| 1 | /* |
| 2 | * AltiVec acceleration for colorspace conversion |
| 3 | * |
| 4 | * copyright (C) 2004 Marc Hoffman <marc.hoffman@analog.com> |
| 5 | * |
| 6 | * This file is part of FFmpeg. |
| 7 | * |
| 8 | * FFmpeg is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; either |
| 11 | * version 2.1 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * FFmpeg is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with FFmpeg; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | /* |
| 24 | * Convert I420 YV12 to RGB in various formats, |
| 25 | * it rejects images that are not in 420 formats, |
| 26 | * it rejects images that don't have widths of multiples of 16, |
| 27 | * it rejects images that don't have heights of multiples of 2. |
| 28 | * Reject defers to C simulation code. |
| 29 | * |
| 30 | * Lots of optimizations to be done here. |
| 31 | * |
| 32 | * 1. Need to fix saturation code. I just couldn't get it to fly with packs |
| 33 | * and adds, so we currently use max/min to clip. |
| 34 | * |
| 35 | * 2. The inefficient use of chroma loading needs a bit of brushing up. |
| 36 | * |
| 37 | * 3. Analysis of pipeline stalls needs to be done. Use shark to identify |
| 38 | * pipeline stalls. |
| 39 | * |
| 40 | * |
| 41 | * MODIFIED to calculate coeffs from currently selected color space. |
| 42 | * MODIFIED core to be a macro where you specify the output format. |
| 43 | * ADDED UYVY conversion which is never called due to some thing in swscale. |
| 44 | * CORRECTED algorithim selection to be strict on input formats. |
| 45 | * ADDED runtime detection of AltiVec. |
| 46 | * |
| 47 | * ADDED altivec_yuv2packedX vertical scl + RGB converter |
| 48 | * |
| 49 | * March 27,2004 |
| 50 | * PERFORMANCE ANALYSIS |
| 51 | * |
| 52 | * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo |
| 53 | * used as test. |
| 54 | * The AltiVec version uses 10% of the processor or ~100Mips for D1 video |
| 55 | * same sequence. |
| 56 | * |
| 57 | * 720 * 480 * 30 ~10MPS |
| 58 | * |
| 59 | * so we have roughly 10 clocks per pixel. This is too high, something has |
| 60 | * to be wrong. |
| 61 | * |
| 62 | * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the |
| 63 | * need for vec_min. |
| 64 | * |
| 65 | * OPTIMIZED DST OUTPUT cache/DMA controls. We are pretty much guaranteed to |
| 66 | * have the input video frame, it was just decompressed so it probably resides |
| 67 | * in L1 caches. However, we are creating the output video stream. This needs |
| 68 | * to use the DSTST instruction to optimize for the cache. We couple this with |
| 69 | * the fact that we are not going to be visiting the input buffer again so we |
| 70 | * mark it Least Recently Used. This shaves 25% of the processor cycles off. |
| 71 | * |
| 72 | * Now memcpy is the largest mips consumer in the system, probably due |
| 73 | * to the inefficient X11 stuff. |
| 74 | * |
| 75 | * GL libraries seem to be very slow on this machine 1.33Ghz PB running |
| 76 | * Jaguar, this is not the case for my 1Ghz PB. I thought it might be |
| 77 | * a versioning issue, however I have libGL.1.2.dylib for both |
| 78 | * machines. (We need to figure this out now.) |
| 79 | * |
| 80 | * GL2 libraries work now with patch for RGB32. |
| 81 | * |
| 82 | * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor. |
| 83 | * |
| 84 | * Integrated luma prescaling adjustment for saturation/contrast/brightness |
| 85 | * adjustment. |
| 86 | */ |
| 87 | |
| 88 | #include <stdio.h> |
| 89 | #include <stdlib.h> |
| 90 | #include <string.h> |
| 91 | #include <inttypes.h> |
| 92 | #include <assert.h> |
| 93 | |
| 94 | #include "config.h" |
| 95 | #include "libswscale/rgb2rgb.h" |
| 96 | #include "libswscale/swscale.h" |
| 97 | #include "libswscale/swscale_internal.h" |
| 98 | #include "libavutil/attributes.h" |
| 99 | #include "libavutil/cpu.h" |
| 100 | #include "libavutil/pixdesc.h" |
| 101 | #include "yuv2rgb_altivec.h" |
| 102 | |
| 103 | #if HAVE_ALTIVEC |
| 104 | |
| 105 | #undef PROFILE_THE_BEAST |
| 106 | #undef INC_SCALING |
| 107 | |
| 108 | typedef unsigned char ubyte; |
| 109 | typedef signed char sbyte; |
| 110 | |
| 111 | /* RGB interleaver, 16 planar pels 8-bit samples per channel in |
| 112 | * homogeneous vector registers x0,x1,x2 are interleaved with the |
| 113 | * following technique: |
| 114 | * |
| 115 | * o0 = vec_mergeh(x0, x1); |
| 116 | * o1 = vec_perm(o0, x2, perm_rgb_0); |
| 117 | * o2 = vec_perm(o0, x2, perm_rgb_1); |
| 118 | * o3 = vec_mergel(x0, x1); |
| 119 | * o4 = vec_perm(o3, o2, perm_rgb_2); |
| 120 | * o5 = vec_perm(o3, o2, perm_rgb_3); |
| 121 | * |
| 122 | * perm_rgb_0: o0(RG).h v1(B) --> o1* |
| 123 | * 0 1 2 3 4 |
| 124 | * rgbr|gbrg|brgb|rgbr |
| 125 | * 0010 0100 1001 0010 |
| 126 | * 0102 3145 2673 894A |
| 127 | * |
| 128 | * perm_rgb_1: o0(RG).h v1(B) --> o2 |
| 129 | * 0 1 2 3 4 |
| 130 | * gbrg|brgb|bbbb|bbbb |
| 131 | * 0100 1001 1111 1111 |
| 132 | * B5CD 6EF7 89AB CDEF |
| 133 | * |
| 134 | * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4* |
| 135 | * 0 1 2 3 4 |
| 136 | * gbrg|brgb|rgbr|gbrg |
| 137 | * 1111 1111 0010 0100 |
| 138 | * 89AB CDEF 0182 3945 |
| 139 | * |
| 140 | * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5* |
| 141 | * 0 1 2 3 4 |
| 142 | * brgb|rgbr|gbrg|brgb |
| 143 | * 1001 0010 0100 1001 |
| 144 | * a67b 89cA BdCD eEFf |
| 145 | * |
| 146 | */ |
| 147 | static const vector unsigned char |
| 148 | perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05, |
| 149 | 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a }, |
| 150 | perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17, |
| 151 | 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f }, |
| 152 | perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, |
| 153 | 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 }, |
| 154 | perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a, |
| 155 | 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f }; |
| 156 | |
| 157 | #define vec_merge3(x2, x1, x0, y0, y1, y2) \ |
| 158 | do { \ |
| 159 | __typeof__(x0) o0, o2, o3; \ |
| 160 | o0 = vec_mergeh(x0, x1); \ |
| 161 | y0 = vec_perm(o0, x2, perm_rgb_0); \ |
| 162 | o2 = vec_perm(o0, x2, perm_rgb_1); \ |
| 163 | o3 = vec_mergel(x0, x1); \ |
| 164 | y1 = vec_perm(o3, o2, perm_rgb_2); \ |
| 165 | y2 = vec_perm(o3, o2, perm_rgb_3); \ |
| 166 | } while (0) |
| 167 | |
| 168 | #define vec_mstbgr24(x0, x1, x2, ptr) \ |
| 169 | do { \ |
| 170 | __typeof__(x0) _0, _1, _2; \ |
| 171 | vec_merge3(x0, x1, x2, _0, _1, _2); \ |
| 172 | vec_st(_0, 0, ptr++); \ |
| 173 | vec_st(_1, 0, ptr++); \ |
| 174 | vec_st(_2, 0, ptr++); \ |
| 175 | } while (0) |
| 176 | |
| 177 | #define vec_mstrgb24(x0, x1, x2, ptr) \ |
| 178 | do { \ |
| 179 | __typeof__(x0) _0, _1, _2; \ |
| 180 | vec_merge3(x2, x1, x0, _0, _1, _2); \ |
| 181 | vec_st(_0, 0, ptr++); \ |
| 182 | vec_st(_1, 0, ptr++); \ |
| 183 | vec_st(_2, 0, ptr++); \ |
| 184 | } while (0) |
| 185 | |
| 186 | /* pack the pixels in rgb0 format |
| 187 | * msb R |
| 188 | * lsb 0 |
| 189 | */ |
| 190 | #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \ |
| 191 | do { \ |
| 192 | T _0, _1, _2, _3; \ |
| 193 | _0 = vec_mergeh(x0, x1); \ |
| 194 | _1 = vec_mergeh(x2, x3); \ |
| 195 | _2 = (T) vec_mergeh((vector unsigned short) _0, \ |
| 196 | (vector unsigned short) _1); \ |
| 197 | _3 = (T) vec_mergel((vector unsigned short) _0, \ |
| 198 | (vector unsigned short) _1); \ |
| 199 | vec_st(_2, 0 * 16, (T *) ptr); \ |
| 200 | vec_st(_3, 1 * 16, (T *) ptr); \ |
| 201 | _0 = vec_mergel(x0, x1); \ |
| 202 | _1 = vec_mergel(x2, x3); \ |
| 203 | _2 = (T) vec_mergeh((vector unsigned short) _0, \ |
| 204 | (vector unsigned short) _1); \ |
| 205 | _3 = (T) vec_mergel((vector unsigned short) _0, \ |
| 206 | (vector unsigned short) _1); \ |
| 207 | vec_st(_2, 2 * 16, (T *) ptr); \ |
