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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
//===----------------------------------------------------------------------===//
///
/// Some workloads may finish operating on RT Memory regions (e.g., SWStack)
/// early on in some of the shaders and do quite a bit more work after.
/// For example:
///
/// [shader("closesthit")]
/// void CHS(inout Payload payload, in BuiltInTriangleIntersectionAttributes attr)
/// {
/// payload.value = (uint)attr.barycentrics.x;
/// // computations ...
/// // texture writes, etc.
/// StackIDRelease(); // inlined from raygen
/// }
///
/// Currently, we inject StackIDRelease calls at the termination of raygen
/// shaders which may then be inlined into other shaders.
///
/// This pass tries to schedule the StackIDRelease as early as possible to
/// free up the stack ID to allow further raygen shaders to execute sooner
/// to improve RT hardware utilization.
///
/// The above would then be:
///
/// [shader("closesthit")]
/// void CHS(inout Payload payload, in BuiltInTriangleIntersectionAttributes attr)
/// {
/// payload.value = (uint)attr.barycentrics.x;
/// StackIDRelease(); // inlined from raygen
/// // computations ...
/// // texture writes, etc.
/// }
///
//===----------------------------------------------------------------------===//
#include "RTBuilder.h"
#include "Compiler/IGCPassSupport.h"
#include "RayTracingAddressSpaceAliasAnalysis.h"
#include "common/LLVMWarningsPush.hpp"
#include <llvm/IR/Dominators.h>
#include <llvm/Analysis/PostDominators.h>
#include <llvm/Analysis/LoopInfo.h>
#include "llvmWrapper/Analysis/AliasAnalysis.h"
#include "llvmWrapper/Analysis/MemoryLocation.h"
#include "common/LLVMWarningsPop.hpp"
using namespace llvm;
using namespace IGC;
class StackIDSchedulingPass : public FunctionPass
{
public:
StackIDSchedulingPass(): FunctionPass(ID)
{
initializeStackIDSchedulingPassPass(*PassRegistry::getPassRegistry());
}
void getAnalysisUsage(AnalysisUsage &AU) const override
{
AU.setPreservesCFG();
AU.addRequired<CodeGenContextWrapper>();
AU.addRequired<DominatorTreeWrapperPass>();
AU.addRequired<PostDominatorTreeWrapperPass>();
AU.addRequired<LoopInfoWrapperPass>();
AU.addRequired<RayTracingAddressSpaceAAWrapperPass>();
}
bool runOnFunction(Function &M) override;
StringRef getPassName() const override
{
return "StackIDSchedulingPass";
}
static char ID;
private:
const DominatorTree* DT = nullptr;
const PostDominatorTree* PDT = nullptr;
const LoopInfo* LI = nullptr;
RayTracingAddressSpaceAAResult* AA = nullptr;
private:
// Maps blocks that contain at least one "hazard instruction" (i.e., an
// instruction for which isSafe(I, SRI) returns false) to the last
// instruction in that block that is a hazard.
using HazardMap = std::unordered_map<const BasicBlock*, Instruction*>;
ModRefInfo getModRefInfo(const CallBase *Call, const MemoryLocation &Loc);
// returns true if the StackIDRelease could safely be scheduled across `I`.
bool isSafe(const Instruction* I, const StackIDReleaseIntrinsic *SRI);
// returns true if there is not path from `From` to its `IDom` that contains
// a hazard
bool isSafeToMove(
const BasicBlock* From, const BasicBlock* IDom, const HazardMap& Hazards) const;
BasicBlock* schedule(
StackIDReleaseIntrinsic* SRI, const HazardMap& Hazards) const;
};
char StackIDSchedulingPass::ID = 0;
// Register pass to igc-opt
#define PASS_FLAG2 "stack-id-scheduling-pass"
#define PASS_DESCRIPTION2 "Schedule Stack ID Release earlier in shader"
#define PASS_CFG_ONLY2 false
#define PASS_ANALYSIS2 false
IGC_INITIALIZE_PASS_BEGIN(StackIDSchedulingPass, PASS_FLAG2, PASS_DESCRIPTION2, PASS_CFG_ONLY2, PASS_ANALYSIS2)
IGC_INITIALIZE_PASS_DEPENDENCY(CodeGenContextWrapper)
IGC_INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
IGC_INITIALIZE_PASS_DEPENDENCY(PostDominatorTreeWrapperPass)
IGC_INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
IGC_INITIALIZE_PASS_DEPENDENCY(RayTracingAddressSpaceAAWrapperPass)
IGC_INITIALIZE_PASS_END(StackIDSchedulingPass, PASS_FLAG2, PASS_DESCRIPTION2, PASS_CFG_ONLY2, PASS_ANALYSIS2)
bool StackIDSchedulingPass::isSafeToMove(
const BasicBlock* From, const BasicBlock* IDom, const HazardMap& Hazards) const
{
SmallPtrSet<const BasicBlock*, 32> Visited;
SmallVector<const BasicBlock*, 8> WorkList{ From };
do {
auto* BB = WorkList.pop_back_val();
if (!Visited.insert(BB).second)
continue;
if (BB == IDom)
continue;
if (Hazards.count(BB) != 0)
return false;
WorkList.append(pred_begin(BB), pred_end(BB));
} while (!WorkList.empty());
return true;
}
BasicBlock* StackIDSchedulingPass::schedule(
StackIDReleaseIntrinsic* SRI, const HazardMap& Hazards) const
{
// Given that there could conceivably be an unreachable BB with a
// StackIDRelease and all nodes in `DT` can dominate it, we just bail here.
