I downloaded it from here:
https://www.epicgames.com/unrealtournament/forums/showthread.php?12011-Unreal-Tournament-Pre-Alpha-Playable-Build

And followed instructions here:
https://www.gamingonlinux.com/articles/unreal-tournament-on-linux-checking-up-on-the-progress-by-epic-games.7241/

I'm running
4.15.0-3315666+++UT+Release-Next 510 0
which seems to be
"UPDATE: New build posted 1/26/2016 - 0.1.7.1"

This continues to crash on LLVM 7-devel and 18.1-devel as of today. Tested with latests Linux build [1].

[1] https://www.epicgames.com/unrealtournament/forums/unreal-tournament-discussion/announcements/387158-0-1-11-update-posted-5-16-2017

(In reply to Timothy Arceri from comment #2)
> This continues to crash on LLVM 7-devel and 18.1-devel as of today. Tested
> with latests Linux build [1].
> 
> [1]
> https://www.epicgames.com/unrealtournament/forums/unreal-tournament-
> discussion/announcements/387158-0-1-11-update-posted-5-16-2017

FWIW that's not the latest.

https://www.epicgames.com/unrealtournament/forums/unreal-tournament-discussion/announcements

0.1.12.1 on tonga current llvm/mesa does segfault on start for me with
R600_DEBUG=sisched

I guess for some unknown reason, for one instruction being scheduled, getInstr() returns NULL, and that causes the crash in isLowLatencyInstruction. Not sure why it happens specifically with this game, I will have to look at the intermediate compiler results.

Similar error for me. Trying to run "Deus Ex Mankind Divided" - later on 16.04, now on 18.04 with same error.
HW: xeon 2665 in hp z420 mobo
SW: mesa 18.04, llvm 6.0, linux 4.16.12
Thread 6 "si_shader:1" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffcf7fe700 (LWP 30129)]
0x00007fffd7f54c00 in llvm::SIInstrInfo::isLowLatencyInstruction(llvm::MachineInstr const&) const () from /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
(gdb) bt
#0  0x00007fffd7f54c00 in llvm::SIInstrInfo::isLowLatencyInstruction(llvm::MachineInstr const&) const () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#1  0x00007fffd7f842ac in llvm::SIScheduleDAGMI::moveLowLatencies() () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#2  0x00007fffd7f8f088 in llvm::SIScheduleDAGMI::schedule() () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#3  0x00007fffd6e2487f in  () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#4  0x00007fffd6e2e3f7 in  () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#5  0x00007fffd6dbbfe0 in llvm::MachineFunctionPass::runOnFunction(llvm::Function&) () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#6  0x00007fffd6bf17f8 in llvm::FPPassManager::runOnFunction(llvm::Function&) () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#7  0x00007fffd78ab1b8 in  () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#8  0x00007fffd6bf108f in llvm::legacy::PassManagerImpl::run(llvm::Module&) () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#9  0x00007fffd7d435b5 in  () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#10 0x00007fffd7d43799 in LLVMTargetMachineEmitToMemoryBuffer () at /usr/lib/x86_64-linux-gnu/libLLVM-6.0.so.1
#11 0x00007fffdaeff264 in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#12 0x00007fffdaef4cdc in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#13 0x00007fffdaef654c in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#14 0x00007fffdaf12a50 in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#15 0x00007fffdab1bc0e in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#16 0x00007fffdab1b8a7 in  () at /usr/lib/x86_64-linux-gnu/dri/radeonsi_dri.so
#17 0x00007ffff476f6db in start_thread (arg=0x7fffcf7fe700) at pthread_create.c:463
#18 0x00007fffede5e88f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

I has modified isLowLatencyInstruction to that:

bool SIInstrInfo::isLowLatencyInstruction(const MachineInstr &MI) const {
  if(&MI != nullptr){
    std::thread::id this_id = std::this_thread::get_id();
    std::cout << getpid() << this_id << " -- Got valid pointer &MI=" << &MI << "\n";
    unsigned Opc = MI.getOpcode();
    printf("Got Opc = %u\n", Opc);
    return isSMRD(Opc);
  } else {
    printf("Got NULL pointer MI\n");
    return false;
  }
}

And got no nullptr passed, but found accessing same pointer from 2 different threads by the same time. 

0x2ce50x7f54094a9700 -- Got valid pointer &MI=0x7f53f03c0d70
Got Opc = 8188
0x2ce50x7f5403fff700 -- Got valid pointer &MI=0x7f53f03c0d70
Got Opc = 8188

All I found - error happens while access to member Desc of class MCInstrInfo. 
isLowLatencyInstruction -> isSMRD(Opcode) -> MCInstrInfo::get(Opcode) -> return Desc[Opcode]

Opcode value is not meaning - it fails even if replace Desc[Opcode] with Desc[0]. But "return *(new MCInstrDesc())" instead of "return Desc[Opcode]" make it working (nuget is only for isSMRD call):

  const MCInstrDesc &get(unsigned Opcode) const {
    assert(Opcode < NumOpcodes && "Invalid opcode!");
    return Desc[Opcode];
  }
  MCInstrDesc &nuget(unsigned Opcode) const {
    assert(Opcode < NumOpcodes && "Invalid opcode!");
    assert(Desc != NULL && "Desc is NULL!");
    assert(Desc != nullptr && "Desc is nullptr!");
    return *(new MCInstrDesc());
  }

of course, it is not a solution, also ambient occlusion isn't working correctly with this.
Declaring Desc as std::vector<MCInstrDesc>, init it with new and fill with Desc->assign(D, D+NO) - also gives nothing, game fails the same way.

Do you have by any chance already extracted the llvm code for the shader that triggers the issue ?

I extracted the shader and looked at the debug info when it compiles.

My understanding is that the bug is triggered when the llvm shader contains a COPY operand as last operation.
I'm not sure why this behaviour is triggered, but a fix is to ignore in this case the successor (which is ExitSU, not valid):

Replace in SIScheduleDAGMI::moveLowLatencies:
    } else if (SU->getInstr()->getOpcode() == AMDGPU::COPY) {
      bool CopyForLowLat = false;
      for (SDep& SuccDep : SU->Succs) {
        SUnit *Succ = SuccDep.getSUnit();
        if (SITII->isLowLatencyInstruction(*Succ->getInstr())) {
          CopyForLowLat = true;
        }
      }


by

    } else if (SU->getInstr()->getOpcode() == AMDGPU::COPY) {
      bool CopyForLowLat = false;
      for (SDep& SuccDep : SU->Succs) {
        SUnit *Succ = SuccDep.getSUnit();
        if (Succ->NodeNum >= DAGSize)
          continue;
        if (SITII->isLowLatencyInstruction(*Succ->getInstr())) {
          CopyForLowLat = true;
        }
      }

The affected shader seems to compile fine with the change.

(In reply to Axel Davy from comment #10)

Thank you, this fix made DX:MD working for me.

I've posted the patch for review last week:
https://reviews.llvm.org/D47984

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/1256.