… | |
… | |
40 | |
40 | |
41 | #ifndef ECB_H |
41 | #ifndef ECB_H |
42 | #define ECB_H |
42 | #define ECB_H |
43 | |
43 | |
44 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
45 | #define ECB_VERSION 0x00010009 |
45 | #define ECB_VERSION 0x0001000c |
46 | |
46 | |
47 | #include <string.h> /* for memcpy */ |
47 | #include <string.h> /* for memcpy */ |
48 | |
48 | |
49 | #if defined (_WIN32) && !defined (__MINGW32__) |
49 | #if defined (_WIN32) && !defined (__MINGW32__) |
50 | typedef signed char int8_t; |
50 | typedef signed char int8_t; |
… | |
… | |
454 | /* count trailing zero bits and count # of one bits */ |
454 | /* count trailing zero bits and count # of one bits */ |
455 | #if ECB_GCC_VERSION(3,4) \ |
455 | #if ECB_GCC_VERSION(3,4) \ |
456 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
456 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
457 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
457 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
458 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
458 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
459 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
|
|
460 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
|
|
461 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
|
|
462 | #define ecb_ctz32(x) __builtin_ctz (x) |
459 | #define ecb_ctz32(x) __builtin_ctz (x) |
|
|
460 | #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x)) |
463 | #define ecb_ctz64(x) __builtin_ctzll (x) |
461 | #define ecb_clz32(x) __builtin_clz (x) |
|
|
462 | #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x)) |
|
|
463 | #define ecb_ld32(x) (ecb_clz32 (x) ^ 31) |
|
|
464 | #define ecb_ld64(x) (ecb_clz64 (x) ^ 63) |
464 | #define ecb_popcount32(x) __builtin_popcount (x) |
465 | #define ecb_popcount32(x) __builtin_popcount (x) |
465 | /* no popcountll */ |
466 | /* ecb_popcount64 is more difficult, see below */ |
466 | #else |
467 | #else |
467 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
468 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
468 | ecb_function_ ecb_const int |
469 | ecb_function_ ecb_const int |
469 | ecb_ctz32 (uint32_t x) |
470 | ecb_ctz32 (uint32_t x) |
470 | { |
471 | { |
471 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
472 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
472 | unsigned long r; |
473 | unsigned long r; |
473 | _BitScanForward (&r, x); |
474 | _BitScanForward (&r, x); |
474 | return (int)r; |
475 | return (int)r; |
475 | #else |
476 | #else |
476 | int r = 0; |
477 | int r; |
477 | |
478 | |
478 | x &= ~x + 1; /* this isolates the lowest bit */ |
479 | x &= ~x + 1; /* this isolates the lowest bit */ |
479 | |
480 | |
480 | #if ECB_branchless_on_i386 |
481 | #if 1 |
|
|
482 | /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */ |
|
|
483 | /* This happens to return 32 for x == 0, but the API does not support this */ |
|
|
484 | |
|
|
485 | /* -0 marks unused entries */ |
|
|
486 | static unsigned char table[64] = |
|
|
487 | { |
|
|
488 | 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14, |
|
|
489 | 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15, |
|
|
490 | 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26, |
|
|
491 | 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0 |
|
|
492 | }; |
|
|
493 | |
|
|
494 | /* magic constant results in 33 unique values in the upper 6 bits */ |
|
|
495 | x *= 0x0450fbafU; /* == 17 * 65 * 65535 */ |
|
|
496 | |
|
|
497 | r = table [x >> 26]; |
|
|
498 | #elif 0 /* branchless on i386, typically */ |
|
|
499 | r = 0; |
481 | r += !!(x & 0xaaaaaaaa) << 0; |
500 | r += !!(x & 0xaaaaaaaa) << 0; |
482 | r += !!(x & 0xcccccccc) << 1; |
501 | r += !!(x & 0xcccccccc) << 1; |
483 | r += !!(x & 0xf0f0f0f0) << 2; |
502 | r += !!(x & 0xf0f0f0f0) << 2; |
484 | r += !!(x & 0xff00ff00) << 3; |
503 | r += !!(x & 0xff00ff00) << 3; |
485 | r += !!(x & 0xffff0000) << 4; |
504 | r += !!(x & 0xffff0000) << 4; |
486 | #else |
505 | #else /* branchless on modern compilers, typically */ |
|
|
506 | r = 0; |
487 | if (x & 0xaaaaaaaa) r += 1; |
507 | if (x & 0xaaaaaaaa) r += 1; |
488 | if (x & 0xcccccccc) r += 2; |
508 | if (x & 0xcccccccc) r += 2; |
489 | if (x & 0xf0f0f0f0) r += 4; |
509 | if (x & 0xf0f0f0f0) r += 4; |
490 | if (x & 0xff00ff00) r += 8; |
510 | if (x & 0xff00ff00) r += 8; |
491 | if (x & 0xffff0000) r += 16; |
511 | if (x & 0xffff0000) r += 16; |
… | |
… | |
507 | int shift = x & 0xffffffff ? 0 : 32; |
527 | int shift = x & 0xffffffff ? 0 : 32; |
508 | return ecb_ctz32 (x >> shift) + shift; |
528 | return ecb_ctz32 (x >> shift) + shift; |
509 | #endif |
529 | #endif |
510 | } |
530 | } |
511 | |
531 | |
|
|
532 | ecb_function_ ecb_const int ecb_clz32 (uint32_t x); |
|
|
533 | ecb_function_ ecb_const int |
|
|
534 | ecb_clz32 (uint32_t x) |
|
|
535 | { |
|
|
536 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
537 | unsigned long r; |
|
|
538 | _BitScanReverse (&r, x); |
|
|
539 | return (int)r; |
|
|
540 | #else |
|
|
541 | |
|
|
542 | /* Robert Harley's algorithm from comp.arch 1996-12-07 */ |
|
|
543 | /* This happens to return 32 for x == 0, but the API does not support this */ |
|
|
544 | |
|
|
545 | /* -0 marks unused table elements */ |
|
|
546 | static unsigned char table[64] = |
|
|
547 | { |
|
|
548 | 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0, |
|
|
549 | -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0, |
|
|
550 | 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18, |
|
|
551 | 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0 |
|
|
552 | }; |
|
|
553 | |
|
|
554 | /* propagate leftmost 1 bit to the right */ |
|
|
555 | x |= x >> 1; |
|
|
556 | x |= x >> 2; |
|
|
557 | x |= x >> 4; |
|
|
558 | x |= x >> 8; |
|
|
559 | x |= x >> 16; |
|
|
560 | |
|
|
561 | /* magic constant results in 33 unique values in the upper 6 bits */ |
|
|
562 | x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */ |
|
|
563 | |
|
|
564 | return table [x >> 26]; |
|
|
565 | #endif |
|
|
566 | } |
|
|
567 | |
|
|
568 | ecb_function_ ecb_const int ecb_clz64 (uint64_t x); |
|
|
569 | ecb_function_ ecb_const int |
|
|
570 | ecb_clz64 (uint64_t x) |
|
|
571 | { |
|
|
572 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
573 | unsigned long r; |
|
|
574 | _BitScanReverse64 (&r, x); |
|
|
575 | return (int)r; |
|
|
576 | #else |
|
|
577 | uint32_t l = x >> 32; |
|
|
578 | int shift = l ? 0 : 32; |
|
|
579 | return ecb_clz32 (l ? l : x) + shift; |
|
|
580 | #endif |
|
|
581 | } |
|
|
582 | |
512 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
583 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
513 | ecb_function_ ecb_const int |
584 | ecb_function_ ecb_const int |
514 | ecb_popcount32 (uint32_t x) |
585 | ecb_popcount32 (uint32_t x) |
515 | { |
586 | { |
516 | x -= (x >> 1) & 0x55555555; |
587 | x -= (x >> 1) & 0x55555555; |
… | |
… | |
591 | x = ( x >> 16 ) | ( x << 16); |
662 | x = ( x >> 16 ) | ( x << 16); |
592 | |
663 | |
593 | return x; |
664 | return x; |
594 | } |
665 | } |
595 | |
666 | |
596 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
|
|
597 | /* so for this version we are lazy */ |
|
|
598 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
667 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
599 | ecb_function_ ecb_const int |
668 | ecb_function_ ecb_const int |
600 | ecb_popcount64 (uint64_t x) |
669 | ecb_popcount64 (uint64_t x) |
601 | { |
670 | { |
|
|
671 | /* popcount64 is only available on 64 bit cpus as gcc builtin. */ |
|
|
672 | /* also, gcc/clang make this surprisingly difficult to use */ |
|
|
673 | #if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl)) |
|
|
674 | return __builtin_popcountl (x); |
|
|
675 | #else |
602 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
676 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
|
|
677 | #endif |
603 | } |
678 | } |
604 | |
679 | |
605 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
680 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
606 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
681 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
607 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
682 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
… | |
… | |
609 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
684 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
610 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
685 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
611 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
686 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
612 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
687 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
613 | |
688 | |
614 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
689 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); } |
615 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
690 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); } |
616 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
691 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); } |
617 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
692 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); } |
618 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
693 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); } |
619 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
694 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); } |
620 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
695 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); } |
621 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
696 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); } |
622 | |
697 | |
623 | #if ECB_CPP |
698 | #if ECB_CPP |
624 | |
699 | |
625 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
700 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
626 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
701 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
… | |
… | |
774 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
849 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
775 | |
850 | |
776 | ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } |
851 | ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } |
777 | ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } |
852 | ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } |
778 | ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } |
853 | ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } |
779 | |
854 | |
780 | ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } |
855 | ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } |
781 | ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } |
856 | ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } |
782 | ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } |
857 | ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } |
783 | |
858 | |
784 | #if ECB_CPP |
859 | #if ECB_CPP |
… | |
… | |
805 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
880 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
806 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
881 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
807 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
882 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
808 | |
883 | |
809 | #endif |
884 | #endif |
|
|
885 | |
|
|
886 | /*****************************************************************************/ |
|
|
887 | /* pointer/integer hashing */ |
|
|
888 | |
|
|
889 | /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */ |
|
|
890 | ecb_function_ uint32_t ecb_mix32 (uint32_t v); |
|
|
891 | ecb_function_ uint32_t ecb_mix32 (uint32_t v) |
