1 //===-- AffinePromotion.cpp -----------------------------------------------===// 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 transformation is a prototype that promote FIR loops operations 10 // to affine dialect operations. 11 // It is not part of the production pipeline and would need more work in order 12 // to be used in production. 13 // More information can be found in this presentation: 14 // https://slides.com/rajanwalia/deck 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "PassDetail.h" 19 #include "flang/Optimizer/Dialect/FIRDialect.h" 20 #include "flang/Optimizer/Dialect/FIROps.h" 21 #include "flang/Optimizer/Dialect/FIRType.h" 22 #include "flang/Optimizer/Transforms/Passes.h" 23 #include "mlir/Dialect/Affine/IR/AffineOps.h" 24 #include "mlir/Dialect/Func/IR/FuncOps.h" 25 #include "mlir/Dialect/SCF/SCF.h" 26 #include "mlir/IR/BuiltinAttributes.h" 27 #include "mlir/IR/IntegerSet.h" 28 #include "mlir/IR/Visitors.h" 29 #include "mlir/Transforms/DialectConversion.h" 30 #include "llvm/ADT/DenseMap.h" 31 #include "llvm/ADT/Optional.h" 32 #include "llvm/Support/Debug.h" 33 34 #define DEBUG_TYPE "flang-affine-promotion" 35 36 using namespace fir; 37 38 namespace { 39 struct AffineLoopAnalysis; 40 struct AffineIfAnalysis; 41 42 /// Stores analysis objects for all loops and if operations inside a function 43 /// these analysis are used twice, first for marking operations for rewrite and 44 /// second when doing rewrite. 45 struct AffineFunctionAnalysis { 46 explicit AffineFunctionAnalysis(mlir::FuncOp funcOp) { 47 for (fir::DoLoopOp op : funcOp.getOps<fir::DoLoopOp>()) 48 loopAnalysisMap.try_emplace(op, op, *this); 49 } 50 51 AffineLoopAnalysis getChildLoopAnalysis(fir::DoLoopOp op) const; 52 53 AffineIfAnalysis getChildIfAnalysis(fir::IfOp op) const; 54 55 llvm::DenseMap<mlir::Operation *, AffineLoopAnalysis> loopAnalysisMap; 56 llvm::DenseMap<mlir::Operation *, AffineIfAnalysis> ifAnalysisMap; 57 }; 58 } // namespace 59 60 static bool analyzeCoordinate(mlir::Value coordinate, mlir::Operation *op) { 61 if (auto blockArg = coordinate.dyn_cast<mlir::BlockArgument>()) { 62 if (isa<fir::DoLoopOp>(blockArg.getOwner()->getParentOp())) 63 return true; 64 LLVM_DEBUG(llvm::dbgs() << "AffineLoopAnalysis: array coordinate is not a " 65 "loop induction variable (owner not loopOp)\n"; 66 op->dump()); 67 return false; 68 } 69 LLVM_DEBUG( 70 llvm::dbgs() << "AffineLoopAnalysis: array coordinate is not a loop " 71 "induction variable (not a block argument)\n"; 72 op->dump(); coordinate.getDefiningOp()->dump()); 73 return false; 74 } 75 76 namespace { 77 struct AffineLoopAnalysis { 78 AffineLoopAnalysis() = default; 79 80 explicit AffineLoopAnalysis(fir::DoLoopOp op, AffineFunctionAnalysis &afa) 81 : legality(analyzeLoop(op, afa)) {} 82 83 bool canPromoteToAffine() { return legality; } 84 85 private: 86 bool analyzeBody(fir::DoLoopOp loopOperation, 87 AffineFunctionAnalysis &functionAnalysis) { 88 for (auto loopOp : loopOperation.getOps<fir::DoLoopOp>()) { 89 auto analysis = functionAnalysis.loopAnalysisMap 90 .try_emplace(loopOp, loopOp, functionAnalysis) 91 .first->getSecond(); 92 if (!analysis.canPromoteToAffine()) 93 return false; 94 } 95 for (auto ifOp : loopOperation.getOps<fir::IfOp>()) 96 functionAnalysis.ifAnalysisMap.try_emplace(ifOp, ifOp, functionAnalysis); 97 return true; 98 } 99 100 bool analyzeLoop(fir::DoLoopOp loopOperation, 101 AffineFunctionAnalysis &functionAnalysis) { 102 LLVM_DEBUG(llvm::dbgs() << "AffineLoopAnalysis: \n"; loopOperation.dump();); 103 return analyzeMemoryAccess(loopOperation) && 104 analyzeBody(loopOperation, functionAnalysis); 105 } 106 107 bool analyzeReference(mlir::Value memref, mlir::Operation *op) { 108 if (auto acoOp = memref.getDefiningOp<ArrayCoorOp>()) { 109 if (acoOp.getMemref().getType().isa<fir::BoxType>()) { 110 // TODO: Look if and how fir.box can be promoted to affine. 111 LLVM_DEBUG(llvm::dbgs() << "AffineLoopAnalysis: cannot promote loop, " 112 "array memory operation uses fir.box\n"; 113 op->dump(); acoOp.dump();); 114 return false; 115 } 116 bool canPromote = true; 117 for (auto coordinate : acoOp.getIndices()) 118 canPromote = canPromote && analyzeCoordinate(coordinate, op); 119 return canPromote; 120 } 121 if (auto coOp = memref.getDefiningOp<CoordinateOp>()) { 122 LLVM_DEBUG(llvm::dbgs() 123 << "AffineLoopAnalysis: cannot promote loop, " 124 "array memory operation uses non ArrayCoorOp\n"; 125 op->dump(); coOp.dump();); 126 127 return false; 128 } 129 LLVM_DEBUG(llvm::dbgs() << "AffineLoopAnalysis: unknown type of memory " 130 "reference for array load\n"; 131 op->dump();); 132 return false; 133 } 134 135 bool analyzeMemoryAccess(fir::DoLoopOp loopOperation) { 136 for (auto loadOp : loopOperation.getOps<fir::LoadOp>()) 137 if (!analyzeReference(loadOp.getMemref(), loadOp)) 138 return false; 139 for (auto storeOp : loopOperation.getOps<fir::StoreOp>()) 140 if (!analyzeReference(storeOp.getMemref(), storeOp)) 141 return false; 142 return true; 143 } 144 145 bool legality{}; 146 }; 147 } // namespace 148 149 AffineLoopAnalysis 150 AffineFunctionAnalysis::getChildLoopAnalysis(fir::DoLoopOp op) const { 151 auto it = loopAnalysisMap.find_as(op); 152 if (it == loopAnalysisMap.end()) { 153 LLVM_DEBUG(llvm::dbgs() << "AffineFunctionAnalysis: not computed for:\n"; 154 op.dump();); 155 op.emitError("error in fetching loop analysis in AffineFunctionAnalysis\n"); 156 return {}; 157 } 158 return it->getSecond(); 159 } 160 161 namespace { 162 /// Calculates arguments for creating an IntegerSet. symCount, dimCount are the 163 /// final number of symbols and dimensions of the affine map. Integer set if 164 /// possible is in Optional IntegerSet. 165 struct AffineIfCondition { 166 using MaybeAffineExpr = llvm::Optional<mlir::AffineExpr>; 167 168 explicit AffineIfCondition(mlir::Value fc) : firCondition(fc) { 169 if (auto condDef = firCondition.getDefiningOp<mlir::arith::CmpIOp>()) 170 fromCmpIOp(condDef); 171 } 172 173 bool hasIntegerSet() const { return integerSet.hasValue(); } 174 175 mlir::IntegerSet getIntegerSet() const { 176 assert(hasIntegerSet() && "integer set is missing"); 177 return integerSet.getValue(); 178 } 179 180 mlir::ValueRange getAffineArgs() const { return affineArgs; } 181 182 private: 183 MaybeAffineExpr affineBinaryOp(mlir::AffineExprKind kind, mlir::Value lhs, 184 mlir::Value rhs) { 185 return affineBinaryOp(kind, toAffineExpr(lhs), toAffineExpr(rhs)); 186 } 187 188 MaybeAffineExpr affineBinaryOp(mlir::AffineExprKind kind, MaybeAffineExpr lhs, 189 MaybeAffineExpr rhs) { 190 if (lhs.hasValue() && rhs.hasValue()) 191 return mlir::getAffineBinaryOpExpr(kind, lhs.getValue(), rhs.getValue()); 192 return {}; 193 } 194 195 MaybeAffineExpr toAffineExpr(MaybeAffineExpr e) { return e; } 196 197 MaybeAffineExpr toAffineExpr(int64_t value) { 198 return {mlir::getAffineConstantExpr(value, firCondition.getContext())}; 199 } 200 201 /// Returns an AffineExpr if it is a result of operations that can be done 202 /// in an affine expression, this includes -, +, *, rem, constant. 203 /// block arguments of a loopOp or forOp are used as dimensions 204 MaybeAffineExpr toAffineExpr(mlir::Value value) { 205 if (auto op = value.getDefiningOp<mlir::arith::SubIOp>()) 206 return affineBinaryOp( 207 mlir::AffineExprKind::Add, toAffineExpr(op.getLhs()), 208 affineBinaryOp(mlir::AffineExprKind::Mul, toAffineExpr(op.getRhs()), 209 toAffineExpr(-1))); 210 if (auto op = value.getDefiningOp<mlir::arith::AddIOp>()) 211 return affineBinaryOp(mlir::AffineExprKind::Add, op.getLhs(), 212 op.getRhs()); 213 if (auto op = value.getDefiningOp<mlir::arith::MulIOp>()) 214 return affineBinaryOp(mlir::AffineExprKind::Mul, op.getLhs(), 215 op.getRhs()); 216 if (auto op = value.getDefiningOp<mlir::arith::RemUIOp>()) 217 return affineBinaryOp(mlir::AffineExprKind::Mod, op.getLhs(), 218 op.getRhs()); 219 if (auto op = value.getDefiningOp<mlir::arith::ConstantOp>()) 220 if (auto intConstant = op.getValue().dyn_cast<IntegerAttr>()) 221 return toAffineExpr(intConstant.getInt()); 222 if (auto blockArg = value.dyn_cast<mlir::BlockArgument>()) { 223 affineArgs.push_back(value); 224 if (isa<fir::DoLoopOp>(blockArg.getOwner()->getParentOp()) || 225 isa<mlir::AffineForOp>(blockArg.getOwner()->getParentOp())) 226 return {mlir::getAffineDimExpr(dimCount++, value.getContext())}; 227 return {mlir::getAffineSymbolExpr(symCount++, value.getContext())}; 228 } 229 return {}; 230 } 231 232 void fromCmpIOp(mlir::arith::CmpIOp cmpOp) { 233 auto lhsAffine = toAffineExpr(cmpOp.getLhs()); 234 auto rhsAffine = toAffineExpr(cmpOp.getRhs()); 235 if (!lhsAffine.hasValue() || !rhsAffine.hasValue()) 236 return; 237 auto constraintPair = constraint( 238 cmpOp.getPredicate(), rhsAffine.getValue() - lhsAffine.getValue()); 239 if (!constraintPair) 240 return; 241 integerSet = mlir::IntegerSet::get(dimCount, symCount, 242 {constraintPair.getValue().first}, 243 {constraintPair.getValue().second}); 244 return; 245 } 246 247 llvm::Optional<std::pair<AffineExpr, bool>> 248 constraint(mlir::arith::CmpIPredicate predicate, mlir::AffineExpr basic) { 249 switch (predicate) { 250 case mlir::arith::CmpIPredicate::slt: 251 return {std::make_pair(basic - 1, false)}; 252 case mlir::arith::CmpIPredicate::sle: 253 return {std::make_pair(basic, false)}; 254 case mlir::arith::CmpIPredicate::sgt: 255 return {std::make_pair(1 - basic, false)}; 256 case mlir::arith::CmpIPredicate::sge: 257 return {std::make_pair(0 - basic, false)}; 258 case mlir::arith::CmpIPredicate::eq: 259 return {std::make_pair(basic, true)}; 260 default: 261 return {}; 262 } 263 } 264 265 llvm::SmallVector<mlir::Value> affineArgs; 266 llvm::Optional<mlir::IntegerSet> integerSet; 267 mlir::Value firCondition; 268 unsigned symCount{0u}; 269 unsigned dimCount{0u}; 270 }; 271 } // namespace 272 273 namespace { 274 /// Analysis for affine promotion of fir.if 275 struct AffineIfAnalysis { 276 AffineIfAnalysis() = default; 277 278 explicit AffineIfAnalysis(fir::IfOp op, AffineFunctionAnalysis &afa) 279 : legality(analyzeIf(op, afa)) {} 280 281 bool canPromoteToAffine() { return legality; } 282 283 private: 284 bool analyzeIf(fir::IfOp op, AffineFunctionAnalysis &afa) { 285 if (op.getNumResults() == 0) 286 return true; 287 LLVM_DEBUG(llvm::dbgs() 288 << "AffineIfAnalysis: not promoting as op has results\n";); 289 return false; 290 } 291 292 bool legality{}; 293 }; 294 } // namespace 295 296 AffineIfAnalysis 297 AffineFunctionAnalysis::getChildIfAnalysis(fir::IfOp op) const { 298 auto it = ifAnalysisMap.find_as(op); 299 if (it == ifAnalysisMap.end()) { 300 LLVM_DEBUG(llvm::dbgs() << "AffineFunctionAnalysis: not computed for:\n"; 301 op.dump();); 302 op.emitError("error in fetching if analysis in AffineFunctionAnalysis\n"); 303 return {}; 304 } 305 return it->getSecond(); 306 } 307 308 /// AffineMap rewriting fir.array_coor operation to affine apply, 309 /// %dim = fir.gendim %lowerBound, %upperBound, %stride 310 /// %a = fir.array_coor %arr(%dim) %i 311 /// returning affineMap = affine_map<(i)[lb, ub, st] -> (i*st - lb)> 312 static mlir::AffineMap createArrayIndexAffineMap(unsigned dimensions, 313 MLIRContext *context) { 314 auto index = mlir::getAffineConstantExpr(0, context); 315 auto accuExtent = mlir::getAffineConstantExpr(1, context); 316 for (unsigned i = 0; i < dimensions; ++i) { 317 mlir::AffineExpr idx = mlir::getAffineDimExpr(i, context), 318 lowerBound = mlir::getAffineSymbolExpr(i * 3, context), 319 currentExtent = 320 mlir::getAffineSymbolExpr(i * 3 + 1, context), 321 stride = mlir::getAffineSymbolExpr(i * 3 + 2, context), 322 currentPart = (idx * stride - lowerBound) * accuExtent; 323 index = currentPart + index; 324 accuExtent = accuExtent * currentExtent; 325 } 326 return mlir::AffineMap::get(dimensions, dimensions * 3, index); 327 } 328 329 static Optional<int64_t> constantIntegerLike(const mlir::Value value) { 330 if (auto definition = value.getDefiningOp<mlir::arith::ConstantOp>()) 331 if (auto stepAttr = definition.getValue().dyn_cast<IntegerAttr>()) 332 return stepAttr.getInt(); 333 return {}; 334 } 335 336 static mlir::Type coordinateArrayElement(fir::ArrayCoorOp op) { 337 if (auto refType = 338 op.getMemref().getType().dyn_cast_or_null<ReferenceType>()) { 339 if (auto seqType = refType.getEleTy().dyn_cast_or_null<SequenceType>()) { 340 return seqType.getEleTy(); 341 } 342 } 343 op.emitError( 344 "AffineLoopConversion: array type in coordinate operation not valid\n"); 345 return mlir::Type(); 346 } 347 348 static void populateIndexArgs(fir::ArrayCoorOp acoOp, fir::ShapeOp shape, 349 SmallVectorImpl<mlir::Value> &indexArgs, 350 mlir::PatternRewriter &rewriter) { 351 auto one = rewriter.create<mlir::arith::ConstantOp>( 352 acoOp.getLoc(), rewriter.getIndexType(), rewriter.getIndexAttr(1)); 353 auto extents = shape.getExtents(); 354 for (auto i = extents.begin(); i < extents.end(); i++) { 355 indexArgs.push_back(one); 356 indexArgs.push_back(*i); 357 indexArgs.push_back(one); 358 } 359 } 360 361 static void populateIndexArgs(fir::ArrayCoorOp acoOp, fir::ShapeShiftOp shape, 362 SmallVectorImpl<mlir::Value> &indexArgs, 363 mlir::PatternRewriter &rewriter) { 364 auto one = rewriter.create<mlir::arith::ConstantOp>( 365 acoOp.getLoc(), rewriter.getIndexType(), rewriter.getIndexAttr(1)); 366 auto extents = shape.getPairs(); 367 for (auto i = extents.begin(); i < extents.end();) { 368 indexArgs.push_back(*i++); 369 indexArgs.push_back(*i++); 370 indexArgs.push_back(one); 371 } 372 } 373 374 static void populateIndexArgs(fir::ArrayCoorOp acoOp, fir::SliceOp slice, 375 SmallVectorImpl<mlir::Value> &indexArgs, 376 mlir::PatternRewriter &rewriter) { 377 auto extents = slice.getTriples(); 378 for (auto i = extents.begin(); i < extents.end();) { 379 indexArgs.push_back(*i++); 380 indexArgs.push_back(*i++); 381 indexArgs.push_back(*i++); 382 } 383 } 384 385 static void populateIndexArgs(fir::ArrayCoorOp acoOp, 386 SmallVectorImpl<mlir::Value> &indexArgs, 387 mlir::PatternRewriter &rewriter) { 388 if (auto shape = acoOp.getShape().getDefiningOp<ShapeOp>()) 389 return populateIndexArgs(acoOp, shape, indexArgs, rewriter); 390 if (auto shapeShift = acoOp.getShape().getDefiningOp<ShapeShiftOp>()) 391 return populateIndexArgs(acoOp, shapeShift, indexArgs, rewriter); 392 if (auto slice = acoOp.getShape().getDefiningOp<SliceOp>()) 393 return populateIndexArgs(acoOp, slice, indexArgs, rewriter); 394 return; 395 } 396 397 /// Returns affine.apply and fir.convert from array_coor and gendims 398 static std::pair<mlir::AffineApplyOp, fir::ConvertOp> 399 createAffineOps(mlir::Value arrayRef, mlir::PatternRewriter &rewriter) { 400 auto acoOp = arrayRef.getDefiningOp<ArrayCoorOp>(); 401 auto affineMap = 402 createArrayIndexAffineMap(acoOp.getIndices().size(), acoOp.getContext()); 403 SmallVector<mlir::Value> indexArgs; 404 indexArgs.append(acoOp.getIndices().begin(), acoOp.getIndices().end()); 405 406 populateIndexArgs(acoOp, indexArgs, rewriter); 407 408 auto affineApply = rewriter.create<mlir::AffineApplyOp>(acoOp.getLoc(), 409 affineMap, indexArgs); 410 auto arrayElementType = coordinateArrayElement(acoOp); 411 auto newType = mlir::MemRefType::get({-1}, arrayElementType); 412 auto arrayConvert = rewriter.create<fir::ConvertOp>(acoOp.getLoc(), newType, 413 acoOp.getMemref()); 414 return std::make_pair(affineApply, arrayConvert); 415 } 416 417 static void rewriteLoad(fir::LoadOp loadOp, mlir::PatternRewriter &rewriter) { 418 rewriter.setInsertionPoint(loadOp); 419 auto affineOps = createAffineOps(loadOp.getMemref(), rewriter); 420 rewriter.replaceOpWithNewOp<mlir::AffineLoadOp>( 421 loadOp, affineOps.second.getResult(), affineOps.first.getResult()); 422 } 423 424 static void rewriteStore(fir::StoreOp storeOp, 425 mlir::PatternRewriter &rewriter) { 426 rewriter.setInsertionPoint(storeOp); 427 auto affineOps = createAffineOps(storeOp.getMemref(), rewriter); 428 rewriter.replaceOpWithNewOp<mlir::AffineStoreOp>(storeOp, storeOp.getValue(), 429 affineOps.second.getResult(), 430 affineOps.first.getResult()); 431 } 432 433 static void rewriteMemoryOps(Block *block, mlir::PatternRewriter &rewriter) { 434 for (auto &bodyOp : block->getOperations()) { 435 if (isa<fir::LoadOp>(bodyOp)) 436 rewriteLoad(cast<fir::LoadOp>(bodyOp), rewriter); 437 if (isa<fir::StoreOp>(bodyOp)) 438 rewriteStore(cast<fir::StoreOp>(bodyOp), rewriter); 439 } 440 } 441 442 namespace { 443 /// Convert `fir.do_loop` to `affine.for`, creates fir.convert for arrays to 444 /// memref, rewrites array_coor to affine.apply with affine_map. Rewrites fir 445 /// loads and stores to affine. 446 class AffineLoopConversion : public mlir::OpRewritePattern<fir::DoLoopOp> { 447 public: 448 using OpRewritePattern::OpRewritePattern; 449 AffineLoopConversion(mlir::MLIRContext *context, AffineFunctionAnalysis &afa) 450 : OpRewritePattern(context), functionAnalysis(afa) {} 451 452 mlir::LogicalResult 453 matchAndRewrite(fir::DoLoopOp loop, 454 mlir::PatternRewriter &rewriter) const override { 455 LLVM_DEBUG(llvm::dbgs() << "AffineLoopConversion: rewriting loop:\n"; 456 loop.dump();); 457 LLVM_ATTRIBUTE_UNUSED auto loopAnalysis = 458 functionAnalysis.getChildLoopAnalysis(loop); 459 auto &loopOps = loop.getBody()->getOperations(); 460 auto loopAndIndex = createAffineFor(loop, rewriter); 461 auto affineFor = loopAndIndex.first; 462 auto inductionVar = loopAndIndex.second; 463 464 rewriter.startRootUpdate(affineFor.getOperation()); 465 affineFor.getBody()->getOperations().splice( 466 std::prev(affineFor.getBody()->end()), loopOps, loopOps.begin(), 467 std::prev(loopOps.end())); 468 rewriter.finalizeRootUpdate(affineFor.getOperation()); 469 470 rewriter.startRootUpdate(loop.getOperation()); 471 loop.getInductionVar().replaceAllUsesWith(inductionVar); 472 rewriter.finalizeRootUpdate(loop.getOperation()); 473 474 rewriteMemoryOps(affineFor.getBody(), rewriter); 475 476 LLVM_DEBUG(llvm::dbgs() << "AffineLoopConversion: loop rewriten to:\n"; 477 affineFor.dump();); 478 rewriter.replaceOp(loop, affineFor.getOperation()->getResults()); 479 return success(); 480 } 481 482 private: 483 std::pair<mlir::AffineForOp, mlir::Value> 484 createAffineFor(fir::DoLoopOp op, mlir::PatternRewriter &rewriter) const { 485 if (auto constantStep = constantIntegerLike(op.getStep())) 486 if (constantStep.getValue() > 0) 487 return positiveConstantStep(op, constantStep.getValue(), rewriter); 488 return genericBounds(op, rewriter); 489 } 490 491 // when step for the loop is positive compile time constant 492 std::pair<mlir::AffineForOp, mlir::Value> 493 positiveConstantStep(fir::DoLoopOp op, int64_t step, 494 mlir::PatternRewriter &rewriter) const { 495 auto affineFor = rewriter.create<mlir::AffineForOp>( 496 op.getLoc(), ValueRange(op.getLowerBound()), 497 mlir::AffineMap::get(0, 1, 498 mlir::getAffineSymbolExpr(0, op.getContext())), 499 ValueRange(op.getUpperBound()), 500 mlir::AffineMap::get(0, 1, 501 1 + mlir::getAffineSymbolExpr(0, op.getContext())), 502 step); 503 return std::make_pair(affineFor, affineFor.getInductionVar()); 504 } 505 506 std::pair<mlir::AffineForOp, mlir::Value> 507 genericBounds(fir::DoLoopOp op, mlir::PatternRewriter &rewriter) const { 508 auto lowerBound = mlir::getAffineSymbolExpr(0, op.getContext()); 509 auto upperBound = mlir::getAffineSymbolExpr(1, op.getContext()); 510 auto step = mlir::getAffineSymbolExpr(2, op.getContext()); 511 mlir::AffineMap upperBoundMap = mlir::AffineMap::get( 512 0, 3, (upperBound - lowerBound + step).floorDiv(step)); 513 auto genericUpperBound = rewriter.create<mlir::AffineApplyOp>( 514 op.getLoc(), upperBoundMap, 515 ValueRange({op.getLowerBound(), op.getUpperBound(), op.getStep()})); 516 auto actualIndexMap = mlir::AffineMap::get( 517 1, 2, 518 (lowerBound + mlir::getAffineDimExpr(0, op.getContext())) * 519 mlir::getAffineSymbolExpr(1, op.getContext())); 520 521 auto affineFor = rewriter.create<mlir::AffineForOp>( 522 op.getLoc(), ValueRange(), 523 AffineMap::getConstantMap(0, op.getContext()), 524 genericUpperBound.getResult(), 525 mlir::AffineMap::get(0, 1, 526 1 + mlir::getAffineSymbolExpr(0, op.getContext())), 527 1); 528 rewriter.setInsertionPointToStart(affineFor.getBody()); 529 auto actualIndex = rewriter.create<mlir::AffineApplyOp>( 530 op.getLoc(), actualIndexMap, 531 ValueRange( 532 {affineFor.getInductionVar(), op.getLowerBound(), op.getStep()})); 533 return std::make_pair(affineFor, actualIndex.getResult()); 534 } 535 536 AffineFunctionAnalysis &functionAnalysis; 537 }; 538 539 /// Convert `fir.if` to `affine.if`. 540 class AffineIfConversion : public mlir::OpRewritePattern<fir::IfOp> { 541 public: 542 using OpRewritePattern::OpRewritePattern; 543 AffineIfConversion(mlir::MLIRContext *context, AffineFunctionAnalysis &afa) 544 : OpRewritePattern(context) {} 545 mlir::LogicalResult 546 matchAndRewrite(fir::IfOp op, 547 mlir::PatternRewriter &rewriter) const override { 548 LLVM_DEBUG(llvm::dbgs() << "AffineIfConversion: rewriting if:\n"; 549 op.dump();); 550 auto &ifOps = op.getThenRegion().front().getOperations(); 551 auto affineCondition = AffineIfCondition(op.getCondition()); 552 if (!affineCondition.hasIntegerSet()) { 553 LLVM_DEBUG( 554 llvm::dbgs() 555 << "AffineIfConversion: couldn't calculate affine condition\n";); 556 return failure(); 557 } 558 auto affineIf = rewriter.create<mlir::AffineIfOp>( 559 op.getLoc(), affineCondition.getIntegerSet(), 560 affineCondition.getAffineArgs(), !op.getElseRegion().empty()); 561 rewriter.startRootUpdate(affineIf); 562 affineIf.getThenBlock()->getOperations().splice( 563 std::prev(affineIf.getThenBlock()->end()), ifOps, ifOps.begin(), 564 std::prev(ifOps.end())); 565 if (!op.getElseRegion().empty()) { 566 auto &otherOps = op.getElseRegion().front().getOperations(); 567 affineIf.getElseBlock()->getOperations().splice( 568 std::prev(affineIf.getElseBlock()->end()), otherOps, otherOps.begin(), 569 std::prev(otherOps.end())); 570 } 571 rewriter.finalizeRootUpdate(affineIf); 572 rewriteMemoryOps(affineIf.getBody(), rewriter); 573 574 LLVM_DEBUG(llvm::dbgs() << "AffineIfConversion: if converted to:\n"; 575 affineIf.dump();); 576 rewriter.replaceOp(op, affineIf.getOperation()->getResults()); 577 return success(); 578 } 579 }; 580 581 /// Promote fir.do_loop and fir.if to affine.for and affine.if, in the cases 582 /// where such a promotion is possible. 583 class AffineDialectPromotion 584 : public AffineDialectPromotionBase<AffineDialectPromotion> { 585 public: 586 void runOnOperation() override { 587 588 auto *context = &getContext(); 589 auto function = getOperation(); 590 markAllAnalysesPreserved(); 591 auto functionAnalysis = AffineFunctionAnalysis(function); 592 mlir::RewritePatternSet patterns(context); 593 patterns.insert<AffineIfConversion>(context, functionAnalysis); 594 patterns.insert<AffineLoopConversion>(context, functionAnalysis); 595 mlir::ConversionTarget target = *context; 596 target.addLegalDialect< 597 mlir::AffineDialect, FIROpsDialect, mlir::scf::SCFDialect, 598 mlir::arith::ArithmeticDialect, mlir::func::FuncDialect>(); 599 target.addDynamicallyLegalOp<IfOp>([&functionAnalysis](fir::IfOp op) { 600 return !(functionAnalysis.getChildIfAnalysis(op).canPromoteToAffine()); 601 }); 602 target.addDynamicallyLegalOp<DoLoopOp>([&functionAnalysis]( 603 fir::DoLoopOp op) { 604 return !(functionAnalysis.getChildLoopAnalysis(op).canPromoteToAffine()); 605 }); 606 607 LLVM_DEBUG(llvm::dbgs() 608 << "AffineDialectPromotion: running promotion on: \n"; 609 function.print(llvm::dbgs());); 610 // apply the patterns 611 if (mlir::failed(mlir::applyPartialConversion(function, target, 612 std::move(patterns)))) { 613 mlir::emitError(mlir::UnknownLoc::get(context), 614 "error in converting to affine dialect\n"); 615 signalPassFailure(); 616 } 617 } 618 }; 619 } // namespace 620 621 /// Convert FIR loop constructs to the Affine dialect 622 std::unique_ptr<mlir::Pass> fir::createPromoteToAffinePass() { 623 return std::make_unique<AffineDialectPromotion>(); 624 } 625