Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
| llvm/lib/Target/AMDGPU/SMInstructions.td | ||
|---|---|---|
| 142 | Looks like we have a use case for classes in multiclasses? | |
| llvm/lib/Target/AMDGPU/SMInstructions.td | ||
|---|---|---|
| 1143 | Is the _COMMON variant defined for aesthetic reasons? I believe the following code should work: def : MnemonicAlias<!cast<SM_Load_Pseudo>(ps#"_IMM").Mnemonic, opName>, | |
| llvm/lib/Target/AMDGPU/SMInstructions.td | ||
|---|---|---|
| 1143 | The aim here is to make the code reflect the fact that the pseudos share the same mnemonic (and I'm not sure casting to SM_Load_Pseudo instead of SM_Pseudo makes it obvious enough). This introduces a pseudo that is not used as such, but that seems a result of using pseudos to describe instruction properties -- a problem that we might want to address separately at some point. | |
| llvm/lib/Target/AMDGPU/SMInstructions.td | ||
|---|---|---|
| 1143 | This is a matter of taste, but for me the _COMMON definition does not make code clearer because the purpose of this definition is not immediately obvious. Maybe the name is a bit misleading and something like _MNEMO would be better? Another (more important) problem is that the _COMMON definition if not for free - it contaminates compiler tables with unused entries. | |
Looks like we have a use case for classes in multiclasses?