[XRay] [compiler-rt] Mark arg1 logging test as failing on !x86_64.

[pelikan]

Summary

rL297000 and rL297003 implemented arg1 logging on amd64 and made
other architectures build. We need to blacklist the new test as well.

[sdardis]

I'm seeing this test fail on x86_64 debian linux; http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-debian-fast/builds/2390/

Should this test be temporarily XFAIL'd for all targets for the moment?

[pelikan]

Apparently the Debian failure is due to http://bugs.llvm.org/show_bug.cgi?id=32148 - you can see it failing on the verbosity=1 expected CHECK printf in all tests, not just arg1-related. It's unfortunate, but unrelated to this.

[dberris]

@sdardis -- that looks like something else, let me have a look. The unit test directory definitely shouldn't be empty.

[dberris]

@timshen -- do you know whether this is the correct architecture for powerpc? Does this need to be powerpc64el?

[timshen]

Should be powerpc64le for ppc.

[pelikan]

• fix powerpc64le to actually work

(is it worth renaming the files so that they match the names used within?)

[sdardis]

test/xray/TestCases/Linux/arg1-logger.cc

12 ↗

(On Diff #90659)

I believe this may require 'mipsel || mips64el' as well.

[pelikan]

• fix XFAIL according to what greps out of the LLVM repo

[pelikan]

test/xray/TestCases/Linux/arg1-logger.cc

12 ↗

(On Diff #90659)

Not according to grep -r. I can see "powerpc64", "mips", "target-is-mips64" and "target-is-mips64el" and then triples. In the LLVM repo, however, I can see "powerpc64-", "mips-" and "mips64-". Fixed according to that.

[dberris]

test/xray/TestCases/Linux/arg1-logger.cc

12 ↗

(On Diff #90659)

Please look in compiler-rt, because the settings are different (these are deduced differently from the CMake set-up).

[pelikan]

• follow compiler-rt XFAIL grep key words, not LLVM's

test/xray/TestCases/Linux/arg1-logger.cc

12 ↗

(On Diff #90659)

Fair enough. I've put the keywords I was able to grep, still not knowing where do they come from.

[pelikan]

• finally use something that will work based on LIT's unit tests

test/xray/TestCases/Linux/arg1-logger.cc

12 ↗

(On Diff #90659)

timshen went digging and found BooleanExpression.py in lit's directory. That has a unit test matching a triple for prefixes witch should work. Therefore, leaving "mips || powerpc" as it's the simplest and most obvious.

Thanks, Tim, for all the effort!

[dberris]

For compiler-rt, the architecture is determined by CMake and is used to inform the lit config in test/lit.common.configured.in -- so you'll find the list of architectures in compiler-rt that are enumerated in cmake/config-ix.cmake. There's a list of architectures there, and you'll see things like the following spelled: mipsel, mips64el, powerpc64, powerpc64le. You then need to know which architectures the XRay tests are actually enabled, and where they're running. As far as I can tell they are in cmake/config-ix.cmake as well. The list is:

set(ALL_XRAY_SUPPORTED_ARCH ${X86_64} ${ARM32} ${ARM64} ${MIPS32} ${MIPS64} powerpc64le)

Does this make it clearer?

[dberris]

I'll land this to see whether it stops the bleeding on PPC at least.

compiler-rt/trunk/test/xray/TestCases/Linux/arg1-logger.cc

 // Check that we can get the first function argument logged
 // using a custom logging function.
 //
 // RUN: %clangxx_xray -std=c++11 %s -o %t
 // RUN: XRAY_OPTIONS="patch_premain=true verbosity=1 xray_logfile_base=arg1-logger-" %run %t 2>&1 | FileCheck %s
 //
 // After all that, clean up the XRay log file.
 //
 // RUN: rm arg1-logger-*
 //
 // At the time of writing, the ARM trampolines weren't written yet.
-// XFAIL: arm || aarch64
+// XFAIL: arm || aarch64 || mips
 // See the mailing list discussion of r296998.
 // UNSUPPORTED: powerpc64le
 
 #include "xray/xray_interface.h"
 
 #include <cinttypes>
 #include <cstdio>