| File: | polly/lib/CodeGen/IslNodeBuilder.cpp |
| Warning: | line 1165, column 10 Value stored to 'size' during its initialization is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- IslNodeBuilder.cpp - Translate an isl AST into a LLVM-IR AST -------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file contains the IslNodeBuilder, a class to translate an isl AST into |
| 10 | // a LLVM-IR AST. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "polly/CodeGen/IslNodeBuilder.h" |
| 15 | #include "polly/CodeGen/BlockGenerators.h" |
| 16 | #include "polly/CodeGen/CodeGeneration.h" |
| 17 | #include "polly/CodeGen/IslAst.h" |
| 18 | #include "polly/CodeGen/IslExprBuilder.h" |
| 19 | #include "polly/CodeGen/LoopGeneratorsGOMP.h" |
| 20 | #include "polly/CodeGen/LoopGeneratorsKMP.h" |
| 21 | #include "polly/CodeGen/RuntimeDebugBuilder.h" |
| 22 | #include "polly/Options.h" |
| 23 | #include "polly/ScopInfo.h" |
| 24 | #include "polly/Support/ISLTools.h" |
| 25 | #include "polly/Support/SCEVValidator.h" |
| 26 | #include "polly/Support/ScopHelper.h" |
| 27 | #include "llvm/ADT/APInt.h" |
| 28 | #include "llvm/ADT/PostOrderIterator.h" |
| 29 | #include "llvm/ADT/SetVector.h" |
| 30 | #include "llvm/ADT/SmallPtrSet.h" |
| 31 | #include "llvm/ADT/Statistic.h" |
| 32 | #include "llvm/Analysis/LoopInfo.h" |
| 33 | #include "llvm/Analysis/RegionInfo.h" |
| 34 | #include "llvm/Analysis/ScalarEvolution.h" |
| 35 | #include "llvm/Analysis/ScalarEvolutionExpressions.h" |
| 36 | #include "llvm/IR/BasicBlock.h" |
| 37 | #include "llvm/IR/Constant.h" |
| 38 | #include "llvm/IR/Constants.h" |
| 39 | #include "llvm/IR/DataLayout.h" |
| 40 | #include "llvm/IR/DerivedTypes.h" |
| 41 | #include "llvm/IR/Dominators.h" |
| 42 | #include "llvm/IR/Function.h" |
| 43 | #include "llvm/IR/InstrTypes.h" |
| 44 | #include "llvm/IR/Instruction.h" |
| 45 | #include "llvm/IR/Instructions.h" |
| 46 | #include "llvm/IR/Type.h" |
| 47 | #include "llvm/IR/Value.h" |
| 48 | #include "llvm/Support/Casting.h" |
| 49 | #include "llvm/Support/CommandLine.h" |
| 50 | #include "llvm/Support/ErrorHandling.h" |
| 51 | #include "llvm/Transforms/Utils/BasicBlockUtils.h" |
| 52 | #include "isl/aff.h" |
| 53 | #include "isl/aff_type.h" |
| 54 | #include "isl/ast.h" |
| 55 | #include "isl/ast_build.h" |
| 56 | #include "isl/isl-noexceptions.h" |
| 57 | #include "isl/map.h" |
| 58 | #include "isl/set.h" |
| 59 | #include "isl/union_map.h" |
| 60 | #include "isl/union_set.h" |
| 61 | #include "isl/val.h" |
| 62 | #include <algorithm> |
| 63 | #include <cassert> |
| 64 | #include <cstdint> |
| 65 | #include <cstring> |
| 66 | #include <string> |
| 67 | #include <utility> |
| 68 | #include <vector> |
| 69 | |
| 70 | using namespace llvm; |
| 71 | using namespace polly; |
| 72 | |
| 73 | #define DEBUG_TYPE"polly-codegen" "polly-codegen" |
| 74 | |
| 75 | STATISTIC(VersionedScops, "Number of SCoPs that required versioning.")static llvm::Statistic VersionedScops = {"polly-codegen", "VersionedScops" , "Number of SCoPs that required versioning."}; |
| 76 | |
| 77 | STATISTIC(SequentialLoops, "Number of generated sequential for-loops")static llvm::Statistic SequentialLoops = {"polly-codegen", "SequentialLoops" , "Number of generated sequential for-loops"}; |
| 78 | STATISTIC(ParallelLoops, "Number of generated parallel for-loops")static llvm::Statistic ParallelLoops = {"polly-codegen", "ParallelLoops" , "Number of generated parallel for-loops"}; |
| 79 | STATISTIC(VectorLoops, "Number of generated vector for-loops")static llvm::Statistic VectorLoops = {"polly-codegen", "VectorLoops" , "Number of generated vector for-loops"}; |
| 80 | STATISTIC(IfConditions, "Number of generated if-conditions")static llvm::Statistic IfConditions = {"polly-codegen", "IfConditions" , "Number of generated if-conditions"}; |
| 81 | |
| 82 | /// OpenMP backend options |
| 83 | enum class OpenMPBackend { GNU, LLVM }; |
| 84 | |
| 85 | static cl::opt<bool> PollyGenerateRTCPrint( |
| 86 | "polly-codegen-emit-rtc-print", |
| 87 | cl::desc("Emit code that prints the runtime check result dynamically."), |
| 88 | cl::Hidden, cl::init(false), cl::ZeroOrMore, cl::cat(PollyCategory)); |
| 89 | |
| 90 | // If this option is set we always use the isl AST generator to regenerate |
| 91 | // memory accesses. Without this option set we regenerate expressions using the |
| 92 | // original SCEV expressions and only generate new expressions in case the |
| 93 | // access relation has been changed and consequently must be regenerated. |
| 94 | static cl::opt<bool> PollyGenerateExpressions( |
| 95 | "polly-codegen-generate-expressions", |
| 96 | cl::desc("Generate AST expressions for unmodified and modified accesses"), |
| 97 | cl::Hidden, cl::init(false), cl::ZeroOrMore, cl::cat(PollyCategory)); |
| 98 | |
| 99 | static cl::opt<int> PollyTargetFirstLevelCacheLineSize( |
| 100 | "polly-target-first-level-cache-line-size", |
| 101 | cl::desc("The size of the first level cache line size specified in bytes."), |
| 102 | cl::Hidden, cl::init(64), cl::ZeroOrMore, cl::cat(PollyCategory)); |
| 103 | |
| 104 | static cl::opt<OpenMPBackend> PollyOmpBackend( |
| 105 | "polly-omp-backend", cl::desc("Choose the OpenMP library to use:"), |
| 106 | cl::values(clEnumValN(OpenMPBackend::GNU, "GNU", "GNU OpenMP")llvm::cl::OptionEnumValue { "GNU", int(OpenMPBackend::GNU), "GNU OpenMP" }, |
| 107 | clEnumValN(OpenMPBackend::LLVM, "LLVM", "LLVM OpenMP")llvm::cl::OptionEnumValue { "LLVM", int(OpenMPBackend::LLVM), "LLVM OpenMP" }), |
| 108 | cl::Hidden, cl::init(OpenMPBackend::GNU), cl::cat(PollyCategory)); |
| 109 | |
| 110 | isl::ast_expr IslNodeBuilder::getUpperBound(isl::ast_node For, |
| 111 | ICmpInst::Predicate &Predicate) { |
| 112 | isl::ast_expr Cond = For.for_get_cond(); |
| 113 | isl::ast_expr Iterator = For.for_get_iterator(); |
| 114 | assert(isl_ast_expr_get_type(Cond.get()) == isl_ast_expr_op &&(static_cast <bool> (isl_ast_expr_get_type(Cond.get()) == isl_ast_expr_op && "conditional expression is not an atomic upper bound" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Cond.get()) == isl_ast_expr_op && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 115, __extension__ __PRETTY_FUNCTION__)) |
| 115 | "conditional expression is not an atomic upper bound")(static_cast <bool> (isl_ast_expr_get_type(Cond.get()) == isl_ast_expr_op && "conditional expression is not an atomic upper bound" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Cond.get()) == isl_ast_expr_op && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 115, __extension__ __PRETTY_FUNCTION__)); |
| 116 | |
| 117 | isl_ast_op_typeisl_ast_expr_op_type OpType = isl_ast_expr_get_op_type(Cond.get()); |
| 118 | |
| 119 | switch (OpType) { |
| 120 | case isl_ast_op_leisl_ast_expr_op_le: |
| 121 | Predicate = ICmpInst::ICMP_SLE; |
| 122 | break; |
| 123 | case isl_ast_op_ltisl_ast_expr_op_lt: |
| 124 | Predicate = ICmpInst::ICMP_SLT; |
| 125 | break; |
| 126 | default: |
| 127 | llvm_unreachable("Unexpected comparison type in loop condition")::llvm::llvm_unreachable_internal("Unexpected comparison type in loop condition" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 127); |
| 128 | } |
| 129 | |
| 130 | isl::ast_expr Arg0 = Cond.get_op_arg(0); |
| 131 | |
| 132 | assert(isl_ast_expr_get_type(Arg0.get()) == isl_ast_expr_id &&(static_cast <bool> (isl_ast_expr_get_type(Arg0.get()) == isl_ast_expr_id && "conditional expression is not an atomic upper bound" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Arg0.get()) == isl_ast_expr_id && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 133, __extension__ __PRETTY_FUNCTION__)) |
| 133 | "conditional expression is not an atomic upper bound")(static_cast <bool> (isl_ast_expr_get_type(Arg0.get()) == isl_ast_expr_id && "conditional expression is not an atomic upper bound" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Arg0.get()) == isl_ast_expr_id && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 133, __extension__ __PRETTY_FUNCTION__)); |
| 134 | |
| 135 | isl::id UBID = Arg0.get_id(); |
| 136 | |
| 137 | assert(isl_ast_expr_get_type(Iterator.get()) == isl_ast_expr_id &&(static_cast <bool> (isl_ast_expr_get_type(Iterator.get ()) == isl_ast_expr_id && "Could not get the iterator" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Iterator.get()) == isl_ast_expr_id && \"Could not get the iterator\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 138, __extension__ __PRETTY_FUNCTION__)) |
| 138 | "Could not get the iterator")(static_cast <bool> (isl_ast_expr_get_type(Iterator.get ()) == isl_ast_expr_id && "Could not get the iterator" ) ? void (0) : __assert_fail ("isl_ast_expr_get_type(Iterator.get()) == isl_ast_expr_id && \"Could not get the iterator\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 138, __extension__ __PRETTY_FUNCTION__)); |
| 139 | |
| 140 | isl::id IteratorID = Iterator.get_id(); |
| 141 | |
| 142 | assert(UBID.get() == IteratorID.get() &&(static_cast <bool> (UBID.get() == IteratorID.get() && "conditional expression is not an atomic upper bound") ? void (0) : __assert_fail ("UBID.get() == IteratorID.get() && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 143, __extension__ __PRETTY_FUNCTION__)) |
| 143 | "conditional expression is not an atomic upper bound")(static_cast <bool> (UBID.get() == IteratorID.get() && "conditional expression is not an atomic upper bound") ? void (0) : __assert_fail ("UBID.get() == IteratorID.get() && \"conditional expression is not an atomic upper bound\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 143, __extension__ __PRETTY_FUNCTION__)); |
| 144 | |
| 145 | return Cond.get_op_arg(1); |
| 146 | } |
| 147 | |
| 148 | /// Return true if a return value of Predicate is true for the value represented |
| 149 | /// by passed isl_ast_expr_int. |
| 150 | static bool checkIslAstExprInt(__isl_take isl_ast_expr *Expr, |
| 151 | isl_bool (*Predicate)(__isl_keep isl_val *)) { |
| 152 | if (isl_ast_expr_get_type(Expr) != isl_ast_expr_int) { |
| 153 | isl_ast_expr_free(Expr); |
| 154 | return false; |
| 155 | } |
| 156 | auto ExprVal = isl_ast_expr_get_val(Expr); |
| 157 | isl_ast_expr_free(Expr); |
| 158 | if (Predicate(ExprVal) != isl_bool_true) { |
| 159 | isl_val_free(ExprVal); |
| 160 | return false; |
| 161 | } |
| 162 | isl_val_free(ExprVal); |
| 163 | return true; |
| 164 | } |
| 165 | |
| 166 | int IslNodeBuilder::getNumberOfIterations(isl::ast_node For) { |
| 167 | assert(isl_ast_node_get_type(For.get()) == isl_ast_node_for)(static_cast <bool> (isl_ast_node_get_type(For.get()) == isl_ast_node_for) ? void (0) : __assert_fail ("isl_ast_node_get_type(For.get()) == isl_ast_node_for" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 167, __extension__ __PRETTY_FUNCTION__)); |
| 168 | isl::ast_node Body = For.for_get_body(); |
| 169 | |
| 170 | // First, check if we can actually handle this code. |
