You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It was pointed out in #1212 that some ops (e.g. LiOp) will lower to have signed types.
This makes little sense, since the RISC-V instruction set itself does not attach sign semantics to the values in its registers. This is the exact reason why LLVM dialect does not use signs, and why using signs in arith does not even verify in mlir.
Therefore I feel like all signs (e.g. si32 types) in riscv dialect should be omitted, since sign semantics are preserved in the riscv ops themselves anyway.
The text was updated successfully, but these errors were encountered:
It was pointed out in #1212 that some ops (e.g.
LiOp
) will lower to have signed types.This makes little sense, since the RISC-V instruction set itself does not attach sign semantics to the values in its registers. This is the exact reason why LLVM dialect does not use signs, and why using signs in arith does not even verify in mlir.
Therefore I feel like all signs (e.g.
si32
types) in riscv dialect should be omitted, since sign semantics are preserved in the riscv ops themselves anyway.The text was updated successfully, but these errors were encountered: