[llvm-commits] Question about /llvm/trunk/lib/CodeGen/MachineScheduler.cpp
Sergei Larin
slarin at codeaurora.org
Mon Feb 6 15:53:50 PST 2012
Andrew,
Asserts are "on", nothing is caught. Debug output offers a single clue -
the BB in question has a non-jump terminator (STrib == store byte).
I guess my question then becomes - can a non control flow instruction serve
as BB terminator (the implicit assumption I was referring to)? If not then
I'll try to find where the "proper" terminator was lost.
Thanks.
Sergei
--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum.
> -----Original Message-----
> From: Andrew Trick [mailto:atrick at apple.com]
> Sent: Monday, February 06, 2012 5:32 PM
> To: Sergei Larin
> Cc: llvm-commits at cs.uiuc.edu
> Subject: Re: Question about
> /llvm/trunk/lib/CodeGen/MachineScheduler.cpp
>
> Hi Sergei,
>
> Please make sure you run with assertions enabled and look at -debug-
> only=misched output. -verify-machineinstrs may also be useful.
>
> I'm not sure what the problem is, but crashing is bad. If there are any
> implicit assumptions, then they should be made explicit.
>
> I'm not sure I understand the CFG output. A block should end in a
> terminator. BB#7 claims to have a successor, but I don't know what it's
> terminator is.
>
> -Andy
>
> On Feb 6, 2012, at 10:56 AM, Sergei Larin <slarin at codeaurora.org>
> wrote:
>
> >
> > Andrew,
> >
> > This is something between a bug report and a question. I do not
> provide
> > enough context to reproduce the failure, but hope it is fresh enough
> in your
> > memory :)
> >
> > In lib/CodeGen/MachineScheduler.cpp
> >
> > In bool MachineScheduler::runOnMachineFunction(MachineFunction &mf)
> {}
> >
> > While iterating over following MBB, for the following fragment
> > ...
> > BB#6: derived from LLVM BB %sw.bb6, ADDRESS TAKEN
> > Predecessors according to CFG: BB#1
> > %vreg10<def> = TFRI 1; IntRegs:%vreg10
> > STrib %vreg1<kill>, 0, %vreg10<kill>; mem:ST1[%Enum_Ref_Par]
> > IntRegs:%vreg1,%vreg10
> > JMP <BB#8>
> > Successors according to CFG: BB#8
> >
> > BB#7: derived from LLVM BB %sw.bb7, ADDRESS TAKEN
> > Predecessors according to CFG: BB#1
> > %vreg9<def> = TFRI 2; IntRegs:%vreg9
> > STrib %vreg1<kill>, 0, %vreg9<kill>; mem:ST1[%Enum_Ref_Par]
> > IntRegs:%vreg1,%vreg9
> > Successors according to CFG: BB#8
> >
> > BB#8: derived from LLVM BB %sw.epilog, ADDRESS TAKEN
> > Predecessors according to CFG: BB#0 BB#1 BB#7 BB#6 BB#5 BB#4 BB#2
> > JMPR %PC<imp-def,dead>, %R31<imp-use>, %R0<imp-use,undef>
> > ...
> >
> > BB#7 seems to cause a problem when iterating over it with RegionEnd
> pointing
> > somewhere wrong. Am I breaking some implicit assumption in BB
> formation?
> > Want MBB->end() suppose to return for the BB#7?
> >
> > *** Final schedule ***
> >
> > MachineScheduling Proc_6:BB#7
> > From: %vreg9<def> = TFRI 2; IntRegs:%vreg9
> > To:
> > Program received signal SIGSEGV, Segmentation fault.
