kazutakahirata wrote: > LGTM. Is DXIL the only remaining user? (We also have > Instruction::isDebugOrPseudoInst(), which should (probably separately) be > replaced with `isa<PseudoProbeInst>()` or removed (with preference to > removal).)
Thanks for the review! Yes. In production `llvm/` code, the DirectX backend is the only remaining caller of `Module::convertFromNewDbgValues()` in `DXILDebugInfo.cpp`, which materializes `DbgRecord`s back into debug intrinsics right before serializing to LLVM 3.7-based DXIL bitcode. Outside of DirectX, the only remaining references to `Dbg*Intrinsic` / `Intrinsic::dbg_*` are: - The bitcode/IR reader upgrade path: - `AutoUpgrade.cpp` - `BitcodeReader.cpp` - `LLParser.cpp` - Dead debug-intrinsic handling in subprojects: - `mlir/lib/Target/LLVMIR/ModuleImport.cpp` - `lldb/source/Expression/IRInterpreter.cpp` - `StripDebugDeclarePass` in `StripSymbols.cpp`, which only checks for `Intrinsic::dbg_declare` and is now a no-op - `LLVM_FOR_EACH_VALUE_SUBCLASS` in `llvm-c/Core.h` - Unit tests: - `BitReaderTest.cpp` - `CloningTest.cpp` - `DebugInfoTest.cpp` - `MetadataTest.cpp` Agreed on `Instruction::isDebugOrPseudoInst()`. I'll follow up with a separate patch to replace its call sites with `isa<PseudoProbeInst>()` and remove `Instruction::isDebugOrPseudoInst()`. https://github.com/llvm/llvm-project/pull/231379 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
