[mli][Linalg] Add a hoistViewAllocOps helper function

[nicolasvasilache]

Summary

This revision adds a helper function to hoist alloc/dealloc pairs and
alloca op out of immediately enclosing scf::ForOp if both conditions are true:

1. all operands are defined outside the loop.
2. all uses are ViewLikeOp or DeallocOp.

This is now considered Linalg-specific and will be generalized on a per-need basis.

[nicolasvasilache]

Rebase

[ftynse]

LGTM module some comments

mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h

521

Spurious whitespace change

mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp

33

Is this loop necessary? It doesn't look like hoisiting only the alloc can put operands of allocs out of the loop.

47

This seems to contradict the code, which fails and returns from the function if any operand is _not_ defined outside of the loop

57

Nit: assert there's one result?

65

Copy-pasta

mlir/test/Dialect/Linalg/hoisting.mlir

23

These are a awkward to read IMO

[nicolasvasilache]

Address review

mlir/include/mlir/Dialect/Linalg/Transforms/Hoisting.h

+//===- Hoisting.h - Linalg hoisting transformations -------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_LINALG_TRANSFORMS_HOISTING_H_
+#define MLIR_DIALECT_LINALG_TRANSFORMS_HOISTING_H_
+
+namespace mlir {
+class FuncOp;
+
+namespace linalg {
+
+/// Hoist alloc/dealloc pairs and alloca op out of immediately enclosing
+/// scf::ForOp if both conditions are true:
+///   1. all operands are defined outside the loop.
+///   2. all uses are ViewLikeOp or DeallocOp.
+// TODO: generalize on a per-need basis.
+void hoistViewAllocOps(FuncOp func);
+
+} // namespace linalg
+} // namespace mlir
+
+#endif // MLIR_DIALECT_LINALG_TRANSFORMS_HOISTING_H_

mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h

 ///   3. the third stage consists of applying a lambda, generally used for
 ///   non-local transformation effects. This allows creating custom fused
 ///   transformations where patterns can be ordered and applied at a finer
 ///   granularity than a sequence of traditional compiler passes.
 LogicalResult applyStagedPatterns(
     Operation *op, ArrayRef<OwningRewritePatternList> stage1Patterns,
     const OwningRewritePatternList &stage2Patterns,
     function_ref<LogicalResult(Operation *)> stage3Lambda = nullptr);
+
 } // namespace linalg
 } // namespace mlir
 
 #endif // DIALECT_LINALG_TRANSFORMS_TRANSFORMS_H_

mlir/lib/Dialect/Linalg/Transforms/CMakeLists.txt

 add_mlir_dialect_library(MLIRLinalgTransforms
   DropUnitDims.cpp
   Fusion.cpp
+  Hoisting.cpp
   Interchange.cpp
   Loops.cpp
   Promotion.cpp
   TensorsToBuffers.cpp
   Tiling.cpp
   Transforms.cpp
   Vectorization.cpp
 

mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp

+//===- Hoisting.cpp - Linalg hoisting transformations ---------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements functions concerned with hoisting invariant operations
+// in the context of Linalg transformations.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Linalg/Transforms/Hoisting.h"
+#include "mlir/Dialect/Linalg/IR/LinalgOps.h"
+#include "mlir/Dialect/SCF/SCF.h"
+#include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/IR/Function.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/Support/Debug.h"
+
+#define DEBUG_TYPE "linalg-hoisting"
+
+#define DBGS() (dbgs() << '[' << DEBUG_TYPE << "] ")
+
+using namespace mlir;
+using namespace mlir::linalg;
+
+using llvm::dbgs;
+
+void mlir::linalg::hoistViewAllocOps(FuncOp func) {
+  bool changed = true;
+  while (changed) {
+    changed = false;
+    func.walk([&changed](Operation *op) {
+      if (!isa<AllocOp>(op) && !isa<AllocaOp>(op) && !isa<DeallocOp>(op))
+        return;
+
+      LLVM_DEBUG(DBGS() << "Candidate for hoisting: " << *op << "\n");
+      auto loop = dyn_cast<scf::ForOp>(op->getParentOp());
+      LLVM_DEBUG(DBGS() << "Parent op: " << *op->getParentOp() << "\n");
+
+      // Only hoist out of immediately enclosing scf::ForOp.
+      if (!loop)
+        return;
+
+      // If any operand is defined outside of the loop don't hoist.
+      if (llvm::any_of(op->getOperands(), [&](Value v) {
+            return !loop.isDefinedOutsideOfLoop(v);
+          }))
+        return;
+
+      LLVM_DEBUG(DBGS() << "All operands defined outside \n");
+
+      // If alloc has other uses than ViewLikeOp and DeallocOp don't hoist.
+      Value v;
+      if (op->getNumResults() > 0)
+        v = op->getResult(0);
+      if (v && !llvm::all_of(v.getUses(), [&](OpOperand &operand) {
+            return isa<ViewLikeOpInterface>(operand.getOwner()) ||
+                   isa<DeallocOp>(operand.getOwner());
+          }))
+        return;
+
+      LLVM_DEBUG(DBGS() << "All operands defined outside \n");
+
+      // Move AllocOp before the loop.
+      if (isa<AllocOp>(op) || isa<AllocaOp>(op))
+        loop.moveOutOfLoop({op});
+      else // Move DeallocOp outside of the loop.
+        op->moveAfter(loop);
+      changed = true;
+    });
+  }
+}

