[PATCH] D80370: [CodeGen] Make CreateStackTemporary choose better alignments

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

See also D29810 <https://reviews.llvm.org/D29810> and D28920 <https://reviews.llvm.org/D28920>. For AMDGPU there's only pain by using a stack object alignment > 4, so completely ignoring the ABI alignment for stack temporaries is a good option. The stack alignment is chosen to be higher to accommodate common types needing a higher ABI alignment



================
Comment at: llvm/test/CodeGen/AMDGPU/scratch-simple.ll:30
+; GCN-DAG: v_add{{_|_nc_}}{{i|u}}32_e32 [[HI_OFF:v[0-9]+]],{{.*}} 0x280, [[CLAMP_IDX]]
+; GCN-DAG: v_add{{_|_nc_}}{{i|u}}32_e32 [[LO_OFF:v[0-9]+]],{{.*}} {{v2|0x80}}, [[CLAMP_IDX]]
 
----------------
I don't understand the new check. This should be either a register with a materialized register, or a constant. Did the total stack usage here increase?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80370/new/

https://reviews.llvm.org/D80370