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
|
/*
* Copyright (C) 2021-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#include "shared/source/command_stream/csr_definitions.h"
#include "shared/source/command_stream/stream_properties.h"
#include "shared/source/debug_settings/debug_settings_manager.h"
#include "shared/source/helpers/hw_helper.h"
#include "shared/source/helpers/pipeline_select_helper.h"
#include "shared/source/helpers/preamble_base.inl"
#include "reg_configs_common.h"
// L3 programming:
// All L3 Client Pool: 320KB
// URB Pool: 64KB
// Use Full ways: true
// SLM: reserved (always enabled)
namespace NEO {
template <typename Family>
void PreambleHelper<Family>::programPipelineSelect(LinearStream *pCommandStream,
const PipelineSelectArgs &pipelineSelectArgs,
const HardwareInfo &hwInfo) {
using PIPELINE_SELECT = typename Family::PIPELINE_SELECT;
PIPELINE_SELECT cmd = Family::cmdInitPipelineSelect;
if (DebugManager.flags.CleanStateInPreamble.get()) {
auto cmdBuffer = pCommandStream->getSpaceForCmd<PIPELINE_SELECT>();
cmd.setPipelineSelection(PIPELINE_SELECT::PIPELINE_SELECTION_3D);
*cmdBuffer = cmd;
PipeControlArgs args = {};
args.stateCacheInvalidationEnable = true;
MemorySynchronizationCommands<Family>::addSingleBarrier(*pCommandStream, args);
}
auto cmdBuffer = pCommandStream->getSpaceForCmd<PIPELINE_SELECT>();
auto mask = pipelineSelectEnablePipelineSelectMaskBits;
cmd.setPipelineSelection(PIPELINE_SELECT::PIPELINE_SELECTION_GPGPU);
if constexpr (Family::isUsingMediaSamplerDopClockGate) {
mask |= pipelineSelectMediaSamplerDopClockGateMaskBits;
cmd.setMediaSamplerDopClockGateEnable(!pipelineSelectArgs.mediaSamplerRequired);
}
bool systolicSupport = pipelineSelectArgs.systolicPipelineSelectSupport;
bool systolicValue = pipelineSelectArgs.systolicPipelineSelectMode;
int32_t overrideSystolic = DebugManager.flags.OverrideSystolicPipelineSelect.get();
if (overrideSystolic != -1) {
systolicSupport = true;
systolicValue = !!overrideSystolic;
}
if (systolicSupport) {
cmd.setSystolicModeEnable(systolicValue);
mask |= pipelineSelectSystolicModeEnableMaskBits;
}
cmd.setMaskBits(mask);
*cmdBuffer = cmd;
if (DebugManager.flags.CleanStateInPreamble.get()) {
PipeControlArgs args = {};
args.stateCacheInvalidationEnable = true;
MemorySynchronizationCommands<Family>::addSingleBarrier(*pCommandStream, args);
}
}
template <>
void PreambleHelper<Family>::addPipeControlBeforeVfeCmd(LinearStream *pCommandStream, const HardwareInfo *hwInfo, EngineGroupType engineGroupType) {
}
template <>
void PreambleHelper<Family>::programL3(LinearStream *pCommandStream, uint32_t l3Config) {
}
template <>
uint32_t PreambleHelper<Family>::getUrbEntryAllocationSize() {
return 0u;
}
template <>
void PreambleHelper<Family>::appendProgramVFEState(const HardwareInfo &hwInfo, const StreamProperties &streamProperties, void *cmd);
template <typename GfxFamily>
void *PreambleHelper<GfxFamily>::getSpaceForVfeState(LinearStream *pCommandStream,
const HardwareInfo &hwInfo,
EngineGroupType engineGroupType) {
using CFE_STATE = typename Family::CFE_STATE;
return pCommandStream->getSpace(sizeof(CFE_STATE));
}
template <typename GfxFamily>
void PreambleHelper<GfxFamily>::programVfeState(void *pVfeState,
const HardwareInfo &hwInfo,
uint32_t scratchSize,
uint64_t scratchAddress,
uint32_t maxFrontEndThreads,
const StreamProperties &streamProperties,
LogicalStateHelper *logicalStateHelper) {
using CFE_STATE = typename Family::CFE_STATE;
auto cfeState = reinterpret_cast<CFE_STATE *>(pVfeState);
CFE_STATE cmd = Family::cmdInitCfeState;
uint32_t lowAddress = uint32_t(0xFFFFFFFF & scratchAddress);
cmd.setScratchSpaceBuffer(lowAddress);
cmd.setMaximumNumberOfThreads(maxFrontEndThreads);
appendProgramVFEState(hwInfo, streamProperties, &cmd);
if (DebugManager.flags.CFEMaximumNumberOfThreads.get() != -1) {
cmd.setMaximumNumberOfThreads(DebugManager.flags.CFEMaximumNumberOfThreads.get());
}
if (DebugManager.flags.CFEOverDispatchControl.get() != -1) {
cmd.setOverDispatchControl(static_cast<typename CFE_STATE::OVER_DISPATCH_CONTROL>(DebugManager.flags.CFEOverDispatchControl.get()));
}
if (DebugManager.flags.CFELargeGRFThreadAdjustDisable.get() != -1) {
cmd.setLargeGRFThreadAdjustDisable(DebugManager.flags.CFELargeGRFThreadAdjustDisable.get());
}
*cfeState = cmd;
}
template <>
uint64_t PreambleHelper<Family>::getScratchSpaceAddressOffsetForVfeState(LinearStream *pCommandStream, void *pVfeState) {
return 0;
}
template <typename GfxFamily>
size_t PreambleHelper<GfxFamily>::getVFECommandsSize() {
using CFE_STATE = typename Family::CFE_STATE;
return sizeof(CFE_STATE);
}
template <>
uint32_t PreambleHelper<Family>::getL3Config(const HardwareInfo &hwInfo, bool useSLM) {
return 0u;
}
template <>
const uint32_t L3CNTLRegisterOffset<Family>::registerOffset = std::numeric_limits<uint32_t>::max();
template <>
struct DebugModeRegisterOffset<Family> {
enum {
registerOffset = 0x20d8,
debugEnabledValue = (1 << 5) | (1 << 21)
};
};
template <>
struct TdDebugControlRegisterOffset<Family> {
enum {
registerOffset = 0xe400,
debugEnabledValue = (1 << 7) | (1 << 4) | (1 << 2) | (1 << 0)
};
};
} // namespace NEO
|
