[PATCH] D70531: AMDGPU: Fix lit test checks with dag option

[David Stuttard via Phabricator via llvm-commits]

dstuttard marked an inline comment as done.
dstuttard added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.ll:1300
+; GCN-DAG:             v_mov_b32_e32 [[ONE:v[0-9]+]], 1.0
+; GCN-DAG-COUNT-8:     global_load_dwordx4
+; GCN-DAG-COUNT-16:    v_accvgpr_write_b32 a{{[0-9]+}}, v{{[0-9]+}}
----------------
arsenm wrote:
> dstuttard wrote:
> > rampitec wrote:
> > > AFAIK DAG-COUNT does not work.
> > Yes, you're right. It just wasn't triggering the check at all - hence passing.
> > I'll re-submit with lots of GCN-DAG checks instead, unless you can think of a better way to do it.
> Repeating -DAG doesn’t work either as far as I know. Might as well just generate this test 
Doesn't it? The other tests in this file do this.

Problem is that I'm seeing different ordering of the mov instructions - hence requiring the DAG variants.
Doesn't the -DAG here mean that FileCheck is expecting to see the starting LABEL, then 2 mov's 8 global_load_dwordx4's and 16 v_accvgpr_write_b32's in any order, followed by the v_mfma instruction?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70531/new/

https://reviews.llvm.org/D70531