> > 0x0000000001a0c814 in llvm::MachineOperand::isReg (this=0x7) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/include/llvm/CodeGen/
> > MachineOperand.h:204
> > 204 bool isReg() const { return OpKind == MO_Register; }
> > (gdb) bt
> > #0 0x0000000001a0c814 in llvm::MachineOperand::isReg (this=0x7) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/include/llvm/CodeGen/
> > MachineOperand.h:204
> > #1 0x00000000021ca720 in llvm::MachineInstr::print (this=0x3e2c9f0,
> OS=...,
> > TM=0x0) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/CodeGen/MachineIn
> > str.cpp:1462
> > #2 0x0000000001a970fe in llvm::operator<< (OS=..., MI=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/include/llvm/CodeGen/
> > MachineInstr.h:934
> > #3 0x00000000022e4c18 in (anonymous
> > namespace)::MachineScheduler::runOnMachineFunction (this=0x3dcc1c0,
> mf=...)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/CodeGen/MachineSc
> > heduler.cpp:297
> > #4 0x00000000021c41e5 in llvm::MachineFunctionPass::runOnFunction
> > (this=0x3dcc1c0, F=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/CodeGen/MachineFu
> > nctionPass.cpp:33
> > #5 0x0000000002789db6 in llvm::FPPassManager::runOnFunction
> > (this=0x3dcb820, F=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/VMCore/PassManage
> > r.cpp:1498
> > #6 0x000000000278a001 in llvm::FPPassManager::runOnModule
> (this=0x3dcb820,
> > M=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/VMCore/PassManage
> > r.cpp:1520
> > #7 0x000000000278a344 in llvm::MPPassManager::runOnModule
> (this=0x3dc8ec0,
> > M=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/VMCore/PassManage
> > r.cpp:1574
> > #8 0x000000000278a85a in llvm::PassManagerImpl::run (this=0x3dc8b70,
> M=...)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/VMCore/PassManage
> > r.cpp:1658
> > #9 0x000000000278ab55 in llvm::PassManager::run (this=0x3d7f140,
> M=...) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/lib/VMCore/PassManage
> > r.cpp:1687
> > #10 0x0000000000c9488d in (anonymous
> > namespace)::EmitAssemblyHelper::EmitAssembly (this=0x7fffffffa2e0,
> > Action=clang::Backend_EmitAssembly, OS=0x3d13660)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/CodeG
> > en/BackendUtil.cpp:497
> > #11 0x0000000000c94968 in clang::EmitBackendOutput (Diags=...,
> CGOpts=...,
> > TOpts=..., LOpts=..., M=0x3d1a3c0,
> Action=clang::Backend_EmitAssembly,
> > OS=0x3d13660)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/CodeG
> > en/BackendUtil.cpp:509
> > #12 0x0000000000c9042c in
> clang::BackendConsumer::HandleTranslationUnit
> > (this=0x3d123a0, C=...)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/CodeG
> > en/CodeGenAction.cpp:155
> > #13 0x0000000000e3cf61 in clang::ParseAST (S=..., PrintStats=false)
> at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/Parse
> > /ParseAST.cpp:110
> > #14 0x0000000000b00725 in clang::ASTFrontendAction::ExecuteAction
> > (this=0x3cf1740) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/Front
> > end/FrontendAction.cpp:414
> > #15 0x0000000000c8ef9c in clang::CodeGenAction::ExecuteAction
> > (this=0x3cf1740) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/CodeG
> > en/CodeGenAction.cpp:407
> > #16 0x0000000000b00377 in clang::FrontendAction::Execute
> (this=0x3cf1740) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/Front
> > end/FrontendAction.cpp:334
> > #17 0x0000000000ad8e63 in clang::CompilerInstance::ExecuteAction
> > (this=0x3ced2d0, Act=...)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/Front
> > end/CompilerInstance.cpp:653
> > #18 0x0000000000aaa8e4 in clang::ExecuteCompilerInvocation
> (Clang=0x3ced2d0)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/lib/Front
> > endTool/ExecuteCompilerInvocation.cpp:175
> > #19 0x0000000000a99618 in cc1_main (ArgBegin=0x7fffffffaf10,
> > ArgEnd=0x7fffffffb108, Argv0=0x3cebaf8
> > "/prj/qct/sunray-austin/scratch/slarin/qdsp6_hex/bin/qc/bin/clang",
> > MainAddr=0xaa4654)
> > at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/tools/dri
> > ver/cc1_main.cpp:165
> > #20 0x0000000000aa5e77 in main (argc_=65, argv_=0x7fffffffc058) at
> > /local/mnt/workspace/slarin/tools/llvm-mainline-
> merged/tools/clang/tools/dri
> > ver/driver.cpp:353
> >
> >
> > Thanks.
> >
> > Sergei
> >
More information about the llvm-commits
mailing list