error: One or more operations with large vector sizes (16384 bytes) were found

[pdhirajkumarprasad]

What happened?

For the given IR

module {
  func.func @torch_jit(%arg0: !torch.vtensor<[1,3,224,224],f32>,  %arg2: !torch.vtensor<[?,768,?],f32>) -> !torch.vtensor<[?,1000],f32> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "1.12.1"} {
    %2 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.0> : tensor<384x192x2x2xf32>} : () -> !torch.vtensor<[384,192,2,2],f32> 
    %3 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.0> : tensor<384xf32>} : () -> !torch.vtensor<[384],f32> 
    %4 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.0> : tensor<1000x768xf32>} : () -> !torch.vtensor<[1000,768],f32> 
    %5 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.0> : tensor<1000xf32>} : () -> !torch.vtensor<[1000],f32> 
    %1 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0> : tensor<1xsi64>} : () -> !torch.vtensor<[1],si64> 
    %6 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<_onnx__Sub_3847> : tensor<si64>} : () -> !torch.vtensor<[],si64> 
    %none = torch.constant.none
    %7 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<_> : tensor<si64>} : () -> !torch.vtensor<[],si64> 
    %8 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__1> : tensor<si64>} : () -> !torch.vtensor<[],si64> 
    %9 = torch.operator "onnx.Mod"(%7, %8) {torch.onnx.fmod = 0 : si64} : (!torch.vtensor<[],si64>, !torch.vtensor<[],si64>) -> !torch.vtensor<[],si64> 
    %10 = torch.operator "onnx.Concat"(%1, %1) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[2],si64> 
    %11 = torch.operator "onnx.Unsqueeze"(%9, %1) : (!torch.vtensor<[],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1],si64> 
    %12 = torch.operator "onnx.Concat"(%1, %11) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[2],si64> 
    %13 = torch.operator "onnx.Shape"(%10) : (!torch.vtensor<[2],si64>) -> !torch.vtensor<[1],si64> 
    %14 = torch.operator "onnx.Gather"(%13, %1) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1],si64> 
    %15 = torch.operator "onnx.Cast"(%12) {torch.onnx.to = 7 : si64} : (!torch.vtensor<[2],si64>) -> !torch.vtensor<[2],si64> 
    %16 = torch.operator "onnx.ConstantOfShape"(%1) {torch.onnx.value = dense_resource<__7> : tensor<1xsi64>} : (!torch.vtensor<[1],si64>) -> !torch.vtensor<[6],si64> 
    %17 = torch.operator "onnx.Concat"(%15, %16) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[2],si64>, !torch.vtensor<[6],si64>) -> !torch.vtensor<[8],si64> 
    %18 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__8> : tensor<2xsi64>} : () -> !torch.vtensor<[2],si64> 
    %19 = torch.operator "onnx.Reshape"(%17, %18) : (!torch.vtensor<[8],si64>, !torch.vtensor<[2],si64>) -> !torch.vtensor<[4,2],si64> 
    %20 = torch.operator "onnx.Slice"(%19, %1, %1, %1, %1) : (!torch.vtensor<[4,2],si64>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[4,2],si64> 
    %21 = torch.operator "onnx.Transpose"(%20) {torch.onnx.perm = [1 : si64, 0 : si64]} : (!torch.vtensor<[4,2],si64>) -> !torch.vtensor<[2,4],si64> 
    %22 = torch.operator "onnx.Reshape"(%21, %1) : (!torch.vtensor<[2,4],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[8],si64> 
    %23 = torch.operator "onnx.Cast"(%22) {torch.onnx.to = 7 : si64} : (!torch.vtensor<[8],si64>) -> !torch.vtensor<[8],si64> 
    %24 = torch.operator "onnx.Pad"(%arg0, %23, %none) {torch.onnx.mode = "constant"} : (!torch.vtensor<[1,3,224,224],f32>, !torch.vtensor<[8],si64>, !torch.none) -> !torch.vtensor<[?,?,?,?],f32> 
    %25 = torch.operator "onnx.Conv"(%24, %2, %3) {torch.onnx.dilations = [1 : si64, 1 : si64], torch.onnx.group = 1 : si64, torch.onnx.kernel_shape = [2 : si64, 2 : si64], torch.onnx.pads = [0 : si64, 0 : si64, 0 : si64, 0 : si64], torch.onnx.strides = [2 : si64, 2 : si64]} : (!torch.vtensor<[?,?,?,?],f32>, !torch.vtensor<[384,192,2,2],f32>, !torch.vtensor<[384],f32>) -> !torch.vtensor<[?,384,?,?],f32> 
    %26 = torch.operator "onnx.Shape"(%25) : (!torch.vtensor<[?,384,?,?],f32>) -> !torch.vtensor<[4],si64> 
    %27 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__364> : tensor<si64>} : () -> !torch.vtensor<[],si64> 
    %28 = torch.operator "onnx.Gather"(%26, %27) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[4],si64>, !torch.vtensor<[],si64>) -> !torch.vtensor<[],si64> 
    %29 = torch.operator "onnx.Unsqueeze"(%28, %1) : (!torch.vtensor<[],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[1],si64> 
    %30 = torch.operator "onnx.Concat"(%29, %1, %1, %29) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>, !torch.vtensor<[1],si64>) -> !torch.vtensor<[4],si64> 
    %31 = torch.operator "onnx.Reshape"(%arg2, %30) : (!torch.vtensor<[?,768,?],f32>, !torch.vtensor<[4],si64>) -> !torch.vtensor<[?,768,?,?],f32> 
    %32 = torch.operator "onnx.GlobalAveragePool"(%31) : (!torch.vtensor<[?,768,?,?],f32>) -> !torch.vtensor<[?,768,1,1],f32> 
    %33 = torch.operator "onnx.Flatten"(%32) {torch.onnx.axis = 1 : si64} : (!torch.vtensor<[?,768,1,1],f32>) -> !torch.vtensor<[?,768],f32> 
    %34 = torch.operator "onnx.Gemm"(%33, %4, %5) {torch.onnx.alpha = 1.000000e+00 : f32, torch.onnx.beta = 1.000000e+00 : f32, torch.onnx.transB = 1 : si64} : (!torch.vtensor<[?,768],f32>, !torch.vtensor<[1000,768],f32>, !torch.vtensor<[1000],f32>) -> !torch.vtensor<[?,1000],f32> 
    return %34 : !torch.vtensor<[?,1000],f32>
  }
}

