1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 11:02:59 +02:00
llvm-mirror/lib/CodeGen/RegisterClassInfo.cpp
Christudasan Devadasan e86b7c773c Support a list of CostPerUse values
This patch allows targets to define multiple cost
values for each register so that the cost model
can be more flexible and better used during the
register allocation as per the target requirements.

For AMDGPU the VGPR allocation will be more efficient
if the register cost can be associated dynamically
based on the calling convention.

Reviewed By: qcolombet

Differential Revision: https://reviews.llvm.org/D86836
2021-01-29 10:14:52 +05:30

204 lines
7.0 KiB
C++

//===- RegisterClassInfo.cpp - Dynamic Register Class Info ----------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file implements the RegisterClassInfo class which provides dynamic
// information about target register classes. Callee-saved vs. caller-saved and
// reserved registers depend on calling conventions and other dynamic
// information, so some things cannot be determined statically.
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/RegisterClassInfo.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
using namespace llvm;
#define DEBUG_TYPE "regalloc"
static cl::opt<unsigned>
StressRA("stress-regalloc", cl::Hidden, cl::init(0), cl::value_desc("N"),
cl::desc("Limit all regclasses to N registers"));
RegisterClassInfo::RegisterClassInfo() = default;
void RegisterClassInfo::runOnMachineFunction(const MachineFunction &mf) {
bool Update = false;
MF = &mf;
// Allocate new array the first time we see a new target.
if (MF->getSubtarget().getRegisterInfo() != TRI) {
TRI = MF->getSubtarget().getRegisterInfo();
RegClass.reset(new RCInfo[TRI->getNumRegClasses()]);
Update = true;
}
// Does this MF have different CSRs?
assert(TRI && "no register info set");
// Get the callee saved registers.
const MCPhysReg *CSR = MF->getRegInfo().getCalleeSavedRegs();
if (Update || CSR != CalleeSavedRegs) {
// Build a CSRAlias map. Every CSR alias saves the last
// overlapping CSR.
CalleeSavedAliases.assign(TRI->getNumRegs(), 0);
for (const MCPhysReg *I = CSR; *I; ++I)
for (MCRegAliasIterator AI(*I, TRI, true); AI.isValid(); ++AI)
CalleeSavedAliases[*AI] = *I;
Update = true;
}
CalleeSavedRegs = CSR;
RegCosts = TRI->getRegisterCosts(*MF);
// Different reserved registers?
const BitVector &RR = MF->getRegInfo().getReservedRegs();
if (Reserved.size() != RR.size() || RR != Reserved) {
Update = true;
Reserved = RR;
}
// Invalidate cached information from previous function.
if (Update) {
unsigned NumPSets = TRI->getNumRegPressureSets();
PSetLimits.reset(new unsigned[NumPSets]);
std::fill(&PSetLimits[0], &PSetLimits[NumPSets], 0);
++Tag;
}
}
/// compute - Compute the preferred allocation order for RC with reserved
/// registers filtered out. Volatile registers come first followed by CSR
/// aliases ordered according to the CSR order specified by the target.
void RegisterClassInfo::compute(const TargetRegisterClass *RC) const {
assert(RC && "no register class given");
RCInfo &RCI = RegClass[RC->getID()];
auto &STI = MF->getSubtarget();
// Raw register count, including all reserved regs.
unsigned NumRegs = RC->getNumRegs();
if (!RCI.Order)
RCI.Order.reset(new MCPhysReg[NumRegs]);
unsigned N = 0;
SmallVector<MCPhysReg, 16> CSRAlias;
uint8_t MinCost = uint8_t(~0u);
uint8_t LastCost = uint8_t(~0u);
unsigned LastCostChange = 0;
// FIXME: Once targets reserve registers instead of removing them from the
// allocation order, we can simply use begin/end here.
ArrayRef<MCPhysReg> RawOrder = RC->getRawAllocationOrder(*MF);
for (unsigned i = 0; i != RawOrder.size(); ++i) {
unsigned PhysReg = RawOrder[i];
// Remove reserved registers from the allocation order.
if (Reserved.test(PhysReg))
continue;
uint8_t Cost = RegCosts[PhysReg];
MinCost = std::min(MinCost, Cost);
if (CalleeSavedAliases[PhysReg] &&
!STI.ignoreCSRForAllocationOrder(*MF, PhysReg))
// PhysReg aliases a CSR, save it for later.
CSRAlias.push_back(PhysReg);
else {
if (Cost != LastCost)
LastCostChange = N;
RCI.Order[N++] = PhysReg;
LastCost = Cost;
}
}
RCI.NumRegs = N + CSRAlias.size();
assert(RCI.NumRegs <= NumRegs && "Allocation order larger than regclass");
// CSR aliases go after the volatile registers, preserve the target's order.
for (unsigned i = 0, e = CSRAlias.size(); i != e; ++i) {
unsigned PhysReg = CSRAlias[i];
uint8_t Cost = RegCosts[PhysReg];
if (Cost != LastCost)
LastCostChange = N;
RCI.Order[N++] = PhysReg;
LastCost = Cost;
}
// Register allocator stress test. Clip register class to N registers.
if (StressRA && RCI.NumRegs > StressRA)
RCI.NumRegs = StressRA;
// Check if RC is a proper sub-class.
if (const TargetRegisterClass *Super =
TRI->getLargestLegalSuperClass(RC, *MF))
if (Super != RC && getNumAllocatableRegs(Super) > RCI.NumRegs)
RCI.ProperSubClass = true;
RCI.MinCost = MinCost;
RCI.LastCostChange = LastCostChange;
LLVM_DEBUG({
dbgs() << "AllocationOrder(" << TRI->getRegClassName(RC) << ") = [";
for (unsigned I = 0; I != RCI.NumRegs; ++I)
dbgs() << ' ' << printReg(RCI.Order[I], TRI);
dbgs() << (RCI.ProperSubClass ? " ] (sub-class)\n" : " ]\n");
});
// RCI is now up-to-date.
RCI.Tag = Tag;
}
/// This is not accurate because two overlapping register sets may have some
/// nonoverlapping reserved registers. However, computing the allocation order
/// for all register classes would be too expensive.
unsigned RegisterClassInfo::computePSetLimit(unsigned Idx) const {
const TargetRegisterClass *RC = nullptr;
unsigned NumRCUnits = 0;
for (const TargetRegisterClass *C : TRI->regclasses()) {
const int *PSetID = TRI->getRegClassPressureSets(C);
for (; *PSetID != -1; ++PSetID) {
if ((unsigned)*PSetID == Idx)
break;
}
if (*PSetID == -1)
continue;
// Found a register class that counts against this pressure set.
// For efficiency, only compute the set order for the largest set.
unsigned NUnits = TRI->getRegClassWeight(C).WeightLimit;
if (!RC || NUnits > NumRCUnits) {
RC = C;
NumRCUnits = NUnits;
}
}
assert(RC && "Failed to find register class");
compute(RC);
unsigned NAllocatableRegs = getNumAllocatableRegs(RC);
unsigned RegPressureSetLimit = TRI->getRegPressureSetLimit(*MF, Idx);
// If all the regs are reserved, return raw RegPressureSetLimit.
// One example is VRSAVERC in PowerPC.
// Avoid returning zero, getRegPressureSetLimit(Idx) assumes computePSetLimit
// return non-zero value.
if (NAllocatableRegs == 0)
return RegPressureSetLimit;
unsigned NReserved = RC->getNumRegs() - NAllocatableRegs;
return RegPressureSetLimit - TRI->getRegClassWeight(RC).RegWeight * NReserved;
}