/usr/include/fflas-ffpack/utils/fflas_memory.h is in fflas-ffpack-common 2.2.2-5.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 | /* -*- mode: C++; tab-width: 8; indent-tabs-mode: t; c-basic-offset: 8 -*- */
// vim:sts=8:sw=8:ts=8:noet:sr:cino=>s,f0,{0,g0,(0,\:0,t0,+0,=s
/* fflas/fflas_memory.h
* Copyright (C) 2014 fflas-ffpack group
*
* Written by Clement Pernet <Clement.Pernet@imag.fr>
*
* The cache size detection has been copied from the Eigen library,
* a lightweight C++ template library for linear algebra, licensed under
* the Mozilla
* Public License v. 2.0. If a copy of the MPL was not distributed
* with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
* Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
* Copyright (C) 2008-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
* Copyright (C) 2009 Kenneth Riddile <kfriddile@yahoo.com>
* Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
* Copyright (C) 2010 Thomas Capricelli <orzel@freehackers.org>
*
*
* ========LICENCE========
* This file is part of the library FFLAS-FFPACK.
*
* FFLAS-FFPACK 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; either
* version 2.1 of the License, or (at your option) any later version.
*
* 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
* ========LICENCE========
*.
*/
#ifndef __FFLASFFPACK_memory_H
#define __FFLASFFPACK_memory_H
#include "fflas-ffpack/utils/align-allocator.h"
#include <givaro/givinteger.h>
namespace FFLAS{
template<class Element>
inline bool alignable() {
return true ;
}
// BB : segfault in Givaro::Integer::logcpy otherwise
template<>
inline bool alignable<Givaro::Integer*>() {
return false;
}
template<class Field>
inline typename Field::Element_ptr fflas_new (const Field& F, const size_t m, const size_t n, const Alignment align = Alignment::DEFAULT)
{
if (alignable<typename Field::Element_ptr>() ) {
return malloc_align<typename Field::Element>(m*n, align);
}
else {
return new typename Field::Element[m*n];
}
}
template<class Element >
inline Element* fflas_new (const size_t m, const Alignment align = Alignment::DEFAULT)
{
if (alignable<Element*>() ) {
return malloc_align<Element>(m, align);
}
else {
return new Element[m];
}
}
template<class Element_ptr>
inline void fflas_delete(Element_ptr A)
{
if (alignable<Element_ptr>() )
free(A);
else
delete[] A;
}
template<class Ptr, class ...Args>
inline void fflas_delete(Ptr p, Args ... args){
fflas_delete(p);
fflas_delete(std::forward<Args>(args)...);
}
#ifdef __FFLASFFPACK_HAVE_SSE4_1_INSTRUCTIONS
inline void prefetch(const int64_t* addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
#else
inline void prefetch(const int64_t*) {}
#endif
#define __CPUID(abcd,func,id) \
__asm__ __volatile__ ("cpuid": "=a" (abcd[0]), "=b" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id) );
inline void getCacheSize(int& l1, int& l2, int& l3)
{
int abcd[4];
l1 = l2 = l3 = 0;
int cache_id = 0;
int cache_type = 0;
do {
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
__CPUID(abcd,0x4,cache_id);
cache_type = (abcd[0] & 0x0F) >> 0;
if(cache_type==1||cache_type==3) // data or unified cache
{
int cache_level = (abcd[0] & 0xE0) >> 5; // A[7:5]
int ways = (abcd[1] & 0xFFC00000) >> 22; // B[31:22]
int partitions = (abcd[1] & 0x003FF000) >> 12; // B[21:12]
int line_size = (abcd[1] & 0x00000FFF) >> 0; // B[11:0]
int sets = (abcd[2]); // C[31:0]
int cache_size = (ways+1) * (partitions+1) * (line_size+1) * (sets+1);
switch(cache_level)
{
case 1: l1 = cache_size; break;
