[RISCV] Merge addi into load/store as there is a ADD between them

[HsiangKai]

Summary

This patch adds peephole optimizations for the following patterns:

(load (add base, (addi src, off1)), off2)
   -> (load (add base, src), off1+off2)
(store val, (add base, (addi src, off1)), off2)
   -> (store val, (add base, src), off1+off2)

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2102

I don't think this is safe for W instructions. We would need to prove the sign extend done by the ADD*W doesn't alter the result of the addition.

[HsiangKai]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2102

You are right. I didn't think about it carefully. I will remove ADD*W in the conditions.

[HsiangKai]

It is not safe to do this peephole for ADD*W.

[HsiangKai]

Update test cases.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2173

I'm a little concerned about the use of UpdateNodeOperands in this function. If the update matches an existing node, it will delete the original node and return a new node. If that happens here, Add will no longer point to a valid node after the call.

I'm going to do some experiments.

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2173

As I feared, this test

define i32 @lw_far_local(i32* %a)  {                                             
  %1 = getelementptr inbounds i32, i32* %a, i64 4095                             
  %2 = load volatile i32, i32* %1                                                
  %3 = getelementptr inbounds i32, i32* %a, i64 4094                             
  %4 = load volatile i32, i32* %3                                                
  %5 = mul i32 %2, %4                                                            
  ret i32 %5                                                                     
}

miscompiles into this with -O0. I needed -O0 to get CodeGenPrepare to not rewrite the GEPs.

mv      a1, a0
lui     a2, 4
addi    a0, a2, -4
add     a0, a1, a0
lw      a0, -4(a0) <- the offset for this is (4 << 12) + -4 + -4 + a1 making it the same offset for the next load. It should be (4 << 12) + -4 + a1
add     a1, a1, a2
lw      a1, -8(a1)
mul     a0, a0, a1
ret

This appears to be fixable by changing.

if (Add)
  CurDAG->UpdateNodeOperands(Add.getNode(), Add.getOperand(AddBaseIdx),
                             Base.getOperand(0));

to

if (Add)
  Add = SDValue(CurDAG->UpdateNodeOperands(Add.getNode(), Add.getOperand(AddBaseIdx),
                                           Base.getOperand(0)), 0);

But I'm not sure that's the only way this can break.

[HsiangKai]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2173

Thanks for your investigation. It looks like UpdateNodeOperands() will return the existing node and it does not update the operands as we expected here.

Does it mean the following other UpdateNodeOperands() has similar problem?

[HsiangKai]

Take the return value of UpdateNodeOperands().

[craig.topper]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

2181

Drop the curly braces here.

2188

And here.

[HsiangKai]

Address comments.

[craig.topper]

LGTM

[nathanchance]

I bisected a crash while building the Linux kernel for RISC-V to this change:

# bad: [68ee5ad0082c0a72ff16a6571c9db7054afd6ea3] [flang] Update Google Doc link for Flang Biweekly Sync call notes
# good: [981ed72a17e4302dfd77ac54d742c08dfb6b35bb] [NFC][SCEV] Refactor `createNodeForSelectViaUMinSeq()` out of `createNodeForSelectOrPHIViaUMinSeq()`
git bisect start '68ee5ad0082c0a72ff16a6571c9db7054afd6ea3' '981ed72a17e4302dfd77ac54d742c08dfb6b35bb'
# bad: [f8463da4a329b839cfd01d7f80ae72e18f3c061e] [lldb] Allow EXE or exe in toolchain-msvc.test
git bisect bad f8463da4a329b839cfd01d7f80ae72e18f3c061e
# bad: [cf90233a67eb513930880ff1b49e80af18de5d5c] [InstCombine] Add additional test for icmp of two ranges (NFC)
git bisect bad cf90233a67eb513930880ff1b49e80af18de5d5c
# bad: [eaca933c59fd61b3df4697b5fae0eeec67acfaeb] [Clang][CodeGen]Fix __builtin_dump_struct missing record type field name
git bisect bad eaca933c59fd61b3df4697b5fae0eeec67acfaeb
# good: [52ce95a1a55424256f0d56e32392396896ed7f76] [NFC] Prevent shadowing a variable declared in `if`
git bisect good 52ce95a1a55424256f0d56e32392396896ed7f76
# good: [78582194361e4454e95d3af23367b81dd63b7943] [asan] Enable detect_stack_use_after_return=1 by default on Linux
git bisect good 78582194361e4454e95d3af23367b81dd63b7943
# bad: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them
git bisect bad c62b014db97953bee25d17b46730f05eea6a3430
# good: [9fc58f1820e3b0883c9759945bca1aa1f1df8e8d] [PowerPC] Support of ppc_fp128 in lowering of llvm.is_fpclass
git bisect good 9fc58f1820e3b0883c9759945bca1aa1f1df8e8d
# first bad commit: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them

I see there was a related follow up change in D124693 but that change does not resolve the issue.

