I am quite interested to obtain access for full version of PVR Tune to optimize my game. We are developing a highly advanced renderer and it will be a handy tool.
Please consider my request.
We only have one version of PVRTune that can be given out to developers. We have another build that is for our DDK licencees only - it contains stats that are useful for hardware bring-up, but doesn’t really include any extra features that would assist application level development.
What information did you want to know about the device you’re profiling?
Thanks for the information. I am interested in performance counters which I am able to go through with the current tool. However, I hit a problem where your UtilitiesPVRScopeDeveloperWindows_x86_32PVRScopeDeveloper.lib does not link with my project which is setup using visual studio 2008. Is this possible that this lib only works for visual studio 2010 projects ?
The additional info I wanted to query is to see if there are counters telling me how much stencil culling, depth culling happened , and if its actually giving me any performance. Also I want to know how I can mark/group draw calls and see it on PVR tune while it captures.
Are you targeting a Windows platform with a PowerVR GPU? If so, is it an x86 or ARM based CPU?
If you’re using Scope on your development machine (non-PowerVR), you will only be able to retrieve the software counters and use the networking features (e.g. adding custom markers to the recording).
We don’t provide any stats for the number of fragments that have passed depth and stencil tests. We’re planning to extend PVRTrace’s functionality to give an offline analysis of this data in a future release.
You can send custom markers to PVRTune, but there are no mechanisms for marking draw calls. The reason for this is that the GPU architecture is very complex and there are no sensible metrics that can be calculated for the cost of a single (or several) draw calls. Unlike some other GPU architectures, our deferred render processes all vertex transformations in one go and colours a tile at a time instead of an object at a time. Additionally, the dynamically load balancing of the GPU gives adds another level of complexity that makes it difficult to determine anything other than system wide statistics.
Quick question: Is there a way to mark each render object and see it in PVRTune capture ?
There’s currently no way to do this. However, hovering your mouse cursor over the timing data will display the frame number. An FBO render will be kicked when there is a draw call issued that uses at least one of the FBOs attachments. With this information, it’s quite straight forward to determine which render target is which.
It may be worth building our PVRScope library into your application as it can send custom markers to PVRTune This will allow you to annotate the captured data from your application in real-time.