[PATCH] D110684: [RISCV] Define _m intrinsics as builtins, instead of macros.

[Craig Topper via Phabricator via cfe-commits]

craig.topper added a comment.

In D110684#3059698 <https://reviews.llvm.org/D110684#3059698>, @aeubanks wrote:

> In D110684#3057498 <https://reviews.llvm.org/D110684#3057498>, @nikic wrote:
>
>> I noticed this a few time in the past already, but this is the worst one yet:
>>
>> Compile-time: https://llvm-compile-time-tracker.com/compare.php?from=6641d29b70993bce6dbd7e0e0f1040753d38842f&to=97f0c63783f52389bd8842df205379ceade7a89d&stat=instructions
>> Max-rss: https://llvm-compile-time-tracker.com/compare.php?from=6641d29b70993bce6dbd7e0e0f1040753d38842f&to=97f0c63783f52389bd8842df205379ceade7a89d&stat=max-rss
>> Plus a >1% increase in clang total binary size.
>>
>> This happens every time a new slew of RISCV intrinsic permutations gets added, which seem to suffer from an extreme degree of redundancy. Is it possible to do something about this?
>
> +1, it seems bad to regress overall Clang memory usage by 1% by adding some intrinsics, especially when we're not targeting RISCV.

Is the max-rss increase due to the binary size increase?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110684/new/

https://reviews.llvm.org/D110684