cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

kd-11
Adept II

GPU PerfStudio 3.4.19 crashes when loading large api trace

GPUPerfStudio cannot analyze a frame capture when the log file is too large. The capture goes well, but when opening the analysis tools, they all fail in one way or another. API trace gives an error about failing to update and the frame debugger crashes with either "Length cannot be less than zero" or something about wrong string format when casting to a number.

This happens when a lot of draw calls are issued. Unfortunately, this is also when the frame debugger is needed the most for me. The generated log file is about 8.5MB for a frame with ~750 draw calls.

Log attached.

EDIT: I've also added the dot net crash message that is shown if you try to open the frame debugger.

0 Likes
2 Replies
john_larkin
Staff
Staff

Re: GPU PerfStudio 3.4.19 crashes when loading large api trace

Hi kd-11,

I've looked at your attached LOG.txt file, which somehow doesn't seem to include all of the data that I'd expect to see.

Are you trying to load the attached LOG.txt back into the GPUPerfStudio client? If so, try exporting/loading traces as .atr files, which should load correctly.

Might be helpful if we were able to debug using the same application that you're testing with, as I've been unable to replicate your attached LOG file locally.

If you're interested in sharing your sample with us to take a look, I've sent a PM with upload details for a private drop.

Thanks,

John

0 Likes
darkradeon
Adept III

Re: GPU PerfStudio 3.4.19 crashes when loading large api trace

Hi,

Looks like that GPU PerfStudio Version 3.4.19 is unable to compile shaders for Carrizo and Stoney GPUs too. Compilation for all other devices works fine.

Outputs:

Error: Couldn't find device named: Carrizo. Run '-s HLSL -l --verbose' to view available devices.

Error: Couldn't find device named: Stoney. Run '-s HLSL -l --verbose' to view available devices.

0 Likes