[Mlir-commits] [mlir] [mlir][vector] Fix invalid IR in `RewriteBitCastOfTruncI` (PR #78146)

[Mehdi Amini]

joker-eph wrote:

> Also, this is perfectly valid in LLVM

Which part?
As far as I can tell, the MLIR error is for:

> %48 = "arith.trunci"(%47) : (vector<3xi16>) -> vector<3xi16>

And looking at some equivalent in LLVM I see:

>error: invalid cast opcode for cast from 'i16' to 'i16'
>  %r8 = trunc i16 undef to i16

or: 

> error: invalid cast opcode for cast from '<3 x i16>' to '<3 x i16>'
>  %r8 = trunc <3 x i16> undef to <3 x i16>



https://github.com/llvm/llvm-project/pull/78146