File: device.cpp

package info (click to toggle)
intel-compute-runtime 25.44.36015.8-1
  • links: PTS, VCS
  • area: main
  • in suites: sid
  • size: 79,632 kB
  • sloc: cpp: 931,547; lisp: 2,074; sh: 719; makefile: 162; python: 21
file content (126 lines) | stat: -rw-r--r-- 5,398 bytes parent folder | download
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
/*
 * Copyright (C) 2023-2025 Intel Corporation
 *
 * SPDX-License-Identifier: MIT
 *
 */

#include "level_zero/core/source/device/device.h"

#include "shared/source/device/device.h"
#include "shared/source/helpers/aligned_memory.h"
#include "shared/source/helpers/fill_pattern_tag_node.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/helpers/in_order_cmd_helpers.h"

#include "level_zero/core/source/gfx_core_helpers/l0_gfx_core_helper.h"

namespace L0 {

uint32_t Device::getRootDeviceIndex() const {
    return neoDevice->getRootDeviceIndex();
}

NEO::DebuggerL0 *Device::getL0Debugger() {
    return getNEODevice()->getL0Debugger();
}

template <typename NodeT>
NEO::TagAllocatorBase *getInOrderCounterAllocator(std::unique_ptr<NEO::TagAllocatorBase> &allocator, std::mutex &inOrderAllocatorMutex, NEO::Device &neoDevice, uint32_t immediateWritePostSyncOffset) {
    if (!allocator.get()) {
        std::unique_lock<std::mutex> lock(inOrderAllocatorMutex);

        if (!allocator.get()) {
            RootDeviceIndicesContainer rootDeviceIndices = {neoDevice.getRootDeviceIndex()};

            const size_t maxPartitionCount = neoDevice.getDeviceBitfield().count();

            const size_t nodeSize = alignUp(immediateWritePostSyncOffset * maxPartitionCount, MemoryConstants::cacheLineSize * 4) * 2; // Multiplied by 2 to handle 32b overflow

            DEBUG_BREAK_IF(alignUp(nodeSize, MemoryConstants::cacheLineSize) * NodeT::defaultAllocatorTagCount > MemoryConstants::pageSize64k);

            allocator = std::make_unique<NEO::TagAllocator<NodeT>>(rootDeviceIndices, neoDevice.getMemoryManager(), NodeT::defaultAllocatorTagCount,
                                                                   MemoryConstants::cacheLineSize, nodeSize, 0, false, false, neoDevice.getDeviceBitfield());
        }
    }

    return allocator.get();
}

NEO::TagAllocatorBase *Device::getDeviceInOrderCounterAllocator() {
    return getInOrderCounterAllocator<NEO::DeviceAllocNodeType<true>>(deviceInOrderCounterAllocator, inOrderAllocatorMutex, *getNEODevice(), getL0GfxCoreHelper().getImmediateWritePostSyncOffset());
}

NEO::TagAllocatorBase *Device::getHostInOrderCounterAllocator() {
    return getInOrderCounterAllocator<NEO::DeviceAllocNodeType<false>>(hostInOrderCounterAllocator, inOrderAllocatorMutex, *getNEODevice(), getL0GfxCoreHelper().getImmediateWritePostSyncOffset());
}

NEO::TagAllocatorBase *Device::getInOrderTimestampAllocator() {
    if (!inOrderTimestampAllocator.get()) {
        std::unique_lock<std::mutex> lock(inOrderAllocatorMutex);

        if (!inOrderTimestampAllocator.get()) {
            RootDeviceIndicesContainer rootDeviceIndices = {getNEODevice()->getRootDeviceIndex()};

            size_t packetsCountPerElement = getEventMaxPacketCount();
            size_t alignment = getGfxCoreHelper().getTimestampPacketAllocatorAlignment();

            inOrderTimestampAllocator = getL0GfxCoreHelper().getInOrderTimestampAllocator(rootDeviceIndices, getNEODevice()->getMemoryManager(), 64, packetsCountPerElement, alignment, getNEODevice()->getDeviceBitfield());
        }
    }

    return inOrderTimestampAllocator.get();
}

NEO::TagAllocatorBase *Device::getFillPatternAllocator() {
    if (!this->fillPatternAllocator.get()) {
        static std::mutex mtx;
        std::unique_lock<std::mutex> lock(mtx);

        if (!this->fillPatternAllocator.get()) {
            RootDeviceIndicesContainer rootDeviceIndices = {getNEODevice()->getRootDeviceIndex()};
            fillPatternAllocator = std::make_unique<NEO::TagAllocator<NEO::FillPaternNodeType>>(rootDeviceIndices, getNEODevice()->getMemoryManager(), MemoryConstants::pageSize2M / MemoryConstants::cacheLineSize,
                                                                                                MemoryConstants::cacheLineSize, MemoryConstants::cacheLineSize, 0, false, false, getNEODevice()->getDeviceBitfield());
        }
    }

    return this->fillPatternAllocator.get();
}

uint32_t Device::getNextSyncDispatchQueueId() {
    auto newValue = syncDispatchQueueIdAllocator.fetch_add(1);

    UNRECOVERABLE_IF(newValue == (std::numeric_limits<uint32_t>::max() - 1));

    ensureSyncDispatchTokenAllocation();

    return newValue;
}

void Device::ensureSyncDispatchTokenAllocation() {
    if (!syncDispatchTokenAllocation) {
        std::unique_lock<std::mutex> lock(syncDispatchTokenMutex);

        if (!syncDispatchTokenAllocation) {

            const NEO::AllocationProperties allocationProperties(getRootDeviceIndex(), true, MemoryConstants::pageSize,
                                                                 NEO::AllocationType::syncDispatchToken,
                                                                 true, false, getNEODevice()->getDeviceBitfield());

            syncDispatchTokenAllocation = getNEODevice()->getMemoryManager()->allocateGraphicsMemoryWithProperties(allocationProperties);
            UNRECOVERABLE_IF(syncDispatchTokenAllocation == nullptr);

            memset(syncDispatchTokenAllocation->getUnderlyingBuffer(), 0, syncDispatchTokenAllocation->getUnderlyingBufferSize());
        }
    }
}

ze_result_t Device::getPriorityLevels(int32_t *lowestPriority, int32_t *highestPriority) {

    *highestPriority = queuePriorityHigh;
    *lowestPriority = queuePriorityLow;

    return ZE_RESULT_SUCCESS;
}

} // namespace L0