asm.h@ 81025

Last change on this file since 81025 was 76723, checked in by vboxsync, 6 years ago
asm.h: do not use the GCC version of ASMAtomicUoWritePtr for old GCC versions to prevent warnings building kernel modules.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 165.6 KB

Line
1	/** @file
2	* IPRT - Assembly Functions.
3	*/
4
5	/*
6	* Copyright (C) 2006-2019 Oracle Corporation
7	*
8	* This file is part of VirtualBox Open Source Edition (OSE), as
9	* available from http://www.virtualbox.org. This file is free software;
10	* you can redistribute it and/or modify it under the terms of the GNU
11	* General Public License (GPL) as published by the Free Software
12	* Foundation, in version 2 as it comes in the "COPYING" file of the
13	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
14	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
15	*
16	* The contents of this file may alternatively be used under the terms
17	* of the Common Development and Distribution License Version 1.0
18	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
19	* VirtualBox OSE distribution, in which case the provisions of the
20	* CDDL are applicable instead of those of the GPL.
21	*
22	* You may elect to license modified versions of this file under the
23	* terms and conditions of either the GPL or the CDDL or both.
24	*/
25
26	#ifndef IPRT_INCLUDED_asm_h
27	#define IPRT_INCLUDED_asm_h
28	#ifndef RT_WITHOUT_PRAGMA_ONCE
29	# pragma once
30	#endif
31
32	#include <iprt/cdefs.h>
33	#include <iprt/types.h>
34	#include <iprt/assert.h>
35	/** @def RT_INLINE_ASM_USES_INTRIN
36	* Defined as 1 if we're using a _MSC_VER 1400.
37	* Otherwise defined as 0.
38	*/
39
40	/* Solaris 10 header ugliness */
41	#ifdef u
42	# undef u
43	#endif
44
45	#if defined(_MSC_VER) && RT_INLINE_ASM_USES_INTRIN
46	# pragma warning(push)
47	# pragma warning(disable:4668) /* Several incorrect __cplusplus uses. */
48	# pragma warning(disable:4255) /* Incorrect __slwpcb prototype. */
49	# include <intrin.h>
50	# pragma warning(pop)
51	/* Emit the intrinsics at all optimization levels. */
52	# pragma intrinsic(_ReadWriteBarrier)
53	# pragma intrinsic(__cpuid)
54	# pragma intrinsic(__stosd)
55	# pragma intrinsic(__stosw)
56	# pragma intrinsic(__stosb)
57	# pragma intrinsic(_BitScanForward)
58	# pragma intrinsic(_BitScanReverse)
59	# pragma intrinsic(_bittest)
60	# pragma intrinsic(_bittestandset)
61	# pragma intrinsic(_bittestandreset)
62	# pragma intrinsic(_bittestandcomplement)
63	# pragma intrinsic(_byteswap_ushort)
64	# pragma intrinsic(_byteswap_ulong)
65	# pragma intrinsic(_interlockedbittestandset)
66	# pragma intrinsic(_interlockedbittestandreset)
67	# pragma intrinsic(_InterlockedAnd)
68	# pragma intrinsic(_InterlockedOr)
69	# pragma intrinsic(_InterlockedIncrement)
70	# pragma intrinsic(_InterlockedDecrement)
71	# pragma intrinsic(_InterlockedExchange)
72	# pragma intrinsic(_InterlockedExchangeAdd)
73	# pragma intrinsic(_InterlockedCompareExchange)
74	# pragma intrinsic(_InterlockedCompareExchange64)
75	# pragma intrinsic(_rotl)
76	# pragma intrinsic(_rotr)
77	# pragma intrinsic(_rotl64)
78	# pragma intrinsic(_rotr64)
79	# ifdef RT_ARCH_AMD64
80	# pragma intrinsic(__stosq)
81	# pragma intrinsic(_byteswap_uint64)
82	# pragma intrinsic(_InterlockedExchange64)
83	# pragma intrinsic(_InterlockedExchangeAdd64)
84	# pragma intrinsic(_InterlockedAnd64)
85	# pragma intrinsic(_InterlockedOr64)
86	# pragma intrinsic(_InterlockedIncrement64)
87	# pragma intrinsic(_InterlockedDecrement64)
88	# endif
89	#endif
90
91	/*
92	* Undefine all symbols we have Watcom C/C++ #pragma aux'es for.
93	*/
94	#if defined(__WATCOMC__) && ARCH_BITS == 16 && defined(RT_ARCH_X86)
95	# include "asm-watcom-x86-16.h"
96	#elif defined(__WATCOMC__) && ARCH_BITS == 32 && defined(RT_ARCH_X86)
97	# include "asm-watcom-x86-32.h"
98	#endif
99
100
101	/** @defgroup grp_rt_asm ASM - Assembly Routines
102	* @ingroup grp_rt
103	*
104	* @remarks The difference between ordered and unordered atomic operations are that
105	* the former will complete outstanding reads and writes before continuing
106	* while the latter doesn't make any promises about the order. Ordered
107	* operations doesn't, it seems, make any 100% promise wrt to whether
108	* the operation will complete before any subsequent memory access.
109	* (please, correct if wrong.)
110	*
111	* ASMAtomicSomething operations are all ordered, while ASMAtomicUoSomething
112	* are unordered (note the Uo).
113	*
114	* @remarks Some remarks about __volatile__: Without this keyword gcc is allowed to reorder
115	* or even optimize assembler instructions away. For instance, in the following code
116	* the second rdmsr instruction is optimized away because gcc treats that instruction
117	* as deterministic:
118	*
119	* @code
120	* static inline uint64_t rdmsr_low(int idx)
121	* {
122	* uint32_t low;
123	* __asm__ ("rdmsr" : "=a"(low) : "c"(idx) : "edx");
124	* }
125	* ...
126	* uint32_t msr1 = rdmsr_low(1);
127	* foo(msr1);
128	* msr1 = rdmsr_low(1);
129	* bar(msr1);
130	* @endcode
131	*
132	* The input parameter of rdmsr_low is the same for both calls and therefore gcc will
133	* use the result of the first call as input parameter for bar() as well. For rdmsr this
134	* is not acceptable as this instruction is _not_ deterministic. This applies to reading
135	* machine status information in general.
136	*
137	* @{
138	*/
139
140
141	/** @def RT_INLINE_ASM_GCC_4_3_X_X86
142	* Used to work around some 4.3.x register allocation issues in this version of
143	* the compiler. So far this workaround is still required for 4.4 and 4.5 but
144	* definitely not for 5.x */
145	#if (RT_GNUC_PREREQ(4, 3) && !RT_GNUC_PREREQ(5, 0) && defined(__i386__))
146	# define RT_INLINE_ASM_GCC_4_3_X_X86 1
147	#else
148	# define RT_INLINE_ASM_GCC_4_3_X_X86 0
149	#endif
150
151	/** @def RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
152	* i686-apple-darwin9-gcc-4.0.1 (GCC) 4.0.1 (Apple Inc. build 5493) screws up
153	* RTSemRWRequestWrite semsemrw-lockless-generic.cpp in release builds. PIC
154	* mode, x86.
155	*
156	* Some gcc 4.3.x versions may have register allocation issues with cmpxchg8b
157	* when in PIC mode on x86.
158	*/
159	#ifndef RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
160	# if defined(DOXYGEN_RUNNING) \|\| defined(__WATCOMC__) /* Watcom has trouble with the expression below */
161	# define RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC 1
162	# elif defined(_MSC_VER) /* Visual C++ has trouble too, but it'll only tell us when C4688 is enabled. */
163	# define RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC 0
164	# elif ( (defined(PIC) \|\| defined(__PIC__)) \
165	&& defined(RT_ARCH_X86) \
166	&& ( RT_INLINE_ASM_GCC_4_3_X_X86 \
167	\|\| defined(RT_OS_DARWIN)) )
168	# define RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC 1
169	# else
170	# define RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC 0
171	# endif
172	#endif
173
174
175	/** @def ASMReturnAddress
176	* Gets the return address of the current (or calling if you like) function or method.
177	*/
178	#ifdef _MSC_VER
179	# ifdef __cplusplus
180	extern "C"
181	# endif
182	void * _ReturnAddress(void);
183	# pragma intrinsic(_ReturnAddress)
184	# define ASMReturnAddress() _ReturnAddress()
185	#elif defined(__GNUC__) \|\| defined(DOXYGEN_RUNNING)
186	# define ASMReturnAddress() __builtin_return_address(0)
187	#elif defined(__WATCOMC__)
188	# define ASMReturnAddress() Watcom_does_not_appear_to_have_intrinsic_return_address_function()
189	#else
190	# error "Unsupported compiler."
191	#endif
192
193
194	/**
195	* Compiler memory barrier.
196	*
197	* Ensure that the compiler does not use any cached (register/tmp stack) memory
198	* values or any outstanding writes when returning from this function.
199	*
200	* This function must be used if non-volatile data is modified by a
201	* device or the VMM. Typical cases are port access, MMIO access,
202	* trapping instruction, etc.
203	*/
204	#if RT_INLINE_ASM_GNU_STYLE
205	# define ASMCompilerBarrier() do { __asm__ __volatile__("" : : : "memory"); } while (0)
206	#elif RT_INLINE_ASM_USES_INTRIN
207	# define ASMCompilerBarrier() do { _ReadWriteBarrier(); } while (0)
208	#elif defined(__WATCOMC__)
209	void ASMCompilerBarrier(void);
210	#else /* 2003 should have _ReadWriteBarrier() but I guess we're at 2002 level then... */
211	DECLINLINE(void) ASMCompilerBarrier(void)
212	{
213	__asm
214	{
215	}
216	}
217	#endif
218
219
220	/** @def ASMBreakpoint
221	* Debugger Breakpoint.
222	* @deprecated Use RT_BREAKPOINT instead.
223	* @internal
224	*/
225	#define ASMBreakpoint() RT_BREAKPOINT()
226
227
228	/**
229	* Spinloop hint for platforms that have these, empty function on the other
230	* platforms.
231	*
232	* x86 & AMD64: The PAUSE variant of NOP for helping hyperthreaded CPUs detecting
233	* spin locks.
234	*/
235	#if RT_INLINE_ASM_EXTERNAL && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86))
236	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMNopPause(void);
237	#else
238	DECLINLINE(void) ASMNopPause(void)
239	{
240	# if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
241	# if RT_INLINE_ASM_GNU_STYLE
242	__asm__ __volatile__(".byte 0xf3,0x90\n\t");
243	# else
244	__asm {
245	_emit 0f3h
246	_emit 090h
247	}
248	# endif
249	# else
250	/* dummy */
251	# endif
252	}
253	#endif
254
255
256	/**
257	* Atomically Exchange an unsigned 8-bit value, ordered.
258	*
259	* @returns Current *pu8 value
260	* @param pu8 Pointer to the 8-bit variable to update.
261	* @param u8 The 8-bit value to assign to *pu8.
262	*/
263	#if RT_INLINE_ASM_EXTERNAL
264	RT_ASM_DECL_PRAGMA_WATCOM(uint8_t) ASMAtomicXchgU8(volatile uint8_t RT_FAR *pu8, uint8_t u8);
265	#else
266	DECLINLINE(uint8_t) ASMAtomicXchgU8(volatile uint8_t RT_FAR *pu8, uint8_t u8)
267	{
268	# if RT_INLINE_ASM_GNU_STYLE
269	__asm__ __volatile__("xchgb %0, %1\n\t"
270	: "=m" (*pu8),
271	"=q" (u8) /* =r - busted on g++ (GCC) 3.4.4 20050721 (Red Hat 3.4.4-2) */
272	: "1" (u8),
273	"m" (*pu8));
274	# else
275	__asm
276	{
277	# ifdef RT_ARCH_AMD64
278	mov rdx, [pu8]
279	mov al, [u8]
280	xchg [rdx], al
281	mov [u8], al
282	# else
283	mov edx, [pu8]
284	mov al, [u8]
285	xchg [edx], al
286	mov [u8], al
287	# endif
288	}
289	# endif
290	return u8;
291	}
292	#endif
293
294
295	/**
296	* Atomically Exchange a signed 8-bit value, ordered.
297	*
298	* @returns Current *pu8 value
299	* @param pi8 Pointer to the 8-bit variable to update.
300	* @param i8 The 8-bit value to assign to *pi8.
301	*/
302	DECLINLINE(int8_t) ASMAtomicXchgS8(volatile int8_t RT_FAR *pi8, int8_t i8)
303	{
304	return (int8_t)ASMAtomicXchgU8((volatile uint8_t RT_FAR *)pi8, (uint8_t)i8);
305	}
306
307
308	/**
309	* Atomically Exchange a bool value, ordered.
310	*
311	* @returns Current *pf value
312	* @param pf Pointer to the 8-bit variable to update.
313	* @param f The 8-bit value to assign to *pi8.
314	*/
315	DECLINLINE(bool) ASMAtomicXchgBool(volatile bool RT_FAR *pf, bool f)
316	{
317	#ifdef _MSC_VER
318	return !!ASMAtomicXchgU8((volatile uint8_t RT_FAR *)pf, (uint8_t)f);
319	#else
320	return (bool)ASMAtomicXchgU8((volatile uint8_t RT_FAR *)pf, (uint8_t)f);
321	#endif
322	}
323
324
325	/**
326	* Atomically Exchange an unsigned 16-bit value, ordered.
327	*
328	* @returns Current *pu16 value
329	* @param pu16 Pointer to the 16-bit variable to update.
330	* @param u16 The 16-bit value to assign to *pu16.
331	*/
332	#if RT_INLINE_ASM_EXTERNAL
333	RT_ASM_DECL_PRAGMA_WATCOM(uint16_t) ASMAtomicXchgU16(volatile uint16_t RT_FAR *pu16, uint16_t u16);
334	#else
335	DECLINLINE(uint16_t) ASMAtomicXchgU16(volatile uint16_t RT_FAR *pu16, uint16_t u16)
336	{
337	# if RT_INLINE_ASM_GNU_STYLE
338	__asm__ __volatile__("xchgw %0, %1\n\t"
339	: "=m" (*pu16),
340	"=r" (u16)
341	: "1" (u16),
342	"m" (*pu16));
343	# else
344	__asm
345	{
346	# ifdef RT_ARCH_AMD64
347	mov rdx, [pu16]
348	mov ax, [u16]
349	xchg [rdx], ax
350	mov [u16], ax
351	# else
352	mov edx, [pu16]
353	mov ax, [u16]
354	xchg [edx], ax
355	mov [u16], ax
356	# endif
357	}
358	# endif
359	return u16;
360	}
361	#endif
362
363
364	/**
365	* Atomically Exchange a signed 16-bit value, ordered.
366	*
367	* @returns Current *pu16 value
368	* @param pi16 Pointer to the 16-bit variable to update.
369	* @param i16 The 16-bit value to assign to *pi16.
370	*/
371	DECLINLINE(int16_t) ASMAtomicXchgS16(volatile int16_t RT_FAR *pi16, int16_t i16)
372	{
373	return (int16_t)ASMAtomicXchgU16((volatile uint16_t RT_FAR *)pi16, (uint16_t)i16);
374	}
375
376
377	/**
378	* Atomically Exchange an unsigned 32-bit value, ordered.
379	*
380	* @returns Current *pu32 value
381	* @param pu32 Pointer to the 32-bit variable to update.
382	* @param u32 The 32-bit value to assign to *pu32.
383	*
384	* @remarks Does not work on 286 and earlier.
385	*/
386	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
387	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicXchgU32(volatile uint32_t RT_FAR *pu32, uint32_t u32);
388	#else
389	DECLINLINE(uint32_t) ASMAtomicXchgU32(volatile uint32_t RT_FAR *pu32, uint32_t u32)
390	{
391	# if RT_INLINE_ASM_GNU_STYLE
392	__asm__ __volatile__("xchgl %0, %1\n\t"
393	: "=m" (*pu32),
394	"=r" (u32)
395	: "1" (u32),
396	"m" (*pu32));
397
398	# elif RT_INLINE_ASM_USES_INTRIN
399	u32 = _InterlockedExchange((long RT_FAR *)pu32, u32);
400
401	# else
402	__asm
403	{
404	# ifdef RT_ARCH_AMD64
405	mov rdx, [pu32]
406	mov eax, u32
407	xchg [rdx], eax
408	mov [u32], eax
409	# else
410	mov edx, [pu32]
411	mov eax, u32
412	xchg [edx], eax
413	mov [u32], eax
414	# endif
415	}
416	# endif
417	return u32;
418	}
419	#endif
420
421
422	/**
423	* Atomically Exchange a signed 32-bit value, ordered.
424	*
425	* @returns Current *pu32 value
426	* @param pi32 Pointer to the 32-bit variable to update.
427	* @param i32 The 32-bit value to assign to *pi32.
428	*/
429	DECLINLINE(int32_t) ASMAtomicXchgS32(volatile int32_t RT_FAR *pi32, int32_t i32)
430	{
431	return (int32_t)ASMAtomicXchgU32((volatile uint32_t RT_FAR *)pi32, (uint32_t)i32);
432	}
433
434
435	/**
436	* Atomically Exchange an unsigned 64-bit value, ordered.
437	*
438	* @returns Current *pu64 value
439	* @param pu64 Pointer to the 64-bit variable to update.
440	* @param u64 The 64-bit value to assign to *pu64.
441	*
442	* @remarks Works on 32-bit x86 CPUs starting with Pentium.
443	*/
444	#if (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \
445	\|\| RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
446	RT_ASM_DECL_PRAGMA_WATCOM(uint64_t) ASMAtomicXchgU64(volatile uint64_t RT_FAR *pu64, uint64_t u64);
447	#else
448	DECLINLINE(uint64_t) ASMAtomicXchgU64(volatile uint64_t RT_FAR *pu64, uint64_t u64)
449	{
450	# if defined(RT_ARCH_AMD64)
451	# if RT_INLINE_ASM_USES_INTRIN
452	u64 = _InterlockedExchange64((__int64 *)pu64, u64);
453
454	# elif RT_INLINE_ASM_GNU_STYLE
455	__asm__ __volatile__("xchgq %0, %1\n\t"
456	: "=m" (*pu64),
457	"=r" (u64)
458	: "1" (u64),
459	"m" (*pu64));
460	# else
461	__asm
462	{
463	mov rdx, [pu64]
464	mov rax, [u64]
465	xchg [rdx], rax
466	mov [u64], rax
467	}
468	# endif
469	# else /* !RT_ARCH_AMD64 */
470	# if RT_INLINE_ASM_GNU_STYLE
471	# if defined(PIC) \|\| defined(__PIC__)
472	uint32_t u32EBX = (uint32_t)u64;
473	__asm__ __volatile__(/"xchgl %%esi, %5\n\t"/
474	"xchgl %%ebx, %3\n\t"
475	"1:\n\t"
476	"lock; cmpxchg8b (%5)\n\t"
477	"jnz 1b\n\t"
478	"movl %3, %%ebx\n\t"
479	/"xchgl %%esi, %5\n\t"/
480	: "=A" (u64),
481	"=m" (*pu64)
482	: "0" (*pu64),
483	"m" ( u32EBX ),
484	"c" ( (uint32_t)(u64 >> 32) ),
485	"S" (pu64));
486	# else /* !PIC */
487	__asm__ __volatile__("1:\n\t"
488	"lock; cmpxchg8b %1\n\t"
489	"jnz 1b\n\t"
490	: "=A" (u64),
491	"=m" (*pu64)
492	: "0" (*pu64),
493	"b" ( (uint32_t)u64 ),
494	"c" ( (uint32_t)(u64 >> 32) ));
495	# endif
496	# else
497	__asm
498	{
499	mov ebx, dword ptr [u64]
500	mov ecx, dword ptr [u64 + 4]
501	mov edi, pu64
502	mov eax, dword ptr [edi]
503	mov edx, dword ptr [edi + 4]
504	retry:
505	lock cmpxchg8b [edi]
506	jnz retry
507	mov dword ptr [u64], eax
508	mov dword ptr [u64 + 4], edx
509	}
510	# endif
511	# endif /* !RT_ARCH_AMD64 */
512	return u64;
513	}
514	#endif
515
516
517	/**
518	* Atomically Exchange an signed 64-bit value, ordered.
519	*
520	* @returns Current *pi64 value
521	* @param pi64 Pointer to the 64-bit variable to update.
522	* @param i64 The 64-bit value to assign to *pi64.
523	*/
524	DECLINLINE(int64_t) ASMAtomicXchgS64(volatile int64_t RT_FAR *pi64, int64_t i64)
525	{
526	return (int64_t)ASMAtomicXchgU64((volatile uint64_t RT_FAR *)pi64, (uint64_t)i64);
527	}
528
529
530	/**
531	* Atomically Exchange a size_t value, ordered.
532	*
533	* @returns Current *ppv value
534	* @param puDst Pointer to the size_t variable to update.
535	* @param uNew The new value to assign to *puDst.
536	*/
537	DECLINLINE(size_t) ASMAtomicXchgZ(size_t volatile RT_FAR *puDst, const size_t uNew)
538	{
539	#if ARCH_BITS == 16
540	AssertCompile(sizeof(size_t) == 2);
541	return ASMAtomicXchgU16((volatile uint16_t RT_FAR *)puDst, uNew);
542	#elif ARCH_BITS == 32
543	return ASMAtomicXchgU32((volatile uint32_t RT_FAR *)puDst, uNew);
544	#elif ARCH_BITS == 64
545	return ASMAtomicXchgU64((volatile uint64_t RT_FAR *)puDst, uNew);
546	#else
547	# error "ARCH_BITS is bogus"
548	#endif
549	}
550
551
552	/**
553	* Atomically Exchange a pointer value, ordered.
554	*
555	* @returns Current *ppv value
556	* @param ppv Pointer to the pointer variable to update.
557	* @param pv The pointer value to assign to *ppv.
558	*/
559	DECLINLINE(void RT_FAR ) ASMAtomicXchgPtr(void RT_FAR volatile RT_FAR ppv, const void RT_FAR pv)
560	{
561	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
562	return (void RT_FAR )ASMAtomicXchgU32((volatile uint32_t RT_FAR )(void RT_FAR *)ppv, (uint32_t)pv);
563	#elif ARCH_BITS == 64
564	return (void RT_FAR )ASMAtomicXchgU64((volatile uint64_t RT_FAR )(void RT_FAR *)ppv, (uint64_t)pv);
565	#else
566	# error "ARCH_BITS is bogus"
567	#endif
568	}
569
570
571	/**
572	* Convenience macro for avoiding the annoying casting with ASMAtomicXchgPtr.
573	*
574	* @returns Current *pv value
575	* @param ppv Pointer to the pointer variable to update.
576	* @param pv The pointer value to assign to *ppv.
577	* @param Type The type of *ppv, sans volatile.
578	*/
579	#ifdef __GNUC__ /* 8.2.0 requires -Wno-ignored-qualifiers */
580	# define ASMAtomicXchgPtrT(ppv, pv, Type) \
581	__extension__ \
582	({\
583	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
584	Type const pvTypeChecked = (pv); \
585	Type pvTypeCheckedRet = (__typeof__((ppv))) ASMAtomicXchgPtr((void volatile )ppvTypeChecked, (void )pvTypeChecked); \
586	pvTypeCheckedRet; \
587	})
588	#else
589	# define ASMAtomicXchgPtrT(ppv, pv, Type) \
590	(Type)ASMAtomicXchgPtr((void RT_FAR * volatile RT_FAR )(ppv), (void RT_FAR )(pv))
591	#endif
592
593
594	/**
595	* Atomically Exchange a raw-mode context pointer value, ordered.
596	*
597	* @returns Current *ppv value
598	* @param ppvRC Pointer to the pointer variable to update.
599	* @param pvRC The pointer value to assign to *ppv.
600	*/
601	DECLINLINE(RTRCPTR) ASMAtomicXchgRCPtr(RTRCPTR volatile RT_FAR *ppvRC, RTRCPTR pvRC)
602	{
603	return (RTRCPTR)ASMAtomicXchgU32((uint32_t volatile RT_FAR )(void RT_FAR )ppvRC, (uint32_t)pvRC);
604	}
605
606
607	/**
608	* Atomically Exchange a ring-0 pointer value, ordered.
609	*
610	* @returns Current *ppv value
611	* @param ppvR0 Pointer to the pointer variable to update.
612	* @param pvR0 The pointer value to assign to *ppv.
613	*/
614	DECLINLINE(RTR0PTR) ASMAtomicXchgR0Ptr(RTR0PTR volatile RT_FAR *ppvR0, RTR0PTR pvR0)
615	{
616	#if R0_ARCH_BITS == 32 \|\| ARCH_BITS == 16
617	return (RTR0PTR)ASMAtomicXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )ppvR0, (uint32_t)pvR0);
618	#elif R0_ARCH_BITS == 64
619	return (RTR0PTR)ASMAtomicXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )ppvR0, (uint64_t)pvR0);
620	#else
621	# error "R0_ARCH_BITS is bogus"
622	#endif
623	}
624
625
626	/**
627	* Atomically Exchange a ring-3 pointer value, ordered.
628	*
629	* @returns Current *ppv value
630	* @param ppvR3 Pointer to the pointer variable to update.
631	* @param pvR3 The pointer value to assign to *ppv.
632	*/
633	DECLINLINE(RTR3PTR) ASMAtomicXchgR3Ptr(RTR3PTR volatile RT_FAR *ppvR3, RTR3PTR pvR3)
634	{
635	#if R3_ARCH_BITS == 32 \|\| ARCH_BITS == 16
636	return (RTR3PTR)ASMAtomicXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )ppvR3, (uint32_t)pvR3);
637	#elif R3_ARCH_BITS == 64
638	return (RTR3PTR)ASMAtomicXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )ppvR3, (uint64_t)pvR3);
639	#else
640	# error "R3_ARCH_BITS is bogus"
641	#endif
642	}
643
644
645	/** @def ASMAtomicXchgHandle
646	* Atomically Exchange a typical IPRT handle value, ordered.
647	*
648	* @param ph Pointer to the value to update.
649	* @param hNew The new value to assigned to *pu.
650	* @param phRes Where to store the current *ph value.
651	*
652	* @remarks This doesn't currently work for all handles (like RTFILE).
653	*/
654	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
655	# define ASMAtomicXchgHandle(ph, hNew, phRes) \
656	do { \
657	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
658	AssertCompile(sizeof(*(phRes)) == sizeof(uint32_t)); \
659	(uint32_t RT_FAR )(phRes) = ASMAtomicXchgU32((uint32_t volatile RT_FAR *)(ph), (const uint32_t)(hNew)); \
660	} while (0)
661	#elif HC_ARCH_BITS == 64
662	# define ASMAtomicXchgHandle(ph, hNew, phRes) \
663	do { \
664	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
665	AssertCompile(sizeof(*(phRes)) == sizeof(uint64_t)); \
666	(uint64_t RT_FAR )(phRes) = ASMAtomicXchgU64((uint64_t volatile RT_FAR *)(ph), (const uint64_t)(hNew)); \
667	} while (0)
668	#else
669	# error HC_ARCH_BITS
670	#endif
671
672
673	/**
674	* Atomically Exchange a value which size might differ
675	* between platforms or compilers, ordered.
676	*
677	* @param pu Pointer to the variable to update.
678	* @param uNew The value to assign to *pu.
679	* @todo This is busted as its missing the result argument.
