride, h);
136 #define HPEL_FILTER(MMSIZE, EXT) \
137 void dirac_hpel_filter_ ## EXT(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, \
138 const uint8_t *src, int stride, int width, int height);
140 #define PIXFUNC(PFX, IDX, EXT) do {} while (0)
142 #endif // HAVE_X86ASM
147 HPEL_FILTER(16, sse2)
149 void ff_diracdsp_init_x86(DiracDSPContext* c)
151 int mm_flags = av_get_cpu_flags();
153 if (EXTERNAL_MMX(mm_flags)) {
154 c->add_dirac_obmc[0] = ff_add_dirac_obmc8_mmx;
156 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_mmx;
157 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
158 c->dirac_hpel_filter = dirac_hpel_filter_mmx;
159 c->add_rect_clamped = ff_add_rect_clamped_mmx;
160 c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_mmx;
162 PIXFUNC(put, 0, mmx);
163 PIXFUNC(avg, 0, mmx);
166 if (EXTERNAL_MMXEXT(mm_flags)) {
167 PIXFUNC(avg, 0, mmxext);
170 if (EXTERNAL_SSE2(mm_flags)) {
171 c->dirac_hpel_filter = dirac_hpel_filter_sse2;
172 c->add_rect_clamped = ff_add_rect_clamped_sse2;
173 c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_sse2;
175 c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
176 c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;
178 c->put_dirac_pixels_tab[1][0] = ff_put_dirac_pixels16_sse2;
179 c->avg_dirac_pixels_tab[1][0] = ff_avg_dirac_pixels16_sse2;
180 c->put_dirac_pixels_tab[2][0] = ff_put_dirac_pixels32_sse2;
181 c->avg_dirac_pixels_tab[2][0] = ff_avg_dirac_pixels32_sse2;
184 if (EXTERNAL_SSE4(mm_flags)) {
185 c->dequant_subband[1] = ff_dequant_subband_32_sse4;
186 c->put_signed_rect_clamped[1] = ff_put_signed_rect_clamped_10_sse4;