[CIR][Lowering] Fix loop lowering for top-level break/continue (llvm#349)

This PR fixes a couple of corner cases connected with the `YieldOp`
lowering in loops.
Previously, in llvm#211 we introduced `lowerNestedBreakContinue` but we
didn't check that `YieldOp` may belong to the same region, i.e. it is
not nested, e.g.
```
while(1) {
   break;
}
```
Hence the error `op already replaced`. 

Next, we fix `yield` lowering for `ifOp` and `switchOp` but didn't cover
`scopeOp`, and the same error occurred. This PR fixes this as well.

I added two tests - with no checks actually, just to make sure no more
crashes happen.

fixes llvm#324

Author: gitoleg

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

@@ -408,8 +408,15 @@ class CIRLoopOpLowering : public mlir::OpConversionPattern<mlir::cir::LoopOp> {
   lowerNestedBreakContinue(mlir::Region &loopBody, mlir::Block *exitBlock,
                            mlir::Block *continueBlock,
                            mlir::ConversionPatternRewriter &rewriter) const {
+    // top-level yields are lowered in matchAndRewrite
+    auto isNested = [&](mlir::Operation *op) {
+      return op->getParentRegion() != &loopBody;
+    };
 
     auto processBreak = [&](mlir::Operation *op) {
+      if (!isNested(op))
+        return mlir::WalkResult::advance();
+
       if (isa<mlir::cir::LoopOp, mlir::cir::SwitchOp>(
               *op)) // don't process breaks in nested loops and switches
         return mlir::WalkResult::skip();
@@ -421,6 +428,9 @@ class CIRLoopOpLowering : public mlir::OpConversionPattern<mlir::cir::LoopOp> {
     };
 
     auto processContinue = [&](mlir::Operation *op) {
+      if (!isNested(op))
+        return mlir::WalkResult::advance();
+
       if (isa<mlir::cir::LoopOp>(
               *op)) // don't process continues in nested loops
         return mlir::WalkResult::skip();
@@ -490,7 +500,10 @@ class CIRLoopOpLowering : public mlir::OpConversionPattern<mlir::cir::LoopOp> {
 
     // Branch from body to condition or to step on for-loop cases.
     rewriter.setInsertionPoint(bodyYield);
-    rewriter.replaceOpWithNewOp<mlir::cir::BrOp>(bodyYield, &stepBlock);
+    auto bodyYieldDest = bodyYield.getKind() == mlir::cir::YieldOpKind::Break
+                             ? continueBlock
+                             : &stepBlock;
+    rewriter.replaceOpWithNewOp<mlir::cir::BrOp>(bodyYield, bodyYieldDest);
 
     // Is a for loop: branch from step to condition.
     if (kind == LoopKind::For) {
@@ -822,11 +835,15 @@ class CIRScopeOpLowering
     // Stack restore before leaving the body region.
     rewriter.setInsertionPointToEnd(afterBody);
     auto yieldOp = cast<mlir::cir::YieldOp>(afterBody->getTerminator());
-    auto branchOp = rewriter.replaceOpWithNewOp<mlir::cir::BrOp>(
-        yieldOp, yieldOp.getArgs(), continueBlock);
 
-    // // Insert stack restore before jumping out of the body of the region.
-    rewriter.setInsertionPoint(branchOp);
+    if (!isLoopYield(yieldOp)) {
+      auto branchOp = rewriter.replaceOpWithNewOp<mlir::cir::BrOp>(
+          yieldOp, yieldOp.getArgs(), continueBlock);
+
+      // // Insert stack restore before jumping out of the body of the region.
+      rewriter.setInsertionPoint(branchOp);
+    }
+
     // TODO(CIR): stackrestore?
     // rewriter.create<mlir::LLVM::StackRestoreOp>(loc, stackSaveOp);
 

clang/test/CIR/Lowering/loop.cir

@@ -217,4 +217,82 @@ module {
   // MLIR-NEXT:     llvm.br ^bb6
   // MLIR-NEXT:   ^bb6:
   // MLIR-NEXT:     llvm.return
+
+  // test corner case 
+  // while (1) {
+  //     break;
+  // }
+  cir.func @whileCornerCase() {
+    cir.scope {
+      cir.loop while(cond : {
+        %0 = cir.const(#cir.int<1> : !s32i) : !s32i
+        %1 = cir.cast(int_to_bool, %0 : !s32i), !cir.bool
+        cir.brcond %1 ^bb1, ^bb2
+      ^bb1:  // pred: ^bb0
+        cir.yield continue
+      ^bb2:  // pred: ^bb0
+        cir.yield
+      }, step : {
+        cir.yield
+      }) {
+        cir.yield break
+      }
+    }
+    cir.return
+  }
+  //      MLIR: llvm.func @whileCornerCase()
+  //      MLIR:   %0 = llvm.mlir.constant(1 : i32) : i32
+  // MLIR-NEXT:   %1 = llvm.mlir.constant(0 : i32) : i32
+  // MLIR-NEXT:   %2 = llvm.icmp "ne" %0, %1 : i32
+  // MLIR-NEXT:   %3 = llvm.zext %2 : i1 to i8
+  // MLIR-NEXT:   %4 = llvm.trunc %3 : i8 to i
+  // MLIR-NEXT:   llvm.cond_br %4, ^bb3, ^bb4
+  // MLIR-NEXT: ^bb3:  // pred: ^bb2
+  // MLIR-NEXT:   llvm.br ^bb5
+  // MLIR-NEXT: ^bb4:  // pred: ^bb2
+  // MLIR-NEXT:   llvm.br ^bb6
+  // MLIR-NEXT: ^bb5:  // pred: ^bb3
+  // MLIR-NEXT:   llvm.br ^bb6
+  // MLIR-NEXT: ^bb6:  // 2 preds: ^bb4, ^bb5
+  // MLIR-NEXT:   llvm.br ^bb7
+  // MLIR-NEXT: ^bb7:  // pred: ^bb6
+  // MLIR-NEXT:   llvm.return
+
+  // test corner case - no fails during the lowering
+  // for (;;) {
+  //     break;
+  // }
+  cir.func @forCornerCase() {
+    cir.scope {
+      cir.loop for(cond : {
+        cir.yield continue
+      }, step : {
+        cir.yield
+      }) {
+        cir.scope {
+          cir.yield break
+        }
+        cir.yield
+      }
+    }
+    cir.return
+  }
+//      MLIR: llvm.func @forCornerCase()
+//      MLIR:     llvm.br ^bb1
+// MLIR-NEXT:   ^bb1:  // pred: ^bb0
+// MLIR-NEXT:     llvm.br ^bb2
+// MLIR-NEXT:   ^bb2:  // 2 preds: ^bb1, ^bb6
+// MLIR-NEXT:     llvm.br ^bb3
+// MLIR-NEXT:   ^bb3:  // pred: ^bb2
+// MLIR-NEXT:     llvm.br ^bb4
+// MLIR-NEXT:   ^bb4:  // pred: ^bb3
+// MLIR-NEXT:     llvm.br ^bb7
+// MLIR-NEXT:   ^bb5:  // no predecessors
+// MLIR-NEXT:     llvm.br ^bb6
+// MLIR-NEXT:   ^bb6:  // pred: ^bb5
+// MLIR-NEXT:     llvm.br ^bb2
+// MLIR-NEXT:   ^bb7:  // pred: ^bb4
+// MLIR-NEXT:     llvm.br ^bb8
+// MLIR-NEXT:   ^bb8:  // pred: ^bb7
+// MLIR-NEXT:     llvm.return
 }