[Mlir] --affine-super-vectorizer-test="vectorize-affine-loop-nest" crashes in VectorUtils.cpp:145: AffineMap makePermutationMap git version: e4372c4454c9

[Emilyaxe]

system: Ubuntu 18.04.6 LTS

reproduce with: mlir-opt a.mlir --affine-super-vectorizer-test="vectorize-affine-loop-nest"

a.mlir:

#map = affine_map<(d0)[s0] -> (d0 mod s0)>
#map1 = affine_map<(d0)[s0] -> (d0 floordiv s0)>
module {
  memref.global "private" constant @__constant_17x18xf32 : memref<17x18xf32> = dense<0.000000e+00> {alignment = 64 : i64}
  memref.global "private" constant @__constant_17x18xi32 : memref<17x18xi32> = dense<0> {alignment = 64 : i64}
  func.func private @func3(%arg0: index) -> (memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>) {
    %c18 = arith.constant 18 : index
    %cst = arith.constant 1.000000e+00 : f32
    %0 = memref.get_global @__constant_17x18xi32 : memref<17x18xi32>
    %1 = memref.get_global @__constant_17x18xf32 : memref<17x18xf32>
    %alloc = memref.alloc() {alignment = 64 : i64} : memref<17x18xf32>
    affine.for %arg1 = 0 to 306 {
      %2 = affine.apply #map(%arg1)[%c18]
      %3 = affine.apply #map1(%arg1)[%c18]
      affine.store %cst, %alloc[%3, %2] : memref<17x18xf32>
    }
    return %1, %0, %alloc : memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>
  }
}

stack trace:

mlir-opt: /data/szy/MLIR/llvm-release/llvm-project/mlir/lib/Dialect/Vector/Utils/VectorUtils.cpp:145: AffineMap makePermutationMap(ArrayRef<Value>, const DenseMap<Operation *, unsigned int> &): Assertion `perm[kvp.second] == getAffineConstantExpr(0, context) && "permutationMap already has an entry along dim"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt a.mlir --affine-super-vectorizer-test=vectorize-affine-loop-nest
 #0 0x00005590fd797ce8 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11f2ce8)
 #1 0x00005590fd7957fe llvm::sys::RunSignalHandlers() (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11f07fe)
 #2 0x00005590fd79867d SignalHandler(int) Signals.cpp:0:0
 #3 0x00007fe98ff96420 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x14420)
 #4 0x00007fe98f5d300b raise /build/glibc-LcI20x/glibc-2.31/signal/../sysdeps/unix/sysv/linux/raise.c:51:1
 #5 0x00007fe98f5b2859 abort /build/glibc-LcI20x/glibc-2.31/stdlib/abort.c:81:7
 #6 0x00007fe98f5b2729 get_sysdep_segment_value /build/glibc-LcI20x/glibc-2.31/intl/loadmsgcat.c:509:8
 #7 0x00007fe98f5b2729 _nl_load_domain /build/glibc-LcI20x/glibc-2.31/intl/loadmsgcat.c:970:34
 #8 0x00007fe98f5c3fd6 (/lib/x86_64-linux-gnu/libc.so.6+0x33fd6)
 #9 0x0000559100095e16 mlir::makePermutationMap(mlir::Block*, llvm::ArrayRef<mlir::Value>, llvm::DenseMap<mlir::Operation*, unsigned int, llvm::DenseMapInfo<mlir::Operation*, void>, llvm::detail::DenseMapPair<mlir::Operation*, unsigned int>> const&) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x3af0e16)
#10 0x00005590fd8d3e0c vectorizeOneOperation(mlir::Operation*, (anonymous namespace)::VectorizationState&) SuperVectorize.cpp:0:0
#11 0x00005590fd8d3549 mlir::WalkResult llvm::function_ref<mlir::WalkResult (mlir::Operation*)>::callback_fn<vectorizeLoopNest(std::vector<llvm::SmallVector<mlir::affine::AffineForOp, 2u>, std::allocator<llvm::SmallVector<mlir::affine::AffineForOp, 2u>>>&, mlir::affine::VectorizationStrategy const&)::$_0>(long, mlir::Operation*) SuperVectorize.cpp:0:0
#12 0x00005590fd8c3228 mlir::WalkResult mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<mlir::WalkResult (mlir::Operation*)>, mlir::WalkOrder) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x131e228)
#13 0x00005590fd8c31d7 mlir::WalkResult mlir::detail::walk<mlir::ForwardIterator>(mlir::Operation*, llvm::function_ref<mlir::WalkResult (mlir::Operation*)>, mlir::WalkOrder) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x131e1d7)
#14 0x00005590fd8d1b6f vectorizeLoopNest(std::vector<llvm::SmallVector<mlir::affine::AffineForOp, 2u>, std::allocator<llvm::SmallVector<mlir::affine::AffineForOp, 2u>>>&, mlir::affine::VectorizationStrategy const&) SuperVectorize.cpp:0:0
#15 0x00005590fd8d18a2 mlir::affine::vectorizeAffineLoopNest(std::vector<llvm::SmallVector<mlir::affine::AffineForOp, 2u>, std::allocator<llvm::SmallVector<mlir::affine::AffineForOp, 2u>>>&, mlir::affine::VectorizationStrategy const&) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x132c8a2)
#16 0x00005591008c6353 (anonymous namespace)::VectorizerTestPass::runOnOperation() TestVectorizationUtils.cpp:0:0
#17 0x00005591006f2496 mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414d496)
#18 0x00005591006f2dc0 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414ddc0)
#19 0x00005591006f831e auto void mlir::parallelForEach<__gnu_cxx::__normal_iterator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo>>>, mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::$_0>(mlir::MLIRContext*, __gnu_cxx::__normal_iterator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo>>>, __gnu_cxx::__normal_iterator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo>>>, mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::$_0&&)::'lambda'(__gnu_cxx::__normal_iterator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo>>>&&)::operator()<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&>(__gnu_cxx::__normal_iterator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator<mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo>>>&&) const Pass.cpp:0:0
#20 0x00005591006f473b mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414f73b)
#21 0x00005591006f25ef mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414d5ef)
#22 0x00005591006f2dc0 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414ddc0)
#23 0x00005591006f56c2 mlir::PassManager::run(mlir::Operation*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41506c2)
#24 0x00005591006edaea performActions(llvm::raw_ostream&, std::shared_ptr<llvm::SourceMgr> const&, mlir::MLIRContext*, mlir::MlirOptMainConfig const&) MlirOptMain.cpp:0:0
#25 0x00005591006ed73d 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
#26 0x0000559100794785 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) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41ef785)
#27 0x00005591006e7755 mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer>>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x4142755)
#28 0x00005591006e79ff mlir::MlirOptMain(int, char**, llvm::StringRef, llvm::StringRef, mlir::DialectRegistry&) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41429ff)
#29 0x00005591006e7d2e mlir::MlirOptMain(int, char**, llvm::StringRef, mlir::DialectRegistry&) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x4142d2e)
#30 0x00005590fd776a67 main (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11d1a67)
#31 0x00007fe98f5b4083 __libc_start_main /build/glibc-LcI20x/glibc-2.31/csu/../csu/libc-start.c:342:3
#32 0x00005590fd7765de _start (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11d15de)

[llvmbot]

@llvm/issue-subscribers-mlir-affine

Author: None (Emilyaxe)

system: `Ubuntu 18.04.6 LTS`

reproduce with: mlir-opt a.mlir --affine-super-vectorizer-test="vectorize-affine-loop-nest"

a.mlir:

#map = affine_map<(d0)[s0] -> (d0 mod s0)>
#map1 = affine_map<(d0)[s0] -> (d0 floordiv s0)>
module {
  memref.global "private" constant @<!-- -->__constant_17x18xf32 : memref&lt;17x18xf32&gt; = dense&lt;0.000000e+00&gt; {alignment = 64 : i64}
  memref.global "private" constant @<!-- -->__constant_17x18xi32 : memref&lt;17x18xi32&gt; = dense&lt;0&gt; {alignment = 64 : i64}
  func.func private @<!-- -->func3(%arg0: index) -&gt; (memref&lt;17x18xf32&gt;, memref&lt;17x18xi32&gt;, memref&lt;17x18xf32&gt;) {
    %c18 = arith.constant 18 : index
    %cst = arith.constant 1.000000e+00 : f32
    %0 = memref.get_global @<!-- -->__constant_17x18xi32 : memref&lt;17x18xi32&gt;
    %1 = memref.get_global @<!-- -->__constant_17x18xf32 : memref&lt;17x18xf32&gt;
    %alloc = memref.alloc() {alignment = 64 : i64} : memref&lt;17x18xf32&gt;
    affine.for %arg1 = 0 to 306 {
      %2 = affine.apply #map(%arg1)[%c18]
      %3 = affine.apply #map1(%arg1)[%c18]
      affine.store %cst, %alloc[%3, %2] : memref&lt;17x18xf32&gt;
    }
    return %1, %0, %alloc : memref&lt;17x18xf32&gt;, memref&lt;17x18xi32&gt;, memref&lt;17x18xf32&gt;
  }
}

