[PATCH] D124125: [NVPTX][tests] add "XFAIL: nvptx" for some tests

Wed Apr 20 15:11:47 PDT 2022

tra accepted this revision.
tra added a comment.
This revision is now accepted and ready to land.

Looks OK, though on a closer look not all the disabled tests are the fault of NVPTX. If those tests are easy to fix, I'd prefer that.
If not, I'm fine with disabling them for NVPTX, they don't seem to be particularly interesting.

================
Comment at: llvm/test/CodeGen/Generic/addr-label.ll:6

+; NVPTX unexpectedly emitted of the global var, while the function is not emitted as expected
+; XFAIL: nvptx
----------------
Wrong description?
AFAICT it fails with a crash on failure to select https://godbolt.org/z/cdfE3bo4W:
`Cannot select: 0x5608e9078c70: i64 = BlockAddress<@test1b, %test_label> 0`

================
Comment at: llvm/test/CodeGen/Generic/inline-asm-mem-clobber.ll:6

+; NVPTX output seems valid, but not what the test expects
+; XFAIL: nvptx
----------------
This test should've used `CHECK-DAG` as the order in which code/data is emitted is not guaranteed.
It should also have used `G` when it was checking for the variable name.  `@G` is only used in  a comment in the x86 assembly.

================
Comment at: llvm/test/CodeGen/Generic/live-debug-label.ll:3
 ;
+; NVPTX produces a different order of the BBs
+; XFAIL: nvptx
----------------
It's the test problem -- it should not have assumed basic block order. 

================
Comment at: llvm/test/CodeGen/Generic/live-debug-label.ll:41-44
 ; CHECK: bye.thread21:
 ; CHECK: DBG_LABEL !14
 ; CHECK: if.then5:
 ; CHECK: DBG_LABEL !14
----------------
Something like this would probably work better:
```
; CHECK: {{bye\.thread21|if\.then5}}:
; CHECK: DBG_LABEL !14
; Skip to the next label.
; CHECK: {{^\s+bb\.}}
; CHECK: {{bye\.thread21|if\.then5}}:
; CHECK: DBG_LABEL !14
; Skip to the next label.
; CHECK: {{^\s+bb\.}}
; CHECK-NOT: DBG_LABEL !14

```

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D124125/new/

https://reviews.llvm.org/D124125