| 208 | vec_st(_3, 3 * 16, (T *) ptr); \ |
| 209 | ptr += 4; \ |
| 210 | } while (0) |
| 211 | |
| 212 | /* |
| 213 | * 1 0 1.4021 | | Y | |
| 214 | * 1 -0.3441 -0.7142 |x| Cb| |
| 215 | * 1 1.7718 0 | | Cr| |
| 216 | * |
| 217 | * |
| 218 | * Y: [-128 127] |
| 219 | * Cb/Cr : [-128 127] |
| 220 | * |
| 221 | * typical YUV conversion works on Y: 0-255 this version has been |
| 222 | * optimized for JPEG decoding. |
| 223 | */ |
| 224 | |
| 225 | #define vec_unh(x) \ |
| 226 | (vector signed short) \ |
| 227 | vec_perm(x, (__typeof__(x)) { 0 }, \ |
| 228 | ((vector unsigned char) { \ |
| 229 | 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \ |
| 230 | 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 })) |
| 231 | |
| 232 | #define vec_unl(x) \ |
| 233 | (vector signed short) \ |
| 234 | vec_perm(x, (__typeof__(x)) { 0 }, \ |
| 235 | ((vector unsigned char) { \ |
| 236 | 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \ |
| 237 | 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F })) |
| 238 | |
| 239 | #define vec_clip_s16(x) \ |
| 240 | vec_max(vec_min(x, ((vector signed short) { \ |
| 241 | 235, 235, 235, 235, 235, 235, 235, 235 })), \ |
| 242 | ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 })) |
| 243 | |
| 244 | #define vec_packclp(x, y) \ |
| 245 | (vector unsigned char) \ |
| 246 | vec_packs((vector unsigned short) \ |
| 247 | vec_max(x, ((vector signed short) { 0 })), \ |
| 248 | (vector unsigned short) \ |
| 249 | vec_max(y, ((vector signed short) { 0 }))) |
| 250 | |
| 251 | static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y, |
| 252 | vector signed short U, vector signed short V, |
| 253 | vector signed short *R, vector signed short *G, |
| 254 | vector signed short *B) |
| 255 | { |
| 256 | vector signed short vx, ux, uvx; |
| 257 | |
| 258 | Y = vec_mradds(Y, c->CY, c->OY); |
| 259 | U = vec_sub(U, (vector signed short) |
| 260 | vec_splat((vector signed short) { 128 }, 0)); |
| 261 | V = vec_sub(V, (vector signed short) |
| 262 | vec_splat((vector signed short) { 128 }, 0)); |
| 263 | |
| 264 | // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15; |
| 265 | ux = vec_sl(U, c->CSHIFT); |
| 266 | *B = vec_mradds(ux, c->CBU, Y); |
| 267 | |
| 268 | // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15; |
| 269 | vx = vec_sl(V, c->CSHIFT); |
| 270 | *R = vec_mradds(vx, c->CRV, Y); |
| 271 | |
| 272 | // uvx = ((CGU * u) + (CGV * v)) >> 15; |
| 273 | uvx = vec_mradds(U, c->CGU, Y); |
| 274 | *G = vec_mradds(V, c->CGV, uvx); |
| 275 | } |
| 276 | |
| 277 | /* |
| 278 | * ------------------------------------------------------------------------------ |
| 279 | * CS converters |
| 280 | * ------------------------------------------------------------------------------ |
| 281 | */ |
| 282 | |
| 283 | #define DEFCSP420_CVT(name, out_pixels) \ |
| 284 | static int altivec_ ## name(SwsContext *c, const unsigned char **in, \ |
| 285 | int *instrides, int srcSliceY, int srcSliceH, \ |
| 286 | unsigned char **oplanes, int *outstrides) \ |
| 287 | { \ |
| 288 | int w = c->srcW; \ |
| 289 | int h = srcSliceH; \ |
| 290 | int i, j; \ |
| 291 | int instrides_scl[3]; \ |
| 292 | vector unsigned char y0, y1; \ |
| 293 | \ |
| 294 | vector signed char u, v; \ |
| 295 | \ |
| 296 | vector signed short Y0, Y1, Y2, Y3; \ |
| 297 | vector signed short U, V; \ |
| 298 | vector signed short vx, ux, uvx; \ |
| 299 | vector signed short vx0, ux0, uvx0; \ |
| 300 | vector signed short vx1, ux1, uvx1; \ |
| 301 | vector signed short R0, G0, B0; \ |
| 302 | vector signed short R1, G1, B1; \ |
| 303 | vector unsigned char R, G, B; \ |
| 304 | \ |
| 305 | const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \ |
| 306 | vector unsigned char align_perm; \ |
| 307 | \ |
| 308 | vector signed short lCY = c->CY; \ |
| 309 | vector signed short lOY = c->OY; \ |
| 310 | vector signed short lCRV = c->CRV; \ |
