Continuously failing on test_conv3.tcl after modifying gen_weigths.py

[LeatherE]

Hi.
I am currently trying to test the conv layer in this library. But when I tried to modify the value(kernel_dim, stride, input_precision, ifm_channels, ofm_channels, ifm_dimension, ofm_dimension, activation_precision, simd, pe, w_precision) in the gen_weigths.py. It often failed at the stage of Instrumenting C test bench. I wonder that if there has any relation between each value?

[fpjentzsch]

Hi,
which errors do you see? Usually, these HLS functions assume that all parameters "match" or divide evenly. Otherwise, the behavior might be undefined or execution will fail outright. For example:

• SIMD must divide ifm_channels
• PE must divide ofm_channels
• ofm_dimension must match the kernel_dim, stride, and ifm_dimension, e.g. for ifm_dim=5, kernel_dim=3, stride=1 ofm_dim will be 3

[LeatherE]

Hi,
thanks for your answers. but i'm still facing the same error. below are my error message.

my parameters are:

• kernel_dim = 5
• stride = 1
• input_precision = 1
• ifm_channels = 6
• ofm_channels = 16
• ifm_dimension = 14
• ofm_dimension = 10
• activation_precision = 16
• expand = 1
• simd = 3
• pe = 2
• w_precision = 1
• mmv=1

thanks!

[LeatherE]

Hi,

Perhaps you can give me one usable parameter combination for one random layer in CNV-w1a1 which you used in bnn-pynq/cnv_end2end_example. So that I can test the throughput of one convolution layer.

Thanks!