简体繁体 English

为什么CuBlas不对三角形矩阵使用一维数组？

[英]Why does CuBlas not use a 1d array for triangular matrices?

原文 2018-07-05 08:44:52 0 1 cuda/ cublas

This might be a throwback to the old BLAS library design, but I was surprised just now to find that CuBlas wastes memory by using regular 2d arrays for triangular matrices. 这可能是对旧的BLAS库设计的一种回想，但我刚才感到惊讶的是，发现CuBlas通过对三角形矩阵使用常规2d数组来浪费内存。 I suppose this makes interfacing with the rest of the API more convenient. 我想这使与API其余部分的接口更加方便。

1 个解决方案

I was surprised just now to find that CuBlas wastes memory by using regular 2d arrays for triangular matrices 刚才我很惊讶地发现CuBlas通过对三角形矩阵使用常规2d数组浪费了内存

That isn't strictly true. 严格说来并非如此。

If you look at the Level 2 BLAS routines, you will see that they operate on triangular or Hermitian matrices stored in a packed format. 如果您查看2级BLAS例程，您会发现它们在以打包格式存储的三角或Hermitian矩阵上运行。

The Level 3 BLAS routines don't, but there are two good reasons why they are stored in full dense format. 3级BLAS例程没有，但是有两个很好的理由将它们以完全密集的格式存储。

BLAS does it that way BLAS就是这样
Those routines were mostly added to BLAS as support for LAPACK solvers. 这些例程大部分已添加到BLAS中，以支持LAPACK求解器。 And those solvers typically store the results of factorizations in-situ in supplied full dense inputs, so it is logical to use that format in BLAS 而且这些求解器通常在提供的全密集输入中原位存储分解的结果，因此在BLAS中使用该格式是合乎逻辑的

I guess if you don't like the design choice you can always try writing to Jack Dongarra to complain. 我想如果您不喜欢这种设计选择，可以随时尝试写信给Jack Dongarra投诉。