mlir/test/Dialect/Linalg/hoisting.mlir

+// RUN: mlir-opt %s -test-linalg-hoisting=test-hoist-view-allocs | FileCheck %s
+
+// CHECK-LABEL: func @hoist(
+//  CHECK-SAME:   %[[VAL:[a-zA-Z0-9]*]]: index,
+//  CHECK-SAME:   %[[LB:[a-zA-Z0-9]*]]: index,
+//  CHECK-SAME:   %[[UB:[a-zA-Z0-9]*]]: index,
+//  CHECK-SAME:   %[[STEP:[a-zA-Z0-9]*]]: index,
+//  CHECK-SAME:   %[[CMP:[a-zA-Z0-9]*]]: i1
+func @hoist(%val: index, %lb : index, %ub : index, %step: index, %cmp: i1) {
+//   CHECK-DAG:   alloca(%[[VAL]]) : memref<?xi8>
+//   CHECK-DAG: %[[A0:.*]] = alloc(%[[VAL]]) : memref<?xi8>
+  scf.for %i = %lb to %ub step %step {
+//       CHECK: scf.for %[[I:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] {
+//       CHECK:   alloca(%[[I]]) : memref<?xi8>
+//       CHECK:   %[[A1:.*]] = alloc(%[[I]]) : memref<?xi8>
+
+    scf.for %j = %lb to %ub step %step {
+//       CHECK:   scf.for %[[J:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] {
+//   CHECK-DAG:     alloca(%[[J]]) : memref<?xi8>
+//   CHECK-DAG:     %[[A2:.*]] = alloc(%[[J]]) : memref<?xi8>
+
+      scf.for %k = %lb to %ub step %step {
+//       CHECK:     scf.for %[[K:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] {
+
+        // Hoist allocs / deallocs outermost, keep view/subview below k.
+        %sa0 = alloca(%val) : memref<? x i8>
+        %a0 = alloc(%val) : memref<? x i8>
+        %v0 = view %a0[%lb][] : memref<? x i8> to memref<16 x f32>
+        %sv0 = subview %a0[0][42][1] : memref<? x i8> to memref<42 x i8>
+        dealloc %a0 : memref<? x i8>
+//       CHECK:       std.view %[[A0]][%[[LB]]][] : memref<?xi8> to memref<16xf32>
+//       CHECK:       subview %[[A0]][0] [42] [1]  : memref<?xi8> to memref<42xi8>
+
+        // Hoist below i.
+        %sa1 = alloca(%i) : memref<? x i8>
+        %a1 = alloc(%i) : memref<? x i8>
+        dealloc %a1 : memref<? x i8>
+
+        // Hoist below j.
+        %sa2 = alloca(%j) : memref<? x i8>
+        %a2 = alloc(%j) : memref<? x i8>
+        dealloc %a2 : memref<? x i8>
+
+        // Don't hoist since k innermost.
+        %sa3 = alloca(%k) : memref<? x i8>
+        %a3 = alloc(%k) : memref<? x i8>
+        dealloc %a3 : memref<? x i8>
+//       CHECK:       alloca(%[[K]]) : memref<?xi8>
+//       CHECK:       %[[A3:.*]] = alloc(%[[K]]) : memref<?xi8>
+//       CHECK:       dealloc %[[A3]] : memref<?xi8>
+
+        // No hoisting due to control flow.
+        scf.if %cmp {
+          %sa4 = alloca(%val) : memref<? x i8>
+          %a4 = alloc(%val) : memref<? x i8>
+          dealloc %a4 : memref<? x i8>
+        }
+//       CHECK:       scf.if %[[CMP]] {
+//       CHECK:         alloca(%[[VAL]]) : memref<?xi8>
+//       CHECK:         %[[A4:.*]] = alloc(%[[VAL]]) : memref<?xi8>
+//       CHECK:         dealloc %[[A4]] : memref<?xi8>
+
+        // No hoisting due to load/store.
+        %sa5 = alloca(%val) : memref<? x i8>
+        %a5 = alloc(%val) : memref<? x i8>
+        %v5 = load %a5[%lb] : memref<? x i8>
+        store %v5, %sa5[%lb] : memref<? x i8>
+        dealloc %a5 : memref<? x i8>
+//       CHECK:       %[[SA5:.*]] = alloca(%[[VAL]]) : memref<?xi8>
+//       CHECK:       %[[A5:.*]] = alloc(%[[VAL]]) : memref<?xi8>
+//       CHECK:       load %[[A5]][%[[LB]]] : memref<?xi8>
+//       CHECK:       store %{{.*}}, %[[SA5]][%[[LB]]] : memref<?xi8>
+//       CHECK:       dealloc %[[A5]] : memref<?xi8>
+
+      }
+//       CHECK:     }
+//       CHECK:     dealloc %[[A2]] : memref<?xi8>
+    }
+//       CHECK:   }
+//       CHECK:   dealloc %[[A1]] : memref<?xi8>
+  }
+//       CHECK: }
+//       CHECK: dealloc %[[A0]] : memref<?xi8>
+  return
+}

