[PATCH] D130096: [Clang][AMDGPU] Emit AMDGPU library control constants in clang

[Joseph Huber via Phabricator via cfe-commits]

jhuber6 added a comment.

In D130096#3850628 <https://reviews.llvm.org/D130096#3850628>, @arsenm wrote:

> In D130096#3850550 <https://reviews.llvm.org/D130096#3850550>, @b-sumner wrote:
>
>> There's the "small matter" of implementing the new device library functions.  Why is all that more likeable than two kinds of control constants?
>
> Different functions providing different behaviors can be handled at link time like any other function, instead of the same functions providing different behaviors per translation unit and requires cloning. The current scheme transfers complexity from the device library build system into the driver and user binaries

Another benefit of this would be that linking could be done only once in a sound manner rather than requiring an instance of the ROCm device libraries to be included for each TU. Although we would probably still need the attribute propagation that `-mlink-builtin-bitcode` offers, so it wouldn't be quite as easy that throwing the device libs into the `lld` invocation.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130096/new/

https://reviews.llvm.org/D130096