[AsmWriter] Fix MIR printing of single constant LLVM IR metadata #165029
Conversation
When LLVM IR is read in and its metadata is canonicalized into Value objects, single constant metadata are not turned into MDNode objects like every other metadata. Instead, they are left as ConstantAsMetadata objects. When the ModuleSlotTracker creates slots for MDNode objects, it thus does not make one for the constant. However, during translation into MIR that constant metadata will be turned into an MDNode because that is the only Metadata object that MachineInstruction objects can house. So when MIRPrintingPass makes use of that ModuleSlotTracker, it will pass it an MDNode corresponding to that constant that never got a slot. This causes the pass to dump a pointer instead. This change fixes this issue by updating the writeAsOperandInternal function to catch the case where it cannot find the slot for the solo constant metadata node and print the constant itself directly instead of dumping the pointer. Signed-off-by: Domenic Nutile <domenic.nutile@gmail.com>
| Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
| @llvm/pr-subscribers-backend-amdgpu @llvm/pr-subscribers-llvm-ir Author: None (saxlungs) ChangesWhen LLVM IR is read in and its metadata are canonicalized into Value objects, single constant metadata are not turned into MDNode objects like every other metadata. Instead, they are left as ConstantAsMetadata objects. When the ModuleSlotTracker creates slots for MDNode objects, it thus does not make one for the constant. However, during translation into MIR that constant metadata will be turned into an MDNode because that is the only Metadata object that MachineInstruction objects can house. So when MIRPrintingPass makes use of that ModuleSlotTracker, it will pass it an MDNode corresponding to that constant that never got a slot. This causes the pass to dump a pointer instead. This change fixes this issue by updating the writeAsOperandInternal function to catch the case where it cannot find the slot for the solo constant metadata node and print the constant itself directly instead of dumping the pointer. Full diff: https://github.com/llvm/llvm-project/pull/165029.diff 2 Files Affected:
diff --git a/llvm/lib/IR/AsmWriter.cpp b/llvm/lib/IR/AsmWriter.cpp index 1096e57632d97..20264f473ad56 100644 --- a/llvm/lib/IR/AsmWriter.cpp +++ b/llvm/lib/IR/AsmWriter.cpp @@ -2810,6 +2810,12 @@ static void writeAsOperandInternal(raw_ostream &Out, const Metadata *MD, writeDILocation(Out, Loc, WriterCtx); return; } + if (N->getNumOperands() == 1) { + if (const auto *CAM = dyn_cast<ConstantAsMetadata>(N->getOperand(0))) { + writeConstantInternal(Out, CAM->getValue(), WriterCtx); + return; + } + } // Give the pointer value instead of "badref", since this comes up all // the time when debugging. Out << "<" << N << ">"; diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll index 101bb6c0ed123..149395830450d 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll @@ -6,8 +6,7 @@ define i32 @reloc_constant() { ; CHECK-LABEL: name: reloc_constant ; CHECK: bb.1 (%ir-block.0): ; CHECK-NEXT: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), !0 - ; We cannot have any specific metadata check here as ConstantAsMetadata is printed as <raw_ptr_val> - ; CHECK-NEXT: [[INT1:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), <0x{{[0-9a-f]+}}> + ; CHECK-NEXT: [[INT1:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), 4 ; CHECK-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[INT]], [[INT1]] ; CHECK-NEXT: $vgpr0 = COPY [[ADD]](s32) ; CHECK-NEXT: SI_RETURN implicit $vgpr0 |
| @llvm/pr-subscribers-llvm-globalisel Author: None (saxlungs) ChangesWhen LLVM IR is read in and its metadata are canonicalized into Value objects, single constant metadata are not turned into MDNode objects like every other metadata. Instead, they are left as ConstantAsMetadata objects. When the ModuleSlotTracker creates slots for MDNode objects, it thus does not make one for the constant. However, during translation into MIR that constant metadata will be turned into an MDNode because that is the only Metadata object that MachineInstruction objects can house. So when MIRPrintingPass makes use of that ModuleSlotTracker, it will pass it an MDNode corresponding to that constant that never got a slot. This causes the pass to dump a pointer instead. This change fixes this issue by updating the writeAsOperandInternal function to catch the case where it cannot find the slot for the solo constant metadata node and print the constant itself directly instead of dumping the pointer. Full diff: https://github.com/llvm/llvm-project/pull/165029.diff 2 Files Affected:
diff --git a/llvm/lib/IR/AsmWriter.cpp b/llvm/lib/IR/AsmWriter.cpp index 1096e57632d97..20264f473ad56 100644 --- a/llvm/lib/IR/AsmWriter.cpp +++ b/llvm/lib/IR/AsmWriter.cpp @@ -2810,6 +2810,12 @@ static void writeAsOperandInternal(raw_ostream &Out, const Metadata *MD, writeDILocation(Out, Loc, WriterCtx); return; } + if (N->getNumOperands() == 1) { + if (const auto *CAM = dyn_cast<ConstantAsMetadata>(N->getOperand(0))) { + writeConstantInternal(Out, CAM->getValue(), WriterCtx); + return; + } + } // Give the pointer value instead of "badref", since this comes up all // the time when debugging. Out << "<" << N << ">"; diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll index 101bb6c0ed123..149395830450d 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-metadata.ll @@ -6,8 +6,7 @@ define i32 @reloc_constant() { ; CHECK-LABEL: name: reloc_constant ; CHECK: bb.1 (%ir-block.0): ; CHECK-NEXT: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), !0 - ; We cannot have any specific metadata check here as ConstantAsMetadata is printed as <raw_ptr_val> - ; CHECK-NEXT: [[INT1:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), <0x{{[0-9a-f]+}}> + ; CHECK-NEXT: [[INT1:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.reloc.constant), 4 ; CHECK-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[INT]], [[INT1]] ; CHECK-NEXT: $vgpr0 = COPY [[ADD]](s32) ; CHECK-NEXT: SI_RETURN implicit $vgpr0 |
2 participants
When LLVM IR is read in and its metadata are canonicalized into Value objects, single constant metadata are not turned into MDNode objects like every other metadata. Instead, they are left as ConstantAsMetadata objects. When the ModuleSlotTracker creates slots for MDNode objects, it thus does not make one for the constant. However, during translation into MIR that constant metadata will be turned into an MDNode because that is the only Metadata object that MachineInstruction objects can house. So when MIRPrintingPass makes use of that ModuleSlotTracker, it will pass it an MDNode corresponding to that constant that never got a slot. This causes the pass to dump a pointer instead.
This change fixes this issue by updating the writeAsOperandInternal function to catch the case where it cannot find the slot for the solo constant metadata node and print the constant itself directly instead of dumping the pointer.