A simplified reproducer:

struct autofs_packet_missing {
  int len;
  char name[]
};
autofs_notify_daemon_wq_0, autofs_notify_daemon_type;
memcpy();
autofs_wait() {
  struct autofs_packet_missing pkt;
  switch (autofs_notify_daemon_type)
  case 0: {
    struct autofs_packet_missing *mp = &pkt;
    memcpy(mp->name);
    mp->name[autofs_notify_daemon_wq_0] = '\0';
  }
}

$ clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
...
clang: /home/nathan/cbl/src/llvm-project/llvm/include/llvm/CodeGen/MachineInstr.h:506: llvm::MachineOperand &llvm::MachineInstr::getOperand(unsigned int): Assertion `i < getNumOperands() && "getOperand() out of range!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
1.      <eof> parser at end of file
2.      Code generation
3.      Running pass 'Function Pass Manager' on module 'waitq.i'.
4.      Running pass 'Prologue/Epilogue Insertion & Frame Finalization' on function '@autofs_wait'
 #0 0x0000aaaac0ce0e8c llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34e0e8c)
 #1 0x0000aaaac0cdf098 llvm::sys::RunSignalHandlers() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34df098)
 #2 0x0000aaaac0c64214 (anonymous namespace)::CrashRecoveryContextImpl::HandleCrash(int, unsigned long) CrashRecoveryContext.cpp:0:0
 #3 0x0000aaaac0c643c4 CrashRecoverySignalHandler(int) CrashRecoveryContext.cpp:0:0
 #4 0x0000ffff9d84d854 (linux-vdso.so.1+0x854)
 #5 0x0000ffff9d3626a8 __pthread_kill_implementation (/lib64/libc.so.6+0x826a8)
 #6 0x0000ffff9d31ab40 gsignal (/lib64/libc.so.6+0x3ab40)
 #7 0x0000ffff9d3070f8 abort (/lib64/libc.so.6+0x270f8)
 #8 0x0000ffff9d314208 __assert_fail_base (/lib64/libc.so.6+0x34208)
 #9 0x0000ffff9d314270 __assert_perror_fail (/lib64/libc.so.6+0x34270)
#10 0x0000aaaabfd84c00 llvm::RISCVRegisterInfo::eliminateFrameIndex(llvm::MachineInstrBundleIterator<llvm::MachineInstr, false>, int, unsigned int, llvm::RegScavenger*) const RISCVRegisterInfo.cpp:0:0
#11 0x0000aaaac040db24 (anonymous namespace)::PEI::replaceFrameIndices(llvm::MachineBasicBlock*, llvm::MachineFunction&, int&) PrologEpilogInserter.cpp:0:0
#12 0x0000aaaac040b5d4 (anonymous namespace)::PEI::runOnMachineFunction(llvm::MachineFunction&) PrologEpilogInserter.cpp:0:0
#13 0x0000aaaac0187f0c llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2987f0c)
#14 0x0000aaaac060cdbc llvm::FPPassManager::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0cdbc)
#15 0x0000aaaac06146d8 llvm::FPPassManager::runOnModule(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e146d8)
#16 0x0000aaaac060d708 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0d708)
#17 0x0000aaaac1433268 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3c33268)
#18 0x0000aaaac178f150 clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) CodeGenAction.cpp:0:0
#19 0x0000aaaac1fb094c clang::ParseAST(clang::Sema&, bool, bool) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x47b094c)
#20 0x0000aaaac16d49e4 clang::FrontendAction::Execute() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3ed49e4)
#21 0x0000aaaac1662438 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3e62438)
#22 0x0000aaaac17895ac clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3f895ac)
#23 0x0000aaaabfb36754 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2336754)
#24 0x0000aaaabfb34b80 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) driver.cpp:0:0
#25 0x0000aaaac153d6e4 void llvm::function_ref<void ()>::callback_fn<clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const::$_1>(long) Job.cpp:0:0
#26 0x0000aaaac0c640bc llvm::CrashRecoveryContext::RunSafely(llvm::function_ref<void ()>) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34640bc)
#27 0x0000aaaac153d1a4 clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d3d1a4)
#28 0x0000aaaac150aa0c clang::driver::Compilation::ExecuteCommand(clang::driver::Command const&, clang::driver::Command const*&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0aa0c)
#29 0x0000aaaac150ac78 clang::driver::Compilation::ExecuteJobs(clang::driver::JobList const&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0ac78)
#30 0x0000aaaac151de28 clang::driver::Driver::ExecuteCompilation(clang::driver::Compilation&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d1de28)
#31 0x0000aaaabfb34310 main (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2334310)
#32 0x0000ffff9d3073c8 __libc_start_call_main (/lib64/libc.so.6+0x273c8)
#33 0x0000ffff9d3074a0 __libc_start_main@GLIBC_2.17 (/lib64/libc.so.6+0x274a0)
#34 0x0000aaaabfb31d30 _start (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2331d30)
clang-15: error: clang frontend command failed with exit code 134 (use -v to see invocation)
ClangBuiltLinux clang version 15.0.0 (https://github.com/llvm/llvm-project 02aa795785379b34f1f82d1c4d852b915c5bfb4a)
Target: riscv64-unknown-linux-gnu
Thread model: posix
InstalledDir: /home/nathan/tmp/build/llvm/stage1/bin
clang-15: note: diagnostic msg: Error generating preprocessed source(s) - no preprocessable inputs.

[craig.topper]

In D124231#3483638, @nathanchance wrote:

I bisected a crash while building the Linux kernel for RISC-V to this change:

# bad: [68ee5ad0082c0a72ff16a6571c9db7054afd6ea3] [flang] Update Google Doc link for Flang Biweekly Sync call notes
# good: [981ed72a17e4302dfd77ac54d742c08dfb6b35bb] [NFC][SCEV] Refactor `createNodeForSelectViaUMinSeq()` out of `createNodeForSelectOrPHIViaUMinSeq()`
git bisect start '68ee5ad0082c0a72ff16a6571c9db7054afd6ea3' '981ed72a17e4302dfd77ac54d742c08dfb6b35bb'
# bad: [f8463da4a329b839cfd01d7f80ae72e18f3c061e] [lldb] Allow EXE or exe in toolchain-msvc.test
git bisect bad f8463da4a329b839cfd01d7f80ae72e18f3c061e
# bad: [cf90233a67eb513930880ff1b49e80af18de5d5c] [InstCombine] Add additional test for icmp of two ranges (NFC)
git bisect bad cf90233a67eb513930880ff1b49e80af18de5d5c
# bad: [eaca933c59fd61b3df4697b5fae0eeec67acfaeb] [Clang][CodeGen]Fix __builtin_dump_struct missing record type field name
git bisect bad eaca933c59fd61b3df4697b5fae0eeec67acfaeb
# good: [52ce95a1a55424256f0d56e32392396896ed7f76] [NFC] Prevent shadowing a variable declared in `if`
git bisect good 52ce95a1a55424256f0d56e32392396896ed7f76
# good: [78582194361e4454e95d3af23367b81dd63b7943] [asan] Enable detect_stack_use_after_return=1 by default on Linux
git bisect good 78582194361e4454e95d3af23367b81dd63b7943
# bad: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them
git bisect bad c62b014db97953bee25d17b46730f05eea6a3430
# good: [9fc58f1820e3b0883c9759945bca1aa1f1df8e8d] [PowerPC] Support of ppc_fp128 in lowering of llvm.is_fpclass
git bisect good 9fc58f1820e3b0883c9759945bca1aa1f1df8e8d
# first bad commit: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them

I see there was a related follow up change in D124693 but that change does not resolve the issue.

A simplified reproducer:

struct autofs_packet_missing {
  int len;
  char name[]
};
autofs_notify_daemon_wq_0, autofs_notify_daemon_type;
memcpy();
autofs_wait() {
  struct autofs_packet_missing pkt;
  switch (autofs_notify_daemon_type)
  case 0: {
    struct autofs_packet_missing *mp = &pkt;
    memcpy(mp->name);
    mp->name[autofs_notify_daemon_wq_0] = '\0';
  }
}

$ clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
...
clang: /home/nathan/cbl/src/llvm-project/llvm/include/llvm/CodeGen/MachineInstr.h:506: llvm::MachineOperand &llvm::MachineInstr::getOperand(unsigned int): Assertion `i < getNumOperands() && "getOperand() out of range!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
1.      <eof> parser at end of file
2.      Code generation
3.      Running pass 'Function Pass Manager' on module 'waitq.i'.
4.      Running pass 'Prologue/Epilogue Insertion & Frame Finalization' on function '@autofs_wait'
 #0 0x0000aaaac0ce0e8c llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34e0e8c)
 #1 0x0000aaaac0cdf098 llvm::sys::RunSignalHandlers() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34df098)
 #2 0x0000aaaac0c64214 (anonymous namespace)::CrashRecoveryContextImpl::HandleCrash(int, unsigned long) CrashRecoveryContext.cpp:0:0
 #3 0x0000aaaac0c643c4 CrashRecoverySignalHandler(int) CrashRecoveryContext.cpp:0:0
 #4 0x0000ffff9d84d854 (linux-vdso.so.1+0x854)
 #5 0x0000ffff9d3626a8 __pthread_kill_implementation (/lib64/libc.so.6+0x826a8)
 #6 0x0000ffff9d31ab40 gsignal (/lib64/libc.so.6+0x3ab40)
 #7 0x0000ffff9d3070f8 abort (/lib64/libc.so.6+0x270f8)
 #8 0x0000ffff9d314208 __assert_fail_base (/lib64/libc.so.6+0x34208)
 #9 0x0000ffff9d314270 __assert_perror_fail (/lib64/libc.so.6+0x34270)
#10 0x0000aaaabfd84c00 llvm::RISCVRegisterInfo::eliminateFrameIndex(llvm::MachineInstrBundleIterator<llvm::MachineInstr, false>, int, unsigned int, llvm::RegScavenger*) const RISCVRegisterInfo.cpp:0:0
#11 0x0000aaaac040db24 (anonymous namespace)::PEI::replaceFrameIndices(llvm::MachineBasicBlock*, llvm::MachineFunction&, int&) PrologEpilogInserter.cpp:0:0
#12 0x0000aaaac040b5d4 (anonymous namespace)::PEI::runOnMachineFunction(llvm::MachineFunction&) PrologEpilogInserter.cpp:0:0
#13 0x0000aaaac0187f0c llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2987f0c)
#14 0x0000aaaac060cdbc llvm::FPPassManager::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0cdbc)
#15 0x0000aaaac06146d8 llvm::FPPassManager::runOnModule(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e146d8)
#16 0x0000aaaac060d708 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0d708)
#17 0x0000aaaac1433268 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3c33268)
#18 0x0000aaaac178f150 clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) CodeGenAction.cpp:0:0
#19 0x0000aaaac1fb094c clang::ParseAST(clang::Sema&, bool, bool) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x47b094c)
#20 0x0000aaaac16d49e4 clang::FrontendAction::Execute() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3ed49e4)
#21 0x0000aaaac1662438 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3e62438)
#22 0x0000aaaac17895ac clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3f895ac)
#23 0x0000aaaabfb36754 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2336754)
#24 0x0000aaaabfb34b80 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) driver.cpp:0:0
#25 0x0000aaaac153d6e4 void llvm::function_ref<void ()>::callback_fn<clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const::$_1>(long) Job.cpp:0:0
#26 0x0000aaaac0c640bc llvm::CrashRecoveryContext::RunSafely(llvm::function_ref<void ()>) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34640bc)
#27 0x0000aaaac153d1a4 clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d3d1a4)
#28 0x0000aaaac150aa0c clang::driver::Compilation::ExecuteCommand(clang::driver::Command const&, clang::driver::Command const*&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0aa0c)
#29 0x0000aaaac150ac78 clang::driver::Compilation::ExecuteJobs(clang::driver::JobList const&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0ac78)
#30 0x0000aaaac151de28 clang::driver::Driver::ExecuteCompilation(clang::driver::Compilation&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d1de28)
#31 0x0000aaaabfb34310 main (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2334310)
#32 0x0000ffff9d3073c8 __libc_start_call_main (/lib64/libc.so.6+0x273c8)
#33 0x0000ffff9d3074a0 __libc_start_main@GLIBC_2.17 (/lib64/libc.so.6+0x274a0)
#34 0x0000aaaabfb31d30 _start (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2331d30)
clang-15: error: clang frontend command failed with exit code 134 (use -v to see invocation)
ClangBuiltLinux clang version 15.0.0 (https://github.com/llvm/llvm-project 02aa795785379b34f1f82d1c4d852b915c5bfb4a)
Target: riscv64-unknown-linux-gnu
Thread model: posix
InstalledDir: /home/nathan/tmp/build/llvm/stage1/bin
clang-15: note: diagnostic msg: Error generating preprocessed source(s) - no preprocessable inputs.

Thanks Nathan I'll take a look. The followup was an enhancement not a bug fix so I wouldn't expect it to resolve anything. If I don't see a quick fix I'll revert.

[craig.topper]

In D124231#3483709, @craig.topper wrote:

In D124231#3483638, @nathanchance wrote:

I bisected a crash while building the Linux kernel for RISC-V to this change:

# bad: [68ee5ad0082c0a72ff16a6571c9db7054afd6ea3] [flang] Update Google Doc link for Flang Biweekly Sync call notes
# good: [981ed72a17e4302dfd77ac54d742c08dfb6b35bb] [NFC][SCEV] Refactor `createNodeForSelectViaUMinSeq()` out of `createNodeForSelectOrPHIViaUMinSeq()`
git bisect start '68ee5ad0082c0a72ff16a6571c9db7054afd6ea3' '981ed72a17e4302dfd77ac54d742c08dfb6b35bb'
# bad: [f8463da4a329b839cfd01d7f80ae72e18f3c061e] [lldb] Allow EXE or exe in toolchain-msvc.test
git bisect bad f8463da4a329b839cfd01d7f80ae72e18f3c061e
# bad: [cf90233a67eb513930880ff1b49e80af18de5d5c] [InstCombine] Add additional test for icmp of two ranges (NFC)
git bisect bad cf90233a67eb513930880ff1b49e80af18de5d5c
# bad: [eaca933c59fd61b3df4697b5fae0eeec67acfaeb] [Clang][CodeGen]Fix __builtin_dump_struct missing record type field name
git bisect bad eaca933c59fd61b3df4697b5fae0eeec67acfaeb
# good: [52ce95a1a55424256f0d56e32392396896ed7f76] [NFC] Prevent shadowing a variable declared in `if`
git bisect good 52ce95a1a55424256f0d56e32392396896ed7f76
# good: [78582194361e4454e95d3af23367b81dd63b7943] [asan] Enable detect_stack_use_after_return=1 by default on Linux
git bisect good 78582194361e4454e95d3af23367b81dd63b7943
# bad: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them
git bisect bad c62b014db97953bee25d17b46730f05eea6a3430
# good: [9fc58f1820e3b0883c9759945bca1aa1f1df8e8d] [PowerPC] Support of ppc_fp128 in lowering of llvm.is_fpclass
git bisect good 9fc58f1820e3b0883c9759945bca1aa1f1df8e8d
# first bad commit: [c62b014db97953bee25d17b46730f05eea6a3430] [RISCV] Merge addi into load/store as there is a ADD between them