if (!DT->isReachableFromEntry(SRI->getParent()))
return nullptr;
auto getIDom = [&](const BasicBlock& BB) -> DomTreeNode*
{
auto* Node = DT->getNode(&BB);
if (!Node)
return nullptr;
return Node->getIDom();
};
BasicBlock* CurBB = SRI->getParent();
while (DomTreeNode* NextNode = getIDom(*CurBB))
{
// These conditions guarantee that the scheduled `SRI` will be executed
// iff it executes in the original program.
auto* NextBB = NextNode->getBlock();
if (!PDT->dominates(SRI->getParent(), NextBB))
return CurBB;
// Ensure that we don't sink `SRI` into a loop since it must execute
// at most once.
if (LI->getLoopFor(NextBB))
return CurBB;
if (!isSafeToMove(CurBB, NextBB, Hazards))
return CurBB;
CurBB = NextBB;
}
return CurBB;
}
ModRefInfo StackIDSchedulingPass::getModRefInfo(
const CallBase* Call, const MemoryLocation& Loc)
{
AAQueryInfo AAQIP = IGCLLVM::makeAAQueryInfo();
return AA->getModRefInfo(Call, Loc, AAQIP);
}
bool StackIDSchedulingPass::isSafe(
const Instruction* I, const StackIDReleaseIntrinsic* SRI)
{
using IGCLLVM::LocationSize;
if (!I->mayReadOrWriteMemory())
return true;
auto checkVal = [&](const Value* V)
{
if (!V->getType()->isPointerTy())
return true;
ModRefInfo Result = getModRefInfo(SRI, MemoryLocation(V, LocationSize::beforeOrAfterPointer()));
return isNoModRef(Result);
};
// Examine all the argument to see if any may alias our `SRI`.
for (auto& Op : I->operands())
{
if (!checkVal(Op))
return false;
}
return checkVal(I);
}
bool StackIDSchedulingPass::runOnFunction(Function &F)
{
auto *Ctx = getAnalysis<CodeGenContextWrapper>().getCodeGenContext();
DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
PDT = &getAnalysis<PostDominatorTreeWrapperPass>().getPostDomTree();
LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
AA = &getAnalysis<RayTracingAddressSpaceAAWrapperPass>().getResult();
RTBuilder RTB{ F.getContext(), *Ctx };
// Insert a dummy release temporarily for the purpose of having something
// to pass to alias analysis.
RTB.SetInsertPoint(&*F.getEntryBlock().begin());
auto* TempSRI = RTB.CreateStackIDRelease(RTB.getInt16(-1));
bool Changed = false;
HazardMap HazardBBs;
// These are the releases that can potentially be scheduled. Any release
// that already has hazards in its BB can't go anywhere but farther up in
// that same BB.
SmallVector<StackIDReleaseIntrinsic*, 4> Releases;
for (auto& BB : F)
{
StackIDReleaseIntrinsic* CurRelease = nullptr;
Instruction* CurHazard = nullptr;
for (auto& I : reverse(BB))
{
if (auto* SRI = dyn_cast<StackIDReleaseIntrinsic>(&I))
{
// We only need to keep track of a single release in a given
// BB because, if there were more than one, we'd already have
// undefined behavior.
if (SRI != TempSRI)
CurRelease = SRI;
}
else if (!isSafe(&I, TempSRI))
{
CurHazard = &I;
break;
}
}
if (CurRelease)
{
if (CurHazard)
{
RTB.SetInsertPoint(CurHazard->getNextNode());
RTB.CreateStackIDRelease();
CurRelease->eraseFromParent();
Changed = true;
}
else
{
Releases.push_back(CurRelease);
}
}
if (CurHazard)
HazardBBs[&BB] = CurHazard;
}
// This doesn't count as `Changed` because we temporarily added this
// instruction then deleted it leaving the function unchanged unless
// otherwise modified.
TempSRI->eraseFromParent();
for (auto* SRI : Releases)
{
if (auto* NewBB = schedule(SRI, HazardBBs))
{
auto I = HazardBBs.find(NewBB);
auto* MovePt = (I == HazardBBs.end()) ?
&*NewBB->getFirstInsertionPt() : I->second->getNextNode();
RTB.SetInsertPoint(MovePt);
RTB.CreateStackIDRelease();
SRI->eraseFromParent();
Changed = true;
}
}
return Changed;
}
namespace IGC
{
Pass* createStackIDSchedulingPass(void)
{
return new StackIDSchedulingPass();
}
} // namespace IGC
|