1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989
|
/* ScummVM - Graphic Adventure Engine
*
* ScummVM is the legal property of its developers, whose names
* are too numerous to list here. Please refer to the COPYRIGHT
* file distributed with this source distribution.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*
*/
#include "ags/ags.h"
// Without this ifdef the iOS backend breaks, please do not remove
#ifdef SCUMMVM_NEON
#include "ags/globals.h"
#include "ags/lib/allegro/color.h"
#include "ags/lib/allegro/flood.h"
#include "ags/lib/allegro/gfx.h"
#include "common/textconsole.h"
#include "graphics/screen.h"
#include <arm_neon.h>
#if !defined(__aarch64__) && !defined(__ARM_NEON)
#if defined(__clang__)
#pragma clang attribute push (__attribute__((target("neon"))), apply_to=function)
#elif defined(__GNUC__)
#pragma GCC push_options
#pragma GCC target("fpu=neon")
#endif
#endif // !defined(__aarch64__) && !defined(__ARM_NEON)
namespace AGS3 {
class DrawInnerImpl_NEON {
static inline uint32x4_t simd2BppTo4Bpp(uint16x4_t pixels) {
uint32x4_t x = vmovl_u16(pixels);
// c is the extracted 5/6 bit color from the image
uint32x4_t c = vshrq_n_u32(x, 11);
// We convert it back to normal by shifting it thrice over, naturally, and then using the 2 most
// sinificant bits in the original color for the least significant bits in the new one
uint32x4_t r = vshlq_n_u32(vorrq_u32(vshlq_n_u32(c, 3), vshrq_n_u32(c, 2)), 16);
c = vshrq_n_u32(vandq_u32(x, vmovq_n_u32(0x07e0)), 5);
uint32x4_t g = vshlq_n_u32(vorrq_u32(vshlq_n_u32(c, 2), vshrq_n_u32(c, 4)), 8);
c = vandq_u32(x, vmovq_n_u32(0x001f));
uint32x4_t b = vorrq_u32(vshlq_n_u32(c, 3), vshrq_n_u32(c, 2));
// By default 2bpp to 4bpp makes the alpha channel 255
return vorrq_u32(vorrq_u32(vorrq_u32(r, g), b), vmovq_n_u32(0xff000000));
}
static inline uint16x4_t simd4BppTo2Bpp(uint32x4_t pixels) {
// x is the final 16 bit rgb pixel
uint32x4_t x = vshrq_n_u32(vandq_u32(pixels, vmovq_n_u32(0x000000ff)), 3);
x = vorrq_u32(x, vshlq_n_u32(vshrq_n_u32(vandq_u32(pixels, vmovq_n_u32(0x0000ff00)), 8+2), 5));
x = vorrq_u32(x, vshlq_n_u32(vshrq_n_u32(vandq_u32(pixels, vmovq_n_u32(0x00ff0000)), 16+3), 11));
return vmovn_u32(x);
}
static inline uint16x8_t rgbBlendSIMD2Bpp(uint16x8_t srcCols, uint16x8_t destCols, uint16x8_t alphas) {
// Here we add 1 to alphas if its 0. This is what the original blender function did
alphas = vaddq_u16(alphas, vandq_u16(vceqq_u16(alphas, vmovq_n_u16(0)), vmovq_n_u16(1)));
// Split the components into rgb
uint16x8_t srcComps[] = {
vandq_u16(srcCols, vmovq_n_u16(0x1f)), // B
vandq_u16(vshrq_n_u16(srcCols, 5), vmovq_n_u16(0x3f)), // G
vshrq_n_u16(srcCols, 11), // R
}, destComps[] = {
vandq_u16(destCols, vmovq_n_u16(0x1f)), // B
vandq_u16(vshrq_n_u16(destCols, 5), vmovq_n_u16(0x3f)), // G
vshrq_n_u16(destCols, 11), // R
};
// At some point I made it so that it would put them into their 8bit depth format
// to keep the function as 1-1 with the original, but it didn't seem to help much
//srcComps[0] = vorrq_u16(vshlq_n_u16(srcComps[0], 3), vshrq_n_u16(srcComps[0], 2));
//srcComps[1] = vorrq_u16(vshlq_n_u16(srcComps[1], 2), vshrq_n_u16(srcComps[1], 4));
//srcComps[2] = vorrq_u16(vshlq_n_u16(srcComps[2], 3), vshrq_n_u16(srcComps[2], 2));
//destComps[0] = vorrq_u16(vshlq_n_u16(destComps[0], 3), vshrq_n_u16(destComps[0], 2));
//destComps[1] = vorrq_u16(vshlq_n_u16(destComps[1], 2), vshrq_n_u16(destComps[1], 4));
//destComps[2] = vorrq_u16(vshlq_n_u16(destComps[2], 3), vshrq_n_u16(destComps[2], 2));
// Calculate the differences between the colors
uint16x8_t diffs[] = {
vsubq_u16(srcComps[0], destComps[0]), // B
vsubq_u16(srcComps[1], destComps[1]), // G
vsubq_u16(srcComps[2], destComps[2]), // R
};
// Multiply by alpha and shift depth bits to the right
// pretty much the same as (int)(((float)component / 255.0f) * ((float)alpha / 255.0f) * 255.0f)
alphas = vshrq_n_u16(alphas, 2);
diffs[1] = vshrq_n_u16(vmulq_u16(diffs[1], alphas), 6);
alphas = vshrq_n_u16(alphas, 1);
diffs[0] = vshrq_n_u16(vmulq_u16(diffs[0], alphas), 5);
diffs[2] = vshrq_n_u16(vmulq_u16(diffs[2], alphas), 5);
// Originally, I converted it back to normal here from the 8bpp form, but don't need to do that anymore
//diffs[0] = vandq_u16(vshrq_n_u16(vaddq_u16(diffs[0], destComps[0]), 3), vmovq_n_u16(0x1f));
//diffs[1] = vandq_u16(vshrq_n_u16(vaddq_u16(diffs[1], destComps[1]), 2), vmovq_n_u16(0x3f));
//diffs[2] = vandq_u16(vshrq_n_u16(vaddq_u16(diffs[2], destComps[2]), 3), vmovq_n_u16(0x1f));
// Here we add the difference between the 2 colors times alpha onto the destination
diffs[0] = vandq_u16(vaddq_u16(diffs[0], destComps[0]), vmovq_n_u16(0x1f));
diffs[1] = vandq_u16(vaddq_u16(diffs[1], destComps[1]), vmovq_n_u16(0x3f));
diffs[2] = vandq_u16(vaddq_u16(diffs[2], destComps[2]), vmovq_n_u16(0x1f));
// We compile all the colors into diffs[0] as a 16 bit rgb pixel
diffs[0] = vorrq_u16(diffs[0], vshlq_n_u16(diffs[1], 5));
return vorrq_u16(diffs[0], vshlq_n_u16(diffs[2], 11));
}
// preserveAlpha:
// false => set destCols's alpha to 0
// true => keep destCols's alpha
static inline uint32x4_t rgbBlendSIMD(uint32x4_t srcCols, uint32x4_t destCols, uint32x4_t alphas, bool preserveAlpha) {
// Here we add 1 to alphas if its 0. This is what the original blender function did
alphas = vaddq_u32(alphas, vandq_u32(vcgtq_u32(alphas, vmovq_n_u32(0)), vmovq_n_u32(1)));
// Get the alpha from the destination
uint32x4_t alpha = vandq_u32(destCols, vmovq_n_u32(0xff000000));
// Get red and blue components
uint32x4_t srcColsCopy = srcCols;
srcColsCopy = vandq_u32(srcColsCopy, vmovq_n_u32(0xff00ff));
uint32x4_t destColsCopy = destCols;
destColsCopy = vandq_u32(destColsCopy, vmovq_n_u32(0xff00ff));
// compute the difference, then multiply by alpha and divide by 255
srcColsCopy = vsubq_u32(srcColsCopy, destColsCopy);
srcColsCopy = vmulq_u32(srcColsCopy, alphas);
srcColsCopy = vshrq_n_u32(srcColsCopy, 8);
srcColsCopy = vaddq_u32(srcColsCopy, destCols); // Add the new red/blue to the old ones
// do the same for the green component
srcCols = vandq_u32(srcCols, vmovq_n_u32(0xff00));
destCols = vandq_u32(destCols, vmovq_n_u32(0xff00));
srcCols = vsubq_u32(srcCols, destCols);
srcCols = vmulq_u32(srcCols, alphas);
srcCols = vshrq_n_u32(srcCols, 8);
srcCols = vaddq_u32(srcCols, destCols); // Add the new green to the old green
// keep values in 8bit range and glue red/blue and green together
srcColsCopy = vandq_u32(srcColsCopy, vmovq_n_u32(0xff00ff));
srcCols = vandq_u32(srcCols, vmovq_n_u32(0xff00));
srcCols = vorrq_u32(srcCols, srcColsCopy);
// Remember that alpha is not alphas, but rather the alpha of destCols
if (preserveAlpha) {
srcCols = vandq_u32(srcCols, vmovq_n_u32(0x00ffffff));
srcCols = vorrq_u32(srcCols, alpha);
}
return srcCols;
}
// uses the alpha from srcCols and destCols
static inline uint32x4_t argbBlendSIMD(uint32x4_t srcCols, uint32x4_t destCols) {
float32x4_t srcA = vcvtq_f32_u32(vshrq_n_u32(srcCols, 24));
srcA = vmulq_n_f32(srcA, 1.0f / 255.0f);
float32x4_t srcR = vcvtq_f32_u32(vandq_u32(vshrq_n_u32(srcCols, 16), vmovq_n_u32(0xff)));
float32x4_t srcG = vcvtq_f32_u32(vandq_u32(vshrq_n_u32(srcCols, 8), vmovq_n_u32(0xff)));
float32x4_t srcB = vcvtq_f32_u32(vandq_u32(srcCols, vmovq_n_u32(0xff)));
float32x4_t destA = vcvtq_f32_u32(vshrq_n_u32(destCols, 24));
destA = vmulq_n_f32(destA, 1.0f / 255.0f);
float32x4_t destR = vcvtq_f32_u32(vandq_u32(vshrq_n_u32(destCols, 16), vmovq_n_u32(0xff)));
float32x4_t destG = vcvtq_f32_u32(vandq_u32(vshrq_n_u32(destCols, 8), vmovq_n_u32(0xff)));
float32x4_t destB = vcvtq_f32_u32(vandq_u32(destCols, vmovq_n_u32(0xff)));
// the destination alpha gets multiplied by 255 - source alpha
destA = vmulq_f32(destA, vsubq_f32(vmovq_n_f32(1.0f), srcA));
// ((src * sAlpha) + (dest * dAlpha)) / (sAlpha + dAlpha)
float32x4_t combA = vaddq_f32(srcA, destA);
float32x4_t combArcp = vrecpeq_f32(combA);
destR = vmulq_f32(vaddq_f32(vmulq_f32(srcR, srcA), vmulq_f32(destR, destA)), combArcp);
destG = vmulq_f32(vaddq_f32(vmulq_f32(srcG, srcA), vmulq_f32(destG, destA)), combArcp);
destB = vmulq_f32(vaddq_f32(vmulq_f32(srcB, srcA), vmulq_f32(destB, destA)), combArcp);
combA = vmulq_n_f32(combA, 255.0);
// Now put it back together
return vorrq_u32(vshlq_n_u32(vcvtq_u32_f32(combA), 24),
vorrq_u32(vshlq_n_u32(vcvtq_u32_f32(destR), 16),
vorrq_u32(vshlq_n_u32(vcvtq_u32_f32(destG), 8),
vcvtq_u32_f32(destB))));
}
static inline uint32x4_t blendTintSpriteSIMD(uint32x4_t srcCols, uint32x4_t destCols, uint32x4_t alphas, bool light) {
// This function is NOT 1 to 1 with the original... It just approximates it
// It gets the value of the HSV of the dest color
// Then it gets the HSV of the srcCols
// how the values are transformed
// from 1 uint32x4_t srcCols with each lane being ARGB uint32
// srcCols[0] = A | R | G | B
// srcCols[1] = A | R | G | B
// srcCols[2] = A | R | G | B
// srcCols[3] = A | R | G | B
// ->
// to 4 float32x4_t's each being a separate channel with each lane
// corresponding to their respective srcCols lane
// dda = { A[0], A[1], A[2], A[3] }
// ddr = { R[0], R[1], R[2], R[3] }
// ddg = { G[0], G[1], G[2], G[3] }
// ddb = { B[0], B[1], B[2], B[3] }
// do the transformation (we don't actually need alpha at all)
float32x4_t ddr, ddg, ddb;
ddr = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(vshrq_n_u32(destCols, 16), vmovq_n_u32(0xff))), 1.0 / 255.0);
ddg = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(vshrq_n_u32(destCols, 8), vmovq_n_u32(0xff))), 1.0 / 255.0);
ddb = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(destCols, vmovq_n_u32(0xff))), 1.0 / 255.0);
float32x4_t ssr, ssg, ssb;
ssr = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(vshrq_n_u32(srcCols, 16), vmovq_n_u32(0xff))), 1.0 / 255.0);
ssg = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(vshrq_n_u32(srcCols, 8), vmovq_n_u32(0xff))), 1.0 / 255.0);
ssb = vmulq_n_f32(vcvtq_f32_u32(vandq_u32(srcCols, vmovq_n_u32(0xff))), 1.0 / 255.0);
// Get the maxes and mins (needed for HSV->RGB and vice-versa)
float32x4_t dmaxes = vmaxq_f32(ddr, vmaxq_f32(ddg, ddb));
float32x4_t smaxes = vmaxq_f32(ssr, vmaxq_f32(ssg, ssb));
float32x4_t smins = vminq_f32(ssr, vminq_f32(ssg, ssb));
// This is here to stop from dividing by 0
const float32x4_t eplison0 = vmovq_n_f32(0.0000001);
float32x4_t chroma = vmaxq_f32(vsubq_f32(smaxes, smins), eplison0);
// RGB to HSV is a piecewise function, so we compute each part of the function first...
