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
|
/*
* Copyright (C) 2018-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/command_encoder.h"
#include "shared/source/command_stream/command_stream_receiver.h"
#include "shared/source/helpers/aligned_memory.h"
#include "shared/source/helpers/engine_node_helper.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/helpers/timestamp_packet.h"
#include "shared/source/os_interface/os_context.h"
#include "shared/source/utilities/tag_allocator.h"
#include "opencl/source/command_queue/command_queue.h"
#include "opencl/source/command_queue/command_queue_hw.h"
#include "opencl/source/command_queue/gpgpu_walker.h"
#include "opencl/source/event/user_event.h"
#include "opencl/source/helpers/queue_helpers.h"
namespace NEO {
template <typename GfxFamily>
template <typename WalkerType>
void GpgpuWalkerHelper<GfxFamily>::setSystolicModeEnable(WalkerType *walkerCmd) {
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsStart(
CommandQueue &commandQueue,
TagNodeBase &hwPerfCounter,
LinearStream *commandStream) {
const auto pPerformanceCounters = commandQueue.getPerfCounters();
const auto commandBufferType = EngineHelpers::isCcs(commandQueue.getGpgpuEngine().osContext->getEngineType())
? MetricsLibraryApi::GpuCommandBufferType::Compute
: MetricsLibraryApi::GpuCommandBufferType::Render;
const uint32_t size = pPerformanceCounters->getGpuCommandsSize(commandBufferType, true);
void *pBuffer = commandStream->getSpace(size);
pPerformanceCounters->getGpuCommands(commandBufferType, hwPerfCounter, true, size, pBuffer);
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsEnd(
CommandQueue &commandQueue,
TagNodeBase &hwPerfCounter,
LinearStream *commandStream) {
const auto pPerformanceCounters = commandQueue.getPerfCounters();
const auto commandBufferType = EngineHelpers::isCcs(commandQueue.getGpgpuEngine().osContext->getEngineType())
? MetricsLibraryApi::GpuCommandBufferType::Compute
: MetricsLibraryApi::GpuCommandBufferType::Render;
const uint32_t size = pPerformanceCounters->getGpuCommandsSize(commandBufferType, false);
void *pBuffer = commandStream->getSpace(size);
pPerformanceCounters->getGpuCommands(commandBufferType, hwPerfCounter, false, size, pBuffer);
}
template <typename GfxFamily>
size_t GpgpuWalkerHelper<GfxFamily>::getSizeForWaDisableRccRhwoOptimization(const Kernel *pKernel) {
return 0u;
}
template <typename GfxFamily>
size_t EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(uint32_t eventType, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounters, bool blitEnqueue, CommandQueue &commandQueue, const MultiDispatchInfo &multiDispatchInfo, bool isMarkerWithProfiling, bool eventsInWaitlist, bool resolveDependenciesByPipecontrol, cl_event *outEvent) {
size_t expectedSizeCS = 0;
auto &gfxCoreHelper = commandQueue.getDevice().getGfxCoreHelper();
auto &commandQueueHw = static_cast<CommandQueueHw<GfxFamily> &>(commandQueue);
auto &rootDeviceEnvironment = commandQueue.getDevice().getRootDeviceEnvironment();
auto isCacheFlushForBcs = commandQueueHw.isCacheFlushForBcsRequired();
if (blitEnqueue) {
size_t expectedSizeCS = TimestampPacketHelper::getRequiredCmdStreamSizeForNodeDependencyWithBlitEnqueue<GfxFamily>();
if (isCacheFlushForBcs || commandQueueHw.isCacheFlushForImageRequired(eventType)) {
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, NEO::PostSyncMode::immediateData);
}
return expectedSizeCS;
}
for (auto &dispatchInfo : multiDispatchInfo) {
expectedSizeCS += EnqueueOperation<GfxFamily>::getSizeRequiredCS(eventType, reserveProfilingCmdsSpace, reservePerfCounters, commandQueue, dispatchInfo.getKernel(), dispatchInfo);
size_t kernelObjAuxCount = multiDispatchInfo.getKernelObjsForAuxTranslation() != nullptr ? multiDispatchInfo.getKernelObjsForAuxTranslation()->size() : 0;
expectedSizeCS += dispatchInfo.dispatchInitCommands.estimateCommandsSize(kernelObjAuxCount, rootDeviceEnvironment, isCacheFlushForBcs);
expectedSizeCS += dispatchInfo.dispatchEpilogueCommands.estimateCommandsSize(kernelObjAuxCount, rootDeviceEnvironment, isCacheFlushForBcs);
}
auto relaxedOrderingEnabled = commandQueue.getGpgpuCommandStreamReceiver().directSubmissionRelaxedOrderingEnabled();
if (relaxedOrderingEnabled) {
expectedSizeCS += 2 * EncodeSetMMIO<GfxFamily>::sizeREG;
}
if (commandQueue.getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled()) {
// add relaxed ordering cond_bb_start
expectedSizeCS += TimestampPacketHelper::getRequiredCmdStreamSize<GfxFamily>(csrDeps, relaxedOrderingEnabled);
expectedSizeCS += EnqueueOperation<GfxFamily>::getSizeRequiredForTimestampPacketWrite();
if (resolveDependenciesByPipecontrol) {
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForStallingBarrier();
}
if (isMarkerWithProfiling) {
if (!eventsInWaitlist) {
expectedSizeCS += commandQueue.getGpgpuCommandStreamReceiver().getCmdsSizeForComputeBarrierCommand();
}
expectedSizeCS += 4 * EncodeStoreMMIO<GfxFamily>::size;
}
} else if (isMarkerWithProfiling) {
expectedSizeCS += 2 * MemorySynchronizationCommands<GfxFamily>::getSizeForSingleBarrier();
if (!gfxCoreHelper.useOnlyGlobalTimestamps()) {
expectedSizeCS += 2 * EncodeStoreMMIO<GfxFamily>::size;
}
}
if (multiDispatchInfo.peekMainKernel()) {
expectedSizeCS += EnqueueOperation<GfxFamily>::getSizeForCacheFlushAfterWalkerCommands(*multiDispatchInfo.peekMainKernel(), commandQueue);
}
if (debugManager.flags.PauseOnEnqueue.get() != -1) {
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForSingleBarrier() * 2;
expectedSizeCS += NEO::EncodeSemaphore<GfxFamily>::getSizeMiSemaphoreWait() * 2;
}
if (debugManager.flags.GpuScratchRegWriteAfterWalker.get() != -1) {
expectedSizeCS += sizeof(typename GfxFamily::MI_LOAD_REGISTER_IMM);
}
expectedSizeCS += TimestampPacketHelper::getRequiredCmdStreamSizeForMultiRootDeviceSyncNodesContainer<GfxFamily>(csrDeps);
if (outEvent) {
auto pEvent = castToObjectOrAbort<Event>(*outEvent);
if ((pEvent->getContext()->getRootDeviceIndices().size() > 1) && (!pEvent->isUserEvent())) {
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, NEO::PostSyncMode::immediateData);
}
}
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForSingleBarrier();
if ((CL_COMMAND_BARRIER == eventType) && !commandQueue.isOOQEnabled() && eventsInWaitlist) {
expectedSizeCS += EncodeStoreMemory<GfxFamily>::getStoreDataImmSize();
}
return expectedSizeCS;
}
template <typename GfxFamily>
size_t EnqueueOperation<GfxFamily>::getSizeRequiredCS(uint32_t cmdType, bool reserveProfilingCmdsSpace, bool reservePerfCounters, CommandQueue &commandQueue, const Kernel *pKernel, const DispatchInfo &dispatchInfo) {
if (isCommandWithoutKernel(cmdType)) {
return EnqueueOperation<GfxFamily>::getSizeRequiredCSNonKernel(reserveProfilingCmdsSpace, reservePerfCounters, commandQueue);
} else {
return EnqueueOperation<GfxFamily>::getSizeRequiredCSKernel<typename GfxFamily::DefaultWalkerType>(reserveProfilingCmdsSpace, reservePerfCounters, commandQueue, pKernel, dispatchInfo);
}
}
template <typename GfxFamily>
size_t EnqueueOperation<GfxFamily>::getSizeRequiredCSNonKernel(bool reserveProfilingCmdsSpace, bool reservePerfCounters, CommandQueue &commandQueue) {
size_t size = 0;
if (reserveProfilingCmdsSpace) {
size += 2 * MemorySynchronizationCommands<GfxFamily>::getSizeForSingleBarrier() + 4 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
}
return size;
}
template <typename GfxFamily>
template <typename WalkerType>
void GpgpuWalkerHelper<GfxFamily>::setupTimestampPacketFlushL3(WalkerType &walkerCmd, CommandQueue &commandQueue, const FlushL3Args &args) {
}
} // namespace NEO
|
