hls

HLS Project

The source code of the HLS-based accelerator is under top subdirection. We also provide an archive of the HLS project containing the CoDeNet accelerator code here.

Please unzip the file unzip CoDeNet_hls.zip, open Vivado HLS GUI, and load the project.

The Vivado Project and Block Design files can be downloaded here.

Please load the archived project and update the HLS accelerator path.

Test Bench usage:

We provide some test bench data for verfication of a single layer

for conv 1x1:

D = 32;        
IC = 64;
OC = 16;
skip1 = 0; 
skip3 = 1; 
stride_2 = 0;
deform = 0;
readfile<FM_W, PA_0>(fmap_in, D*D*IC, "D_32_IC_64_OC_16_k_1_PA_16_PE_16input.txt");
readfilepack<S_W, 1>(quant, OC*2, "D_32_IC_64_OC_16_k_1_PA_16_PE_16quant.txt");
readfile<W_W, PA_0>(wtemp1,  IC * OC, "D_32_IC_64_OC_16_k_1_PA_16_PE_16weight1.txt");

for conv 3x3:

D = 32;       
IC = 64;
OC = 64;
skip1 = 1; 
skip3 = 0; 
stride_2 = 0;
deform = 0;
readfile<FM_W, PA_0>(fmap_in, D * D * IC, "D_32_C_64_k_3_PA_16_PE_16input.txt");
readfilepack<S_W, 1>(quant, OC * 2, "D_32_C_64_k_3_PA_16_PE_16quant.txt");
readfile<W_W, PA_0>(wtemp3,  IC * 9, "D_32_C_64_k_3_PA_16_PE_16weight3.txt");