[test] XFAIL two tests with inlining debug info issues on Sparc

[ro]

Summary

Currently only two test failures remain on Sparc, both sparcv9-sun-solaris2.11 and sparc64-unknown-linux-gnu:

LLVM :: DebugInfo/Generic/debug-label-inline.ll
LLVM :: Linker/subprogram-linkonce-weak.ll

They seem related in that debug info isn't related for inlined functions when LLVM does on other targets.

I've filed separate bugs for both files (Bug 47129 and 47131).

This patch XFAILs the tests.

Tested on sparcv9-sun-solaris2.11 and amd64-pc-solaris2.11.

[dblaikie]

did this regress recently? Did sparc start using instruction bundles recently? (I've commented on the bugs that it looks like they're both caused to the same bundling of add+ret)

If it was recenty, I'd push back on the bundling change & suggest this needs to be fixed rather than XFAILed. If it's been broken for ages - how? who's working on sparc that they didn't notice the breakage when they committed the bundling changes in the first place?

But I guess if it's been broken for a while, it's reasonable to commit this patch & have followup via bugs.

[aprantl]

LGTM + what @dblaikie said :-)

[ro]

In D85827#2214085, @dblaikie wrote:

did this regress recently? Did sparc start using instruction bundles recently? (I've commented on the bugs that it looks like they're both caused to the same bundling of add+ret)

I've just checked: the debug-label-inline.ll has already been broken when it was introduced in LLVM 8. The subprogram-linkonce-weak.ll test OTOH had worked in LLVM 6 and 8, but started to fail with 9.

AFAICT the Sparc target has been using bundles already in LLVM 3.9; introduced at some point after 3.4.

If it was recenty, I'd push back on the bundling change & suggest this needs to be fixed rather than XFAILed. If it's been broken for ages - how? who's working on sparc that they didn't notice the breakage when they committed the bundling changes in the first place?

Good question about the Sparc port: it seems not to have seen much love recently. My current goal is to get the Solaris/Sparc port to a point where there are no unaccounted failures and the buildbot can be moved back to the regular buildmaster so new failures are noticed quickly.

But I guess if it's been broken for a while, it's reasonable to commit this patch & have followup via bugs.

Good. Thanks for your detective work.

llvm/test/DebugInfo/Generic/debug-label-inline.ll

 ; RUN: llc -O0 -filetype=obj -o - %s | llvm-dwarfdump -v - | FileCheck %s
 ;
+; Bug 47129
+; XFAIL: sparc
+;
 ; CHECK: .debug_info contents:
 ; CHECK: [[LABEL_ORIGIN:0x[0-9a-zA-Z]+]]:{{ *}}DW_TAG_label
 ; CHECK-NEXT: DW_AT_name [DW_FORM_strp] {{.*}}"top"
 ; CHECK-NEXT: DW_AT_decl_file [DW_FORM_data1] {{.*}}debug-label-inline.c
 ; CHECK-NEXT: DW_AT_decl_line [DW_FORM_data1] {{.*}}8
 ; CHECK: DW_TAG_label
 ; CHECK-NEXT: DW_AT_abstract_origin [DW_FORM_ref4] {{.*}}{[[LABEL_ORIGIN]]} "top"
 ; CHECK-NEXT: DW_AT_low_pc [DW_FORM_addr] {{.*}}{{0x[0-9a-f]+}}

llvm/test/Linker/subprogram-linkonce-weak.ll

 ; RUN: llvm-link %s %S/Inputs/subprogram-linkonce-weak.ll -S -o %t1
 ; RUN: FileCheck %s -check-prefix=LW -check-prefix=CHECK <%t1
 ; RUN: llvm-link %S/Inputs/subprogram-linkonce-weak.ll %s -S -o %t2
 ; RUN: FileCheck %s -check-prefix=WL -check-prefix=CHECK <%t2
 ; REQUIRES: default_triple
-
+;
+; Bug 47131
+; XFAIL: sparc
+;
 ; This testcase tests the following flow:
 ;  - File A defines a linkonce version of @foo which has inlined into @bar.
 ;  - File B defines a weak version of @foo (different definition).
 ;  - Linkage rules state File B version of @foo wins.
 ;  - @bar still has inlined debug info related to the linkonce @foo.
 ;
 ; This should fix PR22792, although the testcase was hand-written.  There's a
 ; RUN line with a crasher for llc at the end with checks for the DWARF output.