xref: /llvm-project/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp (revision 4c48f016effde67d500fc95290096aec9f3bdb70)
1 //===- LoopSpecialization.cpp - scf.parallel/SCR.for specialization -------===//
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 // Specializes parallel loops and for loops for easier unrolling and
10 // vectorization.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "mlir/Dialect/SCF/Transforms/Passes.h"
15 
16 #include "mlir/Dialect/Affine/Analysis/AffineStructures.h"
17 #include "mlir/Dialect/Affine/IR/AffineOps.h"
18 #include "mlir/Dialect/Arith/IR/Arith.h"
19 #include "mlir/Dialect/SCF/IR/SCF.h"
20 #include "mlir/Dialect/SCF/Transforms/Transforms.h"
21 #include "mlir/Dialect/SCF/Utils/AffineCanonicalizationUtils.h"
22 #include "mlir/Dialect/Utils/StaticValueUtils.h"
23 #include "mlir/IR/AffineExpr.h"
24 #include "mlir/IR/IRMapping.h"
25 #include "mlir/IR/PatternMatch.h"
26 #include "mlir/Transforms/GreedyPatternRewriteDriver.h"
27 #include "llvm/ADT/DenseMap.h"
28 
29 namespace mlir {
30 #define GEN_PASS_DEF_SCFFORLOOPPEELING
31 #define GEN_PASS_DEF_SCFFORLOOPSPECIALIZATION
32 #define GEN_PASS_DEF_SCFPARALLELLOOPSPECIALIZATION
33 #include "mlir/Dialect/SCF/Transforms/Passes.h.inc"
34 } // namespace mlir
35 
36 using namespace mlir;
37 using namespace mlir::affine;
38 using scf::ForOp;
39 using scf::ParallelOp;
40 
41 /// Rewrite a parallel loop with bounds defined by an affine.min with a constant
42 /// into 2 loops after checking if the bounds are equal to that constant. This
43 /// is beneficial if the loop will almost always have the constant bound and
44 /// that version can be fully unrolled and vectorized.
45 static void specializeParallelLoopForUnrolling(ParallelOp op) {
46   SmallVector<int64_t, 2> constantIndices;
47   constantIndices.reserve(op.getUpperBound().size());
48   for (auto bound : op.getUpperBound()) {
49     auto minOp = bound.getDefiningOp<AffineMinOp>();
50     if (!minOp)
51       return;
52     int64_t minConstant = std::numeric_limits<int64_t>::max();
53     for (AffineExpr expr : minOp.getMap().getResults()) {
54       if (auto constantIndex = expr.dyn_cast<AffineConstantExpr>())
55         minConstant = std::min(minConstant, constantIndex.getValue());
56     }
57     if (minConstant == std::numeric_limits<int64_t>::max())
58       return;
59     constantIndices.push_back(minConstant);
60   }
61 
62   OpBuilder b(op);
63   IRMapping map;
64   Value cond;
65   for (auto bound : llvm::zip(op.getUpperBound(), constantIndices)) {
66     Value constant =
67         b.create<arith::ConstantIndexOp>(op.getLoc(), std::get<1>(bound));
68     Value cmp = b.create<arith::CmpIOp>(op.getLoc(), arith::CmpIPredicate::eq,
69                                         std::get<0>(bound), constant);
70     cond = cond ? b.create<arith::AndIOp>(op.getLoc(), cond, cmp) : cmp;
71     map.map(std::get<0>(bound), constant);
72   }
73   auto ifOp = b.create<scf::IfOp>(op.getLoc(), cond, /*withElseRegion=*/true);
74   ifOp.getThenBodyBuilder().clone(*op.getOperation(), map);
75   ifOp.getElseBodyBuilder().clone(*op.getOperation());
76   op.erase();
77 }
78 
79 /// Rewrite a for loop with bounds defined by an affine.min with a constant into
80 /// 2 loops after checking if the bounds are equal to that constant. This is
81 /// beneficial if the loop will almost always have the constant bound and that
82 /// version can be fully unrolled and vectorized.