680	*/
681	#define ASMAtomicXchgSize(pu, uNew) \
682	do { \
683	switch (sizeof(*(pu))) { \
684	case 1: ASMAtomicXchgU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu), (uint8_t)(uNew)); break; \
685	case 2: ASMAtomicXchgU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu), (uint16_t)(uNew)); break; \
686	case 4: ASMAtomicXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
687	case 8: ASMAtomicXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
688	default: AssertMsgFailed(("ASMAtomicXchgSize: size %d is not supported\n", sizeof(*(pu)))); \
689	} \
690	} while (0)
691
692	/**
693	* Atomically Exchange a value which size might differ
694	* between platforms or compilers, ordered.
695	*
696	* @param pu Pointer to the variable to update.
697	* @param uNew The value to assign to *pu.
698	* @param puRes Where to store the current *pu value.
699	*/
700	#define ASMAtomicXchgSizeCorrect(pu, uNew, puRes) \
701	do { \
702	switch (sizeof(*(pu))) { \
703	case 1: (uint8_t RT_FAR )(puRes) = ASMAtomicXchgU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu), (uint8_t)(uNew)); break; \
704	case 2: (uint16_t RT_FAR )(puRes) = ASMAtomicXchgU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu), (uint16_t)(uNew)); break; \
705	case 4: (uint32_t RT_FAR )(puRes) = ASMAtomicXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
706	case 8: (uint64_t RT_FAR )(puRes) = ASMAtomicXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
707	default: AssertMsgFailed(("ASMAtomicXchgSize: size %d is not supported\n", sizeof(*(pu)))); \
708	} \
709	} while (0)
710
711
712
713	/**
714	* Atomically Compare and Exchange an unsigned 8-bit value, ordered.
715	*
716	* @returns true if xchg was done.
717	* @returns false if xchg wasn't done.
718	*
719	* @param pu8 Pointer to the value to update.
720	* @param u8New The new value to assigned to *pu8.
721	* @param u8Old The old value to *pu8 compare with.
722	*
723	* @remarks x86: Requires a 486 or later.
724	*/
725	#if RT_INLINE_ASM_EXTERNAL \|\| !RT_INLINE_ASM_GNU_STYLE
726	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicCmpXchgU8(volatile uint8_t RT_FAR *pu8, const uint8_t u8New, const uint8_t u8Old);
727	#else
728	DECLINLINE(bool) ASMAtomicCmpXchgU8(volatile uint8_t RT_FAR *pu8, const uint8_t u8New, uint8_t u8Old)
729	{
730	uint8_t u8Ret;
731	__asm__ __volatile__("lock; cmpxchgb %3, %0\n\t"
732	"setz %1\n\t"
733	: "=m" (*pu8),
734	"=qm" (u8Ret),
735	"=a" (u8Old)
736	: "q" (u8New),
737	"2" (u8Old),
738	"m" (*pu8));
739	return (bool)u8Ret;
740	}
741	#endif
742
743
744	/**
745	* Atomically Compare and Exchange a signed 8-bit value, ordered.
746	*
747	* @returns true if xchg was done.
748	* @returns false if xchg wasn't done.
749	*
750	* @param pi8 Pointer to the value to update.
751	* @param i8New The new value to assigned to *pi8.
752	* @param i8Old The old value to *pi8 compare with.
753	*
754	* @remarks x86: Requires a 486 or later.
755	*/
756	DECLINLINE(bool) ASMAtomicCmpXchgS8(volatile int8_t RT_FAR *pi8, const int8_t i8New, const int8_t i8Old)
757	{
758	return ASMAtomicCmpXchgU8((volatile uint8_t RT_FAR *)pi8, (uint8_t)i8New, (uint8_t)i8Old);
759	}
760
761
762	/**
763	* Atomically Compare and Exchange a bool value, ordered.
764	*
765	* @returns true if xchg was done.
766	* @returns false if xchg wasn't done.
767	*
768	* @param pf Pointer to the value to update.
769	* @param fNew The new value to assigned to *pf.
770	* @param fOld The old value to *pf compare with.
771	*
772	* @remarks x86: Requires a 486 or later.
773	*/
774	DECLINLINE(bool) ASMAtomicCmpXchgBool(volatile bool RT_FAR *pf, const bool fNew, const bool fOld)
775	{
776	return ASMAtomicCmpXchgU8((volatile uint8_t RT_FAR *)pf, (uint8_t)fNew, (uint8_t)fOld);
777	}
778
779
780	/**
781	* Atomically Compare and Exchange an unsigned 32-bit value, ordered.
782	*
783	* @returns true if xchg was done.
784	* @returns false if xchg wasn't done.
785	*
786	* @param pu32 Pointer to the value to update.
787	* @param u32New The new value to assigned to *pu32.
788	* @param u32Old The old value to *pu32 compare with.
789	*
790	* @remarks x86: Requires a 486 or later.
791	*/
792	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
793	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicCmpXchgU32(volatile uint32_t RT_FAR *pu32, const uint32_t u32New, const uint32_t u32Old);
794	#else
795	DECLINLINE(bool) ASMAtomicCmpXchgU32(volatile uint32_t RT_FAR *pu32, const uint32_t u32New, uint32_t u32Old)
796	{
797	# if RT_INLINE_ASM_GNU_STYLE
798	uint8_t u8Ret;
799	__asm__ __volatile__("lock; cmpxchgl %3, %0\n\t"
800	"setz %1\n\t"
801	: "=m" (*pu32),
802	"=qm" (u8Ret),
803	"=a" (u32Old)
804	: "r" (u32New),
805	"2" (u32Old),
806	"m" (*pu32));
807	return (bool)u8Ret;
808
809	# elif RT_INLINE_ASM_USES_INTRIN
810	return (uint32_t)_InterlockedCompareExchange((long RT_FAR *)pu32, u32New, u32Old) == u32Old;
811
812	# else
813	uint32_t u32Ret;
814	__asm
815	{
816	# ifdef RT_ARCH_AMD64
817	mov rdx, [pu32]
818	# else
819	mov edx, [pu32]
820	# endif
821	mov eax, [u32Old]
822	mov ecx, [u32New]
823	# ifdef RT_ARCH_AMD64
824	lock cmpxchg [rdx], ecx
825	# else
826	lock cmpxchg [edx], ecx
827	# endif
828	setz al
829	movzx eax, al
830	mov [u32Ret], eax
831	}
832	return !!u32Ret;
833	# endif
834	}
835	#endif
836
837
838	/**
839	* Atomically Compare and Exchange a signed 32-bit value, ordered.
840	*
841	* @returns true if xchg was done.
842	* @returns false if xchg wasn't done.
843	*
844	* @param pi32 Pointer to the value to update.
845	* @param i32New The new value to assigned to *pi32.
846	* @param i32Old The old value to *pi32 compare with.
847	*
848	* @remarks x86: Requires a 486 or later.
849	*/
850	DECLINLINE(bool) ASMAtomicCmpXchgS32(volatile int32_t RT_FAR *pi32, const int32_t i32New, const int32_t i32Old)
851	{
852	return ASMAtomicCmpXchgU32((volatile uint32_t RT_FAR *)pi32, (uint32_t)i32New, (uint32_t)i32Old);
853	}
854
855
856	/**
857	* Atomically Compare and exchange an unsigned 64-bit value, ordered.
858	*
859	* @returns true if xchg was done.
860	* @returns false if xchg wasn't done.
861	*
862	* @param pu64 Pointer to the 64-bit variable to update.
863	* @param u64New The 64-bit value to assign to *pu64.
864	* @param u64Old The value to compare with.
865	*
866	* @remarks x86: Requires a Pentium or later.
867	*/
868	#if (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \
869	\|\| RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
870	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicCmpXchgU64(volatile uint64_t RT_FAR *pu64, const uint64_t u64New, const uint64_t u64Old);
871	#else
872	DECLINLINE(bool) ASMAtomicCmpXchgU64(volatile uint64_t RT_FAR *pu64, uint64_t u64New, uint64_t u64Old)
873	{
874	# if RT_INLINE_ASM_USES_INTRIN
875	return (uint64_t)_InterlockedCompareExchange64((__int64 RT_FAR *)pu64, u64New, u64Old) == u64Old;
876
877	# elif defined(RT_ARCH_AMD64)
878	# if RT_INLINE_ASM_GNU_STYLE
879	uint8_t u8Ret;
880	__asm__ __volatile__("lock; cmpxchgq %3, %0\n\t"
881	"setz %1\n\t"
882	: "=m" (*pu64),
883	"=qm" (u8Ret),
884	"=a" (u64Old)
885	: "r" (u64New),
886	"2" (u64Old),
887	"m" (*pu64));
888	return (bool)u8Ret;
889	# else
890	bool fRet;
891	__asm
892	{
893	mov rdx, [pu32]
894	mov rax, [u64Old]
895	mov rcx, [u64New]
896	lock cmpxchg [rdx], rcx
897	setz al
898	mov [fRet], al
899	}
900	return fRet;
901	# endif
902	# else /* !RT_ARCH_AMD64 */
903	uint32_t u32Ret;
904	# if RT_INLINE_ASM_GNU_STYLE
905	# if defined(PIC) \|\| defined(__PIC__)
906	uint32_t u32EBX = (uint32_t)u64New;
907	uint32_t u32Spill;
908	__asm__ __volatile__("xchgl %%ebx, %4\n\t"
909	"lock; cmpxchg8b (%6)\n\t"
910	"setz %%al\n\t"
911	"movl %4, %%ebx\n\t"
912	"movzbl %%al, %%eax\n\t"
913	: "=a" (u32Ret),
914	"=d" (u32Spill),
915	# if RT_GNUC_PREREQ(4, 3)
916	"+m" (*pu64)
917	# else
918	"=m" (*pu64)
919	# endif
920	: "A" (u64Old),
921	"m" ( u32EBX ),
922	"c" ( (uint32_t)(u64New >> 32) ),
923	"S" (pu64));
924	# else /* !PIC */
925	uint32_t u32Spill;
926	__asm__ __volatile__("lock; cmpxchg8b %2\n\t"
927	"setz %%al\n\t"
928	"movzbl %%al, %%eax\n\t"
929	: "=a" (u32Ret),
930	"=d" (u32Spill),
931	"+m" (*pu64)
932	: "A" (u64Old),
933	"b" ( (uint32_t)u64New ),
934	"c" ( (uint32_t)(u64New >> 32) ));
935	# endif
936	return (bool)u32Ret;
937	# else
938	__asm
939	{
940	mov ebx, dword ptr [u64New]
941	mov ecx, dword ptr [u64New + 4]
942	mov edi, [pu64]
943	mov eax, dword ptr [u64Old]
944	mov edx, dword ptr [u64Old + 4]
945	lock cmpxchg8b [edi]
946	setz al
947	movzx eax, al
948	mov dword ptr [u32Ret], eax
949	}
950	return !!u32Ret;
951	# endif
952	# endif /* !RT_ARCH_AMD64 */
953	}
954	#endif
955
956
957	/**
958	* Atomically Compare and exchange a signed 64-bit value, ordered.
959	*
960	* @returns true if xchg was done.
961	* @returns false if xchg wasn't done.
962	*
963	* @param pi64 Pointer to the 64-bit variable to update.
964	* @param i64 The 64-bit value to assign to *pu64.
965	* @param i64Old The value to compare with.
966	*
967	* @remarks x86: Requires a Pentium or later.
968	*/
969	DECLINLINE(bool) ASMAtomicCmpXchgS64(volatile int64_t RT_FAR *pi64, const int64_t i64, const int64_t i64Old)
970	{
971	return ASMAtomicCmpXchgU64((volatile uint64_t RT_FAR *)pi64, (uint64_t)i64, (uint64_t)i64Old);
972	}
973
974
975	/**
976	* Atomically Compare and Exchange a pointer value, ordered.
977	*
978	* @returns true if xchg was done.
979	* @returns false if xchg wasn't done.
980	*
981	* @param ppv Pointer to the value to update.
982	* @param pvNew The new value to assigned to *ppv.
983	* @param pvOld The old value to *ppv compare with.
984	*
985	* @remarks x86: Requires a 486 or later.
986	*/
987	DECLINLINE(bool) ASMAtomicCmpXchgPtrVoid(void RT_FAR * volatile RT_FAR ppv, const void RT_FAR pvNew, const void RT_FAR *pvOld)
988	{
989	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
990	return ASMAtomicCmpXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )ppv, (uint32_t)pvNew, (uint32_t)pvOld);
991	#elif ARCH_BITS == 64
992	return ASMAtomicCmpXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )ppv, (uint64_t)pvNew, (uint64_t)pvOld);
993	#else
994	# error "ARCH_BITS is bogus"
995	#endif
996	}
997
998
999	/**
1000	* Atomically Compare and Exchange a pointer value, ordered.
1001	*
1002	* @returns true if xchg was done.
1003	* @returns false if xchg wasn't done.
1004	*
1005	* @param ppv Pointer to the value to update.
1006	* @param pvNew The new value to assigned to *ppv.
1007	* @param pvOld The old value to *ppv compare with.
1008	*
1009	* @remarks This is relatively type safe on GCC platforms.
1010	* @remarks x86: Requires a 486 or later.
1011	*/
1012	#ifdef __GNUC__
1013	# define ASMAtomicCmpXchgPtr(ppv, pvNew, pvOld) \
1014	__extension__ \
1015	({\
1016	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
1017	__typeof__(*(ppv)) const pvNewTypeChecked = (pvNew); \
1018	__typeof__(*(ppv)) const pvOldTypeChecked = (pvOld); \
1019	bool fMacroRet = ASMAtomicCmpXchgPtrVoid((void * volatile *)ppvTypeChecked, \
1020	(void )pvNewTypeChecked, (void )pvOldTypeChecked); \
1021	fMacroRet; \
1022	})
1023	#else
1024	# define ASMAtomicCmpXchgPtr(ppv, pvNew, pvOld) \
1025	ASMAtomicCmpXchgPtrVoid((void RT_FAR * volatile RT_FAR )(ppv), (void RT_FAR )(pvNew), (void RT_FAR *)(pvOld))
1026	#endif
1027
1028
1029	/** @def ASMAtomicCmpXchgHandle
1030	* Atomically Compare and Exchange a typical IPRT handle value, ordered.
1031	*
1032	* @param ph Pointer to the value to update.
1033	* @param hNew The new value to assigned to *pu.
1034	* @param hOld The old value to *pu compare with.
1035	* @param fRc Where to store the result.
1036	*
1037	* @remarks This doesn't currently work for all handles (like RTFILE).
1038	* @remarks x86: Requires a 486 or later.
1039	*/
1040	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
1041	# define ASMAtomicCmpXchgHandle(ph, hNew, hOld, fRc) \
1042	do { \
1043	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
1044	(fRc) = ASMAtomicCmpXchgU32((uint32_t volatile RT_FAR *)(ph), (const uint32_t)(hNew), (const uint32_t)(hOld)); \
1045	} while (0)
1046	#elif HC_ARCH_BITS == 64
1047	# define ASMAtomicCmpXchgHandle(ph, hNew, hOld, fRc) \
1048	do { \
1049	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
1050	(fRc) = ASMAtomicCmpXchgU64((uint64_t volatile RT_FAR *)(ph), (const uint64_t)(hNew), (const uint64_t)(hOld)); \
1051	} while (0)
1052	#else
1053	# error HC_ARCH_BITS
1054	#endif
1055
1056
1057	/** @def ASMAtomicCmpXchgSize
1058	* Atomically Compare and Exchange a value which size might differ
1059	* between platforms or compilers, ordered.
1060	*
1061	* @param pu Pointer to the value to update.
1062	* @param uNew The new value to assigned to *pu.
1063	* @param uOld The old value to *pu compare with.
1064	* @param fRc Where to store the result.
1065	*
1066	* @remarks x86: Requires a 486 or later.
1067	*/
1068	#define ASMAtomicCmpXchgSize(pu, uNew, uOld, fRc) \
1069	do { \
1070	switch (sizeof(*(pu))) { \
1071	case 4: (fRc) = ASMAtomicCmpXchgU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew), (uint32_t)(uOld)); \
1072	break; \
1073	case 8: (fRc) = ASMAtomicCmpXchgU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew), (uint64_t)(uOld)); \
1074	break; \
1075	default: AssertMsgFailed(("ASMAtomicCmpXchgSize: size %d is not supported\n", sizeof(*(pu)))); \
1076	(fRc) = false; \
1077	break; \
1078	} \
1079	} while (0)
1080
1081
1082	/**
1083	* Atomically Compare and Exchange an unsigned 32-bit value, additionally
1084	* passes back old value, ordered.
1085	*
1086	* @returns true if xchg was done.
1087	* @returns false if xchg wasn't done.
1088	*
1089	* @param pu32 Pointer to the value to update.
1090	* @param u32New The new value to assigned to *pu32.
1091	* @param u32Old The old value to *pu32 compare with.
1092	* @param pu32Old Pointer store the old value at.
1093	*
1094	* @remarks x86: Requires a 486 or later.
1095	*/
1096	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
1097	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicCmpXchgExU32(volatile uint32_t RT_FAR pu32, const uint32_t u32New, const uint32_t u32Old, uint32_t RT_FAR pu32Old);
1098	#else
1099	DECLINLINE(bool) ASMAtomicCmpXchgExU32(volatile uint32_t RT_FAR pu32, const uint32_t u32New, const uint32_t u32Old, uint32_t RT_FAR pu32Old)
1100	{
1101	# if RT_INLINE_ASM_GNU_STYLE
1102	uint8_t u8Ret;
1103	__asm__ __volatile__("lock; cmpxchgl %3, %0\n\t"
1104	"setz %1\n\t"
1105	: "=m" (*pu32),
1106	"=qm" (u8Ret),
1107	"=a" (*pu32Old)
1108	: "r" (u32New),
1109	"a" (u32Old),
1110	"m" (*pu32));
1111	return (bool)u8Ret;
1112
1113	# elif RT_INLINE_ASM_USES_INTRIN
1114	return (pu32Old =_InterlockedCompareExchange((long RT_FAR )pu32, u32New, u32Old)) == u32Old;
1115
1116	# else
1117	uint32_t u32Ret;
1118	__asm
1119	{
1120	# ifdef RT_ARCH_AMD64
1121	mov rdx, [pu32]
1122	# else
1123	mov edx, [pu32]
1124	# endif
1125	mov eax, [u32Old]
1126	mov ecx, [u32New]
1127	# ifdef RT_ARCH_AMD64
1128	lock cmpxchg [rdx], ecx
1129	mov rdx, [pu32Old]
1130	mov [rdx], eax
1131	# else
1132	lock cmpxchg [edx], ecx
1133	mov edx, [pu32Old]
1134	mov [edx], eax
1135	# endif
1136	setz al
1137	movzx eax, al
1138	mov [u32Ret], eax
1139	}
1140	return !!u32Ret;
1141	# endif
1142	}
1143	#endif
1144
1145
1146	/**
1147	* Atomically Compare and Exchange a signed 32-bit value, additionally
1148	* passes back old value, ordered.
1149	*
1150	* @returns true if xchg was done.
1151	* @returns false if xchg wasn't done.
1152	*
1153	* @param pi32 Pointer to the value to update.
1154	* @param i32New The new value to assigned to *pi32.
1155	* @param i32Old The old value to *pi32 compare with.
1156	* @param pi32Old Pointer store the old value at.
1157	*
1158	* @remarks x86: Requires a 486 or later.
1159	*/
1160	DECLINLINE(bool) ASMAtomicCmpXchgExS32(volatile int32_t RT_FAR pi32, const int32_t i32New, const int32_t i32Old, int32_t RT_FAR pi32Old)
1161	{
1162	return ASMAtomicCmpXchgExU32((volatile uint32_t RT_FAR )pi32, (uint32_t)i32New, (uint32_t)i32Old, (uint32_t RT_FAR )pi32Old);
1163	}
1164
1165
1166	/**
1167	* Atomically Compare and exchange an unsigned 64-bit value, additionally
1168	* passing back old value, ordered.
1169	*
1170	* @returns true if xchg was done.
1171	* @returns false if xchg wasn't done.
1172	*
1173	* @param pu64 Pointer to the 64-bit variable to update.
1174	* @param u64New The 64-bit value to assign to *pu64.
1175	* @param u64Old The value to compare with.
1176	* @param pu64Old Pointer store the old value at.
1177	*
1178	* @remarks x86: Requires a Pentium or later.
1179	*/
1180	#if (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \
1181	\|\| RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
1182	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicCmpXchgExU64(volatile uint64_t RT_FAR pu64, const uint64_t u64New, const uint64_t u64Old, uint64_t RT_FAR pu64Old);
1183	#else
1184	DECLINLINE(bool) ASMAtomicCmpXchgExU64(volatile uint64_t RT_FAR pu64, const uint64_t u64New, const uint64_t u64Old, uint64_t RT_FAR pu64Old)
1185	{
1186	# if RT_INLINE_ASM_USES_INTRIN
1187	return (pu64Old =_InterlockedCompareExchange64((__int64 RT_FAR )pu64, u64New, u64Old)) == u64Old;
1188
1189	# elif defined(RT_ARCH_AMD64)
1190	# if RT_INLINE_ASM_GNU_STYLE
1191	uint8_t u8Ret;
1192	__asm__ __volatile__("lock; cmpxchgq %3, %0\n\t"
1193	"setz %1\n\t"
1194	: "=m" (*pu64),
1195	"=qm" (u8Ret),
1196	"=a" (*pu64Old)
1197	: "r" (u64New),
1198	"a" (u64Old),
1199	"m" (*pu64));
1200	return (bool)u8Ret;
1201	# else
1202	bool fRet;
1203	__asm
1204	{
1205	mov rdx, [pu32]
1206	mov rax, [u64Old]
1207	mov rcx, [u64New]
1208	lock cmpxchg [rdx], rcx
1209	mov rdx, [pu64Old]
1210	mov [rdx], rax
1211	setz al
1212	mov [fRet], al
1213	}
1214	return fRet;
1215	# endif
1216	# else /* !RT_ARCH_AMD64 */
1217	# if RT_INLINE_ASM_GNU_STYLE
1218	uint64_t u64Ret;
1219	# if defined(PIC) \|\| defined(__PIC__)
1220	/* NB: this code uses a memory clobber description, because the clean
1221	* solution with an output value for *pu64 makes gcc run out of registers.
1222	* This will cause suboptimal code, and anyone with a better solution is
1223	* welcome to improve this. */
1224	__asm__ __volatile__("xchgl %%ebx, %1\n\t"
1225	"lock; cmpxchg8b %3\n\t"
1226	"xchgl %%ebx, %1\n\t"
1227	: "=A" (u64Ret)
1228	: "DS" ((uint32_t)u64New),
1229	"c" ((uint32_t)(u64New >> 32)),
1230	"m" (*pu64),
1231	"0" (u64Old)
1232	: "memory" );
1233	# else /* !PIC */
1234	__asm__ __volatile__("lock; cmpxchg8b %4\n\t"
1235	: "=A" (u64Ret),
1236	"=m" (*pu64)
1237	: "b" ((uint32_t)u64New),
1238	"c" ((uint32_t)(u64New >> 32)),
1239	"m" (*pu64),
1240	"0" (u64Old));
1241	# endif
1242	*pu64Old = u64Ret;
1243	return u64Ret == u64Old;
1244	# else
1245	uint32_t u32Ret;
1246	__asm
1247	{
1248	mov ebx, dword ptr [u64New]
1249	mov ecx, dword ptr [u64New + 4]
1250	mov edi, [pu64]
1251	mov eax, dword ptr [u64Old]
1252	mov edx, dword ptr [u64Old + 4]
1253	lock cmpxchg8b [edi]
1254	mov ebx, [pu64Old]
1255	mov [ebx], eax
1256	setz al
1257	movzx eax, al
1258	add ebx, 4
1259	mov [ebx], edx
1260	mov dword ptr [u32Ret], eax
1261	}
1262	return !!u32Ret;
1263	# endif
1264	# endif /* !RT_ARCH_AMD64 */
1265	}
1266	#endif
1267
1268
1269	/**
1270	* Atomically Compare and exchange a signed 64-bit value, additionally
1271	* passing back old value, ordered.
1272	*
1273	* @returns true if xchg was done.
1274	* @returns false if xchg wasn't done.
1275	*
1276	* @param pi64 Pointer to the 64-bit variable to update.
1277	* @param i64 The 64-bit value to assign to *pu64.
1278	* @param i64Old The value to compare with.
1279	* @param pi64Old Pointer store the old value at.
1280	*
1281	* @remarks x86: Requires a Pentium or later.
1282	*/
1283	DECLINLINE(bool) ASMAtomicCmpXchgExS64(volatile int64_t RT_FAR pi64, const int64_t i64, const int64_t i64Old, int64_t RT_FAR pi64Old)
1284	{
1285	return ASMAtomicCmpXchgExU64((volatile uint64_t RT_FAR )pi64, (uint64_t)i64, (uint64_t)i64Old, (uint64_t RT_FAR )pi64Old);
1286	}
1287
1288	/** @def ASMAtomicCmpXchgExHandle
1289	* Atomically Compare and Exchange a typical IPRT handle value, ordered.
1290	*
1291	* @param ph Pointer to the value to update.
1292	* @param hNew The new value to assigned to *pu.
1293	* @param hOld The old value to *pu compare with.
1294	* @param fRc Where to store the result.
1295	* @param phOldVal Pointer to where to store the old value.
1296	*
1297	* @remarks This doesn't currently work for all handles (like RTFILE).
1298	*/
1299	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
1300	# define ASMAtomicCmpXchgExHandle(ph, hNew, hOld, fRc, phOldVal) \
1301	do { \
1302	AssertCompile(sizeof(*ph) == sizeof(uint32_t)); \
1303	AssertCompile(sizeof(*phOldVal) == sizeof(uint32_t)); \
1304	(fRc) = ASMAtomicCmpXchgExU32((volatile uint32_t RT_FAR )(pu), (uint32_t)(uNew), (uint32_t)(uOld), (uint32_t RT_FAR )(puOldVal)); \
1305	} while (0)
1306	#elif HC_ARCH_BITS == 64
1307	# define ASMAtomicCmpXchgExHandle(ph, hNew, hOld, fRc, phOldVal) \
1308	do { \
1309	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
1310	AssertCompile(sizeof(*(phOldVal)) == sizeof(uint64_t)); \
1311	(fRc) = ASMAtomicCmpXchgExU64((volatile uint64_t RT_FAR )(pu), (uint64_t)(uNew), (uint64_t)(uOld), (uint64_t RT_FAR )(puOldVal)); \
1312	} while (0)
1313	#else
1314	# error HC_ARCH_BITS
1315	#endif
1316
1317
1318	/** @def ASMAtomicCmpXchgExSize
1319	* Atomically Compare and Exchange a value which size might differ
1320	* between platforms or compilers. Additionally passes back old value.
1321	*
1322	* @param pu Pointer to the value to update.
1323	* @param uNew The new value to assigned to *pu.
1324	* @param uOld The old value to *pu compare with.
1325	* @param fRc Where to store the result.
1326	* @param puOldVal Pointer to where to store the old value.
1327	*
1328	* @remarks x86: Requires a 486 or later.
1329	*/
1330	#define ASMAtomicCmpXchgExSize(pu, uNew, uOld, fRc, puOldVal) \
1331	do { \
1332	switch (sizeof(*(pu))) { \
1333	case 4: (fRc) = ASMAtomicCmpXchgExU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew), (uint32_t)(uOld), (uint32_t RT_FAR *)(uOldVal)); \
1334	break; \
1335	case 8: (fRc) = ASMAtomicCmpXchgExU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew), (uint64_t)(uOld), (uint64_t RT_FAR *)(uOldVal)); \
1336	break; \
1337	default: AssertMsgFailed(("ASMAtomicCmpXchgSize: size %d is not supported\n", sizeof(*(pu)))); \
1338	(fRc) = false; \
1339	(uOldVal) = 0; \
1340	break; \
1341	} \
1342	} while (0)
1343
1344
1345	/**
1346	* Atomically Compare and Exchange a pointer value, additionally
1347	* passing back old value, ordered.
1348	*
1349	* @returns true if xchg was done.
1350	* @returns false if xchg wasn't done.
1351	*
1352	* @param ppv Pointer to the value to update.
1353	* @param pvNew The new value to assigned to *ppv.
1354	* @param pvOld The old value to *ppv compare with.
1355	* @param ppvOld Pointer store the old value at.
1356	*
1357	* @remarks x86: Requires a 486 or later.
1358	*/
1359	DECLINLINE(bool) ASMAtomicCmpXchgExPtrVoid(void RT_FAR * volatile RT_FAR ppv, const void RT_FAR pvNew, const void RT_FAR *pvOld,
1360	void RT_FAR * RT_FAR *ppvOld)
1361	{
1362	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
1363	return ASMAtomicCmpXchgExU32((volatile uint32_t RT_FAR )(void RT_FAR )ppv, (uint32_t)pvNew, (uint32_t)pvOld, (uint32_t RT_FAR *)ppvOld);
1364	#elif ARCH_BITS == 64
1365	return ASMAtomicCmpXchgExU64((volatile uint64_t RT_FAR )(void RT_FAR )ppv, (uint64_t)pvNew, (uint64_t)pvOld, (uint64_t RT_FAR *)ppvOld);
1366	#else
1367	# error "ARCH_BITS is bogus"
1368	#endif
1369	}
1370
1371
1372	/**
1373	* Atomically Compare and Exchange a pointer value, additionally
1374	* passing back old value, ordered.
1375	*
1376	* @returns true if xchg was done.
1377	* @returns false if xchg wasn't done.
1378	*
1379	* @param ppv Pointer to the value to update.
1380	* @param pvNew The new value to assigned to *ppv.
1381	* @param pvOld The old value to *ppv compare with.
1382	* @param ppvOld Pointer store the old value at.
1383	*
1384	* @remarks This is relatively type safe on GCC platforms.
1385	* @remarks x86: Requires a 486 or later.
1386	*/
1387	#ifdef __GNUC__
1388	# define ASMAtomicCmpXchgExPtr(ppv, pvNew, pvOld, ppvOld) \
1389	__extension__ \
1390	({\
1391	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
1392	__typeof__(*(ppv)) const pvNewTypeChecked = (pvNew); \
1393	__typeof__(*(ppv)) const pvOldTypeChecked = (pvOld); \
1394	__typeof__((ppv)) const ppvOldTypeChecked = (ppvOld); \
1395	bool fMacroRet = ASMAtomicCmpXchgExPtrVoid((void * volatile *)ppvTypeChecked, \
1396	(void )pvNewTypeChecked, (void )pvOldTypeChecked, \
1397	(void **)ppvOldTypeChecked); \
1398	fMacroRet; \
1399	})
1400	#else
1401	# define ASMAtomicCmpXchgExPtr(ppv, pvNew, pvOld, ppvOld) \
1402	ASMAtomicCmpXchgExPtrVoid((void RT_FAR * volatile RT_FAR )(ppv), (void RT_FAR )(pvNew), (void RT_FAR )(pvOld), (void RT_FAR RT_FAR *)(ppvOld))
1403	#endif
1404
1405
1406	/**
1407	* Virtualization unfriendly serializing instruction, always exits.
1408	*/
1409	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
1410	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMSerializeInstructionCpuId(void);
1411	#else
1412	DECLINLINE(void) ASMSerializeInstructionCpuId(void)
1413	{
1414	# if RT_INLINE_ASM_GNU_STYLE
1415	RTCCUINTREG xAX = 0;
1416	# ifdef RT_ARCH_AMD64
1417	__asm__ __volatile__ ("cpuid"
1418	: "=a" (xAX)
1419	: "0" (xAX)
1420	: "rbx", "rcx", "rdx", "memory");
1421	# elif (defined(PIC) \|\| defined(__PIC__)) && defined(__i386__)
1422	__asm__ __volatile__ ("push %%ebx\n\t"
1423	"cpuid\n\t"
1424	"pop %%ebx\n\t"
1425	: "=a" (xAX)
1426	: "0" (xAX)
1427	: "ecx", "edx", "memory");
1428	# else
1429	__asm__ __volatile__ ("cpuid"
1430	: "=a" (xAX)
1431	: "0" (xAX)
1432	: "ebx", "ecx", "edx", "memory");
1433	# endif
1434
1435	# elif RT_INLINE_ASM_USES_INTRIN
1436	int aInfo[4];
1437	_ReadWriteBarrier();
1438	__cpuid(aInfo, 0);
1439
1440	# else
1441	__asm
1442	{
1443	push ebx
1444	xor eax, eax
1445	cpuid
1446	pop ebx
1447	}
1448	# endif
1449	}
1450	#endif
1451
1452	/**
1453	* Virtualization friendly serializing instruction, though more expensive.
1454	*/
1455	#if RT_INLINE_ASM_EXTERNAL
1456	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMSerializeInstructionIRet(void);
1457	#else
1458	DECLINLINE(void) ASMSerializeInstructionIRet(void)
1459	{
1460	# if RT_INLINE_ASM_GNU_STYLE
1461	# ifdef RT_ARCH_AMD64
1462	__asm__ __volatile__ ("movq %%rsp,%%r10\n\t"
1463	"subq $128, %%rsp\n\t" /redzone/
1464	"mov %%ss, %%eax\n\t"
1465	"pushq %%rax\n\t"
1466	"pushq %%r10\n\t"
1467	"pushfq\n\t"
1468	"movl %%cs, %%eax\n\t"
1469	"pushq %%rax\n\t"
1470	"leaq 1f(%%rip), %%rax\n\t"
1471	"pushq %%rax\n\t"
1472	"iretq\n\t"
1473	"1:\n\t"
1474	::: "rax", "r10", "memory");
1475	# else
1476	__asm__ __volatile__ ("pushfl\n\t"
1477	"pushl %%cs\n\t"
1478	"pushl $1f\n\t"
1479	"iretl\n\t"
1480	"1:\n\t"
1481	::: "memory");
1482	# endif
1483
1484	# else
1485	__asm
1486	{
1487	pushfd
1488	push cs
1489	push la_ret
1490	iretd
1491	la_ret:
1492	}
1493	# endif
1494	}
1495	#endif
1496
1497	/**
1498	* Virtualization friendlier serializing instruction, may still cause exits.
1499	*/
1500	#if RT_INLINE_ASM_EXTERNAL && RT_INLINE_ASM_USES_INTRIN < 15
1501	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMSerializeInstructionRdTscp(void);
1502	#else
1503	DECLINLINE(void) ASMSerializeInstructionRdTscp(void)
1504	{
1505	# if RT_INLINE_ASM_GNU_STYLE
1506	/* rdtscp is not supported by ancient linux build VM of course :-( */
1507	# ifdef RT_ARCH_AMD64
1508	/__asm__ __volatile__("rdtscp\n\t" ::: "rax", "rdx, "rcx"); /
1509	__asm__ __volatile__(".byte 0x0f,0x01,0xf9\n\t" ::: "rax", "rdx", "rcx", "memory");
1510	# else
1511	/__asm__ __volatile__("rdtscp\n\t" ::: "eax", "edx, "ecx"); /
1512	__asm__ __volatile__(".byte 0x0f,0x01,0xf9\n\t" ::: "eax", "edx", "ecx", "memory");
1513	# endif
1514	# else
1515	# if RT_INLINE_ASM_USES_INTRIN >= 15
1516	uint32_t uIgnore;
1517	_ReadWriteBarrier();
1518	(void)__rdtscp(&uIgnore);
1519	(void)uIgnore;
1520	# else
1521	__asm
1522	{
1523	rdtscp
1524	}
1525	# endif
1526	# endif
1527	}
1528	#endif
1529
1530
1531	/**
1532	* Serialize Instruction.
1533	*/
1534	#if (defined(RT_ARCH_X86) && ARCH_BITS == 16) \|\| defined(IN_GUEST)
1535	# define ASMSerializeInstruction() ASMSerializeInstructionIRet()
1536	#elif defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
1537	# define ASMSerializeInstruction() ASMSerializeInstructionCpuId()
1538	#elif defined(RT_ARCH_SPARC64)
1539	RTDECL(void) ASMSerializeInstruction(void);
1540	#else
1541	# error "Port me"
1542	#endif
1543
1544
1545	/**
1546	* Memory fence, waits for any pending writes and reads to complete.
1547	*/
1548	DECLINLINE(void) ASMMemoryFence(void)
1549	{
1550	#if defined(RT_ARCH_AMD64) \|\| (defined(RT_ARCH_X86) && !defined(RT_WITH_OLD_CPU_SUPPORT))
1551	# if RT_INLINE_ASM_GNU_STYLE
1552	__asm__ __volatile__ (".byte 0x0f,0xae,0xf0\n\t");
1553	# elif RT_INLINE_ASM_USES_INTRIN
1554	_mm_mfence();
1555	# else
1556	__asm
1557	{
1558	_emit 0x0f
1559	_emit 0xae
1560	_emit 0xf0
1561	}
1562	# endif
1563	#elif ARCH_BITS == 16
1564	uint16_t volatile u16;
1565	ASMAtomicXchgU16(&u16, 0);
1566	#else
1567	uint32_t volatile u32;
1568	ASMAtomicXchgU32(&u32, 0);
1569	#endif
1570	}
1571
1572
1573	/**
1574	* Write fence, waits for any pending writes to complete.
1575	*/
1576	DECLINLINE(void) ASMWriteFence(void)
1577	{
1578	#if defined(RT_ARCH_AMD64) \|\| (defined(RT_ARCH_X86) && !defined(RT_WITH_OLD_CPU_SUPPORT))
1579	# if RT_INLINE_ASM_GNU_STYLE
1580	__asm__ __volatile__ (".byte 0x0f,0xae,0xf8\n\t");
1581	# elif RT_INLINE_ASM_USES_INTRIN
1582	_mm_sfence();
1583	# else
1584	__asm
1585	{
1586	_emit 0x0f
1587	_emit 0xae
1588	_emit 0xf8
1589	}
1590	# endif
1591	#else
1592	ASMMemoryFence();
1593	#endif
1594	}
1595
1596
1597	/**
1598	* Read fence, waits for any pending reads to complete.
1599	*/
1600	DECLINLINE(void) ASMReadFence(void)
1601	{
1602	#if defined(RT_ARCH_AMD64) \|\| (defined(RT_ARCH_X86) && !defined(RT_WITH_OLD_CPU_SUPPORT))
1603	# if RT_INLINE_ASM_GNU_STYLE
1604	__asm__ __volatile__ (".byte 0x0f,0xae,0xe8\n\t");
1605	# elif RT_INLINE_ASM_USES_INTRIN
1606	_mm_lfence();
1607	# else
1608	__asm
1609	{
1610	_emit 0x0f
1611	_emit 0xae
1612	_emit 0xe8
1613	}
1614	# endif
1615	#else
1616	ASMMemoryFence();
1617	#endif
1618	}
1619
1620
1621	/**
1622	* Atomically reads an unsigned 8-bit value, ordered.
1623	*
1624	* @returns Current *pu8 value
1625	* @param pu8 Pointer to the 8-bit variable to read.
1626	*/
1627	DECLINLINE(uint8_t) ASMAtomicReadU8(volatile uint8_t RT_FAR *pu8)
1628	{
1629	ASMMemoryFence();
1630	return pu8; / byte reads are atomic on x86 */
1631	}
1632
1633
1634	/**
1635	* Atomically reads an unsigned 8-bit value, unordered.
1636	*
1637	* @returns Current *pu8 value
1638	* @param pu8 Pointer to the 8-bit variable to read.
1639	*/
1640	DECLINLINE(uint8_t) ASMAtomicUoReadU8(volatile uint8_t RT_FAR *pu8)
1641	{
1642	return pu8; / byte reads are atomic on x86 */
1643	}
1644
1645
1646	/**
1647	* Atomically reads a signed 8-bit value, ordered.
1648	*
1649	* @returns Current *pi8 value
1650	* @param pi8 Pointer to the 8-bit variable to read.
1651	*/
1652	DECLINLINE(int8_t) ASMAtomicReadS8(volatile int8_t RT_FAR *pi8)
1653	{
1654	ASMMemoryFence();
1655	return pi8; / byte reads are atomic on x86 */
1656	}
1657
1658
1659	/**
1660	* Atomically reads a signed 8-bit value, unordered.
1661	*
1662	* @returns Current *pi8 value
1663	* @param pi8 Pointer to the 8-bit variable to read.
1664	*/
1665	DECLINLINE(int8_t) ASMAtomicUoReadS8(volatile int8_t RT_FAR *pi8)
1666	{
1667	return pi8; / byte reads are atomic on x86 */
1668	}
1669
1670
1671	/**
1672	* Atomically reads an unsigned 16-bit value, ordered.
1673	*
1674	* @returns Current *pu16 value
1675	* @param pu16 Pointer to the 16-bit variable to read.
1676	*/
1677	DECLINLINE(uint16_t) ASMAtomicReadU16(volatile uint16_t RT_FAR *pu16)
1678	{
1679	ASMMemoryFence();
1680	Assert(!((uintptr_t)pu16 & 1));
1681	return *pu16;
1682	}
1683
1684
1685	/**
1686	* Atomically reads an unsigned 16-bit value, unordered.
1687	*
1688	* @returns Current *pu16 value
1689	* @param pu16 Pointer to the 16-bit variable to read.
1690	*/
1691	DECLINLINE(uint16_t) ASMAtomicUoReadU16(volatile uint16_t RT_FAR *pu16)
1692	{
1693	Assert(!((uintptr_t)pu16 & 1));
1694	return *pu16;
1695	}
1696
1697
1698	/**
1699	* Atomically reads a signed 16-bit value, ordered.
1700	*
1701	* @returns Current *pi16 value
1702	* @param pi16 Pointer to the 16-bit variable to read.
1703	*/
1704	DECLINLINE(int16_t) ASMAtomicReadS16(volatile int16_t RT_FAR *pi16)
1705	{
1706	ASMMemoryFence();
1707	Assert(!((uintptr_t)pi16 & 1));
1708	return *pi16;
1709	}
1710
1711
1712	/**
1713	* Atomically reads a signed 16-bit value, unordered.
1714	*
1715	* @returns Current *pi16 value
1716	* @param pi16 Pointer to the 16-bit variable to read.
1717	*/
1718	DECLINLINE(int16_t) ASMAtomicUoReadS16(volatile int16_t RT_FAR *pi16)
1719	{
1720	Assert(!((uintptr_t)pi16 & 1));
1721	return *pi16;
1722	}
1723
1724
1725	/**
1726	* Atomically reads an unsigned 32-bit value, ordered.
1727	*
1728	* @returns Current *pu32 value
1729	* @param pu32 Pointer to the 32-bit variable to read.
1730	*/
1731	DECLINLINE(uint32_t) ASMAtomicReadU32(volatile uint32_t RT_FAR *pu32)
1732	{
1733	ASMMemoryFence();
1734	Assert(!((uintptr_t)pu32 & 3));
1735	#if ARCH_BITS == 16
1736	AssertFailed(); /** @todo 16-bit */
1737	#endif
1738	return *pu32;
1739	}
1740
1741
1742	/**
1743	* Atomically reads an unsigned 32-bit value, unordered.
1744	*
1745	* @returns Current *pu32 value
1746	* @param pu32 Pointer to the 32-bit variable to read.
1747	*/
1748	DECLINLINE(uint32_t) ASMAtomicUoReadU32(volatile uint32_t RT_FAR *pu32)
1749	{
1750	Assert(!((uintptr_t)pu32 & 3));
1751	#if ARCH_BITS == 16
1752	AssertFailed(); /** @todo 16-bit */
1753	#endif
1754	return *pu32;
1755	}
1756
1757
1758	/**
1759	* Atomically reads a signed 32-bit value, ordered.
1760	*
1761	* @returns Current *pi32 value
1762	* @param pi32 Pointer to the 32-bit variable to read.
1763	*/
1764	DECLINLINE(int32_t) ASMAtomicReadS32(volatile int32_t RT_FAR *pi32)
1765	{
1766	ASMMemoryFence();
1767	Assert(!((uintptr_t)pi32 & 3));
1768	#if ARCH_BITS == 16
1769	AssertFailed(); /** @todo 16-bit */
1770	#endif
1771	return *pi32;
1772	}
1773
1774
1775	/**
1776	* Atomically reads a signed 32-bit value, unordered.
1777	*
1778	* @returns Current *pi32 value
1779	* @param pi32 Pointer to the 32-bit variable to read.
1780	*/
1781	DECLINLINE(int32_t) ASMAtomicUoReadS32(volatile int32_t RT_FAR *pi32)
1782	{
1783	Assert(!((uintptr_t)pi32 & 3));
1784	#if ARCH_BITS == 16
1785	AssertFailed(); /** @todo 16-bit */
1786	#endif
1787	return *pi32;
1788	}
1789
1790
1791	/**
1792	* Atomically reads an unsigned 64-bit value, ordered.
1793	*
1794	* @returns Current *pu64 value
1795	* @param pu64 Pointer to the 64-bit variable to read.
1796	* The memory pointed to must be writable.
1797	*
1798	* @remarks This may fault if the memory is read-only!
1799	* @remarks x86: Requires a Pentium or later.
1800	*/
1801	#if (RT_INLINE_ASM_EXTERNAL && !defined(RT_ARCH_AMD64)) \
1802	\|\| RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC
1803	RT_ASM_DECL_PRAGMA_WATCOM(uint64_t) ASMAtomicReadU64(volatile uint64_t RT_FAR *pu64);
1804	#else
1805	DECLINLINE(uint64_t) ASMAtomicReadU64(volatile uint64_t RT_FAR *pu64)
1806	{
1807	uint64_t u64;
1808	# ifdef RT_ARCH_AMD64
1809	Assert(!((uintptr_t)pu64 & 7));
1810	/*# if RT_INLINE_ASM_GNU_STYLE
1811	__asm__ __volatile__( "mfence\n\t"
1812	"movq %1, %0\n\t"
1813	: "=r" (u64)
1814	: "m" (*pu64));
1815	# else
1816	__asm
1817	{
1818	mfence
1819	mov rdx, [pu64]
1820	mov rax, [rdx]
1821	mov [u64], rax
1822	}
1823	# endif*/
1824	ASMMemoryFence();
1825	u64 = *pu64;
1826	# else /* !RT_ARCH_AMD64 */
1827	# if RT_INLINE_ASM_GNU_STYLE
1828	# if defined(PIC) \|\| defined(__PIC__)
1829	uint32_t u32EBX = 0;
1830	Assert(!((uintptr_t)pu64 & 7));
1831	__asm__ __volatile__("xchgl %%ebx, %3\n\t"
1832	"lock; cmpxchg8b (%5)\n\t"
1833	"movl %3, %%ebx\n\t"
1834	: "=A" (u64),
1835	# if RT_GNUC_PREREQ(4, 3)
1836	"+m" (*pu64)
1837	# else
1838	"=m" (*pu64)
1839	# endif
1840	: "0" (0ULL),
1841	"m" (u32EBX),
1842	"c" (0),
1843	"S" (pu64));
1844	# else /* !PIC */
1845	__asm__ __volatile__("lock; cmpxchg8b %1\n\t"
1846	: "=A" (u64),
1847	"+m" (*pu64)
1848	: "0" (0ULL),
1849	"b" (0),
1850	"c" (0));
1851	# endif
1852	# else
1853	Assert(!((uintptr_t)pu64 & 7));
1854	__asm
1855	{
1856	xor eax, eax
1857	xor edx, edx
1858	mov edi, pu64
1859	xor ecx, ecx
1860	xor ebx, ebx
1861	lock cmpxchg8b [edi]
1862	mov dword ptr [u64], eax
1863	mov dword ptr [u64 + 4], edx
1864	}
1865	# endif
1866	# endif /* !RT_ARCH_AMD64 */
1867	return u64;
1868	}
1869	#endif
1870
1871
1872	/**
1873	* Atomically reads an unsigned 64-bit value, unordered.
1874	*
1875	* @returns Current *pu64 value
1876	* @param pu64 Pointer to the 64-bit variable to read.
1877	* The memory pointed to must be writable.
1878	*
1879	* @remarks This may fault if the memory is read-only!
1880	* @remarks x86: Requires a Pentium or later.
1881	*/
1882	#if !defined(RT_ARCH_AMD64) \
1883	&& ( (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \
1884	\|\| RT_INLINE_DONT_MIX_CMPXCHG8B_AND_PIC)
1885	RT_ASM_DECL_PRAGMA_WATCOM(uint64_t) ASMAtomicUoReadU64(volatile uint64_t RT_FAR *pu64);
1886	#else
1887	DECLINLINE(uint64_t) ASMAtomicUoReadU64(volatile uint64_t RT_FAR *pu64)
1888	{
1889	uint64_t u64;
1890	# ifdef RT_ARCH_AMD64
1891	Assert(!((uintptr_t)pu64 & 7));
1892	/*# if RT_INLINE_ASM_GNU_STYLE
1893	Assert(!((uintptr_t)pu64 & 7));
1894	__asm__ __volatile__("movq %1, %0\n\t"
1895	: "=r" (u64)
1896	: "m" (*pu64));
1897	# else
1898	__asm
1899	{
1900	mov rdx, [pu64]
1901	mov rax, [rdx]
1902	mov [u64], rax
1903	}
1904	# endif */
1905	u64 = *pu64;
1906	# else /* !RT_ARCH_AMD64 */
1907	# if RT_INLINE_ASM_GNU_STYLE
1908	# if defined(PIC) \|\| defined(__PIC__)
1909	uint32_t u32EBX = 0;
1910	uint32_t u32Spill;
1911	Assert(!((uintptr_t)pu64 & 7));
1912	__asm__ __volatile__("xor %%eax,%%eax\n\t"
1913	"xor %%ecx,%%ecx\n\t"
1914	"xor %%edx,%%edx\n\t"
1915	"xchgl %%ebx, %3\n\t"
1916	"lock; cmpxchg8b (%4)\n\t"
1917	"movl %3, %%ebx\n\t"
1918	: "=A" (u64),
1919	# if RT_GNUC_PREREQ(4, 3)
1920	"+m" (*pu64),
1921	# else
1922	"=m" (*pu64),
1923	# endif
1924	"=c" (u32Spill)
1925	: "m" (u32EBX),
1926	"S" (pu64));
1927	# else /* !PIC */
1928	__asm__ __volatile__("lock; cmpxchg8b %1\n\t"
1929	: "=A" (u64),
1930	"+m" (*pu64)
1931	: "0" (0ULL),
1932	"b" (0),
1933	"c" (0));
1934	# endif
1935	# else
1936	Assert(!((uintptr_t)pu64 & 7));
1937	__asm
1938	{
1939	xor eax, eax
1940	xor edx, edx
1941	mov edi, pu64
1942	xor ecx, ecx
1943	xor ebx, ebx
1944	lock cmpxchg8b [edi]
1945	mov dword ptr [u64], eax
1946	mov dword ptr [u64 + 4], edx
1947	}
1948	# endif
1949	# endif /* !RT_ARCH_AMD64 */
1950	return u64;
1951	}
1952	#endif
1953
1954
1955	/**
1956	* Atomically reads a signed 64-bit value, ordered.
1957	*
1958	* @returns Current *pi64 value
1959	* @param pi64 Pointer to the 64-bit variable to read.
1960	* The memory pointed to must be writable.
1961	*
1962	* @remarks This may fault if the memory is read-only!
1963	* @remarks x86: Requires a Pentium or later.
1964	*/
1965	DECLINLINE(int64_t) ASMAtomicReadS64(volatile int64_t RT_FAR *pi64)
1966	{
1967	return (int64_t)ASMAtomicReadU64((volatile uint64_t RT_FAR *)pi64);
1968	}
1969
1970
1971	/**
1972	* Atomically reads a signed 64-bit value, unordered.
1973	*
1974	* @returns Current *pi64 value
1975	* @param pi64 Pointer to the 64-bit variable to read.
1976	* The memory pointed to must be writable.
1977	*
1978	* @remarks This will fault if the memory is read-only!
1979	* @remarks x86: Requires a Pentium or later.
1980	*/
1981	DECLINLINE(int64_t) ASMAtomicUoReadS64(volatile int64_t RT_FAR *pi64)
1982	{
1983	return (int64_t)ASMAtomicUoReadU64((volatile uint64_t RT_FAR *)pi64);
1984	}
1985
1986
1987	/**
1988	* Atomically reads a size_t value, ordered.
1989	*
1990	* @returns Current *pcb value
1991	* @param pcb Pointer to the size_t variable to read.
1992	*/
1993	DECLINLINE(size_t) ASMAtomicReadZ(size_t volatile RT_FAR *pcb)
1994	{
1995	#if ARCH_BITS == 64
1996	return ASMAtomicReadU64((uint64_t volatile RT_FAR *)pcb);
1997	#elif ARCH_BITS == 32
1998	return ASMAtomicReadU32((uint32_t volatile RT_FAR *)pcb);
1999	#elif ARCH_BITS == 16
2000	AssertCompileSize(size_t, 2);
2001	return ASMAtomicReadU16((uint16_t volatile RT_FAR *)pcb);
2002	#else
2003	# error "Unsupported ARCH_BITS value"
2004	#endif
2005	}
2006
2007
2008	/**
2009	* Atomically reads a size_t value, unordered.
2010	*
2011	* @returns Current *pcb value
2012	* @param pcb Pointer to the size_t variable to read.
2013	*/
2014	DECLINLINE(size_t) ASMAtomicUoReadZ(size_t volatile RT_FAR *pcb)
2015	{
2016	#if ARCH_BITS == 64 \|\| ARCH_BITS == 16
2017	return ASMAtomicUoReadU64((uint64_t volatile RT_FAR *)pcb);
2018	#elif ARCH_BITS == 32
2019	return ASMAtomicUoReadU32((uint32_t volatile RT_FAR *)pcb);
2020	#elif ARCH_BITS == 16
2021	AssertCompileSize(size_t, 2);
2022	return ASMAtomicUoReadU16((uint16_t volatile RT_FAR *)pcb);
2023	#else
2024	# error "Unsupported ARCH_BITS value"
2025	#endif
2026	}
2027
2028
2029	/**
2030	* Atomically reads a pointer value, ordered.
2031	*
2032	* @returns Current *pv value
2033	* @param ppv Pointer to the pointer variable to read.
2034	*
2035	* @remarks Please use ASMAtomicReadPtrT, it provides better type safety and
2036	* requires less typing (no casts).