| 311 | vector signed short lCBU = c->CBU; \ |
| 312 | vector signed short lCGU = c->CGU; \ |
| 313 | vector signed short lCGV = c->CGV; \ |
| 314 | vector unsigned short lCSHIFT = c->CSHIFT; \ |
| 315 | \ |
| 316 | const ubyte *y1i = in[0]; \ |
| 317 | const ubyte *y2i = in[0] + instrides[0]; \ |
| 318 | const ubyte *ui = in[1]; \ |
| 319 | const ubyte *vi = in[2]; \ |
| 320 | \ |
| 321 | vector unsigned char *oute, *outo; \ |
| 322 | \ |
| 323 | /* loop moves y{1, 2}i by w */ \ |
| 324 | instrides_scl[0] = instrides[0] * 2 - w; \ |
| 325 | /* loop moves ui by w / 2 */ \ |
| 326 | instrides_scl[1] = instrides[1] - w / 2; \ |
| 327 | /* loop moves vi by w / 2 */ \ |
| 328 | instrides_scl[2] = instrides[2] - w / 2; \ |
| 329 | \ |
| 330 | for (i = 0; i < h / 2; i++) { \ |
| 331 | oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \ |
| 332 | (srcSliceY + i * 2)); \ |
| 333 | outo = oute + (outstrides[0] >> 4); \ |
| 334 | vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \ |
| 335 | vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \ |
| 336 | \ |
| 337 | for (j = 0; j < w / 16; j++) { \ |
| 338 | y1ivP = (const vector unsigned char *) y1i; \ |
| 339 | y2ivP = (const vector unsigned char *) y2i; \ |
| 340 | uivP = (const vector unsigned char *) ui; \ |
| 341 | vivP = (const vector unsigned char *) vi; \ |
| 342 | \ |
| 343 | align_perm = vec_lvsl(0, y1i); \ |
| 344 | y0 = (vector unsigned char) \ |
| 345 | vec_perm(y1ivP[0], y1ivP[1], align_perm); \ |
| 346 | \ |
| 347 | align_perm = vec_lvsl(0, y2i); \ |
| 348 | y1 = (vector unsigned char) \ |
| 349 | vec_perm(y2ivP[0], y2ivP[1], align_perm); \ |
| 350 | \ |
| 351 | align_perm = vec_lvsl(0, ui); \ |
| 352 | u = (vector signed char) \ |
| 353 | vec_perm(uivP[0], uivP[1], align_perm); \ |
| 354 | \ |
| 355 | align_perm = vec_lvsl(0, vi); \ |
| 356 | v = (vector signed char) \ |
| 357 | vec_perm(vivP[0], vivP[1], align_perm); \ |
| 358 | \ |
| 359 | u = (vector signed char) \ |
| 360 | vec_sub(u, \ |
| 361 | (vector signed char) \ |
| 362 | vec_splat((vector signed char) { 128 }, 0)); \ |
| 363 | v = (vector signed char) \ |
| 364 | vec_sub(v, \ |
| 365 | (vector signed char) \ |
| 366 | vec_splat((vector signed char) { 128 }, 0)); \ |
| 367 | \ |
| 368 | U = vec_unpackh(u); \ |
| 369 | V = vec_unpackh(v); \ |
| 370 | \ |
| 371 | Y0 = vec_unh(y0); \ |
| 372 | Y1 = vec_unl(y0); \ |
| 373 | Y2 = vec_unh(y1); \ |
| 374 | Y3 = vec_unl(y1); \ |
| 375 | \ |
| 376 | Y0 = vec_mradds(Y0, lCY, lOY); \ |
| 377 | Y1 = vec_mradds(Y1, lCY, lOY); \ |
| 378 | Y2 = vec_mradds(Y2, lCY, lOY); \ |
| 379 | Y3 = vec_mradds(Y3, lCY, lOY); \ |
| 380 | \ |
| 381 | /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \ |
| 382 | ux = vec_sl(U, lCSHIFT); \ |
| 383 | ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \ |
| 384 | ux0 = vec_mergeh(ux, ux); \ |
| 385 | ux1 = vec_mergel(ux, ux); \ |
| 386 | \ |
| 387 | /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \ |
| 388 | vx = vec_sl(V, lCSHIFT); \ |
| 389 | vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \ |
| 390 | vx0 = vec_mergeh(vx, vx); \ |
| 391 | vx1 = vec_mergel(vx, vx); \ |
| 392 | \ |
| 393 | /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \ |
| 394 | uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \ |
| 395 | uvx = vec_mradds(V, lCGV, uvx); \ |
| 396 | uvx0 = vec_mergeh(uvx, uvx); \ |
| 397 | uvx1 = vec_mergel(uvx, uvx); \ |
| 398 | \ |
| 399 | R0 = vec_add(Y0, vx0); \ |
| 400 | G0 = vec_add(Y0, uvx0); \ |
| 401 | B0 = vec_add(Y0, ux0); \ |
| 402 | R1 = vec_add(Y1, vx1); \ |
| 403 | G1 = vec_add(Y1, uvx1); \ |
| 404 | B1 = vec_add(Y1, ux1); \ |
| 405 | \ |
| 406 | R = vec_packclp(R0, R1); \ |
| 407 | G = vec_packclp(G0, G1); \ |
| 408 | B = vec_packclp(B0, B1); \ |
| 409 | \ |
| 410 | out_pixels(R, G, B, oute); \ |
| 411 | \ |
| 412 | R0 = vec_add(Y2, vx0); \ |
| 413 | G0 = vec_add(Y2, uvx0); \ |
| 414 | B0 = vec_add(Y2, ux0); \ |
| 415 | R1 = vec_add(Y3, vx1); \ |
| 416 | G1 = vec_add(Y3, uvx1); \ |
| 417 | B1 = vec_add(Y3, ux1); \ |
| 418 | R = vec_packclp(R0, R1); \ |
| 419 | G = vec_packclp(G0, G1); \ |
| 420 | B = vec_packclp(B0, B1); \ |
| 421 | \ |
| 422 | \ |
| 423 | out_pixels(R, G, B, outo); \ |
| 424 | \ |
| 425 | y1i += 16; \ |
| 426 | y2i += 16; \ |
| 427 | ui += 8; \ |
| 428 | vi += 8; \ |
| 429 | } \ |
| 430 | \ |
| 431 | ui += instrides_scl[1]; \ |
| 432 | vi += instrides_scl[2]; \ |
| 433 | y1i += instrides_scl[0]; \ |
| 434 | y2i += instrides_scl[0]; \ |
| 435 | } \ |
| 436 | return srcSliceH; \ |
| 437 | } |
| 438 | |
| 439 | #define out_abgr(a, b, c, ptr) \ |
| 440 | vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr) |
| 441 | #define out_bgra(a, b, c, ptr) \ |
| 442 | vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr) |
| 443 | #define out_rgba(a, b, c, ptr) \ |
| 444 | vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr) |
| 445 | #define out_argb(a, b, c, ptr) \ |
| 446 | vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr) |
| 447 | #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr) |
| 448 | #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr) |
| 449 | |
| 450 | DEFCSP420_CVT(yuv2_abgr, out_abgr) |
| 451 | DEFCSP420_CVT(yuv2_bgra, out_bgra) |
| 452 | DEFCSP420_CVT(yuv2_rgba, out_rgba) |
| 453 | DEFCSP420_CVT(yuv2_argb, out_argb) |
| 454 | DEFCSP420_CVT(yuv2_rgb24, out_rgb24) |
| 455 | DEFCSP420_CVT(yuv2_bgr24, out_bgr24) |
| 456 | |
| 457 | // uyvy|uyvy|uyvy|uyvy |
| 458 | // 0123 4567 89ab cdef |
| 459 | static const vector unsigned char |
| 460 | demux_u = { 0x10, 0x00, 0x10, 0x00, |
| 461 | 0x10, 0x04, 0x10, 0x04, |
| 462 | 0x10, 0x08, 0x10, 0x08, |
| 463 | 0x10, 0x0c, 0x10, 0x0c }, |
| 464 | demux_v = { 0x10, 0x02, 0x10, 0x02, |
| 465 | 0x10, 0x06, 0x10, 0x06, |
| 466 | 0x10, 0x0A, 0x10, 0x0A, |
| 467 | 0x10, 0x0E, 0x10, 0x0E }, |
| 468 | demux_y = { 0x10, 0x01, 0x10, 0x03, |
| 469 | 0x10, 0x05, 0x10, 0x07, |
| 470 | 0x10, 0x09, 0x10, 0x0B, |
| 471 | 0x10, 0x0D, 0x10, 0x0F }; |
| 472 | |
| 473 | /* |
| 474 | * this is so I can play live CCIR raw video |
| 475 | */ |
| 476 | static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in, |
| 477 | int *instrides, int srcSliceY, int srcSliceH, |
| 478 | unsigned char **oplanes, int *outstrides) |
| 479 | { |
| 480 | int w = c->srcW; |
| 481 | int h = srcSliceH; |
| 482 | int i, j; |
| 483 | vector unsigned char uyvy; |
| 484 | vector signed short Y, U, V; |
| 485 | vector signed short R0, G0, B0, R1, G1, B1; |
| 486 | vector unsigned char R, G, B; |
| 487 | vector unsigned char *out; |
| 488 | const ubyte *img; |
| 489 | |
| 490 | img = in[0]; |
| 491 | out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]); |
| 492 | |
| 493 | for (i = 0; i < h; i++) |
| 494 | for (j = 0; j < w / 16; j++) { |
| 495 | uyvy = vec_ld(0, img); |
| 496 | |
| 497 | U = (vector signed short) |
| 498 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u); |
| 499 | V = (vector signed short) |
| 500 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v); |
| 501 | Y = (vector signed short) |
| 502 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y); |
| 503 | |
| 504 | cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0); |
| 505 | |
| 506 | uyvy = vec_ld(16, img); |
| 507 | |
| 508 | U = (vector signed short) |
| 509 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u); |
| 510 | V = (vector signed short) |
| 511 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v); |
| 512 | Y = (vector signed short) |
| 513 | vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y); |
| 514 | |
| 515 | cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1); |
| 516 | |
| 517 | R = vec_packclp(R0, R1); |
| 518 | G = vec_packclp(G0, G1); |
| 519 | B = vec_packclp(B0, B1); |
| 520 | |
| 521 | // vec_mstbgr24 (R,G,B, out); |
| 522 | out_rgba(R, G, B, out); |
| 523 | |
| 524 | img += 32; |
| 525 | } |
| 526 | return srcSliceH; |
| 527 | } |
| 528 | |
| 529 | #endif /* HAVE_ALTIVEC */ |
| 530 | |
| 531 | /* Ok currently the acceleration routine only supports |
| 532 | * inputs of widths a multiple of 16 |
| 533 | * and heights a multiple 2 |
| 534 | * |
| 535 | * So we just fall back to the C codes for this. |
| 536 | */ |
| 537 | av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c) |
| 538 | { |
| 539 | #if HAVE_ALTIVEC |
| 540 | if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) |
| 541 | return NULL; |
| 542 | |
| 543 | /* |
| 544 | * and this seems not to matter too much I tried a bunch of |
| 545 | * videos with abnormal widths and MPlayer crashes elsewhere. |
| 546 | * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv |
| 547 | * boom with X11 bad match. |
| 548 | * |
| 549 | */ |
| 550 | if ((c->srcW & 0xf) != 0) |
| 551 | return NULL; |
| 552 | |
| 553 | switch (c->srcFormat) { |
| 554 | case AV_PIX_FMT_YUV410P: |
| 555 | case AV_PIX_FMT_YUV420P: |
| 556 | /*case IMGFMT_CLPL: ??? */ |
| 557 | case AV_PIX_FMT_GRAY8: |
| 558 | case AV_PIX_FMT_NV12: |
| 559 | case AV_PIX_FMT_NV21: |
| 560 | if ((c->srcH & 0x1) != 0) |
| 561 | return NULL; |
| 562 | |
| 563 | switch (c->dstFormat) { |
| 564 | case AV_PIX_FMT_RGB24: |
| 565 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n"); |
| 566 | return altivec_yuv2_rgb24; |
| 567 | case AV_PIX_FMT_BGR24: |
| 568 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n"); |
| 569 | return altivec_yuv2_bgr24; |
| 570 | case AV_PIX_FMT_ARGB: |
| 571 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n"); |
| 572 | return altivec_yuv2_argb; |
| 573 | case AV_PIX_FMT_ABGR: |
| 574 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n"); |
| 575 | return altivec_yuv2_abgr; |
| 576 | case AV_PIX_FMT_RGBA: |
| 577 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n"); |
| 578 | return altivec_yuv2_rgba; |
| 579 | case AV_PIX_FMT_BGRA: |
| 580 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n"); |
| 581 | return altivec_yuv2_bgra; |
| 582 | default: return NULL; |
| 583 | } |
| 584 | break; |
| 585 | |
| 586 | case AV_PIX_FMT_UYVY422: |
| 587 | switch (c->dstFormat) { |
| 588 | case AV_PIX_FMT_BGR32: |
| 589 | av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n"); |
| 590 | return altivec_uyvy_rgb32; |
| 591 | default: return NULL; |
| 592 | } |
| 593 | break; |
| 594 | } |
| 595 | #endif /* HAVE_ALTIVEC */ |
| 596 | |
| 597 | return NULL; |
| 598 | } |
| 599 | |
| 600 | av_cold void ff_yuv2rgb_init_tables_ppc(SwsContext *c, |
| 601 | const int inv_table[4], |
| 602 | int brightness, |
| 603 | int contrast, |
| 604 | int saturation) |
| 605 | { |
| 606 | #if HAVE_ALTIVEC |
| 607 | union { |
| 608 | DECLARE_ALIGNED(16, signed short, tmp)[8]; |
| 609 | vector signed short vec; |
| 610 | } buf; |
| 611 | |
| 612 | if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) |
| 613 | return; |
| 614 | |
| 615 | buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy |
| 616 | buf.tmp[1] = -256 * brightness; // oy |
| 617 | buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv |
| 618 | buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu |
| 619 | buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu |
| 620 | buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv |
| 621 | |
| 622 | c->CSHIFT = (vector unsigned short) vec_splat_u16(2); |
| 623 | c->CY = vec_splat((vector signed short) buf.vec, 0); |
| 624 | c->OY = vec_splat((vector signed short) buf.vec, 1); |
| 625 | c->CRV = vec_splat((vector signed short) buf.vec, 2); |
| 626 | c->CBU = vec_splat((vector signed short) buf.vec, 3); |
| 627 | c->CGU = vec_splat((vector signed short) buf.vec, 4); |
| 628 | c->CGV = vec_splat((vector signed short) buf.vec, 5); |
| 629 | return; |
| 630 | #endif /* HAVE_ALTIVEC */ |
| 631 | } |
| 632 | |
| 633 | #if HAVE_ALTIVEC |
| 634 | |
