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
|
/*
* Copyright (C) 2023-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/helpers/in_order_cmd_helpers.h"
#include "shared/source/command_stream/command_stream_receiver.h"
#include "shared/source/device/device.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/memory_manager/allocation_properties.h"
#include "shared/source/utilities/tag_allocator.h"
#include <cstdint>
#include <string.h>
#include <vector>
namespace NEO {
std::shared_ptr<InOrderExecInfo> InOrderExecInfo::create(TagNodeBase *deviceCounterNode, TagNodeBase *hostCounterNode, NEO::Device &device, uint32_t partitionCount, bool regularCmdList) {
bool atomicDeviceSignalling = device.getGfxCoreHelper().inOrderAtomicSignallingEnabled(device.getRootDeviceEnvironment());
UNRECOVERABLE_IF(!deviceCounterNode);
return std::make_shared<NEO::InOrderExecInfo>(deviceCounterNode, hostCounterNode, device, partitionCount, regularCmdList, atomicDeviceSignalling);
}
std::shared_ptr<InOrderExecInfo> InOrderExecInfo::createFromExternalAllocation(NEO::Device &device, NEO::GraphicsAllocation *deviceAllocation, uint64_t deviceAddress, NEO::GraphicsAllocation *hostAllocation,
uint64_t *hostAddress, uint64_t counterValue, uint32_t devicePartitions, uint32_t hostPartitions) {
auto inOrderExecInfo = std::make_shared<NEO::InOrderExecInfo>(nullptr, nullptr, device, 1, false, true);
inOrderExecInfo->counterValue = counterValue;
inOrderExecInfo->externalHostAllocation = hostAllocation;
inOrderExecInfo->externalDeviceAllocation = deviceAllocation;
inOrderExecInfo->hostAddress = hostAddress;
inOrderExecInfo->deviceAddress = deviceAddress;
inOrderExecInfo->duplicatedHostStorage = (deviceAllocation != hostAllocation);
inOrderExecInfo->numDevicePartitionsToWait = devicePartitions;
inOrderExecInfo->numHostPartitionsToWait = hostPartitions;
return inOrderExecInfo;
}
InOrderExecInfo::~InOrderExecInfo() {
if (deviceCounterNode) {
deviceCounterNode->returnTag();
}
if (hostCounterNode) {
hostCounterNode->returnTag();
}
// forced return - All related objects (CmdList and Events) already destroyed
releaseNotUsedTempTimestampNodes(true);
}
InOrderExecInfo::InOrderExecInfo(TagNodeBase *deviceCounterNode, TagNodeBase *hostCounterNode, NEO::Device &device, uint32_t partitionCount, bool regularCmdList, bool atomicDeviceSignalling)
: device(device), deviceCounterNode(deviceCounterNode), hostCounterNode(hostCounterNode), rootDeviceIndex(device.getRootDeviceIndex()),
regularCmdList(regularCmdList), atomicDeviceSignalling(atomicDeviceSignalling) {
numDevicePartitionsToWait = atomicDeviceSignalling ? 1 : partitionCount;
numHostPartitionsToWait = partitionCount;
if (hostCounterNode) {
hostAddress = reinterpret_cast<uint64_t *>(hostCounterNode->getCpuBase());
duplicatedHostStorage = true;
} else if (deviceCounterNode) {
hostAddress = reinterpret_cast<uint64_t *>(deviceCounterNode->getCpuBase());
}
if (deviceCounterNode) {
deviceAddress = deviceCounterNode->getGpuAddress();
}
isTbx = device.getDefaultEngine().commandStreamReceiver->isTbxMode();
reset();
}
void InOrderExecInfo::uploadToTbx(TagNodeBase &node, size_t size) {
constexpr uint32_t allBanks = std::numeric_limits<uint32_t>::max();
auto csr = device.getDefaultEngine().commandStreamReceiver;
auto allocation = node.getBaseGraphicsAllocation()->getGraphicsAllocation(rootDeviceIndex);
auto offset = ptrDiff(node.getGpuAddress(), allocation->getGpuAddress()) + this->allocationOffset;
if (allocation->isTbxWritable(allBanks)) {
// initialize full page tables for the first time
csr->writeMemory(*allocation, false, 0, 0);
} else {
// chunk write if allocation already initialized
allocation->setTbxWritable(true, allBanks);
csr->writeMemory(*allocation, true, offset, size);
}
allocation->setTbxWritable(false, allBanks);
}
void InOrderExecInfo::initializeAllocationsFromHost() {
if (deviceCounterNode) {
const size_t deviceAllocationWriteSize = sizeof(uint64_t) * numDevicePartitionsToWait;
memset(ptrOffset(deviceCounterNode->getCpuBase(), allocationOffset), 0, deviceAllocationWriteSize);
if (isTbx) {
uploadToTbx(*deviceCounterNode, deviceAllocationWriteSize);
}
}
if (hostCounterNode) {
const size_t hostAllocationWriteSize = sizeof(uint64_t) * numHostPartitionsToWait;
memset(ptrOffset(hostCounterNode->getCpuBase(), allocationOffset), 0, hostAllocationWriteSize);
if (isTbx) {
uploadToTbx(*hostCounterNode, hostAllocationWriteSize);
}
}
}
void InOrderExecInfo::reset() {
resetCounterValue();
regularCmdListSubmissionCounter = 0;
allocationOffset = 0;
initializeAllocationsFromHost();
}
NEO::GraphicsAllocation *InOrderExecInfo::getDeviceCounterAllocation() const {
if (externalDeviceAllocation) {
return externalDeviceAllocation;
}
return deviceCounterNode ? deviceCounterNode->getBaseGraphicsAllocation()->getGraphicsAllocation(rootDeviceIndex) : nullptr;
}
NEO::GraphicsAllocation *InOrderExecInfo::getHostCounterAllocation() const {
if (externalHostAllocation) {
return externalHostAllocation;
}
return hostCounterNode ? hostCounterNode->getBaseGraphicsAllocation()->getGraphicsAllocation(rootDeviceIndex) : nullptr;
}
uint64_t InOrderExecInfo::getBaseHostGpuAddress() const {
return hostCounterNode->getGpuAddress();
}
void InOrderExecInfo::pushTempTimestampNode(TagNodeBase *node, uint64_t value) {
std::unique_lock<std::mutex> lock(mutex);
tempTimestampNodes.emplace_back(node, value);
}
void InOrderExecInfo::releaseNotUsedTempTimestampNodes(bool forceReturn) {
std::unique_lock<std::mutex> lock(mutex);
std::vector<std::pair<TagNodeBase *, uint64_t>> tempVector;
for (auto &node : tempTimestampNodes) {
if (forceReturn || lastWaitedCounterValue >= node.second) {
node.first->returnTag();
} else {
tempVector.push_back(node);
}
}
tempTimestampNodes.swap(tempVector);
}
uint64_t InOrderExecInfo::getHostNodeGpuAddress() const {
if (hostCounterNode) {
return hostCounterNode->getGpuAddress() + allocationOffset;
}
return 0;
}
uint64_t InOrderExecInfo::getDeviceNodeGpuAddress() const {
if (deviceCounterNode) {
return deviceCounterNode->getGpuAddress() + allocationOffset;
}
return 0;
}
} // namespace NEO
|
