Convolution op with scheduling in TVM

[rajshekharM]

For implementing convolution op in TVM, with scheduling, I have followed these examples :

https://docs.tvm.ai/tutorials/optimize/opt_conv_cuda.html

and this discussion link :

Implement Convolution Using TVM

Hi there, I am a first-time TVM user. I am also not very familiar with CUDA GPU programming. I want to write a tensor convolution routine using tvm. I wish the routine to be at least somewhat optimized. Here is a specification of my problem. 1. All the tensors are in NCHW form. 2. Vertical and horizontal padding and stride are required. 3. The code will be executed on an NVIDIA GPU with CUDA, cudnn, cublas etc. support. I am testing a new way of doing convolution, so I really need a base im…

I wanted to know how does the TVM compilation flow work (in terms of scheduling, etc.) on ‘coarse-grained’ operations such as convolution.

Is this code reduced to primitive schedule operators (https://docs.tvm.ai/tutorials/language/schedule_primitives.html), and then code generation done on that ?

Or does it generate code directly from the high-level operations?