[AMDGPU] EXPENSIVE_CHECKS failure: attr-amdgpu-flat-work-group-size-vgpr-limit.ll

[RKSimon]

CodeGen/AMDGPU/attr-amdgpu-flat-work-group-size-vgpr-limit.ll is failing on EXPENSIVE_CHECKS builds:

Command Output (stdout):
--
$ ":" "RUN: at line 3"
$ "e:\llvm\ninja\bin\llc.exe" "-march=amdgcn" "-mcpu=gfx900" "-amdgpu-function-calls=false" "-enable-misched=false" "-sgpr-regalloc=fast" "-vgpr-regalloc=fast"
# command stderr:

# After SI lower SGPR spill instructions
# Machine code for function f512: NoPHIs, TracksLiveness, TiedOpsRewritten, TracksDebugUserValues
Frame Objects:
  fi#0: id=1 dead
Function Live Ins: $vgpr0 in %0, $vgpr1 in %1, $vgpr2 in %2, $sgpr0_sgpr1 in %3, $sgpr2_sgpr3 in %4, $sgpr4_sgpr5 in %5, $sgpr6_sgpr7 in %6, $sgpr8 in %7, $sgpr9 in %8, $sgpr10 in %9

*** Bad machine code: Instruction index out of order ***
- function:    f512
- basic block: %bb.0  (0x142d6b9ea48) [0B;8592B)
- instruction: 80B      $sgpr0 = V_READLANE_B32 $vgpr40, 0, implicit-def $sgpr0_sgpr1
Last instruction was at 112B
LLVM ERROR: Found 1 machine code errors.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: e:\\llvm\\ninja\\bin\\llc.exe -march=amdgcn -mcpu=gfx900 -amdgpu-function-calls=false -enable-misched=false -sgpr-regalloc=fast -vgpr-regalloc=fast
1.      Running pass 'CallGraph Pass Manager' on module '<stdin>'.
2.      Running pass 'Verify generated machine code' on function '@f512'
Exception Code: 0x80000003
 #0 0x00007ff7bdcf6365 HandleAbort E:\llvm\llvm-project\llvm\lib\Support\Windows\Signals.inc:414:0
 #1 0x00007ff9fd521881 (C:\Windows\System32\ucrtbase.dll+0x71881)
 #2 0x00007ff9fd522851 (C:\Windows\System32\ucrtbase.dll+0x72851)
 #3 0x00007ff7bdc266b0 llvm::report_fatal_error(class llvm::Twine const &, bool) E:\llvm\llvm-project\llvm\lib\Support\ErrorHandling.cpp:123:0
 #4 0x00007ff7bce7a501 `anonymous namespace'::MachineVerifierPass::runOnMachineFunction E:\llvm\llvm-project\llvm\lib\CodeGen\MachineVerifier.cpp:311:0
 #5 0x00007ff7bcd1cf76 llvm::MachineFunctionPass::runOnFunction(class llvm::Function &) E:\llvm\llvm-project\llvm\lib\CodeGen\MachineFunctionPass.cpp:85:0
 #6 0x00007ff7bd1b9516 llvm::FPPassManager::runOnFunction(class llvm::Function &) E:\llvm\llvm-project\llvm\lib\IR\LegacyPassManager.cpp:1430:0
 #7 0x00007ff7bc9433a6 `anonymous namespace'::CGPassManager::RunPassOnSCC E:\llvm\llvm-project\llvm\lib\Analysis\CallGraphSCCPass.cpp:179:0
 #8 0x00007ff7bc942e48 `anonymous namespace'::CGPassManager::RunAllPassesOnSCC E:\llvm\llvm-project\llvm\lib\Analysis\CallGraphSCCPass.cpp:479:0
 #9 0x00007ff7bc944d46 `anonymous namespace'::CGPassManager::runOnModule E:\llvm\llvm-project\llvm\lib\Analysis\CallGraphSCCPass.cpp:542:0
#10 0x00007ff7bd1b9c3d `anonymous namespace'::MPPassManager::runOnModule E:\llvm\llvm-project\llvm\lib\IR\LegacyPassManager.cpp:1545:0
#11 0x00007ff7bd1b9167 llvm::legacy::PassManagerImpl::run(class llvm::Module &) E:\llvm\llvm-project\llvm\lib\IR\LegacyPassManager.cpp:536:0
#12 0x00007ff7bac4dd27 compileModule E:\llvm\llvm-project\llvm\tools\llc\llc.cpp:736:0

[llvmbot]

@llvm/issue-subscribers-backend-amdgpu

[RKSimon]

@LuoYuanke Its looks to be related to https://reviews.llvm.org/D131825

https://lab.llvm.org/buildbot/#/builders/104/builds/8348

[RKSimon]

Closing - D131825 was reverted by 853bb192c407f5d9e75a5fd55cc089151530cbd3 and recommitted at 5159be3c9bb0fc9c1d64fab1a8e1fc3640917e0d

• e438ce5694048058a93ead01c7b00e8301aa5719 added "-verify-machineinstrs" checks so non-EXPENSIVE_CHECKS builds would still see any future regression

[LuoYuanke]

Sorry, I forgot to close this issue. Thank @RKSimon for tracing this issue.