1;****************************************************************************** 2;* software YUV to RGB converter 3;* 4;* Copyright (C) 2001-2007 Michael Niedermayer 5;* (c) 2010 Konstantin Shishkov 6;* 7;* This file is part of FFmpeg. 8;* 9;* FFmpeg is free software; you can redistribute it and/or 10;* modify it under the terms of the GNU Lesser General Public 11;* License as published by the Free Software Foundation; either 12;* version 2.1 of the License, or (at your option) any later version. 13;* 14;* FFmpeg is distributed in the hope that it will be useful, 15;* but WITHOUT ANY WARRANTY; without even the implied warranty of 16;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 17;* Lesser General Public License for more details. 18;* 19;* You should have received a copy of the GNU Lesser General Public 20;* License along with FFmpeg; if not, write to the Free Software 21;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 22;****************************************************************************** 23 24%include "libavutil/x86/x86util.asm" 25 26SECTION_RODATA 27 28; below variables are named like mask_dwXY, which means to preserve dword No.X & No.Y 29mask_dw036 : db -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0 30mask_dw147 : db 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1 31mask_dw25 : db 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0 32rgb24_shuf1: db 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15, 4, 5, 10, 11 33rgb24_shuf2: db 10, 11, 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15, 4, 5 34rgb24_shuf3: db 4, 5, 10, 11, 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15 35pw_00ff: times 8 dw 255 36pb_f8: times 16 db 248 37pb_e0: times 16 db 224 38pb_03: times 16 db 3 39pb_07: times 16 db 7 40 41mask_1101: dw -1, -1, 0, -1 42mask_0010: dw 0, 0, -1, 0 43mask_0110: dw 0, -1, -1, 0 44mask_1001: dw -1, 0, 0, -1 45mask_0100: dw 0, -1, 0, 0 46 47SECTION .text 48 49;----------------------------------------------------------------------------- 50; 51; YUV420/YUVA420 to RGB/BGR 15/16/24/32 52; R = Y + ((vrCoff * (v - 128)) >> 8) 53; G = Y - ((ugCoff * (u - 128) + vgCoff * (v - 128)) >> 8) 54; B = Y + ((ubCoff * (u - 128)) >> 8) 55; 56;----------------------------------------------------------------------------- 57 58%macro MOV_H2L 1 59%if mmsize == 8 60 psrlq %1, 32 61%else ; mmsize == 16 62 psrldq %1, 8 63%endif 64%endmacro 65 66%macro yuv2rgb_fn 3 67 68%if %3 == 32 69 %ifidn %1, yuva 70 %define parameters index, image, pu_index, pv_index, pointer_c_dither, py_2index, pa_2index 71 %define GPR_num 7 72 %else 73 %define parameters index, image, pu_index, pv_index, pointer_c_dither, py_2index 74 %define GPR_num 6 75 %endif 76%else 77 %define parameters index, image, pu_index, pv_index, pointer_c_dither, py_2index 78 %define GPR_num 6 79%endif 80 81%define m_green m2 82%define m_alpha m3 83%define m_y m6 84%define m_u m0 85%define m_v m1 86%ifidn %2, rgb 87%define m_red