ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.192 by root, Mon Jun 21 23:59:58 2021 UTC vs.
Revision 1.210 by root, Fri Mar 25 15:28:08 2022 UTC

… …
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;
… …
355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
356#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) 358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
359 359
360/* This marks larger functions that do not neccessarily need to be inlined */
361/* TODO: popssibly static would be best for these at the moment? */
360#define ecb_function_ ecb_inline 362#define ecb_function_ ecb_inline
361 363
362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) 364#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
363 #define ecb_attribute(attrlist) __attribute__ (attrlist) 365 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364#else 366#else
… …
454/* count trailing zero bits and count # of one bits */ 456/* count trailing zero bits and count # of one bits */
455#if ECB_GCC_VERSION(3,4) \ 457#if ECB_GCC_VERSION(3,4) \
456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ 458 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ 459 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458 && ECB_CLANG_BUILTIN(__builtin_popcount)) 460 && 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) 461 #define ecb_ctz32(x) __builtin_ctz (x)
462 #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
463 #define ecb_ctz64(x) __builtin_ctzll (x) 463 #define ecb_clz32(x) __builtin_clz (x)
464 #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
465 #define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
466 #define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
464 #define ecb_popcount32(x) __builtin_popcount (x) 467 #define ecb_popcount32(x) __builtin_popcount (x)
465 /* no popcountll */ 468 /* ecb_popcount64 is more difficult, see below */
466#else 469#else
467 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); 470 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
468 ecb_function_ ecb_const int 471 ecb_function_ ecb_const int
469 ecb_ctz32 (uint32_t x) 472 ecb_ctz32 (uint32_t x)
470 { 473 {
471#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) 474#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
472 unsigned long r; 475 unsigned long r;
473 _BitScanForward (&r, x); 476 _BitScanForward (&r, x);
474 return (int)r; 477 return (int)r;
475#else 478#else
476 int r = 0; 479 int r;
477 480
478 x &= ~x + 1; /* this isolates the lowest bit */ 481 x &= ~x + 1; /* this isolates the lowest bit */
479 482
480#if ECB_branchless_on_i386 483 #if 1
484 /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
485 /* This happens to return 32 for x == 0, but the API does not support this */
486
487 /* -0 marks unused entries */
488 static unsigned char table[64] =
489 {
490 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
491 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
492 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
493 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
494 };
495
496 /* magic constant results in 33 unique values in the upper 6 bits */
497 x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
498
499 r = table [x >> 26];
500 #elif 0 /* branchless on i386, typically */
501 r = 0;
481 r += !!(x & 0xaaaaaaaa) << 0; 502 r += !!(x & 0xaaaaaaaa) << 0;
482 r += !!(x & 0xcccccccc) << 1; 503 r += !!(x & 0xcccccccc) << 1;
483 r += !!(x & 0xf0f0f0f0) << 2; 504 r += !!(x & 0xf0f0f0f0) << 2;
484 r += !!(x & 0xff00ff00) << 3; 505 r += !!(x & 0xff00ff00) << 3;
485 r += !!(x & 0xffff0000) << 4; 506 r += !!(x & 0xffff0000) << 4;
486#else 507 #else /* branchless on modern compilers, typically */
508 r = 0;
487 if (x & 0xaaaaaaaa) r += 1; 509 if (x & 0xaaaaaaaa) r += 1;
488 if (x & 0xcccccccc) r += 2; 510 if (x & 0xcccccccc) r += 2;
489 if (x & 0xf0f0f0f0) r += 4; 511 if (x & 0xf0f0f0f0) r += 4;
490 if (x & 0xff00ff00) r += 8; 512 if (x & 0xff00ff00) r += 8;
491 if (x & 0xffff0000) r += 16; 513 if (x & 0xffff0000) r += 16;
… …
507 int shift = x & 0xffffffff ? 0 : 32; 529 int shift = x & 0xffffffff ? 0 : 32;
508 return ecb_ctz32 (x >> shift) + shift; 530 return ecb_ctz32 (x >> shift) + shift;
509#endif 531#endif
510 } 532 }
511 533
534 ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
535 ecb_function_ ecb_const int
536 ecb_clz32 (uint32_t x)
537 {
538#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
539 unsigned long r;
540 _BitScanReverse (&r, x);
541 return (int)r;
542#else
543
544 /* Robert Harley's algorithm from comp.arch 1996-12-07 */
545 /* This happens to return 32 for x == 0, but the API does not support this */
546
547 /* -0 marks unused table elements */
548 static unsigned char table[64] =
549 {
550 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
551 -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
552 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
553 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
554 };
555
556 /* propagate leftmost 1 bit to the right */
557 x |= x >> 1;
558 x |= x >> 2;
559 x |= x >> 4;
560 x |= x >> 8;
561 x |= x >> 16;
562
563 /* magic constant results in 33 unique values in the upper 6 bits */
564 x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
565
566 return table [x >> 26];
567#endif
568 }
569
570 ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
571 ecb_function_ ecb_const int
572 ecb_clz64 (uint64_t x)
573 {
574#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
575 unsigned long r;
576 _BitScanReverse64 (&r, x);
577 return (int)r;
578#else
579 uint32_t l = x >> 32;
580 int shift = l ? 0 : 32;
581 return ecb_clz32 (l ? l : x) + shift;
582#endif
583 }
584
512 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); 585 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
513 ecb_function_ ecb_const int 586 ecb_function_ ecb_const int
514 ecb_popcount32 (uint32_t x) 587 ecb_popcount32 (uint32_t x)
515 { 588 {
516 x -= (x >> 1) & 0x55555555; 589 x -= (x >> 1) & 0x55555555;
… …
591 x = ( x >> 16 ) | ( x << 16); 664 x = ( x >> 16 ) | ( x << 16);
592 665
593 return x; 666 return x;
594} 667}
595 668
596/* popcount64 is only available on 64 bit cpus as gcc builtin */
597/* so for this version we are lazy */
598ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); 669ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
599ecb_function_ ecb_const int 670ecb_function_ ecb_const int
600ecb_popcount64 (uint64_t x) 671ecb_popcount64 (uint64_t x)
601{ 672{
673 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
674 /* also, gcc/clang make this surprisingly difficult to use */
675#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
676 return __builtin_popcountl (x);
677#else
602 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 678 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
679#endif
603} 680}
604 681
605ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); 682ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
606ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); 683ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
607ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); 684ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
… …
609ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); 686ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
610ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); 687ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
611ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); 688ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
612ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); 689ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
613 690
614ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 691ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
615ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 692ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
616ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 693ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
617ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 694ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
618ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 695ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
619ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 696ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
620ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 697ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
621ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 698ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
622 699
623#if ECB_CPP 700#if ECB_CPP
624 701
625inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } 702inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } 703inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
… …
774ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } 851ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775 852
776ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } 853ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } 854ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } 855ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
779 856
780ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } 857ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } 858ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } 859ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
783 860
784#if ECB_CPP 861#if ECB_CPP
… …
805template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } 882template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } 883template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } 884template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
808 885
809#endif 886#endif
887
888/*****************************************************************************/
889/* pointer/integer hashing */
890
891/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
892ecb_function_ uint32_t ecb_mix32 (uint32_t v);
893ecb_function_ uint32_t ecb_mix32 (uint32_t v)
894{
895 v ^= v >> 16; v *= 0x7feb352dU;
896 v ^= v >> 15; v *= 0x846ca68bU;
897 v ^= v >> 16;
898 return v;
899}
900
901ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
902ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
903{
904 v ^= v >> 16 ; v *= 0x43021123U;
905 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
906 v ^= v >> 16 ;
907 return v;
908}
909
910/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
911ecb_function_ uint64_t ecb_mix64 (uint64_t v);
912ecb_function_ uint64_t ecb_mix64 (uint64_t v)
913{
914 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
915 v ^= v >> 27; v *= 0x94d049bb133111ebU;
