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_DC_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_DC_FUNCS(4x4, mmxext);
74 IDCT_DC_FUNCS(8x8, mmxext);
75 IDCT_DC_FUNCS(8x8, sse2);
76 IDCT_DC_FUNCS(16x16, sse2);
77 IDCT_DC_FUNCS(32x32, sse2);
78 IDCT_DC_FUNCS(16x16, avx2);
79 IDCT_DC_FUNCS(32x32, avx2);
81 #define IDCT_FUNCS(opt) \
82 void ff_hevc_idct_4x4_8_ ## opt(int16_t *coeffs, int col_limit); \
83 void ff_hevc_idct_4x4_10_ ## opt(int16_t *coeffs, int col_limit); \
84 void ff_hevc_idct_8x8_8_ ## opt(int16_t *coeffs, int col_limit); \
85 void ff_hevc_idct_8x8_10_ ## opt(int16_t *coeffs, int col_limit); \
86 void ff_hevc_idct_16x16_8_ ## opt(int16_t *coeffs, int col_limit); \
87 void ff_hevc_idct_16x16_10_ ## opt(int16_t *coeffs, int col_limit); \
88 void ff_hevc_idct_32x32_8_ ## opt(int16_t *coeffs, int col_limit); \
89 void ff_hevc_idct_32x32_10_ ## opt(int16_t *coeffs, int col_limit);
94 void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride);
95 void ff_hevc_add_residual_8_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
96 void ff_hevc_add_residual_16_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
97 void ff_hevc_add_residual_32_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
99 void ff_hevc_add_residual_8_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
100 void ff_hevc_add_residual_16_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
101 void ff_hevc_add_residual_32_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
103 void ff_hevc_add_residual_32_8_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
105 void ff_hevc_add_residual_4_10_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride);
106 void ff_hevc_add_residual_8_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
107 void ff_hevc_add_residual_16_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
108 void ff_hevc_add_residual_32_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
110 void ff_hevc_add_residual_16_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
111 void ff_hevc_add_residual_32_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
113 #define GET_PIXELS(width, depth, cf) \
114 void ff_hevc_get_pixels_ ## width ## _ ## depth ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
115 uint8_t *src, ptrdiff_t srcstride, \
116 int height, int mx, int my, int16_t *mcbuffer);
118 GET_PIXELS(4, 8, sse2)
119 GET_PIXELS(8, 8, sse2)
120 GET_PIXELS(12, 8, sse2)
121 GET_PIXELS(16, 8, sse2)
122 GET_PIXELS(24, 8, sse2)
123 GET_PIXELS(32, 8, sse2)
124 GET_PIXELS(48, 8, sse2)
125 GET_PIXELS(64, 8, sse2)
127 GET_PIXELS(4, 10, sse2)
128 GET_PIXELS(8, 10, sse2)
129 GET_PIXELS(12, 10, sse2)
130 GET_PIXELS(16, 10, sse2)
131 GET_PIXELS(24, 10, sse2)
132 GET_PIXELS(32, 10, sse2)
133 GET_PIXELS(48, 10, sse2)
134 GET_PIXELS(64, 10, sse2)
136 /* those are independent of the bit depth, so declared separately */
137 #define INTERP_HV_FUNC(width, cf) \
138 void ff_hevc_qpel_hv_ ## width ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
139 int16_t *src, ptrdiff_t srcstride, \
140 int height, int mx, int my, int16_t *mcbuffer); \