mlir/test/lib/Transforms/CMakeLists.txt

 # Exclude tests from libMLIR.so
 add_mlir_library(MLIRTestTransforms
   TestAllReduceLowering.cpp
   TestBufferPlacement.cpp
   TestCallGraph.cpp
   TestConstantFold.cpp
   TestConvertGPUKernelToCubin.cpp
   TestConvertGPUKernelToHsaco.cpp
   TestDominance.cpp
   TestLoopFusion.cpp
   TestGpuMemoryPromotion.cpp
   TestGpuParallelLoopMapping.cpp
   TestInlining.cpp
+  TestLinalgHoisting.cpp
   TestLinalgTransforms.cpp
   TestLiveness.cpp
   TestLoopMapping.cpp
   TestLoopParametricTiling.cpp
   TestLoopUnrolling.cpp
   TestOpaqueLoc.cpp
   TestMemRefBoundCheck.cpp
   TestMemRefDependenceCheck.cpp

mlir/test/lib/Transforms/TestLinalgHoisting.cpp

+//===- TestLinalgHoisting.cpp - Test Linalg hoisting functions ------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements logic for testing Linalg hoisting functions.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Linalg/IR/LinalgOps.h"
+#include "mlir/Dialect/Linalg/Transforms/Hoisting.h"
+#include "mlir/Pass/Pass.h"
+
+using namespace mlir;
+using namespace mlir::linalg;
+
+namespace {
+struct TestLinalgHoisting
+    : public PassWrapper<TestLinalgHoisting, FunctionPass> {
+  TestLinalgHoisting() = default;
+  TestLinalgHoisting(const TestLinalgHoisting &pass) {}
+
+  void runOnFunction() override;
+
+  Option<bool> testHoistViewAllocs{
+      *this, "test-hoist-view-allocs",
+      llvm::cl::desc("Test hoisting alloc used by view"),
+      llvm::cl::init(false)};
+};
+} // end anonymous namespace
+
+void TestLinalgHoisting::runOnFunction() {
+  if (testHoistViewAllocs) {
+    hoistViewAllocOps(getFunction());
+    return;
+  }
+}
+
+namespace mlir {
+void registerTestLinalgHoisting() {
+  PassRegistration<TestLinalgHoisting> testTestLinalgHoistingPass(
+      "test-linalg-hoisting", "Test Linalg hoisting functions.");
+}
+} // namespace mlir

mlir/tools/mlir-opt/mlir-opt.cpp

 void registerTestLoopPermutationPass();
 void registerTestCallGraphPass();
 void registerTestConstantFold();
 void registerTestConvertGPUKernelToCubinPass();
 void registerTestConvertGPUKernelToHsacoPass();
 void registerTestDominancePass();
 void registerTestFunc();
 void registerTestGpuMemoryPromotionPass();
+void registerTestLinalgHoisting();
 void registerTestLinalgTransforms();
 void registerTestLivenessPass();
 void registerTestLoopFusion();
 void registerTestLoopMappingPass();
 void registerTestLoopUnrollingPass();
 void registerTestMatchers();
 void registerTestMemRefDependenceCheck();
 void registerTestMemRefStrideCalculation();

 #endif
 #if MLIR_ROCM_CONVERSIONS_ENABLED
   registerTestConvertGPUKernelToHsacoPass();
 #endif
   registerTestBufferPlacementPreparationPass();
   registerTestDominancePass();
   registerTestFunc();
   registerTestGpuMemoryPromotionPass();
+  registerTestLinalgHoisting();
   registerTestLinalgTransforms();
   registerTestLivenessPass();
   registerTestLoopFusion();
   registerTestLoopMappingPass();
   registerTestLoopUnrollingPass();
   registerTestMatchers();
   registerTestMemRefDependenceCheck();
   registerTestMemRefStrideCalculation();