X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=utils%2FTableGen%2FCodeGenRegisters.cpp;h=f2eef4f68f5a7a18119e00329a0b32f64d04d1a0;hb=e778f82a1e33826ab012bb970a406c9acf37349b;hp=bd1eefe36f48897cf039821620af41082dc26614;hpb=c72e08b4a95e494d3bfdf1262ea8b28f614ac40e;p=oota-llvm.git diff --git a/utils/TableGen/CodeGenRegisters.cpp b/utils/TableGen/CodeGenRegisters.cpp index bd1eefe36f4..f2eef4f68f5 100644 --- a/utils/TableGen/CodeGenRegisters.cpp +++ b/utils/TableGen/CodeGenRegisters.cpp @@ -12,13 +12,17 @@ // //===----------------------------------------------------------------------===// +#define DEBUG_TYPE "regalloc-emitter" + #include "CodeGenRegisters.h" #include "CodeGenTarget.h" -#include "llvm/TableGen/Error.h" #include "llvm/ADT/IntEqClasses.h" -#include "llvm/ADT/SmallVector.h" #include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringExtras.h" +#include "llvm/ADT/Twine.h" +#include "llvm/Support/Debug.h" +#include "llvm/TableGen/Error.h" using namespace llvm; @@ -27,19 +31,18 @@ using namespace llvm; //===----------------------------------------------------------------------===// CodeGenSubRegIndex::CodeGenSubRegIndex(Record *R, unsigned Enum) - : TheDef(R), - EnumValue(Enum) -{} - -std::string CodeGenSubRegIndex::getNamespace() const { - if (TheDef->getValue("Namespace")) - return TheDef->getValueAsString("Namespace"); - else - return ""; + : TheDef(R), EnumValue(Enum), LaneMask(0), AllSuperRegsCovered(true) { + Name = R->getName(); + if (R->getValue("Namespace")) + Namespace = R->getValueAsString("Namespace"); + Size = R->getValueAsInt("Size"); + Offset = R->getValueAsInt("Offset"); } -const std::string &CodeGenSubRegIndex::getName() const { - return TheDef->getName(); +CodeGenSubRegIndex::CodeGenSubRegIndex(StringRef N, StringRef Nspace, + unsigned Enum) + : TheDef(0), Name(N), Namespace(Nspace), Size(-1), Offset(-1), + EnumValue(Enum), LaneMask(0), AllSuperRegsCovered(true) { } std::string CodeGenSubRegIndex::getQualifiedName() const { @@ -51,28 +54,51 @@ std::string CodeGenSubRegIndex::getQualifiedName() const { } void CodeGenSubRegIndex::updateComponents(CodeGenRegBank &RegBank) { - std::vector Comps = TheDef->getValueAsListOfDefs("ComposedOf"); - if (Comps.empty()) + if (!TheDef) return; - if (Comps.size() != 2) - throw TGError(TheDef->getLoc(), "ComposedOf must have exactly two entries"); - CodeGenSubRegIndex *A = RegBank.getSubRegIdx(Comps[0]); - CodeGenSubRegIndex *B = RegBank.getSubRegIdx(Comps[1]); - CodeGenSubRegIndex *X = A->addComposite(B, this); - if (X) - throw TGError(TheDef->getLoc(), "Ambiguous ComposedOf entries"); -} -void CodeGenSubRegIndex::cleanComposites() { - // Clean out redundant mappings of the form this+X -> X. - for (CompMap::iterator i = Composed.begin(), e = Composed.end(); i != e;) { - CompMap::iterator j = i; - ++i; - if (j->first == j->second) - Composed.erase(j); + std::vector Comps = TheDef->getValueAsListOfDefs("ComposedOf"); + if (!Comps.empty()) { + if (Comps.size() != 2) + PrintFatalError(TheDef->getLoc(), + "ComposedOf must have exactly two entries"); + CodeGenSubRegIndex *A = RegBank.getSubRegIdx(Comps[0]); + CodeGenSubRegIndex *B = RegBank.getSubRegIdx(Comps[1]); + CodeGenSubRegIndex *X = A->addComposite(B, this); + if (X) + PrintFatalError(TheDef->getLoc(), "Ambiguous ComposedOf entries"); + } + + std::vector Parts = + TheDef->getValueAsListOfDefs("CoveringSubRegIndices"); + if (!Parts.empty()) { + if (Parts.size() < 2) + PrintFatalError(TheDef->getLoc(), + "CoveredBySubRegs must have two or more entries"); + SmallVector IdxParts; + for (unsigned i = 0, e = Parts.size(); i != e; ++i) + IdxParts.push_back(RegBank.getSubRegIdx(Parts[i])); + RegBank.addConcatSubRegIndex(IdxParts, this); } } +unsigned CodeGenSubRegIndex::computeLaneMask() { + // Already computed? + if (LaneMask) + return LaneMask; + + // Recursion guard, shouldn't be required. + LaneMask = ~0u; + + // The lane mask is simply the union of all sub-indices. + unsigned M = 0; + for (CompMap::iterator I = Composed.begin(), E = Composed.end(); I != E; ++I) + M |= I->second->computeLaneMask(); + assert(M && "Missing lane mask, sub-register cycle?"); + LaneMask = M; + return LaneMask; +} + //===----------------------------------------------------------------------===// // CodeGenRegister //===----------------------------------------------------------------------===// @@ -82,9 +108,43 @@ CodeGenRegister::CodeGenRegister(Record *R, unsigned Enum) EnumValue(Enum), CostPerUse(R->getValueAsInt("CostPerUse")), CoveredBySubRegs(R->getValueAsBit("CoveredBySubRegs")), - SubRegsComplete(false) + NumNativeRegUnits(0), + SubRegsComplete(false), + SuperRegsComplete(false), + TopoSig(~0u) {} +void CodeGenRegister::buildObjectGraph(CodeGenRegBank &RegBank) { + std::vector SRIs = TheDef->getValueAsListOfDefs("SubRegIndices"); + std::vector SRs = TheDef->getValueAsListOfDefs("SubRegs"); + + if (SRIs.size() != SRs.size()) + PrintFatalError(TheDef->getLoc(), + "SubRegs and SubRegIndices must have the same size"); + + for (unsigned i = 0, e = SRIs.size(); i != e; ++i) { + ExplicitSubRegIndices.push_back(RegBank.getSubRegIdx(SRIs[i])); + ExplicitSubRegs.push_back(RegBank.getReg(SRs[i])); + } + + // Also compute leading super-registers. Each register has a list of + // covered-by-subregs super-registers where it appears as the first explicit + // sub-register. + // + // This is used by computeSecondarySubRegs() to find candidates. + if (CoveredBySubRegs && !ExplicitSubRegs.empty()) + ExplicitSubRegs.front()->LeadingSuperRegs.push_back(this); + + // Add ad hoc alias links. This is a symmetric relationship between two + // registers, so build a symmetric graph by adding links in both ends. + std::vector Aliases = TheDef->getValueAsListOfDefs("Aliases"); + for (unsigned i = 0, e = Aliases.size(); i != e; ++i) { + CodeGenRegister *Reg = RegBank.getReg(Aliases[i]); + ExplicitAliases.push_back(Reg); + Reg->ExplicitAliases.push_back(this); + } +} + const std::string &CodeGenRegister::getName() const { return TheDef->getName(); } @@ -108,7 +168,7 @@ public: bool isValid() const { return UnitI != UnitE; } - unsigned operator* () const { assert(isValid()); return *UnitI; }; + unsigned operator* () const { assert(isValid()); return *UnitI; } const CodeGenRegister *getReg() const { assert(isValid()); return *RegI; } @@ -152,15 +212,7 @@ bool CodeGenRegister::inheritRegUnits(CodeGenRegBank &RegBank) { unsigned OldNumUnits = RegUnits.size(); for (SubRegMap::const_iterator I = SubRegs.begin(), E = SubRegs.end(); I != E; ++I) { - // Strangely a register may have itself as a subreg (self-cycle) e.g. XMM. - // Only create a unit if no other subregs have units. CodeGenRegister *SR = I->second; - if (SR == this) { - // RegUnits are only empty during getSubRegs, prior to computing weight. - if (RegUnits.empty()) - RegUnits.push_back(RegBank.newRegUnit(0)); - continue; - } // Merge the subregister's units into this register's RegUnits. mergeRegUnits(RegUnits, SR->RegUnits); } @@ -168,27 +220,22 @@ bool CodeGenRegister::inheritRegUnits(CodeGenRegBank &RegBank) { } const CodeGenRegister::SubRegMap & -CodeGenRegister::getSubRegs(CodeGenRegBank &RegBank) { +CodeGenRegister::computeSubRegs(CodeGenRegBank &RegBank) { // Only compute this map once. if (SubRegsComplete) return SubRegs; SubRegsComplete = true; - std::vector SubList = TheDef->getValueAsListOfDefs("SubRegs"); - std::vector IdxList = TheDef->getValueAsListOfDefs("SubRegIndices"); - if (SubList.size() != IdxList.size()) - throw TGError(TheDef->getLoc(), "Register " + getName() + - " SubRegIndices doesn't match SubRegs"); - - // First insert the direct subregs and make sure they are fully indexed. - SmallVector Indices; - for (unsigned i = 0, e = SubList.size(); i != e; ++i) { - CodeGenRegister *SR = RegBank.getReg(SubList[i]); - CodeGenSubRegIndex *Idx = RegBank.getSubRegIdx(IdxList[i]); - Indices.push_back(Idx); + // First insert the explicit subregs and make sure they are fully indexed. + for (unsigned i = 0, e = ExplicitSubRegs.size(); i != e; ++i) { + CodeGenRegister *SR = ExplicitSubRegs[i]; + CodeGenSubRegIndex *Idx = ExplicitSubRegIndices[i]; if (!SubRegs.insert(std::make_pair(Idx, SR)).second) - throw TGError(TheDef->getLoc(), "SubRegIndex " + Idx->getName() + - " appears twice in Register " + getName()); + PrintFatalError(TheDef->getLoc(), "SubRegIndex " + Idx->getName() + + " appears twice in Register " + getName()); + // Map explicit sub-registers first, so the names take precedence. + // The inherited sub-registers are mapped below. + SubReg2Idx.insert(std::make_pair(SR, Idx)); } // Keep track of inherited subregs and how they can be reached. @@ -196,23 +243,14 @@ CodeGenRegister::getSubRegs(CodeGenRegBank &RegBank) { // Clone inherited subregs and place duplicate entries in Orphans. // Here the order is important - earlier subregs take precedence. - for (unsigned i = 0, e = SubList.size(); i != e; ++i) { - CodeGenRegister *SR = RegBank.getReg(SubList[i]); - const SubRegMap &Map = SR->getSubRegs(RegBank); - - // Add this as a super-register of SR now all sub-registers are in the list. - // This creates a topological ordering, the exact order depends on the - // order getSubRegs is called on all registers. - SR->SuperRegs.push_back(this); + for (unsigned i = 0, e = ExplicitSubRegs.size(); i != e; ++i) { + CodeGenRegister *SR = ExplicitSubRegs[i]; + const SubRegMap &Map = SR->computeSubRegs(RegBank); for (SubRegMap::const_iterator SI = Map.begin(), SE = Map.end(); SI != SE; ++SI) { if (!SubRegs.insert(*SI).second) Orphans.insert(SI->second); - - // Noop sub-register indexes are possible, so avoid duplicates. - if (SI->second != SR) - SI->second->SuperRegs.push_back(this); } } @@ -220,11 +258,12 @@ CodeGenRegister::getSubRegs(CodeGenRegBank &RegBank) { // If dsub_2 has ComposedOf = [qsub_1, dsub_0], and this register has a // qsub_1 subreg, add a dsub_2 subreg. Keep growing Indices and process // expanded subreg indices recursively. + SmallVector Indices = ExplicitSubRegIndices; for (unsigned i = 0; i != Indices.size(); ++i) { CodeGenSubRegIndex *Idx = Indices[i]; const CodeGenSubRegIndex::CompMap &Comps = Idx->getComposites(); CodeGenRegister *SR = SubRegs[Idx]; - const SubRegMap &Map = SR->getSubRegs(RegBank); + const SubRegMap &Map = SR->computeSubRegs(RegBank); // Look at the possible compositions of Idx. // They may not all be supported by SR. @@ -243,44 +282,6 @@ CodeGenRegister::getSubRegs(CodeGenRegBank &RegBank) { } } - // Process the composites. - ListInit *Comps = TheDef->getValueAsListInit("CompositeIndices"); - for (unsigned i = 0, e = Comps->size(); i != e; ++i) { - DagInit *Pat = dynamic_cast(Comps->getElement(i)); - if (!Pat) - throw TGError(TheDef->getLoc(), "Invalid dag '" + - Comps->getElement(i)->getAsString() + - "' in CompositeIndices"); - DefInit *BaseIdxInit = dynamic_cast(Pat->getOperator()); - if (!BaseIdxInit || !BaseIdxInit->getDef()->isSubClassOf("SubRegIndex")) - throw TGError(TheDef->getLoc(), "Invalid SubClassIndex in " + - Pat->getAsString()); - CodeGenSubRegIndex *BaseIdx = RegBank.getSubRegIdx(BaseIdxInit->getDef()); - - // Resolve list of subreg indices into R2. - CodeGenRegister *R2 = this; - for (DagInit::const_arg_iterator di = Pat->arg_begin(), - de = Pat->arg_end(); di != de; ++di) { - DefInit *IdxInit = dynamic_cast(*di); - if (!IdxInit || !IdxInit->getDef()->isSubClassOf("SubRegIndex")) - throw TGError(TheDef->getLoc(), "Invalid SubClassIndex in " + - Pat->getAsString()); - CodeGenSubRegIndex *Idx = RegBank.getSubRegIdx(IdxInit->getDef()); - const SubRegMap &R2Subs = R2->getSubRegs(RegBank); - SubRegMap::const_iterator ni = R2Subs.find(Idx); - if (ni == R2Subs.end()) - throw TGError(TheDef->getLoc(), "Composite " + Pat->getAsString() + - " refers to bad index in " + R2->getName()); - R2 = ni->second; - } - - // Insert composite index. Allow overriding inherited indices etc. - SubRegs[BaseIdx] = R2; - - // R2 is no longer an orphan. - Orphans.erase(R2); - } - // Now Orphans contains the inherited subregisters without a direct index. // Create inferred indexes for all missing entries. // Work backwards in the Indices vector in order to compose subregs bottom-up. @@ -295,46 +296,239 @@ CodeGenRegister::getSubRegs(CodeGenRegBank &RegBank) { // dsub_2 -> ssub_0 // // We pick the latter composition because another register may have [dsub_0, - // dsub_1, dsub_2] subregs without neccessarily having a qsub_1 subreg. The + // dsub_1, dsub_2] subregs without necessarily having a qsub_1 subreg. The // dsub_2 -> ssub_0 composition can be shared. while (!Indices.empty() && !Orphans.empty()) { CodeGenSubRegIndex *Idx = Indices.pop_back_val(); CodeGenRegister *SR = SubRegs[Idx]; - const SubRegMap &Map = SR->getSubRegs(RegBank); + const SubRegMap &Map = SR->computeSubRegs(RegBank); for (SubRegMap::const_iterator SI = Map.begin(), SE = Map.end(); SI != SE; ++SI) if (Orphans.erase(SI->second)) SubRegs[RegBank.getCompositeSubRegIndex(Idx, SI->first)] = SI->second; } - // Initialize RegUnitList. A register with no subregisters creates its own - // unit. Otherwise, it inherits all its subregister's units. Because - // getSubRegs is called recursively, this processes the register hierarchy in - // postorder. + // Compute the inverse SubReg -> Idx map. + for (SubRegMap::const_iterator SI = SubRegs.begin(), SE = SubRegs.end(); + SI != SE; ++SI) { + if (SI->second == this) { + ArrayRef Loc; + if (TheDef) + Loc = TheDef->getLoc(); + PrintFatalError(Loc, "Register " + getName() + + " has itself as a sub-register"); + } + + // Compute AllSuperRegsCovered. + if (!CoveredBySubRegs) + SI->first->AllSuperRegsCovered = false; + + // Ensure that every sub-register has a unique name. + DenseMap::iterator Ins = + SubReg2Idx.insert(std::make_pair(SI->second, SI->first)).first; + if (Ins->second == SI->first) + continue; + // Trouble: Two different names for SI->second. + ArrayRef Loc; + if (TheDef) + Loc = TheDef->getLoc(); + PrintFatalError(Loc, "Sub-register can't have two names: " + + SI->second->getName() + " available as " + + SI->first->getName() + " and " + Ins->second->getName()); + } + + // Derive possible names for sub-register concatenations from any explicit + // sub-registers. By doing this before computeSecondarySubRegs(), we ensure + // that getConcatSubRegIndex() won't invent any concatenated indices that the + // user already specified. + for (unsigned i = 0, e = ExplicitSubRegs.size(); i != e; ++i) { + CodeGenRegister *SR = ExplicitSubRegs[i]; + if (!SR->CoveredBySubRegs || SR->ExplicitSubRegs.size() <= 1) + continue; + + // SR is composed of multiple sub-regs. Find their names in this register. + SmallVector Parts; + for (unsigned j = 0, e = SR->ExplicitSubRegs.size(); j != e; ++j) + Parts.push_back(getSubRegIndex(SR->ExplicitSubRegs[j])); + + // Offer this as an existing spelling for the concatenation of Parts. + RegBank.addConcatSubRegIndex(Parts, ExplicitSubRegIndices[i]); + } + + // Initialize RegUnitList. Because getSubRegs is called recursively, this + // processes the register hierarchy in postorder. // - // TODO: We currently assume all register units correspond to a named "leaf" - // register. We should also unify register units for ad-hoc register - // aliases. This can be done by iteratively merging units for aliasing - // registers using a worklist. - assert(RegUnits.empty() && "Should only initialize RegUnits once"); - if (SubRegs.empty()) - RegUnits.push_back(RegBank.newRegUnit(0)); - else - inheritRegUnits(RegBank); + // Inherit all sub-register units. It is good enough to look at the explicit + // sub-registers, the other registers won't contribute any more units. + for (unsigned i = 0, e = ExplicitSubRegs.size(); i != e; ++i) { + CodeGenRegister *SR = ExplicitSubRegs[i]; + // Explicit sub-registers are usually disjoint, so this is a good way of + // computing the union. We may pick up a few duplicates that will be + // eliminated below. + unsigned N = RegUnits.size(); + RegUnits.append(SR->RegUnits.begin(), SR->RegUnits.end()); + std::inplace_merge(RegUnits.begin(), RegUnits.begin() + N, RegUnits.end()); + } + RegUnits.erase(std::unique(RegUnits.begin(), RegUnits.end()), RegUnits.end()); + + // Absent any ad hoc aliasing, we create one register unit per leaf register. + // These units correspond to the maximal cliques in the register overlap + // graph which is optimal. + // + // When there is ad hoc aliasing, we simply create one unit per edge in the + // undirected ad hoc aliasing graph. Technically, we could do better by + // identifying maximal cliques in the ad hoc graph, but cliques larger than 2 + // are extremely rare anyway (I've never seen one), so we don't bother with + // the added complexity. + for (unsigned i = 0, e = ExplicitAliases.size(); i != e; ++i) { + CodeGenRegister *AR = ExplicitAliases[i]; + // Only visit each edge once. + if (AR->SubRegsComplete) + continue; + // Create a RegUnit representing this alias edge, and add it to both + // registers. + unsigned Unit = RegBank.newRegUnit(this, AR); + RegUnits.push_back(Unit); + AR->RegUnits.push_back(Unit); + } + + // Finally, create units for leaf registers without ad hoc aliases. Note that + // a leaf register with ad hoc aliases doesn't get its own unit - it isn't + // necessary. This means the aliasing leaf registers can share a single unit. + if (RegUnits.empty()) + RegUnits.push_back(RegBank.newRegUnit(this)); + + // We have now computed the native register units. More may be adopted later + // for balancing purposes. + NumNativeRegUnits = RegUnits.size(); + return SubRegs; } +// In a register that is covered by its sub-registers, try to find redundant +// sub-registers. For example: +// +// QQ0 = {Q0, Q1} +// Q0 = {D0, D1} +// Q1 = {D2, D3} +// +// We can infer that D1_D2 is also a sub-register, even if it wasn't named in +// the register definition. +// +// The explicitly specified registers form a tree. This function discovers +// sub-register relationships that would force a DAG. +// +void CodeGenRegister::computeSecondarySubRegs(CodeGenRegBank &RegBank) { + // Collect new sub-registers first, add them later. + SmallVector NewSubRegs; + + // Look at the leading super-registers of each sub-register. Those are the + // candidates for new sub-registers, assuming they are fully contained in + // this register. + for (SubRegMap::iterator I = SubRegs.begin(), E = SubRegs.end(); I != E; ++I){ + const CodeGenRegister *SubReg = I->second; + const CodeGenRegister::SuperRegList &Leads = SubReg->LeadingSuperRegs; + for (unsigned i = 0, e = Leads.size(); i != e; ++i) { + CodeGenRegister *Cand = const_cast(Leads[i]); + // Already got this sub-register? + if (Cand == this || getSubRegIndex(Cand)) + continue; + // Check if each component of Cand is already a sub-register. + // We know that the first component is I->second, and is present with the + // name I->first. + SmallVector Parts(1, I->first); + assert(!Cand->ExplicitSubRegs.empty() && + "Super-register has no sub-registers"); + for (unsigned j = 1, e = Cand->ExplicitSubRegs.size(); j != e; ++j) { + if (CodeGenSubRegIndex *Idx = getSubRegIndex(Cand->ExplicitSubRegs[j])) + Parts.push_back(Idx); + else { + // Sub-register doesn't exist. + Parts.clear(); + break; + } + } + // If some Cand sub-register is not part of this register, or if Cand only + // has one sub-register, there is nothing to do. + if (Parts.size() <= 1) + continue; + + // Each part of Cand is a sub-register of this. Make the full Cand also + // a sub-register with a concatenated sub-register index. + CodeGenSubRegIndex *Concat= RegBank.getConcatSubRegIndex(Parts); + NewSubRegs.push_back(std::make_pair(Concat, Cand)); + } + } + + // Now add all the new sub-registers. + for (unsigned i = 0, e = NewSubRegs.size(); i != e; ++i) { + // Don't add Cand if another sub-register is already using the index. + if (!SubRegs.insert(NewSubRegs[i]).second) + continue; + + CodeGenSubRegIndex *NewIdx = NewSubRegs[i].first; + CodeGenRegister *NewSubReg = NewSubRegs[i].second; + SubReg2Idx.insert(std::make_pair(NewSubReg, NewIdx)); + } + + // Create sub-register index composition maps for the synthesized indices. + for (unsigned i = 0, e = NewSubRegs.size(); i != e; ++i) { + CodeGenSubRegIndex *NewIdx = NewSubRegs[i].first; + CodeGenRegister *NewSubReg = NewSubRegs[i].second; + for (SubRegMap::const_iterator SI = NewSubReg->SubRegs.begin(), + SE = NewSubReg->SubRegs.end(); SI != SE; ++SI) { + CodeGenSubRegIndex *SubIdx = getSubRegIndex(SI->second); + if (!SubIdx) + PrintFatalError(TheDef->getLoc(), "No SubRegIndex for " + + SI->second->getName() + " in " + getName()); + NewIdx->addComposite(SI->first, SubIdx); + } + } +} + +void CodeGenRegister::computeSuperRegs(CodeGenRegBank &RegBank) { + // Only visit each register once. + if (SuperRegsComplete) + return; + SuperRegsComplete = true; + + // Make sure all sub-registers have been visited first, so the super-reg + // lists will be topologically ordered. + for (SubRegMap::const_iterator I = SubRegs.begin(), E = SubRegs.end(); + I != E; ++I) + I->second->computeSuperRegs(RegBank); + + // Now add this as a super-register on all sub-registers. + // Also compute the TopoSigId in post-order. + TopoSigId Id; + for (SubRegMap::const_iterator I = SubRegs.begin(), E = SubRegs.end(); + I != E; ++I) { + // Topological signature computed from SubIdx, TopoId(SubReg). + // Loops and idempotent indices have TopoSig = ~0u. + Id.push_back(I->first->EnumValue); + Id.push_back(I->second->TopoSig); + + // Don't add duplicate entries. + if (!I->second->SuperRegs.empty() && I->second->SuperRegs.back() == this) + continue; + I->second->SuperRegs.push_back(this); + } + TopoSig = RegBank.getTopoSig(Id); +} + void CodeGenRegister::addSubRegsPreOrder(SetVector &OSet, CodeGenRegBank &RegBank) const { assert(SubRegsComplete && "Must precompute sub-registers"); - std::vector Indices = TheDef->getValueAsListOfDefs("SubRegIndices"); - for (unsigned i = 0, e = Indices.size(); i != e; ++i) { - CodeGenSubRegIndex *Idx = RegBank.getSubRegIdx(Indices[i]); - CodeGenRegister *SR = SubRegs.find(Idx)->second; + for (unsigned i = 0, e = ExplicitSubRegs.size(); i != e; ++i) { + CodeGenRegister *SR = ExplicitSubRegs[i]; if (OSet.insert(SR)) SR->addSubRegsPreOrder(OSet, RegBank); } + // Add any secondary sub-registers that weren't part of the explicit tree. + for (SubRegMap::const_iterator I = SubRegs.begin(), E = SubRegs.end(); + I != E; ++I) + OSet.insert(I->second); } // Get the sum of this register's unit weights. @@ -342,7 +536,7 @@ unsigned CodeGenRegister::getWeight(const