[llvm] [GlobalISel][AArch64] Legalize G_FABS and G_FNEG for SVE (PR #114784)

[Amara Emerson via llvm-commits]

Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>,
Thorsten =?utf-8?q?Schütt?= <schuett at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/114784 at github.com>


aemerson wrote:

Let me elaborate further. For example, you wrote 7 hours ago:
> First month ping.

... and since this is your PR that you want review on, it's incumbent on you to do that work. It's one of the challenges of working with open source software in that much of the work is about building consensus, evaluating engineering trade-offs and exploring not only the "what" or "how" but the "why". On the flip side, it's also a valuable exercise in finding problematic approaches *before* we invest potentially 100s/1000s of person-hours of effort in a project.

Many things in LLVM follow the same pattern. When we introduced scalable types all those years ago, Paul, myself and the rest of the SVE team went through a bunch of discussions internally at ARM to find what the right approach *could* be. We did a lot of work to have a prototype implementation, but even then once we present to the community it took a long time to actually build consensus about how we should approach the problem. In the mean time the design changed and it was on us to actually do that investigation work since we were the ones trying to add support in LLVM. We didn't just pawn it off to the "community" and send in patches.

SVE in GlobalISel is a much smaller project than general SVE, but if the approach is not yet clear it still needs someone to do that work so that the reviewers can understand the *why* as well.

https://github.com/llvm/llvm-project/pull/114784