squall/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
Duane Merrill 2017-12-05 20:50:15 -05:00 committed by GitHub
parent 8ebd6b06d0
commit f30abfc00a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
README.md
View File

@ -2,13 +2,20 @@
# Introduction
CUTLASS is a CUDA C++ template library for implementing matrix-multiply
procedures that may be instantiated in CUDA device kernels. CUTLASS applies
object-oriented and generic programming techniques to maximize flexibility of
the resulting code and facilitate composition with caller-supplied functionality.
CUDA C++ templates are used to specify policy decisions such as block sizes,
data types of input and accumulator operands, and element-wise operations applied
to the results of matrix multiply.
CUTLASS is a collection of templated CUDA C++ abstractions for implementing
high-performance matrix-multiplication (GEMM) at all levels and scales within CUDA.
It incorporates the same stragies for data movemement and hierarchical decomposition
that are used to implement cuBLAS. CUTLASS decomposes these “moving parts” into
reusabe, modular software components abstracted by C++ template classes. These
thread-wide, warp-wide, block-wide, and device-wide abstractions can be specialized
by custom tiling sizes, data types, and other algorithmic policy. This flexibility
allows them to be used as building blocks within custom kernels and applications.
To support a wide variety of applications, CUTLASS provides extensive support for
mixed-precision computations, providing specialized data-movement and
multiply-accumulate abstractions for 8-bit integer, half-precision floating
point (FP16), single-precision floating point (FP32), and double-precision floating
point (FP64) types.
# Project Structure