float32x4_t hr, hg, hb, hue, chromaReq;
chromaReq = vrecpeq_f32(chroma);
hr = vmulq_f32(vsubq_f32(ssg, ssb), chromaReq);
float32x4_t hrDiv6 = vmulq_n_f32(hr, 1.0 / 6.0);
hrDiv6 = vsubq_f32(hrDiv6, vcvtq_f32_u32(vandq_u32(vcltq_f32(hrDiv6, vmovq_n_f32(0.0)), vmovq_n_u32(1))));
hr = vsubq_f32(hr, vmulq_n_f32(vcvtq_f32_s32(vcvtq_s32_f32(hrDiv6)), 6.0));
hg = vaddq_f32(vmulq_f32(vsubq_f32(ssb, ssr), chromaReq), vmovq_n_f32(2.0));
hb = vaddq_f32(vmulq_f32(vsubq_f32(ssr, ssg), chromaReq), vmovq_n_f32(4.0));
// And then compute which one will be used based on criteria
float32x4_t hrfactors = vcvtq_f32_u32(vandq_u32(vandq_u32(vceqq_f32(ssr, smaxes), vmvnq_u32(vceqq_f32(ssr, ssb))), vmovq_n_u32(1)));
float32x4_t hgfactors = vcvtq_f32_u32(vandq_u32(vandq_u32(vceqq_f32(ssg, smaxes), vmvnq_u32(vceqq_f32(ssg, ssr))), vmovq_n_u32(1)));
float32x4_t hbfactors = vcvtq_f32_u32(vandq_u32(vandq_u32(vceqq_f32(ssb, smaxes), vmvnq_u32(vceqq_f32(ssb, ssg))), vmovq_n_u32(1)));
hue = vmulq_f32(hr, hrfactors);
hue = vaddq_f32(hue, vmulq_f32(hg, hgfactors));
hue = vaddq_f32(hue, vmulq_f32(hb, hbfactors));
// Mess with the light like the original function
float32x4_t val = dmaxes;
if (light) {
val = vsubq_f32(val, vsubq_f32(vmovq_n_f32(1.0), vmulq_n_f32(vcvtq_f32_u32(alphas), 1.0 / 250.0)));
val = vmaxq_f32(val, vmovq_n_f32(0.0));
}
// then it stitches the HSV back together
// the hue and saturation come from the source (tint) color, and the value comes from
// the destination (real source) color
chroma = vmulq_f32(val, vmulq_f32(vsubq_f32(smaxes, smins), vrecpeq_f32(vaddq_f32(smaxes, eplison0))));
float32x4_t hprime_mod2 = vmulq_n_f32(hue, 1.0 / 2.0);
hprime_mod2 = vmulq_n_f32(vsubq_f32(hprime_mod2, vcvtq_f32_s32(vcvtq_s32_f32(hprime_mod2))), 2.0);
float32x4_t x = vmulq_f32(chroma, vsubq_f32(vmovq_n_f32(1.0), vabsq_f32(vsubq_f32(hprime_mod2, vmovq_n_f32(1.0)))));
uint32x4_t hprime_rounded = vcvtq_u32_f32(hue);
uint32x4_t x_int = vcvtq_u32_f32(vmulq_n_f32(x, 255.0));
uint32x4_t c_int = vcvtq_u32_f32(vmulq_n_f32(chroma, 255.0));
// Again HSV->RGB is also a piecewise function
uint32x4_t val0 = vorrq_u32(vshlq_n_u32(x_int, 8), vshlq_n_u32(c_int, 16));
val0 = vandq_u32(val0, vorrq_u32(vceqq_u32(hprime_rounded, vmovq_n_u32(0)), vceqq_u32(hprime_rounded, vmovq_n_u32(6))));
uint32x4_t val1 = vorrq_u32(vshlq_n_u32(c_int, 8), vshlq_n_u32(x_int, 16));
val1 = vandq_u32(val1, vceqq_u32(hprime_rounded, vmovq_n_u32(1)));
uint32x4_t val2 = vorrq_u32(vshlq_n_u32(c_int, 8), x_int);
val2 = vandq_u32(val2, vceqq_u32(hprime_rounded, vmovq_n_u32(2)));
uint32x4_t val3 = vorrq_u32(vshlq_n_u32(x_int, 8), c_int);
val3 = vandq_u32(val3, vceqq_u32(hprime_rounded, vmovq_n_u32(3)));
uint32x4_t val4 = vorrq_u32(vshlq_n_u32(x_int, 16), c_int);
val4 = vandq_u32(val4, vceqq_u32(hprime_rounded, vmovq_n_u32(4)));
uint32x4_t val5 = vorrq_u32(vshlq_n_u32(c_int, 16), x_int);
val5 = vandq_u32(val5, vceqq_u32(hprime_rounded, vmovq_n_u32(5)));
// or the values together
uint32x4_t final = vorrq_u32(val0, vorrq_u32(val1, vorrq_u32(val2, vorrq_u32(val3, vorrq_u32(val4, val5)))));
// add the minimums back in
uint32x4_t val_add = vcvtq_u32_f32(vmulq_n_f32(vsubq_f32(val, chroma), 255.0));
val_add = vorrq_u32(val_add, vorrq_u32(vshlq_n_u32(val_add, 8), vorrq_u32(vshlq_n_u32(val_add, 16), vandq_u32(destCols, vmovq_n_u32(0xff000000)))));
final = vaddq_u32(final, val_add);
return final;
}
static inline uint32x4_t blendPixelSIMD(uint32x4_t srcCols, uint32x4_t destCols, uint32x4_t alphas) {
uint32x4_t srcAlphas, difAlphas, mask, ch1, ch2;
auto setupArgbAlphas = [&]() {
// This acts the same as this in the normal blender functions
// if (alpha == 0)
// alpha = aSrc;
// else
// alpha = aSrc * ((alpha & 0xff) + 1) / 256;
// where alpha is the alpha byte of the srcCols
srcAlphas = vshrq_n_u32(srcCols, 24);
difAlphas = vaddq_u32(vandq_u32(alphas, vmovq_n_u32(0xff)), vmovq_n_u32(1));
difAlphas = vshrq_n_u32(vmulq_u32(srcAlphas, difAlphas), 8);
difAlphas = vshlq_n_u32(difAlphas, 24);
srcAlphas = vshlq_n_u32(srcAlphas, 24);
mask = vceqq_u32(alphas, vmovq_n_u32(0));
srcAlphas = vandq_u32(srcAlphas, mask);
difAlphas = vandq_u32(difAlphas, vmvnq_u32(mask));
srcCols = vandq_u32(srcCols, vmovq_n_u32(0x00ffffff));
srcCols = vorrq_u32(srcCols, vorrq_u32(srcAlphas, difAlphas));
};
switch (_G(_blender_mode)) {
case kSourceAlphaBlender: // see BITMAP member function blendSourceAlpha
alphas = vshrq_n_u32(srcCols, 24);
return rgbBlendSIMD(srcCols, destCols, alphas, false);
case kArgbToArgbBlender: // see BITMAP member function blendArgbToArgb
setupArgbAlphas();
// only blend if alpha isn't 0, otherwise use destCols
mask = vcgtq_u32(vshrq_n_u32(srcCols, 24), vmovq_n_u32(0));
ch1 = vandq_u32(argbBlendSIMD(srcCols, destCols), mask);
ch2 = vandq_u32(destCols, vmvnq_u32(mask));
return vorrq_u32(ch1, ch2);
case kArgbToRgbBlender: // see BITMAP member function blendArgbToRgb
setupArgbAlphas();
return rgbBlendSIMD(srcCols, destCols, vshrq_n_u32(srcCols, 24), false);
case kRgbToArgbBlender: // see BITMAP member function blendRgbToArgb
// if alpha is NOT 0 or 255
ch2 = vandq_u32(srcCols, vmovq_n_u32(0x00ffffff));
ch2 = vorrq_u32(ch2, vshlq_n_u32(alphas, 24));
ch2 = argbBlendSIMD(ch2, destCols);
// if alpha is 0 or 255
ch1 = vorrq_u32(srcCols, vmovq_n_u32(0xff000000));
// mask and or them together
mask = vorrq_u32(vceqq_u32(alphas, vmovq_n_u32(0)), vceqq_u32(alphas, vmovq_n_u32(0xff)));
ch1 = vandq_u32(ch1, mask);
ch2 = vandq_u32(ch2, vmvnq_u32(mask));
return vorrq_u32(ch1, ch2);
case kRgbToRgbBlender: // see BITMAP member function blendRgbToRgb
return rgbBlendSIMD(srcCols, destCols, alphas, false);
case kAlphaPreservedBlenderMode: // see BITMAP member function blendPreserveAlpha
return rgbBlendSIMD(srcCols, destCols, alphas, true);
case kOpaqueBlenderMode: // see BITMAP member function blendOpaque
return vorrq_u32(srcCols, vmovq_n_u32(0xff000000));
case kAdditiveBlenderMode: // see BITMAP member function blendAdditiveAlpha
srcAlphas = vaddq_u32(vshrq_n_u32(srcCols, 24), vshrq_n_u32(destCols, 24));
srcAlphas = vminq_u32(srcAlphas, vmovq_n_u32(0xff));
srcCols = vandq_u32(srcCols, vmovq_n_u32(0x00ffffff));
return vorrq_u32(srcCols, vshlq_n_u32(srcAlphas, 24));
case kTintBlenderMode: // see BITMAP member function blendTintSprite
return blendTintSpriteSIMD(srcCols, destCols, alphas, false);
case kTintLightBlenderMode: // see BITMAP member function blendTintSprite
return blendTintSpriteSIMD(srcCols, destCols, alphas, true);
}
return srcCols;
}
static inline uint16x8_t blendPixelSIMD2Bpp(uint16x8_t srcCols, uint16x8_t destCols, uint16x8_t alphas) {
uint16x8_t mask, ch1, ch2;
switch (_G(_blender_mode)) {
case kSourceAlphaBlender:
case kOpaqueBlenderMode:
case kAdditiveBlenderMode:
return srcCols;
case kArgbToArgbBlender:
case kArgbToRgbBlender:
ch1 = vandq_u16(vmovq_n_u16(0xff), vceqq_u16(alphas, vmovq_n_u16(0)));
ch2 = vandq_u16(alphas, vcgtq_u16(alphas, vmovq_n_u16(0)));
alphas = vorrq_u16(ch1, ch2);
// fall through
case kRgbToRgbBlender:
case kAlphaPreservedBlenderMode:
return rgbBlendSIMD2Bpp(srcCols, destCols, alphas);
case kRgbToArgbBlender:
mask = vorrq_u16(vceqq_u16(alphas, vmovq_n_u16(0)), vceqq_u16(alphas, vmovq_n_u16(255)));
ch1 = vandq_u16(srcCols, mask);
ch2 = vandq_u16(rgbBlendSIMD2Bpp(srcCols, destCols, alphas), vmvnq_u16(mask));
return vorrq_u16(ch1, ch2);
case kTintBlenderMode:
case kTintLightBlenderMode:
uint32x4_t srcColsLo = simd2BppTo4Bpp(vget_low_u16(srcCols));
uint32x4_t srcColsHi = simd2BppTo4Bpp(vget_high_u16(srcCols));
uint32x4_t destColsLo = simd2BppTo4Bpp(vget_low_u16(destCols));
uint32x4_t destColsHi = simd2BppTo4Bpp(vget_high_u16(destCols));
uint32x4_t alphasLo = vmovl_u16(vget_low_u16(alphas));
uint32x4_t alphasHi = vmovl_u16(vget_high_u16(alphas));
uint16x4_t lo = simd4BppTo2Bpp(blendTintSpriteSIMD(srcColsLo, destColsLo, alphasLo, _G(_blender_mode) == kTintLightBlenderMode));
uint16x4_t hi = simd4BppTo2Bpp(blendTintSpriteSIMD(srcColsHi, destColsHi, alphasHi, _G(_blender_mode) == kTintLightBlenderMode));
return vcombine_u16(lo, hi);
}
return srcCols;
}
template<int DestBytesPerPixel, int SrcBytesPerPixel>
static inline void drawPixelSIMD(byte *destPtr, const byte *srcP2, uint32x4_t tint, uint32x4_t alphas, uint32x4_t maskedAlphas, uint32x4_t transColors, int xDir, int xCtrBpp, int srcAlpha, int skipTrans, bool horizFlip, bool useTint, uint32x4_t skipMask) {
uint32x4_t srcCols, destCol;
if (DestBytesPerPixel == 4)
destCol = vld1q_u32((uint32 *)destPtr);
else
destCol = simd2BppTo4Bpp(vld1_u16((uint16 *)destPtr));
if (SrcBytesPerPixel == 4)
srcCols = vld1q_u32((const uint32 *)(srcP2 + xDir * xCtrBpp));
else
srcCols = simd2BppTo4Bpp(vld1_u16((const uint16 *)(srcP2 + xDir * xCtrBpp)));
// we do this here because we need to check if we should skip the pixel before we blend it
uint32x4_t mask1 = skipTrans ? vceqq_u32(vandq_u32(srcCols, maskedAlphas), transColors) : vmovq_n_u32(0);
mask1 = vorrq_u32(mask1, skipMask);
if (srcAlpha != -1) {
// take into account for useTint
if (useTint) {
srcCols = blendPixelSIMD(tint, srcCols, alphas);
} else {
srcCols = blendPixelSIMD(srcCols, destCol, alphas);
}
}
uint32x4_t destCols2 = vandq_u32(destCol, mask1);
uint32x4_t srcCols2 = vandq_u32(srcCols, vmvnq_u32(mask1));
uint32x4_t final = vorrq_u32(destCols2, srcCols2);
if (horizFlip) {
final = vrev64q_u32(final);
final = vcombine_u32(vget_high_u32(final), vget_low_u32(final));
}
if (DestBytesPerPixel == 4) {
vst1q_u32((uint32 *)destPtr, final);
} else {
vst1_u16((uint16 *)destPtr, simd4BppTo2Bpp(final));
}
}
static inline void drawPixelSIMD2Bpp(byte *destPtr, const byte *srcP2, uint16x8_t tint, uint16x8_t alphas, uint16x8_t transColors, int xDir, int xCtrBpp, int srcAlpha, int skipTrans, bool horizFlip, bool useTint, uint16x8_t skipMask) {
uint16x8_t destCol = vld1q_u16((uint16 *)destPtr);
uint16x8_t srcCols = vld1q_u16((const uint16 *)(srcP2 + xDir * xCtrBpp));
uint16x8_t mask1 = skipTrans ? vceqq_u16(srcCols, transColors) : vmovq_n_u16(0);
mask1 = vorrq_u16(mask1, skipMask);
if (srcAlpha != -1) {
// take into account for useTint
if (useTint) {
srcCols = blendPixelSIMD2Bpp(tint, srcCols, alphas);
} else {
srcCols = blendPixelSIMD2Bpp(srcCols, destCol, alphas);
}
}
uint16x8_t destCols2 = vandq_u16(destCol, mask1);
uint16x8_t srcCols2 = vandq_u16(srcCols, vmvnq_u16(mask1));
uint16x8_t final = vorrq_u16(destCols2, srcCols2);
if (horizFlip) {
final = vrev64q_u16(final);
final = vcombine_u16(vget_high_u16(final), vget_low_u16(final));
}
vst1q_u16((uint16 *)destPtr, final);
}
public:
// This template handles 2bpp and 4bpp, the other specializations handle 1bpp and format conversion blits
template<int DestBytesPerPixel, int SrcBytesPerPixel, bool Scale>
static void drawInner4BppWithConv(BITMAP::DrawInnerArgs &args) {
const int xDir = args.horizFlip ? -1 : 1;
byte rSrc, gSrc, bSrc, aSrc;
byte rDest = 0, gDest = 0, bDest = 0, aDest = 0;
uint32x4_t tint = vshlq_n_u32(vdupq_n_u32(args.srcAlpha), 24);
tint = vorrq_u32(tint, vshlq_n_u32(vdupq_n_u32(args.tintRed), 16));
tint = vorrq_u32(tint, vshlq_n_u32(vdupq_n_u32(args.tintGreen), 8));
tint = vorrq_u32(tint, vdupq_n_u32(args.tintBlue));
uint32x4_t maskedAlphas = vmovq_n_u32(args.alphaMask);
uint32x4_t transColors = vmovq_n_u32(args.transColor);
uint32x4_t alphas = vmovq_n_u32(args.srcAlpha);
// This is so that we can calculate what pixels to crop off in a vectorized way
const uint32x4_t addIndexesNormal = {0, 1, 2, 3};
const uint32x4_t addIndexesFlipped = {3, 2, 1, 0};
uint32x4_t addIndexes = args.horizFlip ? addIndexesFlipped : addIndexesNormal;
// This is so that we can calculate in parallel the pixel indexes for scaled drawing
uint32x4_t scaleAdds = {0, (uint32)args.scaleX, (uint32)args.scaleX*2, (uint32)args.scaleX*3};
// Clip the bounds ahead of time (so we don't waste time checking if we are in bounds when
// we are in the inner loop)
int xCtrStart = 0, xCtrBppStart = 0, xCtrWidth = args.dstRect.width();
if (args.xStart + xCtrWidth > args.destArea.w) {
xCtrWidth = args.destArea.w - args.xStart;
}
if (args.xStart < 0) {
xCtrStart = -args.xStart;
xCtrBppStart = xCtrStart * SrcBytesPerPixel;
args.xStart = 0;
}
int destY = args.yStart, srcYCtr = 0, yCtr = 0, scaleYCtr = 0, yCtrHeight = args.dstRect.height();
if (Scale) yCtrHeight = args.dstRect.height();
if (args.yStart < 0) {
yCtr = -args.yStart;
destY = 0;
if (Scale) {
scaleYCtr = yCtr * args.scaleY;
srcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
}
}
if (args.yStart + yCtrHeight > args.destArea.h) {
yCtrHeight = args.destArea.h - args.yStart;
}
/*if (!Scale && xCtrWidth % 4 != 0) {
--yCtrHeight;
}*/
const int secondToLast = xCtrWidth - 4;
byte *destP = (byte *)args.destArea.getBasePtr(0, destY);
const byte *srcP = (const byte *)args.src.getBasePtr(
args.horizFlip ? args.srcArea.right - 4 : args.srcArea.left,
args.vertFlip ? args.srcArea.bottom - 1 - yCtr : args.srcArea.top + yCtr);
for (; yCtr < yCtrHeight; ++destY, ++yCtr, scaleYCtr += args.scaleY) {
uint32x4_t xCtrWidthSIMD = vdupq_n_u32(xCtrWidth); // This is the width of the row
if (!Scale) {
int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart;
for (; xCtr < secondToLast; destX += 4, xCtr += 4, xCtrBpp += SrcBytesPerPixel*4) {
byte *destPtr = &destP[destX * DestBytesPerPixel];
drawPixelSIMD<DestBytesPerPixel, SrcBytesPerPixel>(destPtr, srcP, tint, alphas, maskedAlphas, transColors, xDir, xCtrBpp, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u32(0));
}
byte *destPtr = &destP[destX * DestBytesPerPixel];
uint32x4_t srcCols = vmovq_n_u32(0);
uint32x4_t destCols = vmovq_n_u32(0);
memcpy(&srcCols, srcP + xDir * xCtrBpp, (xCtrWidth - xCtr) * SrcBytesPerPixel);
memcpy(&destCols, destPtr, (xCtrWidth - xCtr) * DestBytesPerPixel);
// Skip pixels that are beyond the row
// uint32x4_t skipMask = vcgeq_u32(vaddq_u32(vdupq_n_u32(xCtr), addIndexes), xCtrWidthSIMD);
drawPixelSIMD<DestBytesPerPixel, SrcBytesPerPixel>((byte *)&destCols, (byte *)&srcCols, tint, alphas, maskedAlphas, transColors, xDir, 0, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u32(0));
memcpy(destPtr, &destCols, (xCtrWidth - xCtr) * DestBytesPerPixel);
// Goto next row in source and destination image
destP += args.destArea.pitch;
srcP += args.vertFlip ? -args.src.pitch : args.src.pitch;
} else {
// Here we are scaling the image
int newSrcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
// Since the source yctr might not update every row of the destination, we have
// to see if we are on a new row...
if (srcYCtr != newSrcYCtr) {
int diffSrcYCtr = newSrcYCtr - srcYCtr; // Have we moved yet
srcP += args.src.pitch * diffSrcYCtr;
srcYCtr = newSrcYCtr;
}
// Now also since we might skip a pixel or 2 or duplicate one to reach the desired
// scaling size, we create a small dummy buffer that we copy the pixels into and then
// call the drawPixelsSIMD function
byte srcBuffer[4*4];
for (int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart, scaleXCtr = xCtrStart * args.scaleX; xCtr < xCtrWidth; destX += 4, xCtr += 4, xCtrBpp += SrcBytesPerPixel*4) {
if (yCtr + 1 == yCtrHeight && xCtr + 4 > xCtrWidth) break; // Don't go past the last 4 pixels
uint32x4_t indexes = vdupq_n_u32(scaleXCtr);
// Calculate in parallel the indexes of the pixels
indexes = vmulq_n_u32(vshrq_n_u32(vaddq_u32(indexes, scaleAdds), BITMAP::SCALE_THRESHOLD_BITS), SrcBytesPerPixel);
// Simply memcpy them in. memcpy has no real performance overhead here
memcpy(&srcBuffer[0*(uintptr_t)SrcBytesPerPixel], srcP + vgetq_lane_u32(indexes, 0), SrcBytesPerPixel);
memcpy(&srcBuffer[1*(uintptr_t)SrcBytesPerPixel], srcP + vgetq_lane_u32(indexes, 1), SrcBytesPerPixel);
memcpy(&srcBuffer[2*(uintptr_t)SrcBytesPerPixel], srcP + vgetq_lane_u32(indexes, 2), SrcBytesPerPixel);
memcpy(&srcBuffer[3*(uintptr_t)SrcBytesPerPixel], srcP + vgetq_lane_u32(indexes, 3), SrcBytesPerPixel);
scaleXCtr += args.scaleX*4;
// Now this is pretty much the same as before with non-scaled code, except that we use
// our dummy source buffer instead of the actual source bitmap
byte *destPtr = &destP[destX * (uintptr_t)DestBytesPerPixel];
uint32x4_t skipMask = vcgeq_u32(vaddq_u32(vdupq_n_u32(xCtr), addIndexes), xCtrWidthSIMD);
drawPixelSIMD<DestBytesPerPixel, SrcBytesPerPixel>(destPtr, (const byte *)srcBuffer, tint, alphas, maskedAlphas, transColors, 1, 0, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, skipMask);
}
// We calculate every row here except the last (because then we need to
// check for if we fall off the edge of the row)
// The only exception here is scaling drawing this is because:
// 1) if statements are costly, and the less we do the faster this loop is
// 2) with this, the only branch in the normal drawing loop is the width check
// 3) the scaling code will actually draw until the last 4 pixels of the image
// and do the extra if checks because the scaling code is already much slower
// than the normal drawing loop, and the less duplicate code helps here.
if (yCtr + 1 != yCtrHeight) destP += args.destArea.pitch;
}
}
// Get the last x values of the last row
int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart;
// We have a picture that is a multiple of 4, so no extra pixels to draw
/*if (xCtrWidth % 4 == 0)*/ return;
// Drawing the last few not scaled pixels here.
// Same as the loop above but now we check if we are going to overflow,
// and thus we don't need to mask out pixels that go over the row.
if (!Scale) {
for (; xCtr + 4 < xCtrWidth; destX += 4, xCtr += 4, xCtrBpp += SrcBytesPerPixel*4) {
byte *destPtr = &destP[(ptrdiff_t)destX * DestBytesPerPixel];
drawPixelSIMD<DestBytesPerPixel, SrcBytesPerPixel>(destPtr, srcP, tint, alphas, maskedAlphas, transColors, xDir, xCtrBpp, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u32(0));
}
// Because we move in 4 pixel units, and horizFlip moves in 1, we have to move
// 1 pixel past the last pixel we did not blit, meaning going forward 3 pixels.
if (args.horizFlip) srcP += SrcBytesPerPixel * 3;
} else {
// So if we are scaling, set up the xCtr to what it was before (AKA the last 4 or so pixels of the image)
xCtr = xCtrWidth - xCtrWidth % 4;
xCtrBpp = xCtr * SrcBytesPerPixel;
destX = args.xStart+xCtr;
}
// For the last 4 pixels, we just do them in serial, nothing special
for (; xCtr < xCtrWidth; ++destX, ++xCtr, xCtrBpp += SrcBytesPerPixel) {
const byte *srcColPtr = (const byte *)(srcP + xDir * xCtrBpp);
if (Scale) {
srcColPtr = (const byte *)(srcP + (xCtr * args.scaleX) / BITMAP::SCALE_THRESHOLD * SrcBytesPerPixel);
}
byte *destVal = (byte *)&destP[destX * DestBytesPerPixel];
uint32 srcCol = args.dstBitmap.getColor(srcColPtr, SrcBytesPerPixel);
// Check if this is a transparent color we should skip
if (args.skipTrans && ((srcCol & args.alphaMask) == args.transColor))
continue;
args.src.format.colorToARGB(srcCol, aSrc, rSrc, gSrc, bSrc);
if (args.srcAlpha != -1) {
if (args.useTint) {
rDest = rSrc;
gDest = gSrc;
bDest = bSrc;
aDest = aSrc;
rSrc = args.tintRed;
gSrc = args.tintGreen;
bSrc = args.tintBlue;
aSrc = args.srcAlpha;
}
args.dstBitmap.blendPixel(aSrc, rSrc, gSrc, bSrc, aDest, rDest, gDest, bDest, args.srcAlpha, args.useTint, destVal);
srcCol = args.dstBitmap.format.ARGBToColor(aDest, rDest, gDest, bDest);
} else {
srcCol = args.dstBitmap.format.ARGBToColor(aSrc, rSrc, gSrc, bSrc);
}
if (DestBytesPerPixel == 4)
*(uint32 *)destVal = srcCol;
else
*(uint16 *)destVal = srcCol;
}
}
template<bool Scale>
static void drawInner2Bpp(BITMAP::DrawInnerArgs &args) {
const int xDir = args.horizFlip ? -1 : 1;
byte rSrc, gSrc, bSrc, aSrc;
byte rDest = 0, gDest = 0, bDest = 0, aDest = 0;
uint16x8_t tint = vdupq_n_u16(args.src.format.ARGBToColor(args.srcAlpha, args.tintRed, args.tintGreen, args.tintBlue));
uint16x8_t transColors = vdupq_n_u16(args.transColor);
uint16x8_t alphas = vdupq_n_u16(args.srcAlpha);
// This is so that we can calculate what pixels to crop off in a vectorized way
uint16x8_t addIndexesNormal = {0, 1, 2, 3, 4, 5, 6, 7};
uint16x8_t addIndexesFlipped = {7, 6, 5, 4, 3, 2, 1, 0};
uint16x8_t addIndexes = args.horizFlip ? addIndexesFlipped : addIndexesNormal;
// This is so that we can calculate in parallel the pixel indices for scaled drawing
uint32x4_t scaleAdds = {0, (uint32)args.scaleX, (uint32)args.scaleX*2, (uint32)args.scaleX*3};
uint32x4_t scaleAdds2 = {(uint32)args.scaleX*4, (uint32)args.scaleX*5, (uint32)args.scaleX*6, (uint32)args.scaleX*7};
// Clip the bounds ahead of time (so we don't waste time checking if we are in bounds when
// we are in the inner loop)
int xCtrStart = 0, xCtrBppStart = 0, xCtrWidth = args.dstRect.width();
if (args.xStart + xCtrWidth > args.destArea.w) {
xCtrWidth = args.destArea.w - args.xStart;
}
if (args.xStart < 0) {
xCtrStart = -args.xStart;
xCtrBppStart = xCtrStart * 2;
args.xStart = 0;
}
int destY = args.yStart, yCtr = 0, srcYCtr = 0, scaleYCtr = 0, yCtrHeight = args.dstRect.height();
if (Scale) yCtrHeight = args.dstRect.height();
if (args.yStart < 0) {
yCtr = -args.yStart;
destY = 0;
if (Scale) {
scaleYCtr = yCtr * args.scaleY;
srcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
}
}
if (args.yStart + yCtrHeight > args.destArea.h) {
yCtrHeight = args.destArea.h - args.yStart;
}
/*if (!Scale && xCtrWidth % 8 != 0) {
--yCtrHeight;
}*/
const int secondToLast = xCtrWidth - 8;
byte *destP = (byte *)args.destArea.getBasePtr(0, destY);
const byte *srcP = (const byte *)args.src.getBasePtr(
args.horizFlip ? args.srcArea.right - 8 : args.srcArea.left,
args.vertFlip ? args.srcArea.bottom - 1 - yCtr : args.srcArea.top + yCtr);
for (; yCtr < yCtrHeight; ++destY, ++yCtr, scaleYCtr += args.scaleY) {
uint16x8_t xCtrWidthSIMD = vmovq_n_u16(xCtrWidth); // This is the width of the row
if (!Scale) {
// If we are not scaling the image
int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart;
for (; xCtr < secondToLast; destX += 8, xCtr += 8, xCtrBpp += 16) {
byte *destPtr = &destP[destX * 2];
drawPixelSIMD2Bpp(destPtr, srcP, tint, alphas, transColors, xDir, xCtrBpp, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u16(0));
}
byte *destPtr = &destP[destX * 2];
uint16x8_t srcCols = vmovq_n_u16(0);
uint16x8_t destCols = vmovq_n_u16(0);
const int copySize = (xCtrWidth - xCtr) * 2;
memcpy(&srcCols, srcP + xDir * xCtrBpp, copySize);
memcpy(&destCols, destPtr, copySize);
// Skip pixels that are beyond the row
// uint16x8_t skipMask = vcgeq_u16(vaddq_u16(vdupq_n_u16(xCtr), addIndexes), xCtrWidthSIMD);
drawPixelSIMD2Bpp((byte *)&destCols, (byte *)&srcCols, tint, alphas, transColors, xDir, 0, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u16(0));
memcpy(destPtr, &destCols, copySize);
// Goto next row in source and destination image
destP += args.destArea.pitch;
srcP += args.vertFlip ? -args.src.pitch : args.src.pitch;
} else {
// Here we are scaling the image
int newSrcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
// Since the source yctr might not update every row of the destination, we have
// to see if we are on a new row...