stack trace:

mlir-opt: /data/szy/MLIR/llvm-release/llvm-project/mlir/lib/Dialect/Vector/Utils/VectorUtils.cpp:145: AffineMap makePermutationMap(ArrayRef<Value>, const DenseMap<Operation *, unsigned int> &): Assertion `perm[kvp.second] == getAffineConstantExpr(0, context) && "permutationMap already has an entry along dim"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt a.mlir --affine-super-vectorizer-test=vectorize-affine-loop-nest
 #<!-- -->0 0x00005590fd797ce8 llvm::sys::PrintStackTrace(llvm::raw_ostream&amp;, int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11f2ce8)
 #<!-- -->1 0x00005590fd7957fe llvm::sys::RunSignalHandlers() (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11f07fe)
 #<!-- -->2 0x00005590fd79867d SignalHandler(int) Signals.cpp:0:0
 #<!-- -->3 0x00007fe98ff96420 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x14420)
 #<!-- -->4 0x00007fe98f5d300b raise /build/glibc-LcI20x/glibc-2.31/signal/../sysdeps/unix/sysv/linux/raise.c:51:1
 #<!-- -->5 0x00007fe98f5b2859 abort /build/glibc-LcI20x/glibc-2.31/stdlib/abort.c:81:7
 #<!-- -->6 0x00007fe98f5b2729 get_sysdep_segment_value /build/glibc-LcI20x/glibc-2.31/intl/loadmsgcat.c:509:8
 #<!-- -->7 0x00007fe98f5b2729 _nl_load_domain /build/glibc-LcI20x/glibc-2.31/intl/loadmsgcat.c:970:34
 #<!-- -->8 0x00007fe98f5c3fd6 (/lib/x86_64-linux-gnu/libc.so.6+0x33fd6)
 #<!-- -->9 0x0000559100095e16 mlir::makePermutationMap(mlir::Block*, llvm::ArrayRef&lt;mlir::Value&gt;, llvm::DenseMap&lt;mlir::Operation*, unsigned int, llvm::DenseMapInfo&lt;mlir::Operation*, void&gt;, llvm::detail::DenseMapPair&lt;mlir::Operation*, unsigned int&gt;&gt; const&amp;) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x3af0e16)
#<!-- -->10 0x00005590fd8d3e0c vectorizeOneOperation(mlir::Operation*, (anonymous namespace)::VectorizationState&amp;) SuperVectorize.cpp:0:0
#<!-- -->11 0x00005590fd8d3549 mlir::WalkResult llvm::function_ref&lt;mlir::WalkResult (mlir::Operation*)&gt;::callback_fn&lt;vectorizeLoopNest(std::vector&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;, std::allocator&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;&gt;&gt;&amp;, mlir::affine::VectorizationStrategy const&amp;)::$_0&gt;(long, mlir::Operation*) SuperVectorize.cpp:0:0
#<!-- -->12 0x00005590fd8c3228 mlir::WalkResult mlir::detail::walk&lt;mlir::ForwardIterator&gt;(mlir::Operation*, llvm::function_ref&lt;mlir::WalkResult (mlir::Operation*)&gt;, mlir::WalkOrder) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x131e228)
#<!-- -->13 0x00005590fd8c31d7 mlir::WalkResult mlir::detail::walk&lt;mlir::ForwardIterator&gt;(mlir::Operation*, llvm::function_ref&lt;mlir::WalkResult (mlir::Operation*)&gt;, mlir::WalkOrder) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x131e1d7)
#<!-- -->14 0x00005590fd8d1b6f vectorizeLoopNest(std::vector&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;, std::allocator&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;&gt;&gt;&amp;, mlir::affine::VectorizationStrategy const&amp;) SuperVectorize.cpp:0:0
#<!-- -->15 0x00005590fd8d18a2 mlir::affine::vectorizeAffineLoopNest(std::vector&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;, std::allocator&lt;llvm::SmallVector&lt;mlir::affine::AffineForOp, 2u&gt;&gt;&gt;&amp;, mlir::affine::VectorizationStrategy const&amp;) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x132c8a2)
#<!-- -->16 0x00005591008c6353 (anonymous namespace)::VectorizerTestPass::runOnOperation() TestVectorizationUtils.cpp:0:0
#<!-- -->17 0x00005591006f2496 mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414d496)
#<!-- -->18 0x00005591006f2dc0 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&amp;, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414ddc0)
#<!-- -->19 0x00005591006f831e auto void mlir::parallelForEach&lt;__gnu_cxx::__normal_iterator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&gt;&gt;&gt;, mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::$_0&gt;(mlir::MLIRContext*, __gnu_cxx::__normal_iterator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&gt;&gt;&gt;, __gnu_cxx::__normal_iterator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&gt;&gt;&gt;, mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::$_0&amp;&amp;)::'lambda'(__gnu_cxx::__normal_iterator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&gt;&gt;&gt;&amp;&amp;)::operator()&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&amp;&gt;(__gnu_cxx::__normal_iterator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo*, std::vector&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo, std::allocator&lt;mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&gt;&gt;&gt;&amp;&amp;) const Pass.cpp:0:0
#<!-- -->20 0x00005591006f473b mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414f73b)
#<!-- -->21 0x00005591006f25ef mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414d5ef)
#<!-- -->22 0x00005591006f2dc0 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&amp;, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x414ddc0)
#<!-- -->23 0x00005591006f56c2 mlir::PassManager::run(mlir::Operation*) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41506c2)
#<!-- -->24 0x00005591006edaea performActions(llvm::raw_ostream&amp;, std::shared_ptr&lt;llvm::SourceMgr&gt; const&amp;, mlir::MLIRContext*, mlir::MlirOptMainConfig const&amp;) MlirOptMain.cpp:0:0
#<!-- -->25 0x00005591006ed73d llvm::LogicalResult llvm::function_ref&lt;llvm::LogicalResult (std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, llvm::raw_ostream&amp;)&gt;::callback_fn&lt;mlir::MlirOptMain(llvm::raw_ostream&amp;, std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, mlir::DialectRegistry&amp;, mlir::MlirOptMainConfig const&amp;)::$_0&gt;(long, std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, llvm::raw_ostream&amp;) MlirOptMain.cpp:0:0
#<!-- -->26 0x0000559100794785 mlir::splitAndProcessBuffer(std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, llvm::function_ref&lt;llvm::LogicalResult (std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, llvm::raw_ostream&amp;)&gt;, llvm::raw_ostream&amp;, llvm::StringRef, llvm::StringRef) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41ef785)
#<!-- -->27 0x00005591006e7755 mlir::MlirOptMain(llvm::raw_ostream&amp;, std::unique_ptr&lt;llvm::MemoryBuffer, std::default_delete&lt;llvm::MemoryBuffer&gt;&gt;, mlir::DialectRegistry&amp;, mlir::MlirOptMainConfig const&amp;) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x4142755)
#<!-- -->28 0x00005591006e79ff mlir::MlirOptMain(int, char**, llvm::StringRef, llvm::StringRef, mlir::DialectRegistry&amp;) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x41429ff)
#<!-- -->29 0x00005591006e7d2e mlir::MlirOptMain(int, char**, llvm::StringRef, mlir::DialectRegistry&amp;) (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x4142d2e)
#<!-- -->30 0x00005590fd776a67 main (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11d1a67)
#<!-- -->31 0x00007fe98f5b4083 __libc_start_main /build/glibc-LcI20x/glibc-2.31/csu/../csu/libc-start.c:342:3
#<!-- -->32 0x00005590fd7765de _start (/data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt+0x11d15de)

[Prakhar-Dixit]

I was looking forward to working on this issue. Could you assign it to me?
@Lewuathe @ftynse @matthias-springer

[Prakhar-Dixit]

Detailed Findings and Request for Suggestions
I observed that when using a single affine.for loop to “unflatten” a 2D index (via two affine.apply operations using mod and floordiv) on a memref, the -affine-super-vectorize pass crashes with the following assertion:

assert(perm[kvp.second] == getAffineConstantExpr(0, context) &&
       "permutationMap already has an entry along dim");

This happens because the pass expects that each vectorized loop contributes exactly one non‑invariant index for constructing the permutation map. In the original IR, a single loop produces two different subscript indices (one computed via mod and one via floordiv) from the same induction variable, which violates that assumption.

Original IR :

#map = affine_map<(d0)[s0] -> (d0 mod s0)>
#map1 = affine_map<(d0)[s0] -> (d0 floordiv s0)>

module {
  memref.global "private" constant @__constant_17x18xf32 : memref<17x18xf32> = dense<0.000000e+00> {alignment = 64 : i64}
  memref.global "private" constant @__constant_17x18xi32 : memref<17x18xi32> = dense<0> {alignment = 64 : i64}
  func.func private @func3(%arg0: index) -> (memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>) {
    %c18 = arith.constant 18 : index
    %cst = arith.constant 1.000000e+00 : f32
    %0 = memref.get_global @__constant_17x18xi32 : memref<17x18xi32>
    %1 = memref.get_global @__constant_17x18xf32 : memref<17x18xf32>
    %alloc = memref.alloc() {alignment = 64 : i64} : memref<17x18xf32>
    affine.for %arg1 = 0 to 306 {
      %col = affine.apply #map(%arg1)[%c18]  
      %row = affine.apply #map1(%arg1)[%c18]  
      affine.store %cst, %alloc[%row, %col] : memref<17x18xf32>
    }
    return %1, %0, %alloc : memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>
  }
}

Modified IR (Using Nested Loops):

module {
  memref.global "private" constant @__constant_17x18xf32 : memref<17x18xf32> = dense<0.000000e+00> {alignment = 64 : i64}
  memref.global "private" constant @__constant_17x18xi32 : memref<17x18xi32> = dense<0> {alignment = 64 : i64}
  func.func private @func3(%arg0: index) -> (memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>) {
    %c17 = arith.constant 17 : index
    %c18 = arith.constant 18 : index
    %cst = arith.constant 1.000000e+00 : f32
    %0 = memref.get_global @__constant_17x18xi32 : memref<17x18xi32>
    %1 = memref.get_global @__constant_17x18xf32 : memref<17x18xf32>
    %alloc = memref.alloc() {alignment = 64 : i64} : memref<17x18xf32>
    affine.for %i = 0 to %c17 {
      affine.for %j = 0 to %c18 {
        affine.store %cst, %alloc[%i, %j] : memref<17x18xf32>
      }
    }
    return %1, %0, %alloc : memref<17x18xf32>, memref<17x18xi32>, memref<17x18xf32>
  }
}

In this version, I replaced the single loop with two nested loops. The outer loop iterates over the rows (0 to 17) and the inner loop iterates over the columns (0 to 18). Because each loop now contributes exactly one non‑invariant index, the vectorization pass can build the permutation map without conflict.

Minimal Example :

module {
  func.func private @func3(%arg0: index) -> memref<2x2xf32> {
    %cst = arith.constant 1.0 : f32
    %alloc = memref.alloc() : memref<2x2xf32>
    
affine.for %i = 0 to 2 {
  affine.for %j = 0 to 2 {
    affine.store %cst, %alloc[%i, %j] : memref<2x2xf32>
  }
}   
    return %alloc : memref<2x2xf32>
  }
}

I am a bit confused about the best long‑term solution. Currently, the affine super‑vectorize pass does not support having a single loop “unpack” into multiple indices. I can avoid the crash by restructuring the IR into nested loops (as shown above), which satisfies the pass’s requirement.

@bondhugula @matthias-springer

[Prakhar-Dixit]

Just a gentle reminder!
@bondhugula @matthias-springer

[matthias-springer]

I am not familiar with this pass. I would look up the recent editors of this file to find a point of contact.

[Prakhar-Dixit]

I am not familiar with this pass. I would look up the recent editors of this file to find a point of contact.

Okay, thanks for helping !

[ftynse]

I am a bit confused about the best long‑term solution. Currently, the affine super‑vectorize pass does not support having a single loop “unpack” into multiple indices. I can avoid the crash by restructuring the IR into nested loops (as shown above), which satisfies the pass’s requirement.

It's unclear to me whether this pass is still being used/maintained by anybody active. We should go for the short-term solution of turning the assertion into a user-visible diagnostic and either not vectorizing the loop (if possible at this point without breaking IR) or just failing the pass entirely in presence of such loops.

[Prakhar-Dixit]

I am a bit confused about the best long‑term solution. Currently, the affine super‑vectorize pass does not support having a single loop “unpack” into multiple indices. I can avoid the crash by restructuring the IR into nested loops (as shown above), which satisfies the pass’s requirement.

It's unclear to me whether this pass is still being used/maintained by anybody active. We should go for the short-term solution of turning the assertion into a user-visible diagnostic and either not vectorizing the loop (if possible at this point without breaking IR) or just failing the pass entirely in presence of such loops.

Thanks for the suggestions, I'll try to make the necessary fixes.