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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
|
/*
Copyright (c) 2014 Intel Corporation. All Rights Reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef OFFLOAD_ENGINE_H_INCLUDED
#define OFFLOAD_ENGINE_H_INCLUDED
#include <limits.h>
#include <list>
#include <set>
#include <map>
#include "offload_common.h"
#include "coi/coi_client.h"
// Address range
class MemRange {
public:
MemRange() : m_start(0), m_length(0) {}
MemRange(const void *addr, uint64_t len) : m_start(addr), m_length(len) {}
const void* start() const {
return m_start;
}
const void* end() const {
return static_cast<const char*>(m_start) + m_length;
}
uint64_t length() const {
return m_length;
}
// returns true if given range overlaps with another one
bool overlaps(const MemRange &o) const {
// Two address ranges A[start, end) and B[start,end) overlap
// if A.start < B.end and A.end > B.start.
return start() < o.end() && end() > o.start();
}
// returns true if given range contains the other range
bool contains(const MemRange &o) const {
return start() <= o.start() && o.end() <= end();
}
private:
const void* m_start;
uint64_t m_length;
};
// Data associated with a pointer variable
class PtrData {
public:
PtrData(const void *addr, uint64_t len) :
cpu_addr(addr, len), cpu_buf(0),
mic_addr(0), alloc_disp(0), mic_buf(0), mic_offset(0),
ref_count(0), is_static(false)
{}
//
// Copy constructor
//
PtrData(const PtrData& ptr):
cpu_addr(ptr.cpu_addr), cpu_buf(ptr.cpu_buf),
mic_addr(ptr.mic_addr), alloc_disp(ptr.alloc_disp),
mic_buf(ptr.mic_buf), mic_offset(ptr.mic_offset),
ref_count(ptr.ref_count), is_static(ptr.is_static)
{}
bool operator<(const PtrData &o) const {
// Variables are sorted by the CPU start address.
// Overlapping memory ranges are considered equal.
return (cpu_addr.start() < o.cpu_addr.start()) &&
!cpu_addr.overlaps(o.cpu_addr);
}
long add_reference() {
if (is_static) {
return LONG_MAX;
}
#ifndef TARGET_WINNT
return __sync_fetch_and_add(&ref_count, 1);
#else // TARGET_WINNT
return _InterlockedIncrement(&ref_count) - 1;
#endif // TARGET_WINNT
}
long remove_reference() {
if (is_static) {
return LONG_MAX;
}
#ifndef TARGET_WINNT
return __sync_sub_and_fetch(&ref_count, 1);
#else // TARGET_WINNT
return _InterlockedDecrement(&ref_count);
#endif // TARGET_WINNT
}
long get_reference() const {
if (is_static) {
return LONG_MAX;
}
return ref_count;
}
public:
// CPU address range
const MemRange cpu_addr;
// CPU and MIC buffers
COIBUFFER cpu_buf;
COIBUFFER mic_buf;
// placeholder for buffer address on mic
uint64_t mic_addr;
uint64_t alloc_disp;
// additional offset to pointer data on MIC for improving bandwidth for
// data which is not 4K aligned
uint32_t mic_offset;
// if true buffers are created from static memory
bool is_static;
mutex_t alloc_ptr_data_lock;
private:
// reference count for the entry
long ref_count;
};
typedef std::list<PtrData*> PtrDataList;
// Data associated with automatic variable
class AutoData {
public:
AutoData(const void *addr, uint64_t len) :
cpu_addr(addr, len), ref_count(0)
{}
bool operator<(const AutoData &o) const {
// Variables are sorted by the CPU start address.