| 171 | switch (isl_ast_node_get_type(Body.get())) { |
| 172 | case isl_ast_node_user: |
| 173 | break; |
| 174 | case isl_ast_node_block: { |
| 175 | isl::ast_node_list List = Body.block_get_children(); |
| 176 | for (isl::ast_node Node : List) { |
| 177 | isl_ast_node_type NodeType = isl_ast_node_get_type(Node.get()); |
| 178 | if (NodeType != isl_ast_node_user) |
| 179 | return -1; |
| 180 | } |
| 181 | break; |
| 182 | } |
| 183 | default: |
| 184 | return -1; |
| 185 | } |
| 186 | |
| 187 | isl::ast_expr Init = For.for_get_init(); |
| 188 | if (!checkIslAstExprInt(Init.release(), isl_val_is_zero)) |
| 189 | return -1; |
| 190 | isl::ast_expr Inc = For.for_get_inc(); |
| 191 | if (!checkIslAstExprInt(Inc.release(), isl_val_is_one)) |
| 192 | return -1; |
| 193 | CmpInst::Predicate Predicate; |
| 194 | isl::ast_expr UB = getUpperBound(For, Predicate); |
| 195 | if (isl_ast_expr_get_type(UB.get()) != isl_ast_expr_int) |
| 196 | return -1; |
| 197 | isl::val UpVal = UB.get_val(); |
| 198 | int NumberIterations = UpVal.get_num_si(); |
| 199 | if (NumberIterations < 0) |
| 200 | return -1; |
| 201 | if (Predicate == CmpInst::ICMP_SLT) |
| 202 | return NumberIterations; |
| 203 | else |
| 204 | return NumberIterations + 1; |
| 205 | } |
| 206 | |
| 207 | /// Extract the values and SCEVs needed to generate code for a block. |
| 208 | static int findReferencesInBlock(struct SubtreeReferences &References, |
| 209 | const ScopStmt *Stmt, BasicBlock *BB) { |
| 210 | for (Instruction &Inst : *BB) { |
| 211 | // Include invariant loads |
| 212 | if (isa<LoadInst>(Inst)) |
| 213 | if (Value *InvariantLoad = References.GlobalMap.lookup(&Inst)) |
| 214 | References.Values.insert(InvariantLoad); |
| 215 | |
| 216 | for (Value *SrcVal : Inst.operands()) { |
| 217 | auto *Scope = References.LI.getLoopFor(BB); |
| 218 | if (canSynthesize(SrcVal, References.S, &References.SE, Scope)) { |
| 219 | References.SCEVs.insert(References.SE.getSCEVAtScope(SrcVal, Scope)); |
| 220 | continue; |
| 221 | } else if (Value *NewVal = References.GlobalMap.lookup(SrcVal)) |
| 222 | References.Values.insert(NewVal); |
| 223 | } |
| 224 | } |
| 225 | return 0; |
| 226 | } |
| 227 | |
| 228 | void polly::addReferencesFromStmt(const ScopStmt *Stmt, void *UserPtr, |
| 229 | bool CreateScalarRefs) { |
| 230 | auto &References = *static_cast<struct SubtreeReferences *>(UserPtr); |
| 231 | |
| 232 | if (Stmt->isBlockStmt()) |
| 233 | findReferencesInBlock(References, Stmt, Stmt->getBasicBlock()); |
| 234 | else if (Stmt->isRegionStmt()) { |
| 235 | for (BasicBlock *BB : Stmt->getRegion()->blocks()) |
| 236 | findReferencesInBlock(References, Stmt, BB); |
| 237 | } else { |
| 238 | assert(Stmt->isCopyStmt())(static_cast <bool> (Stmt->isCopyStmt()) ? void (0) : __assert_fail ("Stmt->isCopyStmt()", "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 238, __extension__ __PRETTY_FUNCTION__)); |
| 239 | // Copy Stmts have no instructions that we need to consider. |
| 240 | } |
| 241 | |
| 242 | for (auto &Access : *Stmt) { |
| 243 | if (References.ParamSpace) { |
| 244 | isl::space ParamSpace = Access->getLatestAccessRelation().get_space(); |
| 245 | (*References.ParamSpace) = |
| 246 | References.ParamSpace->align_params(ParamSpace); |
| 247 | } |
| 248 | |
| 249 | if (Access->isLatestArrayKind()) { |
| 250 | auto *BasePtr = Access->getLatestScopArrayInfo()->getBasePtr(); |
| 251 | if (Instruction *OpInst = dyn_cast<Instruction>(BasePtr)) |
| 252 | if (Stmt->getParent()->contains(OpInst)) |
| 253 | continue; |
| 254 | |
| 255 | References.Values.insert(BasePtr); |
| 256 | continue; |
| 257 | } |
| 258 | |
| 259 | if (CreateScalarRefs) |
| 260 | References.Values.insert(References.BlockGen.getOrCreateAlloca(*Access)); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | /// Extract the out-of-scop values and SCEVs referenced from a set describing |
| 265 | /// a ScopStmt. |
| 266 | /// |
| 267 | /// This includes the SCEVUnknowns referenced by the SCEVs used in the |
| 268 | /// statement and the base pointers of the memory accesses. For scalar |
| 269 | /// statements we force the generation of alloca memory locations and list |
| 270 | /// these locations in the set of out-of-scop values as well. |
| 271 | /// |
| 272 | /// @param Set A set which references the ScopStmt we are interested in. |
| 273 | /// @param UserPtr A void pointer that can be casted to a SubtreeReferences |
| 274 | /// structure. |
| 275 | static void addReferencesFromStmtSet(isl::set Set, |
| 276 | struct SubtreeReferences *UserPtr) { |
| 277 | isl::id Id = Set.get_tuple_id(); |
| 278 | auto *Stmt = static_cast<const ScopStmt *>(Id.get_user()); |
| 279 | return addReferencesFromStmt(Stmt, UserPtr); |
| 280 | } |
| 281 | |
| 282 | /// Extract the out-of-scop values and SCEVs referenced from a union set |
| 283 | /// referencing multiple ScopStmts. |
| 284 | /// |
| 285 | /// This includes the SCEVUnknowns referenced by the SCEVs used in the |
| 286 | /// statement and the base pointers of the memory accesses. For scalar |
| 287 | /// statements we force the generation of alloca memory locations and list |
| 288 | /// these locations in the set of out-of-scop values as well. |
| 289 | /// |
| 290 | /// @param USet A union set referencing the ScopStmts we are interested |
| 291 | /// in. |
| 292 | /// @param References The SubtreeReferences data structure through which |
| 293 | /// results are returned and further information is |
| 294 | /// provided. |
| 295 | static void |
| 296 | addReferencesFromStmtUnionSet(isl::union_set USet, |
| 297 | struct SubtreeReferences &References) { |
| 298 | |
| 299 | for (isl::set Set : USet.get_set_list()) |
| 300 | addReferencesFromStmtSet(Set, &References); |
| 301 | } |
| 302 | |
| 303 | isl::union_map |
| 304 | IslNodeBuilder::getScheduleForAstNode(const isl::ast_node &Node) { |
| 305 | return IslAstInfo::getSchedule(Node); |
| 306 | } |
| 307 | |
| 308 | void IslNodeBuilder::getReferencesInSubtree(const isl::ast_node &For, |
| 309 | SetVector<Value *> &Values, |
| 310 | SetVector<const Loop *> &Loops) { |
| 311 | SetVector<const SCEV *> SCEVs; |
| 312 | struct SubtreeReferences References = { |
| 313 | LI, SE, S, ValueMap, Values, SCEVs, getBlockGenerator(), nullptr}; |
| 314 | |
| 315 | for (const auto &I : IDToValue) |
| 316 | Values.insert(I.second); |
| 317 | |
| 318 | // NOTE: this is populated in IslNodeBuilder::addParameters |
| 319 | for (const auto &I : OutsideLoopIterations) |
| 320 | Values.insert(cast<SCEVUnknown>(I.second)->getValue()); |
| 321 | |
| 322 | isl::union_set Schedule = getScheduleForAstNode(For).domain(); |
| 323 | addReferencesFromStmtUnionSet(Schedule, References); |
| 324 | |
| 325 | for (const SCEV *Expr : SCEVs) { |
| 326 | findValues(Expr, SE, Values); |
| 327 | findLoops(Expr, Loops); |
| 328 | } |
| 329 | |
| 330 | Values.remove_if([](const Value *V) { return isa<GlobalValue>(V); }); |
| 331 | |
| 332 | /// Note: Code generation of induction variables of loops outside Scops |
| 333 | /// |
| 334 | /// Remove loops that contain the scop or that are part of the scop, as they |
| 335 | /// are considered local. This leaves only loops that are before the scop, but |
| 336 | /// do not contain the scop itself. |
| 337 | /// We ignore loops perfectly contained in the Scop because these are already |
| 338 | /// generated at `IslNodeBuilder::addParameters`. These `Loops` are loops |
| 339 | /// whose induction variables are referred to by the Scop, but the Scop is not |
| 340 | /// fully contained in these Loops. Since there can be many of these, |
| 341 | /// we choose to codegen these on-demand. |
| 342 | /// @see IslNodeBuilder::materializeNonScopLoopInductionVariable. |
| 343 | Loops.remove_if([this](const Loop *L) { |
| 344 | return S.contains(L) || L->contains(S.getEntry()); |
| 345 | }); |
| 346 | |
| 347 | // Contains Values that may need to be replaced with other values |
| 348 | // due to replacements from the ValueMap. We should make sure |
| 349 | // that we return correctly remapped values. |
| 350 | // NOTE: this code path is tested by: |
| 351 | // 1. test/Isl/CodeGen/OpenMP/single_loop_with_loop_invariant_baseptr.ll |
| 352 | // 2. test/Isl/CodeGen/OpenMP/loop-body-references-outer-values-3.ll |
| 353 | SetVector<Value *> ReplacedValues; |
| 354 | for (Value *V : Values) { |
| 355 | ReplacedValues.insert(getLatestValue(V)); |
| 356 | } |
| 357 | Values = ReplacedValues; |
| 358 | } |
| 359 | |
| 360 | void IslNodeBuilder::updateValues(ValueMapT &NewValues) { |
| 361 | SmallPtrSet<Value *, 5> Inserted; |
| 362 | |
| 363 | for (const auto &I : IDToValue) { |
| 364 | IDToValue[I.first] = NewValues[I.second]; |
| 365 | Inserted.insert(I.second); |
| 366 | } |
| 367 | |
| 368 | for (const auto &I : NewValues) { |
| 369 | if (Inserted.count(I.first)) |
| 370 | continue; |
| 371 | |
| 372 | ValueMap[I.first] = I.second; |
| 373 | } |
| 374 | } |
| 375 | |
| 376 | Value *IslNodeBuilder::getLatestValue(Value *Original) const { |
| 377 | auto It = ValueMap.find(Original); |
| 378 | if (It == ValueMap.end()) |
| 379 | return Original; |
| 380 | return It->second; |
| 381 | } |
| 382 | |
| 383 | void IslNodeBuilder::createUserVector(__isl_take isl_ast_node *User, |
| 384 | std::vector<Value *> &IVS, |
| 385 | __isl_take isl_id *IteratorID, |
| 386 | __isl_take isl_union_map *Schedule) { |
| 387 | isl_ast_expr *Expr = isl_ast_node_user_get_expr(User); |
| 388 | isl_ast_expr *StmtExpr = isl_ast_expr_get_op_arg(Expr, 0); |
| 389 | isl_id *Id = isl_ast_expr_get_id(StmtExpr); |
| 390 | isl_ast_expr_free(StmtExpr); |
| 391 | ScopStmt *Stmt = (ScopStmt *)isl_id_get_user(Id); |
| 392 | std::vector<LoopToScevMapT> VLTS(IVS.size()); |
| 393 | |
| 394 | isl_union_set *Domain = isl_union_set_from_set(Stmt->getDomain().release()); |
| 395 | Schedule = isl_union_map_intersect_domain(Schedule, Domain); |
| 396 | isl_map *S = isl_map_from_union_map(Schedule); |
| 397 | |
| 398 | auto *NewAccesses = createNewAccesses(Stmt, User); |
| 399 | createSubstitutionsVector(Expr, Stmt, VLTS, IVS, IteratorID); |
| 400 | VectorBlockGenerator::generate(BlockGen, *Stmt, VLTS, S, NewAccesses); |
| 401 | isl_id_to_ast_expr_free(NewAccesses); |
| 402 | isl_map_free(S); |
| 403 | isl_id_free(Id); |
| 404 | isl_ast_node_free(User); |
| 405 | } |
| 406 | |
| 407 | void IslNodeBuilder::createMark(__isl_take isl_ast_node *Node) { |
| 408 | auto *Id = isl_ast_node_mark_get_id(Node); |
| 409 | auto Child = isl_ast_node_mark_get_node(Node); |
| 410 | isl_ast_node_free(Node); |
| 411 | // If a child node of a 'SIMD mark' is a loop that has a single iteration, |
| 412 | // it will be optimized away and we should skip it. |
| 413 | if (strcmp(isl_id_get_name(Id), "SIMD") == 0 && |
| 414 | isl_ast_node_get_type(Child) == isl_ast_node_for) { |
| 415 | bool Vector = PollyVectorizerChoice == VECTORIZER_POLLY; |