916 v ^= v >> 31;
917 return v;
918}
919
920ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
921ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
922{
923 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
924 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
925 v ^= v >> 30 ^ v >> 60;
926 return v;
927}
928
929ecb_function_ uintptr_t ecb_ptrmix (void *p);
930ecb_function_ uintptr_t ecb_ptrmix (void *p)
931{
932 #if ECB_PTRSIZE <= 4
933 return ecb_mix32 ((uint32_t)p);
934 #else
935 return ecb_mix64 ((uint64_t)p);
936 #endif
937}
938
939ecb_function_ void *ecb_ptrunmix (uintptr_t v);
940ecb_function_ void *ecb_ptrunmix (uintptr_t v)
941{
942 #if ECB_PTRSIZE <= 4
943 return (void *)ecb_unmix32 (v);
944 #else
945 return (void *)ecb_unmix64 (v);
946 #endif
947}
948
949#if ECB_CPP
950
951template<typename T>
952inline uintptr_t ecb_ptrmix (T *p)
953{
954 return ecb_ptrmix (static_cast<void *>(p));
955}
956
957template<typename T>
958inline T *ecb_ptrunmix (uintptr_t v)
959{
960 return static_cast<T *>(ecb_ptrunmix (v));
961}
962
963#endif
964
965/*****************************************************************************/
966/* gray code */
967
968ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
969ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
970ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
971ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
972
973ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g);
974ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g)
975{
976 g ^= g >> 1;
977 g ^= g >> 2;
978 g ^= g >> 4;
979
980 return g;
981}
982
983ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g);
984ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g)
985{
986 g ^= g >> 1;
987 g ^= g >> 2;
988 g ^= g >> 4;
989 g ^= g >> 8;
990
991 return g;
992}
993
994ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g);
995ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g)
996{
997 g ^= g >> 1;
998 g ^= g >> 2;
999 g ^= g >> 4;
1000 g ^= g >> 8;
1001 g ^= g >> 16;
1002
1003 return g;
1004}
1005
1006ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g);
1007ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g)
1008{
1009 g ^= g >> 1;
1010 g ^= g >> 2;
1011 g ^= g >> 4;
1012 g ^= g >> 8;
1013 g ^= g >> 16;
1014 g ^= g >> 32;
1015
1016 return g;
1017}
1018
1019#if ECB_CPP
1020
1021ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
1022ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
1023ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
1024ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
1025
1026ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
1027ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
1028ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
1029ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
1030
1031#endif
1032
1033/*****************************************************************************/
1034/* 2d hilbert curves */
1035
1036/* algorithm from the book Hacker's Delight, modified to not */
1037/* run into undefined behaviour for n==16 */
1038static uint32_t
1039ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
1040{
1041 uint32_t comp, swap, cs, t, sr;
1042
1043 /* pad s on the left (unused) bits with 01 (no change groups) */
1044 s |= 0x55555555U << n << n;
1045 /* "s shift right" */
1046 sr = (s >> 1) & 0x55555555U;
1047 /* compute complement and swap info in two-bit groups */
1048 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
1049
1050 /* parallel prefix xor op to propagate both complement
1051 * and swap info together from left to right (there is
1052 * no step "cs ^= cs >> 1", so in effect it computes
1053 * two independent parallel prefix operations on two
1054 * interleaved sets of sixteen bits).
1055 */
1056 cs ^= cs >> 2;
1057 cs ^= cs >> 4;
1058 cs ^= cs >> 8;
1059 cs ^= cs >> 16;
1060
1061 /* separate swap and complement bits */
1062 swap = cs & 0x55555555U;
1063 comp = (cs >> 1) & 0x55555555U;
1064
1065 /* calculate coordinates in odd and even bit positions */
1066 t = (s & swap) ^ comp;
1067 s = s ^ sr ^ t ^ (t << 1);
1068
1069 /* unpad/clear out any junk on the left */
1070 s = s & ((1 << n << n) - 1);
1071
1072 /* Now "unshuffle" to separate the x and y bits. */
1073 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1074 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1075 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1076 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1077
1078 /* now s contains two 16-bit coordinates */
1079 return s;
1080}
1081
1082/* 64 bit, a straightforward extension to the 32 bit case */
1083static uint64_t
1084ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1085{
1086 uint64_t comp, swap, cs, t, sr;
1087
1088 /* pad s on the left (unused) bits with 01 (no change groups) */
1089 s |= 0x5555555555555555U << n << n;
1090 /* "s shift right" */
1091 sr = (s >> 1) & 0x5555555555555555U;
1092 /* compute complement and swap info in two-bit groups */
1093 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1094
1095 /* parallel prefix xor op to propagate both complement
1096 * and swap info together from left to right (there is
1097 * no step "cs ^= cs >> 1", so in effect it computes
1098 * two independent parallel prefix operations on two
1099 * interleaved sets of thirty-two bits).
1100 */
1101 cs ^= cs >> 2;
1102 cs ^= cs >> 4;
1103 cs ^= cs >> 8;
1104 cs ^= cs >> 16;
1105 cs ^= cs >> 32;
1106
1107 /* separate swap and complement bits */
1108 swap = cs & 0x5555555555555555U;
1109 comp = (cs >> 1) & 0x5555555555555555U;
1110
1111 /* calculate coordinates in odd and even bit positions */
1112 t = (s & swap) ^ comp;
1113 s = s ^ sr ^ t ^ (t << 1);
1114
1115 /* unpad/clear out any junk on the left */
1116 s = s & ((1 << n << n) - 1);
1117
1118 /* Now "unshuffle" to separate the x and y bits. */
1119 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1120 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1121 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1122 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1123 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1124
1125 /* now s contains two 32-bit coordinates */
1126 return s;
1127}
1128
1129/* algorithm from the book Hacker's Delight, but a similar algorithm*/
1130/* is given in https://doi.org/10.1002/spe.4380160103 */
1131/* this has been slightly improved over the original version */
1132ecb_function_ uint32_t
1133ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1134{
1135 uint32_t row;
1136 uint32_t state = 0;
1137 uint32_t s = 0;
1138
1139 do
1140 {
1141 --n;
1142
1143 row = 4 * state
1144 | (2 & (xy >> n >> 15))
1145 | (1 & (xy >> n ));
1146
1147 /* these funky constants are lookup tables for two-bit values */
1148 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1149 state = (0x8fe65831U >> 2 * row) & 3;
1150 }
1151 while (n > 0);
1152
1153 return s;
1154}
1155
1156/* 64 bit, essentially the same as 32 bit */
1157ecb_function_ uint64_t
1158ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1159{
1160 uint32_t row;
1161 uint32_t state = 0;
1162 uint64_t s = 0;
1163
1164 do
1165 {
1166 --n;
1167
1168 row = 4 * state
1169 | (2 & (xy >> n >> 31))
1170 | (1 & (xy >> n ));
1171
1172 /* these funky constants are lookup tables for two-bit values */
1173 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1174 state = (0x8fe65831U >> 2 * row) & 3;
1175 }
1176 while (n > 0);
1177
1178 return s;
1179}
810 1180
811/*****************************************************************************/ 1181/*****************************************************************************/
812/* division */ 1182/* division */
813 1183
814#if ECB_GCC_VERSION(3,0) || ECB_C99 1184#if ECB_GCC_VERSION(3,0) || ECB_C99
… …
948} 1318}
949 1319
950/*******************************************************************************/ 1320/*******************************************************************************/
951/* fast integer to ascii */ 1321/* fast integer to ascii */
952 1322
1323/*
1324 * This code is pretty complicated because it is general. The idea behind it,
1325 * however, is pretty simple: first, the number is multiplied with a scaling
1326 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1327 * number with the first digit in the upper bits.
1328 * Then this digit is converted to text and masked out. The resulting number
1329 * is then multiplied by 10, by multiplying the fixed point representation
1330 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1331 * format becomes 5.27, 6.26 and so on.
1332 * The rest involves only advancing the pointer if we already generated a
1333 * non-zero digit, so leading zeroes are overwritten.
1334 */
1335
953// simply return a mask with "bits" bits set 1336/* simply return a mask with "bits" bits set */
954#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) 1337#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
955 1338
956// oputput a single digit. maskvalue is 10**digitidx 1339/* oputput a single digit. maskvalue is 10**digitidx */
957#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ 1340#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
958 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ 1341 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
959 { \ 1342 { \
960 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ 1343 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
961 *ptr = digit + '0'; /* output it */ \ 1344 *ptr = digit + '0'; /* output it */ \
962 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ 1345 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
963 ptr += nz; /* output digit only if non-zero digit seen */ \ 1346 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 */ \ 1347 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
965 } 1348 }
966 1349
967// convert integer to fixed point format and multiply out digits, highest first 1350/* 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 1351/* 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) \ 1352#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
970ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ 1353ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
971{ \ 1354{ \
972 char nz = lz; /* non-zero digit seen? */ \ 1355 char nz = lz; /* non-zero digit seen? */ \
973 /* convert to x.bits fixed-point */ \ 1356 /* convert to x.bits fixed-point */ \
… …
984 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ 1367 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
985 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ 1368 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
986 return ptr; \ 1369 return ptr; \
987} 1370}
988 1371
989// predefined versions of the above, for various digits 1372/* predefined versions of the above, for various digits */
990// ecb_i2a_xN = almost N digits, limit defined by macro 1373/* ecb_i2a_xN = almost N digits, limit defined by macro */
991// ecb_i2a_N = up to N digits, leading zeroes suppressed 1374/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
992// ecb_i2a_0N = exactly N digits, including leading zeroes 1375/* ecb_i2a_0N = exactly N digits, including leading zeroes */
993 1376
994// non-leading-zero versions, limited range 1377/* non-leading-zero versions, limited range */
995#define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5 1378#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
996#define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10 1379#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
997ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) 1380ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
998ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) 1381ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
999 1382
1000// non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit 1383/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1001ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) 1384ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1002ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) 1385ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1003ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) 1386ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1004ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) 1387ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1005ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) 1388ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1006ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) 1389ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1007ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) 1390ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1008ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) 1391ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1009 1392
1010// leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit 1393/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1011ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) 1394ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1012ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) 1395ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1013ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) 1396ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1014ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) 1397ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1015ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) 1398ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1016ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) 1399ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1017ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) 1400ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1018ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) 1401ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1019 1402
1020#define ECB_I2A_I32_DIGITS 11 1403#define ECB_I2A_I32_DIGITS 11
1021#define ECB_I2A_U32_DIGITS 10 1404#define ECB_I2A_U32_DIGITS 10
1022#define ECB_I2A_I64_DIGITS 20 1405#define ECB_I2A_I64_DIGITS 20
1023#define ECB_I2A_U32_DIGITS 21 1406#define ECB_I2A_U64_DIGITS 21
1024#define ECB_I2A_DIGITS 21 1407#define ECB_I2A_MAX_DIGITS 21
1025 1408
1026ecb_inline char * 1409ecb_inline char *
1027ecb_i2a_u32 (char *ptr, uint32_t u) 1410ecb_i2a_u32 (char *ptr, uint32_t u)
1028{ 1411{
1029 #if ECB_64BIT_NATIVE 1412 #if ECB_64BIT_NATIVE
1030 if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) 1413 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1031 ptr = ecb_i2a_x10 (ptr, u); 1414 ptr = ecb_i2a_x10 (ptr, u);
1032 else // x10 almost, but not fully, covers 32 bit 1415 else /* x10 almost, but not fully, covers 32 bit */
1033 { 1416 {
1034 uint32_t u1 = u % 1000000000; 1417 uint32_t u1 = u % 1000000000;
1035 uint32_t u2 = u / 1000000000; 1418 uint32_t u2 = u / 1000000000;
1036 1419
1037 *ptr++ = u2 + '0'; 1420 *ptr++ = u2 + '0';
… …
1069{ 1452{
1070 *ptr = '-'; ptr += v < 0; 1453 *ptr = '-'; ptr += v < 0;
1071 uint32_t u = v < 0 ? -(uint32_t)v : v; 1454 uint32_t u = v < 0 ? -(uint32_t)v : v;
1072 1455
1073 #if ECB_64BIT_NATIVE 1456 #if ECB_64BIT_NATIVE
1074 ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit 1457 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1075 #else 1458 #else
1076 ptr = ecb_i2a_u32 (ptr, u); 1459 ptr = ecb_i2a_u32 (ptr, u);
1077 #endif 1460 #endif
1078 1461
1079 return ptr; 1462 return ptr;
… …
1142 uint64_t u1 = u % 1000000000; 1525 uint64_t u1 = u % 1000000000;
1143 uint64_t ua = u / 1000000000; 1526 uint64_t ua = u / 1000000000;
1144 uint64_t u2 = ua % 1000000000; 1527 uint64_t u2 = ua % 1000000000;
1145 uint64_t u3 = ua / 1000000000; 1528 uint64_t u3 = ua / 1000000000;
1146 1529
1147 // 2**31 is 19 digits, so the top is exactly one digit 1530 /* 2**31 is 19 digits, so the top is exactly one digit */
1148 *ptr++ = u3 + '0'; 1531 *ptr++ = u3 + '0';
1149 ptr = ecb_i2a_09 (ptr, u2); 1532 ptr = ecb_i2a_09 (ptr, u2);
1150 ptr = ecb_i2a_09 (ptr, u1); 1533 ptr = ecb_i2a_09 (ptr, u1);
1151 } 1534 }
1152 #else 1535 #else

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines