1 //===- SCFToGPU.cpp - Convert an affine loop nest to a GPU kernel -------===// 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 implements a straightforward conversion of an loop nest into a GPU 10 // kernel. The caller is expected to guarantee that the conversion is correct 11 // or to further transform the kernel to ensure correctness. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "mlir/Conversion/SCFToGPU/SCFToGPU.h" 16 17 #include "mlir/Conversion/AffineToStandard/AffineToStandard.h" 18 #include "mlir/Dialect/Affine/IR/AffineOps.h" 19 #include "mlir/Dialect/Arith/IR/Arith.h" 20 #include "mlir/Dialect/GPU/IR/GPUDialect.h" 21 #include "mlir/Dialect/GPU/Transforms/ParallelLoopMapper.h" 22 #include "mlir/Dialect/MemRef/IR/MemRef.h" 23 #include "mlir/Dialect/SCF/IR/SCF.h" 24 #include "mlir/IR/AffineExpr.h" 25 #include "mlir/IR/Builders.h" 26 #include "mlir/IR/IRMapping.h" 27 #include "mlir/Interfaces/SideEffectInterfaces.h" 28 #include "mlir/Pass/Pass.h" 29 #include "mlir/Transforms/DialectConversion.h" 30 #include "mlir/Transforms/Passes.h" 31 #include "mlir/Transforms/RegionUtils.h" 32 #include "llvm/ADT/Sequence.h" 33 #include "llvm/Support/Debug.h" 34 #include <optional> 35 36 #define DEBUG_TYPE "loops-to-gpu" 37 38 using namespace mlir; 39 using namespace mlir::affine; 40 using namespace mlir::scf; 41 42 // Name of internal attribute to mark visited operations during conversion. 43 // 44 // NOTE: The conversion originally used the following legality criteria: 45 // `!parallelOp->hasAttr(gpu::getMappingAttrName())` 46 // But the provided pattern might reject some cases based on more detailed 47 // analysis of the `mapping` attribute. 48 // To avoid dialect conversion failure due to non-converted illegal operation 49 // we use this extra Unit attribute as a marker, that the operation was checked 50 // by the pattern and is should be considered as legal in the following legality 51 // checks. The `finalizeParallelLoopToGPUConversion` function performs clean up 52 // of this extra attributes ans is supposed to be called after the dialect 53 // conversion. 54 // 55 // TODO: Implement a cleaner solution, factoring out the "matching" logic 56 // from the pattern and its callees into a separate function that can be called 57 // from both the pattern and the op legality check. 58 static constexpr StringLiteral kVisitedAttrName = "SCFToGPU_visited"; 59 60 // Extract an indexed value from KernelDim3. 61 static Value getDim3Value(const gpu::KernelDim3 &dim3, unsigned pos) { 62 switch (pos) { 63 case 0: 64 return dim3.x; 65 case 1: 66 return dim3.y; 67 case 2: 68 return dim3.z; 69 default: 70 llvm_unreachable("dim3 position out of bounds"); 71 } 72 return nullptr; 73 } 74 75 // Get the lower bound-related operands of a loop operation. 76 static Operation::operand_range getLowerBoundOperands(AffineForOp forOp) { 77 return forOp.getLowerBoundOperands(); 78 } 79 80 // Get the upper bound-related operands of a loop operation. 81 static Operation::operand_range getUpperBoundOperands(AffineForOp forOp) { 82 return forOp.getUpperBoundOperands(); 83 } 84 85 // Get a Value that corresponds to the loop step. If the step is an attribute, 86 // materialize a corresponding constant using builder. 87 static Value getOrCreateStep(AffineForOp forOp, OpBuilder &builder) { 88 return builder.create<arith::ConstantIndexOp>(forOp.getLoc(), 89 forOp.getStep()); 90 } 91 92 // Get a Value for the loop lower bound. If the value requires computation, 93 // materialize the instructions using builder. 