// Overlapping memory ranges are considered equal.
return (cpu_addr.start() < o.cpu_addr.start()) &&
!cpu_addr.overlaps(o.cpu_addr);
}
long add_reference() {
#ifndef TARGET_WINNT
return __sync_fetch_and_add(&ref_count, 1);
#else // TARGET_WINNT
return _InterlockedIncrement(&ref_count) - 1;
#endif // TARGET_WINNT
}
long remove_reference() {
#ifndef TARGET_WINNT
return __sync_sub_and_fetch(&ref_count, 1);
#else // TARGET_WINNT
return _InterlockedDecrement(&ref_count);
#endif // TARGET_WINNT
}
long get_reference() const {
return ref_count;
}
public:
// CPU address range
const MemRange cpu_addr;
private:
// reference count for the entry
long ref_count;
};
// Set of autimatic variables
typedef std::set<AutoData> AutoSet;
// Target image data
struct TargetImage
{
TargetImage(const char *_name, const void *_data, uint64_t _size,
const char *_origin, uint64_t _offset) :
name(_name), data(_data), size(_size),
origin(_origin), offset(_offset)
{}
// library name
const char* name;
// contents and size
const void* data;
uint64_t size;
// file of origin and offset within that file
const char* origin;
uint64_t offset;
};
typedef std::list<TargetImage> TargetImageList;
// Data associated with persistent auto objects
struct PersistData
{
PersistData(const void *addr, uint64_t routine_num, uint64_t size) :
stack_cpu_addr(addr), routine_id(routine_num)
{
stack_ptr_data = new PtrData(0, size);
}
// 1-st key value - begining of the stack at CPU
const void * stack_cpu_addr;
// 2-nd key value - identifier of routine invocation at CPU
uint64_t routine_id;
// corresponded PtrData; only stack_ptr_data->mic_buf is used
PtrData * stack_ptr_data;
// used to get offset of the variable in stack buffer
char * cpu_stack_addr;
};
typedef std::list<PersistData> PersistDataList;
// class representing a single engine
struct Engine {
friend void __offload_init_library_once(void);
friend void __offload_fini_library(void);
#define check_result(res, tag, ...) \
{ \
if (res == COI_PROCESS_DIED) { \
fini_process(true); \
exit(1); \
} \
if (res != COI_SUCCESS) { \
__liboffload_error_support(tag, __VA_ARGS__); \
exit(1); \
} \
}
int get_logical_index() const {
return m_index;
}
int get_physical_index() const {
return m_physical_index;
}
const COIPROCESS& get_process() const {
return m_process;
}
// initialize device
void init(void);
// add new library
void add_lib(const TargetImage &lib)
{
m_lock.lock();
m_ready = false;
m_images.push_back(lib);
m_lock.unlock();
}
COIRESULT compute(
const std::list<COIBUFFER> &buffers,
const void* data,
uint16_t data_size,
void* ret,
uint16_t ret_size,
uint32_t num_deps,
const COIEVENT* deps,
COIEVENT* event
);
#ifdef MYO_SUPPORT
// temporary workaround for blocking behavior for myoiLibInit/Fini calls
void init_myo(COIEVENT *event) {
COIRESULT res;
res = COI::PipelineRunFunction(get_pipeline(),
m_funcs[c_func_myo_init],
0, 0, 0, 0, 0, 0, 0, 0, 0,
event);
check_result(res, c_pipeline_run_func, m_index, res);
}
void fini_myo(COIEVENT *event) {
COIRESULT res;
res = COI::PipelineRunFunction(get_pipeline(),
m_funcs[c_func_myo_fini],
0, 0, 0, 0, 0, 0, 0, 0, 0,
event);
check_result(res, c_pipeline_run_func, m_index, res);
}
#endif // MYO_SUPPORT
//
// Memory association table
//
PtrData* find_ptr_data(const void *ptr) {
m_ptr_lock.lock();
PtrSet::iterator res = m_ptr_set.find(PtrData(ptr, 0));
m_ptr_lock.unlock();
if (res == m_ptr_set.end()) {
return 0;
}
return const_cast<PtrData*>(res.operator->());
}
PtrData* insert_ptr_data(const void *ptr, uint64_t len, bool &is_new) {
m_ptr_lock.lock();
std::pair<PtrSet::iterator, bool> res =
m_ptr_set.insert(PtrData(ptr, len));
PtrData* ptr_data = const_cast<PtrData*>(res.first.operator->());
m_ptr_lock.unlock();
is_new = res.second;
if (is_new) {
// It's necessary to lock as soon as possible.
// unlock must be done at call site of insert_ptr_data at
// branch for is_new
ptr_data->alloc_ptr_data_lock.lock();
}
return ptr_data;
}
void remove_ptr_data(const void *ptr) {
m_ptr_lock.lock();
m_ptr_set.erase(PtrData(ptr, 0));
m_ptr_lock.unlock();
}
//
// Automatic variables
//
AutoData* find_auto_data(const void *ptr) {
AutoSet &auto_vars = get_auto_vars();
AutoSet::iterator res = auto_vars.find(AutoData(ptr, 0));
if (res == auto_vars.end()) {
return 0;
}
return const_cast<AutoData*>(res.operator->());
}
AutoData* insert_auto_data(const void *ptr, uint64_t len) {
AutoSet &auto_vars = get_auto_vars();
std::pair<AutoSet::iterator, bool> res =
auto_vars.insert(AutoData(ptr, len));
return const_cast<AutoData*>(res.first.operator->());
}
void remove_auto_data(const void *ptr) {
get_auto_vars().erase(AutoData(ptr, 0));
}
//
// Signals
//
void add_signal(const void *signal, OffloadDescriptor *desc) {
m_signal_lock.lock();
m_signal_map[signal] = desc;
m_signal_lock.unlock();
}
OffloadDescriptor* find_signal(const void *signal, bool remove) {
OffloadDescriptor *desc = 0;
m_signal_lock.lock();
{
SignalMap::iterator it = m_signal_map.find(signal);
if (it != m_signal_map.end()) {
desc = it->second;
if (remove) {
m_signal_map.erase(it);
}
}
}
m_signal_lock.unlock();
return desc;
}
// stop device process
void fini_process(bool verbose);
// list of stacks active at the engine
PersistDataList m_persist_list;
private:
Engine() : m_index(-1), m_physical_index(-1), m_process(0), m_ready(false),
m_proc_number(0)
{}
~Engine() {
if (m_process != 0) {
fini_process(false);
}
}
// set indexes
void set_indexes(int logical_index, int physical_index) {
m_index = logical_index;
m_physical_index = physical_index;
}
// start process on device
void init_process();
void load_libraries(void);
void init_ptr_data(void);
// performs library intialization on the device side
pid_t init_device(void);
private:
// get pipeline associated with a calling thread
COIPIPELINE get_pipeline(void);
// get automatic vars set associated with the calling thread
AutoSet& get_auto_vars(void);
// destructor for thread data
static void destroy_thread_data(void *data);
private:
typedef std::set<PtrData> PtrSet;
typedef std::map<const void*, OffloadDescriptor*> SignalMap;
// device indexes
int m_index;
int m_physical_index;
// number of COI pipes created for the engine
long m_proc_number;
// process handle
COIPROCESS m_process;
// If false, device either has not been initialized or new libraries
// have been added.
bool m_ready;
mutex_t m_lock;
// List of libraries to be loaded
TargetImageList m_images;
// var table
PtrSet m_ptr_set;
mutex_t m_ptr_lock;
// signals
SignalMap m_signal_map;
mutex_t m_signal_lock;
// constants for accessing device function handles
enum {
c_func_compute = 0,
#ifdef MYO_SUPPORT
c_func_myo_init,
c_func_myo_fini,
#endif // MYO_SUPPORT
c_func_init,
c_func_var_table_size,
c_func_var_table_copy,
c_funcs_total
};
static const char* m_func_names[c_funcs_total];
// device function handles
COIFUNCTION m_funcs[c_funcs_total];
// int -> name mapping for device signals
static const int c_signal_max = 32;
static const char* c_signal_names[c_signal_max];
};
#endif // OFFLOAD_ENGINE_H_INCLUDED
|