Retail vs. Debug vertex shader performance

Retail vs. Debug vertex shader performance

Post by Scott » Sat, 05 Apr 2003 06:02:58



I'm seeing some behavior that seems strange, and I can't put a
finger on what's causing it.

I have a model viewer application that displays skinned mesh
models (my own mesh classes, not d3dx meshes) using either the
fixed-function pipeline or a vertex shader (both using hardware
vertex processing on an GeForce 4 card)

When I run the app using the debug version of DirectX, the
vertex shader is faster than fixed-function rendering, which
is what I expected.  However, when I switch to the retail
version, fixed-function rendering is as fast or faster than the
vertex shader technique, which I did not expect at all.

I'm getting the maximum amount of debug info when running
the debug version, so I don't think there's some validation
failure that's causing the issue when I switch to retail, but while
I continue to try to understand why I'm seeing this, I wanted
to throw it out there to see if anyone had run into anything like
this.

 
 
 

Retail vs. Debug vertex shader performance

Post by Craig Peeper [MS » Sat, 05 Apr 2003 08:41:18


All this says is that the code run for debug vertex shaders is faster than
the debug code for fixed function.  This could be due to any number of
reasons.  i.e. maybe there is more validation in the fixed function code
when debug is enabled.  I don't know and the retail perf is the telltale
factor.  Never use the debug runtime for perf comparisons.

--
Craig Peeper
Microsoft, Direct3D

This posting is provided "AS IS" with no warranties, and confers no rights.

Quote:> I'm seeing some behavior that seems strange, and I can't put a
> finger on what's causing it.

> I have a model viewer application that displays skinned mesh
> models (my own mesh classes, not d3dx meshes) using either the
> fixed-function pipeline or a vertex shader (both using hardware
> vertex processing on an GeForce 4 card)

> When I run the app using the debug version of DirectX, the
> vertex shader is faster than fixed-function rendering, which
> is what I expected.  However, when I switch to the retail
> version, fixed-function rendering is as fast or faster than the
> vertex shader technique, which I did not expect at all.

> I'm getting the maximum amount of debug info when running
> the debug version, so I don't think there's some validation
> failure that's causing the issue when I switch to retail, but while
> I continue to try to understand why I'm seeing this, I wanted
> to throw it out there to see if anyone had run into anything like
> this.