diff --git a/conf.h b/conf.h index 94c1cd9..f07699b 100644 --- a/conf.h +++ b/conf.h @@ -27,6 +27,7 @@ #define CONF_H //------------------------- Compiler ------------------------------------------ #if defined(__GNUC__) +#include #define ALIGNED(t,v,n) t v __attribute__ ((aligned (n))) #define ALWAYS_INLINE inline __attribute__((always_inline)) #define NOINLINE __attribute__((noinline)) @@ -42,16 +43,16 @@ //bsr32: 0:0,1:1,2:2,3:2,4:3,5:3,6:3,7:3,8:4,9:4,10:4,11:4,12:4,13:4,14:4,15:4,16:5,17:5,18:5,19:5,20:5,21:5,22:5,23:5,24:5,25:5,26:5,27:5,28:5,29:5,30:5,31:5,32:6, static inline int __bsr32( int x) { asm("bsr %1,%0" : "=r" (x) : "rm" (x) ); return x; } static inline int bsr32( int x) { int b = -1; asm("bsrl %1,%0" : "+r" (b) : "rm" (x) ); return b + 1; } -static inline int bsr64(unsigned long long x) { return x?64 - __builtin_clzll(x):0; } +static inline int bsr64(uint64_t x) { return x?64 - __builtin_clzll(x):0; } static inline unsigned rol32(unsigned x, int s) { asm ("roll %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } static inline unsigned ror32(unsigned x, int s) { asm ("rorl %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } -static inline unsigned long long rol64(unsigned long long x, int s) { asm ("rolq %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } -static inline unsigned long long ror64(unsigned long long x, int s) { asm ("rorq %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } +static inline uint64_t rol64(uint64_t x, int s) { asm ("rolq %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } +static inline uint64_t ror64(uint64_t x, int s) { asm ("rorq %%cl,%0" :"=r" (x) :"0" (x),"c" (s)); return x; } #else static inline int __bsr32(unsigned x ) { return 31 - __builtin_clz( x); } static inline int bsr32(int x ) { return x?32 - __builtin_clz( x):0; } -static inline int bsr64(unsigned long long x) { return x?64 - __builtin_clzll(x):0; } +static inline int bsr64(uint64_t x) { return x?64 - __builtin_clzll(x):0; } static inline unsigned rol32(unsigned x, int s) { return x << s | x >> (32 - s); } static inline unsigned ror32(unsigned x, int s) { return x >> s | x << (32 - s); } @@ -64,15 +65,24 @@ static inline unsigned ror64(unsigned x, int s) { return x >> s | x << (64 - s); #define clz64(_x_) __builtin_clzll(_x_) #define clz32(_x_) __builtin_clz(_x_) -#if __GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 8 + #if __GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 8 #define bswap16(x) __builtin_bswap16(x) -#else + #else static inline unsigned short bswap16(unsigned short x) { return __builtin_bswap32(x << 16); } -#endif + #endif #define bswap32(x) __builtin_bswap32(x) #define bswap64(x) __builtin_bswap64(x) #elif _MSC_VER //---------------------------------------------------- +#include + #if _MSC_VER < 1600 +#include "vs/stdint.h" +#define __builtin_prefetch(x,a) + #else +#include +#define __builtin_prefetch(x,a) _mm_prefetch(x, _MM_HINT_NTA) + #endif + #define ALIGNED(x) __declspec(align(x)) #define ALWAYS_INLINE __forceinline #define NOINLINE __declspec(noinline) @@ -80,16 +90,16 @@ static inline unsigned short bswap16(unsigned short x) { return __builtin_bswap3 #define THREADLOCAL __declspec(thread) #define likely(x) (x) #define unlikely(x) (x) -#define __builtin_prefetch(x) //_mm_prefetch(x, _MM_HINT_NTA) -static inline int bsr32(int x) { return x ? 