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
|
/*
* Copyright (C) 2022-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/command_stream/transfer_direction.h"
#include "shared/source/helpers/engine_node_helper.h"
#include "shared/source/sku_info/sku_info_base.h"
#include "level_zero/core/source/cmdlist/cmdlist_hw_immediate.h"
#include "level_zero/core/source/cmdqueue/cmdqueue_imp.h"
#include "level_zero/core/source/context/context.h"
#include "level_zero/core/source/event/event.h"
#include <functional>
#include <mutex>
#include <vector>
namespace NEO {
class CommandStreamReceiver;
}
namespace L0 {
struct CommandQueue;
struct DeviceImp;
struct BcsSplit {
DeviceImp &device;
uint32_t clientCount = 0u;
std::mutex mtx;
struct Events {
BcsSplit &bcsSplit;
std::mutex mtx;
std::vector<EventPool *> pools;
std::vector<Event *> barrier;
std::vector<Event *> subcopy;
std::vector<Event *> marker;
size_t createdFromLatestPool = 0u;
std::optional<size_t> obtainForSplit(Context *context, size_t maxEventCountInPool);
std::optional<size_t> allocateNew(Context *context, size_t maxEventCountInPool);
void resetEventPackage(size_t index);
void releaseResources();
Events(BcsSplit &bcsSplit) : bcsSplit(bcsSplit){};
} events;
std::vector<CommandQueue *> cmdQs;
std::vector<CommandQueue *> h2dCmdQs;
std::vector<CommandQueue *> d2hCmdQs;
NEO::BcsInfoMask engines = NEO::EngineHelpers::oddLinkedCopyEnginesMask;
NEO::BcsInfoMask h2dEngines = NEO::EngineHelpers::h2dCopyEngineMask;
NEO::BcsInfoMask d2hEngines = NEO::EngineHelpers::d2hCopyEngineMask;
template <GFXCORE_FAMILY gfxCoreFamily, typename T, typename K>
ze_result_t appendSplitCall(CommandListCoreFamilyImmediate<gfxCoreFamily> *cmdList,
T dstptr,
K srcptr,
size_t size,
ze_event_handle_t hSignalEvent,
uint32_t numWaitEvents,
ze_event_handle_t *phWaitEvents,
bool performMigration,
bool hasRelaxedOrderingDependencies,
NEO::TransferDirection direction,
std::function<ze_result_t(T, K, size_t, ze_event_handle_t)> appendCall) {
ze_result_t result = ZE_RESULT_SUCCESS;
const bool hasStallingCmds = !hasRelaxedOrderingDependencies;
auto markerEventIndexRet = this->events.obtainForSplit(Context::fromHandle(cmdList->getCmdListContext()), MemoryConstants::pageSize64k / sizeof(typename CommandListCoreFamilyImmediate<gfxCoreFamily>::GfxFamily::TimestampPacketType));
if (!markerEventIndexRet.has_value()) {
return ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY;
}
auto markerEventIndex = *markerEventIndexRet;
auto barrierRequired = !cmdList->isInOrderExecutionEnabled() && cmdList->isBarrierRequired();
if (barrierRequired) {
cmdList->appendSignalEvent(this->events.barrier[markerEventIndex]->toHandle());
}
auto subcopyEventIndex = markerEventIndex * this->cmdQs.size();
StackVec<ze_event_handle_t, 4> eventHandles;
auto &cmdQsForSplit = this->getCmdQsForSplit(direction);
auto signalEvent = Event::fromHandle(hSignalEvent);
if (!cmdList->handleCounterBasedEventOperations(signalEvent)) {
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
auto totalSize = size;
auto engineCount = cmdQsForSplit.size();
for (size_t i = 0; i < cmdQsForSplit.size(); i++) {
if (barrierRequired) {
auto barrierEventHandle = this->events.barrier[markerEventIndex]->toHandle();
cmdList->addEventsToCmdList(1u, &barrierEventHandle, nullptr, hasRelaxedOrderingDependencies, false, true, false);
}
cmdList->addEventsToCmdList(numWaitEvents, phWaitEvents, nullptr, hasRelaxedOrderingDependencies, false, true, false);
if (signalEvent && i == 0u) {
cmdList->appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, true, true, false, true);
}
auto localSize = totalSize / engineCount;
auto localDstPtr = ptrOffset(dstptr, size - totalSize);
auto localSrcPtr = ptrOffset(srcptr, size - totalSize);
auto eventHandle = this->events.subcopy[subcopyEventIndex + i]->toHandle();
result = appendCall(localDstPtr, localSrcPtr, localSize, eventHandle);
if (cmdList->flushTaskSubmissionEnabled()) {
cmdList->executeCommandListImmediateWithFlushTaskImpl(performMigration, hasStallingCmds, hasRelaxedOrderingDependencies, false, false, cmdQsForSplit[i]);
} else {
cmdList->executeCommandListImmediateImpl(performMigration, cmdQsForSplit[i]);
}
eventHandles.push_back(eventHandle);
totalSize -= localSize;
engineCount--;
if (signalEvent) {
signalEvent->appendAdditionalCsr(static_cast<CommandQueueImp *>(cmdQsForSplit[i])->getCsr());
}
}
cmdList->addEventsToCmdList(static_cast<uint32_t>(cmdQsForSplit.size()), eventHandles.data(), nullptr, hasRelaxedOrderingDependencies, false, true, false);
if (signalEvent) {
cmdList->appendEventForProfilingAllWalkers(signalEvent, nullptr, nullptr, false, true, false, true);
}
cmdList->appendEventForProfilingAllWalkers(this->events.marker[markerEventIndex], nullptr, nullptr, false, true, false, true);
if (cmdList->isInOrderExecutionEnabled()) {
cmdList->appendSignalInOrderDependencyCounter(signalEvent, false);
}
cmdList->handleInOrderDependencyCounter(signalEvent, false, false);
return result;
}
bool setupDevice(uint32_t productFamily, bool internalUsage, const ze_command_queue_desc_t *desc, NEO::CommandStreamReceiver *csr);
void releaseResources();
std::vector<CommandQueue *> &getCmdQsForSplit(NEO::TransferDirection direction);
BcsSplit(DeviceImp &device) : device(device), events(*this){};
};
} // namespace L0
|