2037	*/
2038	DECLINLINE(void RT_FAR ) ASMAtomicReadPtr(void RT_FAR volatile RT_FAR *ppv)
2039	{
2040	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
2041	return (void RT_FAR )ASMAtomicReadU32((volatile uint32_t RT_FAR )(void RT_FAR *)ppv);
2042	#elif ARCH_BITS == 64
2043	return (void RT_FAR )ASMAtomicReadU64((volatile uint64_t RT_FAR )(void RT_FAR *)ppv);
2044	#else
2045	# error "ARCH_BITS is bogus"
2046	#endif
2047	}
2048
2049	/**
2050	* Convenience macro for avoiding the annoying casting with ASMAtomicReadPtr.
2051	*
2052	* @returns Current *pv value
2053	* @param ppv Pointer to the pointer variable to read.
2054	* @param Type The type of *ppv, sans volatile.
2055	*/
2056	#ifdef __GNUC__ /* 8.2.0 requires -Wno-ignored-qualifiers */
2057	# define ASMAtomicReadPtrT(ppv, Type) \
2058	__extension__ \
2059	({\
2060	__typeof__((ppv)) volatile ppvTypeChecked = (ppv); \
2061	Type pvTypeChecked = (__typeof__((ppv))) ASMAtomicReadPtr((void volatile *)ppvTypeChecked); \
2062	pvTypeChecked; \
2063	})
2064	#else
2065	# define ASMAtomicReadPtrT(ppv, Type) \
2066	(Type)ASMAtomicReadPtr((void RT_FAR * volatile RT_FAR *)(ppv))
2067	#endif
2068
2069
2070	/**
2071	* Atomically reads a pointer value, unordered.
2072	*
2073	* @returns Current *pv value
2074	* @param ppv Pointer to the pointer variable to read.
2075	*
2076	* @remarks Please use ASMAtomicUoReadPtrT, it provides better type safety and
2077	* requires less typing (no casts).
2078	*/
2079	DECLINLINE(void RT_FAR ) ASMAtomicUoReadPtr(void RT_FAR volatile RT_FAR *ppv)
2080	{
2081	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
2082	return (void RT_FAR )ASMAtomicUoReadU32((volatile uint32_t RT_FAR )(void RT_FAR *)ppv);
2083	#elif ARCH_BITS == 64
2084	return (void RT_FAR )ASMAtomicUoReadU64((volatile uint64_t RT_FAR )(void RT_FAR *)ppv);
2085	#else
2086	# error "ARCH_BITS is bogus"
2087	#endif
2088	}
2089
2090
2091	/**
2092	* Convenience macro for avoiding the annoying casting with ASMAtomicUoReadPtr.
2093	*
2094	* @returns Current *pv value
2095	* @param ppv Pointer to the pointer variable to read.
2096	* @param Type The type of *ppv, sans volatile.
2097	*/
2098	#ifdef __GNUC__ /* 8.2.0 requires -Wno-ignored-qualifiers */
2099	# define ASMAtomicUoReadPtrT(ppv, Type) \
2100	__extension__ \
2101	({\
2102	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
2103	Type pvTypeChecked = (__typeof__((ppv))) ASMAtomicUoReadPtr((void volatile *)ppvTypeChecked); \
2104	pvTypeChecked; \
2105	})
2106	#else
2107	# define ASMAtomicUoReadPtrT(ppv, Type) \
2108	(Type)ASMAtomicUoReadPtr((void RT_FAR * volatile RT_FAR *)(ppv))
2109	#endif
2110
2111
2112	/**
2113	* Atomically reads a boolean value, ordered.
2114	*
2115	* @returns Current *pf value
2116	* @param pf Pointer to the boolean variable to read.
2117	*/
2118	DECLINLINE(bool) ASMAtomicReadBool(volatile bool RT_FAR *pf)
2119	{
2120	ASMMemoryFence();
2121	return pf; / byte reads are atomic on x86 */
2122	}
2123
2124
2125	/**
2126	* Atomically reads a boolean value, unordered.
2127	*
2128	* @returns Current *pf value
2129	* @param pf Pointer to the boolean variable to read.
2130	*/
2131	DECLINLINE(bool) ASMAtomicUoReadBool(volatile bool RT_FAR *pf)
2132	{
2133	return pf; / byte reads are atomic on x86 */
2134	}
2135
2136
2137	/**
2138	* Atomically read a typical IPRT handle value, ordered.
2139	*
2140	* @param ph Pointer to the handle variable to read.
2141	* @param phRes Where to store the result.
2142	*
2143	* @remarks This doesn't currently work for all handles (like RTFILE).
2144	*/
2145	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
2146	# define ASMAtomicReadHandle(ph, phRes) \
2147	do { \
2148	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
2149	AssertCompile(sizeof(*(phRes)) == sizeof(uint32_t)); \
2150	(uint32_t RT_FAR )(phRes) = ASMAtomicReadU32((uint32_t volatile RT_FAR *)(ph)); \
2151	} while (0)
2152	#elif HC_ARCH_BITS == 64
2153	# define ASMAtomicReadHandle(ph, phRes) \
2154	do { \
2155	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
2156	AssertCompile(sizeof(*(phRes)) == sizeof(uint64_t)); \
2157	(uint64_t RT_FAR )(phRes) = ASMAtomicReadU64((uint64_t volatile RT_FAR *)(ph)); \
2158	} while (0)
2159	#else
2160	# error HC_ARCH_BITS
2161	#endif
2162
2163
2164	/**
2165	* Atomically read a typical IPRT handle value, unordered.
2166	*
2167	* @param ph Pointer to the handle variable to read.
2168	* @param phRes Where to store the result.
2169	*
2170	* @remarks This doesn't currently work for all handles (like RTFILE).
2171	*/
2172	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
2173	# define ASMAtomicUoReadHandle(ph, phRes) \
2174	do { \
2175	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
2176	AssertCompile(sizeof(*(phRes)) == sizeof(uint32_t)); \
2177	(uint32_t RT_FAR )(phRes) = ASMAtomicUoReadU32((uint32_t volatile RT_FAR *)(ph)); \
2178	} while (0)
2179	#elif HC_ARCH_BITS == 64
2180	# define ASMAtomicUoReadHandle(ph, phRes) \
2181	do { \
2182	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
2183	AssertCompile(sizeof(*(phRes)) == sizeof(uint64_t)); \
2184	(uint64_t RT_FAR )(phRes) = ASMAtomicUoReadU64((uint64_t volatile RT_FAR *)(ph)); \
2185	} while (0)
2186	#else
2187	# error HC_ARCH_BITS
2188	#endif
2189
2190
2191	/**
2192	* Atomically read a value which size might differ
2193	* between platforms or compilers, ordered.
2194	*
2195	* @param pu Pointer to the variable to read.
2196	* @param puRes Where to store the result.
2197	*/
2198	#define ASMAtomicReadSize(pu, puRes) \
2199	do { \
2200	switch (sizeof(*(pu))) { \
2201	case 1: (uint8_t RT_FAR )(puRes) = ASMAtomicReadU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu)); break; \
2202	case 2: (uint16_t RT_FAR )(puRes) = ASMAtomicReadU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu)); break; \
2203	case 4: (uint32_t RT_FAR )(puRes) = ASMAtomicReadU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu)); break; \
2204	case 8: (uint64_t RT_FAR )(puRes) = ASMAtomicReadU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu)); break; \
2205	default: AssertMsgFailed(("ASMAtomicReadSize: size %d is not supported\n", sizeof(*(pu)))); \
2206	} \
2207	} while (0)
2208
2209
2210	/**
2211	* Atomically read a value which size might differ
2212	* between platforms or compilers, unordered.
2213	*
2214	* @param pu Pointer to the variable to read.
2215	* @param puRes Where to store the result.
2216	*/
2217	#define ASMAtomicUoReadSize(pu, puRes) \
2218	do { \
2219	switch (sizeof(*(pu))) { \
2220	case 1: (uint8_t RT_FAR )(puRes) = ASMAtomicUoReadU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu)); break; \
2221	case 2: (uint16_t RT_FAR )(puRes) = ASMAtomicUoReadU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu)); break; \
2222	case 4: (uint32_t RT_FAR )(puRes) = ASMAtomicUoReadU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu)); break; \
2223	case 8: (uint64_t RT_FAR )(puRes) = ASMAtomicUoReadU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu)); break; \
2224	default: AssertMsgFailed(("ASMAtomicReadSize: size %d is not supported\n", sizeof(*(pu)))); \
2225	} \
2226	} while (0)
2227
2228
2229	/**
2230	* Atomically writes an unsigned 8-bit value, ordered.
2231	*
2232	* @param pu8 Pointer to the 8-bit variable.
2233	* @param u8 The 8-bit value to assign to *pu8.
2234	*/
2235	DECLINLINE(void) ASMAtomicWriteU8(volatile uint8_t RT_FAR *pu8, uint8_t u8)
2236	{
2237	ASMAtomicXchgU8(pu8, u8);
2238	}
2239
2240
2241	/**
2242	* Atomically writes an unsigned 8-bit value, unordered.
2243	*
2244	* @param pu8 Pointer to the 8-bit variable.
2245	* @param u8 The 8-bit value to assign to *pu8.
2246	*/
2247	DECLINLINE(void) ASMAtomicUoWriteU8(volatile uint8_t RT_FAR *pu8, uint8_t u8)
2248	{
2249	pu8 = u8; / byte writes are atomic on x86 */
2250	}
2251
2252
2253	/**
2254	* Atomically writes a signed 8-bit value, ordered.
2255	*
2256	* @param pi8 Pointer to the 8-bit variable to read.
2257	* @param i8 The 8-bit value to assign to *pi8.
2258	*/
2259	DECLINLINE(void) ASMAtomicWriteS8(volatile int8_t RT_FAR *pi8, int8_t i8)
2260	{
2261	ASMAtomicXchgS8(pi8, i8);
2262	}
2263
2264
2265	/**
2266	* Atomically writes a signed 8-bit value, unordered.
2267	*
2268	* @param pi8 Pointer to the 8-bit variable to write.
2269	* @param i8 The 8-bit value to assign to *pi8.
2270	*/
2271	DECLINLINE(void) ASMAtomicUoWriteS8(volatile int8_t RT_FAR *pi8, int8_t i8)
2272	{
2273	pi8 = i8; / byte writes are atomic on x86 */
2274	}
2275
2276
2277	/**
2278	* Atomically writes an unsigned 16-bit value, ordered.
2279	*
2280	* @param pu16 Pointer to the 16-bit variable to write.
2281	* @param u16 The 16-bit value to assign to *pu16.
2282	*/
2283	DECLINLINE(void) ASMAtomicWriteU16(volatile uint16_t RT_FAR *pu16, uint16_t u16)
2284	{
2285	ASMAtomicXchgU16(pu16, u16);
2286	}
2287
2288
2289	/**
2290	* Atomically writes an unsigned 16-bit value, unordered.
2291	*
2292	* @param pu16 Pointer to the 16-bit variable to write.
2293	* @param u16 The 16-bit value to assign to *pu16.
2294	*/
2295	DECLINLINE(void) ASMAtomicUoWriteU16(volatile uint16_t RT_FAR *pu16, uint16_t u16)
2296	{
2297	Assert(!((uintptr_t)pu16 & 1));
2298	*pu16 = u16;
2299	}
2300
2301
2302	/**
2303	* Atomically writes a signed 16-bit value, ordered.
2304	*
2305	* @param pi16 Pointer to the 16-bit variable to write.
2306	* @param i16 The 16-bit value to assign to *pi16.
2307	*/
2308	DECLINLINE(void) ASMAtomicWriteS16(volatile int16_t RT_FAR *pi16, int16_t i16)
2309	{
2310	ASMAtomicXchgS16(pi16, i16);
2311	}
2312
2313
2314	/**
2315	* Atomically writes a signed 16-bit value, unordered.
2316	*
2317	* @param pi16 Pointer to the 16-bit variable to write.
2318	* @param i16 The 16-bit value to assign to *pi16.
2319	*/
2320	DECLINLINE(void) ASMAtomicUoWriteS16(volatile int16_t RT_FAR *pi16, int16_t i16)
2321	{
2322	Assert(!((uintptr_t)pi16 & 1));
2323	*pi16 = i16;
2324	}
2325
2326
2327	/**
2328	* Atomically writes an unsigned 32-bit value, ordered.
2329	*
2330	* @param pu32 Pointer to the 32-bit variable to write.
2331	* @param u32 The 32-bit value to assign to *pu32.
2332	*/
2333	DECLINLINE(void) ASMAtomicWriteU32(volatile uint32_t RT_FAR *pu32, uint32_t u32)
2334	{
2335	ASMAtomicXchgU32(pu32, u32);
2336	}
2337
2338
2339	/**
2340	* Atomically writes an unsigned 32-bit value, unordered.
2341	*
2342	* @param pu32 Pointer to the 32-bit variable to write.
2343	* @param u32 The 32-bit value to assign to *pu32.
2344	*/
2345	DECLINLINE(void) ASMAtomicUoWriteU32(volatile uint32_t RT_FAR *pu32, uint32_t u32)
2346	{
2347	Assert(!((uintptr_t)pu32 & 3));
2348	#if ARCH_BITS >= 32
2349	*pu32 = u32;
2350	#else
2351	ASMAtomicXchgU32(pu32, u32);
2352	#endif
2353	}
2354
2355
2356	/**
2357	* Atomically writes a signed 32-bit value, ordered.
2358	*
2359	* @param pi32 Pointer to the 32-bit variable to write.
2360	* @param i32 The 32-bit value to assign to *pi32.
2361	*/
2362	DECLINLINE(void) ASMAtomicWriteS32(volatile int32_t RT_FAR *pi32, int32_t i32)
2363	{
2364	ASMAtomicXchgS32(pi32, i32);
2365	}
2366
2367
2368	/**
2369	* Atomically writes a signed 32-bit value, unordered.
2370	*
2371	* @param pi32 Pointer to the 32-bit variable to write.
2372	* @param i32 The 32-bit value to assign to *pi32.
2373	*/
2374	DECLINLINE(void) ASMAtomicUoWriteS32(volatile int32_t RT_FAR *pi32, int32_t i32)
2375	{
2376	Assert(!((uintptr_t)pi32 & 3));
2377	#if ARCH_BITS >= 32
2378	*pi32 = i32;
2379	#else
2380	ASMAtomicXchgS32(pi32, i32);
2381	#endif
2382	}
2383
2384
2385	/**
2386	* Atomically writes an unsigned 64-bit value, ordered.
2387	*
2388	* @param pu64 Pointer to the 64-bit variable to write.
2389	* @param u64 The 64-bit value to assign to *pu64.
2390	*/
2391	DECLINLINE(void) ASMAtomicWriteU64(volatile uint64_t RT_FAR *pu64, uint64_t u64)
2392	{
2393	ASMAtomicXchgU64(pu64, u64);
2394	}
2395
2396
2397	/**
2398	* Atomically writes an unsigned 64-bit value, unordered.
2399	*
2400	* @param pu64 Pointer to the 64-bit variable to write.
2401	* @param u64 The 64-bit value to assign to *pu64.
2402	*/
2403	DECLINLINE(void) ASMAtomicUoWriteU64(volatile uint64_t RT_FAR *pu64, uint64_t u64)
2404	{
2405	Assert(!((uintptr_t)pu64 & 7));
2406	#if ARCH_BITS == 64
2407	*pu64 = u64;
2408	#else
2409	ASMAtomicXchgU64(pu64, u64);
2410	#endif
2411	}
2412
2413
2414	/**
2415	* Atomically writes a signed 64-bit value, ordered.
2416	*
2417	* @param pi64 Pointer to the 64-bit variable to write.
2418	* @param i64 The 64-bit value to assign to *pi64.
2419	*/
2420	DECLINLINE(void) ASMAtomicWriteS64(volatile int64_t RT_FAR *pi64, int64_t i64)
2421	{
2422	ASMAtomicXchgS64(pi64, i64);
2423	}
2424
2425
2426	/**
2427	* Atomically writes a signed 64-bit value, unordered.
2428	*
2429	* @param pi64 Pointer to the 64-bit variable to write.
2430	* @param i64 The 64-bit value to assign to *pi64.
2431	*/
2432	DECLINLINE(void) ASMAtomicUoWriteS64(volatile int64_t RT_FAR *pi64, int64_t i64)
2433	{
2434	Assert(!((uintptr_t)pi64 & 7));
2435	#if ARCH_BITS == 64
2436	*pi64 = i64;
2437	#else
2438	ASMAtomicXchgS64(pi64, i64);
2439	#endif
2440	}
2441
2442
2443	/**
2444	* Atomically writes a size_t value, ordered.
2445	*
2446	* @returns nothing.
2447	* @param pcb Pointer to the size_t variable to write.
2448	* @param cb The value to assign to *pcb.
2449	*/
2450	DECLINLINE(void) ASMAtomicWriteZ(volatile size_t RT_FAR *pcb, size_t cb)
2451	{
2452	#if ARCH_BITS == 64
2453	ASMAtomicWriteU64((uint64_t volatile *)pcb, cb);
2454	#elif ARCH_BITS == 32
2455	ASMAtomicWriteU32((uint32_t volatile *)pcb, cb);
2456	#elif ARCH_BITS == 16
2457	AssertCompileSize(size_t, 2);
2458	ASMAtomicWriteU16((uint16_t volatile *)pcb, cb);
2459	#else
2460	# error "Unsupported ARCH_BITS value"
2461	#endif
2462	}
2463
2464
2465	/**
2466	* Atomically writes a boolean value, unordered.
2467	*
2468	* @param pf Pointer to the boolean variable to write.
2469	* @param f The boolean value to assign to *pf.
2470	*/
2471	DECLINLINE(void) ASMAtomicWriteBool(volatile bool RT_FAR *pf, bool f)
2472	{
2473	ASMAtomicWriteU8((uint8_t volatile RT_FAR *)pf, f);
2474	}
2475
2476
2477	/**
2478	* Atomically writes a boolean value, unordered.
2479	*
2480	* @param pf Pointer to the boolean variable to write.
2481	* @param f The boolean value to assign to *pf.
2482	*/
2483	DECLINLINE(void) ASMAtomicUoWriteBool(volatile bool RT_FAR *pf, bool f)
2484	{
2485	pf = f; / byte writes are atomic on x86 */
2486	}
2487
2488
2489	/**
2490	* Atomically writes a pointer value, ordered.
2491	*
2492	* @param ppv Pointer to the pointer variable to write.
2493	* @param pv The pointer value to assign to *ppv.
2494	*/
2495	DECLINLINE(void) ASMAtomicWritePtrVoid(void RT_FAR * volatile RT_FAR ppv, const void pv)
2496	{
2497	#if ARCH_BITS == 32 \|\| ARCH_BITS == 16
2498	ASMAtomicWriteU32((volatile uint32_t RT_FAR )(void RT_FAR )ppv, (uint32_t)pv);
2499	#elif ARCH_BITS == 64
2500	ASMAtomicWriteU64((volatile uint64_t RT_FAR )(void RT_FAR )ppv, (uint64_t)pv);
2501	#else
2502	# error "ARCH_BITS is bogus"
2503	#endif
2504	}
2505
2506
2507	/**
2508	* Atomically writes a pointer value, ordered.
2509	*
2510	* @param ppv Pointer to the pointer variable to write.
2511	* @param pv The pointer value to assign to *ppv. If NULL use
2512	* ASMAtomicWriteNullPtr or you'll land in trouble.
2513	*
2514	* @remarks This is relatively type safe on GCC platforms when @a pv isn't
2515	* NULL.
2516	*/
2517	#ifdef __GNUC__
2518	# define ASMAtomicWritePtr(ppv, pv) \
2519	do \
2520	{ \
2521	__typeof__((ppv)) volatile RT_FAR const ppvTypeChecked = (ppv); \
2522	__typeof__(*(ppv)) const pvTypeChecked = (pv); \
2523	\
2524	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2525	AssertCompile(sizeof(pv) == sizeof(void RT_FAR *)); \
2526	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2527	\
2528	ASMAtomicWritePtrVoid((void RT_FAR * volatile RT_FAR )(ppvTypeChecked), (void RT_FAR )(pvTypeChecked)); \
2529	} while (0)
2530	#else
2531	# define ASMAtomicWritePtr(ppv, pv) \
2532	do \
2533	{ \
2534	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2535	AssertCompile(sizeof(pv) == sizeof(void RT_FAR *)); \
2536	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2537	\
2538	ASMAtomicWritePtrVoid((void RT_FAR * volatile RT_FAR )(ppv), (void RT_FAR )(pv)); \
2539	} while (0)
2540	#endif
2541
2542
2543	/**
2544	* Atomically sets a pointer to NULL, ordered.
2545	*
2546	* @param ppv Pointer to the pointer variable that should be set to NULL.
2547	*
2548	* @remarks This is relatively type safe on GCC platforms.
2549	*/
2550	#if RT_GNUC_PREREQ(4, 2)
2551	# define ASMAtomicWriteNullPtr(ppv) \
2552	do \
2553	{ \
2554	__typeof__((ppv)) const ppvTypeChecked = (ppv); \
2555	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2556	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2557	ASMAtomicWritePtrVoid((void RT_FAR * volatile RT_FAR *)(ppvTypeChecked), NULL); \
2558	} while (0)
2559	#else
2560	# define ASMAtomicWriteNullPtr(ppv) \
2561	do \
2562	{ \
2563	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2564	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2565	ASMAtomicWritePtrVoid((void RT_FAR * volatile RT_FAR *)(ppv), NULL); \
2566	} while (0)
2567	#endif
2568
2569
2570	/**
2571	* Atomically writes a pointer value, unordered.
2572	*
2573	* @returns Current *pv value
2574	* @param ppv Pointer to the pointer variable.
2575	* @param pv The pointer value to assign to *ppv. If NULL use
2576	* ASMAtomicUoWriteNullPtr or you'll land in trouble.
2577	*
2578	* @remarks This is relatively type safe on GCC platforms when @a pv isn't
2579	* NULL.
2580	*/
2581	#if RT_GNUC_PREREQ(4, 2)
2582	# define ASMAtomicUoWritePtr(ppv, pv) \
2583	do \
2584	{ \
2585	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
2586	__typeof__(*(ppv)) const pvTypeChecked = (pv); \
2587	\
2588	AssertCompile(sizeof(ppv) == sizeof(void )); \
2589	AssertCompile(sizeof(pv) == sizeof(void *)); \
2590	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2591	\
2592	*(ppvTypeChecked) = pvTypeChecked; \
2593	} while (0)
2594	#else
2595	# define ASMAtomicUoWritePtr(ppv, pv) \
2596	do \
2597	{ \
2598	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2599	AssertCompile(sizeof(pv) == sizeof(void RT_FAR *)); \
2600	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2601	*(ppv) = pv; \
2602	} while (0)
2603	#endif
2604
2605
2606	/**
2607	* Atomically sets a pointer to NULL, unordered.
2608	*
2609	* @param ppv Pointer to the pointer variable that should be set to NULL.
2610	*
2611	* @remarks This is relatively type safe on GCC platforms.
2612	*/
2613	#ifdef __GNUC__
2614	# define ASMAtomicUoWriteNullPtr(ppv) \
2615	do \
2616	{ \
2617	__typeof__((ppv)) volatile const ppvTypeChecked = (ppv); \
2618	AssertCompile(sizeof(ppv) == sizeof(void )); \
2619	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2620	*(ppvTypeChecked) = NULL; \
2621	} while (0)
2622	#else
2623	# define ASMAtomicUoWriteNullPtr(ppv) \
2624	do \
2625	{ \
2626	AssertCompile(sizeof(ppv) == sizeof(void RT_FAR )); \
2627	Assert(!( (uintptr_t)ppv & ((ARCH_BITS / 8) - 1) )); \
2628	*(ppv) = NULL; \
2629	} while (0)
2630	#endif
2631
2632
2633	/**
2634	* Atomically write a typical IPRT handle value, ordered.
2635	*
2636	* @param ph Pointer to the variable to update.
2637	* @param hNew The value to assign to *ph.
2638	*
2639	* @remarks This doesn't currently work for all handles (like RTFILE).
2640	*/
2641	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
2642	# define ASMAtomicWriteHandle(ph, hNew) \
2643	do { \
2644	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
2645	ASMAtomicWriteU32((uint32_t volatile RT_FAR *)(ph), (const uint32_t)(hNew)); \
2646	} while (0)
2647	#elif HC_ARCH_BITS == 64
2648	# define ASMAtomicWriteHandle(ph, hNew) \
2649	do { \
2650	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
2651	ASMAtomicWriteU64((uint64_t volatile RT_FAR *)(ph), (const uint64_t)(hNew)); \
2652	} while (0)
2653	#else
2654	# error HC_ARCH_BITS
2655	#endif
2656
2657
2658	/**
2659	* Atomically write a typical IPRT handle value, unordered.
2660	*
2661	* @param ph Pointer to the variable to update.
2662	* @param hNew The value to assign to *ph.
2663	*
2664	* @remarks This doesn't currently work for all handles (like RTFILE).
2665	*/
2666	#if HC_ARCH_BITS == 32 \|\| ARCH_BITS == 16
2667	# define ASMAtomicUoWriteHandle(ph, hNew) \
2668	do { \
2669	AssertCompile(sizeof(*(ph)) == sizeof(uint32_t)); \
2670	ASMAtomicUoWriteU32((uint32_t volatile RT_FAR *)(ph), (const uint32_t)hNew); \
2671	} while (0)
2672	#elif HC_ARCH_BITS == 64
2673	# define ASMAtomicUoWriteHandle(ph, hNew) \
2674	do { \
2675	AssertCompile(sizeof(*(ph)) == sizeof(uint64_t)); \
2676	ASMAtomicUoWriteU64((uint64_t volatile RT_FAR *)(ph), (const uint64_t)hNew); \
2677	} while (0)
2678	#else
2679	# error HC_ARCH_BITS
2680	#endif
2681
2682
2683	/**
2684	* Atomically write a value which size might differ
2685	* between platforms or compilers, ordered.
2686	*
2687	* @param pu Pointer to the variable to update.
2688	* @param uNew The value to assign to *pu.
2689	*/
2690	#define ASMAtomicWriteSize(pu, uNew) \
2691	do { \
2692	switch (sizeof(*(pu))) { \
2693	case 1: ASMAtomicWriteU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu), (uint8_t )(uNew)); break; \
2694	case 2: ASMAtomicWriteU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu), (uint16_t)(uNew)); break; \
2695	case 4: ASMAtomicWriteU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
2696	case 8: ASMAtomicWriteU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
2697	default: AssertMsgFailed(("ASMAtomicWriteSize: size %d is not supported\n", sizeof(*(pu)))); \
2698	} \
2699	} while (0)
2700
2701	/**
2702	* Atomically write a value which size might differ
2703	* between platforms or compilers, unordered.
2704	*
2705	* @param pu Pointer to the variable to update.
2706	* @param uNew The value to assign to *pu.
2707	*/
2708	#define ASMAtomicUoWriteSize(pu, uNew) \
2709	do { \
2710	switch (sizeof(*(pu))) { \
2711	case 1: ASMAtomicUoWriteU8( (volatile uint8_t RT_FAR )(void RT_FAR )(pu), (uint8_t )(uNew)); break; \
2712	case 2: ASMAtomicUoWriteU16((volatile uint16_t RT_FAR )(void RT_FAR )(pu), (uint16_t)(uNew)); break; \
2713	case 4: ASMAtomicUoWriteU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
2714	case 8: ASMAtomicUoWriteU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
2715	default: AssertMsgFailed(("ASMAtomicWriteSize: size %d is not supported\n", sizeof(*(pu)))); \
2716	} \
2717	} while (0)
2718
2719
2720
2721	/**
2722	* Atomically exchanges and adds to a 16-bit value, ordered.
