[llvm-dev] Are AMDGPU intrinsics available in LLVM IR ?
Reid Kleckner via llvm-dev
llvm-dev at lists.llvm.org
Tue Apr 14 18:22:41 PDT 2020
On Mon, Apr 13, 2020 at 8:43 AM Matt Arsenault via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> These are the definitions, and the standard place for targets to define
> intrinsics. NVPTX is odd in defining intrinsics in the backend. These are
> usable like any other IR intrinsic, through the generated
> Intrinsic::amdgcn_workitem_id_* enums.
This is slightly off topic, but I wanted to point out that intrinsics are
now generated into separate headers (IntrinsicsARM.h, IntrinsicsAMDGPU.h,
...). This opens the door to moving the intrinsics into the backends and
compiling them out when a particular target is disabled. There is obviously
a lot more work to do since there are many mid-level IR transforms that
depend on these headers, but we could sink that logic into targets if we
were motivated to do it.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-dev