94 static Value getOrEmitLowerBound(AffineForOp forOp, OpBuilder &builder) { 95 return lowerAffineLowerBound(forOp, builder); 96 } 97 98 // Get a Value for the loop upper bound. If the value requires computation, 99 // materialize the instructions using builder. 100 static Value getOrEmitUpperBound(AffineForOp forOp, OpBuilder &builder) { 101 return lowerAffineUpperBound(forOp, builder); 102 } 103 104 // Check the structure of the loop nest: 105 // - there are enough loops to map to numDims; 106 // - the loops are perfectly nested; 107 // - the loop bounds can be computed above the outermost loop. 108 // This roughly corresponds to the "matcher" part of the pattern-based 109 // rewriting infrastructure. 110 static LogicalResult checkAffineLoopNestMappableImpl(AffineForOp forOp, 111 unsigned numDims) { 112 Region &limit = forOp.getRegion(); 113 for (unsigned i = 0, e = numDims; i < e; ++i) { 114 Operation *nested = &forOp.getBody()->front(); 115 if (!areValuesDefinedAbove(getLowerBoundOperands(forOp), limit) || 116 !areValuesDefinedAbove(getUpperBoundOperands(forOp), limit)) 117 return forOp.emitError( 118 "loops with bounds depending on other mapped loops " 119 "are not supported"); 120 121 // The innermost loop can have an arbitrary body, skip the perfect nesting 122 // check for it. 123 if (i == e - 1) 124 break; 125 126 auto begin = forOp.getBody()->begin(), end = forOp.getBody()->end(); 127 if (forOp.getBody()->empty() || std::next(begin, 2) != end) 128 return forOp.emitError("expected perfectly nested loops in the body"); 129 130 if (!(forOp = dyn_cast<AffineForOp>(nested))) 131 return nested->emitError("expected a nested loop"); 132 } 133 return success(); 134 } 135 136 static LogicalResult checkAffineLoopNestMappable(AffineForOp forOp, 137 unsigned numBlockDims, 138 unsigned numThreadDims) { 139 if (numBlockDims < 1 || numThreadDims < 1) { 140 LLVM_DEBUG(llvm::dbgs() << "nothing to map"); 141 return success(); 142 } 143 144 if (numBlockDims > 3) { 145 return forOp.emitError("cannot map to more than 3 block dimensions"); 146 } 147 if (numThreadDims > 3) { 148 return forOp.emitError("cannot map to more than 3 thread dimensions"); 149 } 150 return checkAffineLoopNestMappableImpl(forOp, numBlockDims + numThreadDims); 151 } 152 153 namespace { 154 // Helper structure that holds common state of the loop to GPU kernel 155 // conversion. 156 struct AffineLoopToGpuConverter { 157 std::optional<AffineForOp> collectBounds(AffineForOp forOp, 158 unsigned numLoops); 159 160 void createLaunch(AffineForOp rootForOp, AffineForOp innermostForOp, 161 unsigned numBlockDims, unsigned numThreadDims); 162 163 // Ranges of the loops mapped to blocks or threads. 164 SmallVector<Value, 6> dims; 165 // Lower bounds of the loops mapped to blocks or threads. 166 SmallVector<Value, 6> lbs; 167 // Induction variables of the loops mapped to blocks or threads. 168 SmallVector<Value, 6> ivs; 169 // Steps of the loops mapped to blocks or threads. 170 SmallVector<Value, 6> steps; 171 }; 172 } // namespace 173 174 // Collect ranges, bounds, steps and induction variables in preparation for 175 // mapping a loop nest of depth "numLoops" rooted at "forOp" to a GPU kernel. 176 // This may fail if the IR for computing loop bounds cannot be constructed, for 177 // example if an affine loop uses semi-affine maps. Return the last loop to be 178 // mapped on success, std::nullopt on failure. 