[MLIR]`-test-extract-fixed-outer-loops` triggers Assertion Failure `!empty()'

[xisang0]

Test on commit: 6548b63
steps to reproduce:

mlir-opt test.mlir  --test-extract-fixed-outer-loops

test case:

module {
  func.func @main(%arg0: tensor<4xi32>) -> () {
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %2 = bufferization.to_memref %arg0 : memref<4xi32>
    %alloc = memref.alloc() {alignment = 64 : i64} : memref<4xi32>
    scf.for %arg4 = %c0 to %c4 step %c1 {
      %19 = memref.load %2[%arg4] : memref<4xi32>
      %20 = math.ctlz %19 : i32
      memref.store %20, %alloc[%arg4] : memref<4xi32>
    }
    return
  }
}

crash trace:

mlir-opt: /home/workdir/llvm-project-19/llvm/include/llvm/ADT/SmallVector.h:309: reference llvm::SmallVectorTemplateCommon<mlir::scf::ForOp>::back() [T = mlir::scf::ForOp]: Assertion `!empty()' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: ./mlir-opt /home/workdir/test.mlir --test-extract-fixed-outer-loops
^[[A #0 0x000055c40f1e53b8 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (./mlir-opt+0x10723b8)
 #1 0x000055c40f1e2ede llvm::sys::RunSignalHandlers() (./mlir-opt+0x106fede)
 #2 0x000055c40f1e5dc8 SignalHandler(int) Signals.cpp:0:0
 #3 0x000070577cebe520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x000070577cf129fc pthread_kill (/lib/x86_64-linux-gnu/libc.so.6+0x969fc)
 #5 0x000070577cebe476 gsignal (/lib/x86_64-linux-gnu/libc.so.6+0x42476)
 #6 0x000070577cea47f3 abort (/lib/x86_64-linux-gnu/libc.so.6+0x287f3)
 #7 0x000070577cea471b (/lib/x86_64-linux-gnu/libc.so.6+0x2871b)
 #8 0x000070577ceb5e96 (/lib/x86_64-linux-gnu/libc.so.6+0x39e96)
 #9 0x000055c41100ce2a mlir::extractFixedOuterLoops(mlir::scf::ForOp, llvm::ArrayRef<long>) (./mlir-opt+0x2e99e2a)
#10 0x000055c411ef97b5 void llvm::function_ref<void (mlir::Operation*)>::callback_fn<std::enable_if<!llvm::is_one_of<mlir::scf::ForOp, mlir::Operation*, mlir::Region*, mlir::Block*>::value && std::is_same<void, void>::value, void>::type mlir::detail::walk<(mlir::WalkOrder)1, mlir::ForwardIterator, (anonymous namespace)::SimpleParametricLoopTilingPass::runOnOperation()::'lambda'(mlir::scf::ForOp), mlir::scf::ForOp, void>(mlir::Operation*, (anonymous namespace)::SimpleParametricLoopTilingPass::runOnOperation()::'lambda'(mlir::scf::ForOp)&&)::'lambda'(mlir::Operation*)>(long, mlir::Operation*) TestLoopParametricTiling.cpp:0:0
#11 0x000055c40f2f371e void mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>, mlir::WalkOrder) (./mlir-opt+0x118071e)
#12 0x000055c40f2f371e void mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<void (mlir::Operation*)>, mlir::WalkOrder) (./mlir-opt+0x118071e)
#13 0x000055c411ef95dd (anonymous namespace)::SimpleParametricLoopTilingPass::runOnOperation() TestLoopParametricTiling.cpp:0:0
#14 0x000055c4124616df mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (./mlir-opt+0x42ee6df)
#15 0x000055c412461ec2 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (./mlir-opt+0x42eeec2)
#16 0x000055c41246466e mlir::PassManager::run(mlir::Operation*) (./mlir-opt+0x42f166e)
#17 0x000055c41245d192 performActions(llvm::raw_ostream&, std::shared_ptr<llvm::SourceMgr> const&, mlir::MLIRContext*, mlir::MlirOptMainConfig const&) MlirOptMain.cpp:0:0
#18 0x000055c41245cdfb llvm::LogicalResult llvm::function_ref<llvm::LogicalResult (std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&)>::callback_fn<mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&)::$_0>(long, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&) MlirOptMain.cpp:0:0
#19 0x000055c412506ae5 mlir::splitAndProcessBuffer(std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::function_ref<llvm::LogicalResult (std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, llvm::raw_ostream&)>, llvm::raw_ostream&, llvm::StringRef, llvm::StringRef) (./mlir-opt+0x4393ae5)
#20 0x000055c412457ff2 mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&) (./mlir-opt+0x42e4ff2)
#21 0x000055c4124582a3 mlir::MlirOptMain(int, char**, llvm::StringRef, llvm::StringRef, mlir::DialectRegistry&) (./mlir-opt+0x42e52a3)
#22 0x000055c4124584b2 mlir::MlirOptMain(int, char**, llvm::StringRef, mlir::DialectRegistry&) (./mlir-opt+0x42e54b2)
#23 0x000055c40f1c45d7 main (./mlir-opt+0x10515d7)
#24 0x000070577cea5d90 (/lib/x86_64-linux-gnu/libc.so.6+0x29d90)
#25 0x000070577cea5e40 __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x29e40)
#26 0x000055c40f1c4145 _start (./mlir-opt+0x1051145)
Aborted (core dumped)

