196 lines
8.0 KiB
C++
196 lines
8.0 KiB
C++
/*
|
|
* Copyright (C) 2018-2022 Intel Corporation
|
|
*
|
|
* SPDX-License-Identifier: MIT
|
|
*
|
|
*/
|
|
|
|
#include "shared/source/program/kernel_info.h"
|
|
|
|
#include "shared/source/device/device.h"
|
|
#include "shared/source/device_binary_format/patchtokens_decoder.h"
|
|
#include "shared/source/helpers/aligned_memory.h"
|
|
#include "shared/source/helpers/blit_commands_helper.h"
|
|
#include "shared/source/helpers/hw_helper.h"
|
|
#include "shared/source/helpers/kernel_helpers.h"
|
|
#include "shared/source/helpers/ptr_math.h"
|
|
#include "shared/source/helpers/string.h"
|
|
#include "shared/source/memory_manager/memory_manager.h"
|
|
|
|
#include <cstdint>
|
|
#include <cstring>
|
|
#include <map>
|
|
#include <sstream>
|
|
#include <unordered_map>
|
|
|
|
namespace NEO {
|
|
|
|
struct KernelArgumentType {
|
|
const char *argTypeQualifier;
|
|
uint64_t argTypeQualifierValue;
|
|
};
|
|
|
|
WorkSizeInfo::WorkSizeInfo(uint32_t maxWorkGroupSize, bool hasBarriers, uint32_t simdSize, uint32_t slmTotalSize, const HardwareInfo *hwInfo, uint32_t numThreadsPerSubSlice, uint32_t localMemSize, bool imgUsed, bool yTiledSurface, bool disableEUFusion) {
|
|
this->maxWorkGroupSize = maxWorkGroupSize;
|
|
this->hasBarriers = hasBarriers;
|
|
this->simdSize = simdSize;
|
|
this->slmTotalSize = slmTotalSize;
|
|
this->coreFamily = hwInfo->platform.eRenderCoreFamily;
|
|
this->numThreadsPerSubSlice = numThreadsPerSubSlice;
|
|
this->localMemSize = localMemSize;
|
|
this->imgUsed = imgUsed;
|
|
this->yTiledSurfaces = yTiledSurface;
|
|
|
|
setMinWorkGroupSize(hwInfo, disableEUFusion);
|
|
}
|
|
|
|
void WorkSizeInfo::setIfUseImg(const KernelInfo &kernelInfo) {
|
|
for (const auto &arg : kernelInfo.kernelDescriptor.payloadMappings.explicitArgs) {
|
|
if (arg.is<ArgDescriptor::ArgTImage>()) {
|
|
imgUsed = true;
|
|
yTiledSurfaces = true;
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
void WorkSizeInfo::setMinWorkGroupSize(const HardwareInfo *hwInfo, bool disableEUFusion) {
|
|
minWorkGroupSize = 0;
|
|
if (hasBarriers) {
|
|
uint32_t maxBarriersPerHSlice = (coreFamily >= IGFX_GEN9_CORE) ? 32 : 16;
|
|
minWorkGroupSize = numThreadsPerSubSlice * simdSize / maxBarriersPerHSlice;
|
|
}
|
|
if (slmTotalSize > 0) {
|
|
if (localMemSize < slmTotalSize) {
|
|
PRINT_DEBUG_STRING(NEO::DebugManager.flags.PrintDebugMessages.get(), stderr, "Size of SLM (%u) larger than available (%u)\n", slmTotalSize, localMemSize);
|
|
}
|
|
UNRECOVERABLE_IF(localMemSize < slmTotalSize);
|
|
minWorkGroupSize = std::max(maxWorkGroupSize / ((localMemSize / slmTotalSize)), minWorkGroupSize);
|
|
}
|
|
|
|
const auto &hwHelper = HwHelper::get(hwInfo->platform.eRenderCoreFamily);
|
|
if (hwHelper.isFusedEuDispatchEnabled(*hwInfo, disableEUFusion)) {
|
|
minWorkGroupSize *= 2;
|
|
}
|
|
}
|
|
|
|
void WorkSizeInfo::checkRatio(const size_t workItems[3]) {
|
|
if (slmTotalSize > 0) {
|
|
useRatio = true;
|
|
targetRatio = log((float)workItems[0]) - log((float)workItems[1]);
|
|
useStrictRatio = false;
|
|
} else if (yTiledSurfaces == true) {
|
|
useRatio = true;
|
|
targetRatio = YTilingRatioValue;
|
|
useStrictRatio = true;
|
|
}
|
|
}
|
|
|
|
KernelInfo::~KernelInfo() {
|
|
delete[] crossThreadData;
|
|
}
|
|
|
|
size_t KernelInfo::getSamplerStateArrayCount() const {
|
|
return kernelDescriptor.payloadMappings.samplerTable.numSamplers;
|
|
}
|
|
size_t KernelInfo::getSamplerStateArraySize(const HardwareInfo &hwInfo) const {
|
|
size_t samplerStateArraySize = getSamplerStateArrayCount() * HwHelper::get(hwInfo.platform.eRenderCoreFamily).getSamplerStateSize();
|
|
return samplerStateArraySize;
|
|
}
|
|
|
|
size_t KernelInfo::getBorderColorOffset() const {
|
|
size_t borderColorOffset = 0;
|
|
if (kernelDescriptor.payloadMappings.samplerTable.numSamplers > 0U) {
|
|
borderColorOffset = kernelDescriptor.payloadMappings.samplerTable.borderColor;
|
|
}
|
|
return borderColorOffset;
|
|
}
|
|
|
|