32 - __builtin_clz(x) : 0; } +static inline int __bsr32(int x) { unsigned long z; _BitScanReverse(&z, x); return z; } +static inline int bsr32( int x) { unsigned long z; _BitScanReverse(&z, x); return x?z+1:0; } #ifdef _WIN64 -static inline int bsr64(unsigned long long x) { unsigned long z = 0; _BitScanForward64(&z, x); return 64 - z; } -static inline int clz64(unsigned long long x) { unsigned long z = 0; _BitScanForward64(&z, x); return z; } -static inline int ctz64(unsigned long long x) { unsigned long z = 0; _BitScanReverse64(&z, x); return z; } +static inline int bsr64(uint64_t x) { unsigned long z=0; _BitScanForward64(&z, x); return x?z+1:0; } +static inline int ctz64(uint64_t x) { unsigned long z=0; _BitScanForward64(&z, x); return x?z:64;; } +static inline int clz64(uint64_t x) { unsigned long z=0; _BitScanReverse64(&z, x); return x?63-z:64; } #endif -static inline int clz32(unsigned x) { unsigned z = 0; _BitScanForward( &z, x); return 32 - z; } -static inline int ctz32(unsigned x) { unsigned z = 0; _BitScanReverse( &z, x); return z; } +static inline int ctz32(unsigned x) { unsigned long z=0; _BitScanForward( &z, x); return x?z:32; } +static inline int clz32(unsigned x) { unsigned long z=0; _BitScanReverse( &z, x); return x?31-z:32; } #define rol32(x,s) _lrotl(x, s) #define ror32(x,s) _lrotr(x, s) @@ -97,11 +107,16 @@ static inline int ctz32(unsigned x) { unsigned z = 0; _BitScanRev #define bswap32(x) _byteswap_ulong(x) #define bswap64(x) _byteswap_uint64(x) -#define fseeko _fseeki64 -#define ftello _ftelli64 -#define sleep(x) Sleep(x/1000) -#define strcasecmp _stricmp +#define popcnt32(x) __popcnt(x) +#define popcnt64(x) __popcnt64(x) + +#define sleep(x) Sleep(x/1000) +#define fseeko _fseeki64 +#define ftello _ftelli64 +#define strcasecmp _stricmp #define strncasecmp _strnicmp +#define strtoull _strtoui64 +static inline double round(double num) { return (num > 0.0) ? floor(num + 0.5) : ceil(num - 0.5); } #endif #define bsr8(_x_) bsr32(_x_) @@ -114,7 +129,7 @@ static inline int ctz32(unsigned x) { unsigned z = 0; _BitScanRev //--------------- Unaligned memory access ------------------------------------- /*# || defined(i386) || defined(_X86_) || defined(__THW_INTEL)*/ #if defined(__i386__) || defined(__x86_64__) || \ - defined(_M_IX86) || defined(_M_AMD64) || /*MSC_VER*/\ + defined(_M_IX86) || defined(_M_AMD64) || _MSC_VER ||\ defined(__powerpc__) ||\ defined(__ARM_FEATURE_UNALIGNED) || defined(__aarch64__) || defined(__arm__) ||\ defined(__ARM_ARCH_4__) || defined(__ARM_ARCH_4T__) || \ @@ -123,17 +138,17 @@ static inline int ctz32(unsigned x) { unsigned z = 0; _BitScanRev #define ctou16(_cp_) (*(unsigned short *)(_cp_)) #define ctou32(_cp_) (*(unsigned *)(_cp_)) - #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) -#define ctou64(_cp_) (*(unsigned long long *)(_cp_)) + #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || defined(_MSC_VER) +#define ctou64(_cp_) (*(uint64_t *)(_cp_)) #elif defined(__ARM_FEATURE_UNALIGNED) -struct _PACKED longu { unsigned long long l; }; +struct _PACKED longu { uint64_t l; }; #define ctou64(_cp_) ((struct longu *)(_cp_))->l #endif #elif defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7S__) struct _PACKED shortu { unsigned short s; }; struct _PACKED unsignedu { unsigned u; }; -struct _PACKED longu { unsigned long long l; }; +struct _PACKED longu { uint64_t l; }; #define ctou16(_cp_) ((struct shortu *)(_cp_))->s #define ctou32(_cp_) ((struct unsignedu *)(_cp_))->u @@ -145,22 +160,22 @@ struct _PACKED longu { unsigned long long l; }; #ifdef ctou16 //#define utoc16(_x_,_cp_) ctou16(_cp_) = _x_ #else -static inline unsigned short ctou16(const void *cp) { unsigned short x; memcpy(&x, cp, sizeof(x)); return x; } +static inline unsigned short ctou16(void *cp) { unsigned short x; memcpy((void *)&x, cp, (unsigned int)sizeof(x)); return x; } //static inline void utoc16(unsigned short x, void *cp ) { memcpy(cp, &x, sizeof(x)); } #endif #ifdef ctou32 //#define utoc32(_x_,_cp_) ctou32(_cp_) = _x_ #else -static inline unsigned ctou32(const void *cp) { unsigned x; memcpy(&x, cp, sizeof(x)); return x; } +static inline unsigned ctou32(void *cp) { unsigned x; memcpy(void *)&x, cp, (unsigned int)sizeof(x)); return x; } //static inline void utoc32(unsigned x, void *cp ) { memcpy(cp, &x, sizeof(x)); } #endif #ifdef ctou64 //#define utoc64(_x_,_cp_) ctou64(_cp_) = _x_ #else -static inline unsigned long long ctou64(const void *cp) { unsigned long long x; memcpy(&x, cp, sizeof(x)); return x; } -//static inline void utoc64(unsigned long long x, void *cp ) { memcpy(cp, &x, sizeof(x)); } +static inline uint64_t ctou64(void *cp) { uint64_t x; memcpy((void *)&x, cp, (unsigned int)sizeof(x)); return x; } +//static inline void utoc64(uint64_t x, void *cp ) { memcpy(cp, &x, sizeof(x)); } #endif #define ctou24(_cp_) (ctou32(_cp_) & 0xffffff) @@ -181,13 +196,8 @@ static inline unsigned long long ctou64(const void *cp) { unsigned long long x; #endif //---------------------misc --------------------------------------------------- -//#define C64(x) x##ull - #define SIZE_ROUNDUP(_n_, _a_) (((size_t)(_n_) + (size_t)((_a_) - 1)) & ~(size_t)((_a_) - 1)) - #ifndef min -#define min(x,y) (((x)<(y)) ? (x) : (y)) -#define max(x,y) (((x)>(y)) ? (x) : (y)) - #endif +#define ALIGN_DOWN(__ptr, __a) ((void *)((uintptr_t)(__ptr) & ~(uintptr_t)((__a) - 1))) #define TEMPLATE2_(_x_, _y_) _x_##_y_ #define TEMPLATE2(_x_, _y_) TEMPLATE2_(_x_,_y_) @@ -199,13 +209,13 @@ static inline unsigned long long ctou64(const void *cp) { unsigned long long x; #include #ifdef _MSC_VER #ifdef NDEBUG -#define AS(expr, fmt, ...) -#define AC(expr, fmt, ...) if(!(expr)) { fprintf(stderr, fmt, __VA_ARGS__ ); fflush(stderr); abort(); } -#define die(fmt, ...) do { fprintf(stderr, fmt, __VA_ARGS__ ); fflush(stderr); exit(-1); } while(0) +#define AS(expr, fmt, __VA_ARGS__) +#define AC(expr, fmt, __VA_ARGS__) if(!(expr)) { fprintf(stderr, fmt, __VA_ARGS__ ); fflush(stderr); abort(); } +#define die(fmt, __VA_ARGS__) do { fprintf(stderr, fmt, __VA_ARGS__ ); fflush(stderr); exit(-1); } while(0) #else -#define AS(expr, fmt, ...) if(!(expr)) { fflush(stdout);fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, ,__VA_ARGS__ ); fflush(stderr); abort(); } -#define AC(expr, fmt, ...) if(!(expr)) { fflush(stdout);fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, ,__VA_ARGS__ ); fflush(stderr); abort(); } -#define die(fmt, ...) do { fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, ,__VA_ARGS__ ); fflush(stderr); exit(-1); } while(0) +#define AS(expr, fmt, __VA_ARGS__) if(!(expr)) { fflush(stdout);fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, ,__VA_ARGS__ ); fflush(stderr); abort(); } +#define AC(expr, fmt, __VA_ARGS__) if(!(expr)) { fflush(stdout);fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, ,__VA_ARGS__ ); fflush(stderr); abort(); } +#define die(fmt, __VA_ARGS__) do { fprintf(stderr, "%s:%s:%d:", __FILE__, __FUNCTION__, __LINE__); fprintf(stderr, fmt, __VA_ARGS__ ); fflush(stderr); exit(-1); } while(0) #endif #else #ifdef NDEBUG