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
|
/*
* Copyright (C) 2025-2026 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_container/command_encoder.h"
#include "shared/source/command_container/encode_surface_state.h"
#include "shared/source/command_stream/command_stream_receiver.h"
#include "shared/source/command_stream/scratch_space_controller.h"
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "shared/source/helpers/in_order_cmd_helpers.h"
#include "shared/source/helpers/surface_format_info.h"
#include "shared/source/indirect_heap/indirect_heap.h"
#include "shared/source/kernel/kernel_descriptor.h"
#include "shared/source/os_interface/product_helper.h"
#include "shared/source/utilities/tag_allocator.h"
#include "implicit_args.h"
#include <algorithm>
namespace NEO {
template <typename Family>
void EncodeSurfaceState<Family>::setAuxParamsForMCSCCS(R_SURFACE_STATE *surfaceState, const ReleaseHelper *releaseHelper) {
}
template <typename Family>
void EncodeSurfaceState<Family>::setPitchForScratch(R_SURFACE_STATE *surfaceState, uint32_t pitchInBytes, const ProductHelper &productHelper) {
if (!productHelper.isAvailableExtendedScratch()) {
surfaceState->setSurfacePitch(pitchInBytes);
return;
}
auto pitchInUnitsOf64Bytes = pitchInBytes / 64;
surfaceState->setSurfacePitch(pitchInUnitsOf64Bytes);
surfaceState->setExtendScratchSurfaceSize(true);
}
template <typename Family>
uint32_t EncodeSurfaceState<Family>::getPitchForScratchInBytes(R_SURFACE_STATE *surfaceState, const ProductHelper &productHelper) {
if (!productHelper.isAvailableExtendedScratch()) {
return surfaceState->getSurfacePitch();
}
auto pitchInUnitsOf64Bytes = surfaceState->getSurfacePitch();
return 64 * pitchInUnitsOf64Bytes;
}
template <typename Family>
template <bool isHeapless>
void EncodeDispatchKernel<Family>::setScratchAddress(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &submissionCsr) {
if constexpr (isHeapless) {
if (requiredScratchSlot0Size > 0u || requiredScratchSlot1Size > 0u) {
std::unique_lock<NEO::CommandStreamReceiver::MutexType> primaryCsrLock;
auto primaryCsr = submissionCsr.getPrimaryCsr();
if (primaryCsr && primaryCsr != &submissionCsr) {
primaryCsrLock = primaryCsr->obtainUniqueOwnership();
}
auto scratchController = submissionCsr.getPrimaryScratchSpaceController();
UNRECOVERABLE_IF(scratchController == nullptr);
bool sbaStateDirty = false;
bool frontEndStateDirty = false;
scratchController->setRequiredScratchSpace(ssh->getCpuBase(), 0, requiredScratchSlot0Size, requiredScratchSlot1Size,
submissionCsr.getOsContext(), sbaStateDirty, frontEndStateDirty);
if (scratchController->getScratchSpaceSlot0Allocation()) {
submissionCsr.makeResident(*scratchController->getScratchSpaceSlot0Allocation());
}
if (scratchController->getScratchSpaceSlot1Allocation()) {
submissionCsr.makeResident(*scratchController->getScratchSpaceSlot1Allocation());
}
scratchAddress = ssh->getGpuBase() + scratchController->getScratchPatchAddress();
}
}
}
template <typename Family>
template <typename InterfaceDescriptorType>
void EncodeDispatchKernel<Family>::encodeEuSchedulingPolicy(InterfaceDescriptorType *pInterfaceDescriptor, const KernelDescriptor &kernelDesc, int32_t defaultPipelinedThreadArbitrationPolicy) {
auto pipelinedThreadArbitrationPolicy = kernelDesc.kernelAttributes.threadArbitrationPolicy;
if (pipelinedThreadArbitrationPolicy == ThreadArbitrationPolicy::NotPresent) {
pipelinedThreadArbitrationPolicy = static_cast<ThreadArbitrationPolicy>(defaultPipelinedThreadArbitrationPolicy);
}
using EU_THREAD_SCHEDULING_MODE_OVERRIDE = typename InterfaceDescriptorType::EU_THREAD_SCHEDULING_MODE_OVERRIDE;
switch (pipelinedThreadArbitrationPolicy) {
case ThreadArbitrationPolicy::RoundRobin:
pInterfaceDescriptor->setEuThreadSchedulingModeOverride(EU_THREAD_SCHEDULING_MODE_OVERRIDE::EU_THREAD_SCHEDULING_MODE_OVERRIDE_ROUND_ROBIN);
break;
case ThreadArbitrationPolicy::AgeBased:
pInterfaceDescriptor->setEuThreadSchedulingModeOverride(EU_THREAD_SCHEDULING_MODE_OVERRIDE::EU_THREAD_SCHEDULING_MODE_OVERRIDE_OLDEST_FIRST);
break;
case ThreadArbitrationPolicy::RoundRobinAfterDependency:
pInterfaceDescriptor->setEuThreadSchedulingModeOverride(EU_THREAD_SCHEDULING_MODE_OVERRIDE::EU_THREAD_SCHEDULING_MODE_OVERRIDE_STALL_BASED_ROUND_ROBIN);
break;
default:
pInterfaceDescriptor->setEuThreadSchedulingModeOverride(EU_THREAD_SCHEDULING_MODE_OVERRIDE::EU_THREAD_SCHEDULING_MODE_OVERRIDE_STALL_BASED_ROUND_ROBIN);
}
}
template <typename Family>
bool EncodeDispatchKernel<Family>::singleTileExecImplicitScalingRequired(bool cooperativeKernel) {
if (debugManager.flags.SingleTileExecutionForCooperativeKernels.get() == 1) {
return cooperativeKernel;
}
return false;
}
template <typename Family>
template <typename CommandType>
void EncodePostSync<Family>::encodeL3Flush(CommandType &cmd, const EncodePostSyncArgs &args) {
using POSTSYNC_DATA_TYPE = decltype(Family::template getPostSyncType<CommandType>());
if constexpr (std::is_same_v<POSTSYNC_DATA_TYPE, typename Family::POSTSYNC_DATA_2>) {
bool l2Flush = args.dcFlushEnable && args.isFlushL3ForExternalAllocationRequired;
bool l2TransientFlush = args.dcFlushEnable && args.isFlushL3ForHostUsmRequired;
cmd.getPostSync().setL2Flush(l2Flush);
cmd.getPostSync().setL2TransientFlush(l2TransientFlush);
}
}
template <typename Family>
template <typename CommandType>
void EncodePostSync<Family>::setupPostSyncForInOrderExec(CommandType &cmd, const EncodePostSyncArgs &args) {
using POSTSYNC_DATA_TYPE = decltype(Family::template getPostSyncType<CommandType>());
auto gmmHelper = args.device->getGmmHelper();
uint32_t mocs = args.dcFlushEnable ? gmmHelper->getUncachedMOCS() : gmmHelper->getL3EnabledMOCS();
if (debugManager.flags.OverridePostSyncMocs.get() != -1) {
mocs = debugManager.flags.OverridePostSyncMocs.get();
}
auto requiresSystemMemoryFence = args.requiresSystemMemoryFence();
int32_t overrideProgramSystemMemoryFence = debugManager.flags.ProgramGlobalFenceAsPostSyncOperationInComputeWalker.get();
if (overrideProgramSystemMemoryFence != -1) {
requiresSystemMemoryFence = !!overrideProgramSystemMemoryFence;
}
const uint64_t deviceGpuVa = args.inOrderExecInfo->getBaseDeviceAddress() + args.inOrderExecInfo->getAllocationOffset();
const uint64_t data = args.inOrderCounterValue;
if constexpr (Family::template isPostSyncData1<POSTSYNC_DATA_TYPE>()) {
setPostSyncData(cmd.getPostSync(), POSTSYNC_DATA_TYPE::OPERATION_WRITE_IMMEDIATE_DATA, deviceGpuVa, data, 0, mocs, false, requiresSystemMemoryFence);
} else {
uint32_t postSyncId = 0;
const bool deviceInterrupt = (args.interruptEvent && !args.inOrderExecInfo->isHostStorageDuplicated());
if (args.inOrderExecInfo->isAtomicDeviceSignalling()) {
setPostSyncData(getPostSync(cmd, postSyncId++), POSTSYNC_DATA_TYPE::OPERATION_ATOMIC_OPN, deviceGpuVa, 0, static_cast<uint32_t>(POSTSYNC_DATA_TYPE::ATOMIC_OPCODE::ATOMIC_OPCODE_ATOMIC_INC8B),
mocs, deviceInterrupt, requiresSystemMemoryFence);
} else {
setPostSyncData(getPostSync(cmd, postSyncId++), POSTSYNC_DATA_TYPE::OPERATION_WRITE_IMMEDIATE_DATA, deviceGpuVa, data, 0, mocs, deviceInterrupt, requiresSystemMemoryFence);
}
if (args.inOrderExecInfo->isHostStorageDuplicated()) {
setPostSyncData(getPostSync(cmd, postSyncId++), POSTSYNC_DATA_TYPE::OPERATION_WRITE_IMMEDIATE_DATA, args.inOrderExecInfo->getBaseHostGpuAddress(), data, 0, mocs, args.interruptEvent, requiresSystemMemoryFence);
}
if (args.eventAddress) {
setPostSyncData(getPostSync(cmd, postSyncId++), (args.isTimestampEvent ? POSTSYNC_DATA_TYPE::OPERATION_WRITE_TIMESTAMP : POSTSYNC_DATA_TYPE::OPERATION_WRITE_IMMEDIATE_DATA), args.eventAddress, args.postSyncImmValue, 0, mocs, false, requiresSystemMemoryFence);
}
if (args.inOrderIncrementValue > 0) {
setPostSyncData(getPostSync(cmd, postSyncId++), POSTSYNC_DATA_TYPE::OPERATION_ATOMIC_OPN, args.inOrderIncrementGpuAddress, args.inOrderIncrementValue,
static_cast<uint32_t>(POSTSYNC_DATA_TYPE::ATOMIC_OPCODE::ATOMIC_OPCODE_ATOMIC_ADD8B), mocs, false, requiresSystemMemoryFence);
}
setCommandLevelInterrupt(cmd, args.interruptEvent);
encodeL3Flush(cmd, args);
}
}
template <typename Family>
template <typename WalkerType>
void EncodeDispatchKernel<Family>::encodeAdditionalWalkerFields(const RootDeviceEnvironment &rootDeviceEnvironment, WalkerType &walkerCmd, const EncodeWalkerArgs &walkerArgs) {
constexpr bool heaplessModeEnabled = std::is_same_v<WalkerType, typename Family::COMPUTE_WALKER_2>;
if constexpr (heaplessModeEnabled) {
auto maxNumberOfThreads = walkerArgs.maxFrontEndThreads;
maxNumberOfThreads = std::max(64U, maxNumberOfThreads);
if (debugManager.flags.MaximumNumberOfThreads.get() != -1) {
maxNumberOfThreads = static_cast<uint32_t>(debugManager.flags.MaximumNumberOfThreads.get());
}
walkerCmd.setMaximumNumberOfThreads(maxNumberOfThreads);
}
if (walkerArgs.hasSample) {
walkerCmd.setDispatchWalkOrder(WalkerType::DISPATCH_WALK_ORDER::DISPATCH_WALK_ORDER_MORTON_WALK);
walkerCmd.setThreadGroupBatchSize(WalkerType::THREAD_GROUP_BATCH_SIZE::THREAD_GROUP_BATCH_SIZE_TG_BATCH_4);
}
if (walkerArgs.requiredDispatchWalkOrder == NEO::RequiredDispatchWalkOrder::x) {
walkerCmd.setDispatchWalkOrder(WalkerType::DISPATCH_WALK_ORDER::DISPATCH_WALK_ORDER_LINEAR_WALK);
} else if (walkerArgs.requiredDispatchWalkOrder == NEO::RequiredDispatchWalkOrder::y) {
walkerCmd.setDispatchWalkOrder(WalkerType::DISPATCH_WALK_ORDER::DISPATCH_WALK_ORDER_Y_ORDER_WALK);
} else {
UNRECOVERABLE_IF(walkerArgs.requiredDispatchWalkOrder != NEO::RequiredDispatchWalkOrder::none);
}
}
template <typename Family>
void InOrderPatchCommandHelpers::PatchCmd<Family>::patchComputeWalker(uint64_t appendCounterValue) {
if (!deviceAtomicSignaling) {
auto walkerCmd = reinterpret_cast<typename Family::DefaultWalkerType *>(cmd1);
auto &postSync0 = walkerCmd->getPostSync();
postSync0.setImmediateData(baseCounterValue + appendCounterValue);
}
if (duplicatedHostStorage) {
auto walkerCmd = reinterpret_cast<typename Family::COMPUTE_WALKER_2 *>(cmd1);
auto &postSync1 = walkerCmd->getPostSyncOpn1();
postSync1.setImmediateData(baseCounterValue + appendCounterValue);
}
}
template <typename Family>
bool EncodeEnableRayTracing<Family>::is48bResourceNeededForRayTracing() {
return false;
}
template <typename Family>
void EncodeSurfaceState<Family>::convertSurfaceStateToPacked(R_SURFACE_STATE *surfaceState, ImageInfo &imgInfo) {
using SURFACE_FORMAT = typename R_SURFACE_STATE::SURFACE_FORMAT;
using SHADER_CHANNEL_SELECT = typename R_SURFACE_STATE::SHADER_CHANNEL_SELECT;
constexpr uint32_t r32g32b32a32SurfaceFormatImageElementSizeInBytes = 16u;
constexpr auto convertSurfaceFormatToPackedLambda = [](uint32_t bitsPerPixel) -> typename R_SURFACE_STATE::OVERRIDETILEBPP {
using OVERRIDETILEBPP = typename R_SURFACE_STATE::OVERRIDETILEBPP;
switch (bitsPerPixel) {
case 64:
return OVERRIDETILEBPP::OVERRIDETILEBPP_TILE64_64BPP;
case 32:
return OVERRIDETILEBPP::OVERRIDETILEBPP_TILE64_32BPP;
case 16:
return OVERRIDETILEBPP::OVERRIDETILEBPP_TILE64_16BPP;
case 8:
return OVERRIDETILEBPP::OVERRIDETILEBPP_TILE64_8BPP;
default:
return OVERRIDETILEBPP::OVERRIDETILEBPP_TILE64_LEGACY;
}
};
auto scale = r32g32b32a32SurfaceFormatImageElementSizeInBytes / imgInfo.surfaceFormat->imageElementSizeInBytes;
auto newWidth = static_cast<uint32_t>(Math::divideAndRoundUp(imgInfo.imgDesc.imageWidth, scale));
surfaceState->setOverrideTileBPP(convertSurfaceFormatToPackedLambda(static_cast<uint32_t>(imgInfo.surfaceFormat->imageElementSizeInBytes * 8)));
surfaceState->setWidth(newWidth);
surfaceState->setSurfaceFormat(SURFACE_FORMAT::SURFACE_FORMAT_R32G32B32A32_UINT);
surfaceState->setShaderChannelSelectRed(SHADER_CHANNEL_SELECT::SHADER_CHANNEL_SELECT_RED);
surfaceState->setShaderChannelSelectGreen(SHADER_CHANNEL_SELECT::SHADER_CHANNEL_SELECT_GREEN);
surfaceState->setShaderChannelSelectBlue(SHADER_CHANNEL_SELECT::SHADER_CHANNEL_SELECT_BLUE);
surfaceState->setShaderChannelSelectAlpha(SHADER_CHANNEL_SELECT::SHADER_CHANNEL_SELECT_ALPHA);
}
template <typename Family>
void EncodeSurfaceState<Family>::setAdditionalCacheSettings(R_SURFACE_STATE *surfaceState) {
int32_t forceL1P5CacheForRenderSurface = debugManager.flags.ForceL1P5CacheForRenderSurface.get();
if (forceL1P5CacheForRenderSurface != -1) {
surfaceState->setDisableL1P5(!forceL1P5CacheForRenderSurface);
}
}
} // namespace NEO
|