m1 88%define m_blue m0 89%else 90%define m_red m0 91%define m_blue m1 92%endif 93 94%if mmsize == 8 95%define time_num 1 96%define reg_num 8 97%define y_offset [pointer_c_ditherq + 8 * 8] 98%define u_offset [pointer_c_ditherq + 9 * 8] 99%define v_offset [pointer_c_ditherq + 10 * 8] 100%define ug_coff [pointer_c_ditherq + 7 * 8] 101%define vg_coff [pointer_c_ditherq + 6 * 8] 102%define y_coff [pointer_c_ditherq + 3 * 8] 103%define ub_coff [pointer_c_ditherq + 5 * 8] 104%define vr_coff [pointer_c_ditherq + 4 * 8] 105%elif mmsize == 16 106%define time_num 2 107%if ARCH_X86_32 108%define reg_num 8 109%define my_offset [pointer_c_ditherq + 8 * 8] 110%define mu_offset [pointer_c_ditherq + 9 * 8] 111%define mv_offset [pointer_c_ditherq + 10 * 8] 112%define mug_coff [pointer_c_ditherq + 7 * 8] 113%define mvg_coff [pointer_c_ditherq + 6 * 8] 114%define my_coff [pointer_c_ditherq + 3 * 8] 115%define mub_coff [pointer_c_ditherq + 5 * 8] 116%define mvr_coff [pointer_c_ditherq + 4 * 8] 117%else ; ARCH_X86_64 118%define reg_num 16 119%define y_offset m8 120%define u_offset m9 121%define v_offset m10 122%define ug_coff m11 123%define vg_coff m12 124%define y_coff m13 125%define ub_coff m14 126%define vr_coff m15 127%endif ; ARCH_X86_32/64 128%endif ; coeff define mmsize == 8/16 129 130cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters 131 132%if ARCH_X86_64 133 movsxd indexq, indexd 134%if mmsize == 16 135 VBROADCASTSD y_offset, [pointer_c_ditherq + 8 * 8] 136 VBROADCASTSD u_offset, [pointer_c_ditherq + 9 * 8] 137 VBROADCASTSD v_offset, [pointer_c_ditherq + 10 * 8] 138 VBROADCASTSD ug_coff, [pointer_c_ditherq + 7 * 8] 139 VBROADCASTSD vg_coff, [pointer_c_ditherq + 6 * 8] 140 VBROADCASTSD y_coff, [pointer_c_ditherq + 3 * 8] 141 VBROADCASTSD ub_coff, [pointer_c_ditherq + 5 * 8] 142 VBROADCASTSD vr_coff, [pointer_c_ditherq + 4 * 8] 143%endif 144%endif 145.loop0: 146 movu m_y, [py_2indexq + 2 * indexq] 147 movh m_u, [pu_indexq + indexq] 148 movh m_v, [pv_indexq + indexq] 149 pxor m4, m4 150 mova m7, m6 151 punpcklbw m0, m4 152 punpcklbw m1, m4 153 mova m2, [pw_00ff] 154 pand m6, m2 155 psrlw m7, 8 156 psllw m0, 3 157 psllw m1, 3 158 psllw m6, 3 159 psllw m7, 3 160%if (ARCH_X86_32 && mmsize == 16) 161 VBROADCASTSD m2, mu_offset 162 VBROADCASTSD m3, mv_offset 163 VBROADCASTSD m4, my_offset 164 psubsw m0, m2 ; U = U - 128 165 psubsw m1, m3 ; V = V - 128 166 psubw m6, m4 167 psubw m7, m4 168 VBROADCASTSD m2, mug_coff 169 VBROADCASTSD m3, mvg_coff 170 VBROADCASTSD m4, my_coff 171 VBROADCASTSD m5, mub_coff 172 pmulhw m2, m0 173 pmulhw m3, m1 174 pmulhw m6, m4 175 pmulhw m7, m4 176 pmulhw m0, m5 177 VBROADCASTSD m4, mvr_coff 178 pmulhw m1, m4 179%else ; ARCH_X86_64 || mmsize == 8 180 psubsw m0, u_offset ; U = U - 128 181 psubsw m1, v_offset ; V = V - 128 182 psubw m6, y_offset 183 psubw m7, y_offset 184 mova m2, m0 185 mova m3, m1 186 pmulhw m2, ug_coff 187 pmulhw m3, vg_coff 188 pmulhw m6, y_coff 189 pmulhw m7, y_coff 190 pmulhw m0, ub_coff 191 pmulhw m1, vr_coff 192%endif 193 paddsw m2, m3 194 mova m3, m7 195 mova m5, m7 196 paddsw m3, m0 ; B1 B3 B5 B7 ... 197 paddsw m5, m1 ; R1 R3 R5 R7 ... 198 paddsw m7, m2 ; G1 G3 G4 G7 ... 199 paddsw m0, m6 ; B0 B2 B4 B6 ... 200 paddsw m1, m6 ; R0 R2 R4 R6 ... 201 paddsw m2, m6 ; G0 G2 G4 G6 ... 202 203%if %3 == 24 ; PACK RGB24 204%define depth 3 205 packuswb m0, m3 ; R0 R2 R4 R6 ... R1 R3 R5 R7 ... 206 packuswb m1, m5 ; B0 B2 B4 B6 ... B1 B3 B5 B7 ... 207 packuswb m2, m7 ; G0 G2 G4 G6 ... G1 G3 G5 G7 ... 208 mova m3, m_red 209 mova m6, m_blue 210 MOV_H2L m_red 211 punpcklbw m3, m2 ; R0 G0 R2 G2 R4 G4 R6 G6 R8 G8 ... 212 punpcklbw m6, m_red ; B0 R1 B2 R3 B4 R5 B6 R7 B8 R9 ... 213 mova m5, m3 214 punpckhbw m2, m_blue ; G1 B1 G3 B3 G5 B5 G7 B7 G9 B9 ... 215%if mmsize == 8 216 punpcklwd m3 ,m6 ; R0 G0 B0 R1 R2 G2 B2 R3 217 punpckhwd m5, m6 ; R4 G4 B4 R5 R6 G6 B6 R7 218%if cpuflag(mmxext) 219 pshufw m1, m2, 0xc6 220 pshufw m6, m3, 0x84 221 pshufw m7, m5, 0x38 222 pand m6, [mask_1101] ; R0 G0 B0 R1 -- -- R2 G2 223 movq m0, m1 224 pand m7, [mask_0110] ; -- -- R6 G6 B6 R7 -- -- 225 movq m2, m1 226 pand m1, [mask_0100] ; -- -- G3 B3 -- -- -- -- 227 psrlq m3, 48 ; B2 R3 -- -- -- -- -- -- 228 pand m0, [mask_0010] ; -- -- -- -- G1 B1 -- -- 229 psllq m5, 32 ; -- -- -- -- R4 G4 B4 R5 230 pand m2, [mask_1001] ; G5 B5 -- -- -- -- G7 B7 231 por m1, m3 232 por m0, m6 233 por m1, m5 234 por m2, m7 235 movntq [imageq], m0 236 movntq [imageq + 8], m1 237 movntq [imageq + 16], m2 238%else ; cpuflag(mmx) 239 movd [imageq], m3 ; R0 G0 R2 G2 240 movd [imageq + 4], m2 ; G1 B1 241 psrlq m3, 32 242 psrlq m2, 16 243 movd [imageq + 6], m3 ; R2 G2 B2 R3 244 movd [imageq + 10], m2 ; G3 B3 245 psrlq m2, 16 246 movd [imageq + 12], m5 ; R4 G4 B4 R5 247 movd [imageq + 16], m2 ; G5 B5 248 psrlq m5, 32 249 movd [imageq + 20], m2 ; -- -- G7 B7 250 movd [imageq + 18], m5 ; R6 G6 B6 R7 251%endif ; mmsize = 8 252%else ; mmsize == 16 253 pshufb m3, [rgb24_shuf1] ; r0 g0 r6 g6 r12 g12 r2 g2 r8 g8 r14 g14 r4 g4 r10 g10 254 pshufb m6, [rgb24_shuf2] ; b10 r11 b0 r1 b6 r7 b12 r13 b2 r3 b8 r9 b14 r15 b4 r5 255 pshufb m2, [rgb24_shuf3] ; g5 b5 g11 b11 g1 b1 g7 b7 g13 b13 g3 b3 g9 b9 g15 b15 256 mova m7, [mask_dw036] 257 mova m4, [mask_dw147] 258 mova m5, [mask_dw25] 259 pand m0, m7, m3 ; r0 g0 --- --- --- --- r2 g2 --- --- --- --- r4 g4 --- --- 260 pand m1, m4, m6 ; --- --- b0 r1 --- --- --- --- b2 r3 --- --- --- --- b4 r5 261 por m0, m1 262 pand m1, m5, m2 ; --- --- --- --- g1 b1 --- --- --- --- g3 b3 --- --- --- --- 263 por m0, m1 ; r0 g0 b0 r1 g1 b1 r2 g2 b2 r3 g3 b3 r4 g4 b4 r5 264 pand m1, m7, m2 ; g5 b5 --- --- --- --- g7 b7 --- --- --- --- g9 b9 --- --- 265 pand m7, m6 ; b10 r11 --- --- --- --- b12 r13 --- --- --- --- b14 r15 --- --- 266 pand m6, m5 ; --- --- --- --- b6 r7 --- --- --- --- b8 r9 --- --- --- --- 267 por m1, m6 268 pand m6, m4, m3 ; --- --- r6 g6 --- --- --- --- r8 g8 --- --- --- --- r10 g10 269 pand m2, m4 ; --- --- g11 b11 --- --- --- --- g13 b13 --- --- --- --- g15 b15 270 pand m3, m5 ; --- --- --- --- r12 g12 --- --- --- --- r14 g14 --- --- --- --- 271 por m2, m7 272 por m1, m6 ; g5 b5 r6 g6 b6 r7 g7 b7 r8 g8 b8 r9 g9 b9 r10 g10 273 por m2, m3 ; b10 r11 g11 b11 r12 g12 b12 r13 g13 b13 r14 g14 b14 r15 g15 b15 274 movu [imageq], m0 275 movu [imageq + 16], m1 276 movu [imageq + 32], m2 277%endif ; mmsize = 16 278%else ; PACK RGB15/16/32 279 packuswb m0, m1 280 packuswb m3, m5 281 packuswb m2, m2 282 mova m1, m0 283 packuswb m7, m7 284 punpcklbw m0, m3 ; B0 B1 B2 B3 ... B7 285 punpckhbw m1, m3 ; R0 R1 R2 R3 ... R7 286 punpcklbw m2, m7 ; G0 G1 G2 G3 ... G7 287%if %3 == 32 ; PACK RGB32 288%define depth 4 289%ifidn %1, yuv 290 pcmpeqd m3, m3 ; Set alpha empty 291%else 292 movu m3, [pa_2indexq + 2 * indexq] ; Load alpha 293%endif 294 mova m5, m_blue 295 mova m6, m_red 296 punpckhbw m5, m_green 297 punpcklbw m_blue, m_green 298 punpckhbw m6, m_alpha 299 punpcklbw m_red, m_alpha 300 mova m_green, m_blue 301 mova m_alpha, m5 302 punpcklwd m_blue, m_red 303 punpckhwd m_green, m_red 304 punpcklwd m5, m6 305 punpckhwd m_alpha, m6 306 movu [imageq + 0], m_blue 307 movu [imageq + 8 * time_num], m_green 308 movu [imageq + 16 * time_num], m5 309 movu [imageq + 24 * time_num], m_alpha 310%else ; PACK RGB15/16 311%define depth 2 312%if cpuflag(ssse3) 313 %define red_dither m3 314 %define green_dither m4 315 %define blue_dither m5 316 VBROADCASTSD red_dither, [pointer_c_ditherq + 0 * 8] 317 VBROADCASTSD green_dither, [pointer_c_ditherq + 1 * 8] 318 VBROADCASTSD blue_dither, [pointer_c_ditherq + 2 * 8] 319%else ; cpuflag(mmx/mmxext) 320%define blue_dither [pointer_c_ditherq + 2 * 8] 321%define green_dither [pointer_c_ditherq + 1 * 8] 322%define red_dither [pointer_c_ditherq + 0 * 8] 323%endif 324%if %3 == 15 325%define gmask pb_03 326%define isRGB15 1 327%else 328%define gmask pb_07 329%define isRGB15 0 330%endif 331 paddusb m0, blue_dither 332 paddusb m2, green_dither 333 paddusb m1, red_dither 334 pand m0, [pb_f8] 335 pand m1, [pb_f8] 336 mova m3, m2 337 psllw m2, 3 - isRGB15 338 psrlw m3, 5 + isRGB15 339 psrlw m0, 3 340 psrlw m1, isRGB15 341 pand m2, [pb_e0] 342 pand m3, [gmask] 343 por m0, m2 344 por m1, m3 345 mova m2, m0 346 punpcklbw m0, m1 347 punpckhbw m2, m1 348 movu [imageq], m0 349 movu [imageq + 8 * time_num], m2 350%endif ; PACK RGB15/16 351%endif ; PACK RGB15/16/32 352 353add imageq, 8 * depth * time_num 354add indexq, 4 * time_num 355js .loop0 356 357REP_RET 358 359%endmacro 360 361INIT_MMX mmx 362yuv2rgb_fn yuv, rgb, 32 363yuv2rgb_fn yuv, bgr, 32 364yuv2rgb_fn yuva, rgb, 32 365yuv2rgb_fn yuva, bgr, 32 366yuv2rgb_fn yuv, rgb, 15 367yuv2rgb_fn yuv, rgb, 16 368 369INIT_MMX mmxext 370yuv2rgb_fn yuv, rgb, 24 371yuv2rgb_fn yuv, bgr, 24 372 373INIT_XMM ssse3 374yuv2rgb_fn yuv, rgb, 24 375yuv2rgb_fn yuv, bgr, 24 376yuv2rgb_fn yuv, rgb, 32 377yuv2rgb_fn yuv, bgr, 32 378yuv2rgb_fn yuva, rgb, 32 379yuv2rgb_fn yuva, bgr, 32 380yuv2rgb_fn yuv, rgb, 15 381yuv2rgb_fn yuv, rgb, 16 382