[Lewuathe]

It does not throw the assertion failure with the HEAD of main branch.

./bin/mlir-opt test.mlir  --test-extract-fixed-outer-loops
test.mlir:6:42: error: custom op 'bufferization.to_memref' invalid kind of type specified
    %2 = bufferization.to_memref %arg0 : memref<4xi32>

[Lewuathe]

We would rather fix the invalid return type of bufferization.to_memref so that we can reproduce the situation. Here is the code to get this crash.

module {
  func.func @main(%arg0: tensor<4xi32>) -> () {
    %c4 = arith.constant 4 : index
    %c0 = arith.constant 0 : index
    %c1 = arith.constant 1 : index
    %2 = bufferization.to_memref %arg0 : tensor<4xi32> to memref<4xi32>
    %alloc = memref.alloc() {alignment = 64 : i64} : memref<4xi32>
    scf.for %arg4 = %c0 to %c4 step %c1 {
      %19 = memref.load %2[%arg4] : memref<4xi32>
      %20 = math.ctlz %19 : i32
      memref.store %20, %alloc[%arg4] : memref<4xi32>
    }
    return
  }
}

./bin/mlir-opt test.mlir  --test-extract-fixed-outer-loops
Assertion failed: (!empty()), function back, file SmallVector.h, line 309.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.	Program arguments: ./bin/mlir-opt test.mlir --test-extract-fixed-outer-loops
Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it):
0  mlir-opt                 0x0000000100dd5aa8 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) + 56
1  mlir-opt                 0x0000000100dd3b2c llvm::sys::RunSignalHandlers() + 112
2  mlir-opt                 0x0000000100dd6164 SignalHandler(int) + 360
3  libsystem_platform.dylib 0x000000018cfda584 _sigtramp + 56
4  libsystem_pthread.dylib  0x000000018cfa9c20 pthread_kill + 288
5  libsystem_c.dylib        0x000000018ceb6a30 abort + 180
6  libsystem_c.dylib        0x000000018ceb5d20 err + 0

[Lewuathe]

Looks like we need to provide non-zerotest-outer-loop-sizes parameter. The default value is 0 and it seems to cause the crash.

mlir-opt -test-extract-fixed-outer-loops='test-outer-loop-sizes=7

[Lewuathe]

It appears to be the same issue with #61716.

We may need to guard the pass from the zero test-outer-loop-sizes if no option is given.

[xisang0]

Thank you for your answer.