| 416 | int VectorWidth = getNumberOfIterations(isl::manage_copy(Child)); |
| 417 | if (Vector && 1 < VectorWidth && VectorWidth <= 16) |
| 418 | createForVector(Child, VectorWidth); |
| 419 | else |
| 420 | createForSequential(isl::manage(Child), true); |
| 421 | isl_id_free(Id); |
| 422 | return; |
| 423 | } |
| 424 | if (strcmp(isl_id_get_name(Id), "Inter iteration alias-free") == 0) { |
| 425 | auto *BasePtr = static_cast<Value *>(isl_id_get_user(Id)); |
| 426 | Annotator.addInterIterationAliasFreeBasePtr(BasePtr); |
| 427 | } |
| 428 | |
| 429 | BandAttr *ChildLoopAttr = getLoopAttr(isl::manage_copy(Id)); |
| 430 | BandAttr *AncestorLoopAttr; |
| 431 | if (ChildLoopAttr) { |
| 432 | // Save current LoopAttr environment to restore again when leaving this |
| 433 | // subtree. This means there was no loop between the ancestor LoopAttr and |
| 434 | // this mark, i.e. the ancestor LoopAttr did not directly mark a loop. This |
| 435 | // can happen e.g. if the AST build peeled or unrolled the loop. |
| 436 | AncestorLoopAttr = Annotator.getStagingAttrEnv(); |
| 437 | |
| 438 | Annotator.getStagingAttrEnv() = ChildLoopAttr; |
| 439 | } |
| 440 | |
| 441 | create(Child); |
| 442 | |
| 443 | if (ChildLoopAttr) { |
| 444 | assert(Annotator.getStagingAttrEnv() == ChildLoopAttr &&(static_cast <bool> (Annotator.getStagingAttrEnv() == ChildLoopAttr && "Nest must not overwrite loop attr environment") ? void (0) : __assert_fail ("Annotator.getStagingAttrEnv() == ChildLoopAttr && \"Nest must not overwrite loop attr environment\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 445, __extension__ __PRETTY_FUNCTION__)) |
| 445 | "Nest must not overwrite loop attr environment")(static_cast <bool> (Annotator.getStagingAttrEnv() == ChildLoopAttr && "Nest must not overwrite loop attr environment") ? void (0) : __assert_fail ("Annotator.getStagingAttrEnv() == ChildLoopAttr && \"Nest must not overwrite loop attr environment\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 445, __extension__ __PRETTY_FUNCTION__)); |
| 446 | Annotator.getStagingAttrEnv() = AncestorLoopAttr; |
| 447 | } |
| 448 | |
| 449 | isl_id_free(Id); |
| 450 | } |
| 451 | |
| 452 | void IslNodeBuilder::createForVector(__isl_take isl_ast_node *For, |
| 453 | int VectorWidth) { |
| 454 | isl_ast_node *Body = isl_ast_node_for_get_body(For); |
| 455 | isl_ast_expr *Init = isl_ast_node_for_get_init(For); |
| 456 | isl_ast_expr *Inc = isl_ast_node_for_get_inc(For); |
| 457 | isl_ast_expr *Iterator = isl_ast_node_for_get_iterator(For); |
| 458 | isl_id *IteratorID = isl_ast_expr_get_id(Iterator); |
| 459 | |
| 460 | Value *ValueLB = ExprBuilder.create(Init); |
| 461 | Value *ValueInc = ExprBuilder.create(Inc); |
| 462 | |
| 463 | Type *MaxType = ExprBuilder.getType(Iterator); |
| 464 | MaxType = ExprBuilder.getWidestType(MaxType, ValueLB->getType()); |
| 465 | MaxType = ExprBuilder.getWidestType(MaxType, ValueInc->getType()); |
| 466 | |
| 467 | if (MaxType != ValueLB->getType()) |
| 468 | ValueLB = Builder.CreateSExt(ValueLB, MaxType); |
| 469 | if (MaxType != ValueInc->getType()) |
| 470 | ValueInc = Builder.CreateSExt(ValueInc, MaxType); |
| 471 | |
| 472 | std::vector<Value *> IVS(VectorWidth); |
| 473 | IVS[0] = ValueLB; |
| 474 | |
| 475 | for (int i = 1; i < VectorWidth; i++) |
| 476 | IVS[i] = Builder.CreateAdd(IVS[i - 1], ValueInc, "p_vector_iv"); |
| 477 | |
| 478 | isl::union_map Schedule = getScheduleForAstNode(isl::manage_copy(For)); |
| 479 | assert(!Schedule.is_null() &&(static_cast <bool> (!Schedule.is_null() && "For statement annotation does not contain its schedule" ) ? void (0) : __assert_fail ("!Schedule.is_null() && \"For statement annotation does not contain its schedule\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 480, __extension__ __PRETTY_FUNCTION__)) |
| 480 | "For statement annotation does not contain its schedule")(static_cast <bool> (!Schedule.is_null() && "For statement annotation does not contain its schedule" ) ? void (0) : __assert_fail ("!Schedule.is_null() && \"For statement annotation does not contain its schedule\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 480, __extension__ __PRETTY_FUNCTION__)); |
| 481 | |
| 482 | IDToValue[IteratorID] = ValueLB; |
| 483 | |
| 484 | switch (isl_ast_node_get_type(Body)) { |
| 485 | case isl_ast_node_user: |
| 486 | createUserVector(Body, IVS, isl_id_copy(IteratorID), Schedule.copy()); |
| 487 | break; |
| 488 | case isl_ast_node_block: { |
| 489 | isl_ast_node_list *List = isl_ast_node_block_get_children(Body); |
| 490 | |
| 491 | for (int i = 0; i < isl_ast_node_list_n_ast_node(List); ++i) |
| 492 | createUserVector(isl_ast_node_list_get_ast_node(List, i), IVS, |
| 493 | isl_id_copy(IteratorID), Schedule.copy()); |
| 494 | |
| 495 | isl_ast_node_free(Body); |
| 496 | isl_ast_node_list_free(List); |
| 497 | break; |
| 498 | } |
| 499 | default: |
| 500 | isl_ast_node_dump(Body); |
| 501 | llvm_unreachable("Unhandled isl_ast_node in vectorizer")::llvm::llvm_unreachable_internal("Unhandled isl_ast_node in vectorizer" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 501); |
| 502 | } |
| 503 | |
| 504 | IDToValue.erase(IDToValue.find(IteratorID)); |
| 505 | isl_id_free(IteratorID); |
| 506 | |
| 507 | isl_ast_node_free(For); |
| 508 | isl_ast_expr_free(Iterator); |
| 509 | |
| 510 | VectorLoops++; |
| 511 | } |
| 512 | |
| 513 | /// Restore the initial ordering of dimensions of the band node |
| 514 | /// |
| 515 | /// In case the band node represents all the dimensions of the iteration |
| 516 | /// domain, recreate the band node to restore the initial ordering of the |
| 517 | /// dimensions. |
| 518 | /// |
| 519 | /// @param Node The band node to be modified. |
| 520 | /// @return The modified schedule node. |
| 521 | static bool IsLoopVectorizerDisabled(isl::ast_node Node) { |
| 522 | assert(isl_ast_node_get_type(Node.get()) == isl_ast_node_for)(static_cast <bool> (isl_ast_node_get_type(Node.get()) == isl_ast_node_for) ? void (0) : __assert_fail ("isl_ast_node_get_type(Node.get()) == isl_ast_node_for" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 522, __extension__ __PRETTY_FUNCTION__)); |
| 523 | auto Body = Node.for_get_body(); |
| 524 | if (isl_ast_node_get_type(Body.get()) != isl_ast_node_mark) |
| 525 | return false; |
| 526 | auto Id = Body.mark_get_id(); |
| 527 | if (strcmp(Id.get_name().c_str(), "Loop Vectorizer Disabled") == 0) |
| 528 | return true; |
| 529 | return false; |
| 530 | } |
| 531 | |
| 532 | void IslNodeBuilder::createForSequential(isl::ast_node For, bool MarkParallel) { |
| 533 | Value *ValueLB, *ValueUB, *ValueInc; |
| 534 | Type *MaxType; |
| 535 | BasicBlock *ExitBlock; |
| 536 | Value *IV; |
| 537 | CmpInst::Predicate Predicate; |
| 538 | |
| 539 | bool LoopVectorizerDisabled = IsLoopVectorizerDisabled(For); |
| 540 | |
| 541 | isl::ast_node Body = For.for_get_body(); |
| 542 | |
| 543 | // isl_ast_node_for_is_degenerate(For) |
| 544 | // |
| 545 | // TODO: For degenerated loops we could generate a plain assignment. |
| 546 | // However, for now we just reuse the logic for normal loops, which will |
| 547 | // create a loop with a single iteration. |
| 548 | |
| 549 | isl::ast_expr Init = For.for_get_init(); |
| 550 | isl::ast_expr Inc = For.for_get_inc(); |
| 551 | isl::ast_expr Iterator = For.for_get_iterator(); |
| 552 | isl::id IteratorID = Iterator.get_id(); |
| 553 | isl::ast_expr UB = getUpperBound(For, Predicate); |
| 554 | |
| 555 | ValueLB = ExprBuilder.create(Init.release()); |
| 556 | ValueUB = ExprBuilder.create(UB.release()); |
| 557 | ValueInc = ExprBuilder.create(Inc.release()); |
| 558 | |
| 559 | MaxType = ExprBuilder.getType(Iterator.get()); |
| 560 | MaxType = ExprBuilder.getWidestType(MaxType, ValueLB->getType()); |
| 561 | MaxType = ExprBuilder.getWidestType(MaxType, ValueUB->getType()); |
| 562 | MaxType = ExprBuilder.getWidestType(MaxType, ValueInc->getType()); |
| 563 | |
| 564 | if (MaxType != ValueLB->getType()) |
| 565 | ValueLB = Builder.CreateSExt(ValueLB, MaxType); |
| 566 | if (MaxType != ValueUB->getType()) |
| 567 | ValueUB = Builder.CreateSExt(ValueUB, MaxType); |
| 568 | if (MaxType != ValueInc->getType()) |
| 569 | ValueInc = Builder.CreateSExt(ValueInc, MaxType); |
| 570 | |
| 571 | // If we can show that LB <Predicate> UB holds at least once, we can |
| 572 | // omit the GuardBB in front of the loop. |
| 573 | bool UseGuardBB = |
| 574 | !SE.isKnownPredicate(Predicate, SE.getSCEV(ValueLB), SE.getSCEV(ValueUB)); |
| 575 | IV = createLoop(ValueLB, ValueUB, ValueInc, Builder, LI, DT, ExitBlock, |
| 576 | Predicate, &Annotator, MarkParallel, UseGuardBB, |
| 577 | LoopVectorizerDisabled); |
| 578 | IDToValue[IteratorID.get()] = IV; |
| 579 | |
| 580 | create(Body.release()); |
| 581 | |
| 582 | Annotator.popLoop(MarkParallel); |
| 583 | |
| 584 | IDToValue.erase(IDToValue.find(IteratorID.get())); |
| 585 | |
| 586 | Builder.SetInsertPoint(&ExitBlock->front()); |
| 587 | |
| 588 | SequentialLoops++; |
| 589 | } |
| 590 | |
| 591 | /// Remove the BBs contained in a (sub)function from the dominator tree. |
| 592 | /// |
| 593 | /// This function removes the basic blocks that are part of a subfunction from |
| 594 | /// the dominator tree. Specifically, when generating code it may happen that at |
| 595 | /// some point the code generation continues in a new sub-function (e.g., when |
| 596 | /// generating OpenMP code). The basic blocks that are created in this |
| 597 | /// sub-function are then still part of the dominator tree of the original |
| 598 | /// function, such that the dominator tree reaches over function boundaries. |
| 599 | /// This is not only incorrect, but also causes crashes. This function now |
| 600 | /// removes from the dominator tree all basic blocks that are dominated (and |
| 601 | /// consequently reachable) from the entry block of this (sub)function. |
| 602 | /// |
| 603 | /// FIXME: A LLVM (function or region) pass should not touch anything outside of |
| 604 | /// the function/region it runs on. Hence, the pure need for this function shows |
| 605 | /// that we do not comply to this rule. At the moment, this does not cause any |
| 606 | /// issues, but we should be aware that such issues may appear. Unfortunately |
| 607 | /// the current LLVM pass infrastructure does not allow to make Polly a module |
| 608 | /// or call-graph pass to solve this issue, as such a pass would not have access |
| 609 | /// to the per-function analyses passes needed by Polly. A future pass manager |
| 610 | /// infrastructure is supposed to enable such kind of access possibly allowing |
| 611 | /// us to create a cleaner solution here. |
| 612 | /// |
| 613 | /// FIXME: Instead of adding the dominance information and then dropping it |
| 614 | /// later on, we should try to just not add it in the first place. This requires |
| 615 | /// some careful testing to make sure this does not break in interaction with |
| 616 | /// the SCEVBuilder and SplitBlock which may rely on the dominator tree or |
| 617 | /// which may try to update it. |
| 618 | /// |
| 619 | /// @param F The function which contains the BBs to removed. |
| 620 | /// @param DT The dominator tree from which to remove the BBs. |
| 621 | static void removeSubFuncFromDomTree(Function *F, DominatorTree &DT) { |
| 622 | DomTreeNode *N = DT.getNode(&F->getEntryBlock()); |
| 623 | std::vector<BasicBlock *> Nodes; |
| 624 | |
| 625 | // We can only remove an element from the dominator tree, if all its children |
| 626 | // have been removed. To ensure this we obtain the list of nodes to remove |
| 627 | // using a post-order tree traversal. |
| 628 | for (po_iterator<DomTreeNode *> I = po_begin(N), E = po_end(N); I != E; ++I) |
| 629 | Nodes.push_back(I->getBlock()); |
| 630 | |
| 631 | for (BasicBlock *BB : Nodes) |
| 632 | DT.eraseNode(BB); |
| 633 | } |
| 634 | |
| 635 | void IslNodeBuilder::createForParallel(__isl_take isl_ast_node *For) { |
| 636 | isl_ast_node *Body; |
| 637 | isl_ast_expr *Init, *Inc, *Iterator, *UB; |
| 638 | isl_id *IteratorID; |
| 639 | Value *ValueLB, *ValueUB, *ValueInc; |
| 640 | Type *MaxType; |
| 641 | Value *IV; |
| 642 | CmpInst::Predicate Predicate; |
| 643 | |
| 644 | // The preamble of parallel code interacts different than normal code with |
| 645 | // e.g., scalar initialization. Therefore, we ensure the parallel code is |
| 646 | // separated from the last basic block. |
| 647 | BasicBlock *ParBB = SplitBlock(Builder.GetInsertBlock(), |
| 648 | &*Builder.GetInsertPoint(), &DT, &LI); |
| 649 | ParBB->setName("polly.parallel.for"); |
| 650 | Builder.SetInsertPoint(&ParBB->front()); |
| 651 | |
| 652 | Body = isl_ast_node_for_get_body(For); |
| 653 | Init = isl_ast_node_for_get_init(For); |
| 654 | Inc = isl_ast_node_for_get_inc(For); |
| 655 | Iterator = isl_ast_node_for_get_iterator(For); |
| 656 | IteratorID = isl_ast_expr_get_id(Iterator); |
| 657 | UB = getUpperBound(isl::manage_copy(For), Predicate).release(); |
| 658 | |
| 659 | ValueLB = ExprBuilder.create(Init); |
| 660 | ValueUB = ExprBuilder.create(UB); |
| 661 | ValueInc = ExprBuilder.create(Inc); |
| 662 | |
| 663 | // OpenMP always uses SLE. In case the isl generated AST uses a SLT |
| 664 | // expression, we need to adjust the loop bound by one. |
| 665 | if (Predicate == CmpInst::ICMP_SLT) |
| 666 | ValueUB = Builder.CreateAdd( |
| 667 | ValueUB, Builder.CreateSExt(Builder.getTrue(), ValueUB->getType())); |
| 668 | |
| 669 | MaxType = ExprBuilder.getType(Iterator); |
| 670 | MaxType = ExprBuilder.getWidestType(MaxType, ValueLB->getType()); |
| 671 | MaxType = ExprBuilder.getWidestType(MaxType, ValueUB->getType()); |
| 672 | MaxType = ExprBuilder.getWidestType(MaxType, ValueInc->getType()); |
| 673 | |
| 674 | if (MaxType != ValueLB->getType()) |
| 675 | ValueLB = Builder.CreateSExt(ValueLB, MaxType); |
| 676 | if (MaxType != ValueUB->getType()) |
| 677 | ValueUB = Builder.CreateSExt(ValueUB, MaxType); |
| 678 | if (MaxType != ValueInc->getType()) |
| 679 | ValueInc = Builder.CreateSExt(ValueInc, MaxType); |
| 680 | |
| 681 | BasicBlock::iterator LoopBody; |
| 682 | |
| 683 | SetVector<Value *> SubtreeValues; |
| 684 | SetVector<const Loop *> Loops; |
| 685 | |
| 686 | getReferencesInSubtree(isl::manage_copy(For), SubtreeValues, Loops); |
| 687 | |
| 688 | // Create for all loops we depend on values that contain the current loop |
| 689 | // iteration. These values are necessary to generate code for SCEVs that |
| 690 | // depend on such loops. As a result we need to pass them to the subfunction. |
| 691 | // See [Code generation of induction variables of loops outside Scops] |
| 692 | for (const Loop *L : Loops) { |
| 693 | Value *LoopInductionVar = materializeNonScopLoopInductionVariable(L); |
| 694 | SubtreeValues.insert(LoopInductionVar); |
| 695 | } |
| 696 | |
| 697 | ValueMapT NewValues; |
| 698 | |
| 699 | std::unique_ptr<ParallelLoopGenerator> ParallelLoopGenPtr; |
| 700 | |
| 701 | switch (PollyOmpBackend) { |
| 702 | case OpenMPBackend::GNU: |
| 703 | ParallelLoopGenPtr.reset( |
| 704 | new ParallelLoopGeneratorGOMP(Builder, LI, DT, DL)); |
| 705 | break; |
| 706 | case OpenMPBackend::LLVM: |
| 707 | ParallelLoopGenPtr.reset(new ParallelLoopGeneratorKMP(Builder, LI, DT, DL)); |
| 708 | break; |
| 709 | } |
| 710 | |
| 711 | IV = ParallelLoopGenPtr->createParallelLoop( |
| 712 | ValueLB, ValueUB, ValueInc, SubtreeValues, NewValues, &LoopBody); |
| 713 | BasicBlock::iterator AfterLoop = Builder.GetInsertPoint(); |
| 714 | Builder.SetInsertPoint(&*LoopBody); |
| 715 | |
| 716 | // Remember the parallel subfunction |
| 717 | ParallelSubfunctions.push_back(LoopBody->getFunction()); |
| 718 | |
| 719 | // Save the current values. |
| 720 | auto ValueMapCopy = ValueMap; |
| 721 | IslExprBuilder::IDToValueTy IDToValueCopy = IDToValue; |
| 722 | |
| 723 | updateValues(NewValues); |
| 724 | IDToValue[IteratorID] = IV; |
| 725 | |
| 726 | ValueMapT NewValuesReverse; |
| 727 | |
| 728 | for (auto P : NewValues) |
| 729 | NewValuesReverse[P.second] = P.first; |
| 730 | |
| 731 | Annotator.addAlternativeAliasBases(NewValuesReverse); |
| 732 | |
| 733 | create(Body); |
| 734 | |
| 735 | Annotator.resetAlternativeAliasBases(); |
| 736 | // Restore the original values. |
| 737 | ValueMap = ValueMapCopy; |
| 738 | IDToValue = IDToValueCopy; |
| 739 | |
| 740 | Builder.SetInsertPoint(&*AfterLoop); |
| 741 | removeSubFuncFromDomTree((*LoopBody).getParent()->getParent(), DT); |
| 742 | |
| 743 | for (const Loop *L : Loops) |
| 744 | OutsideLoopIterations.erase(L); |
| 745 | |
| 746 | isl_ast_node_free(For); |
| 747 | isl_ast_expr_free(Iterator); |
| 748 | isl_id_free(IteratorID); |
| 749 | |
| 750 | ParallelLoops++; |
| 751 | } |
| 752 | |
| 753 | /// Return whether any of @p Node's statements contain partial accesses. |
| 754 | /// |
| 755 | /// Partial accesses are not supported by Polly's vector code generator. |
| 756 | static bool hasPartialAccesses(__isl_take isl_ast_node *Node) { |
| 757 | return isl_ast_node_foreach_descendant_top_down( |
| 758 | Node, |
| 759 | [](isl_ast_node *Node, void *User) -> isl_bool { |
| 760 | if (isl_ast_node_get_type(Node) != isl_ast_node_user) |
| 761 | return isl_bool_true; |
| 762 | |
| 763 | isl::ast_expr Expr = |
| 764 | isl::manage(isl_ast_node_user_get_expr(Node)); |
| 765 | isl::ast_expr StmtExpr = Expr.get_op_arg(0); |
| 766 | isl::id Id = StmtExpr.get_id(); |
| 767 | |
| 768 | ScopStmt *Stmt = |
| 769 | static_cast<ScopStmt *>(isl_id_get_user(Id.get())); |
| 770 | isl::set StmtDom = Stmt->getDomain(); |
| 771 | for (auto *MA : *Stmt) { |
| 772 | if (MA->isLatestPartialAccess()) |
| 773 | return isl_bool_error; |
| 774 | } |
| 775 | return isl_bool_true; |
| 776 | }, |
| 777 | nullptr) == isl_stat_error; |
| 778 | } |
| 779 | |
| 780 | void IslNodeBuilder::createFor(__isl_take isl_ast_node *For) { |
| 781 | bool Vector = PollyVectorizerChoice == VECTORIZER_POLLY; |
| 782 | |
| 783 | if (Vector && IslAstInfo::isInnermostParallel(isl::manage_copy(For)) && |
| 784 | !IslAstInfo::isReductionParallel(isl::manage_copy(For))) { |
| 785 | int VectorWidth = getNumberOfIterations(isl::manage_copy(For)); |
| 786 | if (1 < VectorWidth && VectorWidth <= 16 && !hasPartialAccesses(For)) { |
| 787 | createForVector(For, VectorWidth); |
| 788 | return; |
| 789 | } |
| 790 | } |
| 791 | |
| 792 | if (IslAstInfo::isExecutedInParallel(isl::manage_copy(For))) { |
| 793 | createForParallel(For); |
| 794 | return; |
| 795 | } |
| 796 | bool Parallel = (IslAstInfo::isParallel(isl::manage_copy(For)) && |
| 797 | !IslAstInfo::isReductionParallel(isl::manage_copy(For))); |
| 798 | createForSequential(isl::manage(For), Parallel); |
| 799 | } |
| 800 | |
| 801 | void IslNodeBuilder::createIf(__isl_take isl_ast_node *If) { |
| 802 | isl_ast_expr *Cond = isl_ast_node_if_get_cond(If); |
| 803 | |
| 804 | Function *F = Builder.GetInsertBlock()->getParent(); |
| 805 | LLVMContext &Context = F->getContext(); |
| 806 | |
| 807 | BasicBlock *CondBB = SplitBlock(Builder.GetInsertBlock(), |
| 808 | &*Builder.GetInsertPoint(), &DT, &LI); |
| 809 | CondBB->setName("polly.cond"); |
| 810 | BasicBlock *MergeBB = SplitBlock(CondBB, &CondBB->front(), &DT, &LI); |
| 811 | MergeBB->setName("polly.merge"); |
| 812 | BasicBlock *ThenBB = BasicBlock::Create(Context, "polly.then", F); |
| 813 | BasicBlock *ElseBB = BasicBlock::Create(Context, "polly.else", F); |
| 814 | |
| 815 | DT.addNewBlock(ThenBB, CondBB); |
| 816 | DT.addNewBlock(ElseBB, CondBB); |
| 817 | DT.changeImmediateDominator(MergeBB, CondBB); |
| 818 | |
| 819 | Loop *L = LI.getLoopFor(CondBB); |
| 820 | if (L) { |
| 821 | L->addBasicBlockToLoop(ThenBB, LI); |
| 822 | L->addBasicBlockToLoop(ElseBB, LI); |
| 823 | } |
| 824 | |
| 825 | CondBB->getTerminator()->eraseFromParent(); |
| 826 | |
| 827 | Builder.SetInsertPoint(CondBB); |
| 828 | Value *Predicate = ExprBuilder.create(Cond); |
| 829 | Builder.CreateCondBr(Predicate, ThenBB, ElseBB); |
| 830 | Builder.SetInsertPoint(ThenBB); |
| 831 | Builder.CreateBr(MergeBB); |
| 832 | Builder.SetInsertPoint(ElseBB); |
| 833 | Builder.CreateBr(MergeBB); |
| 834 | Builder.SetInsertPoint(&ThenBB->front()); |
| 835 | |
| 836 | create(isl_ast_node_if_get_then(If)); |
| 837 | |
| 838 | Builder.SetInsertPoint(&ElseBB->front()); |
| 839 | |
| 840 | if (isl_ast_node_if_has_else(If)) |
| 841 | create(isl_ast_node_if_get_else(If)); |
| 842 | |
| 843 | Builder.SetInsertPoint(&MergeBB->front()); |
| 844 | |
| 845 | isl_ast_node_free(If); |
| 846 | |
| 847 | IfConditions++; |
| 848 | } |
| 849 | |
| 850 | __isl_give isl_id_to_ast_expr * |
| 851 | IslNodeBuilder::createNewAccesses(ScopStmt *Stmt, |
| 852 | __isl_keep isl_ast_node *Node) { |
| 853 | isl::id_to_ast_expr NewAccesses = |
| 854 | isl::id_to_ast_expr::alloc(Stmt->getParent()->getIslCtx(), 0); |
| 855 | |
| 856 | isl::ast_build Build = IslAstInfo::getBuild(isl::manage_copy(Node)); |
| 857 | assert(!Build.is_null() && "Could not obtain isl_ast_build from user node")(static_cast <bool> (!Build.is_null() && "Could not obtain isl_ast_build from user node" ) ? void (0) : __assert_fail ("!Build.is_null() && \"Could not obtain isl_ast_build from user node\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 857, __extension__ __PRETTY_FUNCTION__)); |
| 858 | Stmt->setAstBuild(Build); |
| 859 | |
| 860 | for (auto *MA : *Stmt) { |
| 861 | if (!MA->hasNewAccessRelation()) { |
| 862 | if (PollyGenerateExpressions) { |
| 863 | if (!MA->isAffine()) |
| 864 | continue; |
| 865 | if (MA->getLatestScopArrayInfo()->getBasePtrOriginSAI()) |
| 866 | continue; |
| 867 | |
| 868 | auto *BasePtr = |
| 869 | dyn_cast<Instruction>(MA->getLatestScopArrayInfo()->getBasePtr()); |
| 870 | if (BasePtr && Stmt->getParent()->getRegion().contains(BasePtr)) |
| 871 | continue; |
| 872 | } else { |
| 873 | continue; |
| 874 | } |
| 875 | } |
| 876 | assert(MA->isAffine() &&(static_cast <bool> (MA->isAffine() && "Only affine memory accesses can be code generated" ) ? void (0) : __assert_fail ("MA->isAffine() && \"Only affine memory accesses can be code generated\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 877, __extension__ __PRETTY_FUNCTION__)) |
| 877 | "Only affine memory accesses can be code generated")(static_cast <bool> (MA->isAffine() && "Only affine memory accesses can be code generated" ) ? void (0) : __assert_fail ("MA->isAffine() && \"Only affine memory accesses can be code generated\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 877, __extension__ __PRETTY_FUNCTION__)); |
| 878 | |
| 879 | isl::union_map Schedule = Build.get_schedule(); |
| 880 | |
| 881 | #ifndef NDEBUG |
| 882 | if (MA->isRead()) { |
| 883 | auto Dom = Stmt->getDomain().release(); |
| 884 | auto SchedDom = isl_set_from_union_set(Schedule.domain().release()); |
| 885 | auto AccDom = isl_map_domain(MA->getAccessRelation().release()); |
| 886 | Dom = isl_set_intersect_params(Dom, |
| 887 | Stmt->getParent()->getContext().release()); |
| 888 | SchedDom = isl_set_intersect_params( |
| 889 | SchedDom, Stmt->getParent()->getContext().release()); |
| 890 | assert(isl_set_is_subset(SchedDom, AccDom) &&(static_cast <bool> (isl_set_is_subset(SchedDom, AccDom ) && "Access relation not defined on full schedule domain" ) ? void (0) : __assert_fail ("isl_set_is_subset(SchedDom, AccDom) && \"Access relation not defined on full schedule domain\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 891, __extension__ __PRETTY_FUNCTION__)) |
| 891 | "Access relation not defined on full schedule domain")(static_cast <bool> (isl_set_is_subset(SchedDom, AccDom ) && "Access relation not defined on full schedule domain" ) ? void (0) : __assert_fail ("isl_set_is_subset(SchedDom, AccDom) && \"Access relation not defined on full schedule domain\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 891, __extension__ __PRETTY_FUNCTION__)); |
| 892 | assert(isl_set_is_subset(Dom, AccDom) &&(static_cast <bool> (isl_set_is_subset(Dom, AccDom) && "Access relation not defined on full domain") ? void (0) : __assert_fail ("isl_set_is_subset(Dom, AccDom) && \"Access relation not defined on full domain\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 893, __extension__ __PRETTY_FUNCTION__)) |
| 893 | "Access relation not defined on full domain")(static_cast <bool> (isl_set_is_subset(Dom, AccDom) && "Access relation not defined on full domain") ? void (0) : __assert_fail ("isl_set_is_subset(Dom, AccDom) && \"Access relation not defined on full domain\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 893, __extension__ __PRETTY_FUNCTION__)); |
| 894 | isl_set_free(AccDom); |
| 895 | isl_set_free(SchedDom); |
| 896 | isl_set_free(Dom); |
| 897 | } |
| 898 | #endif |
| 899 | |
| 900 | isl::pw_multi_aff PWAccRel = MA->applyScheduleToAccessRelation(Schedule); |
| 901 | |
| 902 | // isl cannot generate an index expression for access-nothing accesses. |
| 903 | isl::set AccDomain = PWAccRel.domain(); |
| 904 | isl::set Context = S.getContext(); |
| 905 | AccDomain = AccDomain.intersect_params(Context); |
| 906 | if (AccDomain.is_empty()) |
| 907 | continue; |
| 908 | |
| 909 | isl::ast_expr AccessExpr = Build.access_from(PWAccRel); |
| 910 | NewAccesses = NewAccesses.set(MA->getId(), AccessExpr); |
| 911 | } |
| 912 | |
| 913 | return NewAccesses.release(); |
| 914 | } |
| 915 | |
| 916 | void IslNodeBuilder::createSubstitutions(__isl_take isl_ast_expr *Expr, |
| 917 | ScopStmt *Stmt, LoopToScevMapT <S) { |
| 918 | assert(isl_ast_expr_get_type(Expr) == isl_ast_expr_op &&(static_cast <bool> (isl_ast_expr_get_type(Expr) == isl_ast_expr_op && "Expression of type 'op' expected") ? void (0) : __assert_fail ("isl_ast_expr_get_type(Expr) == isl_ast_expr_op && \"Expression of type 'op' expected\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 919, __extension__ __PRETTY_FUNCTION__)) |
| 919 | "Expression of type 'op' expected")(static_cast <bool> (isl_ast_expr_get_type(Expr) == isl_ast_expr_op && "Expression of type 'op' expected") ? void (0) : __assert_fail ("isl_ast_expr_get_type(Expr) == isl_ast_expr_op && \"Expression of type 'op' expected\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 919, __extension__ __PRETTY_FUNCTION__)); |
| 920 | assert(isl_ast_expr_get_op_type(Expr) == isl_ast_op_call &&(static_cast <bool> (isl_ast_expr_get_op_type(Expr) == isl_ast_expr_op_call && "Operation of type 'call' expected") ? void (0) : __assert_fail ("isl_ast_expr_get_op_type(Expr) == isl_ast_op_call && \"Operation of type 'call' expected\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 921, __extension__ __PRETTY_FUNCTION__)) |
| 921 | "Operation of type 'call' expected")(static_cast <bool> (isl_ast_expr_get_op_type(Expr) == isl_ast_expr_op_call && "Operation of type 'call' expected") ? void (0) : __assert_fail ("isl_ast_expr_get_op_type(Expr) == isl_ast_op_call && \"Operation of type 'call' expected\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 921, __extension__ __PRETTY_FUNCTION__)); |
| 922 | for (int i = 0; i < isl_ast_expr_get_op_n_arg(Expr) - 1; ++i) { |
| 923 | isl_ast_expr *SubExpr; |
| 924 | Value *V; |
| 925 | |
| 926 | SubExpr = isl_ast_expr_get_op_arg(Expr, i + 1); |
| 927 | V = ExprBuilder.create(SubExpr); |
| 928 | ScalarEvolution *SE = Stmt->getParent()->getSE(); |
| 929 | LTS[Stmt->getLoopForDimension(i)] = SE->getUnknown(V); |
| 930 | } |
| 931 | |
| 932 | isl_ast_expr_free(Expr); |
| 933 | } |
| 934 | |
| 935 | void IslNodeBuilder::createSubstitutionsVector( |
| 936 | __isl_take isl_ast_expr *Expr, ScopStmt *Stmt, |
| 937 | std::vector<LoopToScevMapT> &VLTS, std::vector<Value *> &IVS, |
| 938 | __isl_take isl_id *IteratorID) { |
| 939 | int i = 0; |
| 940 | |
| 941 | Value *OldValue = IDToValue[IteratorID]; |
| 942 | for (Value *IV : IVS) { |
| 943 | IDToValue[IteratorID] = IV; |
| 944 | createSubstitutions(isl_ast_expr_copy(Expr), Stmt, VLTS[i]); |
| 945 | i++; |
| 946 | } |
| 947 | |
| 948 | IDToValue[IteratorID] = OldValue; |
| 949 | isl_id_free(IteratorID); |
| 950 | isl_ast_expr_free(Expr); |
| 951 | } |
| 952 | |
| 953 | void IslNodeBuilder::generateCopyStmt( |
| 954 | ScopStmt *Stmt, __isl_keep isl_id_to_ast_expr *NewAccesses) { |
| 955 | assert(Stmt->size() == 2)(static_cast <bool> (Stmt->size() == 2) ? void (0) : __assert_fail ("Stmt->size() == 2", "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 955, __extension__ __PRETTY_FUNCTION__)); |
| 956 | auto ReadAccess = Stmt->begin(); |
| 957 | auto WriteAccess = ReadAccess++; |
| 958 | assert((*ReadAccess)->isRead() && (*WriteAccess)->isMustWrite())(static_cast <bool> ((*ReadAccess)->isRead() && (*WriteAccess)->isMustWrite()) ? void (0) : __assert_fail ("(*ReadAccess)->isRead() && (*WriteAccess)->isMustWrite()" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 958, __extension__ __PRETTY_FUNCTION__)); |
| 959 | assert((*ReadAccess)->getElementType() == (*WriteAccess)->getElementType() &&(static_cast <bool> ((*ReadAccess)->getElementType() == (*WriteAccess)->getElementType() && "Accesses use the same data type" ) ? void (0) : __assert_fail ("(*ReadAccess)->getElementType() == (*WriteAccess)->getElementType() && \"Accesses use the same data type\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 960, __extension__ __PRETTY_FUNCTION__)) |
| 960 | "Accesses use the same data type")(static_cast <bool> ((*ReadAccess)->getElementType() == (*WriteAccess)->getElementType() && "Accesses use the same data type" ) ? void (0) : __assert_fail ("(*ReadAccess)->getElementType() == (*WriteAccess)->getElementType() && \"Accesses use the same data type\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 960, __extension__ __PRETTY_FUNCTION__)); |
| 961 | assert((*ReadAccess)->isArrayKind() && (*WriteAccess)->isArrayKind())(static_cast <bool> ((*ReadAccess)->isArrayKind() && (*WriteAccess)->isArrayKind()) ? void (0) : __assert_fail ("(*ReadAccess)->isArrayKind() && (*WriteAccess)->isArrayKind()" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 961, __extension__ __PRETTY_FUNCTION__)); |
| 962 | auto *AccessExpr = |
| 963 | isl_id_to_ast_expr_get(NewAccesses, (*ReadAccess)->getId().release()); |
| 964 | auto *LoadValue = ExprBuilder.create(AccessExpr); |
| 965 | AccessExpr = |
| 966 | isl_id_to_ast_expr_get(NewAccesses, (*WriteAccess)->getId().release()); |
| 967 | auto *StoreAddr = ExprBuilder.createAccessAddress(AccessExpr).first; |
| 968 | Builder.CreateStore(LoadValue, StoreAddr); |
| 969 | } |
| 970 | |
| 971 | Value *IslNodeBuilder::materializeNonScopLoopInductionVariable(const Loop *L) { |
| 972 | assert(OutsideLoopIterations.find(L) == OutsideLoopIterations.end() &&(static_cast <bool> (OutsideLoopIterations.find(L) == OutsideLoopIterations .end() && "trying to materialize loop induction variable twice" ) ? void (0) : __assert_fail ("OutsideLoopIterations.find(L) == OutsideLoopIterations.end() && \"trying to materialize loop induction variable twice\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 973, __extension__ __PRETTY_FUNCTION__)) |
| 973 | "trying to materialize loop induction variable twice")(static_cast <bool> (OutsideLoopIterations.find(L) == OutsideLoopIterations .end() && "trying to materialize loop induction variable twice" ) ? void (0) : __assert_fail ("OutsideLoopIterations.find(L) == OutsideLoopIterations.end() && \"trying to materialize loop induction variable twice\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 973, __extension__ __PRETTY_FUNCTION__)); |
| 974 | const SCEV *OuterLIV = SE.getAddRecExpr(SE.getUnknown(Builder.getInt64(0)), |
| 975 | SE.getUnknown(Builder.getInt64(1)), L, |
| 976 | SCEV::FlagAnyWrap); |
| 977 | Value *V = generateSCEV(OuterLIV); |
| 978 | OutsideLoopIterations[L] = SE.getUnknown(V); |
| 979 | return V; |
| 980 | } |
| 981 | |
| 982 | void IslNodeBuilder::createUser(__isl_take isl_ast_node *User) { |
| 983 | LoopToScevMapT LTS; |
| 984 | isl_id *Id; |
| 985 | ScopStmt *Stmt; |
| 986 | |
| 987 | isl_ast_expr *Expr = isl_ast_node_user_get_expr(User); |
| 988 | isl_ast_expr *StmtExpr = isl_ast_expr_get_op_arg(Expr, 0); |
| 989 | Id = isl_ast_expr_get_id(StmtExpr); |
| 990 | isl_ast_expr_free(StmtExpr); |
| 991 | |
| 992 | LTS.insert(OutsideLoopIterations.begin(), OutsideLoopIterations.end()); |
| 993 | |
| 994 | Stmt = (ScopStmt *)isl_id_get_user(Id); |
| 995 | auto *NewAccesses = createNewAccesses(Stmt, User); |
| 996 | if (Stmt->isCopyStmt()) { |
| 997 | generateCopyStmt(Stmt, NewAccesses); |
| 998 | isl_ast_expr_free(Expr); |
| 999 | } else { |
| 1000 | createSubstitutions(Expr, Stmt, LTS); |
| 1001 | |
| 1002 | if (Stmt->isBlockStmt()) |
| 1003 | BlockGen.copyStmt(*Stmt, LTS, NewAccesses); |
| 1004 | else |
| 1005 | RegionGen.copyStmt(*Stmt, LTS, NewAccesses); |
| 1006 | } |
| 1007 | |
| 1008 | isl_id_to_ast_expr_free(NewAccesses); |
| 1009 | isl_ast_node_free(User); |
| 1010 | isl_id_free(Id); |
| 1011 | } |
| 1012 | |
| 1013 | void IslNodeBuilder::createBlock(__isl_take isl_ast_node *Block) { |
| 1014 | isl_ast_node_list *List = isl_ast_node_block_get_children(Block); |
| 1015 | |
| 1016 | for (int i = 0; i < isl_ast_node_list_n_ast_node(List); ++i) |
| 1017 | create(isl_ast_node_list_get_ast_node(List, i)); |
| 1018 | |
| 1019 | isl_ast_node_free(Block); |
| 1020 | isl_ast_node_list_free(List); |
| 1021 | } |
| 1022 | |
| 1023 | void IslNodeBuilder::create(__isl_take isl_ast_node *Node) { |
| 1024 | switch (isl_ast_node_get_type(Node)) { |
| 1025 | case isl_ast_node_error: |
| 1026 | llvm_unreachable("code generation error")::llvm::llvm_unreachable_internal("code generation error", "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1026); |
| 1027 | case isl_ast_node_mark: |
| 1028 | createMark(Node); |
| 1029 | return; |
| 1030 | case isl_ast_node_for: |
| 1031 | createFor(Node); |
| 1032 | return; |
| 1033 | case isl_ast_node_if: |
| 1034 | createIf(Node); |
| 1035 | return; |
| 1036 | case isl_ast_node_user: |
| 1037 | createUser(Node); |
| 1038 | return; |
| 1039 | case isl_ast_node_block: |
| 1040 | createBlock(Node); |
| 1041 | return; |
| 1042 | } |
| 1043 | |
| 1044 | llvm_unreachable("Unknown isl_ast_node type")::llvm::llvm_unreachable_internal("Unknown isl_ast_node type" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1044); |
| 1045 | } |
| 1046 | |
| 1047 | bool IslNodeBuilder::materializeValue(isl_id *Id) { |
| 1048 | // If the Id is already mapped, skip it. |
| 1049 | if (!IDToValue.count(Id)) { |
| 1050 | auto *ParamSCEV = (const SCEV *)isl_id_get_user(Id); |
| 1051 | Value *V = nullptr; |
| 1052 | |
| 1053 | // Parameters could refer to invariant loads that need to be |
| 1054 | // preloaded before we can generate code for the parameter. Thus, |
| 1055 | // check if any value referred to in ParamSCEV is an invariant load |
| 1056 | // and if so make sure its equivalence class is preloaded. |
| 1057 | SetVector<Value *> Values; |
| 1058 | findValues(ParamSCEV, SE, Values); |
| 1059 | for (auto *Val : Values) { |
| 1060 | // Check if the value is an instruction in a dead block within the SCoP |
| 1061 | // and if so do not code generate it. |
| 1062 | if (auto *Inst = dyn_cast<Instruction>(Val)) { |
| 1063 | if (S.contains(Inst)) { |
| 1064 | bool IsDead = true; |
| 1065 | |
| 1066 | // Check for "undef" loads first, then if there is a statement for |
| 1067 | // the parent of Inst and lastly if the parent of Inst has an empty |
| 1068 | // domain. In the first and last case the instruction is dead but if |
| 1069 | // there is a statement or the domain is not empty Inst is not dead. |
| 1070 | auto MemInst = MemAccInst::dyn_cast(Inst); |
| 1071 | auto Address = MemInst ? MemInst.getPointerOperand() : nullptr; |
| 1072 | if (Address && SE.getUnknown(UndefValue::get(Address->getType())) == |
| 1073 | SE.getPointerBase(SE.getSCEV(Address))) { |
| 1074 | } else if (S.getStmtFor(Inst)) { |
| 1075 | IsDead = false; |
| 1076 | } else { |
| 1077 | auto *Domain = S.getDomainConditions(Inst->getParent()).release(); |
| 1078 | IsDead = isl_set_is_empty(Domain); |
| 1079 | isl_set_free(Domain); |
| 1080 | } |
| 1081 | |
| 1082 | if (IsDead) { |
| 1083 | V = UndefValue::get(ParamSCEV->getType()); |
| 1084 | break; |
| 1085 | } |
| 1086 | } |
| 1087 | } |
| 1088 | |
| 1089 | if (auto *IAClass = S.lookupInvariantEquivClass(Val)) { |
| 1090 | // Check if this invariant access class is empty, hence if we never |
| 1091 | // actually added a loads instruction to it. In that case it has no |
| 1092 | // (meaningful) users and we should not try to code generate it. |
| 1093 | if (IAClass->InvariantAccesses.empty()) |
| 1094 | V = UndefValue::get(ParamSCEV->getType()); |
| 1095 | |
| 1096 | if (!preloadInvariantEquivClass(*IAClass)) { |
| 1097 | isl_id_free(Id); |
| 1098 | return false; |
| 1099 | } |
| 1100 | } |
| 1101 | } |
| 1102 | |
| 1103 | V = V ? V : generateSCEV(ParamSCEV); |
| 1104 | IDToValue[Id] = V; |
| 1105 | } |
| 1106 | |
| 1107 | isl_id_free(Id); |
| 1108 | return true; |
| 1109 | } |
| 1110 | |
| 1111 | bool IslNodeBuilder::materializeParameters(isl_set *Set) { |
| 1112 | for (unsigned i = 0, e = isl_set_dim(Set, isl_dim_param); i < e; ++i) { |
| 1113 | if (!isl_set_involves_dims(Set, isl_dim_param, i, 1)) |
| 1114 | continue; |
| 1115 | isl_id *Id = isl_set_get_dim_id(Set, isl_dim_param, i); |
| 1116 | if (!materializeValue(Id)) |
| 1117 | return false; |
| 1118 | } |
| 1119 | return true; |
| 1120 | } |
| 1121 | |
| 1122 | bool IslNodeBuilder::materializeParameters() { |
| 1123 | for (const SCEV *Param : S.parameters()) { |
| 1124 | isl_id *Id = S.getIdForParam(Param).release(); |
| 1125 | if (!materializeValue(Id)) |
| 1126 | return false; |
| 1127 | } |
| 1128 | return true; |
| 1129 | } |
| 1130 | |
| 1131 | /// Generate the computation of the size of the outermost dimension from the |
| 1132 | /// Fortran array descriptor (in this case, `@g_arr`). The final `%size` |
| 1133 | /// contains the size of the array. |
| 1134 | /// |
| 1135 | /// %arrty = type { i8*, i64, i64, [3 x %desc.dimensionty] } |
| 1136 | /// %desc.dimensionty = type { i64, i64, i64 } |
| 1137 | /// @g_arr = global %arrty zeroinitializer, align 32 |
| 1138 | /// ... |
| 1139 | /// %0 = load i64, i64* getelementptr inbounds |
| 1140 | /// (%arrty, %arrty* @g_arr, i64 0, i32 3, i64 0, i32 2) |
| 1141 | /// %1 = load i64, i64* getelementptr inbounds |
| 1142 | /// (%arrty, %arrty* @g_arr, i64 0, i32 3, i64 0, i32 1) |
| 1143 | /// %2 = sub nsw i64 %0, %1 |
| 1144 | /// %size = add nsw i64 %2, 1 |
| 1145 | static Value *buildFADOutermostDimensionLoad(Value *GlobalDescriptor, |
| 1146 | PollyIRBuilder &Builder, |
| 1147 | std::string ArrayName) { |
| 1148 | assert(GlobalDescriptor && "invalid global descriptor given")(static_cast <bool> (GlobalDescriptor && "invalid global descriptor given" ) ? void (0) : __assert_fail ("GlobalDescriptor && \"invalid global descriptor given\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1148, __extension__ __PRETTY_FUNCTION__)); |
| 1149 | |
| 1150 | Value *endIdx[4] = {Builder.getInt64(0), Builder.getInt32(3), |
| 1151 | Builder.getInt64(0), Builder.getInt32(2)}; |
| 1152 | Value *endPtr = Builder.CreateInBoundsGEP(GlobalDescriptor, endIdx, |
| 1153 | ArrayName + "_end_ptr"); |
| 1154 | Type *type = cast<GEPOperator>(endPtr)->getResultElementType(); |
| 1155 | assert(isa<IntegerType>(type) && "expected type of end to be integral")(static_cast <bool> (isa<IntegerType>(type) && "expected type of end to be integral") ? void (0) : __assert_fail ("isa<IntegerType>(type) && \"expected type of end to be integral\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1155, __extension__ __PRETTY_FUNCTION__)); |
| 1156 | |
| 1157 | Value *end = Builder.CreateLoad(type, endPtr, ArrayName + "_end"); |
| 1158 | |
| 1159 | Value *beginIdx[4] = {Builder.getInt64(0), Builder.getInt32(3), |
| 1160 | Builder.getInt64(0), Builder.getInt32(1)}; |
| 1161 | Value *beginPtr = Builder.CreateInBoundsGEP(GlobalDescriptor, beginIdx, |
| 1162 | ArrayName + "_begin_ptr"); |
| 1163 | Value *begin = Builder.CreateLoad(type, beginPtr, ArrayName + "_begin"); |
| 1164 | |
| 1165 | Value *size = |
Value stored to 'size' during its initialization is never read | |
| 1166 | Builder.CreateNSWSub(end, begin, ArrayName + "_end_begin_delta"); |
| 1167 | |
| 1168 | size = Builder.CreateNSWAdd( |
| 1169 | end, ConstantInt::get(type, 1, /* signed = */ true), ArrayName + "_size"); |
| 1170 | |
| 1171 | return size; |
| 1172 | } |
| 1173 | |
| 1174 | bool IslNodeBuilder::materializeFortranArrayOutermostDimension() { |
| 1175 | for (ScopArrayInfo *Array : S.arrays()) { |
| 1176 | if (Array->getNumberOfDimensions() == 0) |
| 1177 | continue; |
| 1178 | |
| 1179 | Value *FAD = Array->getFortranArrayDescriptor(); |
| 1180 | if (!FAD) |
| 1181 | continue; |
| 1182 | |
| 1183 | isl_pw_aff *ParametricPwAff = Array->getDimensionSizePw(0).release(); |
| 1184 | assert(ParametricPwAff && "parametric pw_aff corresponding "(static_cast <bool> (ParametricPwAff && "parametric pw_aff corresponding " "to outermost dimension does not " "exist") ? void (0) : __assert_fail ("ParametricPwAff && \"parametric pw_aff corresponding \" \"to outermost dimension does not \" \"exist\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1186, __extension__ __PRETTY_FUNCTION__)) |
| 1185 | "to outermost dimension does not "(static_cast <bool> (ParametricPwAff && "parametric pw_aff corresponding " "to outermost dimension does not " "exist") ? void (0) : __assert_fail ("ParametricPwAff && \"parametric pw_aff corresponding \" \"to outermost dimension does not \" \"exist\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1186, __extension__ __PRETTY_FUNCTION__)) |
| 1186 | "exist")(static_cast <bool> (ParametricPwAff && "parametric pw_aff corresponding " "to outermost dimension does not " "exist") ? void (0) : __assert_fail ("ParametricPwAff && \"parametric pw_aff corresponding \" \"to outermost dimension does not \" \"exist\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1186, __extension__ __PRETTY_FUNCTION__)); |
| 1187 | |
| 1188 | isl_id *Id = isl_pw_aff_get_dim_id(ParametricPwAff, isl_dim_param, 0); |
| 1189 | isl_pw_aff_free(ParametricPwAff); |
| 1190 | |
| 1191 | assert(Id && "pw_aff is not parametric")(static_cast <bool> (Id && "pw_aff is not parametric" ) ? void (0) : __assert_fail ("Id && \"pw_aff is not parametric\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1191, __extension__ __PRETTY_FUNCTION__)); |
| 1192 | |
| 1193 | if (IDToValue.count(Id)) { |
| 1194 | isl_id_free(Id); |
| 1195 | continue; |
| 1196 | } |
| 1197 | |
| 1198 | Value *FinalValue = |
| 1199 | buildFADOutermostDimensionLoad(FAD, Builder, Array->getName()); |
| 1200 | assert(FinalValue && "unable to build Fortran array "(static_cast <bool> (FinalValue && "unable to build Fortran array " "descriptor load of outermost dimension") ? void (0) : __assert_fail ("FinalValue && \"unable to build Fortran array \" \"descriptor load of outermost dimension\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1201, __extension__ __PRETTY_FUNCTION__)) |
| 1201 | "descriptor load of outermost dimension")(static_cast <bool> (FinalValue && "unable to build Fortran array " "descriptor load of outermost dimension") ? void (0) : __assert_fail ("FinalValue && \"unable to build Fortran array \" \"descriptor load of outermost dimension\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1201, __extension__ __PRETTY_FUNCTION__)); |
| 1202 | IDToValue[Id] = FinalValue; |
| 1203 | isl_id_free(Id); |
| 1204 | } |
| 1205 | return true; |
| 1206 | } |
| 1207 | |
| 1208 | Value *IslNodeBuilder::preloadUnconditionally(isl_set *AccessRange, |
