Change Details

For generic intrinsics, Intrinsic::getDeclaration expects a unique declaration for each unique set of parameter types. This means, for each set of unique parameter types, we need a unique mangled name. Currently this is not the case, as unnamed types do not have a name, so different unnamed types will get mangled to the same name. This causes a crash when using PredicateInfo with the test case attached. This patch introduces an additional map to LLVMContextImpl, which keeps track of IDs to use when mangling for unnamed types. It is populated on demand when mangling. I am not sure if this is the best approach, but it seems to do the job.

Currently unnamed types cause problems for overloaded intrinsics like ssa_copy, because different unnamed types get mangled to the same string. This patch introduces an additional map to LLVMContextImpl, which keeps track of IDs to use when mangling for unnamed types. The IDs are assigned at type creation time. For textual IR, those types are created by order of appearance in the LL file. This change also changes the bitcode writer to write unnamed types in the order they were created, before other types. This ensures IDs are assigned in the same fashion in both cases.

For generic intrinsics, Intrinsic::getDeclaration expects a uniqueCurrently unnamed types cause problems for overloaded intrinsics like declaration for each unique set of parameter types. This meansssa_copy, for eachbecause different unnamed types get mangled to the same set of unique parameter types, we need a unique mangled namestring. Currently this is not the case, as unnamed types do not have a nameThis patch introduces an additional map to LLVMContextImpl, sowhich keeps different unnamed types will gettrack of IDs to use when mangled to the same nameing for unnamed types. This causes aThe IDs are crash when using PredicateInfo with the test case attached. This patch introduces an additional map to LLVMContextImpl,assigned at type creation time. For textual IR, those types are created by order of appearance in the LL file. which keepsThis change also changes the bitcode writer to write unnamed types in the order they were created, track of IDs to use when mangling for unnamedbefore other types. It is populated onThis ensures IDs are assigned in the same fashion in demand when mangling. I am not sure if this is the best approach, but itboth cases. seems to do the job.