if (srcYCtr != newSrcYCtr) {
int diffSrcYCtr = newSrcYCtr - srcYCtr;
srcP += args.src.pitch * diffSrcYCtr;
srcYCtr = newSrcYCtr;
}
// Now also since we might skip a pixel or 2 or duplicate one to reach the desired
// scaling size, we create a small dummy buffer that we copy the pixels into and then
// call the drawPixelsSIMD function
uint16 srcBuffer[8];
for (int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart, scaleXCtr = xCtrStart * args.scaleX; xCtr < xCtrWidth; destX += 8, xCtr += 8, xCtrBpp += 16) {
if (yCtr + 1 == yCtrHeight && xCtr + 8 > xCtrWidth) break;
uint32x4_t indexes = vdupq_n_u32(scaleXCtr), indexes2 = vdupq_n_u32(scaleXCtr);
// Calculate in parallel the indices of the pixels
indexes = vmulq_n_u32(vshrq_n_u32(vaddq_u32(indexes, scaleAdds), BITMAP::SCALE_THRESHOLD_BITS), 2);
indexes2 = vmulq_n_u32(vshrq_n_u32(vaddq_u32(indexes2, scaleAdds2), BITMAP::SCALE_THRESHOLD_BITS), 2);
// Simply memcpy them in. memcpy has no real performance overhead here
srcBuffer[0] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes, 0));
srcBuffer[1] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes, 1));
srcBuffer[2] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes, 2));
srcBuffer[3] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes, 3));
srcBuffer[4] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes2, 0));
srcBuffer[5] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes2, 1));
srcBuffer[6] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes2, 2));
srcBuffer[7] = *(const uint16 *)(srcP + vgetq_lane_u32(indexes2, 3));
scaleXCtr += args.scaleX*8;
// Now this is pretty much the same as before with non-scaled code, except that we use
// our dummy source buffer instead of the actual source bitmap
byte *destPtr = &destP[destX * 2];
uint16x8_t skipMask = vcgeq_u16(vaddq_u16(vdupq_n_u16(xCtr), addIndexes), xCtrWidthSIMD);
drawPixelSIMD2Bpp(destPtr, (const byte *)srcBuffer, tint, alphas, transColors, 1, 0, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, skipMask);
}
// We calculate every row here except the last (because then we need to
// check for if we fall off the edge of the row)
// The only exception here is scaling drawing this is because:
// 1) if statements are costly, and the less we do the faster this loop is
// 2) with this, the only branch in the normal drawing loop is the width check
// 3) the scaling code will actually draw until the last 4 pixels of the image
// and do the extra if checks because the scaling code is already much slower
// than the normal drawing loop, and the less duplicate code helps here.
if (yCtr + 1 != yCtrHeight) destP += args.destArea.pitch;
}
}
// We have a picture that is a multiple of 8, so no extra pixels to draw
/*if (xCtrWidth % 8 == 0)*/ return;
// Get the last x values of the last row
int xCtr = xCtrStart, xCtrBpp = xCtrBppStart, destX = args.xStart;
// Drawing the last few not scaled pixels here.
// Same as the loop above but now we check if we are going to overflow,
// and thus we don't need to mask out pixels that go over the row.
if (!Scale) {
for (; xCtr + 8 < xCtrWidth; destX += 8, xCtr += 8, xCtrBpp += 16) {
byte *destPtr = &destP[destX * 2];
drawPixelSIMD2Bpp(destPtr, srcP, tint, alphas, transColors, xDir, xCtrBpp, args.srcAlpha, args.skipTrans, args.horizFlip, args.useTint, vmovq_n_u16(0));
}
// Because we move in 8 pixel units, and horizFlip moves in 1, we have to move
// 1 pixel past the last pixel we did not blit, meaning going forward 7 pixels.
if (args.horizFlip) srcP += 2 * 7;
} else {
// So if we are scaling, set up the xCtr to what it was before (AKA the last 8 or so pixels of the image)
xCtr = xCtrWidth - xCtrWidth % 8;
xCtrBpp = xCtr * 2;
destX = args.xStart+xCtr;
}
// For the last 4 pixels, we just do them in serial, nothing special
for (; xCtr < xCtrWidth; ++destX, ++xCtr, xCtrBpp += 2) {
const byte *srcColPtr = (const byte *)(srcP + xDir * xCtrBpp);
if (Scale) {
srcColPtr = (const byte *)(srcP + (xCtr * args.scaleX) / BITMAP::SCALE_THRESHOLD * 2);
}
byte *destVal = (byte *)&destP[destX * 2];
uint32 srcCol = (uint32)(*(const uint16 *)srcColPtr);
// Check if this is a transparent color we should skip
if (args.skipTrans && srcCol == args.transColor)
continue;
args.src.format.colorToARGB(srcCol, aSrc, rSrc, gSrc, bSrc);
if (args.srcAlpha != -1) {
if (args.useTint) {
rDest = rSrc;
gDest = gSrc;
bDest = bSrc;
aDest = aSrc;
rSrc = args.tintRed;
gSrc = args.tintGreen;
bSrc = args.tintBlue;
aSrc = args.srcAlpha;
}/* else {
format.colorToARGB((uint32)(*(uint16 *)destVal), aDest, rDest, gDest, bDest);
}*/
args.dstBitmap.blendPixel(aSrc, rSrc, gSrc, bSrc, aDest, rDest, gDest, bDest, args.srcAlpha, args.useTint, destVal);
srcCol = args.dstBitmap.format.ARGBToColor(aDest, rDest, gDest, bDest);
} else {
srcCol = args.dstBitmap.format.ARGBToColor(aSrc, rSrc, gSrc, bSrc);
}
*(uint16 *)destVal = srcCol;
}
}
template<bool Scale>
static void drawInner1Bpp(BITMAP::DrawInnerArgs &args) {
const int xDir = args.horizFlip ? -1 : 1;
uint8x16_t transColors = vmovq_n_u8(args.transColor);
// This is so that we can calculate in parallel the pixel indices for scaled drawing
uint32x4_t scaleAdds1 = {0, (uint32)args.scaleX, (uint32)args.scaleX*2, (uint32)args.scaleX*3};
uint32x4_t scaleAdds2 = {(uint32)args.scaleX*4, (uint32)args.scaleX*5, (uint32)args.scaleX*6, (uint32)args.scaleX*7};
uint32x4_t scaleAdds3 = {(uint32)args.scaleX*8, (uint32)args.scaleX*9, (uint32)args.scaleX*10, (uint32)args.scaleX*11};
uint32x4_t scaleAdds4 = {(uint32)args.scaleX*12, (uint32)args.scaleX*13, (uint32)args.scaleX*14, (uint32)args.scaleX*15};
// Clip the bounds ahead of time (so we don't waste time checking if we are in bounds when
// we are in the inner loop)
int xCtrStart = 0, xCtrWidth = args.dstRect.width();
if (args.xStart + xCtrWidth > args.destArea.w) {
xCtrWidth = args.destArea.w - args.xStart;
}
if (args.xStart < 0) {
xCtrStart = -args.xStart;
args.xStart = 0;
}
int destY = args.yStart, yCtr = 0, srcYCtr = 0, scaleYCtr = 0, yCtrHeight = args.dstRect.height();
if (Scale) yCtrHeight = args.dstRect.height();
if (args.yStart < 0) {
yCtr = -args.yStart;
destY = 0;
if (Scale) {
scaleYCtr = yCtr * args.scaleY;
srcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
}
}
if (args.yStart + yCtrHeight > args.destArea.h) {
yCtrHeight = args.destArea.h - args.yStart;
}
byte *destP = (byte *)args.destArea.getBasePtr(0, destY);
const byte *srcP = (const byte *)args.src.getBasePtr(
args.horizFlip ? args.srcArea.right - 16 : args.srcArea.left,
args.vertFlip ? args.srcArea.bottom - 1 - yCtr : args.srcArea.top + yCtr);
for (; yCtr < yCtrHeight; ++destY, ++yCtr, scaleYCtr += args.scaleY) {
if (Scale) {
// So here we update the srcYCtr differently due to this being for
// scaling
int newSrcYCtr = scaleYCtr / BITMAP::SCALE_THRESHOLD;
if (srcYCtr != newSrcYCtr) {
// Since the source yctr might not update every row of the destination, we have
// to see if we are on a new row...