| 1209 | isl_ast_build *Build, |
| 1210 | Instruction *AccInst) { |
| 1211 | isl_pw_multi_aff *PWAccRel = isl_pw_multi_aff_from_set(AccessRange); |
| 1212 | isl_ast_expr *Access = |
| 1213 | isl_ast_build_access_from_pw_multi_aff(Build, PWAccRel); |
| 1214 | auto *Address = isl_ast_expr_address_of(Access); |
| 1215 | auto *AddressValue = ExprBuilder.create(Address); |
| 1216 | Value *PreloadVal; |
| 1217 | |
| 1218 | // Correct the type as the SAI might have a different type than the user |
| 1219 | // expects, especially if the base pointer is a struct. |
| 1220 | Type *Ty = AccInst->getType(); |
| 1221 | |
| 1222 | auto *Ptr = AddressValue; |
| 1223 | auto Name = Ptr->getName(); |
| 1224 | auto AS = Ptr->getType()->getPointerAddressSpace(); |
| 1225 | Ptr = Builder.CreatePointerCast(Ptr, Ty->getPointerTo(AS), Name + ".cast"); |
| 1226 | PreloadVal = Builder.CreateLoad(Ty, Ptr, Name + ".load"); |
| 1227 | if (LoadInst *PreloadInst = dyn_cast<LoadInst>(PreloadVal)) |
| 1228 | PreloadInst->setAlignment(cast<LoadInst>(AccInst)->getAlign()); |
| 1229 | |
| 1230 | // TODO: This is only a hot fix for SCoP sequences that use the same load |
| 1231 | // instruction contained and hoisted by one of the SCoPs. |
| 1232 | if (SE.isSCEVable(Ty)) |
| 1233 | SE.forgetValue(AccInst); |
| 1234 | |
| 1235 | return PreloadVal; |
| 1236 | } |
| 1237 | |
| 1238 | Value *IslNodeBuilder::preloadInvariantLoad(const MemoryAccess &MA, |
| 1239 | isl_set *Domain) { |
| 1240 | isl_set *AccessRange = isl_map_range(MA.getAddressFunction().release()); |
| 1241 | AccessRange = isl_set_gist_params(AccessRange, S.getContext().release()); |
| 1242 | |
| 1243 | if (!materializeParameters(AccessRange)) { |
| 1244 | isl_set_free(AccessRange); |
| 1245 | isl_set_free(Domain); |
| 1246 | return nullptr; |
| 1247 | } |
| 1248 | |
| 1249 | auto *Build = |
| 1250 | isl_ast_build_from_context(isl_set_universe(S.getParamSpace().release())); |
| 1251 | isl_set *Universe = isl_set_universe(isl_set_get_space(Domain)); |
| 1252 | bool AlwaysExecuted = isl_set_is_equal(Domain, Universe); |
| 1253 | isl_set_free(Universe); |
| 1254 | |
| 1255 | Instruction *AccInst = MA.getAccessInstruction(); |
| 1256 | Type *AccInstTy = AccInst->getType(); |
| 1257 | |
| 1258 | Value *PreloadVal = nullptr; |
| 1259 | if (AlwaysExecuted) { |
| 1260 | PreloadVal = preloadUnconditionally(AccessRange, Build, AccInst); |
| 1261 | isl_ast_build_free(Build); |
| 1262 | isl_set_free(Domain); |
| 1263 | return PreloadVal; |
| 1264 | } |
| 1265 | |
| 1266 | if (!materializeParameters(Domain)) { |
| 1267 | isl_ast_build_free(Build); |
| 1268 | isl_set_free(AccessRange); |
| 1269 | isl_set_free(Domain); |
| 1270 | return nullptr; |
| 1271 | } |
| 1272 | |
| 1273 | isl_ast_expr *DomainCond = isl_ast_build_expr_from_set(Build, Domain); |
| 1274 | Domain = nullptr; |
| 1275 | |
| 1276 | ExprBuilder.setTrackOverflow(true); |
| 1277 | Value *Cond = ExprBuilder.create(DomainCond); |
| 1278 | Value *OverflowHappened = Builder.CreateNot(ExprBuilder.getOverflowState(), |
| 1279 | "polly.preload.cond.overflown"); |
| 1280 | Cond = Builder.CreateAnd(Cond, OverflowHappened, "polly.preload.cond.result"); |
| 1281 | ExprBuilder.setTrackOverflow(false); |
| 1282 | |
| 1283 | if (!Cond->getType()->isIntegerTy(1)) |
| 1284 | Cond = Builder.CreateIsNotNull(Cond); |
| 1285 | |
| 1286 | BasicBlock *CondBB = SplitBlock(Builder.GetInsertBlock(), |
| 1287 | &*Builder.GetInsertPoint(), &DT, &LI); |
| 1288 | CondBB->setName("polly.preload.cond"); |
| 1289 | |
| 1290 | BasicBlock *MergeBB = SplitBlock(CondBB, &CondBB->front(), &DT, &LI); |
| 1291 | MergeBB->setName("polly.preload.merge"); |
| 1292 | |
| 1293 | Function *F = Builder.GetInsertBlock()->getParent(); |
| 1294 | LLVMContext &Context = F->getContext(); |
| 1295 | BasicBlock *ExecBB = BasicBlock::Create(Context, "polly.preload.exec", F); |
| 1296 | |
| 1297 | DT.addNewBlock(ExecBB, CondBB); |
| 1298 | if (Loop *L = LI.getLoopFor(CondBB)) |
| 1299 | L->addBasicBlockToLoop(ExecBB, LI); |
| 1300 | |
| 1301 | auto *CondBBTerminator = CondBB->getTerminator(); |
| 1302 | Builder.SetInsertPoint(CondBBTerminator); |
| 1303 | Builder.CreateCondBr(Cond, ExecBB, MergeBB); |
| 1304 | CondBBTerminator->eraseFromParent(); |
| 1305 | |
| 1306 | Builder.SetInsertPoint(ExecBB); |
| 1307 | Builder.CreateBr(MergeBB); |
| 1308 | |
| 1309 | Builder.SetInsertPoint(ExecBB->getTerminator()); |
| 1310 | Value *PreAccInst = preloadUnconditionally(AccessRange, Build, AccInst); |
| 1311 | Builder.SetInsertPoint(MergeBB->getTerminator()); |
| 1312 | auto *MergePHI = Builder.CreatePHI( |
| 1313 | AccInstTy, 2, "polly.preload." + AccInst->getName() + ".merge"); |
| 1314 | PreloadVal = MergePHI; |
| 1315 | |
| 1316 | if (!PreAccInst) { |
| 1317 | PreloadVal = nullptr; |
| 1318 | PreAccInst = UndefValue::get(AccInstTy); |
| 1319 | } |
| 1320 | |
| 1321 | MergePHI->addIncoming(PreAccInst, ExecBB); |
| 1322 | MergePHI->addIncoming(Constant::getNullValue(AccInstTy), CondBB); |
| 1323 | |
| 1324 | isl_ast_build_free(Build); |
| 1325 | return PreloadVal; |
| 1326 | } |
| 1327 | |
| 1328 | bool IslNodeBuilder::preloadInvariantEquivClass( |
| 1329 | InvariantEquivClassTy &IAClass) { |
| 1330 | // For an equivalence class of invariant loads we pre-load the representing |
| 1331 | // element with the unified execution context. However, we have to map all |
| 1332 | // elements of the class to the one preloaded load as they are referenced |
| 1333 | // during the code generation and therefor need to be mapped. |
| 1334 | const MemoryAccessList &MAs = IAClass.InvariantAccesses; |
| 1335 | if (MAs.empty()) |
| 1336 | return true; |
| 1337 | |
| 1338 | MemoryAccess *MA = MAs.front(); |
| 1339 | assert(MA->isArrayKind() && MA->isRead())(static_cast <bool> (MA->isArrayKind() && MA ->isRead()) ? void (0) : __assert_fail ("MA->isArrayKind() && MA->isRead()" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1339, __extension__ __PRETTY_FUNCTION__)); |
| 1340 | |
| 1341 | // If the access function was already mapped, the preload of this equivalence |
| 1342 | // class was triggered earlier already and doesn't need to be done again. |
| 1343 | if (ValueMap.count(MA->getAccessInstruction())) |
| 1344 | return true; |
| 1345 | |
| 1346 | // Check for recursion which can be caused by additional constraints, e.g., |
| 1347 | // non-finite loop constraints. In such a case we have to bail out and insert |
| 1348 | // a "false" runtime check that will cause the original code to be executed. |
| 1349 | auto PtrId = std::make_pair(IAClass.IdentifyingPointer, IAClass.AccessType); |
| 1350 | if (!PreloadedPtrs.insert(PtrId).second) |
| 1351 | return false; |
| 1352 | |
| 1353 | // The execution context of the IAClass. |
| 1354 | isl::set &ExecutionCtx = IAClass.ExecutionContext; |
| 1355 | |
| 1356 | // If the base pointer of this class is dependent on another one we have to |
| 1357 | // make sure it was preloaded already. |
| 1358 | auto *SAI = MA->getScopArrayInfo(); |
| 1359 | if (auto *BaseIAClass = S.lookupInvariantEquivClass(SAI->getBasePtr())) { |
| 1360 | if (!preloadInvariantEquivClass(*BaseIAClass)) |
| 1361 | return false; |
| 1362 | |
| 1363 | // After we preloaded the BaseIAClass we adjusted the BaseExecutionCtx and |
| 1364 | // we need to refine the ExecutionCtx. |
| 1365 | isl::set BaseExecutionCtx = BaseIAClass->ExecutionContext; |
| 1366 | ExecutionCtx = ExecutionCtx.intersect(BaseExecutionCtx); |
| 1367 | } |
| 1368 | |
| 1369 | // If the size of a dimension is dependent on another class, make sure it is |
| 1370 | // preloaded. |
| 1371 | for (unsigned i = 1, e = SAI->getNumberOfDimensions(); i < e; ++i) { |
| 1372 | const SCEV *Dim = SAI->getDimensionSize(i); |
| 1373 | SetVector<Value *> Values; |
| 1374 | findValues(Dim, SE, Values); |
| 1375 | for (auto *Val : Values) { |
| 1376 | if (auto *BaseIAClass = S.lookupInvariantEquivClass(Val)) { |
| 1377 | if (!preloadInvariantEquivClass(*BaseIAClass)) |
| 1378 | return false; |
| 1379 | |
| 1380 | // After we preloaded the BaseIAClass we adjusted the BaseExecutionCtx |
| 1381 | // and we need to refine the ExecutionCtx. |
| 1382 | isl::set BaseExecutionCtx = BaseIAClass->ExecutionContext; |
| 1383 | ExecutionCtx = ExecutionCtx.intersect(BaseExecutionCtx); |
| 1384 | } |
| 1385 | } |
| 1386 | } |
| 1387 | |
| 1388 | Instruction *AccInst = MA->getAccessInstruction(); |
| 1389 | Type *AccInstTy = AccInst->getType(); |
| 1390 | |
| 1391 | Value *PreloadVal = preloadInvariantLoad(*MA, ExecutionCtx.copy()); |
| 1392 | if (!PreloadVal) |
| 1393 | return false; |
| 1394 | |
| 1395 | for (const MemoryAccess *MA : MAs) { |
| 1396 | Instruction *MAAccInst = MA->getAccessInstruction(); |
| 1397 | assert(PreloadVal->getType() == MAAccInst->getType())(static_cast <bool> (PreloadVal->getType() == MAAccInst ->getType()) ? void (0) : __assert_fail ("PreloadVal->getType() == MAAccInst->getType()" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1397, __extension__ __PRETTY_FUNCTION__)); |
| 1398 | ValueMap[MAAccInst] = PreloadVal; |
| 1399 | } |
| 1400 | |
| 1401 | if (SE.isSCEVable(AccInstTy)) { |
| 1402 | isl_id *ParamId = S.getIdForParam(SE.getSCEV(AccInst)).release(); |
| 1403 | if (ParamId) |
| 1404 | IDToValue[ParamId] = PreloadVal; |
| 1405 | isl_id_free(ParamId); |
| 1406 | } |
| 1407 | |
| 1408 | BasicBlock *EntryBB = &Builder.GetInsertBlock()->getParent()->getEntryBlock(); |
| 1409 | auto *Alloca = new AllocaInst(AccInstTy, DL.getAllocaAddrSpace(), |
| 1410 | AccInst->getName() + ".preload.s2a", |
| 1411 | &*EntryBB->getFirstInsertionPt()); |
| 1412 | Builder.CreateStore(PreloadVal, Alloca); |
| 1413 | ValueMapT PreloadedPointer; |
| 1414 | PreloadedPointer[PreloadVal] = AccInst; |
| 1415 | Annotator.addAlternativeAliasBases(PreloadedPointer); |
| 1416 | |
| 1417 | for (auto *DerivedSAI : SAI->getDerivedSAIs()) { |
| 1418 | Value *BasePtr = DerivedSAI->getBasePtr(); |
| 1419 | |
| 1420 | for (const MemoryAccess *MA : MAs) { |
| 1421 | // As the derived SAI information is quite coarse, any load from the |
| 1422 | // current SAI could be the base pointer of the derived SAI, however we |
| 1423 | // should only change the base pointer of the derived SAI if we actually |
| 1424 | // preloaded it. |
| 1425 | if (BasePtr == MA->getOriginalBaseAddr()) { |
| 1426 | assert(BasePtr->getType() == PreloadVal->getType())(static_cast <bool> (BasePtr->getType() == PreloadVal ->getType()) ? void (0) : __assert_fail ("BasePtr->getType() == PreloadVal->getType()" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1426, __extension__ __PRETTY_FUNCTION__)); |
| 1427 | DerivedSAI->setBasePtr(PreloadVal); |
| 1428 | } |
| 1429 | |
| 1430 | // For scalar derived SAIs we remap the alloca used for the derived value. |
| 1431 | if (BasePtr == MA->getAccessInstruction()) |
| 1432 | ScalarMap[DerivedSAI] = Alloca; |
| 1433 | } |
| 1434 | } |
| 1435 | |
