[Mlir-commits] [mlir] [mlir][Tensor] Retain discardable attrs in pack(cast) folder (PR #115772)
Mehdi Amini
llvmlistbot at llvm.org
Wed Nov 13 13:07:05 PST 2024
joker-eph wrote:
> An ability for an optimization to leave information for subsequent optimizations to pick up.
To be clear: an optimization **can** do that. You can define some discardable attributes that the optimizations knows about and knows how to preserve through transformations, that's perfectly fine. But this isn't the same as "blind" propagation through transformations.
> but I dont see why just drop the attribute is any better compared to just keep the attribute. Both are equally wrong
I don't think so: one is a miscompile / correctness problem, the other is a conservative / safe options. These are called *discardable* attributes for a reason, and the reason is so that the options are absolutely not "equally wrong".
https://github.com/llvm/llvm-project/pull/115772
More information about the Mlir-commits
mailing list