summaryrefslogtreecommitdiff
path: root/indra/llcommon/llmemory.h
blob: f04ae5f5cbe8b7419a789e090b7cdaed49cc5e5b (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
/**
 * @file llmemory.h
 * @brief Memory allocation/deallocation header-stuff goes here.
 *
 * $LicenseInfo:firstyear=2002&license=viewerlgpl$
 * Second Life Viewer Source Code
 * Copyright (C) 2010, Linden Research, Inc.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation;
 * version 2.1 of the License only.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
 *
 * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
 * $/LicenseInfo$
 */
#ifndef LLMEMORY_H
#define LLMEMORY_H

#include "linden_common.h"
#include "llunits.h"
#include "stdtypes.h"
#if !LL_WINDOWS
#include <stdint.h>
#endif

class LLMutex ;

#if LL_WINDOWS && LL_DEBUG
#define LL_CHECK_MEMORY llassert(_CrtCheckMemory());
#else
#define LL_CHECK_MEMORY 
#endif


#if LL_WINDOWS
#define LL_ALIGN_OF __alignof
#else
#define LL_ALIGN_OF __align_of__
#endif

#if LL_WINDOWS
#define LL_DEFAULT_HEAP_ALIGN 8
#elif LL_DARWIN
#define LL_DEFAULT_HEAP_ALIGN 16
#elif LL_LINUX
#define LL_DEFAULT_HEAP_ALIGN 8
#endif


LL_COMMON_API void ll_assert_aligned_func(uintptr_t ptr,U32 alignment);

#ifdef SHOW_ASSERT
// This is incredibly expensive - in profiling Windows RWD builds, 30%
// of CPU time was in aligment checks.
//#define ASSERT_ALIGNMENT
#endif

#ifdef ASSERT_ALIGNMENT
#define ll_assert_aligned(ptr,alignment) ll_assert_aligned_func(uintptr_t(ptr),((U32)alignment))
#else
#define ll_assert_aligned(ptr,alignment)
#endif

#include <xmmintrin.h>

template <typename T> T* LL_NEXT_ALIGNED_ADDRESS(T* address) 
{ 
	return reinterpret_cast<T*>(
		(uintptr_t(address) + 0xF) & ~0xF);
}

template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) 
{ 
	return reinterpret_cast<T*>(
		(uintptr_t(address) + 0x3F) & ~0x3F);
}

#if LL_LINUX || LL_DARWIN

#define			LL_ALIGN_PREFIX(x)
#define			LL_ALIGN_POSTFIX(x)		__attribute__((aligned(x)))

#elif LL_WINDOWS

#define			LL_ALIGN_PREFIX(x)		__declspec(align(x))
#define			LL_ALIGN_POSTFIX(x)

#else
#error "LL_ALIGN_PREFIX and LL_ALIGN_POSTFIX undefined"
#endif

#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16)

//------------------------------------------------------------------------------------------------
//------------------------------------------------------------------------------------------------
	// for enable buffer overrun detection predefine LL_DEBUG_BUFFER_OVERRUN in current library
	// change preprocessor code to: #if 1 && defined(LL_WINDOWS)

#if 0 && defined(LL_WINDOWS)
	void* ll_aligned_malloc_fallback( size_t size, int align );
	void ll_aligned_free_fallback( void* ptr );
//------------------------------------------------------------------------------------------------
#else
	inline void* ll_aligned_malloc_fallback( size_t size, int align )
	{
	#if defined(LL_WINDOWS)
		return _aligned_malloc(size, align);
	#else
        char* aligned = NULL;
		void* mem = malloc( size + (align - 1) + sizeof(void*) );
        if (mem)
        {
            aligned = ((char*)mem) + sizeof(void*);
            aligned += align - ((uintptr_t)aligned & (align - 1));

            ((void**)aligned)[-1] = mem;
        }
		return aligned;
	#endif
	}

	inline void ll_aligned_free_fallback( void* ptr )
	{
	#if defined(LL_WINDOWS)
		_aligned_free(ptr);
	#else
		if (ptr)
		{
			free( ((void**)ptr)[-1] );
		}
	#endif
	}
#endif
//------------------------------------------------------------------------------------------------
//------------------------------------------------------------------------------------------------

inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed with ll_aligned_free_16().
{
#if defined(LL_WINDOWS)
	return _aligned_malloc(size, 16);
#elif defined(LL_DARWIN)
	return malloc(size); // default osx malloc is 16 byte aligned.
#else
	void *rtn;
	if (LL_LIKELY(0 == posix_memalign(&rtn, 16, size)))
		return rtn;
	else // bad alignment requested, or out of memory
		return NULL;
#endif
}

inline void ll_aligned_free_16(void *p)
{
#if defined(LL_WINDOWS)
	_aligned_free(p);
#elif defined(LL_DARWIN)
	return free(p);
#else
	free(p); // posix_memalign() is compatible with heap deallocator
#endif
}

inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // returned hunk MUST be freed with ll_aligned_free_16().
{
#if defined(LL_WINDOWS)
	return _aligned_realloc(ptr, size, 16);
#elif defined(LL_DARWIN)
	return realloc(ptr,size); // default osx malloc is 16 byte aligned.
#else
	//FIXME: memcpy is SLOW
	void* ret = ll_aligned_malloc_16(size);
	if (ptr)
	{
		if (ret)
		{
			// Only copy the size of the smallest memory block to avoid memory corruption.
			memcpy(ret, ptr, llmin(old_size, size));
		}
		ll_aligned_free_16(ptr);
	}
	return ret;
#endif
}

inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32().
{
#if defined(LL_WINDOWS)
	return _aligned_malloc(size, 32);
#elif defined(LL_DARWIN)
	return ll_aligned_malloc_fallback( size, 32 );
#else
	void *rtn;
	if (LL_LIKELY(0 == posix_memalign(&rtn, 32, size)))
		return rtn;
	else // bad alignment requested, or out of memory
		return NULL;
#endif
}

inline void ll_aligned_free_32(void *p)
{
#if defined(LL_WINDOWS)
	_aligned_free(p);
#elif defined(LL_DARWIN)
	ll_aligned_free_fallback( p );
#else
	free(p); // posix_memalign() is compatible with heap deallocator
#endif
}

// general purpose dispatch functions that are forced inline so they can compile down to a single call
template<size_t ALIGNMENT>
LL_FORCE_INLINE void* ll_aligned_malloc(size_t size)
{
	if (LL_DEFAULT_HEAP_ALIGN % ALIGNMENT == 0)
	{
		return malloc(size);
	}
	else if (ALIGNMENT == 16)
	{
		return ll_aligned_malloc_16(size);
	}
	else if (ALIGNMENT == 32)
	{
		return ll_aligned_malloc_32(size);
	}
	else
	{
		return ll_aligned_malloc_fallback(size, ALIGNMENT);
	}
}

template<size_t ALIGNMENT>
LL_FORCE_INLINE void ll_aligned_free(void* ptr)
{
	if (ALIGNMENT == LL_DEFAULT_HEAP_ALIGN)
	{
		free(ptr);
	}
	else if (ALIGNMENT == 16)
	{
		ll_aligned_free_16(ptr);
	}
	else if (ALIGNMENT == 32)
	{
		return ll_aligned_free_32(ptr);
	}
	else
	{
		return ll_aligned_free_fallback(ptr);
	}
}

// Copy words 16-byte blocks from src to dst. Source and destination MUST NOT OVERLAP. 
// Source and dest must be 16-byte aligned and size must be multiple of 16.
//
inline void ll_memcpy_nonaliased_aligned_16(char* __restrict dst, const char* __restrict src, size_t bytes)
{
	assert(src != NULL);
	assert(dst != NULL);
	assert(bytes > 0);
	assert((bytes % sizeof(F32))== 0); 
	ll_assert_aligned(src,16);
	ll_assert_aligned(dst,16);

	assert((src < dst) ? ((src + bytes) <= dst) : ((dst + bytes) <= src));
	assert(bytes%16==0);

	char* end = dst + bytes;

	if (bytes > 64)
	{

		// Find start of 64b aligned area within block
		//
		void* begin_64 = LL_NEXT_ALIGNED_ADDRESS_64(dst);
		
		//at least 64 bytes before the end of the destination, switch to 16 byte copies
		void* end_64 = end-64;
	
		// Prefetch the head of the 64b area now
		//
		_mm_prefetch((char*)begin_64, _MM_HINT_NTA);
		_mm_prefetch((char*)begin_64 + 64, _MM_HINT_NTA);
		_mm_prefetch((char*)begin_64 + 128, _MM_HINT_NTA);
		_mm_prefetch((char*)begin_64 + 192, _MM_HINT_NTA);
	
		// Copy 16b chunks until we're 64b aligned
		//
		while (dst < begin_64)
		{

			_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
			dst += 16;
			src += 16;
		}
	
		// Copy 64b chunks up to your tail
		//
		// might be good to shmoo the 512b prefetch offset
		// (characterize performance for various values)
		//
		while (dst < end_64)
		{
			_mm_prefetch((char*)src + 512, _MM_HINT_NTA);
			_mm_prefetch((char*)dst + 512, _MM_HINT_NTA);
			_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
			_mm_store_ps((F32*)(dst + 16), _mm_load_ps((F32*)(src + 16)));
			_mm_store_ps((F32*)(dst + 32), _mm_load_ps((F32*)(src + 32)));
			_mm_store_ps((F32*)(dst + 48), _mm_load_ps((F32*)(src + 48)));
			dst += 64;
			src += 64;
		}
	}

	// Copy remainder 16b tail chunks (or ALL 16b chunks for sub-64b copies)
	//
	while (dst < end)
	{
		_mm_store_ps((F32*)dst, _mm_load_ps((F32*)src));
		dst += 16;
		src += 16;
	}
}

#ifndef __DEBUG_PRIVATE_MEM__
#define __DEBUG_PRIVATE_MEM__  0
#endif

class LL_COMMON_API LLMemory
{
public:
	// Return the resident set size of the current process, in bytes.
	// Return value is zero if not known.
	static U64 getCurrentRSS();
	static void* tryToAlloc(void* address, U32 size);
	static void initMaxHeapSizeGB(F32Gigabytes max_heap_size, BOOL prevent_heap_failure);
	static void updateMemoryInfo() ;
	static void logMemoryInfo(BOOL update = FALSE);
	static bool isMemoryPoolLow();

	static U32Kilobytes getAvailableMemKB() ;
	static U32Kilobytes getMaxMemKB() ;
	static U32Kilobytes getAllocatedMemKB() ;
private:
	static U32Kilobytes sAvailPhysicalMemInKB ;
	static U32Kilobytes sMaxPhysicalMemInKB ;
	static U32Kilobytes sAllocatedMemInKB;
	static U32Kilobytes sAllocatedPageSizeInKB ;

	static U32Kilobytes sMaxHeapSizeInKB;
	static BOOL sEnableMemoryFailurePrevention;
};

// LLRefCount moved to llrefcount.h

// LLPointer moved to llpointer.h

// LLSafeHandle moved to llsafehandle.h

// LLSingleton moved to llsingleton.h




#endif