Merge pull request chenzomi12#274 from AndroidHot/patch-1

将 1x1x16 的卷积核进行零填充扩展为 3x3x16 的卷积核，以得到和 1x1x16 同样的卷积输出。

03Compiler/03Frontend/03OPFusion.md

@@ -68,7 +68,8 @@ if __name__ == "__main__":
 
 	# kernel fusion
 	# 将 conv2 的卷积核权重由(1, 1)扩展到(3, 3)
-	weight_expanded = F.interpolate(conv2_weight, size=(3, 3), mode='bilinear', align_corners=False)
+	weight_expanded = torch.zeros(16, 3, 3, 3)
+	weight_expanded[:, :, 1, 1] = conv2_weight[:, :, 0, 0]
 	# conv1 卷积核与 conv2 卷积核融合
 	weight_fusion = torch.concatenate([conv1_weight, weight_expanded], dim=0)
 	# conv1 偏置与 conv2 偏置融合