关于fixed-point

最新推荐文章于 2024-08-24 08:02:05 发布

weixin_30725315

最新推荐文章于 2024-08-24 08:02:05 发布

阅读量1.2k

点赞数

原文链接：http://www.cnblogs.com/yaukey/p/3416290.html

版权

今天又出现了shader的问题，编译到真机效果就没了，后来仔细还是因为浮点数精度的问题，后来仔细查找了些资料，才发现自己太粗心，没有看清楚 fixed-point 数据类型就乱用，这是个范围在 [-1,+1]或者[-1,+2]之间的浮点数，我却当做普通的浮点数用，导致很多数据溢出，所以才丢失原有的效果，得注意了！

不明白的就去看 http://http.developer.nvidia.com/CgTutorial/cg_tutorial_chapter03.html 的3.3.2小节吧

重点：Continuous quantities are not limited to integer values. When programming a CPU, programmers typically use floating-point data types to represent continuous values because floating-point types can represent fractional values. Continuous values processed by GPUs, particularly at the fragment level, have been limited to narrow ranges such as [0, 1] or [-1, +1], rather than supporting the expansive range provided by floating-point. For example, colors are often limited to the [0, 1] range, and normalized vectors are, by definition, confined to the [-1, +1] range. These range-limited data types are known as "fixed-point," rather than floating-point.

转载于:https://www.cnblogs.com/yaukey/p/3416290.html