error: 'func.func' op exceeded stack allocation limit of 32768 bytes for function. Got 229376 bytes

[pdhirajkumarprasad]

What happened?

For the given IR

module {
  func.func @"torch-jit-export"(%arg2: !torch.vtensor<[16,256,56,56],f32>) -> !torch.vtensor<[16,256,56,56],f32>  attributes {torch.onnx_meta.ir_version = 3 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "0.3"} {
    %10 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<256x4x3x3xf32>} : () -> !torch.vtensor<[256,4,3,3],f32> 
    %11 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<256xf32>} : () -> !torch.vtensor<[256],f32> 
    %12 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<256xf32>} : () -> !torch.vtensor<[256],f32> 
    %13 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<256xf32>} : () -> !torch.vtensor<[256],f32> 
    %14 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<256xf32>} : () -> !torch.vtensor<[256],f32> 
    %529 = torch.operator "onnx.Conv"(%arg2, %10) {torch.onnx.dilations = [1 : si64, 1 : si64], torch.onnx.group = 64 : si64, torch.onnx.kernel_shape = [3 : si64, 3 : si64], torch.onnx.pads = [1 : si64, 1 : si64, 1 : si64, 1 : si64], torch.onnx.strides = [1 : si64, 1 : si64]} : (!torch.vtensor<[16,256,56,56],f32>, !torch.vtensor<[256,4,3,3],f32>) -> !torch.vtensor<[16,256,56,56],f32> 
    %530 = torch.operator "onnx.BatchNormalization"(%529, %11, %12, %13, %14) {torch.onnx.epsilon = 9.99999974E-6 : f32, torch.onnx.momentum = 0.899999976 : f32} : (!torch.vtensor<[16,256,56,56],f32>, !torch.vtensor<[256],f32>, !torch.vtensor<[256],f32>, !torch.vtensor<[256],f32>, !torch.vtensor<[256],f32>) -> !torch.vtensor<[16,256,56,56],f32> 
    %531 = torch.operator "onnx.Relu"(%530) : (!torch.vtensor<[16,256,56,56],f32>) -> !torch.vtensor<[16,256,56,56],f32> 
    return %531 : !torch.vtensor<[16,256,56,56],f32>
  }
}

getting error as

../model.torch_onnx.mlir:10:12: error: 'func.func' op exceeded stack allocation limit of 32768 bytes for function. Got 229376 bytes
    %531 = torch.operator "onnx.Relu"(%530) : (!torch.vtensor<[16,256,56,56],f32>) -> !torch.vtensor<[16,256,56,56],f32> 
           ^

Steps to reproduce your issue

command:

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

Version: IREE compiler version 3.1.0rc20241127 @ 615e7ff

What component(s) does this issue relate to?

Compiler

Version information

No response

Additional context

No response

[ScottTodd]

Can you share more information about the program(s) that failed and put that information in the issue title? It's difficult to prioritize issues and expand test coverage without that context.

[pdhirajkumarprasad]

Can you share more information about the program(s) that failed and put that information in the issue title? It's difficult to prioritize issues and expand test coverage without that context.

We have ONNX zoo model regnetz_* which are failing with above error

[ScottTodd]

Okay, RegNetZ, a CNN-based model.

• Paper: https://arxiv.org/abs/2103.06877
• sample model zoo source (contains many for regnet, regnetx, regnety, regnetz, etc.): https://github.com/onnx/models/blob/main/Computer_Vision/regnetz_040_Opset16_timm/regnetz_040_Opset16.onnx