Error converting UPMEM MLIR to UPMEM C

[fengjingge815]

Under Ubuntu 22.04, I encountered an error while installing the official command to compile and run CINM, and when converting UPMEM MLIR to UPMEM C, the specific information is as follows:

source:https://github.com/tud-ccc/Cinnamon
ubuntu：ubuntu22.04
compiler：gcc/g++11.4

cd Cinnamon //ok
just configure -no-torch-mlir //ok
./compile_benches.sh //err

Setting UPMEM_HOME to /home/test/src/llvm18-cinn/tmp/Cinnamon/upmem and updating PATH/LD_LIBRARY_PATH/PYTHONPATH
ninja -Ccinnamon/build cinm-opt
ninja: Entering directory `cinnamon/build'
ninja: no work to do.
Setting UPMEM_HOME to /home/fengjingge/src/llvm18-cinn/tmp/Cinnamon/upmem and updating PATH/LD_LIBRARY_PATH/PYTHONPATH
rm -rf generated2/va
../build/bin/cinm-opt va.mlir --cinm-tiling --affine-loop-unroll='unroll-full unroll-full-threshold=1' > generated2/va/irs/va.tiled.mlir
../build/bin/cinm-opt generated2/va/irs/va.tiled.mlir --convert-cinm-to-cnm --cnm-hoist-workgroups --canonicalize --cse > generated2/va/irs/va.cnm.mlir
../build/bin/cinm-opt generated2/va/irs/va.cnm.mlir --lower-affine --one-shot-bufferize='bufferize-function-boundaries function-boundary-type-conversion=identity-layout-map'
--convert-linalg-to-affine-loops --lower-affine --buffer-loop-hoisting --buffer-hoisting --canonicalize --cse > generated2/va/irs/va.cnm.bufferized.mlir
../build/bin/cinm-opt generated2/va/irs/va.cnm.bufferized.mlir --convert-cnm-to-upmem --cse --upmem-outline-kernel --upmem-dedup-kernels --cse > generated2/va/irs/va.upmem.mlir
../build/bin/cinm-opt generated2/va/irs/va.upmem.mlir --mlir-print-ir-after-failure --canonicalize
--convert-scf-to-cf --convert-cf-to-llvm --fold-memref-alias-ops --lower-affine --convert-arith-to-llvm
--convert-upmem-to-llvm
--expand-strided-metadata --memref-expand --finalize-memref-to-llvm --lower-affine --convert-arith-to-llvm
--convert-func-to-llvm=use-bare-ptr-memref-call-conv=true --cse --reconcile-unrealized-casts --llvm-legalize-for-export --canonicalize --cse \

generated2/va/irs/va.llvm.mlir
../build/bin/cinm-translate --mlir-to-llvmir generated2/va/irs/va.llvm.mlir > generated2/va/irs/va.host.ll
../../llvm/build/bin/opt -O3 -S generated2/va/irs/va.host.ll > generated2/va/irs/va.host.opt.ll
../../llvm/build/bin/llc -O3 generated2/va/irs/va.host.opt.ll -o generated2/va/irs/va.host.o -filetype=obj -relocation-model=pic
../build/bin/cinm-translate --mlir-to-upmem-cpp generated2/va/irs/va.upmem.mlir > generated2/va/irs/va.dpu.c
generated2/va/irs/va.upmem.mlir:4:3: error: Dialect `memref' not found for custom op 'memref.global'
memref.global "private" constant @__constant_2xi64_0 : memref<2xi64> = dense<[16384, 512]> {alignment = 64 : i64}
^
generated2/va/irs/va.upmem.mlir:4:3: note: Registered dialects: builtin, upmem ; for more info on dialect registration see https://mlir.llvm.org/getting_started/Faq/#registered-loaded-dependent-whats-up-with-dialects-management

va.upmem.mlir：
#map = affine_map<(d0, d1, d2) -> (d0 * 1024 + d1 * 16 + d2)>
2 #map1 = affine_map<(d0, d1, d2) -> (d0 * 512 + d1 * 8 + d2)>
3 module {
4 memref.global "private" constant @__constant_2xi64_0 : memref<2xi64> = dense<[16384, 512]> {alignment = 64 : i 64}
5 memref.global "private" constant @__constant_1xi64_1 : memref<1xi64> = dense<8388608> {alignment = 64 : i64}
6 memref.global "private" constant @__constant_16384x512xi32 : memref<16384x512xi32> = dense<0> {alignment = 64 : i64}
7 memref.global "private" constant @__constant_1xi64_0 : memref<1xi64> = dense<16777216> {alignment = 64 : i64}
8 memref.global "private" constant @__constant_2xi64 : memref<2xi64> = dense<[16384, 256]> {alignment = 64 : i64 }
9 memref.global "private" constant @__constant_1xi64 : memref<1xi64> = dense<4194304> {alignment = 64 : i64}
10 memref.global "private" constant @__constant_16384x256xi32 : memref<16384x256xi32> = dense<0> {alignment = 64 : i64}
......

[lvzc0]

I am getting the same issue.

[fengjingge815]

Modify cinnamon/lib/Target/UPMEMCpp/UPMEMTranslateToCpp.cpp
`void mlir::upmem_emitc::registerUPMEMCppTranslation() {
static llvm::cl::opt declareVarsAtTop(
"declare-vars-at-top",
llvm::cl::desc("Declare variables at top when emitting C/C++"),
llvm::cl::init(false));

TranslateFromMLIRRegistration reg(
"mlir-to-upmem-cpp", "translate from upmem's mlir to upmem's cpp",
[](Operation *op, raw_ostream &output) {
return upmem_emitc::UPMEMtranslateToCpp(
op, output,
/declareVariablesAtTop=/declareVarsAtTop);
},
[](DialectRegistry &registry) {
// clang-format off

    //Insert start
    registry.insert<mlir::memref::MemRefDialect>(); //Insert
    registry.insert<mlir::func::FuncDialect>(); //Insert
    registry.insert<mlir::scf::SCFDialect>();  //Insert
    //Insert end


    registry.insert<upmem::UPMEMDialect>();
    // clang-format on
  });

}`

Recompile to fix the above issues