[NewPM][CodeGen] Port regallocfast to new pass manager (#94426)

This pull request port `regallocfast` to new pass manager. It exposes
the parameter `filter` to handle different register classes for AMDGPU.
IIUC AMDGPU need to allocate different register classes separately so it
need implement its own `--<reg-class>-regalloc`. Now users can use e.g.
`-passe=regallocfast<filter=sgpr>` to allocate specific register class.
The command line option `--regalloc-npm` is still in work progress, plan
to reuse the syntax of passes, e.g. use
`--regalloc-npm=regallocfast<filter=sgpr>,greedy<filter=vgpr>` to
replace `--sgpr-regalloc` and `--vgpr-regalloc`.
This commit is contained in:
paperchalice 2024-06-07 12:22:42 +08:00 committed by GitHub
parent e96e7f1f33
commit 1bc8b3258e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
36 changed files with 317 additions and 82 deletions

View File

@ -61,7 +61,7 @@ struct MachinePassModel
#ifndef NDEBUG
if constexpr (is_detected<has_get_required_properties_t, PassT>::value) {
auto &MFProps = IR.getProperties();
auto RequiredProperties = PassT::getRequiredProperties();
auto RequiredProperties = this->Pass.getRequiredProperties();
if (!MFProps.verifyRequiredProperties(RequiredProperties)) {
errs() << "MachineFunctionProperties required by " << PassT::name()
<< " pass are not met by function " << IR.getName() << ".\n"
@ -78,9 +78,9 @@ struct MachinePassModel
auto PA = this->Pass.run(IR, AM);
if constexpr (is_detected<has_get_set_properties_t, PassT>::value)
IR.getProperties().set(PassT::getSetProperties());
IR.getProperties().set(this->Pass.getSetProperties());
if constexpr (is_detected<has_get_cleared_properties_t, PassT>::value)
IR.getProperties().reset(PassT::getClearedProperties());
IR.getProperties().reset(this->Pass.getClearedProperties());
return PA;
}

View File

@ -0,0 +1,57 @@
//==- RegAllocFast.h ----------- fast register allocator ----------*-C++-*-==//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_CODEGEN_REGALLOCFAST_H
#define LLVM_CODEGEN_REGALLOCFAST_H
#include "llvm/CodeGen/MachinePassManager.h"
#include "llvm/CodeGen/RegAllocCommon.h"
namespace llvm {
struct RegAllocFastPassOptions {
RegClassFilterFunc Filter = allocateAllRegClasses;
StringRef FilterName = "all";
bool ClearVRegs = true;
};
class RegAllocFastPass : public PassInfoMixin<RegAllocFastPass> {
RegAllocFastPassOptions Opts;
public:
RegAllocFastPass(RegAllocFastPassOptions Opts = RegAllocFastPassOptions())
: Opts(Opts) {}
MachineFunctionProperties getRequiredProperties() {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoPHIs);
}
MachineFunctionProperties getSetProperties() {
if (Opts.ClearVRegs) {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoVRegs);
}
return MachineFunctionProperties();
}
MachineFunctionProperties getClearedProperties() {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::IsSSA);
}
PreservedAnalyses run(MachineFunction &MF, MachineFunctionAnalysisManager &);
void printPipeline(raw_ostream &OS,
function_ref<StringRef(StringRef)> MapClassName2PassName);
};
} // namespace llvm
#endif // LLVM_CODEGEN_REGALLOCFAST_H

View File

@ -43,6 +43,7 @@
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachinePassManager.h"
#include "llvm/CodeGen/PreISelIntrinsicLowering.h"
#include "llvm/CodeGen/RegAllocFast.h"
#include "llvm/CodeGen/ReplaceWithVeclib.h"
#include "llvm/CodeGen/SafeStack.h"
#include "llvm/CodeGen/SelectOptimize.h"
@ -1038,7 +1039,7 @@ void CodeGenPassBuilder<Derived, TargetMachineT>::addTargetRegisterAllocator(
if (Optimized)
addPass(RAGreedyPass());
else
addPass(RAFastPass());
addPass(RegAllocFastPass());
}
/// Find and instantiate the register allocation pass requested by this target

View File

@ -133,6 +133,19 @@ MACHINE_FUNCTION_PASS("require-all-machine-function-properties",
MACHINE_FUNCTION_PASS("trigger-verifier-error", TriggerVerifierErrorPass())
#undef MACHINE_FUNCTION_PASS
#ifndef MACHINE_FUNCTION_PASS_WITH_PARAMS
#define MACHINE_FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, \
PARAMS)
#endif
MACHINE_FUNCTION_PASS_WITH_PARAMS(
"regallocfast", "RegAllocFast",
[](RegAllocFastPassOptions Opts) { return RegAllocFastPass(Opts); },
[PB = this](StringRef Params) {
return parseRegAllocFastPassOptions(*PB, Params);
},
"filter=reg-filter;no-clear-vregs")
#undef MACHINE_FUNCTION_PASS_WITH_PARAMS
// After a pass is converted to new pass manager, its entry should be moved from
// dummy table to the normal one. For example, for a machine function pass,
// DUMMY_MACHINE_FUNCTION_PASS to MACHINE_FUNCTION_PASS.
@ -211,7 +224,6 @@ DUMMY_MACHINE_FUNCTION_PASS("processimpdefs", ProcessImplicitDefsPass)
DUMMY_MACHINE_FUNCTION_PASS("prologepilog", PrologEpilogInserterPass)
DUMMY_MACHINE_FUNCTION_PASS("prologepilog-code", PrologEpilogCodeInserterPass)
DUMMY_MACHINE_FUNCTION_PASS("ra-basic", RABasicPass)
DUMMY_MACHINE_FUNCTION_PASS("ra-fast", RAFastPass)
DUMMY_MACHINE_FUNCTION_PASS("ra-greedy", RAGreedyPass)
DUMMY_MACHINE_FUNCTION_PASS("ra-pbqp", RAPBQPPass)
DUMMY_MACHINE_FUNCTION_PASS("reg-usage-collector", RegUsageInfoCollectorPass)

View File

@ -17,6 +17,7 @@
#include "llvm/Analysis/CGSCCPassManager.h"
#include "llvm/CodeGen/MachinePassManager.h"
#include "llvm/CodeGen/RegAllocCommon.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Passes/OptimizationLevel.h"
#include "llvm/Support/Error.h"
@ -388,6 +389,9 @@ public:
/// returns false.
Error parseAAPipeline(AAManager &AA, StringRef PipelineText);
/// Parse RegClassFilterName to get RegClassFilterFunc.
RegClassFilterFunc parseRegAllocFilter(StringRef RegClassFilterName);
/// Print pass names.
void printPassNames(raw_ostream &OS);
@ -576,6 +580,14 @@ public:
}
/// @}}
/// Register callbacks to parse target specific filter field if regalloc pass
/// needs it. E.g. AMDGPU requires regalloc passes can handle sgpr and vgpr
/// separately.
void registerRegClassFilterParsingCallback(
const std::function<RegClassFilterFunc(StringRef)> &C) {
RegClassFilterParsingCallbacks.push_back(C);
}
/// Register a callback for a top-level pipeline entry.
///
/// If the PassManager type is not given at the top level of the pipeline
@ -792,6 +804,9 @@ private:
ArrayRef<PipelineElement>)>,
2>
MachineFunctionPipelineParsingCallbacks;
// Callbacks to parse `filter` parameter in register allocation passes
SmallVector<std::function<RegClassFilterFunc(StringRef)>, 2>
RegClassFilterParsingCallbacks;
};
/// This utility template takes care of adding require<> and invalidate<>

View File

@ -12,6 +12,7 @@
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/RegAllocFast.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/IndexedMap.h"
@ -174,14 +175,12 @@ private:
DenseMap<const MachineInstr *, uint64_t> Instr2PosIndex;
};
class RegAllocFast : public MachineFunctionPass {
class RegAllocFastImpl {
public:
static char ID;
RegAllocFast(const RegClassFilterFunc F = allocateAllRegClasses,
bool ClearVirtRegs_ = true)
: MachineFunctionPass(ID), ShouldAllocateClass(F),
StackSlotForVirtReg(-1), ClearVirtRegs(ClearVirtRegs_) {}
RegAllocFastImpl(const RegClassFilterFunc F = allocateAllRegClasses,
bool ClearVirtRegs_ = true)
: ShouldAllocateClass(F), StackSlotForVirtReg(-1),
ClearVirtRegs(ClearVirtRegs_) {}
private:
MachineFrameInfo *MFI = nullptr;
@ -197,8 +196,6 @@ private:
/// Maps virtual regs to the frame index where these values are spilled.
IndexedMap<int, VirtReg2IndexFunctor> StackSlotForVirtReg;
bool ClearVirtRegs;
/// Everything we know about a live virtual register.
struct LiveReg {
MachineInstr *LastUse = nullptr; ///< Last instr to use reg.
@ -318,35 +315,11 @@ private:
};
public:
StringRef getPassName() const override { return "Fast Register Allocator"; }
bool ClearVirtRegs;
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
}
MachineFunctionProperties getRequiredProperties() const override {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoPHIs);
}
MachineFunctionProperties getSetProperties() const override {
if (ClearVirtRegs) {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoVRegs);
}
return MachineFunctionProperties();
}
MachineFunctionProperties getClearedProperties() const override {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::IsSSA);
}
bool runOnMachineFunction(MachineFunction &MF);
private:
bool runOnMachineFunction(MachineFunction &MF) override;
void allocateBasicBlock(MachineBasicBlock &MBB);
void addRegClassDefCounts(std::vector<unsigned> &RegClassDefCounts,
@ -408,6 +381,47 @@ private:
void dumpState() const;
};
class RegAllocFast : public MachineFunctionPass {
RegAllocFastImpl Impl;
public:
static char ID;
RegAllocFast(const RegClassFilterFunc F = allocateAllRegClasses,
bool ClearVirtRegs_ = true)
: MachineFunctionPass(ID), Impl(F, ClearVirtRegs_) {}
bool runOnMachineFunction(MachineFunction &MF) override {
return Impl.runOnMachineFunction(MF);
}
StringRef getPassName() const override { return "Fast Register Allocator"; }
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
}
MachineFunctionProperties getRequiredProperties() const override {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoPHIs);
}
MachineFunctionProperties getSetProperties() const override {
if (Impl.ClearVirtRegs) {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::NoVRegs);
}
return MachineFunctionProperties();
}
MachineFunctionProperties getClearedProperties() const override {
return MachineFunctionProperties().set(
MachineFunctionProperties::Property::IsSSA);
}
};
} // end anonymous namespace
char RegAllocFast::ID = 0;
@ -415,18 +429,18 @@ char RegAllocFast::ID = 0;
INITIALIZE_PASS(RegAllocFast, "regallocfast", "Fast Register Allocator", false,
false)
bool RegAllocFast::shouldAllocateRegister(const Register Reg) const {
bool RegAllocFastImpl::shouldAllocateRegister(const Register Reg) const {
assert(Reg.isVirtual());
const TargetRegisterClass &RC = *MRI->getRegClass(Reg);
return ShouldAllocateClass(*TRI, RC);
}
void RegAllocFast::setPhysRegState(MCPhysReg PhysReg, unsigned NewState) {
void RegAllocFastImpl::setPhysRegState(MCPhysReg PhysReg, unsigned NewState) {
for (MCRegUnit Unit : TRI->regunits(PhysReg))
RegUnitStates[Unit] = NewState;
}
bool RegAllocFast::isPhysRegFree(MCPhysReg PhysReg) const {
bool RegAllocFastImpl::isPhysRegFree(MCPhysReg PhysReg) const {
for (MCRegUnit Unit : TRI->regunits(PhysReg)) {
if (RegUnitStates[Unit] != regFree)
return false;
@ -436,7 +450,7 @@ bool RegAllocFast::isPhysRegFree(MCPhysReg PhysReg) const {
/// This allocates space for the specified virtual register to be held on the
/// stack.
int RegAllocFast::getStackSpaceFor(Register VirtReg) {
int RegAllocFastImpl::getStackSpaceFor(Register VirtReg) {
// Find the location Reg would belong...
int SS = StackSlotForVirtReg[VirtReg];
// Already has space allocated?
@ -464,7 +478,7 @@ static bool dominates(InstrPosIndexes &PosIndexes, const MachineInstr &A,
}
/// Returns false if \p VirtReg is known to not live out of the current block.
bool RegAllocFast::mayLiveOut(Register VirtReg) {
bool RegAllocFastImpl::mayLiveOut(Register VirtReg) {
if (MayLiveAcrossBlocks.test(Register::virtReg2Index(VirtReg))) {
// Cannot be live-out if there are no successors.
return !MBB->succ_empty();
@ -517,7 +531,7 @@ bool RegAllocFast::mayLiveOut(Register VirtReg) {
}
/// Returns false if \p VirtReg is known to not be live into the current block.
bool RegAllocFast::mayLiveIn(Register VirtReg) {
bool RegAllocFastImpl::mayLiveIn(Register VirtReg) {
if (MayLiveAcrossBlocks.test(Register::virtReg2Index(VirtReg)))
return !MBB->pred_empty();
@ -536,8 +550,9 @@ bool RegAllocFast::mayLiveIn(Register VirtReg) {
/// Insert spill instruction for \p AssignedReg before \p Before. Update
/// DBG_VALUEs with \p VirtReg operands with the stack slot.
void RegAllocFast::spill(MachineBasicBlock::iterator Before, Register VirtReg,
MCPhysReg AssignedReg, bool Kill, bool LiveOut) {
void RegAllocFastImpl::spill(MachineBasicBlock::iterator Before,
Register VirtReg, MCPhysReg AssignedReg, bool Kill,
bool LiveOut) {
LLVM_DEBUG(dbgs() << "Spilling " << printReg(VirtReg, TRI) << " in "
<< printReg(AssignedReg, TRI));
int FI = getStackSpaceFor(VirtReg);
@ -596,8 +611,8 @@ void RegAllocFast::spill(MachineBasicBlock::iterator Before, Register VirtReg,
}
/// Insert reload instruction for \p PhysReg before \p Before.
void RegAllocFast::reload(MachineBasicBlock::iterator Before, Register VirtReg,
MCPhysReg PhysReg) {
void RegAllocFastImpl::reload(MachineBasicBlock::iterator Before,
Register VirtReg, MCPhysReg PhysReg) {
LLVM_DEBUG(dbgs() << "Reloading " << printReg(VirtReg, TRI) << " into "
<< printReg(PhysReg, TRI) << '\n');
int FI = getStackSpaceFor(VirtReg);
@ -610,7 +625,7 @@ void RegAllocFast::reload(MachineBasicBlock::iterator Before, Register VirtReg,
/// This is not just MBB.begin() because surprisingly we have EH_LABEL
/// instructions marking the begin of a basic block. This means we must insert
/// new instructions after such labels...
MachineBasicBlock::iterator RegAllocFast::getMBBBeginInsertionPoint(
MachineBasicBlock::iterator RegAllocFastImpl::getMBBBeginInsertionPoint(
MachineBasicBlock &MBB, SmallSet<Register, 2> &PrologLiveIns) const {
MachineBasicBlock::iterator I = MBB.begin();
while (I != MBB.end()) {
@ -637,7 +652,7 @@ MachineBasicBlock::iterator RegAllocFast::getMBBBeginInsertionPoint(
}
/// Reload all currently assigned virtual registers.
void RegAllocFast::reloadAtBegin(MachineBasicBlock &MBB) {
void RegAllocFastImpl::reloadAtBegin(MachineBasicBlock &MBB) {
if (LiveVirtRegs.empty())
return;
@ -680,7 +695,7 @@ void RegAllocFast::reloadAtBegin(MachineBasicBlock &MBB) {
/// Handle the direct use of a physical register. Check that the register is
/// not used by a virtreg. Kill the physreg, marking it free. This may add
/// implicit kills to MO->getParent() and invalidate MO.
bool RegAllocFast::usePhysReg(MachineInstr &MI, MCPhysReg Reg) {
bool RegAllocFastImpl::usePhysReg(MachineInstr &MI, MCPhysReg Reg) {
assert(Register::isPhysicalRegister(Reg) && "expected physreg");
bool displacedAny = displacePhysReg(MI, Reg);
setPhysRegState(Reg, regPreAssigned);
@ -688,7 +703,7 @@ bool RegAllocFast::usePhysReg(MachineInstr &MI, MCPhysReg Reg) {
return displacedAny;
}
bool RegAllocFast::definePhysReg(MachineInstr &MI, MCPhysReg Reg) {
bool RegAllocFastImpl::definePhysReg(MachineInstr &MI, MCPhysReg Reg) {
bool displacedAny = displacePhysReg(MI, Reg);
setPhysRegState(Reg, regPreAssigned);
return displacedAny;
@ -697,7 +712,7 @@ bool RegAllocFast::definePhysReg(MachineInstr &MI, MCPhysReg Reg) {
/// Mark PhysReg as reserved or free after spilling any virtregs. This is very
/// similar to defineVirtReg except the physreg is reserved instead of
/// allocated.
bool RegAllocFast::displacePhysReg(MachineInstr &MI, MCPhysReg PhysReg) {
bool RegAllocFastImpl::displacePhysReg(MachineInstr &MI, MCPhysReg PhysReg) {
bool displacedAny = false;
for (MCRegUnit Unit : TRI->regunits(PhysReg)) {
@ -726,7 +741,7 @@ bool RegAllocFast::displacePhysReg(MachineInstr &MI, MCPhysReg PhysReg) {
return displacedAny;
}
void RegAllocFast::freePhysReg(MCPhysReg PhysReg) {
void RegAllocFastImpl::freePhysReg(MCPhysReg PhysReg) {
LLVM_DEBUG(dbgs() << "Freeing " << printReg(PhysReg, TRI) << ':');
MCRegister FirstUnit = *TRI->regunits(PhysReg).begin();
@ -753,7 +768,7 @@ void RegAllocFast::freePhysReg(MCPhysReg PhysReg) {
/// for allocation. Returns 0 when PhysReg is free or disabled with all aliases
/// disabled - it can be allocated directly.
/// \returns spillImpossible when PhysReg or an alias can't be spilled.
unsigned RegAllocFast::calcSpillCost(MCPhysReg PhysReg) const {
unsigned RegAllocFastImpl::calcSpillCost(MCPhysReg PhysReg) const {
for (MCRegUnit Unit : TRI->regunits(PhysReg)) {
switch (unsigned VirtReg = RegUnitStates[Unit]) {
case regFree:
@ -772,8 +787,9 @@ unsigned RegAllocFast::calcSpillCost(MCPhysReg PhysReg) const {
return 0;
}
void RegAllocFast::assignDanglingDebugValues(MachineInstr &Definition,
Register VirtReg, MCPhysReg Reg) {
void RegAllocFastImpl::assignDanglingDebugValues(MachineInstr &Definition,
Register VirtReg,
MCPhysReg Reg) {
auto UDBGValIter = DanglingDbgValues.find(VirtReg);
if (UDBGValIter == DanglingDbgValues.end())
return;
@ -809,8 +825,8 @@ void RegAllocFast::assignDanglingDebugValues(MachineInstr &Definition,
/// This method updates local state so that we know that PhysReg is the
/// proper container for VirtReg now. The physical register must not be used
/// for anything else when this is called.
void RegAllocFast::assignVirtToPhysReg(MachineInstr &AtMI, LiveReg &LR,
MCPhysReg PhysReg) {
void RegAllocFastImpl::assignVirtToPhysReg(MachineInstr &AtMI, LiveReg &LR,
MCPhysReg PhysReg) {
Register VirtReg = LR.VirtReg;
LLVM_DEBUG(dbgs() << "Assigning " << printReg(VirtReg, TRI) << " to "
<< printReg(PhysReg, TRI) << '\n');
@ -824,7 +840,7 @@ void RegAllocFast::assignVirtToPhysReg(MachineInstr &AtMI, LiveReg &LR,
static bool isCoalescable(const MachineInstr &MI) { return MI.isFullCopy(); }
Register RegAllocFast::traceCopyChain(Register Reg) const {
Register RegAllocFastImpl::traceCopyChain(Register Reg) const {
static const unsigned ChainLengthLimit = 3;
unsigned C = 0;
do {
@ -843,7 +859,7 @@ Register RegAllocFast::traceCopyChain(Register Reg) const {
/// Check if any of \p VirtReg's definitions is a copy. If it is follow the
/// chain of copies to check whether we reach a physical register we can
/// coalesce with.
Register RegAllocFast::traceCopies(Register VirtReg) const {
Register RegAllocFastImpl::traceCopies(Register VirtReg) const {
static const unsigned DefLimit = 3;
unsigned C = 0;
for (const MachineInstr &MI : MRI->def_instructions(VirtReg)) {
@ -861,8 +877,8 @@ Register RegAllocFast::traceCopies(Register VirtReg) const {
}
/// Allocates a physical register for VirtReg.
void RegAllocFast::allocVirtReg(MachineInstr &MI, LiveReg &LR, Register Hint0,
bool LookAtPhysRegUses) {
void RegAllocFastImpl::allocVirtReg(MachineInstr &MI, LiveReg &LR,
Register Hint0, bool LookAtPhysRegUses) {
const Register VirtReg = LR.VirtReg;
assert(LR.PhysReg == 0);
@ -950,7 +966,7 @@ void RegAllocFast::allocVirtReg(MachineInstr &MI, LiveReg &LR, Register Hint0,
assignVirtToPhysReg(MI, LR, BestReg);
}
void RegAllocFast::allocVirtRegUndef(MachineOperand &MO) {
void RegAllocFastImpl::allocVirtRegUndef(MachineOperand &MO) {
assert(MO.isUndef() && "expected undef use");
Register VirtReg = MO.getReg();
assert(VirtReg.isVirtual() && "Expected virtreg");
@ -980,8 +996,9 @@ void RegAllocFast::allocVirtRegUndef(MachineOperand &MO) {
/// Variation of defineVirtReg() with special handling for livethrough regs
/// (tied or earlyclobber) that may interfere with preassigned uses.
/// \return true if MI's MachineOperands were re-arranged/invalidated.
bool RegAllocFast::defineLiveThroughVirtReg(MachineInstr &MI, unsigned OpNum,
Register VirtReg) {
bool RegAllocFastImpl::defineLiveThroughVirtReg(MachineInstr &MI,
unsigned OpNum,
Register VirtReg) {
if (!shouldAllocateRegister(VirtReg))
return false;
LiveRegMap::iterator LRI = findLiveVirtReg(VirtReg);
@ -1016,8 +1033,8 @@ bool RegAllocFast::defineLiveThroughVirtReg(MachineInstr &MI, unsigned OpNum,
/// - The value is live out and all uses are in different basic blocks.
///
/// \return true if MI's MachineOperands were re-arranged/invalidated.
bool RegAllocFast::defineVirtReg(MachineInstr &MI, unsigned OpNum,
Register VirtReg, bool LookAtPhysRegUses) {
bool RegAllocFastImpl::defineVirtReg(MachineInstr &MI, unsigned OpNum,
Register VirtReg, bool LookAtPhysRegUses) {
assert(VirtReg.isVirtual() && "Not a virtual register");
if (!shouldAllocateRegister(VirtReg))
return false;
@ -1094,8 +1111,8 @@ bool RegAllocFast::defineVirtReg(MachineInstr &MI, unsigned OpNum,
/// Allocates a register for a VirtReg use.
/// \return true if MI's MachineOperands were re-arranged/invalidated.
bool RegAllocFast::useVirtReg(MachineInstr &MI, MachineOperand &MO,
Register VirtReg) {
bool RegAllocFastImpl::useVirtReg(MachineInstr &MI, MachineOperand &MO,
Register VirtReg) {
assert(VirtReg.isVirtual() && "Not a virtual register");
if (!shouldAllocateRegister(VirtReg))
return false;
@ -1150,8 +1167,8 @@ bool RegAllocFast::useVirtReg(MachineInstr &MI, MachineOperand &MO,
/// Changes operand OpNum in MI the refer the PhysReg, considering subregs.
/// \return true if MI's MachineOperands were re-arranged/invalidated.
bool RegAllocFast::setPhysReg(MachineInstr &MI, MachineOperand &MO,
MCPhysReg PhysReg) {
bool RegAllocFastImpl::setPhysReg(MachineInstr &MI, MachineOperand &MO,
MCPhysReg PhysReg) {
if (!MO.getSubReg()) {
MO.setReg(PhysReg);
MO.setIsRenamable(true);
@ -1190,7 +1207,7 @@ bool RegAllocFast::setPhysReg(MachineInstr &MI, MachineOperand &MO,
#ifndef NDEBUG
void RegAllocFast::dumpState() const {
void RegAllocFastImpl::dumpState() const {
for (unsigned Unit = 1, UnitE = TRI->getNumRegUnits(); Unit != UnitE;
++Unit) {
switch (unsigned VirtReg = RegUnitStates[Unit]) {
@ -1235,7 +1252,7 @@ void RegAllocFast::dumpState() const {
#endif
/// Count number of defs consumed from each register class by \p Reg
void RegAllocFast::addRegClassDefCounts(
void RegAllocFastImpl::addRegClassDefCounts(
std::vector<unsigned> &RegClassDefCounts, Register Reg) const {
assert(RegClassDefCounts.size() == TRI->getNumRegClasses());
@ -1269,7 +1286,7 @@ void RegAllocFast::addRegClassDefCounts(
/// Compute \ref DefOperandIndexes so it contains the indices of "def" operands
/// that are to be allocated. Those are ordered in a way that small classes,
/// early clobbers and livethroughs are allocated first.
void RegAllocFast::findAndSortDefOperandIndexes(const MachineInstr &MI) {
void RegAllocFastImpl::findAndSortDefOperandIndexes(const MachineInstr &MI) {
DefOperandIndexes.clear();
// Track number of defs which may consume a register from the class.
@ -1343,7 +1360,7 @@ static bool isTiedToNotUndef(const MachineOperand &MO) {
return !TiedMO.isUndef();
}
void RegAllocFast::allocateInstruction(MachineInstr &MI) {
void RegAllocFastImpl::allocateInstruction(MachineInstr &MI) {
// The basic algorithm here is:
// 1. Mark registers of def operands as free
// 2. Allocate registers to use operands and place reload instructions for
@ -1605,7 +1622,7 @@ void RegAllocFast::allocateInstruction(MachineInstr &MI) {
}
}
void RegAllocFast::handleDebugValue(MachineInstr &MI) {
void RegAllocFastImpl::handleDebugValue(MachineInstr &MI) {
// Ignore DBG_VALUEs that aren't based on virtual registers. These are
// mostly constants and frame indices.
assert(MI.isDebugValue() && "not a DBG_VALUE*");
@ -1648,7 +1665,7 @@ void RegAllocFast::handleDebugValue(MachineInstr &MI) {
}
}
void RegAllocFast::handleBundle(MachineInstr &MI) {
void RegAllocFastImpl::handleBundle(MachineInstr &MI) {
MachineBasicBlock::instr_iterator BundledMI = MI.getIterator();
++BundledMI;
while (BundledMI->isBundledWithPred()) {
@ -1671,7 +1688,7 @@ void RegAllocFast::handleBundle(MachineInstr &MI) {
}
}
void RegAllocFast::allocateBasicBlock(MachineBasicBlock &MBB) {
void RegAllocFastImpl::allocateBasicBlock(MachineBasicBlock &MBB) {
this->MBB = &MBB;
LLVM_DEBUG(dbgs() << "\nAllocating " << MBB);
@ -1732,7 +1749,7 @@ void RegAllocFast::allocateBasicBlock(MachineBasicBlock &MBB) {
LLVM_DEBUG(MBB.dump());
}
bool RegAllocFast::runOnMachineFunction(MachineFunction &MF) {
bool RegAllocFastImpl::runOnMachineFunction(MachineFunction &MF) {
LLVM_DEBUG(dbgs() << "********** FAST REGISTER ALLOCATION **********\n"
<< "********** Function: " << MF.getName() << '\n');
MRI = &MF.getRegInfo();
@ -1771,6 +1788,36 @@ bool RegAllocFast::runOnMachineFunction(MachineFunction &MF) {
return true;
}
PreservedAnalyses RegAllocFastPass::run(MachineFunction &MF,
MachineFunctionAnalysisManager &) {
RegAllocFastImpl Impl(Opts.Filter, Opts.ClearVRegs);
bool Changed = Impl.runOnMachineFunction(MF);
if (!Changed)
return PreservedAnalyses::all();
auto PA = getMachineFunctionPassPreservedAnalyses();
PA.preserveSet<CFGAnalyses>();
return PA;
}
void RegAllocFastPass::printPipeline(
raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
bool PrintFilterName = Opts.FilterName != "all";
bool PrintNoClearVRegs = !Opts.ClearVRegs;
bool PrintSemicolon = PrintFilterName && PrintNoClearVRegs;
OS << "regallocfast";
if (PrintFilterName || PrintNoClearVRegs) {
OS << '<';
if (PrintFilterName)
OS << "filter=" << Opts.FilterName;
if (PrintSemicolon)
OS << ';';
if (PrintNoClearVRegs)
OS << "no-clear-vregs";
OS << '>';
}
}
FunctionPass *llvm::createFastRegisterAllocator() { return new RegAllocFast(); }
FunctionPass *llvm::createFastRegisterAllocator(RegClassFilterFunc Ftor,

View File

@ -97,6 +97,7 @@
#include "llvm/CodeGen/MachinePassManager.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PreISelIntrinsicLowering.h"
#include "llvm/CodeGen/RegAllocFast.h"
#include "llvm/CodeGen/SafeStack.h"
#include "llvm/CodeGen/SelectOptimize.h"
#include "llvm/CodeGen/ShadowStackGCLowering.h"
@ -1163,6 +1164,38 @@ Expected<SmallVector<std::string, 0>> parseInternalizeGVs(StringRef Params) {
return Expected<SmallVector<std::string, 0>>(std::move(PreservedGVs));
}
Expected<RegAllocFastPassOptions>
parseRegAllocFastPassOptions(PassBuilder &PB, StringRef Params) {
RegAllocFastPassOptions Opts;
while (!Params.empty()) {
StringRef ParamName;
std::tie(ParamName, Params) = Params.split(';');
if (ParamName.consume_front("filter=")) {
RegClassFilterFunc Filter = PB.parseRegAllocFilter(ParamName);
if (!Filter) {
return make_error<StringError>(
formatv("invalid regallocfast register filter '{0}' ", ParamName)
.str(),
inconvertibleErrorCode());
}
Opts.Filter = Filter;
Opts.FilterName = ParamName;
continue;
}
if (ParamName == "no-clear-vregs") {
Opts.ClearVRegs = false;
continue;
}
return make_error<StringError>(
formatv("invalid regallocfast pass parameter '{0}' ", ParamName).str(),
inconvertibleErrorCode());
}
return Opts;
}
} // namespace
/// Tests whether a pass name starts with a valid prefix for a default pipeline
@ -1296,6 +1329,11 @@ static bool isMachineFunctionPassName(StringRef Name, CallbacksT &Callbacks) {
#define MACHINE_FUNCTION_PASS(NAME, CREATE_PASS) \
if (Name == NAME) \
return true;
#define MACHINE_FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, \
PARAMS) \
if (PassBuilder::checkParametrizedPassName(Name, NAME)) \
return true;
#define MACHINE_FUNCTION_ANALYSIS(NAME, CREATE_PASS) \
if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \
return true;
@ -1925,6 +1963,15 @@ Error PassBuilder::parseMachinePass(MachineFunctionPassManager &MFPM,
MFPM.addPass(CREATE_PASS); \
return Error::success(); \
}
#define MACHINE_FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, \
PARAMS) \
if (checkParametrizedPassName(Name, NAME)) { \
auto Params = parsePassParameters(PARSER, Name, NAME); \
if (!Params) \
return Params.takeError(); \
MFPM.addPass(CREATE_PASS(Params.get())); \
return Error::success(); \
}
#include "llvm/Passes/MachinePassRegistry.def"
for (auto &C : MachineFunctionPipelineParsingCallbacks)
@ -2172,6 +2219,15 @@ Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) {
return Error::success();
}
RegClassFilterFunc PassBuilder::parseRegAllocFilter(StringRef FilterName) {
if (FilterName == "all")
return allocateAllRegClasses;
for (auto &C : RegClassFilterParsingCallbacks)
if (auto F = C(FilterName))
return F;
return nullptr;
}
static void printPassName(StringRef PassName, raw_ostream &OS) {
OS << " " << PassName << "\n";
}

View File

@ -740,6 +740,15 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(
if (EnableLowerModuleLDS)
PM.addPass(AMDGPULowerModuleLDSPass(*this));
});
PB.registerRegClassFilterParsingCallback(
[](StringRef FilterName) -> RegClassFilterFunc {
if (FilterName == "sgpr")
return onlyAllocateSGPRs;
if (FilterName == "vgpr")
return onlyAllocateVGPRs;
return nullptr;
});
}
int64_t AMDGPUTargetMachine::getNullPointerValue(unsigned AddrSpace) {

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple aarch64-apple-ios -run-pass regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple aarch64-apple-ios -passes=regallocfast -o - %s | FileCheck %s
# This test used to crash the fast register alloc.
# Basically, when a basic block has liveins, the fast regalloc
# was deferencing the begin iterator of this block. However,

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=regallocfast -o - %s | FileCheck %s
# Make sure incorrect kills aren't emitted on vcc

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=gfx902 -verify-machineinstrs -run-pass=regallocfast %s -o - | FileCheck --check-prefix=GCN %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx902 -verify-machineinstrs -passes=regallocfast %s -o - | FileCheck --check-prefix=GCN %s
---
name: fast_regalloc_bundle_handling

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -passes=regallocfast -o - %s | FileCheck %s
# This would hit "Illegal subregister index for physical register" verifier error since
# tied operands would skip dropping the subregister index.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=regallocfast -o - %s | FileCheck -check-prefix=GCN %s
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=regallocfast -o - %s | FileCheck -check-prefix=GCN %s
---
name: self_loop_single_def_use

View File

@ -1,7 +1,9 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=gfx908 -run-pass=regallocfast -o - %s | FileCheck -check-prefix=GFX908-SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx908 -passes=regallocfast -o - %s | FileCheck -check-prefix=GFX908-SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx908 -run-pass=regallocfast,prologepilog -o - %s | FileCheck -check-prefix=GFX908-EXPANDED %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=regallocfast -o - %s | FileCheck -check-prefix=GFX90A-SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -passes=regallocfast -o - %s | FileCheck -check-prefix=GFX90A-SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=regallocfast,prologepilog -o - %s | FileCheck -check-prefix=GFX90A-EXPANDED %s
---

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 192 bit registers works. We have to

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 224 bit registers works.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 288 bit registers works.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 320 bit registers works.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 352 bit registers works.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -passes=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 384 bit registers works.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -verify-machineinstrs -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=regallocfast -o - %s | FileCheck %s
---
name: bar

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=armv7-apple-ios -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=armv7-apple-ios -passes=regallocfast -o - %s | FileCheck %s
# tBX_RET uses an implicit vreg with a sub-register. That implicit use will

View File

@ -1,6 +1,7 @@
# RUN: llc -run-pass=greedy -debug-pass=Arguments -o - %s | FileCheck %s
# RUN: llc -run-pass=regallocbasic -debug-pass=Arguments -o - %s | FileCheck %s
# RUN: llc -run-pass=regallocfast -debug-pass=Arguments -o - %s | FileCheck %s
# RUN: llc -passes=regallocfast -o - %s | FileCheck %s
# Check that passes are initialized correctly, so that it's possible to
# use -run-pass.

View File

@ -1,4 +1,5 @@
# RUN: llc -o - %s -mtriple=powerpc64-- -run-pass=regallocfast | FileCheck %s
# RUN: llc -o - %s -mtriple=powerpc64-- -passes=regallocfast | FileCheck %s
---
# CHECK-LABEL: name: func
name: func

View File

@ -1,4 +1,5 @@
# RUN: llc -verify-machineinstrs -run-pass regallocfast -mtriple s390x-ibm-linux -o - %s | FileCheck %s
# RUN: llc -verify-machineinstrs -passes=regallocfast -mtriple s390x-ibm-linux -o - %s | FileCheck %s
--- |
@g_167 = external global [5 x i64], align 8

View File

@ -2,6 +2,7 @@
# RUN: llc -mtriple thumbv6m-arm-none-eabi -run-pass regallocbasic %s -o - | FileCheck %s
# RUN: llc -mtriple thumbv6m-arm-none-eabi -run-pass greedy %s -o - | FileCheck %s
# RUN: llc -mtriple thumbv6m-arm-none-eabi -run-pass regallocfast %s -o - | FileCheck %s --check-prefix=FAST
# RUN: llc -mtriple thumbv6m-arm-none-eabi -passes=regallocfast %s -o - | FileCheck %s --check-prefix=FAST
...
---

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -run-pass regallocfast %s -o - | FileCheck %s
# RUN: llc -passes=regallocfast %s -o - | FileCheck %s
# This test examines register allocation and spilling with Fast Register
# Allocator. The test uses inline assembler that requests an input variable to

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=i386-unknown-linux-musl -verify-machineinstrs -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=i386-unknown-linux-musl -verify-machineinstrs -passes=regallocfast -o - %s | FileCheck %s
# Test for correct management of allocatable and non-allocatable
# live-ins in fastregalloc

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=i386-unknown-linux-musl -verify-machineinstrs -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=i386-unknown-linux-musl -verify-machineinstrs -passes=regallocfast -o - %s | FileCheck %s
# Make sure this case doesn't assert or try to assign $ecx to %1 on
# SHRD32rrCL

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -run-pass=regallocfast -verify-machineinstrs %s -o - | FileCheck %s
# RUN: llc -passes=regallocfast -verify-machineinstrs %s -o - | FileCheck %s
--- |
; ModuleID = 'x.c'
source_filename = "x.c"

View File

@ -1,4 +1,5 @@
# RUN: llc -mtriple=x86_64-- -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=x86_64-- -passes=regallocfast -o - %s | FileCheck %s
...
---

View File

@ -1,4 +1,5 @@
# RUN: llc -mtriple=x86_64-- -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=x86_64-- -passes=regallocfast -o - %s | FileCheck %s
# If the tied use is undef value, fastregalloc should free the def register.
# There is no reload needed for the undef value.

View File

@ -1,5 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=x86_64-grtev4-linux-gnu -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=x86_64-grtev4-linux-gnu -passes=regallocfast -o - %s | FileCheck %s
# Bug 41973. Make sure %12 is detected as live out of %bb.0, even
# though the use is allocated before the def block %bb.3. Previously

View File

@ -1,4 +1,5 @@
# RUN: llc -mtriple=x86_64-- -run-pass=regallocfast -o - %s | FileCheck %s
# RUN: llc -mtriple=x86_64-- -passes=regallocfast -o - %s | FileCheck %s
# Check that fastregalloc does not displace register assigned to tied def when
# RegMask operand is present. STATEPOINT is an example of such instruction.

View File

@ -1,4 +1,5 @@
# RUN: llc -o - -mtriple=x86_64-- -run-pass=regallocfast %s | FileCheck %s
# RUN: llc -o - -mtriple=x86_64-- -passes=regallocfast %s | FileCheck %s
# Fast regalloc used to not collect physical register definitions
# before walking and assigning the virtual definition.
# Therefore it was possible for a virtual definition to end up

View File

@ -0,0 +1,10 @@
# REQUIRES: amdgpu-registered-target
# RUN: llc -mtriple=amdgcn --passes='regallocfast<filter=sgpr>' --print-pipeline-passes %s | FileCheck %s --check-prefix=PASS
# RUN: not llc -mtriple=amdgcn --passes='regallocfast<filter=bad-filter>' --print-pipeline-passes %s 2>&1 | FileCheck %s --check-prefix=BAD-FILTER
# PASS: regallocfast<filter=sgpr>
# BAD-FILTER: invalid regallocfast register filter 'bad-filter'
---
name: f
...