Literature reading (245) roller

• subject ：Roller: Fast and Efficient Tensor Compilation for Deep Learning
• Time ：2022
• meeting ：OSDI
• Research Institute ： Microsoft

The main of this paper motivation Now DNN It takes a long time to explore the compilation of , Especially for Nivida Other hardware platforms such as AMD GPU and Graphcore IPU, So this paper has a different idea , Generate by construction kernel, First, introduce the basic concepts ：

• rTile： The most basic level of abstraction , It's just one. data tile, However, it corresponds to the basic size of calculation and memory access
• rProgram： It includes load、store、compute Based on the rTile The program , You can meet GPU Medium SM perform
• kernel： utilize rProgram structure kernel

1. rTile is a new tile abstraction that encapsulates tensor shapes that align with the key features of the underlying accelerator, thus achieving efficient execution by limiting the shape choices.
2. rProgram: adopts a recursive rTile-based construction algorithm to gradually increase the size of the rTile shape to construct an rProgram that saturates a single execution unit of the accelerator (e.g., an SM, a streaming multi-processor in a NVIDIA GPU)
3. kernel: performs the scale-out process, which simply replicates the resulting rProgram to other parallel execution units