[SandboxVec] Reapply "Add barebones Region class. (#108899)" (#109059)

A `#ifndef NDEBUG` in the wrong place caused an error in release builds.

llvm/include/llvm/Transforms/Vectorize/SandboxVectorizer/Region.h

@@ -0,0 +1,104 @@
+//===- Region.h -------------------------------------------------*- 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 LLVM_TRANSFORMS_VECTORIZE_SANDBOXVECTORIZER_REGION_H
+#define LLVM_TRANSFORMS_VECTORIZE_SANDBOXVECTORIZER_REGION_H
+
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/iterator_range.h"
+#include "llvm/SandboxIR/SandboxIR.h"
+#include "llvm/Support/InstructionCost.h"
+#include "llvm/Support/raw_ostream.h"
+
+namespace llvm::sandboxir {
+
+/// The main job of the Region is to point to new instructions generated by
+/// vectorization passes. It is the unit that RegionPasses operate on with their
+/// runOnRegion() function.
+///
+/// The region allows us to stack transformations horizontally, meaning that
+/// each transformation operates on a single region and the resulting region is
+/// the input to the next transformation, as opposed to vertically, which is the
+/// common way of applying a transformation across the whole function. This
+/// enables us to check for profitability and decide whether we accept or
+/// rollback at a region granularity, which is much better than doing this at
+/// the function level.
+///
+//  Traditional approach: transformations applied vertically for the whole
+//  function
+//    F
+//  +----+
+//  |    |
+//  |    |
+//  |    | -> Transform1 ->  ... -> TransformN -> Check Cost
+//  |    |
+//  |    |
+//  +----+
+//
+//  Region-based approach: transformations applied horizontally, for each Region
+//    F
+//  +----+
+//  |Rgn1| -> Transform1 ->  ... -> TransformN -> Check Cost
+//  |    |
+//  |Rgn2| -> Transform1 ->  ... -> TransformN -> Check Cost
+//  |    |
+//  |Rgn3| -> Transform1 ->  ... -> TransformN -> Check Cost
+//  +----+
+
+class Region {
+  /// All the instructions in the Region. Only new instructions generated during
+  /// vectorization are part of the Region.
+  SetVector<Instruction *> Insts;
+
+  /// A unique ID, used for debugging.
+  unsigned RegionID = 0;
+
+  Context &Ctx;
+
+  // TODO: Add cost modeling.
+  // TODO: Add a way to encode/decode region info to/from metadata.
+
+public:
+  Region(Context &Ctx);
+  ~Region();
+
+  Context &getContext() const { return Ctx; }
+  /// Returns the region's unique ID.
+  unsigned getID() const { return RegionID; }
+
+  /// Adds I to the set.
+  void add(Instruction *I);
+  /// Removes I from the set.
+  void remove(Instruction *I);
+  /// Returns true if I is in the Region.
+  bool contains(Instruction *I) const { return Insts.contains(I); }
+  /// Returns true if the Region has no instructions.
+  bool empty() const { return Insts.empty(); }
+
+  using iterator = decltype(Insts.begin());
+  iterator begin() { return Insts.begin(); }
+  iterator end() { return Insts.end(); }
+  iterator_range<iterator> insts() { return make_range(begin(), end()); }
+
+#ifndef NDEBUG
+  /// This is an expensive check, meant for testing.
+  bool operator==(const Region &Other) const;
+  bool operator!=(const Region &other) const { return !(*this == other); }
+
+  void dump(raw_ostream &OS) const;
+  void dump() const;
+  friend raw_ostream &operator<<(raw_ostream &OS, const Region &Rgn) {
+    Rgn.dump(OS);
+    return OS;
+  }
+#endif
+};
+
+} // namespace llvm::sandboxir
+
+#endif // LLVM_TRANSFORMS_VECTORIZE_SANDBOXVECTORIZER_REGION_H

llvm/lib/Transforms/Vectorize/CMakeLists.txt

@@ -5,6 +5,7 @@ add_llvm_component_library(LLVMVectorize
   LoopVectorize.cpp
   SandboxVectorizer/DependencyGraph.cpp
   SandboxVectorizer/Passes/BottomUpVec.cpp
+  SandboxVectorizer/Region.cpp
   SandboxVectorizer/SandboxVectorizer.cpp
   SLPVectorizer.cpp
   Vectorize.cpp

llvm/lib/Transforms/Vectorize/SandboxVectorizer/Region.cpp

@@ -0,0 +1,45 @@
+//===- Region.cpp ---------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Transforms/Vectorize/SandboxVectorizer/Region.h"
+
+namespace llvm::sandboxir {
+
+Region::Region(Context &Ctx) : Ctx(Ctx) {
+  static unsigned StaticRegionID;
+  RegionID = StaticRegionID++;
+}
+
+Region::~Region() {}
+
+void Region::add(Instruction *I) { Insts.insert(I); }
+
+void Region::remove(Instruction *I) { Insts.remove(I); }
+
+#ifndef NDEBUG
+bool Region::operator==(const Region &Other) const {
+  if (Insts.size() != Other.Insts.size())
+    return false;
+  if (!std::is_permutation(Insts.begin(), Insts.end(), Other.Insts.begin()))
+    return false;
+  return true;
+}
+
+void Region::dump(raw_ostream &OS) const {
+  OS << "RegionID: " << getID() << "\n";
+  for (auto *I : Insts)
+    OS << *I << "\n";
+}
+
+void Region::dump() const {
+  dump(dbgs());
+  dbgs() << "\n";
+}
+#endif // NDEBUG
+
+} // namespace llvm::sandboxir

llvm/unittests/Transforms/Vectorize/SandboxVectorizer/CMakeLists.txt

@@ -10,4 +10,5 @@ set(LLVM_LINK_COMPONENTS
 add_llvm_unittest(SandboxVectorizerTests
   DependencyGraphTest.cpp
   LegalityTest.cpp
+  RegionTest.cpp
   )

llvm/unittests/Transforms/Vectorize/SandboxVectorizer/RegionTest.cpp

@@ -0,0 +1,81 @@
+//===- RegionTest.cpp -----------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Transforms/Vectorize/SandboxVectorizer/Region.h"
+#include "llvm/AsmParser/Parser.h"
+#include "llvm/SandboxIR/SandboxIR.h"
+#include "llvm/Support/SourceMgr.h"
+#include "gmock/gmock-matchers.h"
+#include "gtest/gtest.h"
+
+using namespace llvm;
+
+struct RegionTest : public testing::Test {
+  LLVMContext C;
+  std::unique_ptr<Module> M;
+
+  void parseIR(LLVMContext &C, const char *IR) {
+    SMDiagnostic Err;
+    M = parseAssemblyString(IR, Err, C);
+    if (!M)
+      Err.print("RegionTest", errs());
+  }
+};
+
+TEST_F(RegionTest, Basic) {
+  parseIR(C, R"IR(
+define i8 @foo(i8 %v0, i8 %v1) {
+  %t0 = add i8 %v0, 1
+  %t1 = add i8 %t0, %v1
+  ret i8 %t1
+}
+)IR");
+  llvm::Function *LLVMF = &*M->getFunction("foo");
+  sandboxir::Context Ctx(C);
+  auto *F = Ctx.createFunction(LLVMF);
+  auto *BB = &*F->begin();
+  auto It = BB->begin();
+  auto *T0 = cast<sandboxir::Instruction>(&*It++);
+  auto *T1 = cast<sandboxir::Instruction>(&*It++);
+  auto *Ret = cast<sandboxir::Instruction>(&*It++);
+  sandboxir::Region Rgn(Ctx);
+
+  // Check getters
+  EXPECT_EQ(&Ctx, &Rgn.getContext());
+  EXPECT_EQ(0U, Rgn.getID());
+
+  // Check add / remove / empty.
+  EXPECT_TRUE(Rgn.empty());
+  Rgn.add(T0);
+  EXPECT_FALSE(Rgn.empty());
+  Rgn.remove(T0);
+  EXPECT_TRUE(Rgn.empty());
+
+  // Check iteration.
+  Rgn.add(T0);
+  Rgn.add(T1);
+  Rgn.add(Ret);
+  // Use an ordered matcher because we're supposed to preserve the insertion
+  // order for determinism.
+  EXPECT_THAT(Rgn.insts(), testing::ElementsAre(T0, T1, Ret));
+
+  // Check contains
+  EXPECT_TRUE(Rgn.contains(T0));
+  Rgn.remove(T0);
+  EXPECT_FALSE(Rgn.contains(T0));
+
+#ifndef NDEBUG
+  // Check equality comparison. Insert in reverse order into `Other` to check
+  // that comparison is order-independent.
+  sandboxir::Region Other(Ctx);
+  Other.add(Ret);
+  EXPECT_NE(Rgn, Other);
+  Other.add(T1);
+  EXPECT_EQ(Rgn, Other);
+#endif
+}