Marsjunwang

I try to increase the -1 dim ,the big kernal becomes more and more slower than conv2d

1

if torch.cuda.is_available(): x = torch.randn(64, 384, 256, 31).cuda() m1 = DepthWiseConv2dImplicitGEMM(384, 31, bias=False).cuda() m2 = nn.Conv2d(384, 384, 31, padding=31 // 2, bias=False, groups=384).cuda() m2.load_state_dict(m1.state_dict()) with torch.cuda.amp.autocast(True): import time t1 =...