case 2: l2 = cache_size; break;
case 3: l3 = cache_size; break;
default: break;
}
}
cache_id++;
} while(cache_type>0 && cache_id<16);
}
inline void getTLBSize(int& tlb)
{
int abcd[4]={};
int sTLB=0;
int lTLB;
__CPUID(abcd,0x2,0);
unsigned char * bytes = reinterpret_cast<unsigned char *>(abcd)+2;
for(int i=0; i<14; ++i)
switch(bytes[i]){
case 0x03: sTLB=64; break;
case 0x04: lTLB=8; break;
case 0x05: lTLB=32; break;
case 0x56: lTLB=16; break;
case 0x57: sTLB=16; break;
case 0x59: sTLB=16; break;
case 0x5A: lTLB=32; break;
case 0x5B: sTLB=lTLB=64; break;
case 0x5C: sTLB=lTLB=128; break;
case 0x5D: sTLB=lTLB=256; break;
case 0xB3: sTLB=128; break;
case 0xB4: sTLB=256; break;
case 0xBA: sTLB=64; break;
case 0xC0: sTLB=lTLB=8; break;
case 0xCA: sTLB=512; break;
default: break;
}
//cout<<"small TLB: "<<sTLB<<endl;
//cout<<"large TLB: "<<lTLB<<endl;
tlb=sTLB*4096;
}
//---------- Cache sizes ----------
#if !defined(EIGEN_NO_CPUID)
# if defined(__GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
# if defined(__PIC__) && defined(__i386__)
// Case for x86 with PIC
# define EIGEN_CPUID(abcd,func,id) \
__asm__ __volatile__ ("xchgl %%ebx, %%esi;cpuid; xchgl %%ebx,%%esi": "=a" (abcd[0]), "=S" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id));
# else
// Case for x86_64 or x86 w/o PIC
# define EIGEN_CPUID(abcd,func,id) \
__asm__ __volatile__ ("cpuid": "=a" (abcd[0]), "=b" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id) );
# endif
# elif defined(_MSC_VER)
# if (_MSC_VER > 1500) && ( defined(_M_IX86) || defined(_M_X64) )
# define EIGEN_CPUID(abcd,func,id) __cpuidex((int*)abcd,func,id)
# endif
# endif
#endif
#ifdef EIGEN_CPUID
inline bool cpuid_is_vendor(int abcd[4], const char* vendor)
{
return abcd[1]==(reinterpret_cast<const int*>(vendor))[0] && abcd[3]==(reinterpret_cast<const int*>(vendor))[1] && abcd[2]==(reinterpret_cast<const int*>(vendor))[2];
}
inline void queryCacheSizes_intel_direct(int& l1, int& l2, int& l3)
{
int abcd[4];
l1 = l2 = l3 = 0;
int cache_id = 0;
int cache_type = 0;
do {
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
EIGEN_CPUID(abcd,0x4,cache_id);
cache_type = (abcd[0] & 0x0F) >> 0;
if(cache_type==1||cache_type==3) // data or unified cache
{
int cache_level = (abcd[0] & 0xE0) >> 5; // A[7:5]
int ways = (abcd[1] & 0xFFC00000) >> 22; // B[31:22]
int partitions = (abcd[1] & 0x003FF000) >> 12; // B[21:12]
int line_size = (abcd[1] & 0x00000FFF) >> 0; // B[11:0]
int sets = (abcd[2]); // C[31:0]
int cache_size = (ways+1) * (partitions+1) * (line_size+1) * (sets+1);
switch(cache_level)
{
case 1: l1 = cache_size; break;
case 2: l2 = cache_size; break;
case 3: l3 = cache_size; break;
default: break;
}
}
cache_id++;
} while(cache_type>0 && cache_id<16);
}
inline void queryCacheSizes_intel_codes(int& l1, int& l2, int& l3)
{
int abcd[4];
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
l1 = l2 = l3 = 0;
EIGEN_CPUID(abcd,0x00000002,0);
unsigned char * bytes = reinterpret_cast<unsigned char *>(abcd)+2;
bool check_for_p2_core2 = false;
for(int i=0; i<14; ++i)
{
switch(bytes[i])
{
case 0x0A: l1 = 8; break; // 0Ah data L1 cache, 8 KB, 2 ways, 32 byte lines
case 0x0C: l1 = 16; break; // 0Ch data L1 cache, 16 KB, 4 ways, 32 byte lines
case 0x0E: l1 = 24; break; // 0Eh data L1 cache, 24 KB, 6 ways, 64 byte lines
case 0x10: l1 = 16; break; // 10h data L1 cache, 16 KB, 4 ways, 32 byte lines (IA-64)
case 0x15: l1 = 16; break; // 15h code L1 cache, 16 KB, 4 ways, 32 byte lines (IA-64)
case 0x2C: l1 = 32; break; // 2Ch data L1 cache, 32 KB, 8 ways, 64 byte lines
case 0x30: l1 = 32; break; // 30h code L1 cache, 32 KB, 8 ways, 64 byte lines
case 0x60: l1 = 16; break; // 60h data L1 cache, 16 KB, 8 ways, 64 byte lines, sectored
case 0x66: l1 = 8; break; // 66h data L1 cache, 8 KB, 4 ways, 64 byte lines, sectored
case 0x67: l1 = 16; break; // 67h data L1 cache, 16 KB, 4 ways, 64 byte lines, sectored
case 0x68: l1 = 32; break; // 68h data L1 cache, 32 KB, 4 ways, 64 byte lines, sectored
case 0x1A: l2 = 96; break; // code and data L2 cache, 96 KB, 6 ways, 64 byte lines (IA-64)
case 0x22: l3 = 512; break; // code and data L3 cache, 512 KB, 4 ways (!), 64 byte lines, dual-sectored
case 0x23: l3 = 1024; break; // code and data L3 cache, 1024 KB, 8 ways, 64 byte lines, dual-sectored
case 0x25: l3 = 2048; break; // code and data L3 cache, 2048 KB, 8 ways, 64 byte lines, dual-sectored
case 0x29: l3 = 4096; break; // code and data L3 cache, 4096 KB, 8 ways, 64 byte lines, dual-sectored
case 0x39: l2 = 128; break; // code and data L2 cache, 128 KB, 4 ways, 64 byte lines, sectored
case 0x3A: l2 = 192; break; // code and data L2 cache, 192 KB, 6 ways, 64 byte lines, sectored
case 0x3B: l2 = 128; break; // code and data L2 cache, 128 KB, 2 ways, 64 byte lines, sectored
case 0x3C: l2 = 256; break; // code and data L2 cache, 256 KB, 4 ways, 64 byte lines, sectored
case 0x3D: l2 = 384; break; // code and data L2 cache, 384 KB, 6 ways, 64 byte lines, sectored
case 0x3E: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 64 byte lines, sectored
case 0x40: l2 = 0; break; // no integrated L2 cache (P6 core) or L3 cache (P4 core)
case 0x41: l2 = 128; break; // code and data L2 cache, 128 KB, 4 ways, 32 byte lines
case 0x42: l2 = 256; break; // code and data L2 cache, 256 KB, 4 ways, 32 byte lines
case 0x43: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 32 byte lines
case 0x44: l2 = 1024; break; // code and data L2 cache, 1024 KB, 4 ways, 32 byte lines
case 0x45: l2 = 2048; break; // code and data L2 cache, 2048 KB, 4 ways, 32 byte lines
case 0x46: l3 = 4096; break; // code and data L3 cache, 4096 KB, 4 ways, 64 byte lines
case 0x47: l3 = 8192; break; // code and data L3 cache, 8192 KB, 8 ways, 64 byte lines
case 0x48: l2 = 3072; break; // code and data L2 cache, 3072 KB, 12 ways, 64 byte lines
case 0x49: if(l2!=0) l3 = 4096; else {check_for_p2_core2=true; l3 = l2 = 4096;} break;// code and data L3 cache, 4096 KB, 16 ways, 64 byte lines (P4) or L2 for core2
case 0x4A: l3 = 6144; break; // code and data L3 cache, 6144 KB, 12 ways, 64 byte lines
case 0x4B: l3 = 8192; break; // code and data L3 cache, 8192 KB, 16 ways, 64 byte lines
case 0x4C: l3 = 12288; break; // code and data L3 cache, 12288 KB, 12 ways, 64 byte lines
case 0x4D: l3 = 16384; break; // code and data L3 cache, 16384 KB, 16 ways, 64 byte lines
case 0x4E: l2 = 6144; break; // code and data L2 cache, 6144 KB, 24 ways, 64 byte lines
case 0x78: l2 = 1024; break; // code and data L2 cache, 1024 KB, 4 ways, 64 byte lines
case 0x79: l2 = 128; break; // code and data L2 cache, 128 KB, 8 ways, 64 byte lines, dual-sectored
case 0x7A: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 64 byte lines, dual-sectored
case 0x7B: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 64 byte lines, dual-sectored
case 0x7C: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 64 byte lines, dual-sectored
case 0x7D: l2 = 2048; break; // code and data L2 cache, 2048 KB, 8 ways, 64 byte lines
case 0x7E: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 128 byte lines, sect. (IA-64)
case 0x7F: l2 = 512; break; // code and data L2 cache, 512 KB, 2 ways, 64 byte lines
case 0x80: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 64 byte lines
case 0x81: l2 = 128; break; // code and data L2 cache, 128 KB, 8 ways, 32 byte lines
case 0x82: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 32 byte lines
case 0x83: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 32 byte lines
case 0x84: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 32 byte lines
case 0x85: l2 = 2048; break; // code and data L2 cache, 2048 KB, 8 ways, 32 byte lines
case 0x86: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 64 byte lines
case 0x87: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 64 byte lines
case 0x88: l3 = 2048; break; // code and data L3 cache, 2048 KB, 4 ways, 64 byte lines (IA-64)
case 0x89: l3 = 4096; break; // code and data L3 cache, 4096 KB, 4 ways, 64 byte lines (IA-64)
case 0x8A: l3 = 8192; break; // code and data L3 cache, 8192 KB, 4 ways, 64 byte lines (IA-64)
case 0x8D: l3 = 3072; break; // code and data L3 cache, 3072 KB, 12 ways, 128 byte lines (IA-64)
default: break;
}
}
if(check_for_p2_core2 && l2 == l3)
l3 = 0;
l1 *= 1024;
l2 *= 1024;
l3 *= 1024;
}
inline void queryCacheSizes_intel(int& l1, int& l2, int& l3, int max_std_funcs)
{
if(max_std_funcs>=4)
queryCacheSizes_intel_direct(l1,l2,l3);
else
queryCacheSizes_intel_codes(l1,l2,l3);
}
inline void queryCacheSizes_amd(int& l1, int& l2, int& l3)
{
int abcd[4];
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
EIGEN_CPUID(abcd,0x80000005,0);
l1 = (abcd[2] >> 24) * 1024; // C[31:24] = L1 size in KB
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
EIGEN_CPUID(abcd,0x80000006,0);
l2 = (abcd[2] >> 16) * 1024; // C[31;16] = l2 cache size in KB
l3 = ((abcd[3] & 0xFFFC000) >> 18) * 512 * 1024; // D[31;18] = l3 cache size in 512KB
}
#endif
/** \internal
* Queries and returns the cache sizes in Bytes of the L1, L2, and L3 data caches respectively */
inline void queryCacheSizes(int& l1, int& l2, int& l3)
{
#ifdef EIGEN_CPUID
int abcd[4];
// identify the CPU vendor
EIGEN_CPUID(abcd,0x0,0);
int max_std_funcs = abcd[1];
if(cpuid_is_vendor(abcd,"GenuineIntel"))
queryCacheSizes_intel(l1,l2,l3,max_std_funcs);
else if(cpuid_is_vendor(abcd,"AuthenticAMD") || cpuid_is_vendor(abcd,"AMDisbetter!"))
queryCacheSizes_amd(l1,l2,l3);
else
// by default let's use Intel's API
queryCacheSizes_intel(l1,l2,l3,max_std_funcs);
// here is the list of other vendors:
// ||cpuid_is_vendor(abcd,"VIA VIA VIA ")
// ||cpuid_is_vendor(abcd,"CyrixInstead")
// ||cpuid_is_vendor(abcd,"CentaurHauls")
// ||cpuid_is_vendor(abcd,"GenuineTMx86")
// ||cpuid_is_vendor(abcd,"TransmetaCPU")
// ||cpuid_is_vendor(abcd,"RiseRiseRise")
// ||cpuid_is_vendor(abcd,"Geode by NSC")
// ||cpuid_is_vendor(abcd,"SiS SiS SiS ")
// ||cpuid_is_vendor(abcd,"UMC UMC UMC ")
// ||cpuid_is_vendor(abcd,"NexGenDriven")
#else
l1 = l2 = l3 = -1;
#endif
}
/** \internal
* \returns the size in Bytes of the L1 data cache */
inline int queryL1CacheSize()
{
int l1(-1), l2, l3;
queryCacheSizes(l1,l2,l3);
return l1;
}
/** \internal
* \returns the size in Bytes of the L2 or L3 cache if this later is present */
inline int queryTopLevelCacheSize()
{
int l1, l2(-1), l3(-1);
queryCacheSizes(l1,l2,l3);
return (std::max)(l2,l3);
}
} // namespace FFLAS
#endif // __FFLASFFPACK_memory_H
|