83 static void specializeForLoopForUnrolling(ForOp op) {
84   auto bound = op.getUpperBound();
85   auto minOp = bound.getDefiningOp<AffineMinOp>();
86   if (!minOp)
87     return;
88   int64_t minConstant = std::numeric_limits<int64_t>::max();
89   for (AffineExpr expr : minOp.getMap().getResults()) {
90     if (auto constantIndex = expr.dyn_cast<AffineConstantExpr>())
91       minConstant = std::min(minConstant, constantIndex.getValue());
92   }
93   if (minConstant == std::numeric_limits<int64_t>::max())
94     return;
95 
96   OpBuilder b(op);
97   IRMapping map;
98   Value constant = b.create<arith::ConstantIndexOp>(op.getLoc(), minConstant);
99   Value cond = b.create<arith::CmpIOp>(op.getLoc(), arith::CmpIPredicate::eq,
100                                        bound, constant);
101   map.map(bound, constant);
102   auto ifOp = b.create<scf::IfOp>(op.getLoc(), cond, /*withElseRegion=*/true);
103   ifOp.getThenBodyBuilder().clone(*op.getOperation(), map);
104   ifOp.getElseBodyBuilder().clone(*op.getOperation());
105   op.erase();
106 }
107 
108 /// Rewrite a for loop with bounds/step that potentially do not divide evenly
109 /// into a for loop where the step divides the iteration space evenly, followed
110 /// by an scf.if for the last (partial) iteration (if any).
111 ///
112 /// This function rewrites the given scf.for loop in-place and creates a new
113 /// scf.if operation for the last iteration. It replaces all uses of the
114 /// unpeeled loop with the results of the newly generated scf.if.
115 ///
116 /// The newly generated scf.if operation is returned via `ifOp`. The boundary
117 /// at which the loop is split (new upper bound) is returned via `splitBound`.
118 /// The return value indicates whether the loop was rewritten or not.
119 static LogicalResult peelForLoop(RewriterBase &b, ForOp forOp,
120                                  ForOp &partialIteration, Value &splitBound) {
121   RewriterBase::InsertionGuard guard(b);
122   auto lbInt = getConstantIntValue(forOp.getLowerBound());
123   auto ubInt = getConstantIntValue(forOp.getUpperBound());
124   auto stepInt = getConstantIntValue(forOp.getStep());
125 
126   // No specialization necessary if step already divides upper bound evenly.
127   if (lbInt && ubInt && stepInt && (*ubInt - *lbInt) % *stepInt == 0)
128     return failure();
129   // No specialization necessary if step size is 1.
130   if (stepInt == static_cast<int64_t>(1))
131     return failure();
132 
133   auto loc = forOp.getLoc();
134   AffineExpr sym0, sym1, sym2;
135   bindSymbols(b.getContext(), sym0, sym1, sym2);
136   // New upper bound: %ub - (%ub - %lb) mod %step
137   auto modMap = AffineMap::get(0, 3, {sym1 - ((sym1 - sym0) % sym2)});
138   b.setInsertionPoint(forOp);
139   splitBound = b.createOrFold<AffineApplyOp>(loc, modMap,
140                                              ValueRange{forOp.getLowerBound(),
141                                                         forOp.getUpperBound(),
142                                                         forOp.getStep()});
143 
144   // Create ForOp for partial iteration.
145   b.setInsertionPointAfter(forOp);
146   partialIteration = cast<ForOp>(b.clone(*forOp.getOperation()));
147   partialIteration.getLowerBoundMutable().assign(splitBound);
148   b.replaceAllUsesWith(forOp.getResults(), partialIteration->getResults());
149   partialIteration.getInitArgsMutable().assign(forOp->getResults());
150 
151   // Set new upper loop bound.
152   b.updateRootInPlace(
153       forOp, [&]() { forOp.getUpperBoundMutable().assign(splitBound); });
154 
155   return success();
156 }
157 
158 static void rewriteAffineOpAfterPeeling(RewriterBase &rewriter, ForOp forOp,
159                                         ForOp partialIteration,
160                                         Value previousUb) {
161   Value mainIv = forOp.getInductionVar();
162   Value partialIv = partialIteration.getInductionVar();
163   assert(forOp.getStep() == partialIteration.getStep() &&
164          "expected same step in main and partial loop");
165   Value step = forOp.getStep();
166 
167   forOp.walk([&](Operation *affineOp) {
168     if (!isa<AffineMinOp, AffineMaxOp>(affineOp))
169       return WalkResult::advance();
170     (void)scf::rewritePeeledMinMaxOp(rewriter, affineOp, mainIv, previousUb,
171                                      step,
172                                      /*insideLoop=*/true);
173     return WalkResult::advance();
174   });
175   partialIteration.walk([&](Operation *affineOp) {
176     if (!isa<AffineMinOp, AffineMaxOp>(affineOp))
177       return WalkResult::advance();
178     (void)scf::rewritePeeledMinMaxOp(rewriter, affineOp, partialIv, previousUb,
179                                      step, /*insideLoop=*/false);
180     return WalkResult::advance();
181   });
182 }
183 
184 LogicalResult mlir::scf::peelForLoopAndSimplifyBounds(RewriterBase &rewriter,
185                                                       ForOp forOp,
186                                                       ForOp &partialIteration) {
187   Value previousUb = forOp.getUpperBound();
188   Value splitBound;
189   if (failed(peelForLoop(rewriter, forOp, partialIteration, splitBound)))
190     return failure();
191 
192   // Rewrite affine.min and affine.max ops.
193   rewriteAffineOpAfterPeeling(rewriter, forOp, partialIteration, previousUb);
194 
195   return success();
196 }
197 
198 static constexpr char kPeeledLoopLabel[] = "__peeled_loop__";
199 static constexpr char kPartialIterationLabel[] = "__partial_iteration__";
200 
201 namespace {
202 struct ForLoopPeelingPattern : public OpRewritePattern<ForOp> {
203   ForLoopPeelingPattern(MLIRContext *ctx, bool skipPartial)
204       : OpRewritePattern<ForOp>(ctx), skipPartial(skipPartial) {}
205 
206   LogicalResult matchAndRewrite(ForOp forOp,
207                                 PatternRewriter &rewriter) const override {
208     // Do not peel already peeled loops.
209     if (forOp->hasAttr(kPeeledLoopLabel))
210       return failure();
211     if (skipPartial) {
212       // No peeling of loops inside the partial iteration of another peeled
213       // loop.
214       Operation *op = forOp.getOperation();
215       while ((op = op->getParentOfType<scf::ForOp>())) {
216         if (op->hasAttr(kPartialIterationLabel))
217           return failure();
218       }
219     }
220     // Apply loop peeling.
221     scf::ForOp partialIteration;
222     if (failed(peelForLoopAndSimplifyBounds(rewriter, forOp, partialIteration)))
223       return failure();
224     // Apply label, so that the same loop is not rewritten a second time.
225     rewriter.updateRootInPlace(partialIteration, [&]() {
226       partialIteration->setAttr(kPeeledLoopLabel, rewriter.getUnitAttr());
227       partialIteration->setAttr(kPartialIterationLabel, rewriter.getUnitAttr());
228     });
229     rewriter.updateRootInPlace(forOp, [&]() {
230       forOp->setAttr(kPeeledLoopLabel, rewriter.getUnitAttr());
231     });
232     return success();
233   }
234 
235   /// If set to true, loops inside partial iterations of another peeled loop
236   /// are not peeled. This reduces the size of the generated code. Partial
237   /// iterations are not usually performance critical.
238   /// Note: Takes into account the entire chain of parent operations, not just
239   /// the direct parent.
240   bool skipPartial;
241 };
242 } // namespace
243 
244 namespace {
245 struct ParallelLoopSpecialization
246     : public impl::SCFParallelLoopSpecializationBase<
247           ParallelLoopSpecialization> {
248   void runOnOperation() override {
249     getOperation()->walk(
250         [](ParallelOp op) { specializeParallelLoopForUnrolling(op); });
251   }
252 };
253 
254 struct ForLoopSpecialization
255     : public impl::SCFForLoopSpecializationBase<ForLoopSpecialization> {
256   void runOnOperation() override {
257     getOperation()->walk([](ForOp op) { specializeForLoopForUnrolling(op); });
258   }
259 };
260 
261 struct ForLoopPeeling : public impl::SCFForLoopPeelingBase<ForLoopPeeling> {
262   void runOnOperation() override {
263     auto *parentOp = getOperation();
264     MLIRContext *ctx = parentOp->getContext();
265     RewritePatternSet patterns(ctx);
266     patterns.add<ForLoopPeelingPattern>(ctx, skipPartial);
267     (void)applyPatternsAndFoldGreedily(parentOp, std::move(patterns));
268 
269     // Drop the markers.
270     parentOp->walk([](Operation *op) {
271       op->removeAttr(kPeeledLoopLabel);
272       op->removeAttr(kPartialIterationLabel);
273     });
274   }
275 };
276 } // namespace
277 
278 std::unique_ptr<Pass> mlir::createParallelLoopSpecializationPass() {
279   return std::make_unique<ParallelLoopSpecialization>();
280 }
281 
282 std::unique_ptr<Pass> mlir::createForLoopSpecializationPass() {
283   return std::make_unique<ForLoopSpecialization>();
284 }
285 
286 std::unique_ptr<Pass> mlir::createForLoopPeelingPass() {
287   return std::make_unique<ForLoopPeeling>();
288 }
289