[RISCV] Make XFAIL test UNSUPPORTED.

[mgudim]

Currently the test cfi-multiple-location.mir is marked as XFAIL. This causes failures on some build bots because the test unexpectedly passes.

Mark this test as UNSUPPORTED for now. Later I plan to merge an MR which fixes an issue in CFIInstrInserter and this test will be enabled.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Mikhail Gudim (mgudim)

Changes

Currently the test cfi-multiple-location.mir is marked as XFAIL. This causes failures on some build bots because the test unexpectedly passes.

Mark this test as UNSUPPORTED for now. Later I plan to merge an MR which fixes an issue in CFIInstrInserter and this test will be enabled.

---

Full diff: https://github.com/llvm/llvm-project/pull/168525.diff

1 Files Affected:

• (modified) llvm/test/CodeGen/RISCV/cfi-multiple-locations.mir (+1-1)

diff --git a/llvm/test/CodeGen/RISCV/cfi-multiple-locations.mir b/llvm/test/CodeGen/RISCV/cfi-multiple-locations.mir
index 7844589e3f93c..6a447f5fff174 100644
--- a/llvm/test/CodeGen/RISCV/cfi-multiple-locations.mir
+++ b/llvm/test/CodeGen/RISCV/cfi-multiple-locations.mir
@@ -1,7 +1,7 @@
 # RUN: llc %s -mtriple=riscv64 \
 # RUN: -run-pass=cfi-instr-inserter \
 # RUN: -riscv-enable-cfi-instr-inserter=true
-# XFAIL: *
+# UNSUPPORTED: *
 
 # Technically, it is possible that a callee-saved register is saved in multiple different locations.
 # CFIInstrInserter should handle this, but currently it does not.

[jrtc27]

Why is this non-deterministic?

[mgudim]

@lenary @ppenzin alternatively, we can just merge #168531

[mgudim]

Why is this non-deterministic?

Actually I don't know why it successfully XFAIL on some buildbots, but not on others. When the test runs llc will hit an assertion in the code. Maybe there is a difference in what happens next?

[mgudim]

I ran the test locally and it worked as I expected. I suspect it will be a lot of work to see why exactly it fails on some buildbots. Since this is only XFAIL temporarily, I hope it would be OK just to make this UNSUPPORTED.

[github-actions]

🐧 Linux x64 Test Results

• 186287 tests passed
• 4854 tests skipped

[lenary]

Approving so this can fix the bots.

I presume the issue is some bots have asserts enabled and some don't

[mgudim]

Approving so this can fix the bots.

I presume the issue is some bots have asserts enabled and some don't

@lenary oh, right. I guess the proper fix would have been to add "requires asserts".

[dyung]

Approving so this can fix the bots.
I presume the issue is some bots have asserts enabled and some don't

@lenary oh, right. I guess the proper fix would have been to add "requires asserts".

This is probably the correct fix. I also saw this test failure on my bot that specifically builds without asserts: https://lab.llvm.org/buildbot/#/builders/202/builds/4233

[topperc]

Approving so this can fix the bots.
I presume the issue is some bots have asserts enabled and some don't

@lenary oh, right. I guess the proper fix would have been to add "requires asserts".

This is probably the correct fix. I also saw this test failure on my bot that specifically builds without asserts: https://lab.llvm.org/buildbot/#/builders/202/builds/4233

The correct fix is to change the llvm_unreachable to a reportFatalError. If there is an input that can reach an llvm_unreachable, then it is not "unreachable"