179 std::optional<AffineForOp> 180 AffineLoopToGpuConverter::collectBounds(AffineForOp forOp, unsigned numLoops) { 181 OpBuilder builder(forOp.getOperation()); 182 dims.reserve(numLoops); 183 lbs.reserve(numLoops); 184 ivs.reserve(numLoops); 185 steps.reserve(numLoops); 186 AffineForOp currentLoop = forOp; 187 for (unsigned i = 0; i < numLoops; ++i) { 188 Value lowerBound = getOrEmitLowerBound(currentLoop, builder); 189 Value upperBound = getOrEmitUpperBound(currentLoop, builder); 190 if (!lowerBound || !upperBound) { 191 return std::nullopt; 192 } 193 194 Value range = builder.create<arith::SubIOp>(currentLoop.getLoc(), 195 upperBound, lowerBound); 196 Value step = getOrCreateStep(currentLoop, builder); 197 if (getConstantIntValue(step) != static_cast<int64_t>(1)) 198 range = builder.create<arith::DivSIOp>(currentLoop.getLoc(), range, step); 199 dims.push_back(range); 200 201 lbs.push_back(lowerBound); 202 ivs.push_back(currentLoop.getInductionVar()); 203 steps.push_back(step); 204 205 if (i != numLoops - 1) 206 currentLoop = cast<AffineForOp>(¤tLoop.getBody()->front()); 207 } 208 return currentLoop; 209 } 210 211 // Replace the rooted at "rootForOp" with a GPU launch operation. This expects 212 // "innermostForOp" to point to the last loop to be transformed to the kernel, 213 // and to have (numBlockDims + numThreadDims) perfectly nested loops between 214 // "rootForOp" and "innermostForOp". 215 void AffineLoopToGpuConverter::createLaunch(AffineForOp rootForOp, 216 AffineForOp innermostForOp, 217 unsigned numBlockDims, 218 unsigned numThreadDims) { 219 OpBuilder builder(rootForOp.getOperation()); 220 // Prepare the grid and block sizes for the launch operation. If there is 221 // no loop mapped to a specific dimension, use constant "1" as its size. 222 Value constOne = 223 (numBlockDims < 3 || numThreadDims < 3) 224 ? builder.create<arith::ConstantIndexOp>(rootForOp.getLoc(), 1) 225 : nullptr; 226 Value gridSizeX = numBlockDims > 0 ? dims[0] : constOne; 227 Value gridSizeY = numBlockDims > 1 ? dims[1] : constOne; 228 Value gridSizeZ = numBlockDims > 2 ? dims[2] : constOne; 229 Value blockSizeX = numThreadDims > 0 ? dims[numBlockDims] : constOne; 230 Value blockSizeY = numThreadDims > 1 ? dims[numBlockDims + 1] : constOne; 231 Value blockSizeZ = numThreadDims > 2 ? dims[numBlockDims + 2] : constOne; 232 233 // Create a launch op and move the body region of the innermost loop to the 234 // launch op. 235 auto launchOp = builder.create<gpu::LaunchOp>( 236 rootForOp.getLoc(), gridSizeX, gridSizeY, gridSizeZ, blockSizeX, 237 blockSizeY, blockSizeZ); 238 239 // Replace the loop terminator (loops contain only a single block) with the 240 // gpu terminator and move the operations from the loop body block to the gpu 241 // launch body block. Do not move the entire block because of the difference 242 // in block arguments. 243 Operation &terminator = innermostForOp.getBody()->back(); 244 Location terminatorLoc = terminator.getLoc(); 245 terminator.erase(); 246 builder.setInsertionPointToEnd(innermostForOp.getBody()); 247 builder.create<gpu::TerminatorOp>(terminatorLoc, std::nullopt); 248 launchOp.getBody().front().getOperations().splice( 249 launchOp.getBody().front().begin(), 250 innermostForOp.getBody()->getOperations()); 251 252 // Remap the loop iterators to use block/thread identifiers instead. Loops 253 // may iterate from LB with step S whereas GPU thread/block ids always iterate 254 // from 0 to N with step 1. Therefore, loop induction variables are replaced 255 // with (gpu-thread/block-id * S) + LB. 256 builder.setInsertionPointToStart(&launchOp.getBody().front()); 257 auto *lbArgumentIt = lbs.begin(); 258 auto *stepArgumentIt = steps.begin(); 259 for (const auto &en : llvm::enumerate(ivs)) { 260 Value id = 261 en.index() < numBlockDims 262 ? getDim3Value(launchOp.getBlockIds(), en.index()) 263 : getDim3Value(launchOp.getThreadIds(), en.index() - numBlockDims); 264 Value step = steps[en.index()]; 265 if (getConstantIntValue(step) != static_cast<int64_t>(1)) 266 id = builder.create<arith::MulIOp>(rootForOp.getLoc(), step, id); 267 268 Value ivReplacement = 269 builder.create<arith::AddIOp>(rootForOp.getLoc(), *lbArgumentIt, id); 270 en.value().replaceAllUsesWith(ivReplacement); 271 std::advance(lbArgumentIt, 1); 272 std::advance(stepArgumentIt, 1); 273 } 274 275 // We are done and can erase the original outermost loop. 276 rootForOp.erase(); 277 } 278 279 // Generic loop to GPU kernel conversion function. 280 static LogicalResult convertAffineLoopNestToGPULaunch(AffineForOp forOp, 281 unsigned numBlockDims, 282 unsigned numThreadDims) { 283 if (failed(checkAffineLoopNestMappable(forOp, numBlockDims, numThreadDims))) 284 return failure(); 285 286 AffineLoopToGpuConverter converter; 287 auto maybeInnerLoop = 288 converter.collectBounds(forOp, numBlockDims + numThreadDims); 289 if (!maybeInnerLoop) 290 return failure(); 291 converter.createLaunch(forOp, *maybeInnerLoop, numBlockDims, numThreadDims); 292 293 return success(); 294 } 295 296 LogicalResult mlir::convertAffineLoopNestToGPULaunch(AffineForOp forOp, 297 unsigned numBlockDims, 298 unsigned numThreadDims) { 299 return ::convertAffineLoopNestToGPULaunch(forOp, numBlockDims, numThreadDims); 300 } 301 302 namespace { 303 struct ParallelToGpuLaunchLowering : public OpRewritePattern<ParallelOp> { 304 using OpRewritePattern<ParallelOp>::OpRewritePattern; 305 306 LogicalResult matchAndRewrite(ParallelOp parallelOp, 307 PatternRewriter &rewriter) const override; 308 }; 309 } // namespace 310 311 /// Tries to derive a static upper bound from the defining operation of 312 /// `upperBound`. 313 static Value deriveStaticUpperBound(Value upperBound, 314 PatternRewriter &rewriter) { 315 if (auto op = upperBound.getDefiningOp<arith::ConstantIndexOp>()) { 316 return op; 317 } 318 319 if (auto minOp = upperBound.getDefiningOp<AffineMinOp>()) { 320 for (const AffineExpr &result : minOp.getMap().getResults()) { 321 if (auto constExpr = result.dyn_cast<AffineConstantExpr>()) { 322 return rewriter.create<arith::ConstantIndexOp>(minOp.getLoc(), 323 constExpr.getValue()); 324 } 325 } 326 } 327 328 if (auto minOp = upperBound.getDefiningOp<arith::MinSIOp>()) { 329 for (Value operand : {minOp.getLhs(), minOp.getRhs()}) { 330 if (auto staticBound = deriveStaticUpperBound(operand, rewriter)) 331 return staticBound; 332 } 333 } 334 335 if (auto multiplyOp = upperBound.getDefiningOp<arith::MulIOp>()) { 336 if (auto lhs = dyn_cast_or_null<arith::ConstantIndexOp>( 337 deriveStaticUpperBound(multiplyOp.getOperand(0), rewriter) 338 .getDefiningOp())) 339 if (auto rhs = dyn_cast_or_null<arith::ConstantIndexOp>( 340 deriveStaticUpperBound(multiplyOp.getOperand(1), rewriter) 341 .getDefiningOp())) { 342 // Assumptions about the upper bound of minimum computations no longer 343 // work if multiplied by mixed signs, so abort in this case. 344 if ((lhs.value() < 0) != (rhs.value() < 0)) 345 return {}; 346 347 return rewriter.create<arith::ConstantIndexOp>( 348 multiplyOp.getLoc(), lhs.value() * rhs.value()); 349 } 350 } 351 352 return {}; 353 } 354 355 static bool isMappedToProcessor(gpu::Processor processor) { 356 return processor != gpu::Processor::Sequential; 357 } 358 359 static unsigned getLaunchOpArgumentNum(gpu::Processor processor) { 360 switch (processor) { 361 case gpu::Processor::BlockX: 362 return 0; 363 case gpu::Processor::BlockY: 364 return 1; 365 case gpu::Processor::BlockZ: 366 return 2; 367 case gpu::Processor::ThreadX: 368 return 3; 369 case gpu::Processor::ThreadY: 370 return 4; 371 case gpu::Processor::ThreadZ: 372 return 5; 373 default:; 374 } 375 llvm_unreachable( 376 "invalid processor type while retrieving launch op argument number"); 377 } 378 379 /// Modifies the current transformation state to capture the effect of the given 380 /// `scf.parallel` operation on index substitutions and the operations to be 381 /// inserted. 382 /// Specifically, if a dimension of a parallel loop is mapped to a hardware id, 383 /// this function will 384 /// - compute the loop index based on the hardware id and affine map from the 385 /// mapping and update `cloningMap` to substitute all uses. 386 /// - derive a new upper bound for the hardware id and augment the provided 387 /// `gpu.launch operation` accordingly. 388 /// - if the upper bound is imprecise, insert a conditional in the `gpu.launch` 389 /// and update the rewriter to insert into the conditional's body. 390 /// If the dimension is mapped to sequential, 391 /// - insert a for loop into the body and update the rewriter to insert into 392 /// the for loop's body. 393 /// - update the `cloningMap` to replace uses of the index with the index of 394 /// the new for loop. 395 /// In either case, 396 /// - append the instructions from the loops body to worklist, in reverse order. 397 /// To note the end of the current scope in case a loop or conditional was 398 /// inserted, a sentinel (the `gpu.launch` operation) is inserted into the 399 /// worklist. This signals the processor of the worklist to pop the rewriter 400 /// one scope-level up. 401 static LogicalResult processParallelLoop( 402 ParallelOp parallelOp, gpu::LaunchOp launchOp, IRMapping &cloningMap, 403 SmallVectorImpl<Operation *> &worklist, 404 DenseMap<gpu::Processor, Value> &bounds, PatternRewriter &rewriter) { 405 // TODO: Verify that this is a valid GPU mapping. 406 // processor ids: 0-2 block [x/y/z], 3-5 -> thread [x/y/z], 6-> sequential 407 ArrayAttr mapping = 408 parallelOp->getAttrOfType<ArrayAttr>(gpu::getMappingAttrName()); 409 410 // TODO: Support reductions. 411 if (!mapping || parallelOp.getNumResults() != 0) 412 return failure(); 413 414 Location loc = parallelOp.getLoc(); 415 416 auto launchIndependent = [&launchOp](Value val) { 417 return val.getParentRegion()->isAncestor(launchOp->getParentRegion()); 418 }; 419 420 auto ensureLaunchIndependent = [&rewriter, 421 launchIndependent](Value val) -> Value { 422 if (launchIndependent(val)) 423 return val; 424 if (auto constOp = val.getDefiningOp<arith::ConstantOp>()) 425 return rewriter.create<arith::ConstantOp>(constOp.getLoc(), 426 constOp.getValue()); 427 return {}; 428 }; 429 430 for (auto config : llvm::zip( 431 mapping, parallelOp.getInductionVars(), parallelOp.getLowerBound(), 432 parallelOp.getUpperBound(), parallelOp.getStep())) { 433 Attribute mappingAttribute; 434 Value iv, lowerBound, upperBound, step; 435 std::tie(mappingAttribute, iv, lowerBound, upperBound, step) = config; 436 auto annotation = 437 dyn_cast<gpu::ParallelLoopDimMappingAttr>(mappingAttribute); 438 if (!annotation) 439 return parallelOp.emitOpError() 440 << "expected mapping attribute for lowering to GPU"; 441 Value newIndex; 442 gpu::Processor processor = annotation.getProcessor(); 443 444 if (isMappedToProcessor(processor)) { 445 // Use the corresponding thread/grid index as replacement for the loop iv. 446 Value operand = 447 launchOp.getBody().getArgument(getLaunchOpArgumentNum(processor)); 448 // Take the indexmap and add the lower bound and step computations in. 449 // This computes operand * step + lowerBound. 450 // Use an affine map here so that it composes nicely with the provided 451 // annotation. 452 AffineMap lowerAndStep = AffineMap::get( 453 1, 2, 454 rewriter.getAffineDimExpr(0) * rewriter.getAffineSymbolExpr(0) + 455 rewriter.getAffineSymbolExpr(1)); 456 newIndex = rewriter.create<AffineApplyOp>( 457 loc, annotation.getMap().compose(lowerAndStep), 458 ValueRange{operand, step, lowerBound}); 459 // If there was also a bound, insert that, too. 460 // TODO: Check that we do not assign bounds twice. 461 if (annotation.getBound()) { 462 // We pass as the single operand to the bound-map the number of 463 // iterations, which is (upperBound - lowerBound) ceilDiv step. To 464 // support inner loops with dynamic upper bounds (as generated by e.g. 465 // tiling), try to derive a max for the bounds. If the used bound for 466 // the hardware id is imprecise, wrap the contained code into a 467 // conditional. If the lower-bound is constant or defined before the 468 // launch, we can use it in the launch bounds. Otherwise fail. 469 if (!launchIndependent(lowerBound) && 470 !isa_and_nonnull<arith::ConstantOp>(lowerBound.getDefiningOp())) 471 return failure(); 472 // The step must also be constant or defined outside of the loop nest. 473 if (!launchIndependent(step) && 474 !isa_and_nonnull<arith::ConstantOp>(step.getDefiningOp())) 475 return failure(); 476 // If the upper-bound is constant or defined before the launch, we can 477 // use it in the launch bounds directly. Otherwise try derive a bound. 478 bool boundIsPrecise = 479 launchIndependent(upperBound) || 480 isa_and_nonnull<arith::ConstantOp>(upperBound.getDefiningOp()); 481 { 482 PatternRewriter::InsertionGuard guard(rewriter); 483 rewriter.setInsertionPoint(launchOp); 484 if (!boundIsPrecise) { 485 upperBound = deriveStaticUpperBound(upperBound, rewriter); 486 if (!upperBound) { 487 return rewriter.notifyMatchFailure( 488 parallelOp, 489 "cannot derive loop-invariant upper bound for number of" 490 "iterations"); 491 } 492 } 493 // Compute the number of iterations needed. We compute this as an 494 // affine expression ceilDiv (upperBound - lowerBound) step. We use 495 // affine.apply here so that it composes nicely with the provided map. 496 AffineMap stepMap = AffineMap::get( 497 1, 2, 498 ((rewriter.getAffineDimExpr(0) - rewriter.getAffineSymbolExpr(0)) 499 .ceilDiv(rewriter.getAffineSymbolExpr(1)))); 500 Value launchBound = rewriter.create<AffineApplyOp>( 501 loc, annotation.getBound().compose(stepMap), 502 ValueRange{ 503 ensureLaunchIndependent( 504 cloningMap.lookupOrDefault(upperBound)), 505 ensureLaunchIndependent( 506 cloningMap.lookupOrDefault(lowerBound)), 507 ensureLaunchIndependent(cloningMap.lookupOrDefault(step))}); 508 // todo(herhut,ravishankarm): Update the behavior of setMappingAttr 509 // when this condition is relaxed. 510 if (bounds.contains(processor)) { 511 return rewriter.notifyMatchFailure( 512 parallelOp, "cannot redefine the bound for processor " + 513 Twine(static_cast<int64_t>(processor))); 514 } 515 bounds[processor] = launchBound; 516 } 517 if (!boundIsPrecise) { 518 // We are using an approximation, create a surrounding conditional. 519 Value originalBound = std::get<3>(config); 520 arith::CmpIOp pred = rewriter.create<arith::CmpIOp>( 521 loc, arith::CmpIPredicate::slt, newIndex, 522 cloningMap.lookupOrDefault(originalBound)); 523 scf::IfOp ifOp = rewriter.create<scf::IfOp>(loc, pred, false); 524 rewriter.setInsertionPointToStart(&ifOp.getThenRegion().front()); 525 // Put a sentinel into the worklist so we know when to pop out of the 526 // if body again. We use the launchOp here, as that cannot be part of 527 // the bodies instruction. 528 worklist.push_back(launchOp.getOperation()); 529 } 530 } 531 } else { 532 // Create a sequential for loop. 533 auto loopOp = rewriter.create<scf::ForOp>( 534 loc, cloningMap.lookupOrDefault(lowerBound), 535 cloningMap.lookupOrDefault(upperBound), 536 cloningMap.lookupOrDefault(step)); 537 newIndex = loopOp.getInductionVar(); 538 rewriter.setInsertionPointToStart(loopOp.getBody()); 539 // Put a sentinel into the worklist so we know when to pop out of the loop 540 // body again. We use the launchOp here, as that cannot be part of the 541 // bodies instruction. 542 worklist.push_back(launchOp.getOperation()); 543 } 544 cloningMap.map(iv, newIndex); 545 } 546 547 // Propagate custom user defined optional attributes, that can be used at 548 // later stage, such as extension data for GPU kernel dispatch 549 for (const auto &namedAttr : parallelOp->getAttrs()) { 550 if (namedAttr.getName() == gpu::getMappingAttrName() || 551 namedAttr.getName() == ParallelOp::getOperandSegmentSizeAttr()) 552 continue; 553 launchOp->setAttr(namedAttr.getName(), namedAttr.getValue()); 554 } 555 556 Block *body = parallelOp.getBody(); 557 worklist.reserve(worklist.size() + body->getOperations().size()); 558 for (Operation &op : llvm::reverse(body->without_terminator())) 559 worklist.push_back(&op); 560 return success(); 561 } 562 563 /// Lower a `scf.parallel` operation into a corresponding `gpu.launch` 564 /// operation. 565 /// 566 /// This essentially transforms a loop nest into a corresponding SIMT function. 567 /// The conversion is driven by mapping annotations on the `scf.parallel` 568 /// operations. The mapping is provided via a `DictionaryAttribute` named 569 /// `mapping`, which has three entries: 570 /// - processor: the hardware id to map to. 0-2 are block dimensions, 3-5 are 571 /// thread dimensions and 6 is sequential. 572 /// - map : An affine map that is used to pre-process hardware ids before 573 /// substitution. 574 /// - bound : An affine map that is used to compute the bound of the hardware 575 /// id based on an upper bound of the number of iterations. 576 /// If the `scf.parallel` contains nested `scf.parallel` operations, those 577 /// need to be annotated, as well. Structurally, the transformation works by 578 /// splicing all operations from nested `scf.parallel` operations into a single 579 /// sequence. Indices mapped to hardware ids are substituted with those ids, 580 /// wheras sequential mappings result in a sequential for-loop. To have more 581 /// flexibility when mapping code to hardware ids, the transform supports two 582 /// affine maps. The first `map` is used to compute the actual index for 583 /// substitution from the hardware id. The second `bound` is used to compute the 584 /// launch dimension for the hardware id from the number of iterations the 585 /// mapped loop is performing. Note that the number of iterations might be 586 /// imprecise if the corresponding loop-bounds are loop-dependent. In such case, 587 /// the hardware id might iterate over additional indices. The transformation 588 /// caters for this by predicating the created sequence of instructions on 589 /// the actual loop bound. This only works if an static upper bound for the 590 /// dynamic loop bound can be derived, currently via analyzing `affine.min` 591 /// operations. 592 LogicalResult 593 ParallelToGpuLaunchLowering::matchAndRewrite(ParallelOp parallelOp, 594 PatternRewriter &rewriter) const { 595 // Mark the operation as visited for recursive legality check. 596 parallelOp->setAttr(kVisitedAttrName, rewriter.getUnitAttr()); 597 598 // We can only transform starting at the outer-most loop. Launches inside of 599 // parallel loops are not supported. 600 if (auto parentLoop = parallelOp->getParentOfType<ParallelOp>()) 601 return failure(); 602 // Create a launch operation. We start with bound one for all grid/block 603 // sizes. Those will be refined later as we discover them from mappings. 604 Location loc = parallelOp.getLoc(); 605 Value constantOne = 606 rewriter.create<arith::ConstantIndexOp>(parallelOp.getLoc(), 1); 607 gpu::LaunchOp launchOp = rewriter.create<gpu::LaunchOp>( 608 parallelOp.getLoc(), constantOne, constantOne, constantOne, constantOne, 609 constantOne, constantOne); 610 rewriter.setInsertionPointToEnd(&launchOp.getBody().front()); 611 rewriter.create<gpu::TerminatorOp>(loc); 612 rewriter.setInsertionPointToStart(&launchOp.getBody().front()); 613 614 IRMapping cloningMap; 615 llvm::DenseMap<gpu::Processor, Value> launchBounds; 616 SmallVector<Operation *, 16> worklist; 617 if (failed(processParallelLoop(parallelOp, launchOp, cloningMap, worklist, 618 launchBounds, rewriter))) 619 return failure(); 620 621 // Whether we have seen any side-effects. Reset when leaving an inner scope. 622 bool seenSideeffects = false; 623 // Whether we have left a nesting scope (and hence are no longer innermost). 624 bool leftNestingScope = false; 625 while (!worklist.empty()) { 626 Operation *op = worklist.pop_back_val(); 627 // Now walk over the body and clone it. 628 // TODO: This is only correct if there either is no further scf.parallel 629 // nested or this code is side-effect free. Otherwise we might need 630 // predication. We are overly conservative for now and only allow 631 // side-effects in the innermost scope. 632 if (auto nestedParallel = dyn_cast<ParallelOp>(op)) { 633 // Before entering a nested scope, make sure there have been no 634 // sideeffects until now. 635 if (seenSideeffects) 636 return failure(); 637 // A nested scf.parallel needs insertion of code to compute indices. 638 // Insert that now. This will also update the worklist with the loops 639 // body. 640 if (failed(processParallelLoop(nestedParallel, launchOp, cloningMap, 641 worklist, launchBounds, rewriter))) 642 return failure(); 643 } else if (op == launchOp.getOperation()) { 644 // Found our sentinel value. We have finished the operations from one 645 // nesting level, pop one level back up. 646 auto *parent = rewriter.getInsertionPoint()->getParentOp(); 647 rewriter.setInsertionPointAfter(parent); 648 leftNestingScope = true; 649 seenSideeffects = false; 650 } else { 651 // Otherwise we copy it over. 652 Operation *clone = rewriter.clone(*op, cloningMap); 653 cloningMap.map(op->getResults(), clone->getResults()); 654 // Check for side effects. 655 // TODO: Handle region side effects properly. 656 seenSideeffects |= 657 !isMemoryEffectFree(clone) || clone->getNumRegions() != 0; 658 // If we are no longer in the innermost scope, sideeffects are disallowed. 659 if (seenSideeffects && leftNestingScope) 660 return failure(); 661 } 662 } 663 664 // Now that we succeeded creating the launch operation, also update the 665 // bounds. 666 for (auto bound : launchBounds) 667 launchOp.setOperand(getLaunchOpArgumentNum(std::get<0>(bound)), 668 std::get<1>(bound)); 669 670 rewriter.eraseOp(parallelOp); 671 return success(); 672 } 673 674 void mlir::populateParallelLoopToGPUPatterns(RewritePatternSet &patterns) { 675 patterns.add<ParallelToGpuLaunchLowering>(patterns.getContext()); 676 } 677 678 void mlir::configureParallelLoopToGPULegality(ConversionTarget &target) { 679 target.addLegalDialect<memref::MemRefDialect>(); 680 target.addDynamicallyLegalOp<scf::ParallelOp>([](scf::ParallelOp parallelOp) { 681 return !parallelOp->hasAttr(gpu::getMappingAttrName()) || 682 parallelOp->hasAttr(kVisitedAttrName); 683 }); 684 } 685 686 void mlir::finalizeParallelLoopToGPUConversion(Operation *op) { 687 op->walk([](scf::ParallelOp parallelOp) { 688 parallelOp->removeAttr(kVisitedAttrName); 689 }); 690 } 691