141 void ff_hevc_epel_hv_ ## width ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
142 int16_t *src, ptrdiff_t srcstride, \
143 int height, int mx, int my, int16_t *mcbuffer);
145 INTERP_HV_FUNC(4, avx)
146 INTERP_HV_FUNC(8, avx)
147 INTERP_HV_FUNC(12, avx)
148 INTERP_HV_FUNC(16, avx)
149 INTERP_HV_FUNC(24, avx)
150 INTERP_HV_FUNC(32, avx)
151 INTERP_HV_FUNC(48, avx)
152 INTERP_HV_FUNC(64, avx)
154 #if ARCH_X86_64 && HAVE_AVX_EXTERNAL
155 #define QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv) \
156 static void hevc_qpel_hv_ ## width ## _ ## depth ## _ ## cf_hv(int16_t *dst, ptrdiff_t dststride, \
157 uint8_t *src, ptrdiff_t srcstride, \
158 int height, int mx, int my, int16_t *mcbuffer) \
160 const ptrdiff_t stride = FFALIGN(width + 7, 8); \
161 ff_hevc_qpel_h_ ## width ## _ ## depth ## _ ## cf_h(mcbuffer, 2 * stride, src - 3 * srcstride, srcstride, \
162 height + 7, mx, my, mcbuffer); \
163 ff_hevc_qpel_hv_ ## width ## _ ## cf_hv(dst, dststride, mcbuffer + 3 * stride, 2 * stride, \
164 height, mx, my, mcbuffer); \
167 #define QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
168 #endif /* ARCH_X86_64 && HAVE_AVX_EXTERNAL */
170 #define QPEL_FUNCS(width, depth, cf_h, cf_v, cf_hv) \
171 void ff_hevc_qpel_h_ ## width ## _ ## depth ## _ ## cf_h(int16_t *dst, ptrdiff_t dststride, \
172 uint8_t *src, ptrdiff_t srcstride, \
173 int height, int mx, int my, int16_t *mcbuffer); \
174 void ff_hevc_qpel_v_ ## width ## _ ## depth ## _ ## cf_v(int16_t *dst, ptrdiff_t dststride, \
175 uint8_t *src, ptrdiff_t srcstride, \
176 int height, int mx, int my, int16_t *mcbuffer); \
177 QPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
179 QPEL_FUNCS(4, 8, ssse3, ssse3, avx)
180 QPEL_FUNCS(8, 8, ssse3, ssse3, avx)
181 QPEL_FUNCS(12, 8, ssse3, ssse3, avx)
182 QPEL_FUNCS(16, 8, ssse3, ssse3, avx)
183 QPEL_FUNCS(24, 8, ssse3, ssse3, avx)
184 QPEL_FUNCS(32, 8, ssse3, ssse3, avx)
185 QPEL_FUNCS(48, 8, ssse3, ssse3, avx)
186 QPEL_FUNCS(64, 8, ssse3, ssse3, avx)
188 QPEL_FUNCS(4, 10, avx, avx, avx)
189 QPEL_FUNCS(8, 10, avx, avx, avx)
190 QPEL_FUNCS(12, 10, avx, avx, avx)
191 QPEL_FUNCS(16, 10, avx, avx, avx)
192 QPEL_FUNCS(24, 10, avx, avx, avx)
193 QPEL_FUNCS(32, 10, avx, avx, avx)
194 QPEL_FUNCS(48, 10, avx, avx, avx)
195 QPEL_FUNCS(64, 10, avx, avx, avx)
197 #if ARCH_X86_64 && HAVE_AVX_EXTERNAL
198 #define EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv) \
199 static void hevc_epel_hv_ ## width ## _ ## depth ## _ ## cf_hv(int16_t *dst, ptrdiff_t dststride, \
200 uint8_t *src, ptrdiff_t srcstride, \
201 int height, int mx, int my, int16_t *mcbuffer) \
203 const ptrdiff_t stride = FFALIGN(width + 3, 8); \
204 ff_hevc_epel_h_ ## width ## _ ## depth ## _ ## cf_h(mcbuffer, 2 * stride, src - srcstride, srcstride, \
205 height + 3, mx, my, mcbuffer); \
206 ff_hevc_epel_hv_ ## width ## _ ## cf_hv(dst, dststride, mcbuffer + stride, 2 * stride, \
207 height, mx, my, mcbuffer); \
210 #define EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
211 #endif /* ARCH_X86_64 && HAVE_AVX_EXTERNAL */
213 #define EPEL_FUNCS(width, depth, cf_h, cf_v, cf_hv) \
214 void ff_hevc_epel_h_ ## width ## _ ## depth ## _ ## cf_h(int16_t *dst, ptrdiff_t dststride, \
215 uint8_t *src, ptrdiff_t srcstride, \
216 int height, int mx, int my, int16_t *mcbuffer); \
217 void ff_hevc_epel_v_ ## width ## _ ## depth ## _ ## cf_v(int16_t *dst, ptrdiff_t dststride, \
218 uint8_t *src, ptrdiff_t srcstride, \
219 int height, int mx, int my, int16_t *mcbuffer); \
220 EPEL_FUNC_HV(width, depth, cf_h, cf_v, cf_hv)
222 EPEL_FUNCS(4, 8, ssse3, ssse3, avx)
223 EPEL_FUNCS(8, 8, ssse3, ssse3, avx)
224 EPEL_FUNCS(12, 8, ssse3, ssse3, avx)
225 EPEL_FUNCS(16, 8, ssse3, ssse3, avx)
226 EPEL_FUNCS(24, 8, ssse3, ssse3, avx)
227 EPEL_FUNCS(32, 8, ssse3, ssse3, avx)
229 EPEL_FUNCS(4, 10, avx, avx, avx)
230 EPEL_FUNCS(8, 10, avx, avx, avx)
231 EPEL_FUNCS(12, 10, avx, avx, avx)
232 EPEL_FUNCS(16, 10, avx, avx, avx)
233 EPEL_FUNCS(24, 10, avx, avx, avx)
234 EPEL_FUNCS(32, 10, avx, avx, avx)
236 #define PUT_PRED(width, depth, cf_uw, cf_w) \
237 void ff_hevc_put_unweighted_pred_ ## width ## _ ## depth ## _ ## cf_uw(uint8_t *dst, ptrdiff_t dststride, \
238 int16_t *src, ptrdiff_t srcstride, \
240 void ff_hevc_put_unweighted_pred_avg_ ## width ## _ ## depth ## _ ## cf_uw(uint8_t *dst, ptrdiff_t dststride, \
241 int16_t *src1, int16_t *src2, \
242 ptrdiff_t srcstride, int height); \
243 void ff_hevc_put_weighted_pred_ ## width ## _ ## depth ## _ ## cf_w(uint8_t denom, int16_t weight, int16_t offset, \
244 uint8_t *dst, ptrdiff_t dststride, \
245 int16_t *src, ptrdiff_t srcstride, \
247 void ff_hevc_put_weighted_pred_avg_ ## width ## _ ## depth ## _ ## cf_w(uint8_t denom, int16_t weight0, int16_t weight1, \
248 int16_t offset0, int16_t offset1, \
249 uint8_t *dst, ptrdiff_t dststride, \
250 int16_t *src0, int16_t *src1, ptrdiff_t srcstride, \
253 PUT_PRED(4, 8, sse2, sse4)
254 PUT_PRED(8, 8, sse2, sse4)
255 PUT_PRED(12, 8, sse2, sse4)
256 PUT_PRED(16, 8, sse2, sse4)
257 PUT_PRED(24, 8, sse2, sse4)
258 PUT_PRED(32, 8, sse2, sse4)
259 PUT_PRED(48, 8, sse2, sse4)
260 PUT_PRED(64, 8, sse2, sse4)
262 PUT_PRED(4, 10, sse2, sse4)
263 PUT_PRED(8, 10, sse2, sse4)
264 PUT_PRED(12, 10, sse2, sse4)
265 PUT_PRED(16, 10, sse2, sse4)
266 PUT_PRED(24, 10, sse2, sse4)
267 PUT_PRED(32, 10, sse2, sse4)
268 PUT_PRED(48, 10, sse2, sse4)
269 PUT_PRED(64, 10, sse2, sse4)
271 void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
273 int cpu_flags = av_get_cpu_flags();
275 #define SET_LUMA_FUNCS(tabname, funcname, depth, cf) \
276 c->tabname[0] = funcname ## _4_ ## depth ## _ ## cf; \
277 c->tabname[1] = funcname ## _8_ ## depth ## _ ## cf; \
278 c->tabname[2] = funcname ## _12_ ## depth ## _ ## cf; \
279 c->tabname[3] = funcname ## _16_ ## depth ## _ ## cf; \
280 c->tabname[4] = funcname ## _24_ ## depth ## _ ## cf; \
281 c->tabname[5] = funcname ## _32_ ## depth ## _ ## cf; \
282 c->tabname[6] = funcname ## _48_ ## depth ## _ ## cf; \
283 c->tabname[7] = funcname ## _64_ ## depth ## _ ## cf;
285 #define SET_CHROMA_FUNCS(tabname, funcname, depth, cf) \
286 c->tabname[1] = funcname ## _4_ ## depth ## _ ## cf; \
287 c->tabname[3] = funcname ## _8_ ## depth ## _ ## cf; \
288 c->tabname[4] = funcname ## _12_ ## depth ## _ ## cf; \
289 c->tabname[5] = funcname ## _16_ ## depth ## _ ## cf; \
290 c->tabname[6] = funcname ## _24_ ## depth ## _ ## cf; \
291 c->tabname[7] = funcname ## _32_ ## depth ## _ ## cf;
293 #define SET_QPEL_FUNCS(v, h, depth, cf, name) SET_LUMA_FUNCS (put_hevc_qpel[v][h], name, depth, cf)
294 #define SET_EPEL_FUNCS(v, h, depth, cf, name) SET_CHROMA_FUNCS(put_hevc_epel[v][h], name, depth, cf)
296 if (bit_depth == 8) {
297 if (EXTERNAL_MMXEXT(cpu_flags)) {
298 c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_mmxext;
299 c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_mmxext;
301 c->add_residual[0] = ff_hevc_add_residual_4_8_mmxext;
303 if (EXTERNAL_SSE2(cpu_flags)) {
304 c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
305 c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
307 c->add_residual[1] = ff_hevc_add_residual_8_8_sse2;
308 c->add_residual[2] = ff_hevc_add_residual_16_8_sse2;
309 c->add_residual[3] = ff_hevc_add_residual_32_8_sse2;
311 c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_sse2;
312 c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_sse2;
313 c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_sse2;
315 c->idct[0] = ff_hevc_idct_4x4_8_sse2;
316 c->idct[1] = ff_hevc_idct_8x8_8_sse2;
318 SET_QPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
319 SET_EPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
321 SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 8, sse2);
322 SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 8, sse2);
323 SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 8, sse2);
324 SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 8, sse2);
326 if (EXTERNAL_SSSE3(cpu_flags)) {
327 SET_QPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_qpel_h);
328 SET_QPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_qpel_v);
329 SET_EPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_epel_h);
330 SET_EPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_epel_v);
333 if (EXTERNAL_AVX(cpu_flags)) {
334 c->idct[0] = ff_hevc_idct_4x4_8_avx;
335 c->idct[1] = ff_hevc_idct_8x8_8_avx;
336 c->add_residual[1] = ff_hevc_add_residual_8_8_avx;
337 c->add_residual[2] = ff_hevc_add_residual_16_8_avx;
338 c->add_residual[3] = ff_hevc_add_residual_32_8_avx;
340 if (EXTERNAL_AVX2(cpu_flags)) {
341 c->add_residual[3] = ff_hevc_add_residual_32_8_avx2;
343 } else if (bit_depth == 10) {
344 if (EXTERNAL_MMXEXT(cpu_flags)) {
345 c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
346 c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
348 c->add_residual[0] = ff_hevc_add_residual_4_10_mmxext;
350 if (EXTERNAL_SSE2(cpu_flags)) {
351 c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
352 c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
354 c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_sse2;
355 c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_sse2;
356 c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_sse2;
358 c->idct[0] = ff_hevc_idct_4x4_10_sse2;
359 c->idct[1] = ff_hevc_idct_8x8_10_sse2;
360 SET_QPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
