BLIS Retreat 2019

Contributed talks

• Cris Cecka, NVIDIA
  Title: Programming GPUs for speed-of-light linear algebra

• Marat Dukhan, Google Research
  Title: Indirect GEMM and Indirect Convolution Algorithm
  
  Abstract: Deep learning frameworks commonly implement convolution operators with GEMM-based algorithms. In these algorithms, convolution is implemented on top of GEMM primitive, provided by highly optimized BLAS libraries. Convolutions with 1x1 kernels can be directly represented as a GEMM call, but convolutions with larger kernels require a special memory layout transformation - im2col or im2row - to fit into GEMM interface. The Indirect Convolution algorithm provides the efficiency of the GEMM primitive without the overhead of im2col transformation. In contrast to GEMM-based algorithms, the Indirect Convolution does not reshuffle the data to fit into the GEMM primitive but introduces an indirection buffer --a buffer of pointers to the start of each row of image pixels. This broadens the application of our modified GEMM function to convolutions with arbitrary kernel size, padding, stride, and dilation.
  Paper on ArXiv .

• Albert Cohen, Google
  Title: What MLIR has to offer as a toolkit for building and leveraging numerical libraries in ML and HPC

• Greg Henry (Intel)
  Title: TBD

• Thomas Hines, Tennessee Tech
  Title: Issues with fat by thin matrix multiplication

• Jianyu Huang, Facebook
  Title: FBGEMM: High-Performance Low-Precision Library for Deep Learning Inference

• Tze Meng Low, CMU
  Analytical models for MMM-like problems on GPUs

• Devin Matthews, SMU
  Title: GEMM-Based Kernels for Tensor Hypercontraction

• John McCalpin, TACC
  Title: What You Don’t Know Can Hurt Performance — Snoop Filters in Intel Xeon Scalable Processors

• Maggie Myers, UT-Austin
  Title: Educational outreach

• Devangi Parikh, UT-Austin
  Title: What BLIS brings to the table--performance, portability, and productivity

• Christos Psarras, RWTH-Aachen
  Title: The Linear Algebra Mapping Problem

• Martin Schatz, Facebook
  Title: FLAME in Machine Learning (ML) Applications

• Brian Shi (Intel)
  Title: TBD

• Tyler Smith, ETH-Zurich
  Title: I/O Lower Bounds for Small MMM

• Nicholai Tukanov, UT-Austin
  Title: Mapping BLIS to the IBM Power9 architecture

• Field Van Zee, UT-Austin
  Title: The BLIS Approach to Skinny Matrix Multiplication

• Kiran Varaganti, AMD
  Title: BLIS optimizations and results on AMD Rome (tentative title)
  

---