2 * Copyright (c) 2013 Seppo Tomperi
3 * Copyright (c) 2013 - 2014 Pierre-Edouard Lepere
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/cpu.h"
25 #include "libavutil/x86/cpu.h"
27 #include "libavcodec/hevcdsp.h"
29 #define LFC_FUNC(DIR, DEPTH, OPT) \
30 void ff_hevc_ ## DIR ## _loop_filter_chroma_ ## DEPTH ## _ ## OPT(uint8_t *pix, ptrdiff_t stride, int *tc, uint8_t *no_p, uint8_t *no_q);
32 #define LFL_FUNC(DIR, DEPTH, OPT) \
33 void ff_hevc_ ## DIR ## _loop_filter_luma_ ## DEPTH ## _ ## OPT(uint8_t *pix, ptrdiff_t stride, int beta, int *tc, uint8_t *no_p, uint8_t *no_q);
35 #define LFC_FUNCS(type, depth) \
36 LFC_FUNC(h, depth, sse2) \
37 LFC_FUNC(v, depth, sse2)
39 #define LFL_FUNCS(type, depth) \
40 LFL_FUNC(h, depth, ssse3) \
41 LFL_FUNC(v, depth, ssse3)
44 LFC_FUNCS(uint8_t, 10)
46 LFL_FUNCS(uint8_t, 10)
48 #define idct_dc_proto(size, bitd, opt) \
49 void ff_hevc_idct_ ## size ## _dc_add_ ## bitd ## _ ## opt(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
51 idct_dc_proto(4, 8,mmxext);
52 idct_dc_proto(8, 8,mmxext);
53 idct_dc_proto(16,8, sse2);
54 idct_dc_proto(32,8, sse2);
56 idct_dc_proto(32,8, avx2);
58 idct_dc_proto(4, 10,mmxext);
59 idct_dc_proto(8, 10, sse2);
60 idct_dc_proto(16,10, sse2);
61 idct_dc_proto(32,10, sse2);
62 idct_dc_proto(8, 10, avx);
63 idct_dc_proto(16,10, avx);
64 idct_dc_proto(32,10, avx);
66 idct_dc_proto(16,10, avx2);
67 idct_dc_proto(32,10, avx2);
69 #define IDCT_FUNCS(W, opt) \
70 void ff_hevc_idct_ ## W ## _dc_8_ ## opt(int16_t *coeffs); \
71 void ff_hevc_idct_ ## W ## _dc_10_ ## opt(int16_t *coeffs)
73 IDCT_FUNCS(4x4, mmxext);
74 IDCT_FUNCS(8x8, mmxext);
75 IDCT_FUNCS(8x8, sse2);
76 IDCT_FUNCS(16x16, sse2);
77 IDCT_FUNCS(32x32, sse2);
78 IDCT_FUNCS(16x16, avx2);
79 IDCT_FUNCS(32x32, avx2);
81 #define GET_PIXELS(width, depth, cf) \
82 void ff_hevc_get_pixels_ ## width ## _ ## depth ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
83 uint8_t *src, ptrdiff_t srcstride, \
84 int height, int mx, int my, int16_t *mcbuffer);
86 GET_PIXELS(4, 8, sse2)
87 GET_PIXELS(8, 8, sse2)
88 GET_PIXELS(12, 8, sse2)
89 GET_PIXELS(16, 8, sse2)
90 GET_PIXELS(24, 8, sse2)
91 GET_PIXELS(32, 8, sse2)
92 GET_PIXELS(48, 8, sse2)
93 GET_PIXELS(64, 8, sse2)
95 GET_PIXELS(4, 10, sse2)
96 GET_PIXELS(8, 10, sse2)
97 GET_PIXELS(12, 10, sse2)
98 GET_PIXELS(16, 10, sse2)
99 GET_PIXELS(24, 10, sse2)
100 GET_PIXELS(32, 10, sse2)
101 GET_PIXELS(48, 10, sse2)
102 GET_PIXELS(64, 10, sse2)
104 /* those are independent of the bit depth, so declared separately */
105 #define INTERP_HV_FUNC(width, cf) \
106 void ff_hevc_qpel_hv_ ## width ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
107 int16_t *src, ptrdiff_t srcstride, \
108 int height, int mx, int my, int16_t *mcbuffer); \
109 void ff_hevc_epel_hv_ ## width ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
110 int16_t *src, ptrdiff_t srcstride, \
111 int height, int mx, int my, int16_t *mcbuffer);
113 INTERP_HV_FUNC(4, avx)
114 INTERP_HV_FUNC(8, avx)
115 INTERP_HV_FUNC(12, avx)
116 INTERP_HV_FUNC(16, avx)
117 INTERP_HV_FUNC(24, avx)
118 INTERP_HV_FUNC(32, avx)
119 INTERP_HV_FUNC(48, avx)
120 INTERP_HV_FUNC(64, avx)
122 #if ARCH_X86_64 && HAVE_AVX_EXTERNAL
123 #define QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv) \
124 static void hevc_qpel_hv_ ## width ## _ ## depth ## _ ## cf_hv(int16_t *dst, ptrdiff_t dststride, \
125 uint8_t *src, ptrdiff_t srcstride, \