361 SET_EPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
363 SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 10, sse2);
364 SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 10, sse2);
365 SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 10, sse2);
366 SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 10, sse2);
368 c->add_residual[1] = ff_hevc_add_residual_8_10_sse2;
369 c->add_residual[2] = ff_hevc_add_residual_16_10_sse2;
370 c->add_residual[3] = ff_hevc_add_residual_32_10_sse2;
372 if (EXTERNAL_AVX(cpu_flags)) {
373 c->idct[0] = ff_hevc_idct_4x4_10_avx;
374 c->idct[1] = ff_hevc_idct_8x8_10_avx;
376 if (EXTERNAL_AVX2(cpu_flags)) {
377 c->add_residual[2] = ff_hevc_add_residual_16_10_avx2;
378 c->add_residual[3] = ff_hevc_add_residual_32_10_avx2;
383 if (bit_depth == 8) {
384 if (EXTERNAL_SSE2(cpu_flags)) {
385 c->idct[2] = ff_hevc_idct_16x16_8_sse2;
386 c->idct[3] = ff_hevc_idct_32x32_8_sse2;
388 if (EXTERNAL_SSSE3(cpu_flags)) {
389 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
390 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
393 if (EXTERNAL_SSE4(cpu_flags)) {
394 SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 8, sse4);
395 SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 8, sse4);
396 SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 8, sse4);
397 SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 8, sse4);
400 if (EXTERNAL_AVX(cpu_flags)) {
401 #if HAVE_AVX_EXTERNAL
402 SET_QPEL_FUNCS(1, 1, 8, avx, hevc_qpel_hv);
403 SET_EPEL_FUNCS(1, 1, 8, avx, hevc_epel_hv);
404 #endif /* HAVE_AVX_EXTERNAL */
405 c->idct[2] = ff_hevc_idct_16x16_8_avx;
406 c->idct[3] = ff_hevc_idct_32x32_8_avx;
408 if (EXTERNAL_AVX2(cpu_flags)) {
409 c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_avx2;
410 c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_avx2;
412 } else if (bit_depth == 10) {
413 if (EXTERNAL_SSE2(cpu_flags)) {
414 c->idct[2] = ff_hevc_idct_16x16_10_sse2;
415 c->idct[3] = ff_hevc_idct_32x32_10_sse2;
417 if (EXTERNAL_SSSE3(cpu_flags)) {
418 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
419 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3;
421 if (EXTERNAL_SSE4(cpu_flags)) {
422 SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 10, sse4);
423 SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 10, sse4);
424 SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 10, sse4);
425 SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 10, sse4);
427 if (EXTERNAL_AVX(cpu_flags)) {
428 #if HAVE_AVX_EXTERNAL
429 SET_QPEL_FUNCS(0, 1, 10, avx, ff_hevc_qpel_h);
430 SET_QPEL_FUNCS(1, 0, 10, avx, ff_hevc_qpel_v);
431 SET_QPEL_FUNCS(1, 1, 10, avx, hevc_qpel_hv);
432 SET_EPEL_FUNCS(0, 1, 10, avx, ff_hevc_epel_h);
433 SET_EPEL_FUNCS(1, 0, 10, avx, ff_hevc_epel_v);
434 SET_EPEL_FUNCS(1, 1, 10, avx, hevc_epel_hv);
435 #endif /* HAVE_AVX_EXTERNAL */
436 c->idct[2] = ff_hevc_idct_16x16_10_avx;
437 c->idct[3] = ff_hevc_idct_32x32_10_avx;
439 if (EXTERNAL_AVX2(cpu_flags)) {
440 c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_avx2;
441 c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_avx2;
444 #endif /* ARCH_X86_64 */