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
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2019-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
//===----------------------------------------------------------------------===//
///
/// Any transformations required for functional correctness prior to splitting
/// shaders into continuations.
///
//===----------------------------------------------------------------------===//
#include "RTBuilder.h"
#include "RTStackFormat.h"
#include "RTArgs.h"
#include "Compiler/IGCPassSupport.h"
#include "Compiler/CodeGenPublicEnums.h"
#include "common/LLVMWarningsPush.hpp"
#include <llvm/IR/InstIterator.h>
#include "common/LLVMWarningsPop.hpp"
#include "Probe/Assertion.h"
#include "Utils.h"
using namespace llvm;
using namespace IGC;
using namespace RTStackFormat;
class SplitPreparePass : public ModulePass
{
public:
SplitPreparePass() : ModulePass(ID)
{
initializeSplitPreparePassPass(*PassRegistry::getPassRegistry());
}
bool runOnModule(Module &M) override;
StringRef getPassName() const override
{
return "SplitPreparePass";
}
void getAnalysisUsage(llvm::AnalysisUsage &AU) const override
{
AU.addRequired<CodeGenContextWrapper>();
}
static char ID;
private:
void hoistValues(Function &F, CallableShaderTypeMD Ty) const;
bool processShader(Function& F, FunctionMetaData &FMD) const;
private:
RayDispatchShaderContext *m_CGCtx = nullptr;
const UnifiedBits* RayFlagBits = nullptr;
};
char SplitPreparePass::ID = 0;
// Register pass to igc-opt
#define PASS_FLAG "split-prepare-pass"
#define PASS_DESCRIPTION "transformations required before continuation splitting to ensure correctness"
#define PASS_CFG_ONLY false
#define PASS_ANALYSIS false
IGC_INITIALIZE_PASS_BEGIN(SplitPreparePass, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
IGC_INITIALIZE_PASS_DEPENDENCY(CodeGenContextWrapper)
IGC_INITIALIZE_PASS_END(SplitPreparePass, PASS_FLAG, PASS_DESCRIPTION, PASS_CFG_ONLY, PASS_ANALYSIS)
bool SplitPreparePass::runOnModule(Module &M)
{
m_CGCtx = static_cast<RayDispatchShaderContext*>(
getAnalysis<CodeGenContextWrapper>().getCodeGenContext());
ModuleMetaData* modMD = m_CGCtx->getModuleMetaData();
auto &FuncMD = modMD->FuncMD;
auto FlagBits = examineRayFlags(*m_CGCtx);
RayFlagBits = &FlagBits;
bool Changed = false;
for (auto& F : M)
{
if (F.isDeclaration())
continue;
auto MD = FuncMD.find(&F);
IGC_ASSERT_MESSAGE((MD != FuncMD.end()), "Missing metadata?");
Changed |= processShader(F, MD->second);
}
return Changed;
}
bool SplitPreparePass::processShader(
Function& F, FunctionMetaData &FMD) const
{
auto ShaderTy = FMD.rtInfo.callableShaderType;
hoistValues(F, ShaderTy);
switch (ShaderTy)
{
case ClosestHit:
case AnyHit:
{
// Lower triangle intersection attributes here so we can spill the
// barycentrics if necessary. The procedural intersection attributes
// are allocated in a stack frame so don't need to be spilled. They
// will be processed later.
auto HitTy = m_CGCtx->getHitGroupType(F.getName().str());
if (HitTy == HIT_GROUP_TYPE::TRIANGLES)
{
RTArgs Args(&F, ShaderTy,
HitTy,
m_CGCtx, FMD, FMD.rtInfo.Types);
RTBuilder RTB(&*F.getEntryBlock().getFirstInsertionPt(), *m_CGCtx);
auto* StackPointer = RTB.getAsyncStackPointer();
RTBuilder::lowerIntersectionAttributeFromMemHit(
F, Args, StackPointer);
}
break;
}
default:
break;
}
return true;
}
// For example, say that WorldRayOrigin() is invoked after a TraceRay() call:
// [shader("closesthit")]
// void ClosestHitShader(...)
// {
// TraceRay(...)
// ... = WorldRayOrigin()
// }
// in a closest-hit shader. Given that we set up the RTStack MemRay::org value
// when doing a TraceRay(), if we try to read it in the continuation after the
// TraceRay() we'll read the value from the ray passed to the TraceRay() rather
// than the ray that we're still processing.
//
// Here, we hoist system values to the entry block (at least for now) so that
// they will spill and we can use the correct value in the continuation.
//
// In addition, local pointers aren't valid in continuations since the private
// shader table just contains shaders. We spill the pointer so it can be
// refilled in the continuation if needed.
void SplitPreparePass::hoistValues(
Function &F,
CallableShaderTypeMD Ty) const
{
IGC_ASSERT(RayFlagBits);
bool HoistSysVal = (Ty == ClosestHit || Ty == Miss);
IRBuilder<> IRB(F.getContext());
auto *IP = &*F.getEntryBlock().getFirstInsertionPt();
for (auto II = inst_begin(F), IE = inst_end(F); II != IE; /* empty */)
{
auto* I = &*II++;
if (auto *RIQ = dyn_cast<RayInfoIntrinsic>(I))
{
if (RIQ->getInfoKind() == IGC::RAY_FLAGS)
{
if (auto V = RayFlagBits->getConstant())
{
auto* C = IRB.getInt32((uint32_t)V->getZExtValue());
RIQ->replaceAllUsesWith(C);
RIQ->eraseFromParent();
continue;
}
}
if (HoistSysVal)
RIQ->moveBefore(IP);
}
else if (auto *HK = dyn_cast<HitKindIntrinsic>(I))
{
if (HoistSysVal)
HK->moveBefore(IP);
}
else if (auto *LocalPtr = dyn_cast<LocalBufferPointerIntrinsic>(I))
{
LocalPtr->moveBefore(IP);
}
}
}
namespace IGC
{
Pass* createSplitPreparePass(void)
{
return new SplitPreparePass();
}
} // namespace IGC
|
