NVIDIA Corporation
Monitoring execution in a graphics processing unit

Abstract:

Marker commands are added to a stream of commands that are executed by a graphics processing unit (GPU) in a computing system. While the GPU executes the commands, information is written to a memory location each time a marker is reached in the pipeline. The memory location is accessible to the central processing unit (CPU), and the information identifies a command executed by the GPU. If the CPU receives an indication that the GPU is in an invalid state, then the CPU responds by accessing the memory location to identify the command executed in the pipeline of the GPU. Consequently, a command that was executing when the GPU entered the invalid state can be identified. This information can be used to pinpoint the cause of the invalid state.