|
|
@@ -186,10 +186,11 @@ Using lower precision for some operations can speed up the math involved (at the
|
|
|
This is rarely needed in the vertex processor function (where full precision is needed most of the time),
|
|
|
but is often useful in the fragment processor.
|
|
|
|
|
|
-Keep in mind that some architectures (mainly mobile) benefit a lot from this, but are also restricted
|
|
|
-(conversion between precisions has a cost). Please read the relevant documentation on the target architecture
|
|
|
-to find out more. In all honesty though, mobile drivers are buggy, so, to stay out of trouble, make simple
|
|
|
-shaders without specifying precision unless you *really* need to.
|
|
|
+Some architectures (mainly mobile) can benefit significantly from this, but
|
|
|
+there are downsides such as the additional overhead of conversion between
|
|
|
+precisions. Refer to the documentation of the target architecture for further
|
|
|
+information. In many cases, mobile drivers cause inconsistent or unexpected
|
|
|
+behavior and it is best to avoid specifying precision unless necessary.
|
|
|
|
|
|
Arrays
|
|
|
------
|
Nathan Lovato