int diffSrcYCtr = newSrcYCtr - srcYCtr;
srcP += args.src.pitch * diffSrcYCtr;
srcYCtr = newSrcYCtr;
}
}
int xCtr = xCtrStart, destX = args.xStart, scaleXCtr = xCtrStart * args.scaleX;
for (; xCtr + 16 < xCtrWidth; destX += 16, xCtr += 16) {
byte *destPtr = &destP[destX];
// Here we don't use the drawPixelSIMD function because 1bpp bitmaps in allegro
// can't have any blending applied to them
uint8x16_t destCols = vld1q_u8(destPtr);
uint8x16_t srcCols = vld1q_u8(srcP + xDir * xCtr);
if (Scale) {
// If we are scaling, we have to set each pixel individually
uint32x4_t indexes1 = vdupq_n_u32(scaleXCtr), indexes2 = vdupq_n_u32(scaleXCtr);
uint32x4_t indexes3 = vdupq_n_u32(scaleXCtr), indexes4 = vdupq_n_u32(scaleXCtr);
indexes1 = vshrq_n_u32(vaddq_u32(indexes1, scaleAdds1), BITMAP::SCALE_THRESHOLD_BITS);
indexes2 = vshrq_n_u32(vaddq_u32(indexes2, scaleAdds2), BITMAP::SCALE_THRESHOLD_BITS);
indexes3 = vshrq_n_u32(vaddq_u32(indexes3, scaleAdds3), BITMAP::SCALE_THRESHOLD_BITS);
indexes4 = vshrq_n_u32(vaddq_u32(indexes4, scaleAdds4), BITMAP::SCALE_THRESHOLD_BITS);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes1, 0)], srcCols, 0);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes1, 1)], srcCols, 1);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes1, 2)], srcCols, 2);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes1, 3)], srcCols, 3);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes2, 0)], srcCols, 4);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes2, 1)], srcCols, 5);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes2, 2)], srcCols, 6);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes2, 3)], srcCols, 7);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes3, 0)], srcCols, 8);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes3, 1)], srcCols, 9);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes3, 2)], srcCols, 10);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes3, 3)], srcCols, 11);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes4, 0)], srcCols, 12);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes4, 1)], srcCols, 13);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes4, 2)], srcCols, 14);
srcCols = vsetq_lane_u8(srcP[vgetq_lane_u32(indexes4, 3)], srcCols, 15);
scaleXCtr += args.scaleX*16;
}
// Mask out transparent pixels
uint8x16_t mask1 = args.skipTrans ? vceqq_u8(srcCols, transColors) : vmovq_n_u8(0);
uint8x16_t final = vorrq_u8(vandq_u8(srcCols, vmvnq_u8(mask1)), vandq_u8(destCols, mask1));
if (args.horizFlip) {
final = vrev64q_u8(final);
final = vcombine_u8(vget_high_u8(final), vget_low_u8(final));
}
vst1q_u8(destPtr, final);
}
// Get the last x values
// Because we move in 16 pixel units, and horizFlip moves in 1, we have to move
// 1 pixel past the last pixel we did not blit, meaning going forward 15 pixels.
if (args.horizFlip) srcP += 15;
for (; xCtr < xCtrWidth; ++destX, ++xCtr, scaleXCtr += args.scaleX) {
const byte *srcCol = (const byte *)(srcP + xDir * xCtr);
if (Scale) {
srcCol = (const byte *)(srcP + scaleXCtr / BITMAP::SCALE_THRESHOLD);
}
// Check if this is a transparent color we should skip
if (args.skipTrans && *srcCol == args.transColor)
continue;
byte *destVal = (byte *)&destP[destX];
*destVal = *srcCol;
}
if (args.horizFlip) srcP -= 15; // Undo what we did up there
destP += args.destArea.pitch; // Go to next row
// Only advance the src row by 1 every time like this if we don't scale
if (!Scale) srcP += args.vertFlip ? -args.src.pitch : args.src.pitch;
}
}
}; // end of class DrawInnerImpl_NEON
template<bool Scale>
void BITMAP::drawNEON(DrawInnerArgs &args) {
if (args.sameFormat) {
switch (format.bytesPerPixel) {
case 1: DrawInnerImpl_NEON::drawInner1Bpp<Scale>(args); break;
case 2: DrawInnerImpl_NEON::drawInner2Bpp<Scale>(args); break;
case 4: DrawInnerImpl_NEON::drawInner4BppWithConv<4, 4, Scale>(args); break;
}
} else if (format.bytesPerPixel == 4 && args.src.format.bytesPerPixel == 2) {
DrawInnerImpl_NEON::drawInner4BppWithConv<4, 2, Scale>(args);
} else if (format.bytesPerPixel == 2 && args.src.format.bytesPerPixel == 4) {
DrawInnerImpl_NEON::drawInner4BppWithConv<2, 4, Scale>(args);
}
}
template void BITMAP::drawNEON<false>(DrawInnerArgs &);
template void BITMAP::drawNEON<true>(DrawInnerArgs &);
} // namespace AGS3
#if !defined(__aarch64__) && !defined(__ARM_NEON)
#if defined(__clang__)
#pragma clang attribute pop
#elif defined(__GNUC__)
#pragma GCC pop_options
#endif
#endif // !defined(__aarch64__) && !defined(__ARM_NEON)
#endif // SCUMMVM_NEON
|