[llvm-branch-commits] [llvm] [AMDGPU] Add FUNC_WAVE32 for object linking info flag (PR #207633)

Krzysztof Drewniak via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Fri Nov 20 15:00:12 PST 2026


================
@@ -3250,6 +3250,7 @@ if needed.
      0x2   ``FUNC_USES_FLAT_SCRATCH``  Function uses flat scratch addressing
      0x4   ``FUNC_HAS_DYN_STACK``      Function has dynamic stack allocation
      0x8   ``FUNC_WGP_MODE``           Function uses WGP execution mode
+     0x10  ``FUNC_WAVE32``             Function uses wave32; otherwise wave64
----------------
krzysz00 wrote:

Yeah - usually, if you don't specify a wave size, you get the target default, and so  0 would mean wave64 on a GCN or something else that doesn't have a wave32, and then wave32 on gfx10+ ... so it'd be something like `FUNC_ALTERNATE_WAVESIZE` maybe?

Or at least that's my thought on this discussion

https://github.com/llvm/llvm-project/pull/207633


More information about the llvm-branch-commits mailing list