Matt Arsenault
c198f775cd
AMDGPU: Remove flat/global fmin/fmax intrinsics ( #105642 )
...
These have been replaced with atomicrmw
2024-10-09 09:27:28 +04:00
Matt Arsenault
9dca83f2e1
AMDGPU: Add noalias.addrspace metadata when autoupgrading atomic intrinsics ( #102599 )
...
This will be needed to continue generating the raw instruction in the flat case.
2024-10-08 00:13:28 +04:00
Matt Arsenault
ee08d9cba5
AMDGPU: Remove global/flat atomic fadd intrinics ( #97051 )
...
These have been replaced with atomicrmw.
2024-08-22 23:27:33 +04:00
Matt Arsenault
9d364286f3
AMDGPU: Remove flat/global atomic fadd v2bf16 intrinsics ( #97050 )
...
These are now fully covered by atomicrmw.
2024-08-21 14:26:42 +04:00
Matt Arsenault
f55bcc5dbe
AMDGPU: Add amdgpu.no.fine.grained.memory when upgrading old atomic intrinsics ( #89655 )
...
This should replicate the old intrinsic behavior better when codegen of
the raw instruction will require metadata in the future.
2024-06-27 19:52:23 +02:00
Matt Arsenault
4477ff6836
AMDGPU: Remove ds_fmin/ds_fmax intrinsics ( #96739 )
...
These have been replaced with atomicrmw.
2024-06-27 15:35:24 +02:00
Matt Arsenault
70c8b9c24a
AMDGPU: Remove ds atomic fadd intrinsics ( #95396 )
...
These have been replaced with atomicrmw fadd
2024-06-23 10:30:20 +02:00
Matt Arsenault
204a417d51
AutoUpgrade: Use syncscope("agent") atomic.inc/dec intrinsic upgrade
...
The old syncscope parameter never really worked correctly, but
effectively gave "workgroup" scope. Use something faster than system
but more correct than before.
https://reviews.llvm.org/D157389
2023-08-10 17:38:25 -04:00
Matt Arsenault
92ee60b66f
AMDGPU: Drop and upgrade llvm.amdgcn.atomic.inc/dec to atomicrmw
2023-06-21 21:20:26 -04:00
