本文共 873 字，大约阅读时间需要 2 分钟。

CUDA devices use several memory spaces, which have different characteristics that reflect their distinct usages in CUDA applications. These memory spaces include global,

writing to its underlying global memory array in the same kernel launch should be avoided because the texture caches are read-only and are not invalidated when the associated global memory is modified.

转载地址：http://nmzqi.baihongyu.com/