CodeGenRegBank &RegBank) const { unsigned Weight = 0; for (RegUnitList::const_iterator I = RegUnits.begin(), E = RegUnits.end(); I != E; ++I) { - Weight += RegBank.getRegUnitWeight(*I); + Weight += RegBank.getRegUnit(*I).Weight; } return Weight; } @@ -361,15 +555,16 @@ struct TupleExpander : SetTheory::Expander { unsigned Dim = Indices.size(); ListInit *SubRegs = Def->getValueAsListInit("SubRegs"); if (Dim != SubRegs->getSize()) - throw TGError(Def->getLoc(), "SubRegIndices and SubRegs size mismatch"); + PrintFatalError(Def->getLoc(), "SubRegIndices and SubRegs size mismatch"); if (Dim < 2) - throw TGError(Def->getLoc(), "Tuples must have at least 2 sub-registers"); + PrintFatalError(Def->getLoc(), + "Tuples must have at least 2 sub-registers"); // Evaluate the sub-register lists to be zipped. unsigned Length = ~0u; SmallVector Lists(Dim); for (unsigned i = 0; i != Dim; ++i) { - ST.evaluate(SubRegs->getElement(i), Lists[i]); + ST.evaluate(SubRegs->getElement(i), Lists[i], Def->getLoc()); Length = std::min(Length, unsigned(Lists[i].size())); } @@ -403,8 +598,10 @@ struct TupleExpander : SetTheory::Expander { Elts.insert(NewReg); // Copy Proto super-classes. - for (unsigned i = 0, e = Proto->getSuperClasses().size(); i != e; ++i) - NewReg->addSuperClass(Proto->getSuperClasses()[i]); + ArrayRef Supers = Proto->getSuperClasses(); + ArrayRef Ranges = Proto->getSuperClassRanges(); + for (unsigned i = 0, e = Supers.size(); i != e; ++i) + NewReg->addSuperClass(Supers[i], Ranges[i]); // Copy Proto fields. for (unsigned i = 0, e = Proto->getValues().size(); i != e; ++i) { @@ -461,19 +658,24 @@ struct TupleExpander : SetTheory::Expander { //===----------------------------------------------------------------------===// CodeGenRegisterClass::CodeGenRegisterClass(CodeGenRegBank &RegBank, Record *R) - : TheDef(R), Name(R->getName()), EnumValue(-1) { + : TheDef(R), + Name(R->getName()), + TopoSigs(RegBank.getNumTopoSigs()), + EnumValue(-1) { // Rename anonymous register classes. if (R->getName().size() > 9 && R->getName()[9] == '.') { static unsigned AnonCounter = 0; - R->setName("AnonRegClass_"+utostr(AnonCounter++)); + R->setName("AnonRegClass_" + utostr(AnonCounter)); + // MSVC2012 ICEs if AnonCounter++ is directly passed to utostr. + ++AnonCounter; } std::vector TypeList = R->getValueAsListOfDefs("RegTypes"); for (unsigned i = 0, e = TypeList.size(); i != e; ++i) { Record *Type = TypeList[i]; if (!Type->isSubClassOf("ValueType")) - throw "RegTypes list member '" + Type->getName() + - "' does not derive from the ValueType class!"; + PrintFatalError("RegTypes list member '" + Type->getName() + + "' does not derive from the ValueType class!"); VTs.push_back(getValueType(Type)); } assert(!VTs.empty() && "RegisterClass must contain at least one ValueType!"); @@ -486,47 +688,26 @@ CodeGenRegisterClass::CodeGenRegisterClass(CodeGenRegBank &RegBank, Record *R) // Default allocation order always contains all registers. for (unsigned i = 0, e = Elements->size(); i != e; ++i) { Orders[0].push_back((*Elements)[i]); - Members.insert(RegBank.getReg((*Elements)[i])); + const CodeGenRegister *Reg = RegBank.getReg((*Elements)[i]); + Members.insert(Reg); + TopoSigs.set(Reg->getTopoSig()); } // Alternative allocation orders may be subsets. SetTheory::RecSet Order; for (unsigned i = 0, e = AltOrders->size(); i != e; ++i) { - RegBank.getSets().evaluate(AltOrders->getElement(i), Order); + RegBank.getSets().evaluate(AltOrders->getElement(i), Order, R->getLoc()); Orders[1 + i].append(Order.begin(), Order.end()); // Verify that all altorder members are regclass members. while (!Order.empty()) { CodeGenRegister *Reg = RegBank.getReg(Order.back()); Order.pop_back(); if (!contains(Reg)) - throw TGError(R->getLoc(), " AltOrder register " + Reg->getName() + + PrintFatalError(R->getLoc(), " AltOrder register " + Reg->getName() + " is not a class member"); } } - // SubRegClasses is a list containing (RC, subregindex, ...) dags. - ListInit *SRC = R->getValueAsListInit("SubRegClasses"); - for (ListInit::const_iterator i = SRC->begin(), e = SRC->end(); i != e; ++i) { - DagInit *DAG = dynamic_cast(*i); - if (!DAG) throw "SubRegClasses must contain DAGs"; - DefInit *DAGOp = dynamic_cast(DAG->getOperator()); - Record *RCRec; - if (!DAGOp || !(RCRec = DAGOp->getDef())->isSubClassOf("RegisterClass")) - throw "Operator '" + DAG->getOperator()->getAsString() + - "' in SubRegClasses is not a RegisterClass"; - // Iterate over args, all SubRegIndex instances. - for (DagInit::const_arg_iterator ai = DAG->arg_begin(), ae = DAG->arg_end(); - ai != ae; ++ai) { - DefInit *Idx = dynamic_cast(*ai); - Record *IdxRec; - if (!Idx || !(IdxRec = Idx->getDef())->isSubClassOf("SubRegIndex")) - throw "Argument '" + (*ai)->getAsString() + - "' in SubRegClasses is not a SubRegIndex"; - if (!SubRegClasses.insert(std::make_pair(IdxRec, RCRec)).second) - throw "SubRegIndex '" + IdxRec->getName() + "' mentioned twice"; - } - } - // Allow targets to override the size in bits of the RegisterClass. unsigned Size = R->getValueAsInt("Size"); @@ -541,15 +722,20 @@ CodeGenRegisterClass::CodeGenRegisterClass(CodeGenRegBank &RegBank, Record *R) // Create an inferred register class that was missing from the .td files. // Most properties will be inherited from the closest super-class after the // class structure has been computed. -CodeGenRegisterClass::CodeGenRegisterClass(StringRef Name, Key Props) +CodeGenRegisterClass::CodeGenRegisterClass(CodeGenRegBank &RegBank, + StringRef Name, Key Props) : Members(*Props.Members), TheDef(0), Name(Name), + TopoSigs(RegBank.getNumTopoSigs()), EnumValue(-1), SpillSize(Props.SpillSize), SpillAlignment(Props.SpillAlignment), CopyCost(0), Allocatable(true) { + for (CodeGenRegister::Set::iterator I = Members.begin(), E = Members.end(); + I != E; ++I) + TopoSigs.set((*I)->getTopoSig()); } // Compute inherited propertied for a synthesized register class. @@ -627,19 +813,13 @@ static bool testSubClass(const CodeGenRegisterClass *A, /// Register classes with the same registers, spill size, and alignment form a /// clique. They will be ordered alphabetically. /// -static int TopoOrderRC(const void *PA, const void *PB) { - const CodeGenRegisterClass *A = *(const CodeGenRegisterClass* const*)PA; - const CodeGenRegisterClass *B = *(const CodeGenRegisterClass* const*)PB; +static int TopoOrderRC(CodeGenRegisterClass *const *PA, + CodeGenRegisterClass *const *PB) { + const CodeGenRegisterClass *A = *PA; + const CodeGenRegisterClass *B = *PB; if (A == B) return 0; - // Order by descending set size. Note that the classes' allocation order may - // not have been computed yet. The Members set is always vaild. - if (A->getMembers().size() > B->getMembers().size()) - return -1; - if (A->getMembers().size() < B->getMembers().size()) - return 1; - // Order by ascending spill size. if (A->SpillSize < B->SpillSize) return -1; @@ -652,6 +832,13 @@ static int TopoOrderRC(const void *PA, const void *PB) { if (A->SpillAlignment > B->SpillAlignment) return 1; + // Order by descending set size. Note that the classes' allocation order may + // not have been computed yet. The Members set is always vaild. + if (A->getMembers().size() > B->getMembers().size()) + return -1; + if (A->getMembers().size() < B->getMembers().size()) + return 1; + // Finally order by name as a tie breaker. return StringRef(A->getName()).compare(B->getName()); } @@ -686,7 +873,7 @@ void CodeGenRegisterClass::computeSubClasses(CodeGenRegBank &RegBank) { RC.SubClasses |= SubRC->SubClasses; } - // Sweep up missed clique members. They will be immediately preceeding RC. + // Sweep up missed clique members. They will be immediately preceding RC. for (unsigned s = rci - 1; s && testSubClass(&RC, RegClasses[s - 1]); --s) RC.SubClasses.set(s - 1); } @@ -722,12 +909,22 @@ CodeGenRegisterClass::getSuperRegClasses(CodeGenSubRegIndex *SubIdx, Out.set((*I)->EnumValue); } +// Populate a unique sorted list of units from a register set. +void CodeGenRegisterClass::buildRegUnitSet( + std::vector &RegUnits) const { + std::vector TmpUnits; + for (RegUnitIterator UnitI(Members); UnitI.isValid(); ++UnitI) + TmpUnits.push_back(*UnitI); + std::sort(TmpUnits.begin(), TmpUnits.end()); + std::unique_copy(TmpUnits.begin(), TmpUnits.end(), + std::back_inserter(RegUnits)); +} //===----------------------------------------------------------------------===// // CodeGenRegBank //===----------------------------------------------------------------------===// -CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) : Records(Records) { +CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) { // Configure register Sets to understand register classes and tuples. Sets.addFieldExpander("RegisterClass", "MemberList"); Sets.addFieldExpander("CalleeSavedRegs", "SaveList"); @@ -737,7 +934,6 @@ CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) : Records(Records) { // More indices will be synthesized later. std::vector SRIs = Records.getAllDerivedDefinitions("SubRegIndex"); std::sort(SRIs.begin(), SRIs.end(), LessRecord()); - NumNamedIndices = SRIs.size(); for (unsigned i = 0, e = SRIs.size(); i != e; ++i) getSubRegIdx(SRIs[i]); // Build composite maps from ComposedOf fields. @@ -746,7 +942,7 @@ CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) : Records(Records) { // Read in the register definitions. std::vector Regs = Records.getAllDerivedDefinitions("Register"); - std::sort(Regs.begin(), Regs.end(), LessRecord()); + std::sort(Regs.begin(), Regs.end(), LessRecordRegister()); Registers.reserve(Regs.size()); // Assign the enumeration values. for (unsigned i = 0, e = Regs.size(); i != e; ++i) @@ -755,26 +951,52 @@ CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) : Records(Records) { // Expand tuples and number the new registers. std::vector Tups = Records.getAllDerivedDefinitions("RegisterTuples"); + + std::vector TupRegsCopy; for (unsigned i = 0, e = Tups.size(); i != e; ++i) { const std::vector *TupRegs = Sets.expand(Tups[i]); - for (unsigned j = 0, je = TupRegs->size(); j != je; ++j) - getReg((*TupRegs)[j]); + TupRegsCopy.reserve(TupRegs->size()); + TupRegsCopy.assign(TupRegs->begin(), TupRegs->end()); + std::sort(TupRegsCopy.begin(), TupRegsCopy.end(), LessRecordRegister()); + for (unsigned j = 0, je = TupRegsCopy.size(); j != je; ++j) + getReg((TupRegsCopy)[j]); + TupRegsCopy.clear(); } - // Precompute all sub-register maps now all the registers are known. + // Now all the registers are known. Build the object graph of explicit + // register-register references. + for (unsigned i = 0, e = Registers.size(); i != e; ++i) + Registers[i]->buildObjectGraph(*this); + + // Compute register name map. + for (unsigned i = 0, e = Registers.size(); i != e; ++i) + RegistersByName.GetOrCreateValue( + Registers[i]->TheDef->getValueAsString("AsmName"), + Registers[i]); + + // Precompute all sub-register maps. // This will create Composite entries for all inferred sub-register indices. - NumRegUnits = 0; for (unsigned i = 0, e = Registers.size(); i != e; ++i) - Registers[i]->getSubRegs(*this); + Registers[i]->computeSubRegs(*this); + + // Infer even more sub-registers by combining leading super-registers. + for (unsigned i = 0, e = Registers.size(); i != e; ++i) + if (Registers[i]->CoveredBySubRegs) + Registers[i]->computeSecondarySubRegs(*this); + + // After the sub-register graph is complete, compute the topologically + // ordered SuperRegs list. + for (unsigned i = 0, e = Registers.size(); i != e; ++i) + Registers[i]->computeSuperRegs(*this); // Native register units are associated with a leaf register. They've all been // discovered now. - NumNativeRegUnits = NumRegUnits; + NumNativeRegUnits = RegUnits.size(); // Read in register class definitions. std::vector RCs = Records.getAllDerivedDefinitions("RegisterClass"); if (RCs.empty()) - throw std::string("No 'RegisterClass' subclasses defined!"); + PrintFatalError(std::string("No 'RegisterClass' subclasses defined!")); // Allocate user-defined register classes. RegClasses.reserve(RCs.size()); @@ -791,6 +1013,15 @@ CodeGenRegBank::CodeGenRegBank(RecordKeeper &Records) : Records(Records) { CodeGenRegisterClass::computeSubClasses(*this); } +// Create a synthetic CodeGenSubRegIndex without a corresponding Record. +CodeGenSubRegIndex* +CodeGenRegBank::createSubRegIndex(StringRef Name, StringRef Namespace) { + CodeGenSubRegIndex *Idx = new CodeGenSubRegIndex(Name, Namespace, + SubRegIndices.size() + 1); + SubRegIndices.push_back(Idx); + return Idx; +} + CodeGenSubRegIndex *CodeGenRegBank::getSubRegIdx(Record *Def) { CodeGenSubRegIndex *&Idx = Def2SubRegIdx[Def]; if (Idx) @@ -833,7 +1064,7 @@ CodeGenRegBank::getOrCreateSubClass(const CodeGenRegisterClass *RC, return FoundI->second; // Sub-class doesn't exist, create a new one. - CodeGenRegisterClass *NewRC = new CodeGenRegisterClass(Name, K); + CodeGenRegisterClass *NewRC = new CodeGenRegisterClass(*this, Name, K); addToMaps(NewRC); return NewRC; } @@ -842,7 +1073,7 @@ CodeGenRegisterClass *CodeGenRegBank::getRegClass(Record *Def) { if (CodeGenRegisterClass *RC = Def2RC[Def]) return RC; - throw TGError(Def->getLoc(), "Not a known RegisterClass!"); + PrintFatalError(Def->getLoc(), "Not a known RegisterClass!"); } CodeGenSubRegIndex* @@ -855,14 +1086,55 @@ CodeGenRegBank::getCompositeSubRegIndex(CodeGenSubRegIndex *A, // None exists, synthesize one. std::string Name = A->getName() + "_then_" + B->getName(); - Comp = getSubRegIdx(new Record(Name, SMLoc(), Records)); + Comp = createSubRegIndex(Name, A->getNamespace()); A->addComposite(B, Comp); return Comp; } +CodeGenSubRegIndex *CodeGenRegBank:: +getConcatSubRegIndex(const SmallVector &Parts) { + assert(Parts.size() > 1 && "Need two parts to concatenate"); + + // Look for an existing entry. + CodeGenSubRegIndex *&Idx = ConcatIdx[Parts]; + if (Idx) + return Idx; + + // None exists, synthesize one. + std::string Name = Parts.front()->getName(); + // Determine whether all parts are contiguous. + bool isContinuous = true; + unsigned Size = Parts.front()->Size; + unsigned LastOffset = Parts.front()->Offset; + unsigned LastSize = Parts.front()->Size; + for (unsigned i = 1, e = Parts.size(); i != e; ++i) { + Name += '_'; + Name += Parts[i]->getName(); + Size += Parts[i]->Size; + if (Parts[i]->Offset != (LastOffset + LastSize)) + isContinuous = false; + LastOffset = Parts[i]->Offset; + LastSize = Parts[i]->Size; + } + Idx = createSubRegIndex(Name, Parts.front()->getNamespace()); + Idx->Size = Size; + Idx->Offset = isContinuous ? Parts.front()->Offset : -1; + return Idx; +} + void CodeGenRegBank::computeComposites() { + // Keep track of TopoSigs visited. We only need to visit each TopoSig once, + // and many registers will share TopoSigs on regular architectures. + BitVector TopoSigs(getNumTopoSigs()); + for (unsigned i = 0, e = Registers.size(); i != e; ++i) { CodeGenRegister *Reg1 = Registers[i]; + + // Skip identical subreg structures already processed. + if (TopoSigs.test(Reg1->getTopoSig())) + continue; + TopoSigs.set(Reg1->getTopoSig()); + const CodeGenRegister::SubRegMap &SRM1 = Reg1->getSubRegs(); for (CodeGenRegister::SubRegMap::const_iterator i1 = SRM1.begin(), e1 = SRM1.end(); i1 != e1; ++i1) { @@ -875,32 +1147,71 @@ void CodeGenRegBank::computeComposites() { // Try composing Idx1 with another SubRegIndex. for (CodeGenRegister::SubRegMap::const_iterator i2 = SRM2.begin(), e2 = SRM2.end(); i2 != e2; ++i2) { - CodeGenSubRegIndex *Idx2 = i2->first; + CodeGenSubRegIndex *Idx2 = i2->first; CodeGenRegister *Reg3 = i2->second; // Ignore identity compositions. if (Reg2 == Reg3) continue; // OK Reg1:IdxPair == Reg3. Find the index with Reg:Idx == Reg3. - for (CodeGenRegister::SubRegMap::const_iterator i1d = SRM1.begin(), - e1d = SRM1.end(); i1d != e1d; ++i1d) { - if (i1d->second == Reg3) { - // Conflicting composition? Emit a warning but allow it. - if (CodeGenSubRegIndex *Prev = Idx1->addComposite(Idx2, i1d->first)) - errs() << "Warning: SubRegIndex " << Idx1->getQualifiedName() - << " and " << Idx2->getQualifiedName() - << " compose ambiguously as " - << Prev->getQualifiedName() << " or " - << i1d->first->getQualifiedName() << "\n"; - } - } + CodeGenSubRegIndex *Idx3 = Reg1->getSubRegIndex(Reg3); + assert(Idx3 && "Sub-register doesn't have an index"); + + // Conflicting composition? Emit a warning but allow it. + if (CodeGenSubRegIndex *Prev = Idx1->addComposite(Idx2, Idx3)) + PrintWarning(Twine("SubRegIndex ") + Idx1->getQualifiedName() + + " and " + Idx2->getQualifiedName() + + " compose ambiguously as " + Prev->getQualifiedName() + + " or " + Idx3->getQualifiedName()); } } } +} - // We don't care about the difference between (Idx1, Idx2) -> Idx2 and invalid - // compositions, so remove any mappings of that form. - for (unsigned i = 0, e = SubRegIndices.size(); i != e; ++i) - SubRegIndices[i]->cleanComposites(); +// Compute lane masks. This is similar to register units, but at the +// sub-register index level. Each bit in the lane mask is like a register unit +// class, and two lane masks will have a bit in common if two sub-register +// indices overlap in some register. +// +// Conservatively share a lane mask bit if two sub-register indices overlap in +// some registers, but not in others. That shouldn't happen a lot. +void CodeGenRegBank::computeSubRegIndexLaneMasks() { + // First assign individual bits to all the leaf indices. + unsigned Bit = 0; + // Determine mask of lanes that cover their registers. + CoveringLanes = ~0u; + for (unsigned i = 0, e = SubRegIndices.size(); i != e; ++i) { + CodeGenSubRegIndex *Idx = SubRegIndices[i]; + if (Idx->getComposites().empty()) { + Idx->LaneMask = 1u << Bit; + // Share bit 31 in the unlikely case there are more than 32 leafs. + // + // Sharing bits is harmless; it allows graceful degradation in targets + // with more than 32 vector lanes. They simply get a limited resolution + // view of lanes beyond the 32nd. + // + // See also the comment for getSubRegIndexLaneMask(). + if (Bit < 31) + ++Bit; + else + // Once bit 31 is shared among multiple leafs, the 'lane' it represents + // is no longer covering its registers. + CoveringLanes &= ~(1u << Bit); + } else { + Idx->LaneMask = 0; + } + } + + // FIXME: What if ad-hoc aliasing introduces overlaps that aren't represented + // by the sub-register graph? This doesn't occur in any known targets. + + // Inherit lanes from composites. + for (unsigned i = 0, e = SubRegIndices.size(); i != e; ++i) { + unsigned Mask = SubRegIndices[i]->computeLaneMask(); + // If some super-registers without CoveredBySubRegs use this index, we can + // no longer assume that the lanes are covering their registers. + if (!SubRegIndices[i]->AllSuperRegsCovered) + CoveringLanes &= ~Mask; + } } namespace { @@ -945,22 +1256,34 @@ static void computeUberSets(std::vector &UberSets, // For simplicitly make the SetID the same as EnumValue. IntEqClasses UberSetIDs(Registers.size()+1); + std::set AllocatableRegs; for (unsigned i = 0, e = RegBank.getRegClasses().size(); i != e; ++i) { + CodeGenRegisterClass *RegClass = RegBank.getRegClasses()[i]; + if (!RegClass->Allocatable) + continue; + const CodeGenRegister::Set &Regs = RegClass->getMembers(); - if (Regs.empty()) continue; + if (Regs.empty()) + continue; unsigned USetID = UberSetIDs.findLeader((*Regs.begin())->EnumValue); assert(USetID && "register number 0 is invalid"); - // combine non-allocatable classes - if (!RegClass->Allocatable) { - UberSetIDs.join(0, USetID); - USetID = 0; - } + AllocatableRegs.insert((*Regs.begin())->EnumValue); for (CodeGenRegister::Set::const_iterator I = llvm::next(Regs.begin()), - E = Regs.end(); I != E; ++I) + E = Regs.end(); I != E; ++I) { + AllocatableRegs.insert((*I)->EnumValue); UberSetIDs.join(USetID, (*I)->EnumValue); + } + } + // Combine non-allocatable regs. + for (unsigned i = 0, e = Registers.size(); i != e; ++i) { + unsigned RegNum = Registers[i]->EnumValue; + if (AllocatableRegs.count(RegNum)) + continue; + + UberSetIDs.join(0, RegNum); } UberSetIDs.compress(); @@ -1001,7 +1324,7 @@ static void computeUberWeights(std::vector &UberSets, Reg = UnitI.getReg(); Weight = 0; } - unsigned UWeight = RegBank.getRegUnitWeight(*UnitI); + unsigned UWeight = RegBank.getRegUnit(*UnitI).Weight; if (!UWeight) { UWeight = 1; RegBank.increaseRegUnitWeight(*UnitI, UWeight); @@ -1010,9 +1333,18 @@ static void computeUberWeights(std::vector &UberSets, } if (Weight > MaxWeight) MaxWeight = Weight; - - // Update the set weight. - I->Weight = MaxWeight; + if (I->Weight != MaxWeight) { + DEBUG( + dbgs() << "UberSet " << I - UberSets.begin() << " Weight " << MaxWeight; + for (CodeGenRegister::Set::iterator + UnitI = I->Regs.begin(), UnitE = I->Regs.end(); + UnitI != UnitE; ++UnitI) { + dbgs() << " " << (*UnitI)->getName(); + } + dbgs() << "\n"); + // Update the set weight. + I->Weight = MaxWeight; + } // Find singular determinants. for (CodeGenRegister::Set::iterator RegI = I->Regs.begin(), @@ -1037,17 +1369,21 @@ static void computeUberWeights(std::vector &UberSets, static bool normalizeWeight(CodeGenRegister *Reg, std::vector &UberSets, std::vector &RegSets, + std::set &NormalRegs, CodeGenRegister::RegUnitList &NormalUnits, CodeGenRegBank &RegBank) { bool Changed = false; + if (!NormalRegs.insert(Reg->EnumValue).second) + return Changed; + const CodeGenRegister::SubRegMap &SRM = Reg->getSubRegs(); for (CodeGenRegister::SubRegMap::const_iterator SRI = SRM.begin(), SRE = SRM.end(); SRI != SRE; ++SRI) { if (SRI->second == Reg) continue; // self-cycles happen - Changed |= - normalizeWeight(SRI->second, UberSets, RegSets, NormalUnits, RegBank); + Changed |= normalizeWeight(SRI->second, UberSets, RegSets, + NormalRegs, NormalUnits, RegBank); } // Postorder register normalization. @@ -1092,11 +1428,6 @@ static bool normalizeWeight(CodeGenRegister *Reg, // The goal is that two registers in the same class will have the same weight, // where each register's weight is defined as sum of its units' weights. void CodeGenRegBank::computeRegUnitWeights() { - assert(RegUnitWeights.empty() && "Only initialize RegUnitWeights once"); - - // Only allocatable units will be initialized to nonzero weight. - RegUnitWeights.resize(NumRegUnits); - std::vector UberSets; std::vector RegSets(Registers.size()); computeUberSets(UberSets, RegSets, *this); @@ -1112,8 +1443,9 @@ void CodeGenRegBank::computeRegUnitWeights() { Changed = false; for (unsigned i = 0, e = Registers.size(); i != e; ++i) { CodeGenRegister::RegUnitList NormalUnits; - Changed |= - normalizeWeight(Registers[i], UberSets, RegSets, NormalUnits, *this); + std::set NormalRegs; + Changed |= normalizeWeight(Registers[i], UberSets, RegSets, + NormalRegs, NormalUnits, *this); } } } @@ -1139,34 +1471,56 @@ static bool isRegUnitSubSet(const std::vector &RUSubSet, RUSubSet.begin(), RUSubSet.end()); } -// Iteratively prune unit sets. +/// Iteratively prune unit sets. Prune subsets that are close to the superset, +/// but with one or two registers removed. We occasionally have registers like +/// APSR and PC thrown in with the general registers. We also see many +/// special-purpose register subsets, such as tail-call and Thumb +/// encodings. Generating all possible overlapping sets is combinatorial and +/// overkill for modeling pressure. Ideally we could fix this statically in +/// tablegen by (1) having the target define register classes that only include +/// the allocatable registers and marking other classes as non-allocatable and +/// (2) having a way to mark special purpose classes as "don't-care" classes for +/// the purpose of pressure. However, we make an attempt to handle targets that +/// are not nicely defined by merging nearly identical register unit sets +/// statically. This generates smaller tables. Then, dynamically, we adjust the +/// set limit by filtering the reserved registers. +/// +/// Merge sets only if the units have the same weight. For example, on ARM, +/// Q-tuples with ssub index 0 include all S regs but also include D16+. We +/// should not expand the S set to include D regs. void CodeGenRegBank::pruneUnitSets() { assert(RegClassUnitSets.empty() && "this invalidates RegClassUnitSets"); // Form an equivalence class of UnitSets with no significant difference. - IntEqClasses RepUnitSetIDs(RegUnitSets.size()); + std::vector SuperSetIDs; for (unsigned SubIdx = 0, EndIdx = RegUnitSets.size(); SubIdx != EndIdx; ++SubIdx) { const RegUnitSet &SubSet = RegUnitSets[SubIdx]; - for (unsigned SuperIdx = 0; SuperIdx != EndIdx; ++SuperIdx) { + unsigned SuperIdx = 0; + for (; SuperIdx != EndIdx; ++SuperIdx) { if (SuperIdx == SubIdx) continue; + unsigned UnitWeight = RegUnits[SubSet.Units[0]].Weight; const RegUnitSet &SuperSet = RegUnitSets[SuperIdx]; if (isRegUnitSubSet(SubSet.Units, SuperSet.Units) - && (SubSet.Units.size() + 3 > SuperSet.Units.size())) { - RepUnitSetIDs.join(SubIdx, SuperIdx); + && (SubSet.Units.size() + 3 > SuperSet.Units.size()) + && UnitWeight == RegUnits[SuperSet.Units[0]].Weight + && UnitWeight == RegUnits[SuperSet.Units.back()].Weight) { + DEBUG(dbgs() << "UnitSet " << SubIdx << " subsumed by " << SuperIdx + << "\n"); + break; } } + if (SuperIdx == EndIdx) + SuperSetIDs.push_back(SubIdx); } - RepUnitSetIDs.compress(); - // Populate PrunedUnitSets with each equivalence class's superset. - std::vector PrunedUnitSets(RepUnitSetIDs.getNumClasses()); - for (unsigned i = 0, e = RegUnitSets.size(); i != e; ++i) { - RegUnitSet &SuperSet = PrunedUnitSets[RepUnitSetIDs[i]]; - if (SuperSet.Units.size() < RegUnitSets[i].Units.size()) - SuperSet = RegUnitSets[i]; + std::vector PrunedUnitSets(SuperSetIDs.size()); + for (unsigned i = 0, e = SuperSetIDs.size(); i != e; ++i) { + unsigned SuperIdx = SuperSetIDs[i]; + PrunedUnitSets[i].Name = RegUnitSets[SuperIdx].Name; + PrunedUnitSets[i].Units.swap(RegUnitSets[SuperIdx].Units); } RegUnitSets.swap(PrunedUnitSets); } @@ -1179,24 +1533,21 @@ void CodeGenRegBank::pruneUnitSets() { // RegisterInfoEmitter will map each RegClass to its RegUnitClass and any // RegUnitSet that is a superset of that RegUnitClass. void CodeGenRegBank::computeRegUnitSets() { + assert(RegUnitSets.empty() && "dirty RegUnitSets"); // Compute a unique RegUnitSet for each RegClass. const ArrayRef &RegClasses = getRegClasses(); unsigned NumRegClasses = RegClasses.size(); for (unsigned RCIdx = 0, RCEnd = NumRegClasses; RCIdx != RCEnd; ++RCIdx) { - - // Compute a sorted list of units in this class. - std::vector RegUnits; - const CodeGenRegister::Set &Regs = RegClasses[RCIdx]->getMembers(); - for (RegUnitIterator UnitI(Regs); UnitI.isValid(); ++UnitI) - RegUnits.push_back(*UnitI); - std::sort(RegUnits.begin(), RegUnits.end()); + if (!RegClasses[RCIdx]->Allocatable) + continue; // Speculatively grow the RegUnitSets to hold the new set. RegUnitSets.resize(RegUnitSets.size() + 1); RegUnitSets.back().Name = RegClasses[RCIdx]->getName(); - std::unique_copy(RegUnits.begin(), RegUnits.end(), - std::back_inserter(RegUnitSets.back().Units)); + + // Compute a sorted list of units in this class. + RegClasses[RCIdx]->buildRegUnitSet(RegUnitSets.back().Units); // Find an existing RegUnitSet. std::vector::const_iterator SetI = @@ -1205,9 +1556,32 @@ void CodeGenRegBank::computeRegUnitSets() { RegUnitSets.pop_back(); } + DEBUG(dbgs() << "\nBefore pruning:\n"; + for (unsigned USIdx = 0, USEnd = RegUnitSets.size(); + USIdx < USEnd; ++USIdx) { + dbgs() << "UnitSet " << USIdx << " " << RegUnitSets[USIdx].Name + << ":"; + ArrayRef Units = RegUnitSets[USIdx].Units; + for (unsigned i = 0, e = Units.size(); i < e; ++i) + dbgs() << " " << RegUnits[Units[i]].Roots[0]->getName(); + dbgs() << "\n"; + }); + // Iteratively prune unit sets. pruneUnitSets(); + DEBUG(dbgs() << "\nBefore union:\n"; + for (unsigned USIdx = 0, USEnd = RegUnitSets.size(); + USIdx < USEnd; ++USIdx) { + dbgs() << "UnitSet " << USIdx << " " << RegUnitSets[USIdx].Name + << ":"; + ArrayRef Units = RegUnitSets[USIdx].Units; + for (unsigned i = 0, e = Units.size(); i < e; ++i) + dbgs() << " " << RegUnits[Units[i]].Roots[0]->getName(); + dbgs() << "\n"; + } + dbgs() << "\nUnion sets:\n"); + // Iterate over all unit sets, including new ones added by this loop. unsigned NumRegUnitSubSets = RegUnitSets.size(); for (unsigned Idx = 0, EndIdx = RegUnitSets.size(); Idx != EndIdx; ++Idx) { @@ -1217,7 +1591,7 @@ void CodeGenRegBank::computeRegUnitSets() { assert(Idx < (2*NumRegUnitSubSets) && "runaway unit set inference"); // Compare new sets with all original classes. - for (unsigned SearchIdx = (SearchIdx >= NumRegUnitSubSets) ? 0 : Idx+1; + for (unsigned SearchIdx = (Idx >= NumRegUnitSubSets) ? 0 : Idx+1; SearchIdx != EndIdx; ++SearchIdx) { std::set Intersection; std::set_intersection(RegUnitSets[Idx].Units.begin(), @@ -1245,111 +1619,105 @@ void CodeGenRegBank::computeRegUnitSets() { findRegUnitSet(RegUnitSets, RegUnitSets.back()); if (SetI != llvm::prior(RegUnitSets.end())) RegUnitSets.pop_back(); + else { + DEBUG(dbgs() << "UnitSet " << RegUnitSets.size()-1 + << " " << RegUnitSets.back().Name << ":"; + ArrayRef Units = RegUnitSets.back().Units; + for (unsigned i = 0, e = Units.size(); i < e; ++i) + dbgs() << " " << RegUnits[Units[i]].Roots[0]->getName(); + dbgs() << "\n";); + } } } - // Iteratively prune unit sets again after inferring supersets. + // Iteratively prune unit sets after inferring supersets. pruneUnitSets(); + DEBUG(dbgs() << "\n"; + for (unsigned USIdx = 0, USEnd = RegUnitSets.size(); + USIdx < USEnd; ++USIdx) { + dbgs() << "UnitSet " << USIdx << " " << RegUnitSets[USIdx].Name + << ":"; + ArrayRef Units = RegUnitSets[USIdx].Units; + for (unsigned i = 0, e = Units.size(); i < e; ++i) + dbgs() << " " << RegUnits[Units[i]].Roots[0]->getName(); + dbgs() << "\n"; + }); + // For each register class, list the UnitSets that are supersets. RegClassUnitSets.resize(NumRegClasses); for (unsigned RCIdx = 0, RCEnd = NumRegClasses; RCIdx != RCEnd; ++RCIdx) { + if (!RegClasses[RCIdx]->Allocatable) + continue; + // Recompute the sorted list of units in this class. - std::vector RegUnits; - const CodeGenRegister::Set &Regs = RegClasses[RCIdx]->getMembers(); - for (RegUnitIterator UnitI(Regs); UnitI.isValid(); ++UnitI) - RegUnits.push_back(*UnitI); - std::sort(RegUnits.begin(), RegUnits.end()); + std::vector RCRegUnits; + RegClasses[RCIdx]->buildRegUnitSet(RCRegUnits); // Don't increase pressure for unallocatable regclasses. - if (RegUnits.empty()) + if (RCRegUnits.empty()) continue; + DEBUG(dbgs() << "RC " << RegClasses[RCIdx]->getName() << " Units: \n"; + for (unsigned i = 0, e = RCRegUnits.size(); i < e; ++i) + dbgs() << RegUnits[RCRegUnits[i]].getRoots()[0]->getName() << " "; + dbgs() << "\n UnitSetIDs:"); + // Find all supersets. for (unsigned USIdx = 0, USEnd = RegUnitSets.size(); USIdx != USEnd; ++USIdx) { - if (isRegUnitSubSet(RegUnits, RegUnitSets[USIdx].Units)) + if (isRegUnitSubSet(RCRegUnits, RegUnitSets[USIdx].Units)) { + DEBUG(dbgs() << " " << USIdx); RegClassUnitSets[RCIdx].push_back(USIdx); + } } + DEBUG(dbgs() << "\n"); + assert(!RegClassUnitSets[RCIdx].empty() && "missing unit set for regclass"); } -} -// Compute sets of overlapping registers. -// -// The standard set is all super-registers and all sub-registers, but the -// target description can add arbitrary overlapping registers via the 'Aliases' -// field. This complicates things, but we can compute overlapping sets using -// the following rules: -// -// 1. The relation overlap(A, B) is reflexive and symmetric but not transitive. -// -// 2. overlap(A, B) implies overlap(A, S) for all S in supers(B). -// -// Alternatively: -// -// overlap(A, B) iff there exists: -// A' in { A, subregs(A) } and B' in { B, subregs(B) } such that: -// A' = B' or A' in aliases(B') or B' in aliases(A'). -// -// Here subregs(A) is the full flattened sub-register set returned by -// A.getSubRegs() while aliases(A) is simply the special 'Aliases' field in the -// description of register A. -// -// This also implies that registers with a common sub-register are considered -// overlapping. This can happen when forming register pairs: -// -// P0 = (R0, R1) -// P1 = (R1, R2) -// P2 = (R2, R3) -// -// In this case, we will infer an overlap between P0 and P1 because of the -// shared sub-register R1. There is no overlap between P0 and P2. -// -void CodeGenRegBank:: -computeOverlaps(std::map &Map) { - assert(Map.empty()); - - // Collect overlaps that don't follow from rule 2. - for (unsigned i = 0, e = Registers.size(); i != e; ++i) { - CodeGenRegister *Reg = Registers[i]; - CodeGenRegister::Set &Overlaps = Map[Reg]; - - // Reg overlaps itself. - Overlaps.insert(Reg); - - // All super-registers overlap. - const CodeGenRegister::SuperRegList &Supers = Reg->getSuperRegs(); - Overlaps.insert(Supers.begin(), Supers.end()); - - // Form symmetrical relations from the special Aliases[] lists. - std::vector RegList = Reg->TheDef->getValueAsListOfDefs("Aliases"); - for (unsigned i2 = 0, e2 = RegList.size(); i2 != e2; ++i2) { - CodeGenRegister *Reg2 = getReg(RegList[i2]); - CodeGenRegister::Set &Overlaps2 = Map[Reg2]; - const CodeGenRegister::SuperRegList &Supers2 = Reg2->getSuperRegs(); - // Reg overlaps Reg2 which implies it overlaps supers(Reg2). - Overlaps.insert(Reg2); - Overlaps.insert(Supers2.begin(), Supers2.end()); - Overlaps2.insert(Reg); - Overlaps2.insert(Supers.begin(), Supers.end()); + // For each register unit, ensure that we have the list of UnitSets that + // contain the unit. Normally, this matches an existing list of UnitSets for a + // register class. If not, we create a new entry in RegClassUnitSets as a + // "fake" register class. + for (unsigned UnitIdx = 0, UnitEnd = NumNativeRegUnits; + UnitIdx < UnitEnd; ++UnitIdx) { + std::vector RUSets; + for (unsigned i = 0, e = RegUnitSets.size(); i != e; ++i) { + RegUnitSet &RUSet = RegUnitSets[i]; + if (std::find(RUSet.Units.begin(), RUSet.Units.end(), UnitIdx) + == RUSet.Units.end()) + continue; + RUSets.push_back(i); } - } - - // Apply rule 2. and inherit all sub-register overlaps. - for (unsigned i = 0, e = Registers.size(); i != e; ++i) { - CodeGenRegister *Reg = Registers[i]; - CodeGenRegister::Set &Overlaps = Map[Reg]; - const CodeGenRegister::SubRegMap &SRM = Reg->getSubRegs(); - for (CodeGenRegister::SubRegMap::const_iterator i2 = SRM.begin(), - e2 = SRM.end(); i2 != e2; ++i2) { - CodeGenRegister::Set &Overlaps2 = Map[i2->second]; - Overlaps.insert(Overlaps2.begin(), Overlaps2.end()); + unsigned RCUnitSetsIdx = 0; + for (unsigned e = RegClassUnitSets.size(); + RCUnitSetsIdx != e; ++RCUnitSetsIdx) { + if (RegClassUnitSets[RCUnitSetsIdx] == RUSets) { + break; + } + } + RegUnits[UnitIdx].RegClassUnitSetsIdx = RCUnitSetsIdx; + if (RCUnitSetsIdx == RegClassUnitSets.size()) { + // Create a new list of UnitSets as a "fake" register class. + RegClassUnitSets.resize(RCUnitSetsIdx + 1); + RegClassUnitSets[RCUnitSetsIdx].swap(RUSets); } } } +struct LessUnits { + const CodeGenRegBank &RegBank; + LessUnits(const CodeGenRegBank &RB): RegBank(RB) {} + + bool operator()(unsigned ID1, unsigned ID2) { + return RegBank.getRegPressureSet(ID1).Units.size() + < RegBank.getRegPressureSet(ID2).Units.size(); + } +}; + void CodeGenRegBank::computeDerivedInfo() { computeComposites(); + computeSubRegIndexLaneMasks(); // Compute a weight for each register unit created during getSubRegs. // This may create adopted register units (with unit # >= NumNativeRegUnits). @@ -1358,6 +1726,21 @@ void CodeGenRegBank::computeDerivedInfo() { // Compute a unique set of RegUnitSets. One for each RegClass and inferred // supersets for the union of overlapping sets. computeRegUnitSets(); + + // Get the weight of each set. + for (unsigned Idx = 0, EndIdx = RegUnitSets.size(); Idx != EndIdx; ++Idx) + RegUnitSets[Idx].Weight = getRegUnitSetWeight(RegUnitSets[Idx].Units); + + // Find the order of each set. + RegUnitSetOrder.reserve(RegUnitSets.size()); + for (unsigned Idx = 0, EndIdx = RegUnitSets.size(); Idx != EndIdx; ++Idx) + RegUnitSetOrder.push_back(Idx); + + std::stable_sort(RegUnitSetOrder.begin(), RegUnitSetOrder.end(), + LessUnits(*this)); + for (unsigned Idx = 0, EndIdx = RegUnitSets.size(); Idx != EndIdx; ++Idx) { + RegUnitSets[RegUnitSetOrder[Idx]].Order = Idx; + } } // @@ -1451,6 +1834,7 @@ void CodeGenRegBank::inferMatchingSuperRegClass(CodeGenRegisterClass *RC, unsigned FirstSubRegRC) { SmallVector, 16> SSPairs; + BitVector TopoSigs(getNumTopoSigs()); // Iterate in SubRegIndex numerical order to visit synthetic indices last. for (unsigned sri = 0, sre = SubRegIndices.size(); sri != sre; ++sri) { @@ -1463,12 +1847,14 @@ void CodeGenRegBank::inferMatchingSuperRegClass(CodeGenRegisterClass *RC, // Build list of (Super, Sub) pairs for this SubIdx. SSPairs.clear(); + TopoSigs.reset(); for (CodeGenRegister::Set::const_iterator RI = RC->getMembers().begin(), RE = RC->getMembers().end(); RI != RE; ++RI) { const CodeGenRegister *Super = *RI; const CodeGenRegister *Sub = Super->getSubRegs().find(SubIdx)->second; assert(Sub && "Missing sub-register"); SSPairs.push_back(std::make_pair(Super, Sub)); + TopoSigs.set(Sub->getTopoSig()); } // Iterate over sub-register class candidates. Ignore classes created by @@ -1476,6 +1862,9 @@ void CodeGenRegBank::inferMatchingSuperRegClass(CodeGenRegisterClass *RC, for (unsigned rci = FirstSubRegRC, rce = RegClasses.size(); rci != rce; ++rci) { CodeGenRegisterClass *SubRC = RegClasses[rci]; + // Topological shortcut: SubRC members have the wrong shape. + if (!TopoSigs.anyCommon(SubRC->getTopoSigs())) + continue; // Compute the subset of RC that maps into SubRC. CodeGenRegister::Set SubSet; for (unsigned i = 0, e = SSPairs.size(); i != e; ++i)