I see there was a related follow up change in D124693 but that change does not resolve the issue.

A simplified reproducer:

struct autofs_packet_missing {
  int len;
  char name[]
};
autofs_notify_daemon_wq_0, autofs_notify_daemon_type;
memcpy();
autofs_wait() {
  struct autofs_packet_missing pkt;
  switch (autofs_notify_daemon_type)
  case 0: {
    struct autofs_packet_missing *mp = &pkt;
    memcpy(mp->name);
    mp->name[autofs_notify_daemon_wq_0] = '\0';
  }
}

$ clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
...
clang: /home/nathan/cbl/src/llvm-project/llvm/include/llvm/CodeGen/MachineInstr.h:506: llvm::MachineOperand &llvm::MachineInstr::getOperand(unsigned int): Assertion `i < getNumOperands() && "getOperand() out of range!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: clang --target=riscv64-linux-gnu -O2 -c -o /dev/null waitq.i
1.      <eof> parser at end of file
2.      Code generation
3.      Running pass 'Function Pass Manager' on module 'waitq.i'.
4.      Running pass 'Prologue/Epilogue Insertion & Frame Finalization' on function '@autofs_wait'
 #0 0x0000aaaac0ce0e8c llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34e0e8c)
 #1 0x0000aaaac0cdf098 llvm::sys::RunSignalHandlers() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34df098)
 #2 0x0000aaaac0c64214 (anonymous namespace)::CrashRecoveryContextImpl::HandleCrash(int, unsigned long) CrashRecoveryContext.cpp:0:0
 #3 0x0000aaaac0c643c4 CrashRecoverySignalHandler(int) CrashRecoveryContext.cpp:0:0
 #4 0x0000ffff9d84d854 (linux-vdso.so.1+0x854)
 #5 0x0000ffff9d3626a8 __pthread_kill_implementation (/lib64/libc.so.6+0x826a8)
 #6 0x0000ffff9d31ab40 gsignal (/lib64/libc.so.6+0x3ab40)
 #7 0x0000ffff9d3070f8 abort (/lib64/libc.so.6+0x270f8)
 #8 0x0000ffff9d314208 __assert_fail_base (/lib64/libc.so.6+0x34208)
 #9 0x0000ffff9d314270 __assert_perror_fail (/lib64/libc.so.6+0x34270)
#10 0x0000aaaabfd84c00 llvm::RISCVRegisterInfo::eliminateFrameIndex(llvm::MachineInstrBundleIterator<llvm::MachineInstr, false>, int, unsigned int, llvm::RegScavenger*) const RISCVRegisterInfo.cpp:0:0
#11 0x0000aaaac040db24 (anonymous namespace)::PEI::replaceFrameIndices(llvm::MachineBasicBlock*, llvm::MachineFunction&, int&) PrologEpilogInserter.cpp:0:0
#12 0x0000aaaac040b5d4 (anonymous namespace)::PEI::runOnMachineFunction(llvm::MachineFunction&) PrologEpilogInserter.cpp:0:0
#13 0x0000aaaac0187f0c llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2987f0c)
#14 0x0000aaaac060cdbc llvm::FPPassManager::runOnFunction(llvm::Function&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0cdbc)
#15 0x0000aaaac06146d8 llvm::FPPassManager::runOnModule(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e146d8)
#16 0x0000aaaac060d708 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2e0d708)
#17 0x0000aaaac1433268 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, std::unique_ptr<llvm::raw_pwrite_stream, std::default_delete<llvm::raw_pwrite_stream> >) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3c33268)
#18 0x0000aaaac178f150 clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) CodeGenAction.cpp:0:0
#19 0x0000aaaac1fb094c clang::ParseAST(clang::Sema&, bool, bool) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x47b094c)
#20 0x0000aaaac16d49e4 clang::FrontendAction::Execute() (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3ed49e4)
#21 0x0000aaaac1662438 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3e62438)
#22 0x0000aaaac17895ac clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3f895ac)
#23 0x0000aaaabfb36754 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2336754)
#24 0x0000aaaabfb34b80 ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) driver.cpp:0:0
#25 0x0000aaaac153d6e4 void llvm::function_ref<void ()>::callback_fn<clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const::$_1>(long) Job.cpp:0:0
#26 0x0000aaaac0c640bc llvm::CrashRecoveryContext::RunSafely(llvm::function_ref<void ()>) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x34640bc)
#27 0x0000aaaac153d1a4 clang::driver::CC1Command::Execute(llvm::ArrayRef<llvm::Optional<llvm::StringRef> >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, bool*) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d3d1a4)
#28 0x0000aaaac150aa0c clang::driver::Compilation::ExecuteCommand(clang::driver::Command const&, clang::driver::Command const*&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0aa0c)
#29 0x0000aaaac150ac78 clang::driver::Compilation::ExecuteJobs(clang::driver::JobList const&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) const (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d0ac78)
#30 0x0000aaaac151de28 clang::driver::Driver::ExecuteCompilation(clang::driver::Compilation&, llvm::SmallVectorImpl<std::pair<int, clang::driver::Command const*> >&) (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x3d1de28)
#31 0x0000aaaabfb34310 main (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2334310)
#32 0x0000ffff9d3073c8 __libc_start_call_main (/lib64/libc.so.6+0x273c8)
#33 0x0000ffff9d3074a0 __libc_start_main@GLIBC_2.17 (/lib64/libc.so.6+0x274a0)
#34 0x0000aaaabfb31d30 _start (/home/nathan/tmp/build/llvm/stage1/bin/clang-15+0x2331d30)
clang-15: error: clang frontend command failed with exit code 134 (use -v to see invocation)
ClangBuiltLinux clang version 15.0.0 (https://github.com/llvm/llvm-project 02aa795785379b34f1f82d1c4d852b915c5bfb4a)
Target: riscv64-unknown-linux-gnu
Thread model: posix
InstalledDir: /home/nathan/tmp/build/llvm/stage1/bin
clang-15: note: diagnostic msg: Error generating preprocessed source(s) - no preprocessable inputs.