|
|
892 | { |
|
|
893 | v ^= v >> 16; v *= 0x7feb352dU; |
|
|
894 | v ^= v >> 15; v *= 0x846ca68bU; |
|
|
895 | v ^= v >> 16; |
|
|
896 | return v; |
|
|
897 | } |
|
|
898 | |
|
|
899 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v); |
|
|
900 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v) |
|
|
901 | { |
|
|
902 | v ^= v >> 16 ; v *= 0x43021123U; |
|
|
903 | v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U; |
|
|
904 | v ^= v >> 16 ; |
|
|
905 | return v; |
|
|
906 | } |
|
|
907 | |
|
|
908 | /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */ |
|
|
909 | ecb_function_ uint64_t ecb_mix64 (uint64_t v); |
|
|
910 | ecb_function_ uint64_t ecb_mix64 (uint64_t v) |
|
|
911 | { |
|
|
912 | v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U; |
|
|
913 | v ^= v >> 27; v *= 0x94d049bb133111ebU; |
|
|
914 | v ^= v >> 31; |
|
|
915 | return v; |
|
|
916 | } |
|
|
917 | |
|
|
918 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v); |
|
|
919 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v) |
|
|
920 | { |
|
|
921 | v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U; |
|
|
922 | v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U; |
|
|
923 | v ^= v >> 30 ^ v >> 60; |
|
|
924 | return v; |
|
|
925 | } |
|
|
926 | |
|
|
927 | ecb_function_ uintptr_t ecb_ptrmix (void *p); |
|
|
928 | ecb_function_ uintptr_t ecb_ptrmix (void *p) |
|
|
929 | { |
|
|
930 | #if ECB_PTRSIZE <= 4 |
|
|
931 | return ecb_mix32 ((uint32_t)p); |
|
|
932 | #else |
|
|
933 | return ecb_mix64 ((uint64_t)p); |
|
|
934 | #endif |
|
|
935 | } |
|
|
936 | |
|
|
937 | ecb_function_ void *ecb_ptrunmix (uintptr_t v); |
|
|
938 | ecb_function_ void *ecb_ptrunmix (uintptr_t v) |
|
|
939 | { |
|
|
940 | #if ECB_PTRSIZE <= 4 |
|
|
941 | return (void *)ecb_unmix32 (v); |
|
|
942 | #else |
|
|
943 | return (void *)ecb_unmix64 (v); |
|
|
944 | #endif |
|
|
945 | } |
|
|
946 | |
|
|
947 | #if ECB_CPP |
|
|
948 | |
|
|
949 | template<typename T> |
|
|
950 | inline uintptr_t ecb_ptrmix (T *p) |
|
|
951 | { |
|
|
952 | return ecb_ptrmix (static_cast<void *>(p)); |
|
|
953 | } |
|
|
954 | |
|
|
955 | template<typename T> |
|
|
956 | inline T *ecb_ptrunmix (uintptr_t v) |
|
|
957 | { |
|
|
958 | return static_cast<T *>(ecb_ptrunmix (v)); |
|
|
959 | } |
|
|
960 | |
|
|
961 | #endif |
|
|
962 | |
|
|
963 | /*****************************************************************************/ |
|
|
964 | /* gray code */ |
|
|
965 | |
|
|
966 | ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); } |
|
|
967 | ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); } |
|
|
968 | ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); } |
|
|
969 | ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); } |
|
|
970 | |
|
|
971 | ecb_function_ uint8_t ecb_gray8_decode (uint8_t g) |
|
|
972 | { |
|
|
973 | g ^= g >> 1; |
|
|
974 | g ^= g >> 2; |
|
|
975 | g ^= g >> 4; |
|
|
976 | |
|
|
977 | return g; |
|
|
978 | } |
|
|
979 | |
|
|
980 | ecb_function_ uint16_t ecb_gray16_decode (uint16_t g) |
|
|
981 | { |
|
|
982 | g ^= g >> 1; |
|
|
983 | g ^= g >> 2; |
|
|
984 | g ^= g >> 4; |
|
|
985 | g ^= g >> 8; |
|
|
986 | |
|
|
987 | return g; |
|
|
988 | } |
|
|
989 | |
|
|
990 | ecb_function_ uint32_t ecb_gray32_decode (uint32_t g) |
|
|
991 | { |
|
|
992 | g ^= g >> 1; |
|
|
993 | g ^= g >> 2; |
|
|
994 | g ^= g >> 4; |
|
|
995 | g ^= g >> 8; |
|
|
996 | g ^= g >> 16; |
|
|
997 | |
|
|
998 | return g; |
|
|
999 | } |
|
|
1000 | |
|
|
1001 | ecb_function_ uint64_t ecb_gray64_decode (uint64_t g) |
|
|
1002 | { |
|
|
1003 | g ^= g >> 1; |
|
|
1004 | g ^= g >> 2; |
|
|
1005 | g ^= g >> 4; |
|
|
1006 | g ^= g >> 8; |
|
|
1007 | g ^= g >> 16; |
|
|
1008 | g ^= g >> 32; |
|
|
1009 | |
|
|
1010 | return g; |
|
|
1011 | } |
|
|
1012 | |
|
|
1013 | #if ECB_CPP |
|
|
1014 | |
|
|
1015 | ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); } |
|
|
