[CIR][MLIR] Add scf.scope lowering to standard dialects

Author: Kuree

clang/lib/CIR/Lowering/ThroughMLIR/LowerCIRToMLIR.cpp

@@ -488,12 +488,43 @@ class CIRBrOpLowering : public mlir::OpRewritePattern<mlir::cir::BrOp> {
   }
 };
 
+class CIRScopeOpLowering : public mlir::OpRewritePattern<mlir::cir::ScopeOp> {
+  using mlir::OpRewritePattern<mlir::cir::ScopeOp>::OpRewritePattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(mlir::cir::ScopeOp scopeOp,
+                  mlir::PatternRewriter &rewriter) const override {
+    // Empty scope: just remove it.
+    if (scopeOp.getRegion().empty()) {
+      rewriter.eraseOp(scopeOp);
+      return mlir::success();
+    }
+
+    for (auto &block : scopeOp.getRegion()) {
+      rewriter.setInsertionPointToEnd(&block);
+      auto *terminator = block.getTerminator();
+      rewriter.replaceOpWithNewOp<mlir::memref::AllocaScopeReturnOp>(
+          terminator, terminator->getOperands());
+    }
+
+    rewriter.setInsertionPoint(scopeOp);
+    auto newScopeOp = rewriter.create<mlir::memref::AllocaScopeOp>(
+        scopeOp.getLoc(), scopeOp.getResultTypes());
+    rewriter.inlineRegionBefore(scopeOp.getScopeRegion(),
+                                newScopeOp.getBodyRegion(),
+                                newScopeOp.getBodyRegion().end());
+    rewriter.replaceOp(scopeOp, newScopeOp);
+
+    return mlir::LogicalResult::success();
+  }
+};
+
 void populateCIRToMLIRConversionPatterns(mlir::RewritePatternSet &patterns,
                                          mlir::TypeConverter &converter) {
   patterns.add<CIRAllocaLowering, CIRLoadLowering, CIRStoreLowering,
                CIRConstantLowering, CIRUnaryOpLowering, CIRBinOpLowering,
                CIRCmpOpLowering, CIRBrOpLowering, CIRCallLowering,
-               CIRReturnLowering>(patterns.getContext());
+               CIRReturnLowering, CIRScopeOpLowering>(patterns.getContext());
   patterns.add<CIRFuncLowering>(converter, patterns.getContext());
 }
 

clang/test/CIR/Lowering/ThroughMLIR/scope.cir

@@ -0,0 +1,52 @@
+// RUN: cir-opt %s -cir-to-mlir -o - | FileCheck %s -check-prefix=MLIR
+// RUN: cir-opt %s -cir-to-mlir -cir-mlir-to-llvm -o - | mlir-translate -mlir-to-llvmir | FileCheck %s -check-prefix=LLVM
+
+module {
+  cir.func @foo() {
+    cir.scope {
+      %0 = cir.alloca i32, cir.ptr <i32>, ["a", init] {alignment = 4 : i64}
+      %1 = cir.const(4 : i32) : i32
+      cir.store %1, %0 : i32, cir.ptr <i32>
+    }
+    cir.return
+  }
+
+//      MLIR: func.func @foo()
+// MLIR-NEXT:   memref.alloca_scope
+// MLIR-NEXT:     %alloca = memref.alloca() {alignment = 4 : i64} : memref<i32>
+// MLIR-NEXT:     %c4_i32 = arith.constant 4 : i32
+// MLIR-NEXT:     memref.store %c4_i32, %alloca[] : memref<i32>
+// MLIR-NEXT:   }
+// MLIR-NEXT:   return
+
+
+//       LLVM: define void @foo()
+//  LLVM-NEXT:   %1 = call ptr @llvm.stacksave()
+//  LLVM-NEXT:   br label %2
+// LLVM-EMPTY:
+//  LLVM-NEXT: 2:
+//  LLVM-NEXT:   %3 = alloca i32, i64 1, align 4
+//  LLVM-NEXT:   %4 = insertvalue { ptr, ptr, i64 } undef, ptr %3, 0
+//  LLVM-NEXT:   %5 = insertvalue { ptr, ptr, i64 } %4, ptr %3, 1
+//  LLVM-NEXT:   %6 = insertvalue { ptr, ptr, i64 } %5, i64 0, 2
+//  LLVM-NEXT:   %7 = extractvalue { ptr, ptr, i64 } %6, 1
+//  LLVM-NEXT:   store i32 4, ptr %7, align 4
+//  LLVM-NEXT:   call void @llvm.stackrestore(ptr %1)
+//  LLVM-NEXT:   br label %8
+// LLVM-EMPTY:
+//  LLVM-NEXT: 8:
+//  LLVM-NEXT:   ret void
+//  LLVM-NEXT: }
+
+
+  // Should drop empty scopes.
+  cir.func @empty_scope() {
+    cir.scope {
+    }
+    cir.return
+  }
+  //      MLIR: func.func @empty_scope()
+  // MLIR-NEXT:   return
+  // MLIR-NEXT: }
+
+}