[PATCH] D112890: headers: optionalise some generated resource headers

Nico Weber via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Mon Dec 13 04:02:28 PST 2021


thakis added a comment.

In D112890#3119897 <https://reviews.llvm.org/D112890#3119897>, @jrtc27 wrote:

> This change seems pretty counter to Clang's ability to spit out IR for anything regardless of what backends you have. The right fix should be to shrink the headers, not mask the problem by changing the principles of how Clang works.

For what it's worth, I agree with this. "RISC-V did it already" seems like a weak justification given relative popularity of RISC-V and ARM.

What's more, I haven't seen the intrinsic header generation take up a lot of time, and 10 MB isn't a lot of storage.

(Intrinsic _tests_ on the other hand did take quite a lot of time on RISC-V, which is why they got disabled if the RISC-V is off. This doesn't help bots much, since bots need to enable all backends. We had a discussion about this when the intrinsics got added, but in the end we sadly just accepted a big increase in test time instead of figuring out some less expensive approach.)

Anyways, I agree with @jrtc27 about this change.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112890/new/

https://reviews.llvm.org/D112890



More information about the cfe-commits mailing list