| 635 | static av_always_inline void yuv2packedX_altivec(SwsContext *c, |
| 636 | const int16_t *lumFilter, |
| 637 | const int16_t **lumSrc, |
| 638 | int lumFilterSize, |
| 639 | const int16_t *chrFilter, |
| 640 | const int16_t **chrUSrc, |
| 641 | const int16_t **chrVSrc, |
| 642 | int chrFilterSize, |
| 643 | const int16_t **alpSrc, |
| 644 | uint8_t *dest, |
| 645 | int dstW, int dstY, |
| 646 | enum AVPixelFormat target) |
| 647 | { |
| 648 | int i, j; |
| 649 | vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V; |
| 650 | vector signed short R0, G0, B0, R1, G1, B1; |
| 651 | |
| 652 | vector unsigned char R, G, B; |
| 653 | vector unsigned char *out, *nout; |
| 654 | |
| 655 | vector signed short RND = vec_splat_s16(1 << 3); |
| 656 | vector unsigned short SCL = vec_splat_u16(4); |
| 657 | DECLARE_ALIGNED(16, unsigned int, scratch)[16]; |
| 658 | |
| 659 | vector signed short *YCoeffs, *CCoeffs; |
| 660 | |
| 661 | YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize; |
| 662 | CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize; |
| 663 | |
| 664 | out = (vector unsigned char *) dest; |
| 665 | |
| 666 | for (i = 0; i < dstW; i += 16) { |
| 667 | Y0 = RND; |
| 668 | Y1 = RND; |
| 669 | /* extract 16 coeffs from lumSrc */ |
| 670 | for (j = 0; j < lumFilterSize; j++) { |
| 671 | X0 = vec_ld(0, &lumSrc[j][i]); |
| 672 | X1 = vec_ld(16, &lumSrc[j][i]); |
| 673 | Y0 = vec_mradds(X0, YCoeffs[j], Y0); |
| 674 | Y1 = vec_mradds(X1, YCoeffs[j], Y1); |
| 675 | } |
| 676 | |
| 677 | U = RND; |
| 678 | V = RND; |
| 679 | /* extract 8 coeffs from U,V */ |
| 680 | for (j = 0; j < chrFilterSize; j++) { |
| 681 | X = vec_ld(0, &chrUSrc[j][i / 2]); |
| 682 | U = vec_mradds(X, CCoeffs[j], U); |
| 683 | X = vec_ld(0, &chrVSrc[j][i / 2]); |
| 684 | V = vec_mradds(X, CCoeffs[j], V); |
| 685 | } |
| 686 | |
| 687 | /* scale and clip signals */ |
| 688 | Y0 = vec_sra(Y0, SCL); |
| 689 | Y1 = vec_sra(Y1, SCL); |
| 690 | U = vec_sra(U, SCL); |
| 691 | V = vec_sra(V, SCL); |
| 692 | |
| 693 | Y0 = vec_clip_s16(Y0); |
| 694 | Y1 = vec_clip_s16(Y1); |
| 695 | U = vec_clip_s16(U); |
| 696 | V = vec_clip_s16(V); |
| 697 | |
| 698 | /* now we have |
| 699 | * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 |
| 700 | * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7 |
| 701 | * |
| 702 | * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 |
| 703 | * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7 |
| 704 | * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7 |
| 705 | */ |
| 706 | |
| 707 | U0 = vec_mergeh(U, U); |
| 708 | V0 = vec_mergeh(V, V); |
| 709 | |
| 710 | U1 = vec_mergel(U, U); |
| 711 | V1 = vec_mergel(V, V); |
| 712 | |
| 713 | cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0); |
| 714 | cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1); |
| 715 | |
| 716 | R = vec_packclp(R0, R1); |
| 717 | G = vec_packclp(G0, G1); |
| 718 | B = vec_packclp(B0, B1); |
| 719 | |
| 720 | switch (target) { |
| 721 | case AV_PIX_FMT_ABGR: |
| 722 | out_abgr(R, G, B, out); |
| 723 | break; |
| 724 | case AV_PIX_FMT_BGRA: |
| 725 | out_bgra(R, G, B, out); |
| 726 | break; |
| 727 | case AV_PIX_FMT_RGBA: |
| 728 | out_rgba(R, G, B, out); |
| 729 | break; |
| 730 | case AV_PIX_FMT_ARGB: |
| 731 | out_argb(R, G, B, out); |
| 732 | break; |
| 733 | case AV_PIX_FMT_RGB24: |
| 734 | out_rgb24(R, G, B, out); |
| 735 | break; |
| 736 | case AV_PIX_FMT_BGR24: |
| 737 | out_bgr24(R, G, B, out); |
| 738 | break; |
| 739 | default: |
| 740 | { |
| 741 | /* If this is reached, the caller should have called yuv2packedXinC |
| 742 | * instead. */ |
| 743 | static int printed_error_message; |
| 744 | if (!printed_error_message) { |
| 745 | av_log(c, AV_LOG_ERROR, |
| 746 | "altivec_yuv2packedX doesn't support %s output\n", |
| 747 | av_get_pix_fmt_name(c->dstFormat)); |
| 748 | printed_error_message = 1; |
| 749 | } |
| 750 | return; |
| 751 | } |
| 752 | } |
| 753 | } |
| 754 | |
| 755 | if (i < dstW) { |
| 756 | i -= 16; |
| 757 | |
| 758 | Y0 = RND; |
| 759 | Y1 = RND; |
| 760 | /* extract 16 coeffs from lumSrc */ |
| 761 | for (j = 0; j < lumFilterSize; j++) { |
| 762 | X0 = vec_ld(0, &lumSrc[j][i]); |
| 763 | X1 = vec_ld(16, &lumSrc[j][i]); |
| 764 | Y0 = vec_mradds(X0, YCoeffs[j], Y0); |
| 765 | Y1 = vec_mradds(X1, YCoeffs[j], Y1); |
| 766 | } |
| 767 | |
| 768 | U = RND; |
| 769 | V = RND; |
| 770 | /* extract 8 coeffs from U,V */ |
| 771 | for (j = 0; j < chrFilterSize; j++) { |
| 772 | X = vec_ld(0, &chrUSrc[j][i / 2]); |
| 773 | U = vec_mradds(X, CCoeffs[j], U); |
| 774 | X = vec_ld(0, &chrVSrc[j][i / 2]); |
| 775 | V = vec_mradds(X, CCoeffs[j], V); |
| 776 | } |
| 777 | |
| 778 | /* scale and clip signals */ |
| 779 | Y0 = vec_sra(Y0, SCL); |
| 780 | Y1 = vec_sra(Y1, SCL); |
| 781 | U = vec_sra(U, SCL); |
| 782 | V = vec_sra(V, SCL); |
| 783 | |
| 784 | Y0 = vec_clip_s16(Y0); |
| 785 | Y1 = vec_clip_s16(Y1); |
| 786 | U = vec_clip_s16(U); |
| 787 | V = vec_clip_s16(V); |
| 788 | |
| 789 | /* now we have |
| 790 | * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 |
| 791 | * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7 |
| 792 | * |
| 793 | * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 |
| 794 | * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7 |
| 795 | * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7 |
| 796 | */ |
| 797 | |
| 798 | U0 = vec_mergeh(U, U); |
| 799 | V0 = vec_mergeh(V, V); |
| 800 | |
| 801 | U1 = vec_mergel(U, U); |
| 802 | V1 = vec_mergel(V, V); |
| 803 | |
| 804 | cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0); |
| 805 | cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1); |
| 806 | |
| 807 | R = vec_packclp(R0, R1); |
| 808 | G = vec_packclp(G0, G1); |
| 809 | B = vec_packclp(B0, B1); |
| 810 | |
| 811 | nout = (vector unsigned char *) scratch; |
| 812 | switch (target) { |
| 813 | case AV_PIX_FMT_ABGR: |
| 814 | out_abgr(R, G, B, nout); |
| 815 | break; |
| 816 | case AV_PIX_FMT_BGRA: |
| 817 | out_bgra(R, G, B, nout); |
| 818 | break; |
| 819 | case AV_PIX_FMT_RGBA: |
| 820 | out_rgba(R, G, B, nout); |
| 821 | break; |
| 822 | case AV_PIX_FMT_ARGB: |
| 823 | out_argb(R, G, B, nout); |
| 824 | break; |
| 825 | case AV_PIX_FMT_RGB24: |
| 826 | out_rgb24(R, G, B, nout); |
| 827 | break; |
| 828 | case AV_PIX_FMT_BGR24: |
| 829 | out_bgr24(R, G, B, nout); |
| 830 | break; |
| 831 | default: |
| 832 | /* Unreachable, I think. */ |
| 833 | av_log(c, AV_LOG_ERROR, |
| 834 | "altivec_yuv2packedX doesn't support %s output\n", |
| 835 | av_get_pix_fmt_name(c->dstFormat)); |
| 836 | return; |
| 837 | } |
| 838 | |
| 839 | memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4); |
| 840 | } |
| 841 | } |
| 842 | |
| 843 | #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \ |
| 844 | void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \ |
| 845 | const int16_t *lumFilter, \ |
| 846 | const int16_t **lumSrc, \ |
| 847 | int lumFilterSize, \ |
| 848 | const int16_t *chrFilter, \ |
| 849 | const int16_t **chrUSrc, \ |
| 850 | const int16_t **chrVSrc, \ |
| 851 | int chrFilterSize, \ |
| 852 | const int16_t **alpSrc, \ |
| 853 | uint8_t *dest, int dstW, int dstY) \ |
| 854 | { \ |
| 855 | yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \ |
| 856 | chrFilter, chrUSrc, chrVSrc, \ |
| 857 | chrFilterSize, alpSrc, \ |
| 858 | dest, dstW, dstY, pixfmt); \ |
| 859 | } |
| 860 | |
| 861 | YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR); |
| 862 | YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA); |
| 863 | YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB); |
| 864 | YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA); |
| 865 | YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24); |
| 866 | YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24); |
| 867 | |
| 868 | #endif /* HAVE_ALTIVEC */ |