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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
|
/*
* Copyright (C) 2023-2025 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/helpers/common_types.h"
#include "shared/source/helpers/non_copyable_or_moveable.h"
#include "shared/source/helpers/ptr_math.h"
#include "shared/source/memory_manager/allocation_type.h"
#include <cstdint>
#include <memory>
#include <mutex>
#include <vector>
namespace NEO {
class GraphicsAllocation;
class MemoryManager;
class Device;
class TagNodeBase;
template <bool deviceAlloc>
class DeviceAllocNodeType {
public:
using ValueT = uint64_t;
static constexpr size_t defaultAllocatorTagCount = 128;
static constexpr AllocationType getAllocationType() { return deviceAlloc ? NEO::AllocationType::gpuTimestampDeviceBuffer : NEO::AllocationType::timestampPacketTagBuffer; }
static constexpr TagNodeType getTagNodeType() { return TagNodeType::counter64b; }
static constexpr size_t getSinglePacketSize() { return sizeof(uint64_t); }
void initialize(uint64_t initValue) { data = initValue; }
protected:
uint64_t data = {};
};
static_assert(sizeof(uint64_t) == sizeof(DeviceAllocNodeType<true>), "This structure is consumed by GPU and has to follow specific restrictions for padding and size");
static_assert(sizeof(uint64_t) == sizeof(DeviceAllocNodeType<false>), "This structure is consumed by GPU and has to follow specific restrictions for padding and size");
class InOrderExecInfo : public NEO::NonCopyableClass {
public:
~InOrderExecInfo();
InOrderExecInfo() = delete;
static std::shared_ptr<InOrderExecInfo> create(TagNodeBase *deviceCounterNode, TagNodeBase *hostCounterNode, NEO::Device &device, uint32_t partitionCount, bool regularCmdList);
static std::shared_ptr<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);
InOrderExecInfo(TagNodeBase *deviceCounterNode, TagNodeBase *hostCounterNode, NEO::Device &device, uint32_t partitionCount, bool regularCmdList, bool atomicDeviceSignalling);
NEO::GraphicsAllocation *getDeviceCounterAllocation() const;
NEO::GraphicsAllocation *getHostCounterAllocation() const;
uint64_t *getBaseHostAddress() const { return hostAddress; }
uint64_t getBaseDeviceAddress() const { return deviceAddress; }
uint64_t getBaseHostGpuAddress() const;
uint64_t getDeviceNodeGpuAddress() const;
uint64_t getHostNodeGpuAddress() const;
size_t getDeviceNodeWriteSize() const {
if (deviceCounterNode) {
const size_t deviceAllocationWriteSize = sizeof(uint64_t) * numDevicePartitionsToWait;
return deviceAllocationWriteSize;
}
return 0;
}
size_t getHostNodeWriteSize() const {
if (hostCounterNode) {
const size_t hostAllocationWriteSize = sizeof(uint64_t) * numHostPartitionsToWait;
return hostAllocationWriteSize;
}
return 0;
}
uint64_t getCounterValue() const { return counterValue; }
void addCounterValue(uint64_t addValue) { counterValue += addValue; }
void resetCounterValue() { counterValue = 0; }
uint64_t getRegularCmdListSubmissionCounter() const { return regularCmdListSubmissionCounter; }
void addRegularCmdListSubmissionCounter(uint64_t addValue) { regularCmdListSubmissionCounter += addValue; }
bool isRegularCmdList() const { return regularCmdList; }
bool isHostStorageDuplicated() const { return duplicatedHostStorage; }
bool isAtomicDeviceSignalling() const { return atomicDeviceSignalling; }
uint32_t getNumDevicePartitionsToWait() const { return numDevicePartitionsToWait; }
uint32_t getNumHostPartitionsToWait() const { return numHostPartitionsToWait; }
void setAllocationOffset(uint32_t newOffset) { allocationOffset = newOffset; }
void initializeAllocationsFromHost();
uint32_t getAllocationOffset() const { return allocationOffset; }
void reset();
bool isExternalMemoryExecInfo() const { return deviceCounterNode == nullptr; }
void setLastWaitedCounterValue(uint64_t value) {
if (!isExternalMemoryExecInfo()) {
lastWaitedCounterValue = std::max(value, lastWaitedCounterValue);
}
}
bool isCounterAlreadyDone(uint64_t waitValue) const {
return lastWaitedCounterValue >= waitValue && this->allocationOffset == 0u;
}
NEO::GraphicsAllocation *getExternalHostAllocation() const { return externalHostAllocation; }
NEO::GraphicsAllocation *getExternalDeviceAllocation() const { return externalDeviceAllocation; }
void pushTempTimestampNode(TagNodeBase *node, uint64_t value);
void releaseNotUsedTempTimestampNodes(bool forceReturn);
protected:
void uploadToTbx(TagNodeBase &node, size_t size);
NEO::Device &device;
NEO::TagNodeBase *deviceCounterNode = nullptr;
NEO::TagNodeBase *hostCounterNode = nullptr;
NEO::GraphicsAllocation *externalHostAllocation = nullptr;
