asm-mem.h

Last change on this file was 106559, checked in by vboxsync, 5 weeks ago
IPRT: Sort out export/import trouble with 'inline-assembly' functions implemented in C++ on windows by introducing a RT_DECL_ASM() macro that combines RTDECL with DECLASM. jiraref:VBP-1171
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 10.6 KB

Line
1	/** @file
2	* IPRT - Assembly Memory Functions.
3	*/
4
5	/*
6	* Copyright (C) 2006-2024 Oracle and/or its affiliates.
7	*
8	* This file is part of VirtualBox base platform packages, as
9	* available from https://www.virtualbox.org.
10	*
11	* This program is free software; you can redistribute it and/or
12	* modify it under the terms of the GNU General Public License
13	* as published by the Free Software Foundation, in version 3 of the
14	* License.
15	*
16	* This program is distributed in the hope that it will be useful, but
17	* WITHOUT ANY WARRANTY; without even the implied warranty of
18	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19	* General Public License for more details.
20	*
21	* You should have received a copy of the GNU General Public License
22	* along with this program; if not, see <https://www.gnu.org/licenses>.
23	*
24	* The contents of this file may alternatively be used under the terms
25	* of the Common Development and Distribution License Version 1.0
26	* (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
27	* in the VirtualBox distribution, in which case the provisions of the
28	* CDDL are applicable instead of those of the GPL.
29	*
30	* You may elect to license modified versions of this file under the
31	* terms and conditions of either the GPL or the CDDL or both.
32	*
33	* SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
34	*/
35
36	#ifndef IPRT_INCLUDED_asm_mem_h
37	#define IPRT_INCLUDED_asm_mem_h
38	#ifndef RT_WITHOUT_PRAGMA_ONCE
39	# pragma once
40	#endif
41
42	#include <iprt/cdefs.h>
43	#include <iprt/types.h>
44	#include <iprt/assert.h>
45
46	#if defined(_MSC_VER) && RT_INLINE_ASM_USES_INTRIN
47	/* Emit the intrinsics at all optimization levels. */
48	# include <iprt/sanitized/intrin.h>
49	# if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
50	# pragma intrinsic(__cpuid)
51	# pragma intrinsic(__stosd)
52	# pragma intrinsic(__stosw)
53	# pragma intrinsic(__stosb)
54	# ifdef RT_ARCH_AMD64
55	# pragma intrinsic(__stosq)
56	# endif
57	# elif defined(RT_ARCH_ARM64) \|\| defined(RT_ARCH_ARM32)
58	# pragma intrinsic(__iso_volatile_load8)
59	# endif
60	#endif
61
62
63	/*
64	* Undefine all symbols we have Watcom C/C++ #pragma aux'es for.
65	*/
66	#if defined(__WATCOMC__) && ARCH_BITS == 16 && defined(RT_ARCH_X86)
67	# include "asm-mem-watcom-x86-16.h"
68	#elif defined(__WATCOMC__) && ARCH_BITS == 32 && defined(RT_ARCH_X86)
69	# include "asm-mem-watcom-x86-32.h"
70	#endif
71
72
73
74	/** @defgroup grp_rt_asm_mem ASM - Memory Assembly Routines
75	* @ingroup grp_rt_asm
76	* @{
77	*/
78
79	/**
80	* Zeros a memory block with a 32-bit aligned size.
81	*
82	* @param pv Pointer to the memory block.
83	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
84	*/
85	#if (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \|\| (!defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86))
86	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMMemZero32(volatile void RT_FAR *pv, size_t cb) RT_NOTHROW_PROTO;
87	#else
88	DECLINLINE(void) ASMMemZero32(volatile void RT_FAR *pv, size_t cb) RT_NOTHROW_DEF
89	{
90	# if RT_INLINE_ASM_USES_INTRIN
91	# ifdef RT_ARCH_AMD64
92	if (!(cb & 7))
93	__stosq((unsigned __int64 RT_FAR *)pv, 0, cb / 8);
94	else
95	# endif
96	__stosd((unsigned long RT_FAR *)pv, 0, cb / 4);
97
98	# elif RT_INLINE_ASM_GNU_STYLE
99	__asm__ __volatile__("rep stosl"
100	: "=D" (pv),
101	"=c" (cb)
102	: "0" (pv),
103	"1" (cb >> 2),
104	"a" (0)
105	: "memory");
106	# else
107	__asm
108	{
109	xor eax, eax
110	# ifdef RT_ARCH_AMD64
111	mov rcx, [cb]
112	shr rcx, 2
113	mov rdi, [pv]
114	# else
115	mov ecx, [cb]
116	shr ecx, 2
117	mov edi, [pv]
118	# endif
119	rep stosd
120	}
121	# endif
122	}
123	#endif
124
125
126	/**
127	* Fills a memory block with a 32-bit aligned size.
128	*
129	* @param pv Pointer to the memory block.
130	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
131	* @param u32 The value to fill with.
132	*/
133	#if (RT_INLINE_ASM_EXTERNAL && !RT_INLINE_ASM_USES_INTRIN) \|\| (!defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86))
134	RT_ASM_DECL_PRAGMA_WATCOM(void) ASMMemFill32(volatile void RT_FAR *pv, size_t cb, uint32_t u32) RT_NOTHROW_PROTO;
135	#else
136	DECLINLINE(void) ASMMemFill32(volatile void RT_FAR *pv, size_t cb, uint32_t u32) RT_NOTHROW_DEF
137	{
138	Assert(!(cb & 3));
139	Assert(cb > 0);
140	# if RT_INLINE_ASM_USES_INTRIN
141	# ifdef RT_ARCH_AMD64
142	if (!(cb & 7))
143	__stosq((unsigned __int64 RT_FAR *)pv, RT_MAKE_U64(u32, u32), cb / 8);
144	else
145	# endif
146	__stosd((unsigned long RT_FAR *)pv, u32, cb / 4);
147
148	# elif RT_INLINE_ASM_GNU_STYLE
149	__asm__ __volatile__("rep stosl"
150	: "=D" (pv),
151	"=c" (cb)
152	: "0" (pv),
153	"1" (cb >> 2),
154	"a" (u32)
155	: "memory");
156	# else
157	__asm
158	{
159	# ifdef RT_ARCH_AMD64
160	mov rcx, [cb]
161	shr rcx, 2
162	mov rdi, [pv]
163	# else
164	mov ecx, [cb]
165	shr ecx, 2
166	mov edi, [pv]
167	# endif
168	mov eax, [u32]
169	rep stosd
170	}
171	# endif
172	}
173	#endif
174
175
176	/**
177	* Checks if a memory block is all zeros.
178	*
179	* @returns Pointer to the first non-zero byte.
180	* @returns NULL if all zero.
181	*
182	* @param pv Pointer to the memory block.
183	* @param cb Number of bytes in the block.
184	*/
185	#if !defined(RDESKTOP) && (!defined(RT_OS_LINUX) \|\| !defined(__KERNEL__))
186	RT_DECL_ASM(void RT_FAR ) ASMMemFirstNonZero(void const RT_FAR pv, size_t cb) RT_NOTHROW_PROTO;
187	#else
188	DECLINLINE(void RT_FAR ) ASMMemFirstNonZero(void const RT_FAR pv, size_t cb) RT_NOTHROW_DEF
189	{
190	/** @todo replace with ASMMemFirstNonZero-generic.cpp in kernel modules. */
191	uint8_t const pb = (uint8_t const RT_FAR )pv;
192	for (; cb; cb--, pb++)
193	if (RT_LIKELY(*pb == 0))
194	{ /* likely */ }
195	else
196	return (void RT_FAR *)pb;
197	return NULL;
198	}
199	#endif
200
201
202	/**
203	* Checks if a memory block is all zeros.
204	*
205	* @returns true if zero, false if not.
206	*
207	* @param pv Pointer to the memory block.
208	* @param cb Number of bytes in the block.
209	*
210	* @sa ASMMemFirstNonZero
211	*/
212	DECLINLINE(bool) ASMMemIsZero(void const RT_FAR *pv, size_t cb) RT_NOTHROW_DEF
213	{
214	return ASMMemFirstNonZero(pv, cb) == NULL;
215	}
216
217
218	/**
219	* Checks if a memory block is filled with the specified byte, returning the
220	* first mismatch.
221	*
222	* This is sort of an inverted memchr.
223	*
224	* @returns Pointer to the byte which doesn't equal u8.
225	* @returns NULL if all equal to u8.
226	*
227	* @param pv Pointer to the memory block.
228	* @param cb Number of bytes in the block.
229	* @param u8 The value it's supposed to be filled with.
230	*
231	* @remarks No alignment requirements.
232	*/
233	#if (!defined(RT_OS_LINUX) \|\| !defined(__KERNEL__)) \
234	&& (!defined(RT_OS_FREEBSD) \|\| !defined(_KERNEL))
235	RT_DECL_ASM(void ) ASMMemFirstMismatchingU8(void const RT_FAR pv, size_t cb, uint8_t u8) RT_NOTHROW_PROTO;
236	#else
237	DECLINLINE(void ) ASMMemFirstMismatchingU8(void const RT_FAR pv, size_t cb, uint8_t u8) RT_NOTHROW_DEF
238	{
239	/** @todo replace with ASMMemFirstMismatchingU8-generic.cpp in kernel modules. */
240	uint8_t const pb = (uint8_t const RT_FAR )pv;
241	for (; cb; cb--, pb++)
242	if (RT_LIKELY(*pb == u8))
243	{ /* likely */ }
244	else
245	return (void *)pb;
246	return NULL;
247	}
248	#endif
249
250
251	/**
252	* Checks if a memory block is filled with the specified byte.
253	*
254	* @returns true if all matching, false if not.
255	*
256	* @param pv Pointer to the memory block.
257	* @param cb Number of bytes in the block.
258	* @param u8 The value it's supposed to be filled with.
259	*
260	* @remarks No alignment requirements.
261	*/
262	DECLINLINE(bool) ASMMemIsAllU8(void const RT_FAR *pv, size_t cb, uint8_t u8) RT_NOTHROW_DEF
263	{
264	return ASMMemFirstMismatchingU8(pv, cb, u8) == NULL;
265	}
266
267
268	/**
269	* Checks if a memory block is filled with the specified 32-bit value.
270	*
271	* This is a sort of inverted memchr.
272	*
273	* @returns Pointer to the first value which doesn't equal u32.
274	* @returns NULL if all equal to u32.
275	*
276	* @param pv Pointer to the memory block.
277	* @param cb Number of bytes in the block. This MUST be aligned on 32-bit!
278	* @param u32 The value it's supposed to be filled with.
279	*/
280	DECLINLINE(uint32_t RT_FAR ) ASMMemFirstMismatchingU32(void const RT_FAR pv, size_t cb, uint32_t u32) RT_NOTHROW_DEF
281	{
282	/** @todo rewrite this in inline assembly? */
283	uint32_t const RT_FAR pu32 = (uint32_t const RT_FAR )pv;
284	for (; cb; cb -= 4, pu32++)
285	if (RT_LIKELY(*pu32 == u32))
286	{ /* likely */ }
287	else
288	return (uint32_t RT_FAR *)pu32;
289	return NULL;
290	}
291
292
293	/**
294	* Probes a byte pointer for read access.
295	*
296	* While the function will not fault if the byte is not read accessible,
297	* the idea is to do this in a safe place like before acquiring locks
298	* and such like.
299	*
300	* Also, this functions guarantees that an eager compiler is not going
301	* to optimize the probing away.
302	*
303	* @param pvByte Pointer to the byte.
304	*/
305	#if RT_INLINE_ASM_EXTERNAL_TMP_ARM
306	RT_ASM_DECL_PRAGMA_WATCOM(uint8_t) ASMProbeReadByte(const void RT_FAR *pvByte) RT_NOTHROW_PROTO;
307	#else
308	DECLINLINE(uint8_t) ASMProbeReadByte(const void RT_FAR *pvByte) RT_NOTHROW_DEF
309	{
310	# if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
311	uint8_t u8;
312	# if RT_INLINE_ASM_GNU_STYLE
313	__asm__ __volatile__("movb %1, %0\n\t"
314	: "=q" (u8)
315	: "m" ((const uint8_t )pvByte));
316	# else
317	__asm
318	{
319	# ifdef RT_ARCH_AMD64
320	mov rax, [pvByte]
321	mov al, [rax]
322	# else
323	mov eax, [pvByte]
324	mov al, [eax]
325	# endif
326	mov [u8], al
327	}
328	# endif
329	return u8;
330
331	# elif defined(RT_ARCH_ARM64) \|\| defined(RT_ARCH_ARM32)
332	# if RT_INLINE_ASM_USES_INTRIN
333	return (uint8_t)__iso_volatile_load8((volatile const char *)pvByte);
334	# else
335	uint32_t u32;
336	__asm__ __volatile__("Lstart_ASMProbeReadByte_%=:\n\t"
337	# if defined(RT_ARCH_ARM64)
338	"ldxrb %w[uDst], %[pMem]\n\t"
339	# else
340	"ldrexb %[uDst], %[pMem]\n\t"
341	# endif
342	: [uDst] "=&r" (u32)
343	: [pMem] "Q" ((uint8_t const )pvByte));
344	return (uint8_t)u32;
345	# endif
346
347	# else
348	# error "Port me"
349	# endif
350	}
351	#endif
352
353	/** @} */
354
355	/*
356	* Include #pragma aux definitions for Watcom C/C++.
357	*/
358	#if defined(__WATCOMC__) && ARCH_BITS == 16 && defined(RT_ARCH_X86)
359	# define IPRT_ASM_WATCOM_X86_16_WITH_PRAGMAS
360	# undef IPRT_INCLUDED_asm_mem_watcom_x86_16_h
361	# include "asm-mem-watcom-x86-16.h"
362	#elif defined(__WATCOMC__) && ARCH_BITS == 32 && defined(RT_ARCH_X86)
363	# define IPRT_ASM_WATCOM_X86_32_WITH_PRAGMAS
364	# undef IPRT_INCLUDED_asm_mem_watcom_x86_32_h
365	# include "asm-mem-watcom-x86-32.h"
366	#endif
367
368	#endif /* !IPRT_INCLUDED_asm_mem_h */
369

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

source: vbox/trunk/include/iprt/asm-mem.h

Download in other formats: