Deep Learning HDL Toolbox - Error using dnnfpga.co​mpiler.cod​egenfpga Index exceeds the number of array elements. Index must not exceed 0.

Hello,

I have a Nerual Network imported from keras (TF 2.6). The model imported is working and I would like to configure an Intel Arria 10 FPGA using the Deep Learning HDL Toolbox. However, I get the following error while model is compiling with hW.compile:

### Compiling network for Deep Learning FPGA prototyping ...
### Targeting FPGA bitstream arria10soc_single.
### Optimizing network: Flatten layer ('flatten_1'), removed and fused into FC layer ('dense_1'). Weights of FC layer were modified to account for the flatten layer removal.
### Optimizing network: Fused 'nnet.keras.layer.ZeroPadding2dLayer' into 'nnet.cnn.layer.Convolution2DLayer'
### Notice: The layer 'conv2d_transpose' of type 'nnet.cnn.layer.TransposedConvolution2DLayer' is split into 'conv2d_transpose_insertZeros' and 'conv2d_transpose'.
### Notice: The layer 'conv2d_transpose_1' of type 'nnet.cnn.layer.TransposedConvolution2DLayer' is split into 'conv2d_transpose_1_insertZeros' and 'conv2d_transpose_1'.
### Notice: The layer 'conv2d_transpose_2' of type 'nnet.cnn.layer.TransposedConvolution2DLayer' is split into 'conv2d_transpose_2_insertZeros' and 'conv2d_transpose_2'.
### The network includes the following layers:
     1   'input_1'                 Image Input                                   1×1×4 images                                                                       (SW Layer)
     2   'dense'                   Fully Connected                               30 fully connected layer                                                           (HW Layer)
     3   'batch_normalization'     Batch Normalization                           Batch normalization with 30 channels                                               (HW Layer)
     4   'Sigmoid1'                Sigmoid                                       sigmoid                                                                            (SW Layer)
     5   'multiply'                nnet.internal.cnn.coder.MultiplicationLayer   nnet.internal.cnn.coder.MultiplicationLayer                                        (HW Layer)
     6   'conv2d_transpose'        2-D Transposed Convolution                    1600 4×4×30 transposed convolutions with stride [1  1] and cropping [0  0  0  0]   (HW Layer)
     7   'batch_normalization_1'   Batch Normalization                           Batch normalization with 1600 channels                                             (HW Layer)
     8   'Sigmoid2'                Sigmoid                                       sigmoid                                                                            (SW Layer)
     9   'multiply_1'              nnet.internal.cnn.coder.MultiplicationLayer   nnet.internal.cnn.coder.MultiplicationLayer                                        (HW Layer)
    10   'conv2d_transpose_1'      2-D Transposed Convolution                    800 4×4×1600 transposed convolutions with stride [1  1] and cropping [0  0  0  0]  (HW Layer)
    11   'batch_normalization_2'   Batch Normalization                           Batch normalization with 800 channels                                              (HW Layer)
    12   'Sigmoid3'                Sigmoid                                       sigmoid                                                                            (SW Layer)
    13   'multiply_2'              nnet.internal.cnn.coder.MultiplicationLayer   nnet.internal.cnn.coder.MultiplicationLayer                                        (HW Layer)
    14   'conv2d_transpose_2'      2-D Transposed Convolution                    400 4×4×800 transposed convolutions with stride [1  1] and cropping [0  0  0  0]   (HW Layer)
    15   'batch_normalization_3'   Batch Normalization                           Batch normalization with 400 channels                                              (HW Layer)
    16   'Sigmoid4'                Sigmoid                                       sigmoid                                                                            (SW Layer)
    17   'multiply_3'              nnet.internal.cnn.coder.MultiplicationLayer   nnet.internal.cnn.coder.MultiplicationLayer                                        (HW Layer)
    18   'conv2d'                  2-D Convolution                               200 3×3×400 convolutions with stride [1  1] and padding [1  1  1  1]               (HW Layer)
    19   'Sigmoid5'                Sigmoid                                       sigmoid                                                                            (SW Layer)
    20   'multiply_4'              nnet.internal.cnn.coder.MultiplicationLayer   nnet.internal.cnn.coder.MultiplicationLayer                                        (HW Layer)
    21   'conv2d_1'                2-D Convolution                               1 3×3×200 convolutions with stride [1  1] and padding [1  1  1  1]                 (HW Layer)
    22   'dense_1'                 Fully Connected                               100 fully connected layer                                                          (HW Layer)
    23   'Sigmoid'                 Sigmoid                                       sigmoid                                                                            (SW Layer)
    24   'output'                  Classification Output                         crossentropyex with '1' and 99 other classes                                       (SW Layer)
                                                                                                                                                                  
### Notice: The layer 'input_1' with type 'nnet.cnn.layer.ImageInputLayer' is implemented in software.
### Notice: The layer 'Sigmoid1' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'Sigmoid2' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'Sigmoid3' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'Sigmoid4' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'Sigmoid5' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'Sigmoid' with type 'nnet.cnn.layer.SigmoidLayer' is implemented in software.
### Notice: The layer 'output' with type 'nnet.cnn.layer.ClassificationOutputLayer' is implemented in software.
### Compiling layer group: dense ...
### Compiling layer group: dense ... complete.
### Compiling layer group: batch_normalization ...
### Optimizing network: Replaced Batch Normalization layer batch_normalization with an equivalent Convolution layer.
### Compiling layer group: batch_normalization ... complete.
### Compiling layer group: conv2d_transpose_insertZeros ...
### Compiling layer group: conv2d_transpose_insertZeros ... complete.
### Compiling layer group: conv2d_transpose>>batch_normalization_1 ...
### Optimizing network: Fused 'nnet.cnn.layer.BatchNormalizationLayer' into 'nnet.cnn.layer.Convolution2DLayer'
### Compiling layer group: conv2d_transpose>>batch_normalization_1 ... complete.
### Compiling layer group: conv2d_transpose_1_insertZeros ...
### Compiling layer group: conv2d_transpose_1_insertZeros ... complete.
### Compiling layer group: conv2d_transpose_1>>batch_normalization_2 ...
### Optimizing network: Fused 'nnet.cnn.layer.BatchNormalizationLayer' into 'nnet.cnn.layer.Convolution2DLayer'
### Compiling layer group: conv2d_transpose_1>>batch_normalization_2 ... complete.
### Compiling layer group: conv2d_transpose_2_insertZeros ...
### Compiling layer group: conv2d_transpose_2_insertZeros ... complete.
### Compiling layer group: conv2d_transpose_2>>batch_normalization_3 ...
### Optimizing network: Fused 'nnet.cnn.layer.BatchNormalizationLayer' into 'nnet.cnn.layer.Convolution2DLayer'
### Compiling layer group: conv2d_transpose_2>>batch_normalization_3 ... complete.
### Compiling layer group: conv2d ...
### Compiling layer group: conv2d ... complete.
### Compiling layer group: conv2d_1 ...
### Compiling layer group: conv2d_1 ... complete.
### Compiling layer group: dense_1 ...
### Compiling layer group: dense_1 ... complete.
Error using dnnfpga.compiler.codegenfpga
Index exceeds the number of array elements. Index must not exceed 0.

Error in dnnfpga.apis.Workflow/compileNetwork

Error in dnnfpga.apis.Workflow/compile

Error in generate_HDL_from_NN (line 18)
    hW.compile;

Could anyone help me to find the rootcause?

You can find the model imported here nn.mat

---------------------------

Software version:

Deep Learning HDL Toolbox Support Package For Intel FPGA And SoC Devices 23.1.1

Matlab R2023a Update 2 (9.14.0.2254940)

Ubuntu 20.04

---------------------------

Answer 1

[Kiran Kintali]

This is not an expected error message. Please reach out to tech support for help and any available workaround.