126 int height, int mx, int my, int16_t *mcbuffer) \
128 const ptrdiff_t stride = FFALIGN(width + 7, 8); \
129 ff_hevc_qpel_h_ ## width ## _ ## depth ## _ ## cf_h(mcbuffer, 2 * stride, src - 3 * srcstride, srcstride, \
130 height + 7, mx, my, mcbuffer); \
131 ff_hevc_qpel_hv_ ## width ## _ ## cf_hv(dst, dststride, mcbuffer + 3 * stride, 2 * stride, \
132 height, mx, my, mcbuffer); \
135 #define QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
136 #endif /* ARCH_X86_64 && HAVE_AVX_EXTERNAL */
138 #define QPEL_FUNCS(width, depth, cf_h, cf_v, cf_hv) \
139 void ff_hevc_qpel_h_ ## width ## _ ## depth ## _ ## cf_h(int16_t *dst, ptrdiff_t dststride, \
140 uint8_t *src, ptrdiff_t srcstride, \
141 int height, int mx, int my, int16_t *mcbuffer); \
142 void ff_hevc_qpel_v_ ## width ## _ ## depth ## _ ## cf_v(int16_t *dst, ptrdiff_t dststride, \
143 uint8_t *src, ptrdiff_t srcstride, \
144 int height, int mx, int my, int16_t *mcbuffer); \
145 QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
147 QPEL_FUNCS(4, 8, ssse3, ssse3, avx)
148 QPEL_FUNCS(8, 8, ssse3, ssse3, avx)
149 QPEL_FUNCS(12, 8, ssse3, ssse3, avx)
150 QPEL_FUNCS(16, 8, ssse3, ssse3, avx)
151 QPEL_FUNCS(24, 8, ssse3, ssse3, avx)
152 QPEL_FUNCS(32, 8, ssse3, ssse3, avx)
153 QPEL_FUNCS(48, 8, ssse3, ssse3, avx)
154 QPEL_FUNCS(64, 8, ssse3, ssse3, avx)
156 QPEL_FUNCS(4, 10, avx, avx, avx)
157 QPEL_FUNCS(8, 10, avx, avx, avx)
158 QPEL_FUNCS(12, 10, avx, avx, avx)
159 QPEL_FUNCS(16, 10, avx, avx, avx)
160 QPEL_FUNCS(24, 10, avx, avx, avx)
161 QPEL_FUNCS(32, 10, avx, avx, avx)
162 QPEL_FUNCS(48, 10, avx, avx, avx)
163 QPEL_FUNCS(64, 10, avx, avx, avx)
165 #if ARCH_X86_64 && HAVE_AVX_EXTERNAL
166 #define EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv) \
167 static void hevc_epel_hv_ ## width ## _ ## depth ## _ ## cf_hv(int16_t *dst, ptrdiff_t dststride, \
168 uint8_t *src, ptrdiff_t srcstride, \
169 int height, int mx, int my, int16_t *mcbuffer) \
171 const ptrdiff_t stride = FFALIGN(width + 3, 8); \
172 ff_hevc_epel_h_ ## width ## _ ## depth ## _ ## cf_h(mcbuffer, 2 * stride, src - srcstride, srcstride, \
173 height + 3, mx, my, mcbuffer); \
174 ff_hevc_epel_hv_ ## width ## _ ## cf_hv(dst, dststride, mcbuffer + stride, 2 * stride, \
175 height, mx, my, mcbuffer); \
178 #define EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
179 #endif /* ARCH_X86_64 && HAVE_AVX_EXTERNAL */
181 #define EPEL_FUNCS(width, depth, cf_h, cf_v, cf_hv) \
182 void ff_hevc_epel_h_ ## width ## _ ## depth ## _ ## cf_h(int16_t *dst, ptrdiff_t dststride, \
183 uint8_t *src, ptrdiff_t srcstride, \
184 int height, int mx, int my, int16_t *mcbuffer); \
185 void ff_hevc_epel_v_ ## width ## _ ## depth ## _ ## cf_v(int16_t *dst, ptrdiff_t dststride, \
186 uint8_t *src, ptrdiff_t srcstride, \
187 int height, int mx, int my, int16_t *mcbuffer); \
188 EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
190 EPEL_FUNCS(4, 8, ssse3, ssse3, avx)
191 EPEL_FUNCS(8, 8, ssse3, ssse3, avx)
192 EPEL_FUNCS(12, 8, ssse3, ssse3, avx)
193 EPEL_FUNCS(16, 8, ssse3, ssse3, avx)
194 EPEL_FUNCS(24, 8, ssse3, ssse3, avx)
195 EPEL_FUNCS(32, 8, ssse3, ssse3, avx)
197 EPEL_FUNCS(4, 10, avx, avx, avx)
198 EPEL_FUNCS(8, 10, avx, avx, avx)
199 EPEL_FUNCS(12, 10, avx, avx, avx)
200 EPEL_FUNCS(16, 10, avx, avx, avx)
201 EPEL_FUNCS(24, 10, avx, avx, avx)
202 EPEL_FUNCS(32, 10, avx, avx, avx)
204 #define PUT_PRED(width, depth, cf_uw, cf_w) \
