[PATCH] D64567: AMDGPU/GISel: Add support for vector shader inputs

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUCallingConv.td:64
+
+  CCIfNotInReg<CCIfType<[v2i32, v2f32], CCCustom<"allocateVGPRTuple">>>
 ]>;
----------------
I think this should be handled by decomposing the incoming types into 32-bit pieces, and repacking. We do this for functions in the DAG and it's less constraining on the lowering. This also avoids SGPR alignment, since they will be repacked if necessary


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64567/new/

https://reviews.llvm.org/D64567