Abstract: Examples described herein generally relate to reporting faults in a graphics processing unit (GPU). A list of rendering instructions for causing the GPU to render graphics can be captured, and occurrence of a fault in executing the list of rendering instructions can be detected. A portion of the list of rendering instructions can be executed, in isolation from a second portion of the list of rendering instructions, based on detecting occurrence of the fault to identify the fault in the portion of the list of rendering instructions or the second portion of the list of rendering instructions. The portion of the list of rendering instructions or the second portion of the list of rendering instructions can be indicated as causing the fault.

Abstract: Examples described herein generally relate to reporting faults in a graphics processing unit (GPU). A list of rendering instructions for causing the GPU to render graphics can be captured, and occurrence of a fault in executing the list of rendering instructions can be detected. A portion of the list of rendering instructions can be executed, in isolation from a second portion of the list of rendering instructions, based on detecting occurrence of the fault to identify the fault in the portion of the list of rendering instructions or the second portion of the list of rendering instructions. The portion of the list of rendering instructions or the second portion of the list of rendering instructions can be indicated as causing the fault.

Abstract: Embodiments of the invention rely on simple time measurements to extract rich performance analysis from a graphics program. The invention times the program as it repeatedly executes on a selected platform, while a series of variations are applied to the program with each execution. As a result, without requiring any specific profiling tools for the platform, a developer can analyze and understand the performance characteristics of the program based upon these elapsed time measurements. The program can then be optimized to incorporate variations that improve execution time.

Abstract: Embodiments of the invention rely on simple time measurements to extract rich performance analysis from a graphics program. The invention times the program as it repeatedly executes on a selected platform, while a series of variations are applied to the program with each execution. As a result, without requiring any specific profiling tools for the platform, a developer can analyze and understand the performance characteristics of the program based upon these elapsed time measurements. The program can then be optimized to incorporate variations that improve execution time.