[Mlir-commits] [mlir] [mlir][SPIRVToLLVM] Set valid insertion point after op erasure (PR #146551)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir-spirv

@llvm/pr-subscribers-mlir

Author: Matthias Springer (matthias-springer)

<details>
<summary>Changes</summary>

Erasing/replacing an op, which is also the current insertion point, invalidates the insertion point. Explicitly set the insertion point, so that `copy` does not crash after the One-Shot Dialect Conversion refactoring.

---
Full diff: https://github.com/llvm/llvm-project/pull/146551.diff


1 Files Affected:

- (modified) mlir/lib/Conversion/SPIRVToLLVM/ConvertLaunchFuncToLLVMCalls.cpp (+3-2) 


``````````diff
diff --git a/mlir/lib/Conversion/SPIRVToLLVM/ConvertLaunchFuncToLLVMCalls.cpp b/mlir/lib/Conversion/SPIRVToLLVM/ConvertLaunchFuncToLLVMCalls.cpp
index 0e9eb9799c3e0..2a87f78643486 100644
--- a/mlir/lib/Conversion/SPIRVToLLVM/ConvertLaunchFuncToLLVMCalls.cpp
+++ b/mlir/lib/Conversion/SPIRVToLLVM/ConvertLaunchFuncToLLVMCalls.cpp
@@ -269,8 +269,9 @@ class GPULaunchLowering : public ConvertOpToLLVMPattern<gpu::LaunchFuncOp> {
       copyInfo.push_back(info);
     }
     // Create a call to the kernel and copy the data back.
-    rewriter.replaceOpWithNewOp<LLVM::CallOp>(op, kernelFunc,
-                                              ArrayRef<Value>());
+    Operation *callOp = rewriter.replaceOpWithNewOp<LLVM::CallOp>(
+        op, kernelFunc, ArrayRef<Value>());
+    rewriter.setInsertionPointAfter(callOp);
     for (CopyInfo info : copyInfo)
       copy(loc, info.src, info.dst, info.size, rewriter);
     return success();

``````````

</details>


https://github.com/llvm/llvm-project/pull/146551