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
|
/*
* Copyright (C) 2018-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/built_ins/sip.h"
#include "shared/source/command_stream/preemption.h"
#include "shared/source/device/device.h"
#include "shared/source/helpers/hw_helper.h"
#include "shared/source/helpers/pipe_control_args.h"
#include "shared/source/helpers/preamble.h"
#include "shared/source/memory_manager/graphics_allocation.h"
namespace NEO {
template <typename GfxFamily>
void PreemptionHelper::programCsrBaseAddress(LinearStream &preambleCmdStream, Device &device, const GraphicsAllocation *preemptionCsr, LogicalStateHelper *logicalStateHelper) {
if (device.getPreemptionMode() == PreemptionMode::MidThread) {
UNRECOVERABLE_IF(nullptr == preemptionCsr);
programCsrBaseAddressCmd<GfxFamily>(preambleCmdStream, preemptionCsr, logicalStateHelper);
}
}
template <typename GfxFamily>
void PreemptionHelper::programCsrBaseAddressCmd(LinearStream &preambleCmdStream, const GraphicsAllocation *preemptionCsr, LogicalStateHelper *logicalStateHelper) {
using GPGPU_CSR_BASE_ADDRESS = typename GfxFamily::GPGPU_CSR_BASE_ADDRESS;
auto csr = reinterpret_cast<GPGPU_CSR_BASE_ADDRESS *>(preambleCmdStream.getSpace(sizeof(GPGPU_CSR_BASE_ADDRESS)));
GPGPU_CSR_BASE_ADDRESS cmd = GfxFamily::cmdInitGpgpuCsrBaseAddress;
cmd.setGpgpuCsrBaseAddress(preemptionCsr->getGpuAddressToPatch());
*csr = cmd;
}
template <typename GfxFamily>
void PreemptionHelper::programStateSip(LinearStream &preambleCmdStream, Device &device, LogicalStateHelper *logicalStateHelper) {
using STATE_SIP = typename GfxFamily::STATE_SIP;
bool debuggingEnabled = device.getDebugger() != nullptr || device.isDebuggerActive();
bool isMidThreadPreemption = device.getPreemptionMode() == PreemptionMode::MidThread;
if (isMidThreadPreemption || debuggingEnabled) {
auto sipAllocation = SipKernel::getSipKernel(device).getSipAllocation();
programStateSipCmd<GfxFamily>(preambleCmdStream, sipAllocation, logicalStateHelper);
}
}
template <typename GfxFamily>
void PreemptionHelper::programStateSipCmd(LinearStream &preambleCmdStream, GraphicsAllocation *sipAllocation, LogicalStateHelper *logicalStateHelper) {
using STATE_SIP = typename GfxFamily::STATE_SIP;
auto sip = reinterpret_cast<STATE_SIP *>(preambleCmdStream.getSpace(sizeof(STATE_SIP)));
STATE_SIP cmd = GfxFamily::cmdInitStateSip;
cmd.setSystemInstructionPointer(sipAllocation->getGpuAddressToPatch());
*sip = cmd;
}
template <typename GfxFamily>
void PreemptionHelper::programStateSipEndWa(LinearStream &cmdStream, Device &device) {}
template <typename GfxFamily>
void PreemptionHelper::programCmdStream(LinearStream &cmdStream, PreemptionMode newPreemptionMode,
PreemptionMode oldPreemptionMode, GraphicsAllocation *preemptionCsr) {
if (oldPreemptionMode == newPreemptionMode) {
return;
}
uint32_t regVal = 0;
if (newPreemptionMode == PreemptionMode::MidThread) {
regVal = PreemptionConfig<GfxFamily>::midThreadVal | PreemptionConfig<GfxFamily>::mask;
} else if (newPreemptionMode == PreemptionMode::ThreadGroup) {
regVal = PreemptionConfig<GfxFamily>::threadGroupVal | PreemptionConfig<GfxFamily>::mask;
} else {
regVal = PreemptionConfig<GfxFamily>::cmdLevelVal | PreemptionConfig<GfxFamily>::mask;
}
LriHelper<GfxFamily>::program(&cmdStream, PreemptionConfig<GfxFamily>::mmioAddress, regVal, true);
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredCmdStreamSize(PreemptionMode newPreemptionMode, PreemptionMode oldPreemptionMode) {
if (newPreemptionMode == oldPreemptionMode) {
return 0;
}
return sizeof(typename GfxFamily::MI_LOAD_REGISTER_IMM);
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredPreambleSize(const Device &device) {
if (device.getPreemptionMode() == PreemptionMode::MidThread) {
return sizeof(typename GfxFamily::GPGPU_CSR_BASE_ADDRESS);
}
return 0;
}
template <typename GfxFamily>
size_t PreemptionHelper::getRequiredStateSipCmdSize(Device &device, bool isRcs) {
size_t size = 0;
bool isMidThreadPreemption = device.getPreemptionMode() == PreemptionMode::MidThread;
bool debuggingEnabled = device.getDebugger() != nullptr || device.isDebuggerActive();
if (isMidThreadPreemption || debuggingEnabled) {
size += sizeof(typename GfxFamily::STATE_SIP);
}
return size;
}
template <typename GfxFamily>
size_t PreemptionHelper::getPreemptionWaCsSize(const Device &device) {
return 0u;
}
template <typename GfxFamily>
void PreemptionHelper::applyPreemptionWaCmdsBegin(LinearStream *pCommandStream, const Device &device) {
}
template <typename GfxFamily>
void PreemptionHelper::applyPreemptionWaCmdsEnd(LinearStream *pCommandStream, const Device &device) {
}
template <typename GfxFamily>
void PreemptionHelper::programInterfaceDescriptorDataPreemption(INTERFACE_DESCRIPTOR_DATA<GfxFamily> *idd, PreemptionMode preemptionMode) {
using INTERFACE_DESCRIPTOR_DATA = typename GfxFamily::INTERFACE_DESCRIPTOR_DATA;
if (preemptionMode == PreemptionMode::MidThread) {
idd->setThreadPreemptionDisable(INTERFACE_DESCRIPTOR_DATA::THREAD_PREEMPTION_DISABLE_DISABLE);
} else {
idd->setThreadPreemptionDisable(INTERFACE_DESCRIPTOR_DATA::THREAD_PREEMPTION_DISABLE_ENABLE);
}
}
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::mmioAddress = 0x2580;
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::mask = ((1 << 1) | (1 << 2)) << 16;
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::threadGroupVal = (1 << 1);
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::cmdLevelVal = (1 << 2);
template <typename GfxFamily>
constexpr uint32_t PreemptionConfig<GfxFamily>::midThreadVal = 0;
} // namespace NEO
|