{-#
  dialect_resources: {
    builtin: {
      _onnx__Sub_3847: "0x080000000800000000000000",
      __364: "0x080000000000000000000000",
      __7: "0x080000000000000000000000",
      __8: "0x08000000FFFFFFFFFFFFFFFF0200000000000000",
      __1: "0x080000000400000000000000",
      _: "0x08000000E000000000000000"
    }
  }
#-}

getting error as

model.torch_onnx.mlir:37:11: error: One or more operations with large vector sizes (16384 bytes) were found:

    %34 = torch.operator "onnx.Gemm"(%33, %4, %5) {torch.onnx.alpha = 1.000000e+00 : f32, torch.onnx.beta = 1.000000e+00 : f32, torch.onnx.transB = 1 : si64} : (!torch.vtensor<[?,768],f32>, !torch.vtensor<[1000,768],f32>, !torch.vtensor<[1000],f32>) -> !torch.vtensor<[?,1000],f32> 

iree version:

IREE (https://iree.dev):
IREE compiler version 20241105.1069 @ 17bba14
LLVM version 20.0.0git
Optimized build

Steps to reproduce your issue

command:

iree-compile --iree-hal-target-backends=llvm-cpu model.torch_onnx.mlir -o abc.vmfb --iree-llvmcpu-target-cpu=host

dump with '--mlir-print-ir-after-all --mlir-print-ir-before-all --mlir-disable-threading --mlir-elide-elementsattrs-if-larger=4'

dump.zip

What component(s) does this issue relate to?

Compiler

Version information

No response

Additional context

No response

[pashu123]

Fails with a different error in the latest main

<unknown>:0: error: invalid size -9223372036854775808 for !torch.tensor type
iree-compile: iree/third_party/llvm-project/mlir/include/mlir/IR/StorageUniquerSupport.h:180: static ConcreteT mlir::detail::StorageUserBase<mlir::torch::Torch::ValueTensorType, mlir::torch::Torch::BaseTensorType, mlir::torch::Torch::detail::ValueTensorTypeStorage, mlir::detail::TypeUniquer>::get(MLIRContext *, Args &&...) [ConcreteT = mlir::torch::Torch::ValueTensorType, BaseT = mlir::torch::Torch::BaseTensorType, StorageT = mlir::torch::Torch::detail::ValueTensorTypeStorage, UniquerT = mlir::detail::TypeUniquer, Traits = <>, Args = <std::optional<llvm::ArrayRef<long>> &, mlir::Type &, mlir::Attribute &>]: Assertion `succeeded( ConcreteT::verifyInvariants(getDefaultDiagnosticEmitFn(ctx), args...))' failed.
Please report issues to https://github.com/iree-org/iree/issues and include the crash backtrace.
Stack dump:
0.	Program arguments: iree-compile --iree-hal-target-backends=llvm-cpu new_onnx.mlir -o abc.vmfb --iree-llvmcpu-target-cpu=host
 #0 0x000076aef4672168 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /home/prashant/iree/third_party/llvm-project/llvm/lib/Support/Unix/Signals.inc:723:13
 #1 0x000076aef4670380 llvm::sys::RunSignalHandlers() /home/prashant/iree/third_party/llvm-project/llvm/lib/Support/Signals.cpp:106:18
 #2 0x000076aef4672808 SignalHandler(int) /home/prashant/iree/third_party/llvm-project/llvm/lib/Support/Unix/Signals.inc:413:1
 #3 0x000076aeee842520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x000076aeee8969fc __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x000076aeee8969fc __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x000076aeee8969fc pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x000076aeee842476 gsignal ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x000076aeee8287f3 abort ./stdlib/abort.c:81:7
 #9 0x000076aeee82871b _nl_load_domain ./intl/loadmsgcat.c:1177:9
#10 0x000076aeee839e96 (/lib/x86_64-linux-gnu/libc.so.6+0x39e96)
#11 0x000076aef5669f3c (/home/prashant/iree-build/lib/libIREECompiler.so+0x6669f3c)
#12 0x000076aef5666bb1 mlir::torch::Torch::ValueTensorType::get(mlir::MLIRContext*, std::optional<llvm::ArrayRef<long>>, mlir::Type, mlir::Attribute) /home/prashant/iree-build/compiler/plugins/input/Torch/torch-mlir/Dialect/Torch/IR/TorchTypes.cpp.inc:382:3
#13 0x000076aef5434fb9 mlir::TypeStorage::getAbstractType() /home/prashant/iree/third_party/llvm-project/mlir/include/mlir/IR/TypeSupport.h:173:5
#14 0x000076aef5434fb9 mlir::Type::getTypeID() /home/prashant/iree/third_party/llvm-project/mlir/include/mlir/IR/Types.h:117:37
#15 0x000076aef5434fb9 bool mlir::detail::StorageUserBase<mlir::torch::Torch::NonValueTensorType, mlir::torch::Torch::BaseTensorType, mlir::torch::Torch::detail::NonValueTensorTypeStorage, mlir::detail::TypeUniquer>::classof<mlir::Type>(mlir::Type) /home/prashant/iree/third_party/llvm-project/mlir/include/mlir/IR/StorageUniquerSupport.h:113:16
#16 0x000076aef5434fb9 llvm::CastInfo<mlir::torch::Torch::NonValueTensorType, mlir::Type const, void>::isPossible(mlir::Type) /home/prashant/iree/third_party/llvm-project/mlir/include/mlir/IR/Types.h:423:14
#17 0x000076aef5434fb9 bool llvm::isa<mlir::torch::Torch::NonValueTensorType, mlir::Type>(mlir::Type const&) /home/prashant/iree/third_party/llvm-project/llvm/include/llvm/Support/Casting.h:549:10
#18 0x000076aef5434fb9 bool llvm::isa<mlir::torch::Torch::NonValueTensorType, mlir::torch::Torch::ValueTensorType, mlir::Type>(mlir::Type const&) /home/prashant/iree/third_party/llvm-project/llvm/include/llvm/Support/Casting.h:554:10
#19 0x000076aef5434fb9 mlir::torch::Torch::BaseTensorType::classof(mlir::Type) /home/prashant/iree/third_party/torch-mlir/include/torch-mlir/Dialect/Torch/IR/TorchTypes.h:163:10
#20 0x000076aef5434fb9 llvm::CastInfo<mlir::torch::Torch::BaseTensorType, mlir::Type const, void>::isPossible(mlir::Type) /home/prashant/iree/third_party/llvm-project/mlir/include/mlir/IR/Types.h:423:14
#21 0x000076aef5434fb9 bool llvm::isa<mlir::torch::Torch::BaseTensorType, mlir::Type>(mlir::Type const&) /home/prashant/iree/third_party/llvm-project/llvm/include/llvm/Support/Casting.h:549:10
#22 0x000076aef5434fb9 decltype(auto) llvm::cast<mlir::torch::Torch::BaseTensorType, mlir::Type>(mlir::Type const&) /home/prashant/iree/third_party/llvm-project/llvm/include/llvm/Support/Casting.h:566:3
#23 0x000076aef5434fb9 refineShapeCalculateResult(mlir::torch::Torch

[pdhirajkumarprasad]

@pashu123 this issue is there even with today's build so likely some local change in your sandbox?

IREE (https://iree.dev):
IREE compiler version 20241107.1071 @ e4f3960
LLVM version 20.0.0git
Optimized build