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
|
/*
* Copyright (C) 2020-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/command_container/command_encoder.h"
#include "shared/source/device/device.h"
#include "shared/source/helpers/cache_policy.h"
#include "shared/source/kernel/implicit_args.h"
#include "shared/source/memory_manager/unified_memory_manager.h"
#include "level_zero/core/source/device/device.h"
#include "level_zero/core/source/driver/driver_handle.h"
#include "level_zero/core/source/kernel/kernel_imp.h"
#include "level_zero/core/source/module/module.h"
#include "igfxfmid.h"
namespace L0 {
template <GFXCORE_FAMILY gfxCoreFamily>
struct KernelHw : public KernelImp {
using KernelImp::KernelImp;
using GfxFamily = typename NEO::GfxFamilyMapper<gfxCoreFamily>::GfxFamily;
void setBufferSurfaceState(uint32_t argIndex, void *address, NEO::GraphicsAllocation *alloc) override {
uint64_t baseAddress = alloc->getGpuAddressToPatch();
auto sshAlignmentMask = NEO::EncodeSurfaceState<GfxFamily>::getSurfaceBaseAddressAlignmentMask();
// Remove misaligned bytes, accounted for in bufferOffset patch token
baseAddress &= sshAlignmentMask;
auto misalignedSize = ptrDiff(alloc->getGpuAddressToPatch(), baseAddress);
auto offset = ptrDiff(address, reinterpret_cast<void *>(baseAddress));
size_t bufferSizeForSsh = alloc->getUnderlyingBufferSize();
auto argInfo = kernelImmData->getDescriptor().payloadMappings.explicitArgs[argIndex].as<NEO::ArgDescPointer>();
bool offsetWasPatched = NEO::patchNonPointer<uint32_t, uint32_t>(ArrayRef<uint8_t>(this->crossThreadData.get(), this->crossThreadDataSize),
argInfo.bufferOffset, static_cast<uint32_t>(offset));
if (false == offsetWasPatched) {
// fallback to handling offset in surface state
baseAddress = reinterpret_cast<uintptr_t>(address);
bufferSizeForSsh -= offset;
DEBUG_BREAK_IF(baseAddress != (baseAddress & sshAlignmentMask));
offset = 0;
}
void *surfaceStateAddress = nullptr;
auto surfaceState = GfxFamily::cmdInitRenderSurfaceState;
if (NEO::isValidOffset(argInfo.bindless)) {
surfaceStateAddress = patchBindlessSurfaceState(alloc, argInfo.bindless);
} else {
surfaceStateAddress = ptrOffset(surfaceStateHeapData.get(), argInfo.bindful);
surfaceState = *reinterpret_cast<typename GfxFamily::RENDER_SURFACE_STATE *>(surfaceStateAddress);
}
uint64_t bufferAddressForSsh = baseAddress;
auto alignment = NEO::EncodeSurfaceState<GfxFamily>::getSurfaceBaseAddressAlignment();
bufferSizeForSsh += misalignedSize;
bufferSizeForSsh = alignUp(bufferSizeForSsh, alignment);
bool l3Enabled = true;
// Allocation MUST be cacheline (64 byte) aligned in order to enable L3 caching otherwise Heap corruption will occur coming from the KMD.
// Most commonly this issue will occur with Host Point Allocations from customers.
l3Enabled = isL3Capable(*alloc);
Device *device = module->getDevice();
NEO::Device *neoDevice = device->getNEODevice();
auto allocData = device->getDriverHandle()->getSvmAllocsManager()->getSVMAlloc(reinterpret_cast<void *>(alloc->getGpuAddress()));
if (allocData && allocData->allocationFlagsProperty.flags.locallyUncachedResource) {
l3Enabled = false;
}
if (l3Enabled == false) {
this->kernelRequiresQueueUncachedMocsCount++;
}
auto isDebuggerActive = neoDevice->isDebuggerActive() || neoDevice->getDebugger() != nullptr;
NEO::EncodeSurfaceStateArgs args;
args.outMemory = &surfaceState;
args.graphicsAddress = bufferAddressForSsh;
args.size = bufferSizeForSsh;
args.mocs = device->getMOCS(l3Enabled, false);
args.numAvailableDevices = neoDevice->getNumGenericSubDevices();
args.allocation = alloc;
args.gmmHelper = neoDevice->getGmmHelper();
args.useGlobalAtomics = kernelImmData->getDescriptor().kernelAttributes.flags.useGlobalAtomics;
args.areMultipleSubDevicesInContext = args.numAvailableDevices > 1;
args.implicitScaling = device->isImplicitScalingCapable();
args.isDebuggerActive = isDebuggerActive;
NEO::EncodeSurfaceState<GfxFamily>::encodeBuffer(args);
*reinterpret_cast<typename GfxFamily::RENDER_SURFACE_STATE *>(surfaceStateAddress) = surfaceState;
}
void evaluateIfRequiresGenerationOfLocalIdsByRuntime(const NEO::KernelDescriptor &kernelDescriptor) override {
size_t localWorkSizes[3];
localWorkSizes[0] = this->groupSize[0];
localWorkSizes[1] = this->groupSize[1];
localWorkSizes[2] = this->groupSize[2];
kernelRequiresGenerationOfLocalIdsByRuntime = NEO::EncodeDispatchKernel<GfxFamily>::isRuntimeLocalIdsGenerationRequired(
kernelDescriptor.kernelAttributes.numLocalIdChannels,
localWorkSizes,
std::array<uint8_t, 3>{
{kernelDescriptor.kernelAttributes.workgroupWalkOrder[0],
kernelDescriptor.kernelAttributes.workgroupWalkOrder[1],
kernelDescriptor.kernelAttributes.workgroupWalkOrder[2]}},
kernelDescriptor.kernelAttributes.flags.requiresWorkgroupWalkOrder,
requiredWorkgroupOrder,
kernelDescriptor.kernelAttributes.simdSize);
}
};
} // namespace L0
|