uint32_t KernelInfo::getConstantBufferSize() const {
|
|
return kernelDescriptor.kernelAttributes.crossThreadDataSize;
|
|
}
|
|
int32_t KernelInfo::getArgNumByName(const char *name) const {
|
|
int32_t argNum = 0;
|
|
for (const auto &argMeta : kernelDescriptor.explicitArgsExtendedMetadata) {
|
|
if (argMeta.argName.compare(name) == 0) {
|
|
return argNum;
|
|
}
|
|
++argNum;
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
bool KernelInfo::createKernelAllocation(const Device &device, bool internalIsa) {
|
|
UNRECOVERABLE_IF(kernelAllocation);
|
|
auto kernelIsaSize = heapInfo.KernelHeapSize;
|
|
const auto allocType = internalIsa ? AllocationType::KERNEL_ISA_INTERNAL : AllocationType::KERNEL_ISA;
|
|
|
|
if (device.getMemoryManager()->isKernelBinaryReuseEnabled()) {
|
|
auto lock = device.getMemoryManager()->lockKernelAllocationMap();
|
|
auto kernelName = this->kernelDescriptor.kernelMetadata.kernelName;
|
|
auto &storedAllocations = device.getMemoryManager()->getKernelAllocationMap();
|
|
auto kernelAllocations = storedAllocations.find(kernelName);
|
|
if (kernelAllocations != storedAllocations.end()) {
|
|
kernelAllocation = kernelAllocations->second.kernelAllocation;
|
|
kernelAllocations->second.reuseCounter++;
|
|
auto &hwInfo = device.getHardwareInfo();
|
|
auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
|
|
|
|
return MemoryTransferHelper::transferMemoryToAllocation(hwInfoConfig.isBlitCopyRequiredForLocalMemory(hwInfo, *kernelAllocation),
|
|
device, kernelAllocation, 0, heapInfo.pKernelHeap,
|
|
static_cast<size_t>(kernelIsaSize));
|
|
} else {
|
|
kernelAllocation = device.getMemoryManager()->allocateGraphicsMemoryWithProperties({device.getRootDeviceIndex(), kernelIsaSize, allocType, device.getDeviceBitfield()});
|
|
storedAllocations.insert(std::make_pair(kernelName, MemoryManager::KernelAllocationInfo(kernelAllocation, 1u)));
|
|
}
|
|
} else {
|
|
kernelAllocation = device.getMemoryManager()->allocateGraphicsMemoryWithProperties({device.getRootDeviceIndex(), kernelIsaSize, allocType, device.getDeviceBitfield()});
|
|
}
|
|
|
|
if (!kernelAllocation) {
|
|
return false;
|
|
}
|
|
|
|
auto &hwInfo = device.getHardwareInfo();
|
|
auto &hwInfoConfig = *HwInfoConfig::get(hwInfo.platform.eProductFamily);
|
|
|
|
return MemoryTransferHelper::transferMemoryToAllocation(hwInfoConfig.isBlitCopyRequiredForLocalMemory(hwInfo, *kernelAllocation),
|
|
device, kernelAllocation, 0, heapInfo.pKernelHeap,
|
|
static_cast<size_t>(kernelIsaSize));
|
|
}
|
|
|
|
void KernelInfo::apply(const DeviceInfoKernelPayloadConstants &constants) {
|
|
if (nullptr == this->crossThreadData) {
|
|
return;
|
|
}
|
|
|
|
const auto &implicitArgs = kernelDescriptor.payloadMappings.implicitArgs;
|
|
const auto privateMemorySize = static_cast<uint32_t>(KernelHelper::getPrivateSurfaceSize(kernelDescriptor.kernelAttributes.perHwThreadPrivateMemorySize,
|
|
constants.computeUnitsUsedForScratch));
|
|
|
|
auto setIfValidOffset = [&](auto value, NEO::CrossThreadDataOffset offset) {
|
|
if (isValidOffset(offset)) {
|
|
*ptrOffset(reinterpret_cast<decltype(value) *>(crossThreadData), offset) = value;
|
|
}
|
|
};
|
|
setIfValidOffset(reinterpret_cast<uintptr_t>(constants.slmWindow), implicitArgs.localMemoryStatelessWindowStartAddres);
|
|
setIfValidOffset(constants.slmWindowSize, implicitArgs.localMemoryStatelessWindowSize);
|
|
setIfValidOffset(privateMemorySize, implicitArgs.privateMemorySize);
|
|
setIfValidOffset(constants.maxWorkGroupSize, implicitArgs.maxWorkGroupSize);
|
|
}
|
|
|
|
std::string concatenateKernelNames(ArrayRef<KernelInfo *> kernelInfos) {
|
|
std::string semiColonDelimitedKernelNameStr;
|
|
|
|
for (const auto &kernelInfo : kernelInfos) {
|
|
if (!semiColonDelimitedKernelNameStr.empty()) {
|
|
semiColonDelimitedKernelNameStr += ';';
|
|
}
|
|
semiColonDelimitedKernelNameStr += kernelInfo->kernelDescriptor.kernelMetadata.kernelName;
|
|
}
|
|
|
|
return semiColonDelimitedKernelNameStr;
|
|
}
|
|
|
|
} // namespace NEO
|