2723	*
2724	* @returns The old value.
2725	* @param pu16 Pointer to the value.
2726	* @param u16 Number to add.
2727	*
2728	* @remarks Currently not implemented, just to make 16-bit code happy.
2729	* @remarks x86: Requires a 486 or later.
2730	*/
2731	RT_ASM_DECL_PRAGMA_WATCOM(uint16_t) ASMAtomicAddU16(uint16_t volatile RT_FAR *pu16, uint32_t u16);
2732
2733
2734	/**
2735	* Atomically exchanges and adds to a 32-bit value, ordered.
2736	*
2737	* @returns The old value.
2738	* @param pu32 Pointer to the value.
2739	* @param u32 Number to add.
2740	*
2741	* @remarks x86: Requires a 486 or later.
2742	*/
2743	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
2744	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicAddU32(uint32_t volatile RT_FAR *pu32, uint32_t u32);
2745	#else
2746	DECLINLINE(uint32_t) ASMAtomicAddU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
2747	{
2748	# if RT_INLINE_ASM_USES_INTRIN
2749	u32 = _InterlockedExchangeAdd((long RT_FAR *)pu32, u32);
2750	return u32;
2751
2752	# elif RT_INLINE_ASM_GNU_STYLE
2753	__asm__ __volatile__("lock; xaddl %0, %1\n\t"
2754	: "=r" (u32),
2755	"=m" (*pu32)
2756	: "0" (u32),
2757	"m" (*pu32)
2758	: "memory");
2759	return u32;
2760	# else
2761	__asm
2762	{
2763	mov eax, [u32]
2764	# ifdef RT_ARCH_AMD64
2765	mov rdx, [pu32]
2766	lock xadd [rdx], eax
2767	# else
2768	mov edx, [pu32]
2769	lock xadd [edx], eax
2770	# endif
2771	mov [u32], eax
2772	}
2773	return u32;
2774	# endif
2775	}
2776	#endif
2777
2778
2779	/**
2780	* Atomically exchanges and adds to a signed 32-bit value, ordered.
2781	*
2782	* @returns The old value.
2783	* @param pi32 Pointer to the value.
2784	* @param i32 Number to add.
2785	*
2786	* @remarks x86: Requires a 486 or later.
2787	*/
2788	DECLINLINE(int32_t) ASMAtomicAddS32(int32_t volatile RT_FAR *pi32, int32_t i32)
2789	{
2790	return (int32_t)ASMAtomicAddU32((uint32_t volatile RT_FAR *)pi32, (uint32_t)i32);
2791	}
2792
2793
2794	/**
2795	* Atomically exchanges and adds to a 64-bit value, ordered.
2796	*
2797	* @returns The old value.
2798	* @param pu64 Pointer to the value.
2799	* @param u64 Number to add.
2800	*
2801	* @remarks x86: Requires a Pentium or later.
2802	*/
2803	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
2804	DECLASM(uint64_t) ASMAtomicAddU64(uint64_t volatile RT_FAR *pu64, uint64_t u64);
2805	#else
2806	DECLINLINE(uint64_t) ASMAtomicAddU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
2807	{
2808	# if RT_INLINE_ASM_USES_INTRIN && defined(RT_ARCH_AMD64)
2809	u64 = _InterlockedExchangeAdd64((__int64 RT_FAR *)pu64, u64);
2810	return u64;
2811
2812	# elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
2813	__asm__ __volatile__("lock; xaddq %0, %1\n\t"
2814	: "=r" (u64),
2815	"=m" (*pu64)
2816	: "0" (u64),
2817	"m" (*pu64)
2818	: "memory");
2819	return u64;
2820	# else
2821	uint64_t u64Old;
2822	for (;;)
2823	{
2824	uint64_t u64New;
2825	u64Old = ASMAtomicUoReadU64(pu64);
2826	u64New = u64Old + u64;
2827	if (ASMAtomicCmpXchgU64(pu64, u64New, u64Old))
2828	break;
2829	ASMNopPause();
2830	}
2831	return u64Old;
2832	# endif
2833	}
2834	#endif
2835
2836
2837	/**
2838	* Atomically exchanges and adds to a signed 64-bit value, ordered.
2839	*
2840	* @returns The old value.
2841	* @param pi64 Pointer to the value.
2842	* @param i64 Number to add.
2843	*
2844	* @remarks x86: Requires a Pentium or later.
2845	*/
2846	DECLINLINE(int64_t) ASMAtomicAddS64(int64_t volatile RT_FAR *pi64, int64_t i64)
2847	{
2848	return (int64_t)ASMAtomicAddU64((uint64_t volatile RT_FAR *)pi64, (uint64_t)i64);
2849	}
2850
2851
2852	/**
2853	* Atomically exchanges and adds to a size_t value, ordered.
2854	*
2855	* @returns The old value.
2856	* @param pcb Pointer to the size_t value.
2857	* @param cb Number to add.
2858	*/
2859	DECLINLINE(size_t) ASMAtomicAddZ(size_t volatile RT_FAR *pcb, size_t cb)
2860	{
2861	#if ARCH_BITS == 64
2862	AssertCompileSize(size_t, 8);
2863	return ASMAtomicAddU64((uint64_t volatile RT_FAR *)pcb, cb);
2864	#elif ARCH_BITS == 32
2865	AssertCompileSize(size_t, 4);
2866	return ASMAtomicAddU32((uint32_t volatile RT_FAR *)pcb, cb);
2867	#elif ARCH_BITS == 16
2868	AssertCompileSize(size_t, 2);
2869	return ASMAtomicAddU16((uint16_t volatile RT_FAR *)pcb, cb);
2870	#else
2871	# error "Unsupported ARCH_BITS value"
2872	#endif
2873	}
2874
2875
2876	/**
2877	* Atomically exchanges and adds a value which size might differ between
2878	* platforms or compilers, ordered.
2879	*
2880	* @param pu Pointer to the variable to update.
2881	* @param uNew The value to add to *pu.
2882	* @param puOld Where to store the old value.
2883	*/
2884	#define ASMAtomicAddSize(pu, uNew, puOld) \
2885	do { \
2886	switch (sizeof(*(pu))) { \
2887	case 4: (uint32_t )(puOld) = ASMAtomicAddU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
2888	case 8: (uint64_t )(puOld) = ASMAtomicAddU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
2889	default: AssertMsgFailed(("ASMAtomicAddSize: size %d is not supported\n", sizeof(*(pu)))); \
2890	} \
2891	} while (0)
2892
2893
2894
2895	/**
2896	* Atomically exchanges and subtracts to an unsigned 16-bit value, ordered.
2897	*
2898	* @returns The old value.
2899	* @param pu16 Pointer to the value.
2900	* @param u16 Number to subtract.
2901	*
2902	* @remarks x86: Requires a 486 or later.
2903	*/
2904	DECLINLINE(uint16_t) ASMAtomicSubU16(uint16_t volatile RT_FAR *pu16, uint32_t u16)
2905	{
2906	return ASMAtomicAddU16(pu16, (uint16_t)-(int16_t)u16);
2907	}
2908
2909
2910	/**
2911	* Atomically exchanges and subtracts to a signed 16-bit value, ordered.
2912	*
2913	* @returns The old value.
2914	* @param pi16 Pointer to the value.
2915	* @param i16 Number to subtract.
2916	*
2917	* @remarks x86: Requires a 486 or later.
2918	*/
2919	DECLINLINE(int16_t) ASMAtomicSubS16(int16_t volatile RT_FAR *pi16, int16_t i16)
2920	{
2921	return (int16_t)ASMAtomicAddU16((uint16_t volatile RT_FAR *)pi16, (uint16_t)-i16);
2922	}
2923
2924
2925	/**
2926	* Atomically exchanges and subtracts to an unsigned 32-bit value, ordered.
2927	*
2928	* @returns The old value.
2929	* @param pu32 Pointer to the value.
2930	* @param u32 Number to subtract.
2931	*
2932	* @remarks x86: Requires a 486 or later.
2933	*/
2934	DECLINLINE(uint32_t) ASMAtomicSubU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
2935	{
2936	return ASMAtomicAddU32(pu32, (uint32_t)-(int32_t)u32);
2937	}
2938
2939
2940	/**
2941	* Atomically exchanges and subtracts to a signed 32-bit value, ordered.
2942	*
2943	* @returns The old value.
2944	* @param pi32 Pointer to the value.
2945	* @param i32 Number to subtract.
2946	*
2947	* @remarks x86: Requires a 486 or later.
2948	*/
2949	DECLINLINE(int32_t) ASMAtomicSubS32(int32_t volatile RT_FAR *pi32, int32_t i32)
2950	{
2951	return (int32_t)ASMAtomicAddU32((uint32_t volatile RT_FAR *)pi32, (uint32_t)-i32);
2952	}
2953
2954
2955	/**
2956	* Atomically exchanges and subtracts to an unsigned 64-bit value, ordered.
2957	*
2958	* @returns The old value.
2959	* @param pu64 Pointer to the value.
2960	* @param u64 Number to subtract.
2961	*
2962	* @remarks x86: Requires a Pentium or later.
2963	*/
2964	DECLINLINE(uint64_t) ASMAtomicSubU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
2965	{
2966	return ASMAtomicAddU64(pu64, (uint64_t)-(int64_t)u64);
2967	}
2968
2969
2970	/**
2971	* Atomically exchanges and subtracts to a signed 64-bit value, ordered.
2972	*
2973	* @returns The old value.
2974	* @param pi64 Pointer to the value.
2975	* @param i64 Number to subtract.
2976	*
2977	* @remarks x86: Requires a Pentium or later.
2978	*/
2979	DECLINLINE(int64_t) ASMAtomicSubS64(int64_t volatile RT_FAR *pi64, int64_t i64)
2980	{
2981	return (int64_t)ASMAtomicAddU64((uint64_t volatile RT_FAR *)pi64, (uint64_t)-i64);
2982	}
2983
2984
2985	/**
2986	* Atomically exchanges and subtracts to a size_t value, ordered.
2987	*
2988	* @returns The old value.
2989	* @param pcb Pointer to the size_t value.
2990	* @param cb Number to subtract.
2991	*
2992	* @remarks x86: Requires a 486 or later.
2993	*/
2994	DECLINLINE(size_t) ASMAtomicSubZ(size_t volatile RT_FAR *pcb, size_t cb)
2995	{
2996	#if ARCH_BITS == 64
2997	return ASMAtomicSubU64((uint64_t volatile RT_FAR *)pcb, cb);
2998	#elif ARCH_BITS == 32
2999	return ASMAtomicSubU32((uint32_t volatile RT_FAR *)pcb, cb);
3000	#elif ARCH_BITS == 16
3001	AssertCompileSize(size_t, 2);
3002	return ASMAtomicSubU16((uint16_t volatile RT_FAR *)pcb, cb);
3003	#else
3004	# error "Unsupported ARCH_BITS value"
3005	#endif
3006	}
3007
3008
3009	/**
3010	* Atomically exchanges and subtracts a value which size might differ between
3011	* platforms or compilers, ordered.
3012	*
3013	* @param pu Pointer to the variable to update.
3014	* @param uNew The value to subtract to *pu.
3015	* @param puOld Where to store the old value.
3016	*
3017	* @remarks x86: Requires a 486 or later.
3018	*/
3019	#define ASMAtomicSubSize(pu, uNew, puOld) \
3020	do { \
3021	switch (sizeof(*(pu))) { \
3022	case 4: (uint32_t RT_FAR )(puOld) = ASMAtomicSubU32((volatile uint32_t RT_FAR )(void RT_FAR )(pu), (uint32_t)(uNew)); break; \
3023	case 8: (uint64_t RT_FAR )(puOld) = ASMAtomicSubU64((volatile uint64_t RT_FAR )(void RT_FAR )(pu), (uint64_t)(uNew)); break; \
3024	default: AssertMsgFailed(("ASMAtomicSubSize: size %d is not supported\n", sizeof(*(pu)))); \
3025	} \
3026	} while (0)
3027
3028
3029
3030	/**
3031	* Atomically increment a 16-bit value, ordered.
3032	*
3033	* @returns The new value.
3034	* @param pu16 Pointer to the value to increment.
3035	* @remarks Not implemented. Just to make 16-bit code happy.
3036	*
3037	* @remarks x86: Requires a 486 or later.
3038	*/
3039	RT_ASM_DECL_PRAGMA_WATCOM(uint16_t) ASMAtomicIncU16(uint16_t volatile RT_FAR *pu16);
3040
3041
3042	/**
3043	* Atomically increment a 32-bit value, ordered.
3044	*
3045	* @returns The new value.
3046	* @param pu32 Pointer to the value to increment.
3047	*
3048	* @remarks x86: Requires a 486 or later.
3049	*/
3050	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3051	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicIncU32(uint32_t volatile RT_FAR *pu32);
3052	#else
3053	DECLINLINE(uint32_t) ASMAtomicIncU32(uint32_t volatile RT_FAR *pu32)
3054	{
3055	uint32_t u32;
3056	# if RT_INLINE_ASM_USES_INTRIN
3057	u32 = _InterlockedIncrement((long RT_FAR *)pu32);
3058	return u32;
3059
3060	# elif RT_INLINE_ASM_GNU_STYLE
3061	__asm__ __volatile__("lock; xaddl %0, %1\n\t"
3062	: "=r" (u32),
3063	"=m" (*pu32)
3064	: "0" (1),
3065	"m" (*pu32)
3066	: "memory");
3067	return u32+1;
3068	# else
3069	__asm
3070	{
3071	mov eax, 1
3072	# ifdef RT_ARCH_AMD64
3073	mov rdx, [pu32]
3074	lock xadd [rdx], eax
3075	# else
3076	mov edx, [pu32]
3077	lock xadd [edx], eax
3078	# endif
3079	mov u32, eax
3080	}
3081	return u32+1;
3082	# endif
3083	}
3084	#endif
3085
3086
3087	/**
3088	* Atomically increment a signed 32-bit value, ordered.
3089	*
3090	* @returns The new value.
3091	* @param pi32 Pointer to the value to increment.
3092	*
3093	* @remarks x86: Requires a 486 or later.
3094	*/
3095	DECLINLINE(int32_t) ASMAtomicIncS32(int32_t volatile RT_FAR *pi32)
3096	{
3097	return (int32_t)ASMAtomicIncU32((uint32_t volatile RT_FAR *)pi32);
3098	}
3099
3100
3101	/**
3102	* Atomically increment a 64-bit value, ordered.
3103	*
3104	* @returns The new value.
3105	* @param pu64 Pointer to the value to increment.
3106	*
3107	* @remarks x86: Requires a Pentium or later.
3108	*/
3109	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3110	DECLASM(uint64_t) ASMAtomicIncU64(uint64_t volatile RT_FAR *pu64);
3111	#else
3112	DECLINLINE(uint64_t) ASMAtomicIncU64(uint64_t volatile RT_FAR *pu64)
3113	{
3114	# if RT_INLINE_ASM_USES_INTRIN && defined(RT_ARCH_AMD64)
3115	uint64_t u64;
3116	u64 = _InterlockedIncrement64((__int64 RT_FAR *)pu64);
3117	return u64;
3118
3119	# elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3120	uint64_t u64;
3121	__asm__ __volatile__("lock; xaddq %0, %1\n\t"
3122	: "=r" (u64),
3123	"=m" (*pu64)
3124	: "0" (1),
3125	"m" (*pu64)
3126	: "memory");
3127	return u64 + 1;
3128	# else
3129	return ASMAtomicAddU64(pu64, 1) + 1;
3130	# endif
3131	}
3132	#endif
3133
3134
3135	/**
3136	* Atomically increment a signed 64-bit value, ordered.
3137	*
3138	* @returns The new value.
3139	* @param pi64 Pointer to the value to increment.
3140	*
3141	* @remarks x86: Requires a Pentium or later.
3142	*/
3143	DECLINLINE(int64_t) ASMAtomicIncS64(int64_t volatile RT_FAR *pi64)
3144	{
3145	return (int64_t)ASMAtomicIncU64((uint64_t volatile RT_FAR *)pi64);
3146	}
3147
3148
3149	/**
3150	* Atomically increment a size_t value, ordered.
3151	*
3152	* @returns The new value.
3153	* @param pcb Pointer to the value to increment.
3154	*
3155	* @remarks x86: Requires a 486 or later.
3156	*/
3157	DECLINLINE(int64_t) ASMAtomicIncZ(size_t volatile RT_FAR *pcb)
3158	{
3159	#if ARCH_BITS == 64
3160	return ASMAtomicIncU64((uint64_t volatile RT_FAR *)pcb);
3161	#elif ARCH_BITS == 32
3162	return ASMAtomicIncU32((uint32_t volatile RT_FAR *)pcb);
3163	#elif ARCH_BITS == 16
3164	return ASMAtomicIncU16((uint16_t volatile RT_FAR *)pcb);
3165	#else
3166	# error "Unsupported ARCH_BITS value"
3167	#endif
3168	}
3169
3170
3171
3172	/**
3173	* Atomically decrement an unsigned 32-bit value, ordered.
3174	*
3175	* @returns The new value.
3176	* @param pu16 Pointer to the value to decrement.
3177	* @remarks Not implemented. Just to make 16-bit code happy.
3178	*
3179	* @remarks x86: Requires a 486 or later.
3180	*/
3181	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicDecU16(uint16_t volatile RT_FAR *pu16);
3182
3183
3184	/**
3185	* Atomically decrement an unsigned 32-bit value, ordered.
3186	*
3187	* @returns The new value.
3188	* @param pu32 Pointer to the value to decrement.
3189	*
3190	* @remarks x86: Requires a 486 or later.
3191	*/
3192	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3193	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicDecU32(uint32_t volatile RT_FAR *pu32);
3194	#else
3195	DECLINLINE(uint32_t) ASMAtomicDecU32(uint32_t volatile RT_FAR *pu32)
3196	{
3197	uint32_t u32;
3198	# if RT_INLINE_ASM_USES_INTRIN
3199	u32 = _InterlockedDecrement((long RT_FAR *)pu32);
3200	return u32;
3201
3202	# elif RT_INLINE_ASM_GNU_STYLE
3203	__asm__ __volatile__("lock; xaddl %0, %1\n\t"
3204	: "=r" (u32),
3205	"=m" (*pu32)
3206	: "0" (-1),
3207	"m" (*pu32)
3208	: "memory");
3209	return u32-1;
3210	# else
3211	__asm
3212	{
3213	mov eax, -1
3214	# ifdef RT_ARCH_AMD64
3215	mov rdx, [pu32]
3216	lock xadd [rdx], eax
3217	# else
3218	mov edx, [pu32]
3219	lock xadd [edx], eax
3220	# endif
3221	mov u32, eax
3222	}
3223	return u32-1;
3224	# endif
3225	}
3226	#endif
3227
3228
3229	/**
3230	* Atomically decrement a signed 32-bit value, ordered.
3231	*
3232	* @returns The new value.
3233	* @param pi32 Pointer to the value to decrement.
3234	*
3235	* @remarks x86: Requires a 486 or later.
3236	*/
3237	DECLINLINE(int32_t) ASMAtomicDecS32(int32_t volatile RT_FAR *pi32)
3238	{
3239	return (int32_t)ASMAtomicDecU32((uint32_t volatile RT_FAR *)pi32);
3240	}
3241
3242
3243	/**
3244	* Atomically decrement an unsigned 64-bit value, ordered.
3245	*
3246	* @returns The new value.
3247	* @param pu64 Pointer to the value to decrement.
3248	*
3249	* @remarks x86: Requires a Pentium or later.
3250	*/
3251	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3252	RT_ASM_DECL_PRAGMA_WATCOM(uint64_t) ASMAtomicDecU64(uint64_t volatile RT_FAR *pu64);
3253	#else
3254	DECLINLINE(uint64_t) ASMAtomicDecU64(uint64_t volatile RT_FAR *pu64)
3255	{
3256	# if RT_INLINE_ASM_USES_INTRIN && defined(RT_ARCH_AMD64)
3257	uint64_t u64 = _InterlockedDecrement64((__int64 volatile RT_FAR *)pu64);
3258	return u64;
3259
3260	# elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3261	uint64_t u64;
3262	__asm__ __volatile__("lock; xaddq %q0, %1\n\t"
3263	: "=r" (u64),
3264	"=m" (*pu64)
3265	: "0" (~(uint64_t)0),
3266	"m" (*pu64)
3267	: "memory");
3268	return u64-1;
3269	# else
3270	return ASMAtomicAddU64(pu64, UINT64_MAX) - 1;
3271	# endif
3272	}
3273	#endif
3274
3275
3276	/**
3277	* Atomically decrement a signed 64-bit value, ordered.
3278	*
3279	* @returns The new value.
3280	* @param pi64 Pointer to the value to decrement.
3281	*
3282	* @remarks x86: Requires a Pentium or later.
3283	*/
3284	DECLINLINE(int64_t) ASMAtomicDecS64(int64_t volatile RT_FAR *pi64)
3285	{
3286	return (int64_t)ASMAtomicDecU64((uint64_t volatile RT_FAR *)pi64);
3287	}
3288
3289
3290	/**
3291	* Atomically decrement a size_t value, ordered.
3292	*
3293	* @returns The new value.
3294	* @param pcb Pointer to the value to decrement.
3295	*
3296	* @remarks x86: Requires a 486 or later.
3297	*/
3298	DECLINLINE(int64_t) ASMAtomicDecZ(size_t volatile RT_FAR *pcb)
3299	{
3300	#if ARCH_BITS == 64
3301	return ASMAtomicDecU64((uint64_t volatile RT_FAR *)pcb);
3302	#elif ARCH_BITS == 32
3303	return ASMAtomicDecU32((uint32_t volatile RT_FAR *)pcb);
3304	#elif ARCH_BITS == 16
3305	return ASMAtomicDecU16((uint16_t volatile RT_FAR *)pcb);
3306	#else
3307	# error "Unsupported ARCH_BITS value"
3308	#endif
3309	}
3310
3311
3312	/**
3313	* Atomically Or an unsigned 32-bit value, ordered.
3314	*
3315	* @param pu32 Pointer to the pointer variable to OR u32 with.
3316	* @param u32 The value to OR *pu32 with.
3317	*
3318	* @remarks x86: Requires a 386 or later.
3319	*/
3320	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3321	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicOrU32(uint32_t volatile RT_FAR *pu32, uint32_t u32);
3322	#else
3323	DECLINLINE(void) ASMAtomicOrU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
3324	{
3325	# if RT_INLINE_ASM_USES_INTRIN
3326	_InterlockedOr((long volatile RT_FAR *)pu32, (long)u32);
3327
3328	# elif RT_INLINE_ASM_GNU_STYLE
3329	__asm__ __volatile__("lock; orl %1, %0\n\t"
3330	: "=m" (*pu32)
3331	: "ir" (u32),
3332	"m" (*pu32));
3333	# else
3334	__asm
3335	{
3336	mov eax, [u32]
3337	# ifdef RT_ARCH_AMD64
3338	mov rdx, [pu32]
3339	lock or [rdx], eax
3340	# else
3341	mov edx, [pu32]
3342	lock or [edx], eax
3343	# endif
3344	}
3345	# endif
3346	}
3347	#endif
3348
3349
3350	/**
3351	* Atomically Or a signed 32-bit value, ordered.
3352	*
3353	* @param pi32 Pointer to the pointer variable to OR u32 with.
3354	* @param i32 The value to OR *pu32 with.
3355	*
3356	* @remarks x86: Requires a 386 or later.
3357	*/
3358	DECLINLINE(void) ASMAtomicOrS32(int32_t volatile RT_FAR *pi32, int32_t i32)
3359	{
3360	ASMAtomicOrU32((uint32_t volatile RT_FAR *)pi32, i32);
3361	}
3362
3363
3364	/**
3365	* Atomically Or an unsigned 64-bit value, ordered.
3366	*
3367	* @param pu64 Pointer to the pointer variable to OR u64 with.
3368	* @param u64 The value to OR *pu64 with.
3369	*
3370	* @remarks x86: Requires a Pentium or later.
3371	*/
3372	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3373	DECLASM(void) ASMAtomicOrU64(uint64_t volatile RT_FAR *pu64, uint64_t u64);
3374	#else
3375	DECLINLINE(void) ASMAtomicOrU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
3376	{
3377	# if RT_INLINE_ASM_USES_INTRIN && defined(RT_ARCH_AMD64)
3378	_InterlockedOr64((__int64 volatile RT_FAR *)pu64, (__int64)u64);
3379
3380	# elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3381	__asm__ __volatile__("lock; orq %1, %q0\n\t"
3382	: "=m" (*pu64)
3383	: "r" (u64),
3384	"m" (*pu64));
3385	# else
3386	for (;;)
3387	{
3388	uint64_t u64Old = ASMAtomicUoReadU64(pu64);
3389	uint64_t u64New = u64Old \| u64;
3390	if (ASMAtomicCmpXchgU64(pu64, u64New, u64Old))
3391	break;
3392	ASMNopPause();
3393	}
3394	# endif
3395	}
3396	#endif
3397
3398
3399	/**
3400	* Atomically Or a signed 64-bit value, ordered.
3401	*
3402	* @param pi64 Pointer to the pointer variable to OR u64 with.
3403	* @param i64 The value to OR *pu64 with.
3404	*
3405	* @remarks x86: Requires a Pentium or later.
3406	*/
3407	DECLINLINE(void) ASMAtomicOrS64(int64_t volatile RT_FAR *pi64, int64_t i64)
3408	{
3409	ASMAtomicOrU64((uint64_t volatile RT_FAR *)pi64, i64);
3410	}
3411
3412
3413	/**
3414	* Atomically And an unsigned 32-bit value, ordered.
3415	*
3416	* @param pu32 Pointer to the pointer variable to AND u32 with.
3417	* @param u32 The value to AND *pu32 with.
3418	*
3419	* @remarks x86: Requires a 386 or later.
3420	*/
3421	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3422	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicAndU32(uint32_t volatile RT_FAR *pu32, uint32_t u32);
3423	#else
3424	DECLINLINE(void) ASMAtomicAndU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
3425	{
3426	# if RT_INLINE_ASM_USES_INTRIN
3427	_InterlockedAnd((long volatile RT_FAR *)pu32, u32);
3428
3429	# elif RT_INLINE_ASM_GNU_STYLE
3430	__asm__ __volatile__("lock; andl %1, %0\n\t"
3431	: "=m" (*pu32)
3432	: "ir" (u32),
3433	"m" (*pu32));
3434	# else
3435	__asm
3436	{
3437	mov eax, [u32]
3438	# ifdef RT_ARCH_AMD64
3439	mov rdx, [pu32]
3440	lock and [rdx], eax
3441	# else
3442	mov edx, [pu32]
3443	lock and [edx], eax
3444	# endif
3445	}
3446	# endif
3447	}
3448	#endif
3449
3450
3451	/**
3452	* Atomically And a signed 32-bit value, ordered.
3453	*
3454	* @param pi32 Pointer to the pointer variable to AND i32 with.
3455	* @param i32 The value to AND *pi32 with.
3456	*
3457	* @remarks x86: Requires a 386 or later.
3458	*/
3459	DECLINLINE(void) ASMAtomicAndS32(int32_t volatile RT_FAR *pi32, int32_t i32)
3460	{
3461	ASMAtomicAndU32((uint32_t volatile RT_FAR *)pi32, (uint32_t)i32);
3462	}
3463
3464
3465	/**
3466	* Atomically And an unsigned 64-bit value, ordered.
3467	*
3468	* @param pu64 Pointer to the pointer variable to AND u64 with.
3469	* @param u64 The value to AND *pu64 with.
3470	*
3471	* @remarks x86: Requires a Pentium or later.
3472	*/
3473	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3474	DECLASM(void) ASMAtomicAndU64(uint64_t volatile RT_FAR *pu64, uint64_t u64);
3475	#else
3476	DECLINLINE(void) ASMAtomicAndU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
3477	{
3478	# if RT_INLINE_ASM_USES_INTRIN && defined(RT_ARCH_AMD64)
3479	_InterlockedAnd64((__int64 volatile RT_FAR *)pu64, u64);
3480
3481	# elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3482	__asm__ __volatile__("lock; andq %1, %0\n\t"
3483	: "=m" (*pu64)
3484	: "r" (u64),
3485	"m" (*pu64));
3486	# else
3487	for (;;)
3488	{
3489	uint64_t u64Old = ASMAtomicUoReadU64(pu64);
3490	uint64_t u64New = u64Old & u64;
3491	if (ASMAtomicCmpXchgU64(pu64, u64New, u64Old))
3492	break;
3493	ASMNopPause();
3494	}
3495	# endif
3496	}
3497	#endif
3498
3499
3500	/**
3501	* Atomically And a signed 64-bit value, ordered.
3502	*
3503	* @param pi64 Pointer to the pointer variable to AND i64 with.
3504	* @param i64 The value to AND *pi64 with.
3505	*
3506	* @remarks x86: Requires a Pentium or later.
3507	*/
3508	DECLINLINE(void) ASMAtomicAndS64(int64_t volatile RT_FAR *pi64, int64_t i64)
3509	{
3510	ASMAtomicAndU64((uint64_t volatile RT_FAR *)pi64, (uint64_t)i64);
3511	}
3512
3513
3514	/**
3515	* Atomically OR an unsigned 32-bit value, unordered but interrupt safe.
3516	*
3517	* @param pu32 Pointer to the pointer variable to OR u32 with.
3518	* @param u32 The value to OR *pu32 with.
3519	*
3520	* @remarks x86: Requires a 386 or later.
3521	*/
3522	#if RT_INLINE_ASM_EXTERNAL
3523	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicUoOrU32(uint32_t volatile RT_FAR *pu32, uint32_t u32);
3524	#else
3525	DECLINLINE(void) ASMAtomicUoOrU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
3526	{
3527	# if RT_INLINE_ASM_GNU_STYLE
3528	__asm__ __volatile__("orl %1, %0\n\t"
3529	: "=m" (*pu32)
3530	: "ir" (u32),
3531	"m" (*pu32));
3532	# else
3533	__asm
3534	{
3535	mov eax, [u32]
3536	# ifdef RT_ARCH_AMD64
3537	mov rdx, [pu32]
3538	or [rdx], eax
3539	# else
3540	mov edx, [pu32]
3541	or [edx], eax
3542	# endif
3543	}
3544	# endif
3545	}
3546	#endif
3547
3548
3549	/**
3550	* Atomically OR a signed 32-bit value, unordered.
3551	*
3552	* @param pi32 Pointer to the pointer variable to OR u32 with.
3553	* @param i32 The value to OR *pu32 with.
3554	*
3555	* @remarks x86: Requires a 386 or later.
3556	*/
3557	DECLINLINE(void) ASMAtomicUoOrS32(int32_t volatile RT_FAR *pi32, int32_t i32)
3558	{
3559	ASMAtomicUoOrU32((uint32_t volatile RT_FAR *)pi32, i32);
3560	}
3561
3562
3563	/**
3564	* Atomically OR an unsigned 64-bit value, unordered.
3565	*
3566	* @param pu64 Pointer to the pointer variable to OR u64 with.
3567	* @param u64 The value to OR *pu64 with.
3568	*
3569	* @remarks x86: Requires a Pentium or later.
3570	*/
3571	#if RT_INLINE_ASM_EXTERNAL
3572	DECLASM(void) ASMAtomicUoOrU64(uint64_t volatile RT_FAR *pu64, uint64_t u64);
3573	#else
3574	DECLINLINE(void) ASMAtomicUoOrU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
3575	{
3576	# if RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3577	__asm__ __volatile__("orq %1, %q0\n\t"
3578	: "=m" (*pu64)
3579	: "r" (u64),
3580	"m" (*pu64));
3581	# else
3582	for (;;)
3583	{
3584	uint64_t u64Old = ASMAtomicUoReadU64(pu64);
3585	uint64_t u64New = u64Old \| u64;
3586	if (ASMAtomicCmpXchgU64(pu64, u64New, u64Old))
3587	break;
3588	ASMNopPause();
3589	}
3590	# endif
3591	}
3592	#endif
3593
3594
3595	/**
3596	* Atomically Or a signed 64-bit value, unordered.
3597	*
3598	* @param pi64 Pointer to the pointer variable to OR u64 with.
3599	* @param i64 The value to OR *pu64 with.
3600	*
3601	* @remarks x86: Requires a Pentium or later.
3602	*/
3603	DECLINLINE(void) ASMAtomicUoOrS64(int64_t volatile RT_FAR *pi64, int64_t i64)
3604	{
3605	ASMAtomicUoOrU64((uint64_t volatile RT_FAR *)pi64, i64);
3606	}
3607
3608
3609	/**
3610	* Atomically And an unsigned 32-bit value, unordered.
3611	*
3612	* @param pu32 Pointer to the pointer variable to AND u32 with.
3613	* @param u32 The value to AND *pu32 with.
3614	*
3615	* @remarks x86: Requires a 386 or later.
3616	*/
3617	#if RT_INLINE_ASM_EXTERNAL
3618	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicUoAndU32(uint32_t volatile RT_FAR *pu32, uint32_t u32);
3619	#else
3620	DECLINLINE(void) ASMAtomicUoAndU32(uint32_t volatile RT_FAR *pu32, uint32_t u32)
3621	{
3622	# if RT_INLINE_ASM_GNU_STYLE
3623	__asm__ __volatile__("andl %1, %0\n\t"
3624	: "=m" (*pu32)
3625	: "ir" (u32),
3626	"m" (*pu32));
3627	# else
3628	__asm
3629	{
3630	mov eax, [u32]
3631	# ifdef RT_ARCH_AMD64
3632	mov rdx, [pu32]
3633	and [rdx], eax
3634	# else
3635	mov edx, [pu32]
3636	and [edx], eax
3637	# endif
3638	}
3639	# endif
3640	}
3641	#endif
3642
3643
3644	/**
3645	* Atomically And a signed 32-bit value, unordered.
3646	*
3647	* @param pi32 Pointer to the pointer variable to AND i32 with.
3648	* @param i32 The value to AND *pi32 with.
3649	*
3650	* @remarks x86: Requires a 386 or later.
3651	*/
3652	DECLINLINE(void) ASMAtomicUoAndS32(int32_t volatile RT_FAR *pi32, int32_t i32)
3653	{
3654	ASMAtomicUoAndU32((uint32_t volatile RT_FAR *)pi32, (uint32_t)i32);
3655	}
3656
3657
3658	/**
3659	* Atomically And an unsigned 64-bit value, unordered.
3660	*
3661	* @param pu64 Pointer to the pointer variable to AND u64 with.
3662	* @param u64 The value to AND *pu64 with.
3663	*
3664	* @remarks x86: Requires a Pentium or later.
3665	*/
3666	#if RT_INLINE_ASM_EXTERNAL
3667	DECLASM(void) ASMAtomicUoAndU64(uint64_t volatile RT_FAR *pu64, uint64_t u64);
3668	#else
3669	DECLINLINE(void) ASMAtomicUoAndU64(uint64_t volatile RT_FAR *pu64, uint64_t u64)
3670	{
3671	# if RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
3672	__asm__ __volatile__("andq %1, %0\n\t"
3673	: "=m" (*pu64)
3674	: "r" (u64),
3675	"m" (*pu64));
3676	# else
3677	for (;;)
3678	{
3679	uint64_t u64Old = ASMAtomicUoReadU64(pu64);
3680	uint64_t u64New = u64Old & u64;
3681	if (ASMAtomicCmpXchgU64(pu64, u64New, u64Old))
3682	break;
3683	ASMNopPause();
3684	}
3685	# endif
3686	}
3687	#endif
3688
3689
3690	/**
3691	* Atomically And a signed 64-bit value, unordered.
3692	*
3693	* @param pi64 Pointer to the pointer variable to AND i64 with.
3694	* @param i64 The value to AND *pi64 with.
3695	*
3696	* @remarks x86: Requires a Pentium or later.
3697	*/
3698	DECLINLINE(void) ASMAtomicUoAndS64(int64_t volatile RT_FAR *pi64, int64_t i64)
3699	{
3700	ASMAtomicUoAndU64((uint64_t volatile RT_FAR *)pi64, (uint64_t)i64);
3701	}
3702
3703
3704	/**
3705	* Atomically increment an unsigned 32-bit value, unordered.
3706	*
3707	* @returns the new value.
3708	* @param pu32 Pointer to the variable to increment.
3709	*
3710	* @remarks x86: Requires a 486 or later.
3711	*/
3712	#if RT_INLINE_ASM_EXTERNAL
3713	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicUoIncU32(uint32_t volatile RT_FAR *pu32);
3714	#else
3715	DECLINLINE(uint32_t) ASMAtomicUoIncU32(uint32_t volatile RT_FAR *pu32)
3716	{
3717	uint32_t u32;
3718	# if RT_INLINE_ASM_GNU_STYLE
3719	__asm__ __volatile__("xaddl %0, %1\n\t"
3720	: "=r" (u32),
3721	"=m" (*pu32)
3722	: "0" (1),
3723	"m" (*pu32)
3724	: "memory");
3725	return u32 + 1;
3726	# else
3727	__asm
3728	{
3729	mov eax, 1
3730	# ifdef RT_ARCH_AMD64
3731	mov rdx, [pu32]
3732	xadd [rdx], eax
3733	# else
3734	mov edx, [pu32]
3735	xadd [edx], eax
3736	# endif
3737	mov u32, eax
3738	}
3739	return u32 + 1;
3740	# endif
3741	}
3742	#endif
3743
3744
3745	/**
3746	* Atomically decrement an unsigned 32-bit value, unordered.
3747	*
3748	* @returns the new value.
3749	* @param pu32 Pointer to the variable to decrement.
3750	*
3751	* @remarks x86: Requires a 486 or later.
3752	*/
3753	#if RT_INLINE_ASM_EXTERNAL
3754	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMAtomicUoDecU32(uint32_t volatile RT_FAR *pu32);
3755	#else
3756	DECLINLINE(uint32_t) ASMAtomicUoDecU32(uint32_t volatile RT_FAR *pu32)
3757	{
3758	uint32_t u32;
3759	# if RT_INLINE_ASM_GNU_STYLE
3760	__asm__ __volatile__("lock; xaddl %0, %1\n\t"
3761	: "=r" (u32),
3762	"=m" (*pu32)
3763	: "0" (-1),
3764	"m" (*pu32)
3765	: "memory");
3766	return u32 - 1;
3767	# else
3768	__asm
3769	{
3770	mov eax, -1
3771	# ifdef RT_ARCH_AMD64
3772	mov rdx, [pu32]
3773	xadd [rdx], eax
3774	# else
3775	mov edx, [pu32]
3776	xadd [edx], eax
3777	# endif
3778	mov u32, eax
3779	}
3780	return u32 - 1;
3781	# endif
3782	}
3783	#endif
3784
3785
3786	/** @def RT_ASM_PAGE_SIZE
3787	* We try avoid dragging in iprt/param.h here.
3788	* @internal
3789	*/
3790	#if defined(RT_ARCH_SPARC64)
3791	# define RT_ASM_PAGE_SIZE 0x2000
3792	# if defined(PAGE_SIZE) && !defined(NT_INCLUDED)
3793	# if PAGE_SIZE != 0x2000
3794	# error "PAGE_SIZE is not 0x2000!"
3795	# endif
3796	# endif
3797	#else
3798	# define RT_ASM_PAGE_SIZE 0x1000
3799	# if defined(PAGE_SIZE) && !defined(NT_INCLUDED)
3800	# if PAGE_SIZE != 0x1000
3801	# error "PAGE_SIZE is not 0x1000!"
3802	# endif
3803	# endif
3804	#endif
3805
3806	/**
3807	* Zeros a 4K memory page.
3808	*
3809	* @param pv Pointer to the memory block. This must be page aligned.
3810	*/
3811	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3812	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMMemZeroPage(volatile void RT_FAR *pv);
3813	# else
3814	DECLINLINE(void) ASMMemZeroPage(volatile void RT_FAR *pv)
3815	{
3816	# if RT_INLINE_ASM_USES_INTRIN
3817	# ifdef RT_ARCH_AMD64
3818	__stosq((unsigned __int64 *)pv, 0, RT_ASM_PAGE_SIZE / 8);
3819	# else
3820	__stosd((unsigned long *)pv, 0, RT_ASM_PAGE_SIZE / 4);
3821	# endif
3822
3823	# elif RT_INLINE_ASM_GNU_STYLE
3824	RTCCUINTREG uDummy;
3825	# ifdef RT_ARCH_AMD64
3826	__asm__ __volatile__("rep stosq"
3827	: "=D" (pv),
3828	"=c" (uDummy)
3829	: "0" (pv),
3830	"c" (RT_ASM_PAGE_SIZE >> 3),
3831	"a" (0)
3832	: "memory");
3833	# else
3834	__asm__ __volatile__("rep stosl"
3835	: "=D" (pv),
3836	"=c" (uDummy)
3837	: "0" (pv),
3838	"c" (RT_ASM_PAGE_SIZE >> 2),
3839	"a" (0)
3840	: "memory");
3841	# endif
3842	# else
3843	__asm
3844	{
3845	# ifdef RT_ARCH_AMD64
3846	xor rax, rax
3847	mov ecx, 0200h
3848	mov rdi, [pv]
3849	rep stosq
3850	# else
3851	xor eax, eax
3852	mov ecx, 0400h
3853	mov edi, [pv]
3854	rep stosd
3855	# endif
3856	}
3857	# endif
3858	}
3859	# endif
3860
3861
3862	/**
3863	* Zeros a memory block with a 32-bit aligned size.
3864	*
3865	* @param pv Pointer to the memory block.
3866	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
3867	*/
3868	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3869	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMMemZero32(volatile void RT_FAR *pv, size_t cb);
3870	#else
3871	DECLINLINE(void) ASMMemZero32(volatile void RT_FAR *pv, size_t cb)
3872	{
3873	# if RT_INLINE_ASM_USES_INTRIN
3874	# ifdef RT_ARCH_AMD64
3875	if (!(cb & 7))
3876	__stosq((unsigned __int64 RT_FAR *)pv, 0, cb / 8);
3877	else
3878	# endif
3879	__stosd((unsigned long RT_FAR *)pv, 0, cb / 4);
3880
3881	# elif RT_INLINE_ASM_GNU_STYLE
3882	__asm__ __volatile__("rep stosl"
3883	: "=D" (pv),
3884	"=c" (cb)
3885	: "0" (pv),
3886	"1" (cb >> 2),
3887	"a" (0)
3888	: "memory");
3889	# else
3890	__asm
3891	{
3892	xor eax, eax
3893	# ifdef RT_ARCH_AMD64
3894	mov rcx, [cb]
3895	shr rcx, 2
3896	mov rdi, [pv]
3897	# else
3898	mov ecx, [cb]
3899	shr ecx, 2
3900	mov edi, [pv]
3901	# endif
3902	rep stosd
3903	}
3904	# endif
3905	}
3906	#endif
3907
3908
3909	/**
3910	* Fills a memory block with a 32-bit aligned size.
3911	*
3912	* @param pv Pointer to the memory block.
3913	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
3914	* @param u32 The value to fill with.
3915	*/
3916	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
3917	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMMemFill32(volatile void RT_FAR *pv, size_t cb, uint32_t u32);
3918	#else
3919	DECLINLINE(void) ASMMemFill32(volatile void RT_FAR *pv, size_t cb, uint32_t u32)
3920	{
3921	# if RT_INLINE_ASM_USES_INTRIN
3922	# ifdef RT_ARCH_AMD64
3923	if (!(cb & 7))
3924	__stosq((unsigned __int64 RT_FAR *)pv, RT_MAKE_U64(u32, u32), cb / 8);
3925	else
3926	# endif
3927	__stosd((unsigned long RT_FAR *)pv, u32, cb / 4);
3928
3929	# elif RT_INLINE_ASM_GNU_STYLE
3930	__asm__ __volatile__("rep stosl"
3931	: "=D" (pv),
3932	"=c" (cb)
3933	: "0" (pv),
3934	"1" (cb >> 2),
3935	"a" (u32)
3936	: "memory");
3937	# else
3938	__asm
3939	{
3940	# ifdef RT_ARCH_AMD64
3941	mov rcx, [cb]
3942	shr rcx, 2
3943	mov rdi, [pv]
3944	# else
3945	mov ecx, [cb]
3946	shr ecx, 2
3947	mov edi, [pv]
3948	# endif
3949	mov eax, [u32]
3950	rep stosd
3951	}
3952	# endif
3953	}
3954	#endif
3955
3956
3957	/**
3958	* Checks if a memory block is all zeros.
3959	*
3960	* @returns Pointer to the first non-zero byte.
3961	* @returns NULL if all zero.
3962	*
3963	* @param pv Pointer to the memory block.
3964	* @param cb Number of bytes in the block.
3965	*
3966	* @todo Fix name, it is a predicate function but it's not returning boolean!
3967	*/
3968	#if !defined(RDESKTOP) && (!defined(RT_OS_LINUX) \|\| !defined(__KERNEL__)) \
3969	&& !defined(RT_ARCH_SPARC64) \
3970	&& !defined(RT_ARCH_SPARC)
3971	DECLASM(void RT_FAR ) ASMMemFirstNonZero(void const RT_FAR pv, size_t cb);
3972	#else
3973	DECLINLINE(void RT_FAR ) ASMMemFirstNonZero(void const RT_FAR pv, size_t cb)
3974	{
3975	uint8_t const pb = (uint8_t const RT_FAR )pv;
3976	for (; cb; cb--, pb++)
3977	if (RT_LIKELY(*pb == 0))
3978	{ /* likely */ }
3979	else
3980	return (void RT_FAR *)pb;
3981	return NULL;
3982	}
3983	#endif
3984
3985
3986	/**
3987	* Checks if a memory block is all zeros.
3988	*
3989	* @returns true if zero, false if not.
3990	*
3991	* @param pv Pointer to the memory block.
3992	* @param cb Number of bytes in the block.
3993	*
3994	* @sa ASMMemFirstNonZero
3995	*/
3996	DECLINLINE(bool) ASMMemIsZero(void const RT_FAR *pv, size_t cb)
3997	{
3998	return ASMMemFirstNonZero(pv, cb) == NULL;
3999	}
4000
4001
4002	/**
4003	* Checks if a memory page is all zeros.
4004	*
4005	* @returns true / false.
4006	*
4007	* @param pvPage Pointer to the page. Must be aligned on 16 byte
4008	* boundary
4009	*/
4010	DECLINLINE(bool) ASMMemIsZeroPage(void const RT_FAR *pvPage)
4011	{
4012	# if 0 /RT_INLINE_ASM_GNU_STYLE - this is actually slower... /
4013	union { RTCCUINTREG r; bool f; } uAX;
4014	RTCCUINTREG xCX, xDI;
4015	Assert(!((uintptr_t)pvPage & 15));
4016	__asm__ __volatile__("repe; "
4017	# ifdef RT_ARCH_AMD64
4018	"scasq\n\t"
4019	# else
4020	"scasl\n\t"
4021	# endif
4022	"setnc %%al\n\t"
4023	: "=&c" (xCX),
4024	"=&D" (xDI),
4025	"=&a" (uAX.r)
4026	: "mr" (pvPage),
4027	# ifdef RT_ARCH_AMD64
4028	"0" (RT_ASM_PAGE_SIZE/8),
4029	# else
4030	"0" (RT_ASM_PAGE_SIZE/4),
4031	# endif
4032	"1" (pvPage),
4033	"2" (0));
4034	return uAX.f;
4035	# else
4036	uintptr_t const RT_FAR puPtr = (uintptr_t const RT_FAR )pvPage;
4037	size_t cLeft = RT_ASM_PAGE_SIZE / sizeof(uintptr_t) / 8;
4038	Assert(!((uintptr_t)pvPage & 15));
4039	for (;;)
4040	{
4041	if (puPtr[0]) return false;
4042	if (puPtr[4]) return false;
4043
4044	if (puPtr[2]) return false;
4045	if (puPtr[6]) return false;
4046
4047	if (puPtr[1]) return false;
4048	if (puPtr[5]) return false;
4049
4050	if (puPtr[3]) return false;
4051	if (puPtr[7]) return false;
4052
4053	if (!--cLeft)
4054	return true;
4055	puPtr += 8;
4056	}
4057	# endif
4058	}
4059
4060
4061	/**
4062	* Checks if a memory block is filled with the specified byte, returning the
4063	* first mismatch.
4064	*
4065	* This is sort of an inverted memchr.
4066	*
4067	* @returns Pointer to the byte which doesn't equal u8.
4068	* @returns NULL if all equal to u8.
4069	*
4070	* @param pv Pointer to the memory block.
4071	* @param cb Number of bytes in the block.
4072	* @param u8 The value it's supposed to be filled with.
4073	*
4074	* @remarks No alignment requirements.
4075	*/
4076	#if (!defined(RT_OS_LINUX) \|\| !defined(__KERNEL__)) \
4077	&& (!defined(RT_OS_FREEBSD) \|\| !defined(_KERNEL)) \
4078	&& !defined(RT_ARCH_SPARC64) \
4079	&& !defined(RT_ARCH_SPARC)
4080	DECLASM(void ) ASMMemFirstMismatchingU8(void const RT_FAR pv, size_t cb, uint8_t u8);
4081	#else
4082	DECLINLINE(void ) ASMMemFirstMismatchingU8(void const RT_FAR pv, size_t cb, uint8_t u8)
4083	{
4084	uint8_t const pb = (uint8_t const RT_FAR )pv;
4085	for (; cb; cb--, pb++)
4086	if (RT_LIKELY(*pb == u8))
4087	{ /* likely */ }
4088	else
4089	return (void *)pb;
4090	return NULL;
4091	}
4092	#endif
4093
4094
4095	/**
4096	* Checks if a memory block is filled with the specified byte.
4097	*
4098	* @returns true if all matching, false if not.
4099	*
4100	* @param pv Pointer to the memory block.
4101	* @param cb Number of bytes in the block.
4102	* @param u8 The value it's supposed to be filled with.
4103	*
4104	* @remarks No alignment requirements.
4105	*/
4106	DECLINLINE(bool) ASMMemIsAllU8(void const RT_FAR *pv, size_t cb, uint8_t u8)
4107	{
4108	return ASMMemFirstMismatchingU8(pv, cb, u8) == NULL;
4109	}
4110
4111
4112	/**
4113	* Checks if a memory block is filled with the specified 32-bit value.
4114	*
4115	* This is a sort of inverted memchr.
4116	*
4117	* @returns Pointer to the first value which doesn't equal u32.
4118	* @returns NULL if all equal to u32.
4119	*
4120	* @param pv Pointer to the memory block.
4121	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
4122	* @param u32 The value it's supposed to be filled with.
4123	*/
4124	DECLINLINE(uint32_t RT_FAR ) ASMMemFirstMismatchingU32(void const RT_FAR pv, size_t cb, uint32_t u32)
4125	{
4126	/** @todo rewrite this in inline assembly? */
4127	uint32_t const RT_FAR pu32 = (uint32_t const RT_FAR )pv;
4128	for (; cb; cb -= 4, pu32++)
4129	if (RT_LIKELY(*pu32 == u32))
4130	{ /* likely */ }
4131	else
4132	return (uint32_t RT_FAR *)pu32;
4133	return NULL;
4134	}
4135
4136
4137	/**
4138	* Probes a byte pointer for read access.
4139	*
4140	* While the function will not fault if the byte is not read accessible,
4141	* the idea is to do this in a safe place like before acquiring locks
4142	* and such like.
4143	*
4144	* Also, this functions guarantees that an eager compiler is not going
4145	* to optimize the probing away.
4146	*
4147	* @param pvByte Pointer to the byte.
4148	*/
4149	#if RT_INLINE_ASM_EXTERNAL
4150	RT_ASM_DECL_PRAGMA_WATCOM(uint8_t) ASMProbeReadByte(const void RT_FAR *pvByte);
4151	#else
4152	DECLINLINE(uint8_t) ASMProbeReadByte(const void RT_FAR *pvByte)
4153	{
4154	/** @todo verify that the compiler actually doesn't optimize this away. (intel & gcc) */
4155	uint8_t u8;
4156	# if RT_INLINE_ASM_GNU_STYLE
4157	__asm__ __volatile__("movb (%1), %0\n\t"
4158	: "=r" (u8)
4159	: "r" (pvByte));
4160	# else
4161	__asm
4162	{
4163	# ifdef RT_ARCH_AMD64
4164	mov rax, [pvByte]
4165	mov al, [rax]
4166	# else
4167	mov eax, [pvByte]
4168	mov al, [eax]
4169	# endif
4170	mov [u8], al
4171	}
4172	# endif
4173	return u8;
4174	}
4175	#endif
4176
4177	/**
4178	* Probes a buffer for read access page by page.
4179	*
4180	* While the function will fault if the buffer is not fully read
4181	* accessible, the idea is to do this in a safe place like before
4182	* acquiring locks and such like.
4183	*
4184	* Also, this functions guarantees that an eager compiler is not going
4185	* to optimize the probing away.
4186	*
4187	* @param pvBuf Pointer to the buffer.
4188	* @param cbBuf The size of the buffer in bytes. Must be >= 1.
4189	*/
4190	DECLINLINE(void) ASMProbeReadBuffer(const void RT_FAR *pvBuf, size_t cbBuf)
4191	{
4192	/** @todo verify that the compiler actually doesn't optimize this away. (intel & gcc) */
4193	/* the first byte */
4194	const uint8_t RT_FAR pu8 = (const uint8_t RT_FAR )pvBuf;
4195	ASMProbeReadByte(pu8);
4196
4197	/* the pages in between pages. */
4198	while (cbBuf > RT_ASM_PAGE_SIZE)
4199	{
4200	ASMProbeReadByte(pu8);
4201	cbBuf -= RT_ASM_PAGE_SIZE;
4202	pu8 += RT_ASM_PAGE_SIZE;
4203	}
4204
4205	/* the last byte */
4206	ASMProbeReadByte(pu8 + cbBuf - 1);
4207	}
4208
4209
4210
4211	/** @defgroup grp_inline_bits Bit Operations
4212	* @{
4213	*/
4214
4215
4216	/**
4217	* Sets a bit in a bitmap.
4218	*
4219	* @param pvBitmap Pointer to the bitmap. This should be 32-bit aligned.
4220	* @param iBit The bit to set.
4221	*
4222	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4223	* However, doing so will yield better performance as well as avoiding
4224	* traps accessing the last bits in the bitmap.
4225	*/
4226	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4227	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMBitSet(volatile void RT_FAR *pvBitmap, int32_t iBit);
4228	#else
4229	DECLINLINE(void) ASMBitSet(volatile void RT_FAR *pvBitmap, int32_t iBit)
4230	{
4231	# if RT_INLINE_ASM_USES_INTRIN
4232	_bittestandset((long RT_FAR *)pvBitmap, iBit);
4233
4234	# elif RT_INLINE_ASM_GNU_STYLE
4235	__asm__ __volatile__("btsl %1, %0"
4236	: "=m" ((volatile long RT_FAR )pvBitmap)
4237	: "Ir" (iBit),
4238	"m" ((volatile long RT_FAR )pvBitmap)
4239	: "memory");
4240	# else
4241	__asm
4242	{
4243	# ifdef RT_ARCH_AMD64
4244	mov rax, [pvBitmap]
4245	mov edx, [iBit]
4246	bts [rax], edx
4247	# else
4248	mov eax, [pvBitmap]
4249	mov edx, [iBit]
4250	bts [eax], edx
4251	# endif
4252	}
4253	# endif
4254	}
4255	#endif
4256
4257
4258	/**
4259	* Atomically sets a bit in a bitmap, ordered.
4260	*
4261	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4262	* the memory access isn't atomic!
4263	* @param iBit The bit to set.
4264	*
4265	* @remarks x86: Requires a 386 or later.
4266	*/
4267	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4268	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicBitSet(volatile void RT_FAR *pvBitmap, int32_t iBit);
4269	#else
4270	DECLINLINE(void) ASMAtomicBitSet(volatile void RT_FAR *pvBitmap, int32_t iBit)
4271	{
4272	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4273	# if RT_INLINE_ASM_USES_INTRIN
4274	_interlockedbittestandset((long RT_FAR *)pvBitmap, iBit);
4275	# elif RT_INLINE_ASM_GNU_STYLE
4276	__asm__ __volatile__("lock; btsl %1, %0"
4277	: "=m" ((volatile long )pvBitmap)
4278	: "Ir" (iBit),
4279	"m" ((volatile long )pvBitmap)
4280	: "memory");
4281	# else
4282	__asm
4283	{
4284	# ifdef RT_ARCH_AMD64
4285	mov rax, [pvBitmap]
4286	mov edx, [iBit]
4287	lock bts [rax], edx
4288	# else
4289	mov eax, [pvBitmap]
4290	mov edx, [iBit]
4291	lock bts [eax], edx
4292	# endif
4293	}
4294	# endif
4295	}
4296	#endif
4297
4298
4299	/**
4300	* Clears a bit in a bitmap.
4301	*
4302	* @param pvBitmap Pointer to the bitmap.
4303	* @param iBit The bit to clear.
4304	*
4305	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4306	* However, doing so will yield better performance as well as avoiding
4307	* traps accessing the last bits in the bitmap.
4308	*/
4309	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4310	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMBitClear(volatile void RT_FAR *pvBitmap, int32_t iBit);
4311	#else
4312	DECLINLINE(void) ASMBitClear(volatile void RT_FAR *pvBitmap, int32_t iBit)
4313	{
4314	# if RT_INLINE_ASM_USES_INTRIN
4315	_bittestandreset((long RT_FAR *)pvBitmap, iBit);
4316
4317	# elif RT_INLINE_ASM_GNU_STYLE
4318	__asm__ __volatile__("btrl %1, %0"
4319	: "=m" ((volatile long RT_FAR )pvBitmap)
4320	: "Ir" (iBit),
4321	"m" ((volatile long RT_FAR )pvBitmap)
4322	: "memory");
4323	# else
4324	__asm
4325	{
4326	# ifdef RT_ARCH_AMD64
4327	mov rax, [pvBitmap]
4328	mov edx, [iBit]
4329	btr [rax], edx
4330	# else
4331	mov eax, [pvBitmap]
4332	mov edx, [iBit]
4333	btr [eax], edx
4334	# endif
4335	}
4336	# endif
4337	}
4338	#endif
4339
4340
4341	/**
4342	* Atomically clears a bit in a bitmap, ordered.
4343	*
4344	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4345	* the memory access isn't atomic!
4346	* @param iBit The bit to toggle set.
4347	*
4348	* @remarks No memory barrier, take care on smp.
4349	* @remarks x86: Requires a 386 or later.
4350	*/
4351	#if RT_INLINE_ASM_EXTERNAL
4352	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicBitClear(volatile void RT_FAR *pvBitmap, int32_t iBit);
4353	#else
4354	DECLINLINE(void) ASMAtomicBitClear(volatile void RT_FAR *pvBitmap, int32_t iBit)
4355	{
4356	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4357	# if RT_INLINE_ASM_GNU_STYLE
4358	__asm__ __volatile__("lock; btrl %1, %0"
4359	: "=m" ((volatile long RT_FAR )pvBitmap)
4360	: "Ir" (iBit),
4361	"m" ((volatile long RT_FAR )pvBitmap)
4362	: "memory");
4363	# else
4364	__asm
4365	{
4366	# ifdef RT_ARCH_AMD64
4367	mov rax, [pvBitmap]
4368	mov edx, [iBit]
4369	lock btr [rax], edx
4370	# else
4371	mov eax, [pvBitmap]
4372	mov edx, [iBit]
4373	lock btr [eax], edx
4374	# endif
4375	}
4376	# endif
4377	}
4378	#endif
4379
4380
4381	/**
4382	* Toggles a bit in a bitmap.
4383	*
4384	* @param pvBitmap Pointer to the bitmap.
4385	* @param iBit The bit to toggle.
4386	*
4387	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4388	* However, doing so will yield better performance as well as avoiding
4389	* traps accessing the last bits in the bitmap.
4390	*/
4391	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4392	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMBitToggle(volatile void RT_FAR *pvBitmap, int32_t iBit);
4393	#else
4394	DECLINLINE(void) ASMBitToggle(volatile void RT_FAR *pvBitmap, int32_t iBit)
4395	{
4396	# if RT_INLINE_ASM_USES_INTRIN
4397	_bittestandcomplement((long RT_FAR *)pvBitmap, iBit);
4398	# elif RT_INLINE_ASM_GNU_STYLE
4399	__asm__ __volatile__("btcl %1, %0"
4400	: "=m" ((volatile long )pvBitmap)
4401	: "Ir" (iBit),
4402	"m" ((volatile long )pvBitmap)
4403	: "memory");
4404	# else
4405	__asm
4406	{
4407	# ifdef RT_ARCH_AMD64
4408	mov rax, [pvBitmap]
4409	mov edx, [iBit]
4410	btc [rax], edx
4411	# else
4412	mov eax, [pvBitmap]
4413	mov edx, [iBit]
4414	btc [eax], edx
4415	# endif
4416	}
4417	# endif
4418	}
4419	#endif
4420
4421
4422	/**
4423	* Atomically toggles a bit in a bitmap, ordered.
4424	*
4425	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4426	* the memory access isn't atomic!
4427	* @param iBit The bit to test and set.
4428	*
4429	* @remarks x86: Requires a 386 or later.
4430	*/
4431	#if RT_INLINE_ASM_EXTERNAL
4432	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMAtomicBitToggle(volatile void RT_FAR *pvBitmap, int32_t iBit);
4433	#else
4434	DECLINLINE(void) ASMAtomicBitToggle(volatile void RT_FAR *pvBitmap, int32_t iBit)
4435	{
4436	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4437	# if RT_INLINE_ASM_GNU_STYLE
4438	__asm__ __volatile__("lock; btcl %1, %0"
4439	: "=m" ((volatile long RT_FAR )pvBitmap)
4440	: "Ir" (iBit),
4441	"m" ((volatile long RT_FAR )pvBitmap)
4442	: "memory");
4443	# else
4444	__asm
4445	{
4446	# ifdef RT_ARCH_AMD64
4447	mov rax, [pvBitmap]
4448	mov edx, [iBit]
4449	lock btc [rax], edx
4450	# else
4451	mov eax, [pvBitmap]
4452	mov edx, [iBit]
4453	lock btc [eax], edx
4454	# endif
4455	}
4456	# endif
4457	}
4458	#endif
4459
4460
4461	/**
4462	* Tests and sets a bit in a bitmap.
4463	*
4464	* @returns true if the bit was set.
4465	* @returns false if the bit was clear.
4466	*
4467	* @param pvBitmap Pointer to the bitmap.
4468	* @param iBit The bit to test and set.
4469	*
4470	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4471	* However, doing so will yield better performance as well as avoiding
4472	* traps accessing the last bits in the bitmap.
4473	*/
4474	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4475	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMBitTestAndSet(volatile void RT_FAR *pvBitmap, int32_t iBit);
4476	#else
4477	DECLINLINE(bool) ASMBitTestAndSet(volatile void RT_FAR *pvBitmap, int32_t iBit)
4478	{
4479	union { bool f; uint32_t u32; uint8_t u8; } rc;
4480	# if RT_INLINE_ASM_USES_INTRIN
4481	rc.u8 = _bittestandset((long RT_FAR *)pvBitmap, iBit);
4482
4483	# elif RT_INLINE_ASM_GNU_STYLE
4484	__asm__ __volatile__("btsl %2, %1\n\t"
4485	"setc %b0\n\t"
4486	"andl $1, %0\n\t"
4487	: "=q" (rc.u32),
4488	"=m" ((volatile long RT_FAR )pvBitmap)
4489	: "Ir" (iBit),
4490	"m" ((volatile long RT_FAR )pvBitmap)
4491	: "memory");
4492	# else
4493	__asm
4494	{
4495	mov edx, [iBit]
4496	# ifdef RT_ARCH_AMD64
4497	mov rax, [pvBitmap]
4498	bts [rax], edx
4499	# else
4500	mov eax, [pvBitmap]
4501	bts [eax], edx
4502	# endif
4503	setc al
4504	and eax, 1
4505	mov [rc.u32], eax
4506	}
4507	# endif
4508	return rc.f;
4509	}
4510	#endif
4511
4512
4513	/**
4514	* Atomically tests and sets a bit in a bitmap, ordered.
4515	*
4516	* @returns true if the bit was set.
4517	* @returns false if the bit was clear.
4518	*
4519	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4520	* the memory access isn't atomic!
4521	* @param iBit The bit to set.
4522	*
4523	* @remarks x86: Requires a 386 or later.
4524	*/
4525	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4526	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicBitTestAndSet(volatile void RT_FAR *pvBitmap, int32_t iBit);
4527	#else
4528	DECLINLINE(bool) ASMAtomicBitTestAndSet(volatile void RT_FAR *pvBitmap, int32_t iBit)
4529	{
4530	union { bool f; uint32_t u32; uint8_t u8; } rc;
4531	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4532	# if RT_INLINE_ASM_USES_INTRIN
4533	rc.u8 = _interlockedbittestandset((long RT_FAR *)pvBitmap, iBit);
4534	# elif RT_INLINE_ASM_GNU_STYLE
4535	__asm__ __volatile__("lock; btsl %2, %1\n\t"
4536	"setc %b0\n\t"
4537	"andl $1, %0\n\t"
4538	: "=q" (rc.u32),
4539	"=m" ((volatile long RT_FAR )pvBitmap)
4540	: "Ir" (iBit),
4541	"m" ((volatile long RT_FAR )pvBitmap)
4542	: "memory");
4543	# else
4544	__asm
4545	{
4546	mov edx, [iBit]
4547	# ifdef RT_ARCH_AMD64
4548	mov rax, [pvBitmap]
4549	lock bts [rax], edx
4550	# else
4551	mov eax, [pvBitmap]
4552	lock bts [eax], edx
4553	# endif
4554	setc al
4555	and eax, 1
4556	mov [rc.u32], eax
4557	}
4558	# endif
4559	return rc.f;
4560	}
4561	#endif
4562
4563
4564	/**
4565	* Tests and clears a bit in a bitmap.
4566	*
4567	* @returns true if the bit was set.
4568	* @returns false if the bit was clear.
4569	*
4570	* @param pvBitmap Pointer to the bitmap.
4571	* @param iBit The bit to test and clear.
4572	*
4573	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4574	* However, doing so will yield better performance as well as avoiding
4575	* traps accessing the last bits in the bitmap.
4576	*/
4577	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4578	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMBitTestAndClear(volatile void RT_FAR *pvBitmap, int32_t iBit);
4579	#else
4580	DECLINLINE(bool) ASMBitTestAndClear(volatile void RT_FAR *pvBitmap, int32_t iBit)
4581	{
4582	union { bool f; uint32_t u32; uint8_t u8; } rc;
4583	# if RT_INLINE_ASM_USES_INTRIN
4584	rc.u8 = _bittestandreset((long RT_FAR *)pvBitmap, iBit);
4585
4586	# elif RT_INLINE_ASM_GNU_STYLE
4587	__asm__ __volatile__("btrl %2, %1\n\t"
4588	"setc %b0\n\t"
4589	"andl $1, %0\n\t"
4590	: "=q" (rc.u32),
4591	"=m" ((volatile long RT_FAR )pvBitmap)
4592	: "Ir" (iBit),
4593	"m" ((volatile long RT_FAR )pvBitmap)
4594	: "memory");
4595	# else
4596	__asm
4597	{
4598	mov edx, [iBit]
4599	# ifdef RT_ARCH_AMD64
4600	mov rax, [pvBitmap]
4601	btr [rax], edx
4602	# else
4603	mov eax, [pvBitmap]
4604	btr [eax], edx
4605	# endif
4606	setc al
4607	and eax, 1
4608	mov [rc.u32], eax
4609	}
4610	# endif
4611	return rc.f;
4612	}
4613	#endif
4614
4615
4616	/**
4617	* Atomically tests and clears a bit in a bitmap, ordered.
4618	*
4619	* @returns true if the bit was set.
4620	* @returns false if the bit was clear.
4621	*
4622	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4623	* the memory access isn't atomic!
4624	* @param iBit The bit to test and clear.
4625	*
4626	* @remarks No memory barrier, take care on smp.
4627	* @remarks x86: Requires a 386 or later.
4628	*/
4629	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4630	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicBitTestAndClear(volatile void RT_FAR *pvBitmap, int32_t iBit);
4631	#else
4632	DECLINLINE(bool) ASMAtomicBitTestAndClear(volatile void RT_FAR *pvBitmap, int32_t iBit)
4633	{
4634	union { bool f; uint32_t u32; uint8_t u8; } rc;
4635	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4636	# if RT_INLINE_ASM_USES_INTRIN
4637	rc.u8 = _interlockedbittestandreset((long RT_FAR *)pvBitmap, iBit);
4638
4639	# elif RT_INLINE_ASM_GNU_STYLE
4640	__asm__ __volatile__("lock; btrl %2, %1\n\t"
4641	"setc %b0\n\t"
4642	"andl $1, %0\n\t"
4643	: "=q" (rc.u32),
4644	"=m" ((volatile long RT_FAR )pvBitmap)
4645	: "Ir" (iBit),
4646	"m" ((volatile long RT_FAR )pvBitmap)
4647	: "memory");
4648	# else
4649	__asm
4650	{
4651	mov edx, [iBit]
4652	# ifdef RT_ARCH_AMD64
4653	mov rax, [pvBitmap]
4654	lock btr [rax], edx
4655	# else
4656	mov eax, [pvBitmap]
4657	lock btr [eax], edx
4658	# endif
4659	setc al
4660	and eax, 1
4661	mov [rc.u32], eax
4662	}
4663	# endif
4664	return rc.f;
4665	}
4666	#endif
4667
4668
4669	/**
4670	* Tests and toggles a bit in a bitmap.
4671	*
4672	* @returns true if the bit was set.
4673	* @returns false if the bit was clear.
4674	*
4675	* @param pvBitmap Pointer to the bitmap.
4676	* @param iBit The bit to test and toggle.
4677	*
4678	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4679	* However, doing so will yield better performance as well as avoiding
4680	* traps accessing the last bits in the bitmap.
4681	*/
4682	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4683	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMBitTestAndToggle(volatile void RT_FAR *pvBitmap, int32_t iBit);
4684	#else
4685	DECLINLINE(bool) ASMBitTestAndToggle(volatile void RT_FAR *pvBitmap, int32_t iBit)
4686	{
4687	union { bool f; uint32_t u32; uint8_t u8; } rc;
4688	# if RT_INLINE_ASM_USES_INTRIN
4689	rc.u8 = _bittestandcomplement((long RT_FAR *)pvBitmap, iBit);
4690
4691	# elif RT_INLINE_ASM_GNU_STYLE
4692	__asm__ __volatile__("btcl %2, %1\n\t"
4693	"setc %b0\n\t"
4694	"andl $1, %0\n\t"
4695	: "=q" (rc.u32),
4696	"=m" ((volatile long RT_FAR )pvBitmap)
4697	: "Ir" (iBit),
4698	"m" ((volatile long RT_FAR )pvBitmap)
4699	: "memory");
4700	# else
4701	__asm
4702	{
4703	mov edx, [iBit]
4704	# ifdef RT_ARCH_AMD64
4705	mov rax, [pvBitmap]
4706	btc [rax], edx
4707	# else
4708	mov eax, [pvBitmap]
4709	btc [eax], edx
4710	# endif
4711	setc al
4712	and eax, 1
4713	mov [rc.u32], eax
4714	}
4715	# endif
4716	return rc.f;
4717	}
4718	#endif
4719
4720
4721	/**
4722	* Atomically tests and toggles a bit in a bitmap, ordered.
4723	*
4724	* @returns true if the bit was set.
4725	* @returns false if the bit was clear.
4726	*
4727	* @param pvBitmap Pointer to the bitmap. Must be 32-bit aligned, otherwise
4728	* the memory access isn't atomic!
4729	* @param iBit The bit to test and toggle.
4730	*
4731	* @remarks x86: Requires a 386 or later.
4732	*/
4733	#if RT_INLINE_ASM_EXTERNAL
4734	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMAtomicBitTestAndToggle(volatile void RT_FAR *pvBitmap, int32_t iBit);
4735	#else
4736	DECLINLINE(bool) ASMAtomicBitTestAndToggle(volatile void RT_FAR *pvBitmap, int32_t iBit)
4737	{
4738	union { bool f; uint32_t u32; uint8_t u8; } rc;
4739	AssertMsg(!((uintptr_t)pvBitmap & 3), ("address %p not 32-bit aligned", pvBitmap));
4740	# if RT_INLINE_ASM_GNU_STYLE
4741	__asm__ __volatile__("lock; btcl %2, %1\n\t"
4742	"setc %b0\n\t"
4743	"andl $1, %0\n\t"
4744	: "=q" (rc.u32),
4745	"=m" ((volatile long RT_FAR )pvBitmap)
4746	: "Ir" (iBit),
4747	"m" ((volatile long RT_FAR )pvBitmap)
4748	: "memory");
4749	# else
4750	__asm
4751	{
4752	mov edx, [iBit]
4753	# ifdef RT_ARCH_AMD64
4754	mov rax, [pvBitmap]
4755	lock btc [rax], edx
4756	# else
4757	mov eax, [pvBitmap]
4758	lock btc [eax], edx
4759	# endif
4760	setc al
4761	and eax, 1
4762	mov [rc.u32], eax
4763	}
4764	# endif
4765	return rc.f;
4766	}
4767	#endif
4768
4769
4770	/**
4771	* Tests if a bit in a bitmap is set.
4772	*
4773	* @returns true if the bit is set.
4774	* @returns false if the bit is clear.
4775	*
4776	* @param pvBitmap Pointer to the bitmap.
4777	* @param iBit The bit to test.
4778	*
4779	* @remarks The 32-bit aligning of pvBitmap is not a strict requirement.
4780	* However, doing so will yield better performance as well as avoiding
4781	* traps accessing the last bits in the bitmap.
4782	*/
4783	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
4784	RT_ASM_DECL_PRAGMA_WATCOM(bool) ASMBitTest(const volatile void RT_FAR *pvBitmap, int32_t iBit);
4785	#else
4786	DECLINLINE(bool) ASMBitTest(const volatile void RT_FAR *pvBitmap, int32_t iBit)
4787	{
4788	union { bool f; uint32_t u32; uint8_t u8; } rc;
4789	# if RT_INLINE_ASM_USES_INTRIN
4790	rc.u32 = _bittest((long *)pvBitmap, iBit);
4791	# elif RT_INLINE_ASM_GNU_STYLE
4792
4793	__asm__ __volatile__("btl %2, %1\n\t"
4794	"setc %b0\n\t"
4795	"andl $1, %0\n\t"
4796	: "=q" (rc.u32)
4797	: "m" ((const volatile long RT_FAR )pvBitmap),
4798	"Ir" (iBit)
4799	: "memory");
4800	# else
4801	__asm
4802	{
4803	mov edx, [iBit]
4804	# ifdef RT_ARCH_AMD64
4805	mov rax, [pvBitmap]
4806	bt [rax], edx
4807	# else
4808	mov eax, [pvBitmap]
4809	bt [eax], edx
4810	# endif
4811	setc al
4812	and eax, 1
4813	mov [rc.u32], eax
4814	}
4815	# endif
4816	return rc.f;
4817	}
4818	#endif
4819
4820
4821	/**
4822	* Clears a bit range within a bitmap.
4823	*
4824	* @param pvBitmap Pointer to the bitmap.
4825	* @param iBitStart The First bit to clear.
4826	* @param iBitEnd The first bit not to clear.
4827	*/
4828	DECLINLINE(void) ASMBitClearRange(volatile void RT_FAR *pvBitmap, int32_t iBitStart, int32_t iBitEnd)
4829	{
4830	if (iBitStart < iBitEnd)
4831	{
4832	volatile uint32_t RT_FAR pu32 = (volatile uint32_t RT_FAR )pvBitmap + (iBitStart >> 5);
4833	int32_t iStart = iBitStart & ~31;
4834	int32_t iEnd = iBitEnd & ~31;
4835	if (iStart == iEnd)
4836	*pu32 &= ((UINT32_C(1) << (iBitStart & 31)) - 1) \| ~((UINT32_C(1) << (iBitEnd & 31)) - 1);
4837	else
4838	{
4839	/* bits in first dword. */
4840	if (iBitStart & 31)
4841	{
4842	*pu32 &= (UINT32_C(1) << (iBitStart & 31)) - 1;
4843	pu32++;
4844	iBitStart = iStart + 32;
4845	}
4846
4847	/* whole dword. */
4848	if (iBitStart != iEnd)
4849	ASMMemZero32(pu32, (iEnd - iBitStart) >> 3);
4850
4851	/* bits in last dword. */
4852	if (iBitEnd & 31)
4853	{
4854	pu32 = (volatile uint32_t *)pvBitmap + (iBitEnd >> 5);
4855	*pu32 &= ~((UINT32_C(1) << (iBitEnd & 31)) - 1);
4856	}
4857	}
4858	}
4859	}
4860
4861
4862	/**
4863	* Sets a bit range within a bitmap.
4864	*
4865	* @param pvBitmap Pointer to the bitmap.
4866	* @param iBitStart The First bit to set.
4867	* @param iBitEnd The first bit not to set.
4868	*/
4869	DECLINLINE(void) ASMBitSetRange(volatile void RT_FAR *pvBitmap, int32_t iBitStart, int32_t iBitEnd)
4870	{
4871	if (iBitStart < iBitEnd)
4872	{
4873	volatile uint32_t RT_FAR pu32 = (volatile uint32_t RT_FAR )pvBitmap + (iBitStart >> 5);
4874	int32_t iStart = iBitStart & ~31;
4875	int32_t iEnd = iBitEnd & ~31;
4876	if (iStart == iEnd)
4877	*pu32 \|= ((UINT32_C(1) << (iBitEnd - iBitStart)) - 1) << (iBitStart & 31);
4878	else
4879	{
4880	/* bits in first dword. */
4881	if (iBitStart & 31)
4882	{
4883	*pu32 \|= ~((UINT32_C(1) << (iBitStart & 31)) - 1);
4884	pu32++;
4885	iBitStart = iStart + 32;
4886	}
4887
4888	/* whole dword. */
4889	if (iBitStart != iEnd)
4890	ASMMemFill32(pu32, (iEnd - iBitStart) >> 3, ~UINT32_C(0));
4891
4892	/* bits in last dword. */
4893	if (iBitEnd & 31)
4894	{
4895	pu32 = (volatile uint32_t RT_FAR *)pvBitmap + (iBitEnd >> 5);
4896	*pu32 \|= (UINT32_C(1) << (iBitEnd & 31)) - 1;
4897	}
4898	}
4899	}
4900	}
4901
4902
4903	/**
4904	* Finds the first clear bit in a bitmap.
4905	*
4906	* @returns Index of the first zero bit.
4907	* @returns -1 if no clear bit was found.
4908	* @param pvBitmap Pointer to the bitmap.
4909	* @param cBits The number of bits in the bitmap. Multiple of 32.
4910	*/
4911	#if RT_INLINE_ASM_EXTERNAL
4912	DECLASM(int32_t) ASMBitFirstClear(const volatile void RT_FAR *pvBitmap, uint32_t cBits);
4913	#else
4914	DECLINLINE(int32_t) ASMBitFirstClear(const volatile void RT_FAR *pvBitmap, uint32_t cBits)
4915	{
4916	if (cBits)
4917	{
4918	int32_t iBit;
4919	# if RT_INLINE_ASM_GNU_STYLE
4920	RTCCUINTREG uEAX, uECX, uEDI;
4921	cBits = RT_ALIGN_32(cBits, 32);
4922	__asm__ __volatile__("repe; scasl\n\t"
4923	"je 1f\n\t"
4924	# ifdef RT_ARCH_AMD64
4925	"lea -4(%%rdi), %%rdi\n\t"
4926	"xorl (%%rdi), %%eax\n\t"
4927	"subq %5, %%rdi\n\t"
4928	# else
4929	"lea -4(%%edi), %%edi\n\t"
4930	"xorl (%%edi), %%eax\n\t"
4931	"subl %5, %%edi\n\t"
4932	# endif
4933	"shll $3, %%edi\n\t"
4934	"bsfl %%eax, %%edx\n\t"
4935	"addl %%edi, %%edx\n\t"
4936	"1:\t\n"
4937	: "=d" (iBit),
4938	"=&c" (uECX),
4939	"=&D" (uEDI),
4940	"=&a" (uEAX)
4941	: "0" (0xffffffff),
4942	"mr" (pvBitmap),
4943	"1" (cBits >> 5),
4944	"2" (pvBitmap),
4945	"3" (0xffffffff));
4946	# else
4947	cBits = RT_ALIGN_32(cBits, 32);
4948	__asm
4949	{
4950	# ifdef RT_ARCH_AMD64
4951	mov rdi, [pvBitmap]
4952	mov rbx, rdi
4953	# else
4954	mov edi, [pvBitmap]
4955	mov ebx, edi
4956	# endif
4957	mov edx, 0ffffffffh
4958	mov eax, edx
4959	mov ecx, [cBits]
4960	shr ecx, 5
4961	repe scasd
4962	je done
4963
4964	# ifdef RT_ARCH_AMD64
4965	lea rdi, [rdi - 4]
4966	xor eax, [rdi]
4967	sub rdi, rbx
4968	# else
4969	lea edi, [edi - 4]
4970	xor eax, [edi]
4971	sub edi, ebx
4972	# endif
4973	shl edi, 3
4974	bsf edx, eax
4975	add edx, edi
4976	done:
4977	mov [iBit], edx
4978	}
4979	# endif
4980	return iBit;
4981	}
4982	return -1;
4983	}
4984	#endif
4985
4986
4987	/**
4988	* Finds the next clear bit in a bitmap.
4989	*
4990	* @returns Index of the first zero bit.
4991	* @returns -1 if no clear bit was found.
4992	* @param pvBitmap Pointer to the bitmap.
4993	* @param cBits The number of bits in the bitmap. Multiple of 32.
4994	* @param iBitPrev The bit returned from the last search.
4995	* The search will start at iBitPrev + 1.
4996	*/
4997	#if RT_INLINE_ASM_EXTERNAL
4998	DECLASM(int) ASMBitNextClear(const volatile void RT_FAR *pvBitmap, uint32_t cBits, uint32_t iBitPrev);
4999	#else
5000	DECLINLINE(int) ASMBitNextClear(const volatile void RT_FAR *pvBitmap, uint32_t cBits, uint32_t iBitPrev)
5001	{
5002	const volatile uint32_t RT_FAR pau32Bitmap = (const volatile uint32_t RT_FAR )pvBitmap;
5003	int iBit = ++iBitPrev & 31;
5004	if (iBit)
5005	{
5006	/*
5007	* Inspect the 32-bit word containing the unaligned bit.
5008	*/
5009	uint32_t u32 = ~pau32Bitmap[iBitPrev / 32] >> iBit;
5010
5011	# if RT_INLINE_ASM_USES_INTRIN
5012	unsigned long ulBit = 0;
5013	if (_BitScanForward(&ulBit, u32))
5014	return ulBit + iBitPrev;
5015	# else
5016	# if RT_INLINE_ASM_GNU_STYLE
5017	__asm__ __volatile__("bsf %1, %0\n\t"
5018	"jnz 1f\n\t"
5019	"movl $-1, %0\n\t"
5020	"1:\n\t"
5021	: "=r" (iBit)
5022	: "r" (u32));
5023	# else
5024	__asm
5025	{
5026	mov edx, [u32]
5027	bsf eax, edx
5028	jnz done
5029	mov eax, 0ffffffffh
5030	done:
5031	mov [iBit], eax
5032	}
5033	# endif
5034	if (iBit >= 0)
5035	return iBit + iBitPrev;
5036	# endif
5037
5038	/*
5039	* Skip ahead and see if there is anything left to search.
5040	*/
5041	iBitPrev \|= 31;
5042	iBitPrev++;
5043	if (cBits <= (uint32_t)iBitPrev)
5044	return -1;
5045	}
5046
5047	/*
5048	* 32-bit aligned search, let ASMBitFirstClear do the dirty work.
5049	*/
5050	iBit = ASMBitFirstClear(&pau32Bitmap[iBitPrev / 32], cBits - iBitPrev);
5051	if (iBit >= 0)
5052	iBit += iBitPrev;
5053	return iBit;
5054	}
5055	#endif
5056
5057
5058	/**
5059	* Finds the first set bit in a bitmap.
5060	*
5061	* @returns Index of the first set bit.
5062	* @returns -1 if no clear bit was found.
5063	* @param pvBitmap Pointer to the bitmap.
5064	* @param cBits The number of bits in the bitmap. Multiple of 32.
5065	*/
5066	#if RT_INLINE_ASM_EXTERNAL
5067	DECLASM(int32_t) ASMBitFirstSet(const volatile void RT_FAR *pvBitmap, uint32_t cBits);
5068	#else
5069	DECLINLINE(int32_t) ASMBitFirstSet(const volatile void RT_FAR *pvBitmap, uint32_t cBits)
5070	{
5071	if (cBits)
5072	{
5073	int32_t iBit;
5074	# if RT_INLINE_ASM_GNU_STYLE
5075	RTCCUINTREG uEAX, uECX, uEDI;
5076	cBits = RT_ALIGN_32(cBits, 32);
5077	__asm__ __volatile__("repe; scasl\n\t"
5078	"je 1f\n\t"
5079	# ifdef RT_ARCH_AMD64
5080	"lea -4(%%rdi), %%rdi\n\t"
5081	"movl (%%rdi), %%eax\n\t"
5082	"subq %5, %%rdi\n\t"
5083	# else
5084	"lea -4(%%edi), %%edi\n\t"
5085	"movl (%%edi), %%eax\n\t"
5086	"subl %5, %%edi\n\t"
5087	# endif
5088	"shll $3, %%edi\n\t"
5089	"bsfl %%eax, %%edx\n\t"
5090	"addl %%edi, %%edx\n\t"
5091	"1:\t\n"
5092	: "=d" (iBit),
5093	"=&c" (uECX),
5094	"=&D" (uEDI),
5095	"=&a" (uEAX)
5096	: "0" (0xffffffff),
5097	"mr" (pvBitmap),
5098	"1" (cBits >> 5),
5099	"2" (pvBitmap),
5100	"3" (0));
5101	# else
5102	cBits = RT_ALIGN_32(cBits, 32);
5103	__asm
5104	{
5105	# ifdef RT_ARCH_AMD64
5106	mov rdi, [pvBitmap]
5107	mov rbx, rdi
5108	# else
5109	mov edi, [pvBitmap]
5110	mov ebx, edi
5111	# endif
5112	mov edx, 0ffffffffh
5113	xor eax, eax
5114	mov ecx, [cBits]
5115	shr ecx, 5
5116	repe scasd
5117	je done
5118	# ifdef RT_ARCH_AMD64
5119	lea rdi, [rdi - 4]
5120	mov eax, [rdi]
5121	sub rdi, rbx
5122	# else
5123	lea edi, [edi - 4]
5124	mov eax, [edi]
5125	sub edi, ebx
5126	# endif
5127	shl edi, 3
5128	bsf edx, eax
5129	add edx, edi
5130	done:
5131	mov [iBit], edx
5132	}
5133	# endif
5134	return iBit;
5135	}
5136	return -1;
5137	}
5138	#endif
5139
5140
5141	/**
5142	* Finds the next set bit in a bitmap.
5143	*
5144	* @returns Index of the next set bit.
5145	* @returns -1 if no set bit was found.
5146	* @param pvBitmap Pointer to the bitmap.
5147	* @param cBits The number of bits in the bitmap. Multiple of 32.
5148	* @param iBitPrev The bit returned from the last search.
5149	* The search will start at iBitPrev + 1.
5150	*/
5151	#if RT_INLINE_ASM_EXTERNAL
5152	DECLASM(int) ASMBitNextSet(const volatile void RT_FAR *pvBitmap, uint32_t cBits, uint32_t iBitPrev);
5153	#else
5154	DECLINLINE(int) ASMBitNextSet(const volatile void RT_FAR *pvBitmap, uint32_t cBits, uint32_t iBitPrev)
5155	{
5156	const volatile uint32_t RT_FAR pau32Bitmap = (const volatile uint32_t RT_FAR )pvBitmap;
5157	int iBit = ++iBitPrev & 31;
5158	if (iBit)
5159	{
5160	/*
5161	* Inspect the 32-bit word containing the unaligned bit.
5162	*/
5163	uint32_t u32 = pau32Bitmap[iBitPrev / 32] >> iBit;
5164
5165	# if RT_INLINE_ASM_USES_INTRIN
5166	unsigned long ulBit = 0;
5167	if (_BitScanForward(&ulBit, u32))
5168	return ulBit + iBitPrev;
5169	# else
5170	# if RT_INLINE_ASM_GNU_STYLE
5171	__asm__ __volatile__("bsf %1, %0\n\t"
5172	"jnz 1f\n\t"
5173	"movl $-1, %0\n\t"
5174	"1:\n\t"
5175	: "=r" (iBit)
5176	: "r" (u32));
5177	# else
5178	__asm
5179	{
5180	mov edx, [u32]
5181	bsf eax, edx
5182	jnz done
5183	mov eax, 0ffffffffh
5184	done:
5185	mov [iBit], eax
5186	}
5187	# endif
5188	if (iBit >= 0)
5189	return iBit + iBitPrev;
5190	# endif
5191
5192	/*
5193	* Skip ahead and see if there is anything left to search.
5194	*/
5195	iBitPrev \|= 31;
5196	iBitPrev++;
5197	if (cBits <= (uint32_t)iBitPrev)
5198	return -1;
5199	}
5200
5201	/*
5202	* 32-bit aligned search, let ASMBitFirstClear do the dirty work.
5203	*/
5204	iBit = ASMBitFirstSet(&pau32Bitmap[iBitPrev / 32], cBits - iBitPrev);
5205	if (iBit >= 0)
5206	iBit += iBitPrev;
5207	return iBit;
5208	}
5209	#endif
5210
5211
5212	/**
5213	* Finds the first bit which is set in the given 32-bit integer.
5214	* Bits are numbered from 1 (least significant) to 32.
5215	*
5216	* @returns index [1..32] of the first set bit.
5217	* @returns 0 if all bits are cleared.
5218	* @param u32 Integer to search for set bits.
5219	* @remarks Similar to ffs() in BSD.
5220	*/
5221	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5222	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitFirstSetU32(uint32_t u32);
5223	#else
5224	DECLINLINE(unsigned) ASMBitFirstSetU32(uint32_t u32)
5225	{
5226	# if RT_INLINE_ASM_USES_INTRIN
5227	unsigned long iBit;
5228	if (_BitScanForward(&iBit, u32))
5229	iBit++;
5230	else
5231	iBit = 0;
5232	# elif RT_INLINE_ASM_GNU_STYLE
5233	uint32_t iBit;
5234	__asm__ __volatile__("bsf %1, %0\n\t"
5235	"jnz 1f\n\t"
5236	"xorl %0, %0\n\t"
5237	"jmp 2f\n"
5238	"1:\n\t"
5239	"incl %0\n"
5240	"2:\n\t"
5241	: "=r" (iBit)
5242	: "rm" (u32));
5243	# else
5244	uint32_t iBit;
5245	_asm
5246	{
5247	bsf eax, [u32]
5248	jnz found
5249	xor eax, eax
5250	jmp done
5251	found:
5252	inc eax
5253	done:
5254	mov [iBit], eax
5255	}
5256	# endif
5257	return iBit;
5258	}
5259	#endif
5260
5261
5262	/**
5263	* Finds the first bit which is set in the given 32-bit integer.
5264	* Bits are numbered from 1 (least significant) to 32.
5265	*
5266	* @returns index [1..32] of the first set bit.
5267	* @returns 0 if all bits are cleared.
5268	* @param i32 Integer to search for set bits.
5269	* @remark Similar to ffs() in BSD.
5270	*/
5271	DECLINLINE(unsigned) ASMBitFirstSetS32(int32_t i32)
5272	{
5273	return ASMBitFirstSetU32((uint32_t)i32);
5274	}
5275
5276
5277	/**
5278	* Finds the first bit which is set in the given 64-bit integer.
5279	*
5280	* Bits are numbered from 1 (least significant) to 64.
5281	*
5282	* @returns index [1..64] of the first set bit.
5283	* @returns 0 if all bits are cleared.
5284	* @param u64 Integer to search for set bits.
5285	* @remarks Similar to ffs() in BSD.
5286	*/
5287	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5288	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitFirstSetU64(uint64_t u64);
5289	#else
5290	DECLINLINE(unsigned) ASMBitFirstSetU64(uint64_t u64)
5291	{
5292	# if RT_INLINE_ASM_USES_INTRIN
5293	unsigned long iBit;
5294	# if ARCH_BITS == 64
5295	if (_BitScanForward64(&iBit, u64))
5296	iBit++;
5297	else
5298	iBit = 0;
5299	# else
5300	if (_BitScanForward(&iBit, (uint32_t)u64))
5301	iBit++;
5302	else if (_BitScanForward(&iBit, (uint32_t)(u64 >> 32)))
5303	iBit += 33;
5304	else
5305	iBit = 0;
5306	# endif
5307	# elif RT_INLINE_ASM_GNU_STYLE && ARCH_BITS == 64
5308	uint64_t iBit;
5309	__asm__ __volatile__("bsfq %1, %0\n\t"
5310	"jnz 1f\n\t"
5311	"xorl %k0, %k0\n\t"
5312	"jmp 2f\n"
5313	"1:\n\t"
5314	"incl %k0\n"
5315	"2:\n\t"
5316	: "=r" (iBit)
5317	: "rm" (u64));
5318	# else
5319	unsigned iBit = ASMBitFirstSetU32((uint32_t)u64);
5320	if (!iBit)
5321	{
5322	iBit = ASMBitFirstSetU32((uint32_t)(u64 >> 32));
5323	if (iBit)
5324	iBit += 32;
5325	}
5326	# endif
5327	return (unsigned)iBit;
5328	}
5329	#endif
5330
5331
5332	/**
5333	* Finds the first bit which is set in the given 16-bit integer.
5334	*
5335	* Bits are numbered from 1 (least significant) to 16.
5336	*
5337	* @returns index [1..16] of the first set bit.
5338	* @returns 0 if all bits are cleared.
5339	* @param u16 Integer to search for set bits.
5340	* @remarks For 16-bit bs3kit code.
5341	*/
5342	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5343	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitFirstSetU16(uint16_t u16);
5344	#else
5345	DECLINLINE(unsigned) ASMBitFirstSetU16(uint16_t u16)
5346	{
5347	return ASMBitFirstSetU32((uint32_t)u16);
5348	}
5349	#endif
5350
5351
5352	/**
5353	* Finds the last bit which is set in the given 32-bit integer.
5354	* Bits are numbered from 1 (least significant) to 32.
5355	*
5356	* @returns index [1..32] of the last set bit.
5357	* @returns 0 if all bits are cleared.
5358	* @param u32 Integer to search for set bits.
5359	* @remark Similar to fls() in BSD.
5360	*/
5361	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5362	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitLastSetU32(uint32_t u32);
5363	#else
5364	DECLINLINE(unsigned) ASMBitLastSetU32(uint32_t u32)
5365	{
5366	# if RT_INLINE_ASM_USES_INTRIN
5367	unsigned long iBit;
5368	if (_BitScanReverse(&iBit, u32))
5369	iBit++;
5370	else
5371	iBit = 0;
5372	# elif RT_INLINE_ASM_GNU_STYLE
5373	uint32_t iBit;
5374	__asm__ __volatile__("bsrl %1, %0\n\t"
5375	"jnz 1f\n\t"
5376	"xorl %0, %0\n\t"
5377	"jmp 2f\n"
5378	"1:\n\t"
5379	"incl %0\n"
5380	"2:\n\t"
5381	: "=r" (iBit)
5382	: "rm" (u32));
5383	# else
5384	uint32_t iBit;
5385	_asm
5386	{
5387	bsr eax, [u32]
5388	jnz found
5389	xor eax, eax
5390	jmp done
5391	found:
5392	inc eax
5393	done:
5394	mov [iBit], eax
5395	}
5396	# endif
5397	return iBit;
5398	}
5399	#endif
5400
5401
5402	/**
5403	* Finds the last bit which is set in the given 32-bit integer.
5404	* Bits are numbered from 1 (least significant) to 32.
5405	*
5406	* @returns index [1..32] of the last set bit.
5407	* @returns 0 if all bits are cleared.
5408	* @param i32 Integer to search for set bits.
5409	* @remark Similar to fls() in BSD.
5410	*/
5411	DECLINLINE(unsigned) ASMBitLastSetS32(int32_t i32)
5412	{
5413	return ASMBitLastSetU32((uint32_t)i32);
5414	}
5415
5416
5417	/**
5418	* Finds the last bit which is set in the given 64-bit integer.
5419	*
5420	* Bits are numbered from 1 (least significant) to 64.
5421	*
5422	* @returns index [1..64] of the last set bit.
5423	* @returns 0 if all bits are cleared.
5424	* @param u64 Integer to search for set bits.
5425	* @remark Similar to fls() in BSD.
5426	*/
5427	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5428	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitLastSetU64(uint64_t u64);
5429	#else
5430	DECLINLINE(unsigned) ASMBitLastSetU64(uint64_t u64)
5431	{
5432	# if RT_INLINE_ASM_USES_INTRIN
5433	unsigned long iBit;
5434	# if ARCH_BITS == 64
5435	if (_BitScanReverse64(&iBit, u64))
5436	iBit++;
5437	else
5438	iBit = 0;
5439	# else
5440	if (_BitScanReverse(&iBit, (uint32_t)(u64 >> 32)))
5441	iBit += 33;
5442	else if (_BitScanReverse(&iBit, (uint32_t)u64))
5443	iBit++;
5444	else
5445	iBit = 0;
5446	# endif
5447	# elif RT_INLINE_ASM_GNU_STYLE && ARCH_BITS == 64
5448	uint64_t iBit;
5449	__asm__ __volatile__("bsrq %1, %0\n\t"
5450	"jnz 1f\n\t"
5451	"xorl %k0, %k0\n\t"
5452	"jmp 2f\n"
5453	"1:\n\t"
5454	"incl %k0\n"
5455	"2:\n\t"
5456	: "=r" (iBit)
5457	: "rm" (u64));
5458	# else
5459	unsigned iBit = ASMBitLastSetU32((uint32_t)(u64 >> 32));
5460	if (iBit)
5461	iBit += 32;
5462	else
5463	iBit = ASMBitLastSetU32((uint32_t)u64);
5464	#endif
5465	return (unsigned)iBit;
5466	}
5467	#endif
5468
5469
5470	/**
5471	* Finds the last bit which is set in the given 16-bit integer.
5472	*
5473	* Bits are numbered from 1 (least significant) to 16.
5474	*
5475	* @returns index [1..16] of the last set bit.
5476	* @returns 0 if all bits are cleared.
5477	* @param u16 Integer to search for set bits.
5478	* @remarks For 16-bit bs3kit code.
5479	*/
5480	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5481	RT_ASM_DECL_PRAGMA_WATCOM_386(unsigned) ASMBitLastSetU16(uint16_t u16);
5482	#else
5483	DECLINLINE(unsigned) ASMBitLastSetU16(uint16_t u16)
5484	{
5485	return ASMBitLastSetU32((uint32_t)u16);
5486	}
5487	#endif
5488
5489
5490	/**
5491	* Reverse the byte order of the given 16-bit integer.
5492	*
5493	* @returns Revert
5494	* @param u16 16-bit integer value.
5495	*/
5496	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5497	RT_ASM_DECL_PRAGMA_WATCOM(uint16_t) ASMByteSwapU16(uint16_t u16);
5498	#else
5499	DECLINLINE(uint16_t) ASMByteSwapU16(uint16_t u16)
5500	{
5501	# if RT_INLINE_ASM_USES_INTRIN
5502	u16 = _byteswap_ushort(u16);
5503	# elif RT_INLINE_ASM_GNU_STYLE
5504	__asm__ ("rorw $8, %0" : "=r" (u16) : "0" (u16));
5505	# else
5506	_asm
5507	{
5508	mov ax, [u16]
5509	ror ax, 8
5510	mov [u16], ax
5511	}
5512	# endif
5513	return u16;
5514	}
5515	#endif
5516
5517
5518	/**
5519	* Reverse the byte order of the given 32-bit integer.
5520	*
5521	* @returns Revert
5522	* @param u32 32-bit integer value.
5523	*/
5524	#if RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN
5525	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMByteSwapU32(uint32_t u32);
5526	#else
5527	DECLINLINE(uint32_t) ASMByteSwapU32(uint32_t u32)
5528	{
5529	# if RT_INLINE_ASM_USES_INTRIN
5530	u32 = _byteswap_ulong(u32);
5531	# elif RT_INLINE_ASM_GNU_STYLE
5532	__asm__ ("bswapl %0" : "=r" (u32) : "0" (u32));
5533	# else
5534	_asm
5535	{
5536	mov eax, [u32]
5537	bswap eax
5538	mov [u32], eax
5539	}
5540	# endif
5541	return u32;
5542	}
5543	#endif
5544
5545
5546	/**
5547	* Reverse the byte order of the given 64-bit integer.
5548	*
5549	* @returns Revert
5550	* @param u64 64-bit integer value.
5551	*/
5552	DECLINLINE(uint64_t) ASMByteSwapU64(uint64_t u64)
5553	{
5554	#if defined(RT_ARCH_AMD64) && RT_INLINE_ASM_USES_INTRIN
5555	u64 = _byteswap_uint64(u64);
5556	#else
5557	u64 = (uint64_t)ASMByteSwapU32((uint32_t)u64) << 32
5558	\| (uint64_t)ASMByteSwapU32((uint32_t)(u64 >> 32));
5559	#endif
5560	return u64;
5561	}
5562
5563
5564	/**
5565	* Rotate 32-bit unsigned value to the left by @a cShift.
5566	*
5567	* @returns Rotated value.
5568	* @param u32 The value to rotate.
5569	* @param cShift How many bits to rotate by.
5570	*/
5571	#ifdef __WATCOMC__
5572	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMRotateLeftU32(uint32_t u32, unsigned cShift);
5573	#else
5574	DECLINLINE(uint32_t) ASMRotateLeftU32(uint32_t u32, uint32_t cShift)
5575	{
5576	# if RT_INLINE_ASM_USES_INTRIN
5577	return _rotl(u32, cShift);
5578	# elif RT_INLINE_ASM_GNU_STYLE && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86))
5579	__asm__ __volatile__("roll %b1, %0" : "=g" (u32) : "Ic" (cShift), "0" (u32));
5580	return u32;
5581	# else
5582	cShift &= 31;
5583	return (u32 << cShift) \| (u32 >> (32 - cShift));
5584	# endif
5585	}
5586	#endif
5587
5588
5589	/**
5590	* Rotate 32-bit unsigned value to the right by @a cShift.
5591	*
5592	* @returns Rotated value.
5593	* @param u32 The value to rotate.
5594	* @param cShift How many bits to rotate by.
5595	*/
5596	#ifdef __WATCOMC__
5597	RT_ASM_DECL_PRAGMA_WATCOM(uint32_t) ASMRotateRightU32(uint32_t u32, unsigned cShift);
5598	#else
5599	DECLINLINE(uint32_t) ASMRotateRightU32(uint32_t u32, uint32_t cShift)
5600	{
5601	# if RT_INLINE_ASM_USES_INTRIN
5602	return _rotr(u32, cShift);
5603	# elif RT_INLINE_ASM_GNU_STYLE && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86))
5604	__asm__ __volatile__("rorl %b1, %0" : "=g" (u32) : "Ic" (cShift), "0" (u32));
5605	return u32;
5606	# else
5607	cShift &= 31;
5608	return (u32 >> cShift) \| (u32 << (32 - cShift));
5609	# endif
5610	}
5611	#endif
5612
5613
5614	/**
5615	* Rotate 64-bit unsigned value to the left by @a cShift.
5616	*
5617	* @returns Rotated value.
5618	* @param u64 The value to rotate.
5619	* @param cShift How many bits to rotate by.
5620	*/
5621	DECLINLINE(uint64_t) ASMRotateLeftU64(uint64_t u64, uint32_t cShift)
5622	{
5623	#if RT_INLINE_ASM_USES_INTRIN
5624	return _rotl64(u64, cShift);
5625	#elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
5626	__asm__ __volatile__("rolq %b1, %0" : "=g" (u64) : "Jc" (cShift), "0" (u64));
5627	return u64;
5628	#elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_X86)
5629	uint32_t uSpill;
5630	__asm__ __volatile__("testb $0x20, %%cl\n\t" /* if (cShift >= 0x20) { swap(u64.hi, u64lo); cShift -= 0x20; } */
5631	"jz 1f\n\t"
5632	"xchgl %%eax, %%edx\n\t"
5633	"1:\n\t"
5634	"andb $0x1f, %%cl\n\t" /* if (cShift & 0x1f) { */
5635	"jz 2f\n\t"
5636	"movl %%edx, %2\n\t" /* save the hi value in %3. */
5637	"shldl %%cl,%%eax,%%edx\n\t" /* shift the hi value left, feeding MSBits from the low value. */
5638	"shldl %%cl,%2,%%eax\n\t" /* shift the lo value left, feeding MSBits from the saved hi value. */
5639	"2:\n\t" /* } */
5640	: "=A" (u64), "=c" (cShift), "=r" (uSpill)
5641	: "0" (u64),
5642	"1" (cShift));
5643	return u64;
5644	#else
5645	cShift &= 63;
5646	return (u64 << cShift) \| (u64 >> (64 - cShift));
5647	#endif
5648	}
5649
5650
5651	/**
5652	* Rotate 64-bit unsigned value to the right by @a cShift.
5653	*
5654	* @returns Rotated value.
5655	* @param u64 The value to rotate.
5656	* @param cShift How many bits to rotate by.
5657	*/
5658	DECLINLINE(uint64_t) ASMRotateRightU64(uint64_t u64, uint32_t cShift)
5659	{
5660	#if RT_INLINE_ASM_USES_INTRIN
5661	return _rotr64(u64, cShift);
5662	#elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_AMD64)
5663	__asm__ __volatile__("rorq %b1, %0" : "=g" (u64) : "Jc" (cShift), "0" (u64));
5664	return u64;
5665	#elif RT_INLINE_ASM_GNU_STYLE && defined(RT_ARCH_X86)
5666	uint32_t uSpill;
5667	__asm__ __volatile__("testb $0x20, %%cl\n\t" /* if (cShift >= 0x20) { swap(u64.hi, u64lo); cShift -= 0x20; } */
5668	"jz 1f\n\t"
5669	"xchgl %%eax, %%edx\n\t"
5670	"1:\n\t"
5671	"andb $0x1f, %%cl\n\t" /* if (cShift & 0x1f) { */
5672	"jz 2f\n\t"
5673	"movl %%edx, %2\n\t" /* save the hi value in %3. */
5674	"shrdl %%cl,%%eax,%%edx\n\t" /* shift the hi value right, feeding LSBits from the low value. */
5675	"shrdl %%cl,%2,%%eax\n\t" /* shift the lo value right, feeding LSBits from the saved hi value. */
5676	"2:\n\t" /* } */
5677	: "=A" (u64), "=c" (cShift), "=r" (uSpill)
5678	: "0" (u64),
5679	"1" (cShift));
5680	return u64;
5681	#else
5682	cShift &= 63;
5683	return (u64 >> cShift) \| (u64 << (64 - cShift));
5684	#endif
5685	}
5686
5687	/** @} */
5688
5689
5690	/** @} */
5691
5692	/*
5693	* Include #pragma aux definitions for Watcom C/C++.
5694	*/
5695	#if defined(__WATCOMC__) && ARCH_BITS == 16 && defined(RT_ARCH_X86)
5696	# define IPRT_ASM_WATCOM_X86_16_WITH_PRAGMAS
5697	# undef IPRT_INCLUDED_asm_watcom_x86_16_h
5698	# include "asm-watcom-x86-16.h"
5699	#elif defined(__WATCOMC__) && ARCH_BITS == 32 && defined(RT_ARCH_X86)
5700	# define IPRT_ASM_WATCOM_X86_32_WITH_PRAGMAS
5701	# undef IPRT_INCLUDED_asm_watcom_x86_32_h
5702	# include "asm-watcom-x86-32.h"
5703	#endif
5704
5705	#endif /* !IPRT_INCLUDED_asm_h */
5706

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/include/iprt/asm.h@ 81025

Download in other formats: