1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2013 Marc Alexander Lehmann <libecb@schmorp.de> |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
6 | * All rights reserved. |
6 | * All rights reserved. |
7 | * |
7 | * |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
… | |
… | |
64 | #define ECB_PTRSIZE 4 |
64 | #define ECB_PTRSIZE 4 |
65 | #endif |
65 | #endif |
66 | #endif |
66 | #endif |
67 | |
67 | |
68 | /* work around x32 idiocy by defining proper macros */ |
68 | /* work around x32 idiocy by defining proper macros */ |
69 | #if __x86_64 || _M_AMD64 |
69 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
70 | #if __ILP32 |
70 | #if _ILP32 |
71 | #define ECB_AMD64_X32 1 |
71 | #define ECB_AMD64_X32 1 |
72 | #else |
72 | #else |
73 | #define ECB_AMD64 1 |
73 | #define ECB_AMD64 1 |
74 | #endif |
74 | #endif |
75 | #endif |
75 | #endif |
… | |
… | |
134 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
134 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
136 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
136 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
137 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
137 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
139 | #elif __sparc || __sparc__ |
139 | #elif (__sparc || __sparc__) && !__sparcv8 |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
143 | #elif defined __s390__ || defined __s390x__ |
143 | #elif defined __s390__ || defined __s390x__ |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
145 | #elif defined __mips__ |
145 | #elif defined __mips__ |
|
|
146 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
147 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
147 | #elif defined __alpha__ |
149 | #elif defined __alpha__ |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
149 | #elif defined __hppa__ |
151 | #elif defined __hppa__ |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
152 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
151 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
153 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
152 | #elif defined __ia64__ |
154 | #elif defined __ia64__ |
153 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
155 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
156 | #elif defined __m68k__ |
|
|
157 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
158 | #elif defined __m88k__ |
|
|
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
160 | #elif defined __sh__ |
|
|
161 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
154 | #endif |
162 | #endif |
155 | #endif |
163 | #endif |
156 | #endif |
164 | #endif |
157 | |
165 | |
158 | #ifndef ECB_MEMORY_FENCE |
166 | #ifndef ECB_MEMORY_FENCE |
… | |
… | |
169 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
177 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
170 | */ |
178 | */ |
171 | |
179 | |
172 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
173 | #define ECB_MEMORY_FENCE __sync_synchronize () |
181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
182 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
183 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
184 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
185 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
186 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
187 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
174 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
188 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
175 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
189 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
176 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
190 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
177 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
191 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
178 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
192 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
565 | || defined __s390__ || defined __s390x__ \ |
579 | || defined __s390__ || defined __s390x__ \ |
566 | || defined __mips__ \ |
580 | || defined __mips__ \ |
567 | || defined __alpha__ \ |
581 | || defined __alpha__ \ |
568 | || defined __hppa__ \ |
582 | || defined __hppa__ \ |
569 | || defined __ia64__ \ |
583 | || defined __ia64__ \ |
|
|
584 | || defined __m68k__ \ |
|
|
585 | || defined __m88k__ \ |
|
|
586 | || defined __sh__ \ |
570 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
587 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
571 | #define ECB_STDFP 1 |
588 | #define ECB_STDFP 1 |
572 | #include <string.h> /* for memcpy */ |
589 | #include <string.h> /* for memcpy */ |
573 | #else |
590 | #else |
574 | #define ECB_STDFP 0 |
591 | #define ECB_STDFP 0 |
575 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
576 | #endif |
592 | #endif |
577 | |
593 | |
578 | #ifndef ECB_NO_LIBM |
594 | #ifndef ECB_NO_LIBM |
|
|
595 | |
|
|
596 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
597 | |
|
|
598 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
599 | #ifdef INFINITY |
|
|
600 | #define ECB_INFINITY INFINITY |
|
|
601 | #else |
|
|
602 | #define ECB_INFINITY HUGE_VAL |
|
|
603 | #endif |
|
|
604 | |
|
|
605 | #ifdef NAN |
|
|
606 | #define ECB_NAN NAN |
|
|
607 | #else |
|
|
608 | #define ECB_NAN ECB_INFINITY |
|
|
609 | #endif |
|
|
610 | |
|
|
611 | /* converts an ieee half/binary16 to a float */ |
|
|
612 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
613 | ecb_function_ float |
|
|
614 | ecb_binary16_to_float (uint16_t x) |
|
|
615 | { |
|
|
616 | int e = (x >> 10) & 0x1f; |
|
|
617 | int m = x & 0x3ff; |
|
|
618 | float r; |
|
|
619 | |
|
|
620 | if (!e ) r = ldexpf (m , -24); |
|
|
621 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
622 | else if (m ) r = ECB_NAN; |
|
|
623 | else r = ECB_INFINITY; |
|
|
624 | |
|
|
625 | return x & 0x8000 ? -r : r; |
|
|
626 | } |
579 | |
627 | |
580 | /* convert a float to ieee single/binary32 */ |
628 | /* convert a float to ieee single/binary32 */ |
581 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
629 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
582 | ecb_function_ uint32_t |
630 | ecb_function_ uint32_t |
583 | ecb_float_to_binary32 (float x) |
631 | ecb_float_to_binary32 (float x) |