Resolve implicit trunc TODOs

[nikic]

#106524 introduces a consistency assertion in the APInt constructor (currently still disabled by default), which verifies that the passed value is indeed a signed/unsigned value of the indicated bit width.

To allow this assertion to be enabled by default, a number of // TODO: Avoid implicit trunc? comments were left behind, in cases where it was not immediately clear whether/how the implicit truncation in the APInt constructor could be avoided.

Long term, these should be resolved in one three ways:

• Determine that the implicit truncation is actually desirable/acceptable in the given context. Replace the TODO with a comment explaining why.
• Change the code to avoid passing values that require implicit truncation.
• Make an API change, e.g. by propagating the implicitTrunc parameter up, so that some users can use it and some not. Or by accepting an APInt instead of a plain integer and leaving the proper construction to callers.

[llvmbot]

@llvm/issue-subscribers-mlir

Author: Nikita Popov (nikic)

https://github.com//pull/106524 introduces a consistency assertion in the APInt constructor (currently still disabled by default), which verifies that the passed value is indeed a signed/unsigned value of the indicated bit width.

To allow this assertion to be enabled by default, a number of // TODO: Avoid implicit trunc? comments were left behind, in cases where it was not immediately clear whether/how the implicit truncation in the APInt constructor could be avoided.

Long term, these should be resolved in one three ways:

• Determine that the implicit truncation is actually desirable/acceptable in the given context. Replace the TODO with a comment explaining why.
• Change the code to avoid passing values that require implicit truncation.
• Make an API change, e.g. by propagating the implicitTrunc parameter up, so that some users can use it and some not. Or by accepting an APInt instead of a plain integer and leaving the proper construction to callers.