Move 100 ImageNet val class (each with 50 sampels) into data directory
Save preprocessed data and labels in input, and compute torchvision's ResNet18's acc
python torch_baseline.py
torch's acc: ~ 0.75.
Save the parameters of torchvision's ResNet18 to input/param.bin
python param.py
nvcc test.cu matrix.cu im2col.cu conv.cu layers.cu utils.cu sim.cu -o test -arch=sm_70
Run inference on 5000 images with batch size 16
./test 16
WMMA: warp-level matrix multiply and accumulate
GEMM: general matrix multiplication