205 void ff_hevc_put_unweighted_pred_ ## width ## _ ## depth ## _ ## cf_uw(uint8_t *dst, ptrdiff_t dststride, \
206 int16_t *src, ptrdiff_t srcstride, \
208 void ff_hevc_put_unweighted_pred_avg_ ## width ## _ ## depth ## _ ## cf_uw(uint8_t *dst, ptrdiff_t dststride, \
209 int16_t *src1, int16_t *src2, \
210 ptrdiff_t srcstride, int height); \
211 void ff_hevc_put_weighted_pred_ ## width ## _ ## depth ## _ ## cf_w(uint8_t denom, int16_t weight, int16_t offset, \
212 uint8_t *dst, ptrdiff_t dststride, \
213 int16_t *src, ptrdiff_t srcstride, \
215 void ff_hevc_put_weighted_pred_avg_ ## width ## _ ## depth ## _ ## cf_w(uint8_t denom, int16_t weight0, int16_t weight1, \
216 int16_t offset0, int16_t offset1, \
217 uint8_t *dst, ptrdiff_t dststride, \
218 int16_t *src0, int16_t *src1, ptrdiff_t srcstride, \
221 PUT_PRED(4, 8, sse2, sse4)
222 PUT_PRED(8, 8, sse2, sse4)
223 PUT_PRED(12, 8, sse2, sse4)
224 PUT_PRED(16, 8, sse2, sse4)
225 PUT_PRED(24, 8, sse2, sse4)
226 PUT_PRED(32, 8, sse2, sse4)
227 PUT_PRED(48, 8, sse2, sse4)
228 PUT_PRED(64, 8, sse2, sse4)
230 PUT_PRED(4, 10, sse2, sse4)
231 PUT_PRED(8, 10, sse2, sse4)
232 PUT_PRED(12, 10, sse2, sse4)
233 PUT_PRED(16, 10, sse2, sse4)
234 PUT_PRED(24, 10, sse2, sse4)
235 PUT_PRED(32, 10, sse2, sse4)
236 PUT_PRED(48, 10, sse2, sse4)
237 PUT_PRED(64, 10, sse2, sse4)
239 void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
241 int cpu_flags = av_get_cpu_flags();
243 #define SET_LUMA_FUNCS(tabname, funcname, depth, cf) \
244 c->tabname[0] = funcname ## _4_ ## depth ## _ ## cf; \
245 c->tabname[1] = funcname ## _8_ ## depth ## _ ## cf; \
246 c->tabname[2] = funcname ## _12_ ## depth ## _ ## cf; \
247 c->tabname[3] = funcname ## _16_ ## depth ## _ ## cf; \
248 c->tabname[4] = funcname ## _24_ ## depth ## _ ## cf; \
249 c->tabname[5] = funcname ## _32_ ## depth ## _ ## cf; \
250 c->tabname[6] = funcname ## _48_ ## depth ## _ ## cf; \
251 c->tabname[7] = funcname ## _64_ ## depth ## _ ## cf;
253 #define SET_CHROMA_FUNCS(tabname, funcname, depth, cf) \
254 c->tabname[1] = funcname ## _4_ ## depth ## _ ## cf; \
255 c->tabname[3] = funcname ## _8_ ## depth ## _ ## cf; \
256 c->tabname[4] = funcname ## _12_ ## depth ## _ ## cf; \
257 c->tabname[5] = funcname ## _16_ ## depth ## _ ## cf; \
258 c->tabname[6] = funcname ## _24_ ## depth ## _ ## cf; \
259 c->tabname[7] = funcname ## _32_ ## depth ## _ ## cf;
261 #define SET_QPEL_FUNCS(v, h, depth, cf, name) SET_LUMA_FUNCS (put_hevc_qpel[v][h], name, depth, cf)
262 #define SET_EPEL_FUNCS(v, h, depth, cf, name) SET_CHROMA_FUNCS(put_hevc_epel[v][h], name, depth, cf)
264 if (bit_depth == 8) {
265 if (EXTERNAL_MMXEXT(cpu_flags)) {
266 c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_mmxext;
267 c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_mmxext;
269 if (EXTERNAL_SSE2(cpu_flags)) {
270 c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
271 c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
273 c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_sse2;
274 c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_sse2;
275 c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_sse2;
276 SET_QPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
277 SET_EPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
279 SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 8, sse2);
