failed to legalize operation 'torch.operator' that was explicitly marked illegal: onnx.QLinearConv

[pdhirajkumarprasad]

For given IR, it failed to legalize: onnx.QLinearConv

module {
  func.func @mxnet_converted_model(%arg0: !torch.vtensor<[?,3,224,224],f32>) -> !torch.vtensor<[?,64,112,112],ui8>  attributes {torch.onnx_meta.ir_version = 4 : si64, torch.onnx_meta.opset_version = 21 : si64, torch.onnx_meta.opset_versions = {ai.onnx.ml = 2 : si64, ai.onnx.preview.training = 1 : si64, ai.onnx.training = 1 : si64, com.microsoft = 1 : si64, com.microsoft.experimental = 1 : si64, com.microsoft.mlfeaturizers = 1 : si64, com.microsoft.nchwc = 1 : si64}, torch.onnx_meta.producer_name = "onnx.quantize", torch.onnx_meta.producer_version = "0.1.0"} {
    %0 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<114> : tensor<ui8>} : () -> !torch.vtensor<[],ui8> 
    %1 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.018658448> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
    %2 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1> : tensor<64x3x7x7xsi8>} : () -> !torch.vtensor<[64,3,7,7],si8> 
    %3 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1.0> : tensor<64xf32>} : () -> !torch.vtensor<[64],f32> 
    %4 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0> : tensor<64xsi8>} : () -> !torch.vtensor<[64],si8> 
    %5 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<1> : tensor<64xsi32>} : () -> !torch.vtensor<[64],si32> 
    %6 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0> : tensor<ui8>} : () -> !torch.vtensor<[],ui8> 
    %7 = torch.operator "onnx.Constant"() {torch.onnx.value = dense<0.0215634871> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
    %8 = torch.operator "onnx.QuantizeLinear"(%arg0, %1, %0) : (!torch.vtensor<[?,3,224,224],f32>, !torch.vtensor<[],f32>, !torch.vtensor<[],ui8>) -> !torch.vtensor<[?,3,224,224],ui8> 
    %9 = torch.operator "onnx.QLinearConv"(%8, %1, %0, %2, %3, %4, %7, %6, %5) {torch.onnx.auto_pad = "NOTSET", torch.onnx.dilations = [1 : si64, 1 : si64], torch.onnx.group = 1 : si64, torch.onnx.kernel_shape = [7 : si64, 7 : si64], torch.onnx.pads = [3 : si64, 3 : si64, 3 : si64, 3 : si64], torch.onnx.strides = [2 : si64, 2 : si64]} : (!torch.vtensor<[?,3,224,224],ui8>, !torch.vtensor<[],f32>, !torch.vtensor<[],ui8>, !torch.vtensor<[64,3,7,7],si8>, !torch.vtensor<[64],f32>, !torch.vtensor<[64],si8>, !torch.vtensor<[],f32>, !torch.vtensor<[],ui8>, !torch.vtensor<[64],si32>) -> !torch.vtensor<[?,64,112,112],ui8> 
    return %9 : !torch.vtensor<[?,64,112,112],ui8>
  }
}

model impacted: 19

mobilenetv2-12-int8
vgg16-12-int8
resnet50-v1-12-int8
zfnet512-12-int8
squeezenet1.0-12-int8
bvlcalexnet-12-int8
caffenet-12-int8
densenet-12-int8
googlenet-12-int8
inception-v1-12-int8
mnist-12-int8
shufflenet-v2-12-int8
MaskRCNN-12-int8
ResNet101-DUC-12-int8
ssd-12-int8
FasterRCNN-12-int8
fcn-resnet50-12-int8
version-RFB-320-int8
arcfaceresnet100-11-int8

[vivekkhandelwal1]

Partially fixed by llvm/torch-mlir#3917. The Onnx->Torch path is fixed by this PR.