Thanks Nathan I'll take a look. The followup was an enhancement not a bug fix so I wouldn't expect it to resolve anything. If I don't see a quick fix I'll revert.

I have a fix. Just need to tidy up a test case.

[craig.topper]

Should be fixed after f91690f7db96aae16425241a17fe6649f5d156b4

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

bool RISCVDAGToDAGISel::selectRVVSimm5(SDValue N, unsigned Width,
   }
 
   return false;
 }
 
 // Merge an ADDI into the offset of a load/store instruction where possible.
 // (load (addi base, off1), off2) -> (load base, off1+off2)
 // (store val, (addi base, off1), off2) -> (store val, base, off1+off2)
+// (load (add base, (addi src, off1)), off2)
+//    -> (load (add base, src), off1+off2)
+// (store val, (add base, (addi src, off1)), off2)
+//    -> (store val, (add base, src), off1+off2)
 // This is possible when off1+off2 fits a 12-bit immediate.
 bool RISCVDAGToDAGISel::doPeepholeLoadStoreADDI(SDNode *N) {
   int OffsetOpIdx;
   int BaseOpIdx;
 
   // Only attempt this optimisation for I-type loads and S-type stores.
   switch (N->getMachineOpcode()) {
   default:
case RISCV::FSD:
     break;
   }
 
   if (!isa<ConstantSDNode>(N->getOperand(OffsetOpIdx)))
     return false;
 
   SDValue Base = N->getOperand(BaseOpIdx);
 
+  if (!Base.isMachineOpcode())
+    return false;
+
+  // There is a ADD between ADDI and load/store.
+  SDValue Add;
+  int AddBaseIdx;
+  if (Base.getMachineOpcode() == RISCV::ADD) {
+    if (!Base.hasOneUse())
+      return false;
+    Add = Base;
+    SDValue Op0 = Base.getOperand(0);
+    SDValue Op1 = Base.getOperand(1);
+    if (Op0.isMachineOpcode() && Op0.getMachineOpcode() == RISCV::ADDI &&
+        isa<ConstantSDNode>(Op0.getOperand(1)) &&
+        cast<ConstantSDNode>(Op0.getOperand(1))->getSExtValue() != 0) {
+      AddBaseIdx = 1;
+      Base = Op0;
+    } else if (Op1.isMachineOpcode() && Op1.getMachineOpcode() == RISCV::ADDI &&
+               isa<ConstantSDNode>(Op1.getOperand(1)) &&
+               cast<ConstantSDNode>(Op1.getOperand(1))->getSExtValue() != 0) {
+      AddBaseIdx = 0;
+      Base = Op1;
+    } else
+      return false;
+  }
+
   // If the base is an ADDI, we can merge it in to the load/store.
-  if (!Base.isMachineOpcode() || Base.getMachineOpcode() != RISCV::ADDI)
+  if (Base.getMachineOpcode() != RISCV::ADDI)
     return false;
 
   SDValue ImmOperand = Base.getOperand(1);
   uint64_t Offset2 = N->getConstantOperandVal(OffsetOpIdx);
 
   if (auto *Const = dyn_cast<ConstantSDNode>(ImmOperand)) {
     int64_t Offset1 = Const->getSExtValue();
     int64_t CombinedOffset = Offset1 + Offset2;
bool RISCVDAGToDAGISel::doPeepholeLoadStoreADDI(SDNode *N) {
   }
 
   LLVM_DEBUG(dbgs() << "Folding add-immediate into mem-op:\nBase:    ");
   LLVM_DEBUG(Base->dump(CurDAG));
   LLVM_DEBUG(dbgs() << "\nN: ");
   LLVM_DEBUG(N->dump(CurDAG));
   LLVM_DEBUG(dbgs() << "\n");
 
+  if (Add)
+    Add = SDValue(CurDAG->UpdateNodeOperands(Add.getNode(),
+                                             Add.getOperand(AddBaseIdx),
+                                             Base.getOperand(0)),
+                  0);
+
   // Modify the offset operand of the load/store.
-  if (BaseOpIdx == 0) // Load
-    CurDAG->UpdateNodeOperands(N, Base.getOperand(0), ImmOperand,
+  if (BaseOpIdx == 0) { // Load
+    if (Add)
+      N = CurDAG->UpdateNodeOperands(N, Add, ImmOperand, N->getOperand(2));
+    else
+      N = CurDAG->UpdateNodeOperands(N, Base.getOperand(0), ImmOperand,
                                      N->getOperand(2));
-  else // Store
-    CurDAG->UpdateNodeOperands(N, N->getOperand(0), Base.getOperand(0),
+  } else { // Store
+    if (Add)
+      N = CurDAG->UpdateNodeOperands(N, N->getOperand(0), Add, ImmOperand,
+                                     N->getOperand(3));
+    else
+      N = CurDAG->UpdateNodeOperands(N, N->getOperand(0), Base.getOperand(0),
                                      ImmOperand, N->getOperand(3));
+  }
 
   return true;
 }
 
 // Try to remove sext.w if the input is a W instruction or can be made into
 // a W instruction cheaply.
 bool RISCVDAGToDAGISel::doPeepholeSExtW(SDNode *N) {
   // Look for the sext.w pattern, addiw rd, rs1, 0.

llvm/test/CodeGen/RISCV/large-stack.ll

 ; RV32I-FPELIM-NEXT:    sw s1, 2024(sp) # 4-byte Folded Spill
 ; RV32I-FPELIM-NEXT:    .cfi_offset s0, -4
 ; RV32I-FPELIM-NEXT:    .cfi_offset s1, -8
 ; RV32I-FPELIM-NEXT:    lui a1, 97
 ; RV32I-FPELIM-NEXT:    addi a1, a1, 672
 ; RV32I-FPELIM-NEXT:    sub sp, sp, a1
 ; RV32I-FPELIM-NEXT:    .cfi_def_cfa_offset 400016
 ; RV32I-FPELIM-NEXT:    lui a1, 78
-; RV32I-FPELIM-NEXT:    addi a1, a1, 512
 ; RV32I-FPELIM-NEXT:    addi a2, sp, 8
 ; RV32I-FPELIM-NEXT:    add a1, a2, a1
 ; RV32I-FPELIM-NEXT:    #APP
 ; RV32I-FPELIM-NEXT:    nop
 ; RV32I-FPELIM-NEXT:    #NO_APP
-; RV32I-FPELIM-NEXT:    sw a0, 0(a1)
+; RV32I-FPELIM-NEXT:    sw a0, 512(a1)
 ; RV32I-FPELIM-NEXT:    #APP
 ; RV32I-FPELIM-NEXT:    nop
 ; RV32I-FPELIM-NEXT:    #NO_APP
 ; RV32I-FPELIM-NEXT:    lui a0, 97
 ; RV32I-FPELIM-NEXT:    addi a0, a0, 672
 ; RV32I-FPELIM-NEXT:    add sp, sp, a0
 ; RV32I-FPELIM-NEXT:    lw s0, 2028(sp) # 4-byte Folded Reload
 ; RV32I-FPELIM-NEXT:    lw s1, 2024(sp) # 4-byte Folded Reload

 ; RV32I-WITHFP-NEXT:    .cfi_offset s1, -12
 ; RV32I-WITHFP-NEXT:    .cfi_offset s2, -16
 ; RV32I-WITHFP-NEXT:    addi s0, sp, 2032
 ; RV32I-WITHFP-NEXT:    .cfi_def_cfa s0, 0
 ; RV32I-WITHFP-NEXT:    lui a1, 97
 ; RV32I-WITHFP-NEXT:    addi a1, a1, 688
 ; RV32I-WITHFP-NEXT:    sub sp, sp, a1
 ; RV32I-WITHFP-NEXT:    lui a1, 78
-; RV32I-WITHFP-NEXT:    addi a1, a1, 512
 ; RV32I-WITHFP-NEXT:    lui a2, 1048478
 ; RV32I-WITHFP-NEXT:    addi a2, a2, 1388
 ; RV32I-WITHFP-NEXT:    add a2, s0, a2
 ; RV32I-WITHFP-NEXT:    add a1, a2, a1
 ; RV32I-WITHFP-NEXT:    #APP
 ; RV32I-WITHFP-NEXT:    nop
 ; RV32I-WITHFP-NEXT:    #NO_APP
-; RV32I-WITHFP-NEXT:    sw a0, 0(a1)
+; RV32I-WITHFP-NEXT:    sw a0, 512(a1)
 ; RV32I-WITHFP-NEXT:    #APP
 ; RV32I-WITHFP-NEXT:    nop
 ; RV32I-WITHFP-NEXT:    #NO_APP
 ; RV32I-WITHFP-NEXT:    lui a0, 97
 ; RV32I-WITHFP-NEXT:    addi a0, a0, 688
 ; RV32I-WITHFP-NEXT:    add sp, sp, a0
 ; RV32I-WITHFP-NEXT:    lw ra, 2028(sp) # 4-byte Folded Reload
 ; RV32I-WITHFP-NEXT:    lw s0, 2024(sp) # 4-byte Folded Reload

llvm/test/CodeGen/RISCV/mem.ll

 ; RV32I-NEXT:    sw a0, -273(a2)
 ; RV32I-NEXT:    mv a0, a1
 ; RV32I-NEXT:    ret
   %1 = inttoptr i32 3735928559 to i32*
   %2 = load volatile i32, i32* %1
   store i32 %a, i32* %1
   ret i32 %2
 }
+
+define i32 @lw_far_local(i32* %a)  {
+; RV32I-LABEL: lw_far_local:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 4
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    lw a0, -4(a0)
+; RV32I-NEXT:    ret
+  %1 = getelementptr inbounds i32, i32* %a, i64 4095
+  %2 = load volatile i32, i32* %1
+  ret i32 %2
+}
+
+define void @st_far_local(i32* %a, i32 %b)  {
+; RV32I-LABEL: st_far_local:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 4
+; RV32I-NEXT:    add a0, a0, a2
+; RV32I-NEXT:    sw a1, -4(a0)
+; RV32I-NEXT:    ret
+  %1 = getelementptr inbounds i32, i32* %a, i64 4095
+  store i32 %b, i32* %1
+  ret void
+}
+
+define i32 @lw_sw_far_local(i32* %a, i32 %b)  {
+; RV32I-LABEL: lw_sw_far_local:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 4
+; RV32I-NEXT:    addi a2, a2, -4
+; RV32I-NEXT:    add a2, a0, a2
+; RV32I-NEXT:    lw a0, 0(a2)
+; RV32I-NEXT:    sw a1, 0(a2)
+; RV32I-NEXT:    ret
+  %1 = getelementptr inbounds i32, i32* %a, i64 4095
+  %2 = load volatile i32, i32* %1
+  store i32 %b, i32* %1
+  ret i32 %2
+}

llvm/test/CodeGen/RISCV/mem64.ll

 ; RV64I-NEXT:    ret
   %1 = load volatile i64, i64* @G
   store i64 %a, i64* @G
   %2 = getelementptr i64, i64* @G, i64 9
   %3 = load volatile i64, i64* %2
   store i64 %a, i64* %2
   ret i64 %1
 }
+
+define i64 @lw_far_local(i64* %a)  {
+; RV64I-LABEL: lw_far_local:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    lui a1, 8
+; RV64I-NEXT:    addiw a1, a1, -8
+; RV64I-NEXT:    add a0, a0, a1
+; RV64I-NEXT:    ld a0, 0(a0)
+; RV64I-NEXT:    ret
+  %1 = getelementptr inbounds i64, i64* %a, i64 4095
+  %2 = load volatile i64, i64* %1
+  ret i64 %2
+}
+
+define void @st_far_local(i64* %a, i64 %b)  {
+; RV64I-LABEL: st_far_local:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    lui a2, 8
+; RV64I-NEXT:    addiw a2, a2, -8
+; RV64I-NEXT:    add a0, a0, a2
+; RV64I-NEXT:    sd a1, 0(a0)
+; RV64I-NEXT:    ret
+  %1 = getelementptr inbounds i64, i64* %a, i64 4095
+  store i64 %b, i64* %1
+  ret void
+}
+
+define i64 @lw_sw_far_local(i64* %a, i64 %b)  {
+; RV64I-LABEL: lw_sw_far_local:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    lui a2, 8
+; RV64I-NEXT:    addiw a2, a2, -8
+; RV64I-NEXT:    add a2, a0, a2
+; RV64I-NEXT:    ld a0, 0(a2)
+; RV64I-NEXT:    sd a1, 0(a2)
+; RV64I-NEXT:    ret
+  %1 = getelementptr inbounds i64, i64* %a, i64 4095
+  %2 = load volatile i64, i64* %1
+  store i64 %b, i64* %1
+  ret i64 %2
+}