1016 | ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); } |
|
|
1017 | ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); } |
|
|
1018 | ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); } |
|
|
1019 | |
|
|
1020 | ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); } |
|
|
1021 | ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); } |
|
|
1022 | ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); } |
|
|
1023 | ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); } |
|
|
1024 | |
|
|
1025 | #endif |
|
|
1026 | |
|
|
1027 | /*****************************************************************************/ |
|
|
1028 | /* 2d hilbert curves */ |
|
|
1029 | |
|
|
1030 | /* algorithm from the book Hacker's Delight, modified to not */ |
|
|
1031 | /* run into undefined behaviour for n==16 */ |
|
|
1032 | static uint32_t |
|
|
1033 | ecb_hilbert2d_index_to_coord32 (int n, uint32_t s) |
|
|
1034 | { |
|
|
1035 | uint32_t comp, swap, cs, t, sr; |
|
|
1036 | |
|
|
1037 | /* pad s on the left (unused) bits with 01 (no change groups) */ |
|
|
1038 | s |= 0x55555555U << n << n; |
|
|
1039 | /* "s shift right" */ |
|
|
1040 | sr = (s >> 1) & 0x55555555U; |
|
|
1041 | /* compute complement and swap info in two-bit groups */ |
|
|
1042 | cs = ((s & 0x55555555U) + sr) ^ 0x55555555U; |
|
|
1043 | |
|
|
1044 | /* parallel prefix xor op to propagate both complement |
|
|
1045 | * and swap info together from left to right (there is |
|
|
1046 | * no step "cs ^= cs >> 1", so in effect it computes |
|
|
1047 | * two independent parallel prefix operations on two |
|
|
1048 | * interleaved sets of sixteen bits). |
|
|
1049 | */ |
|
|
1050 | cs ^= cs >> 2; |
|
|
1051 | cs ^= cs >> 4; |
|
|
1052 | cs ^= cs >> 8; |
|
|
1053 | cs ^= cs >> 16; |
|
|
1054 | |
|
|
1055 | /* separate swap and complement bits */ |
|
|
1056 | swap = cs & 0x55555555U; |
|
|
1057 | comp = (cs >> 1) & 0x55555555U; |
|
|
1058 | |
|
|
1059 | /* calculate coordinates in odd and even bit positions */ |
|
|
1060 | t = (s & swap) ^ comp; |
|
|
1061 | s = s ^ sr ^ t ^ (t << 1); |
|
|
1062 | |
|
|
1063 | /* unpad/clear out any junk on the left */ |
|
|
1064 | s = s & ((1 << n << n) - 1); |
|
|
1065 | |
|
|
1066 | /* Now "unshuffle" to separate the x and y bits. */ |
|
|
1067 | t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1); |
|
|
1068 | t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2); |
|
|
1069 | t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4); |
|
|
1070 | t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8); |
|
|
1071 | |
|
|
1072 | /* now s contains two 16-bit coordinates */ |
|
|
1073 | return s; |
|
|
1074 | } |
|
|
1075 | |
|
|
1076 | /* 64 bit, a straightforward extension to the 32 bit case */ |
|
|
1077 | static uint64_t |
|
|
1078 | ecb_hilbert2d_index_to_coord64 (int n, uint64_t s) |
|
|
1079 | { |
|
|
1080 | uint64_t comp, swap, cs, t, sr; |
|
|
1081 | |
|
|
1082 | /* pad s on the left (unused) bits with 01 (no change groups) */ |
|
|
1083 | s |= 0x5555555555555555U << n << n; |
|
|
1084 | /* "s shift right" */ |
|
|
1085 | sr = (s >> 1) & 0x5555555555555555U; |
|
|
1086 | /* compute complement and swap info in two-bit groups */ |
|
|
1087 | cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U; |
|
|
1088 | |
|
|
1089 | /* parallel prefix xor op to propagate both complement |
|
|
1090 | * and swap info together from left to right (there is |
|
|
1091 | * no step "cs ^= cs >> 1", so in effect it computes |
|
|
1092 | * two independent parallel prefix operations on two |
|
|
1093 | * interleaved sets of thirty-two bits). |
|
|
1094 | */ |
|
|
1095 | cs ^= cs >> 2; |
|
|
1096 | cs ^= cs >> 4; |
|
|
1097 | cs ^= cs >> 8; |
|
|
1098 | cs ^= cs >> 16; |
|
|
1099 | cs ^= cs >> 32; |
|
|
1100 | |
|
|
1101 | /* separate swap and complement bits */ |
|
|
1102 | swap = cs & 0x5555555555555555U; |
|
|
1103 | comp = (cs >> 1) & 0x5555555555555555U; |
|
|
1104 | |
|
|