| 1436 | for (const MemoryAccess *MA : MAs) { |
| 1437 | Instruction *MAAccInst = MA->getAccessInstruction(); |
| 1438 | // Use the escape system to get the correct value to users outside the SCoP. |
| 1439 | BlockGenerator::EscapeUserVectorTy EscapeUsers; |
| 1440 | for (auto *U : MAAccInst->users()) |
| 1441 | if (Instruction *UI = dyn_cast<Instruction>(U)) |
| 1442 | if (!S.contains(UI)) |
| 1443 | EscapeUsers.push_back(UI); |
| 1444 | |
| 1445 | if (EscapeUsers.empty()) |
| 1446 | continue; |
| 1447 | |
| 1448 | EscapeMap[MA->getAccessInstruction()] = |
| 1449 | std::make_pair(Alloca, std::move(EscapeUsers)); |
| 1450 | } |
| 1451 | |
| 1452 | return true; |
| 1453 | } |
| 1454 | |
| 1455 | void IslNodeBuilder::allocateNewArrays(BBPair StartExitBlocks) { |
| 1456 | for (auto &SAI : S.arrays()) { |
| 1457 | if (SAI->getBasePtr()) |
| 1458 | continue; |
| 1459 | |
| 1460 | assert(SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) &&(static_cast <bool> (SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && "The size of the outermost dimension is used to declare newly " "created arrays that require memory allocation.") ? void (0) : __assert_fail ("SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && \"The size of the outermost dimension is used to declare newly \" \"created arrays that require memory allocation.\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1462, __extension__ __PRETTY_FUNCTION__)) |
| 1461 | "The size of the outermost dimension is used to declare newly "(static_cast <bool> (SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && "The size of the outermost dimension is used to declare newly " "created arrays that require memory allocation.") ? void (0) : __assert_fail ("SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && \"The size of the outermost dimension is used to declare newly \" \"created arrays that require memory allocation.\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1462, __extension__ __PRETTY_FUNCTION__)) |
| 1462 | "created arrays that require memory allocation.")(static_cast <bool> (SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && "The size of the outermost dimension is used to declare newly " "created arrays that require memory allocation.") ? void (0) : __assert_fail ("SAI->getNumberOfDimensions() > 0 && SAI->getDimensionSize(0) && \"The size of the outermost dimension is used to declare newly \" \"created arrays that require memory allocation.\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1462, __extension__ __PRETTY_FUNCTION__)); |
| 1463 | |
| 1464 | Type *NewArrayType = nullptr; |
| 1465 | |
| 1466 | // Get the size of the array = size(dim_1)*...*size(dim_n) |
| 1467 | uint64_t ArraySizeInt = 1; |
| 1468 | for (int i = SAI->getNumberOfDimensions() - 1; i >= 0; i--) { |
| 1469 | auto *DimSize = SAI->getDimensionSize(i); |
| 1470 | unsigned UnsignedDimSize = static_cast<const SCEVConstant *>(DimSize) |
| 1471 | ->getAPInt() |
| 1472 | .getLimitedValue(); |
| 1473 | |
| 1474 | if (!NewArrayType) |
| 1475 | NewArrayType = SAI->getElementType(); |
| 1476 | |
| 1477 | NewArrayType = ArrayType::get(NewArrayType, UnsignedDimSize); |
| 1478 | ArraySizeInt *= UnsignedDimSize; |
| 1479 | } |
| 1480 | |
| 1481 | if (SAI->isOnHeap()) { |
| 1482 | LLVMContext &Ctx = NewArrayType->getContext(); |
| 1483 | |
| 1484 | // Get the IntPtrTy from the Datalayout |
| 1485 | auto IntPtrTy = DL.getIntPtrType(Ctx); |
| 1486 | |
| 1487 | // Get the size of the element type in bits |
| 1488 | unsigned Size = SAI->getElemSizeInBytes(); |
| 1489 | |
| 1490 | // Insert the malloc call at polly.start |
| 1491 | auto InstIt = std::get<0>(StartExitBlocks)->getTerminator(); |
| 1492 | auto *CreatedArray = CallInst::CreateMalloc( |
| 1493 | &*InstIt, IntPtrTy, SAI->getElementType(), |
| 1494 | ConstantInt::get(Type::getInt64Ty(Ctx), Size), |
| 1495 | ConstantInt::get(Type::getInt64Ty(Ctx), ArraySizeInt), nullptr, |
| 1496 | SAI->getName()); |
| 1497 | |
| 1498 | SAI->setBasePtr(CreatedArray); |
| 1499 | |
| 1500 | // Insert the free call at polly.exiting |
| 1501 | CallInst::CreateFree(CreatedArray, |
| 1502 | std::get<1>(StartExitBlocks)->getTerminator()); |
| 1503 | } else { |
| 1504 | auto InstIt = Builder.GetInsertBlock() |
| 1505 | ->getParent() |
| 1506 | ->getEntryBlock() |
| 1507 | .getTerminator(); |
| 1508 | |
| 1509 | auto *CreatedArray = new AllocaInst(NewArrayType, DL.getAllocaAddrSpace(), |
| 1510 | SAI->getName(), &*InstIt); |
| 1511 | if (PollyTargetFirstLevelCacheLineSize) |
| 1512 | CreatedArray->setAlignment(Align(PollyTargetFirstLevelCacheLineSize)); |
| 1513 | SAI->setBasePtr(CreatedArray); |
| 1514 | } |
| 1515 | } |
| 1516 | } |
| 1517 | |
| 1518 | bool IslNodeBuilder::preloadInvariantLoads() { |
| 1519 | auto &InvariantEquivClasses = S.getInvariantAccesses(); |
| 1520 | if (InvariantEquivClasses.empty()) |
| 1521 | return true; |
| 1522 | |
| 1523 | BasicBlock *PreLoadBB = SplitBlock(Builder.GetInsertBlock(), |
| 1524 | &*Builder.GetInsertPoint(), &DT, &LI); |
| 1525 | PreLoadBB->setName("polly.preload.begin"); |
| 1526 | Builder.SetInsertPoint(&PreLoadBB->front()); |
| 1527 | |
| 1528 | for (auto &IAClass : InvariantEquivClasses) |
| 1529 | if (!preloadInvariantEquivClass(IAClass)) |
| 1530 | return false; |
| 1531 | |
| 1532 | return true; |
| 1533 | } |
| 1534 | |
| 1535 | void IslNodeBuilder::addParameters(__isl_take isl_set *Context) { |
| 1536 | // Materialize values for the parameters of the SCoP. |
| 1537 | materializeParameters(); |
| 1538 | |
| 1539 | // materialize the outermost dimension parameters for a Fortran array. |
| 1540 | // NOTE: materializeParameters() does not work since it looks through |
| 1541 | // the SCEVs. We don't have a corresponding SCEV for the array size |
| 1542 | // parameter |
| 1543 | materializeFortranArrayOutermostDimension(); |
| 1544 | |
| 1545 | // Generate values for the current loop iteration for all surrounding loops. |
| 1546 | // |
| 1547 | // We may also reference loops outside of the scop which do not contain the |
| 1548 | // scop itself, but as the number of such scops may be arbitrarily large we do |
| 1549 | // not generate code for them here, but only at the point of code generation |
| 1550 | // where these values are needed. |
| 1551 | Loop *L = LI.getLoopFor(S.getEntry()); |
| 1552 | |
| 1553 | while (L != nullptr && S.contains(L)) |
| 1554 | L = L->getParentLoop(); |
| 1555 | |
| 1556 | while (L != nullptr) { |
| 1557 | materializeNonScopLoopInductionVariable(L); |
| 1558 | L = L->getParentLoop(); |
| 1559 | } |
| 1560 | |
| 1561 | isl_set_free(Context); |
| 1562 | } |
| 1563 | |
| 1564 | Value *IslNodeBuilder::generateSCEV(const SCEV *Expr) { |
| 1565 | /// We pass the insert location of our Builder, as Polly ensures during IR |
| 1566 | /// generation that there is always a valid CFG into which instructions are |
| 1567 | /// inserted. As a result, the insertpoint is known to be always followed by a |
| 1568 | /// terminator instruction. This means the insert point may be specified by a |
| 1569 | /// terminator instruction, but it can never point to an ->end() iterator |
| 1570 | /// which does not have a corresponding instruction. Hence, dereferencing |
| 1571 | /// the insertpoint to obtain an instruction is known to be save. |
| 1572 | /// |
| 1573 | /// We also do not need to update the Builder here, as new instructions are |
| 1574 | /// always inserted _before_ the given InsertLocation. As a result, the |
| 1575 | /// insert location remains valid. |
| 1576 | assert(Builder.GetInsertBlock()->end() != Builder.GetInsertPoint() &&(static_cast <bool> (Builder.GetInsertBlock()->end() != Builder.GetInsertPoint() && "Insert location points after last valid instruction" ) ? void (0) : __assert_fail ("Builder.GetInsertBlock()->end() != Builder.GetInsertPoint() && \"Insert location points after last valid instruction\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1577, __extension__ __PRETTY_FUNCTION__)) |
| 1577 | "Insert location points after last valid instruction")(static_cast <bool> (Builder.GetInsertBlock()->end() != Builder.GetInsertPoint() && "Insert location points after last valid instruction" ) ? void (0) : __assert_fail ("Builder.GetInsertBlock()->end() != Builder.GetInsertPoint() && \"Insert location points after last valid instruction\"" , "/build/llvm-toolchain-snapshot-13~++20210621111111+acefe0eaaf82/polly/lib/CodeGen/IslNodeBuilder.cpp" , 1577, __extension__ __PRETTY_FUNCTION__)); |
| 1578 | Instruction *InsertLocation = &*Builder.GetInsertPoint(); |
| 1579 | return expandCodeFor(S, SE, DL, "polly", Expr, Expr->getType(), |
| 1580 | InsertLocation, &ValueMap, |
| 1581 | StartBlock->getSinglePredecessor()); |
| 1582 | } |
| 1583 | |
| 1584 | /// The AST expression we generate to perform the run-time check assumes |
| 1585 | /// computations on integer types of infinite size. As we only use 64-bit |
| 1586 | /// arithmetic we check for overflows, in case of which we set the result |
| 1587 | /// of this run-time check to false to be conservatively correct, |
| 1588 | Value *IslNodeBuilder::createRTC(isl_ast_expr *Condition) { |
| 1589 | auto ExprBuilder = getExprBuilder(); |
| 1590 | |
| 1591 | // In case the AST expression has integers larger than 64 bit, bail out. The |
| 1592 | // resulting LLVM-IR will contain operations on types that use more than 64 |
| 1593 | // bits. These are -- in case wrapping intrinsics are used -- translated to |
| 1594 | // runtime library calls that are not available on all systems (e.g., Android) |
| 1595 | // and consequently will result in linker errors. |
| 1596 | if (ExprBuilder.hasLargeInts(isl::manage_copy(Condition))) { |
| 1597 | isl_ast_expr_free(Condition); |
| 1598 | return Builder.getFalse(); |
| 1599 | } |
| 1600 | |
| 1601 | ExprBuilder.setTrackOverflow(true); |
| 1602 | Value *RTC = ExprBuilder.create(Condition); |
| 1603 | if (!RTC->getType()->isIntegerTy(1)) |
| 1604 | RTC = Builder.CreateIsNotNull(RTC); |
| 1605 | Value *OverflowHappened = |
| 1606 | Builder.CreateNot(ExprBuilder.getOverflowState(), "polly.rtc.overflown"); |
| 1607 | |
| 1608 | if (PollyGenerateRTCPrint) { |
| 1609 | auto *F = Builder.GetInsertBlock()->getParent(); |
| 1610 | RuntimeDebugBuilder::createCPUPrinter( |
| 1611 | Builder, |
| 1612 | "F: " + F->getName().str() + " R: " + S.getRegion().getNameStr() + |
| 1613 | "RTC: ", |
| 1614 | RTC, " Overflow: ", OverflowHappened, |
| 1615 | "\n" |
| 1616 | " (0 failed, -1 succeeded)\n" |
| 1617 | " (if one or both are 0 falling back to original code, if both are -1 " |
| 1618 | "executing Polly code)\n"); |
| 1619 | } |
| 1620 | |
| 1621 | RTC = Builder.CreateAnd(RTC, OverflowHappened, "polly.rtc.result"); |
| 1622 | ExprBuilder.setTrackOverflow(false); |
| 1623 | |
| 1624 | if (!isa<ConstantInt>(RTC)) |
| 1625 | VersionedScops++; |
| 1626 | |
| 1627 | return RTC; |
| 1628 | } |