280 SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 8, sse2);
281 SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 8, sse2);
282 SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 8, sse2);
284 if (EXTERNAL_SSSE3(cpu_flags)) {
285 SET_QPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_qpel_h);
286 SET_QPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_qpel_v);
287 SET_EPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_epel_h);
288 SET_EPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_epel_v);
291 } else if (bit_depth == 10) {
292 if (EXTERNAL_MMXEXT(cpu_flags)) {
293 c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
294 c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
296 if (EXTERNAL_SSE2(cpu_flags)) {
297 c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
298 c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
300 c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_sse2;
301 c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_sse2;
302 c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_sse2;
304 SET_QPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
305 SET_EPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
307 SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 10, sse2);
308 SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 10, sse2);
309 SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 10, sse2);
310 SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 10, sse2);
315 if (bit_depth == 8) {
316 if (EXTERNAL_SSSE3(cpu_flags)) {
317 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
318 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
321 if (EXTERNAL_SSE4(cpu_flags)) {
322 SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 8, sse4);
323 SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 8, sse4);
324 SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 8, sse4);
325 SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 8, sse4);
328 if (EXTERNAL_AVX(cpu_flags)) {
329 #if HAVE_AVX_EXTERNAL
330 SET_QPEL_FUNCS(1, 1, 8, avx, hevc_qpel_hv);
331 SET_EPEL_FUNCS(1, 1, 8, avx, hevc_epel_hv);
332 #endif /* HAVE_AVX_EXTERNAL */
334 if (EXTERNAL_AVX2(cpu_flags)) {
335 c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_avx2;
336 c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_avx2;
338 } else if (bit_depth == 10) {
339 if (EXTERNAL_SSSE3(cpu_flags)) {
340 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
341 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3;
343 if (EXTERNAL_SSE4(cpu_flags)) {
344 SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 10, sse4);
345 SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 10, sse4);
346 SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 10, sse4);
347 SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 10, sse4);
349 if (EXTERNAL_AVX(cpu_flags)) {
350 #if HAVE_AVX_EXTERNAL
351 SET_QPEL_FUNCS(0, 1, 10, avx, ff_hevc_qpel_h);
352 SET_QPEL_FUNCS(1, 0, 10, avx, ff_hevc_qpel_v);
353 SET_QPEL_FUNCS(1, 1, 10, avx, hevc_qpel_hv);
354 SET_EPEL_FUNCS(0, 1, 10, avx, ff_hevc_epel_h);
355 SET_EPEL_FUNCS(1, 0, 10, avx, ff_hevc_epel_v);
356 SET_EPEL_FUNCS(1, 1, 10, avx, hevc_epel_hv);
357 #endif /* HAVE_AVX_EXTERNAL */
359 if (EXTERNAL_AVX2(cpu_flags)) {
360 c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_avx2;
361 c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_avx2;
364 #endif /* ARCH_X86_64 */