Dynamic Object-FIFO lowering not functional

[andrej]

Hi @AndraBisca @hunhoffe, I hope you guys are well.

I'm looking at the new Object-FIFO tests. I believe none of these tests use the dynamic lowering, as the --dynamic-objFifos option is missing from the aiecc.py call. This is confirmed by looking at the lowered MLIR, which does not contain any of the scf.index_switches that the dynamic lowering is supposed to emit.

If I do try to enable the dynamic lowering by adding the command line option, compilation fails at the buffer address assignment stage (--aie-assign-buffer-addresses) with this error:

aie-opt: /project/llvm-project/mlir/lib/IR/Types.cpp:134: unsigned int mlir::Type::getIntOrFloatBitWidth() const: Assertion `isIntOrFloat() && "only integers and floats have a bitwidth"' failed.

Backtrace

PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: aie-opt "--pass-pipeline=builtin.module(lower-affine,aie-canonicalize-device,aie.device(aie-assign-lock-ids,aie-register-objectFifos,aie-objectFifo-stateful-transform{ dynamic-objFifos=1 },aie-assign-bd-ids,aie-lower-cascade-flows,aie-lower-broadcast-packet,aie-lower-multicast,aie-assign-tile-controller-ids,aie-generate-column-control-overlay{ route-shim-to-tile-ctrl=0 },aie-assign-buffer-addresses{ basic-alloc=0 }),convert-scf-to-cf)" ./aie2.mlir
 #0 0x00005642438bbf7b llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) Signals.cpp:0:0
 #1 0x00005642438b947b llvm::sys::RunSignalHandlers() Signals.cpp:0:0
 #2 0x00005642438b95a5 SignalHandler(int) Signals.cpp:0:0
 #3 0x00007f071e442520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x00007f071e4969fc __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x00007f071e4969fc __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x00007f071e4969fc pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x00007f071e442476 gsignal ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x00007f071e4287f3 abort ./stdlib/abort.c:81:7
 #9 0x00007f071e42871b _nl_load_domain ./intl/loadmsgcat.c:1177:9
#10 0x00007f071e439e96 (/lib/x86_64-linux-gnu/libc.so.6+0x39e96)
#11 0x00005642437d875d mlir::Type::getIntOrFloatBitWidth() const Types.cpp:0:0
#12 0x0000564241763e3a xilinx::AIE::BufferOp::getAllocationSize() AIEDialect.cpp:0:0
#13 0x00005642417e8e1d void std::__insertion_sort>(xilinx::AIE::BufferOp*, xilinx::AIE::BufferOp*, __gnu_cxx::__ops::_Iter_comp_iter) AIEAssignBuffers.cpp:0:0
#14 0x00005642417f2135 simpleBankAwareAllocation(xilinx::AIE::TileOp) AIEAssignBuffers.cpp:0:0
#15 0x00005642417f28b3 AIEAssignBufferAddressesPass::runOnOperation() AIEAssignBuffers.cpp:0:0
#16 0x00005642435f15ae mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) Pass.cpp:0:0
#17 0x00005642435f1a78 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) Pass.cpp:0:0
#18 0x00005642435f2123 mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::'lambda'(mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&)::operator()(mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool)::OpPMInfo&) const Pass.cpp:0:0
#19 0x00005642435f0aed mlir::detail::OpToOpPassAdaptor::runOnOperationAsyncImpl(bool) Pass.cpp:0:0
#20 0x00005642435f13df mlir::detail::OpToOpPassAdaptor::run(mlir::Pass*, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int) Pass.cpp:0:0
#21 0x00005642435f1a78 mlir::detail::OpToOpPassAdaptor::runPipeline(mlir::OpPassManager&, mlir::Operation*, mlir::AnalysisManager, bool, unsigned int, mlir::PassInstrumentor*, mlir::PassInstrumentation::PipelineParentInfo const*) Pass.cpp:0:0
#22 0x00005642435f2a05 mlir::PassManager::run(mlir::Operation*) Pass.cpp:0:0
#23 0x0000564241fe6ac7 performActions(llvm::raw_ostream&, std::shared_ptr const&, mlir::MLIRContext*, mlir::MlirOptMainConfig const&) MlirOptMain.cpp:0:0
#24 0x0000564241fe72d3 processBuffer(llvm::raw_ostream&, std::unique_ptr>, mlir::MlirOptMainConfig const&, mlir::DialectRegistry&, llvm::ThreadPoolInterface*) MlirOptMain.cpp:0:0
#25 0x0000564241fe7409 llvm::LogicalResult llvm::function_ref>, llvm::raw_ostream&)>::callback_fn>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&)::'lambda'(std::unique_ptr>, llvm::raw_ostream&)>(long, std::unique_ptr>, llvm::raw_ostream&) MlirOptMain.cpp:0:0
#26 0x00005642437f912f mlir::splitAndProcessBuffer(std::unique_ptr>, llvm::function_ref>, llvm::raw_ostream&)>, llvm::raw_ostream&, llvm::StringRef, llvm::StringRef) ToolUtilities.cpp:0:0
#27 0x0000564241fdf78c mlir::MlirOptMain(llvm::raw_ostream&, std::unique_ptr>, mlir::DialectRegistry&, mlir::MlirOptMainConfig const&) MlirOptMain.cpp:0:0
#28 0x0000564241fe7540 mlir::MlirOptMain(int, char**, llvm::StringRef, llvm::StringRef, mlir::DialectRegistry&) MlirOptMain.cpp:0:0
#29 0x0000564241fe7a2e mlir::MlirOptMain(int, char**, llvm::StringRef, mlir::DialectRegistry&) MlirOptMain.cpp:0:0
#30 0x000056423f8a8709 main aie-opt.cpp:0:0
#31 0x00007f071e429d90 __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:58:16
#32 0x00007f071e429e40 call_init ./csu/../csu/libc-start.c:128:20
#33 0x00007f071e429e40 __libc_start_main ./csu/../csu/libc-start.c:379:5
#34 0x000056423f896625 _start (/home/andre/mlir-aie/install/bin/aie-opt+0x197c625)

Please let me know if this is something I'm doing wrong. I'd be very interested in using this feature for my run-time-parametrized matrix multiplication.

[AndraBisca]

Fixed by PR above.