NEO::GraphicsAllocation *externalDeviceAllocation = nullptr;
std::vector<std::pair<NEO::TagNodeBase *, uint64_t>> tempTimestampNodes;
std::mutex mutex;
uint64_t counterValue = 0;
uint64_t lastWaitedCounterValue = 0;
uint64_t regularCmdListSubmissionCounter = 0;
uint64_t deviceAddress = 0;
uint64_t *hostAddress = nullptr;
uint32_t numDevicePartitionsToWait = 0;
uint32_t numHostPartitionsToWait = 0;
uint32_t allocationOffset = 0;
uint32_t rootDeviceIndex = 0;
bool regularCmdList = false;
bool duplicatedHostStorage = false;
bool atomicDeviceSignalling = false;
bool isTbx = false;
};
static_assert(NEO::NonCopyable<InOrderExecInfo>);
namespace InOrderPatchCommandHelpers {
inline uint64_t getAppendCounterValue(const InOrderExecInfo &inOrderExecInfo) {
if (inOrderExecInfo.isRegularCmdList() && inOrderExecInfo.getRegularCmdListSubmissionCounter() > 1) {
return inOrderExecInfo.getCounterValue() * (inOrderExecInfo.getRegularCmdListSubmissionCounter() - 1);
}
return 0;
}
enum class PatchCmdType {
none,
lri64b,
sdi,
semaphore,
walker,
pipeControl,
xyCopyBlt,
xyBlockCopyBlt,
xyColorBlt,
memSet
};
template <typename GfxFamily>
struct PatchCmd {
PatchCmd(std::shared_ptr<InOrderExecInfo> *inOrderExecInfo, void *cmd1, void *cmd2, uint64_t baseCounterValue, PatchCmdType patchCmdType, bool deviceAtomicSignaling, bool duplicatedHostStorage)
: cmd1(cmd1), cmd2(cmd2), baseCounterValue(baseCounterValue), patchCmdType(patchCmdType), deviceAtomicSignaling(deviceAtomicSignaling), duplicatedHostStorage(duplicatedHostStorage) {
if (inOrderExecInfo) {
this->inOrderExecInfo = *inOrderExecInfo;
}
}
void patch(uint64_t appendCounterValue) {
if (skipPatching) {
return;
}
switch (patchCmdType) {
case PatchCmdType::sdi:
patchSdi(appendCounterValue);
break;
case PatchCmdType::semaphore:
patchSemaphore(appendCounterValue);
break;
case PatchCmdType::walker:
patchComputeWalker(appendCounterValue);
break;
case PatchCmdType::lri64b:
patchLri64b(appendCounterValue);
break;
case PatchCmdType::pipeControl:
patchPipeControl(appendCounterValue);
break;
case PatchCmdType::xyCopyBlt:
case PatchCmdType::xyBlockCopyBlt:
case PatchCmdType::xyColorBlt:
case PatchCmdType::memSet:
patchBlitterCommand(appendCounterValue, patchCmdType);
break;
default:
UNRECOVERABLE_IF(true);
break;
}
}
void updateInOrderExecInfo(std::shared_ptr<InOrderExecInfo> *inOrderExecInfo) {
this->inOrderExecInfo = *inOrderExecInfo;
}
void setSkipPatching(bool value) {
skipPatching = value;
}
bool isExternalDependency() const { return inOrderExecInfo.get(); }
std::shared_ptr<InOrderExecInfo> inOrderExecInfo;
void *cmd1 = nullptr;
void *cmd2 = nullptr;
const uint64_t baseCounterValue = 0;
const PatchCmdType patchCmdType = PatchCmdType::none;
bool deviceAtomicSignaling = false;
bool duplicatedHostStorage = false;
bool skipPatching = false;
protected:
void patchSdi(uint64_t appendCounterValue) {
auto sdiCmd = reinterpret_cast<typename GfxFamily::MI_STORE_DATA_IMM *>(cmd1);
sdiCmd->setDataDword0(getLowPart(baseCounterValue + appendCounterValue));
sdiCmd->setDataDword1(getHighPart(baseCounterValue + appendCounterValue));
}
void patchSemaphore(uint64_t appendCounterValue) {
if (isExternalDependency()) {
appendCounterValue = InOrderPatchCommandHelpers::getAppendCounterValue(*inOrderExecInfo);
if (appendCounterValue == 0) {
return;
}
}
auto semaphoreCmd = reinterpret_cast<typename GfxFamily::MI_SEMAPHORE_WAIT *>(cmd1);
semaphoreCmd->setSemaphoreDataDword(static_cast<uint32_t>(baseCounterValue + appendCounterValue));
}
void patchComputeWalker(uint64_t appendCounterValue);
void patchBlitterCommand(uint64_t appendCounterValue, PatchCmdType patchCmdType);
void patchPipeControl(uint64_t appendCounterValue) {
auto pcCmd = reinterpret_cast<typename GfxFamily::PIPE_CONTROL *>(cmd1);
pcCmd->setImmediateData(static_cast<uint64_t>(baseCounterValue + appendCounterValue));
}
void patchLri64b(uint64_t appendCounterValue) {
if (isExternalDependency()) {
appendCounterValue = InOrderPatchCommandHelpers::getAppendCounterValue(*inOrderExecInfo);
if (appendCounterValue == 0) {
return;
}
}
const uint64_t counterValue = baseCounterValue + appendCounterValue;
auto lri1 = reinterpret_cast<typename GfxFamily::MI_LOAD_REGISTER_IMM *>(cmd1);
lri1->setDataDword(getLowPart(counterValue));
auto lri2 = reinterpret_cast<typename GfxFamily::MI_LOAD_REGISTER_IMM *>(cmd2);
lri2->setDataDword(getHighPart(counterValue));
}
PatchCmd() = delete;
};
} // namespace InOrderPatchCommandHelpers
template <typename GfxFamily>
using InOrderPatchCommandsContainer = std::vector<NEO::InOrderPatchCommandHelpers::PatchCmd<GfxFamily>>;
} // namespace NEO
|