1105 | /* calculate coordinates in odd and even bit positions */ |
|
|
1106 | t = (s & swap) ^ comp; |
|
|
1107 | s = s ^ sr ^ t ^ (t << 1); |
|
|
1108 | |
|
|
1109 | /* unpad/clear out any junk on the left */ |
|
|
1110 | s = s & ((1 << n << n) - 1); |
|
|
1111 | |
|
|
1112 | /* Now "unshuffle" to separate the x and y bits. */ |
|
|
1113 | t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1); |
|
|
1114 | t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2); |
|
|
1115 | t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4); |
|
|
1116 | t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8); |
|
|
1117 | t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16); |
|
|
1118 | |
|
|
1119 | /* now s contains two 32-bit coordinates */ |
|
|
1120 | return s; |
|
|
1121 | } |
|
|
1122 | |
|
|
1123 | /* algorithm from the book Hacker's Delight, but a similar algorithm*/ |
|
|
1124 | /* is given in https://doi.org/10.1002/spe.4380160103 */ |
|
|
1125 | /* this has been slightly improved over the original version */ |
|
|
1126 | ecb_function_ uint32_t |
|
|
1127 | ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy) |
|
|
1128 | { |
|
|
1129 | uint32_t row; |
|
|
1130 | uint32_t state = 0; |
|
|
1131 | uint32_t s = 0; |
|
|
1132 | |
|
|
1133 | do |
|
|
1134 | { |
|
|
1135 | --n; |
|
|
1136 | |
|
|
1137 | row = 4 * state |
|
|
1138 | | (2 & (xy >> n >> 15)) |
|
|
1139 | | (1 & (xy >> n )); |
|
|
1140 | |
|
|
1141 | /* these funky constants are lookup tables for two-bit values */ |
|
|
1142 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
|
|
1143 | state = (0x8fe65831U >> 2 * row) & 3; |
|
|
1144 | } |
|
|
1145 | while (n > 0); |
|
|
1146 | |
|
|
1147 | return s; |
|
|
1148 | } |
|
|
1149 | |
|
|
1150 | /* 64 bit, essentially the same as 32 bit */ |
|
|
1151 | ecb_function_ uint64_t |
|
|
1152 | ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy) |
|
|
1153 | { |
|
|
1154 | uint32_t row; |
|
|
1155 | uint32_t state = 0; |
|
|
1156 | uint64_t s = 0; |
|
|
1157 | |
|
|
1158 | do |
|
|
1159 | { |
|
|
1160 | --n; |
|
|
1161 | |
|
|
1162 | row = 4 * state |
|
|
1163 | | (2 & (xy >> n >> 31)) |
|
|
1164 | | (1 & (xy >> n )); |
|
|
1165 | |
|
|
1166 | /* these funky constants are lookup tables for two-bit values */ |
|
|
1167 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
|
|
1168 | state = (0x8fe65831U >> 2 * row) & 3; |
|
|
1169 | } |
|
|
1170 | while (n > 0); |
|
|
1171 | |
|
|
1172 | return s; |
|
|
1173 | } |
810 | |
1174 | |
811 | /*****************************************************************************/ |
1175 | /*****************************************************************************/ |
812 | /* division */ |
1176 | /* division */ |
813 | |
1177 | |
814 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
1178 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
… | |
… | |
948 | } |
1312 | } |
949 | |
1313 | |
950 | /*******************************************************************************/ |
1314 | /*******************************************************************************/ |
951 | /* fast integer to ascii */ |
1315 | /* fast integer to ascii */ |
952 | |
1316 | |
|
|
1317 | /* |
|
|
1318 | * This code is pretty complicated because it is general. The idea behind it, |
|
|
1319 | * however, is pretty simple: first, the number is multiplied with a scaling |
|
|
1320 | * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point |
|
|
1321 | * number with the first digit in the upper bits. |
|
|
1322 | * Then this digit is converted to text and masked out. The resulting number |
|
|
1323 | * is then multiplied by 10, by multiplying the fixed point representation |
|
|
1324 | * by 5 and shifting the (binary) decimal point one to the right, so a 4.28 |
|
|
1325 | * format becomes 5.27, 6.26 and so on. |
|
|
1326 | * The rest involves only advancing the pointer if we already generated a |
|
|
1327 | * non-zero digit, so leading zeroes are overwritten. |
|
|
1328 | */ |
|
|
1329 | |
953 | // simply return a mask with "bits" bits set |
1330 | /* simply return a mask with "bits" bits set */ |
954 | #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) |
1331 | #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) |
955 | |
1332 | |
956 | // oputput a single digit. maskvalue is 10**digitidx |
1333 | /* oputput a single digit. maskvalue is 10**digitidx */ |
957 | #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ |
1334 | #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ |
958 | if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ |
1335 | if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ |
959 | { \ |
1336 | { \ |
960 | char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ |
1337 | char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ |
961 | *ptr = digit + '0'; /* output it */ \ |
1338 | *ptr = digit + '0'; /* output it */ \ |
962 | nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ |
1339 | nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ |
963 | ptr += nz; /* output digit only if non-zero digit seen */ \ |
1340 | ptr += nz; /* output digit only if non-zero digit seen */ \ |
964 | x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ |
1341 | x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ |
965 | } |
1342 | } |
966 | |
1343 | |
967 | // convert integer to fixed point format and multiply out digits, highest first |
1344 | /* convert integer to fixed point format and multiply out digits, highest first */ |
968 | // requires magic constants: max. digits and number of bits after the decimal point |
1345 | /* requires magic constants: max. digits and number of bits after the decimal point */ |
969 | #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ |
1346 | #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ |
970 | ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ |
1347 | ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ |
971 | { \ |
1348 | { \ |
972 | char nz = lz; /* non-zero digit seen? */ \ |
1349 | char nz = lz; /* non-zero digit seen? */ \ |
973 | /* convert to x.bits fixed-point */ \ |
1350 | /* convert to x.bits fixed-point */ \ |
… | |
… | |
984 | ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ |
1361 | ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ |
985 | ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ |
1362 | ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ |
986 | return ptr; \ |
1363 | return ptr; \ |
987 | } |
1364 | } |
988 | |
1365 | |
989 | // predefined versions of the above, for various digits |
1366 | /* predefined versions of the above, for various digits */ |
990 | // ecb_i2a_xN = almost N digits, limit defined by macro |
1367 | /* ecb_i2a_xN = almost N digits, limit defined by macro */ |
991 | // ecb_i2a_N = up to N digits, leading zeroes suppressed |
1368 | /* ecb_i2a_N = up to N digits, leading zeroes suppressed */ |
992 | // ecb_i2a_0N = exactly N digits, including leading zeroes |
1369 | /* ecb_i2a_0N = exactly N digits, including leading zeroes */ |
993 | |
1370 | |
994 | // non-leading-zero versions, limited range |
1371 | /* non-leading-zero versions, limited range */ |
995 | #define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5 |
1372 | #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */ |
996 | #define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10 |
1373 | #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */ |
997 | ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) |
1374 | ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) |
998 | ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) |
1375 | ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) |
999 | |
1376 | |
1000 | // non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit |
1377 | /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */ |
1001 | ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) |
1378 | ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) |
1002 | ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) |
1379 | ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) |
1003 | ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) |
1380 | ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) |
1004 | ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) |
1381 | ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) |
1005 | ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) |
1382 | ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) |
1006 | ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) |
1383 | ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) |
1007 | ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) |
1384 | ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) |
1008 | ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) |
1385 | ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) |
1009 | |
1386 | |
1010 | // leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit |
1387 | /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */ |
1011 | ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) |
1388 | ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) |
1012 | ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) |
1389 | ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) |
1013 | ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) |
1390 | ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) |
1014 | ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) |
1391 | ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) |
1015 | ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) |
1392 | ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) |
… | |
… | |
1027 | ecb_i2a_u32 (char *ptr, uint32_t u) |
1404 | ecb_i2a_u32 (char *ptr, uint32_t u) |
1028 | { |
1405 | { |
1029 | #if ECB_64BIT_NATIVE |
1406 | #if ECB_64BIT_NATIVE |
1030 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
1407 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
1031 | ptr = ecb_i2a_x10 (ptr, u); |
1408 | ptr = ecb_i2a_x10 (ptr, u); |
1032 | else // x10 almost, but not fully, covers 32 bit |
1409 | else /* x10 almost, but not fully, covers 32 bit */ |
1033 | { |
1410 | { |
1034 | uint32_t u1 = u % 1000000000; |
1411 | uint32_t u1 = u % 1000000000; |
1035 | uint32_t u2 = u / 1000000000; |
1412 | uint32_t u2 = u / 1000000000; |
1036 | |
1413 | |
1037 | *ptr++ = u2 + '0'; |
1414 | *ptr++ = u2 + '0'; |
… | |
… | |
1069 | { |
1446 | { |
1070 | *ptr = '-'; ptr += v < 0; |
1447 | *ptr = '-'; ptr += v < 0; |
1071 | uint32_t u = v < 0 ? -(uint32_t)v : v; |
1448 | uint32_t u = v < 0 ? -(uint32_t)v : v; |
1072 | |
1449 | |
1073 | #if ECB_64BIT_NATIVE |
1450 | #if ECB_64BIT_NATIVE |
1074 | ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit |
1451 | ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */ |
1075 | #else |
1452 | #else |
1076 | ptr = ecb_i2a_u32 (ptr, u); |
1453 | ptr = ecb_i2a_u32 (ptr, u); |
1077 | #endif |
1454 | #endif |
1078 | |
1455 | |
1079 | return ptr; |
1456 | return ptr; |
… | |
… | |
1142 | uint64_t u1 = u % 1000000000; |
1519 | uint64_t u1 = u % 1000000000; |
1143 | uint64_t ua = u / 1000000000; |
1520 | uint64_t ua = u / 1000000000; |
1144 | uint64_t u2 = ua % 1000000000; |
1521 | uint64_t u2 = ua % 1000000000; |
1145 | uint64_t u3 = ua / 1000000000; |
1522 | uint64_t u3 = ua / 1000000000; |
1146 | |
1523 | |
1147 | // 2**31 is 19 digits, so the top is exactly one digit |
1524 | /* 2**31 is 19 digits, so the top is exactly one digit */ |
1148 | *ptr++ = u3 + '0'; |
1525 | *ptr++ = u3 + '0'; |
1149 | ptr = ecb_i2a_09 (ptr, u2); |
1526 | ptr = ecb_i2a_09 (ptr, u2); |
1150 | ptr = ecb_i2a_09 (ptr, u1); |
1527 | ptr = ecb_i2a_09 (ptr, u1); |
1151 | } |
1528 | } |
1152 | #else |
1529 | #else |