Performance overhead optimization in GPU scoping

A method of graphics processing, comprising:

• capturing a plurality of performance counters at each stage of a graphics processing unit (GPU) pipeline;
• processing a first workload of a plurality of workloads at each of multiple clusters in the GPU pipeline;
• incrementing the plurality of performance counters during the processing of the first workload at each of the multiple clusters;
• determining, at each of the multiple clusters, whether the first workload is finished processing;
• scoping the GPU pipeline by reading, upon determining that the first workload is finished processing, a value of each of the multiple clusters for each of the plurality of performance counters at defined workload boundaries;
• transmitting an indication of the read value of each of the multiple clusters for all of the plurality of performance counters;
• processing a second workload of the plurality of workloads at each of the multiple clusters when, for the first workload, the reading of the plurality of performance counters is completed for each of the multiple clusters,
• wherein the scoping the GPU pipeline by reading the value of each of the multiple clusters for each of the plurality of performance counters provides accurate capturing of the performance counters without any mixing or modifying of the workloads while allowing multiple workloads to be present in the GPU pipeline at a time and capturing the performance counters at the defined workload boundaries.

An apparatus for graphics processing, comprising:

• means for capturing a plurality of performance counters at each stage of a graphics processing unit (GPU) pipeline;
• means for processing a first workload of a plurality of workloads at each of multiple clusters in the GPU pipeline;
• means for incrementing the plurality of performance counters during the processing of the first workload at each of the multiple clusters;
• means for determining, at each of the multiple clusters, whether the first workload is finished processing;
• means for scoping the GPU pipeline by reading, upon determining that the first workload is finished processing, a value of each of the multiple clusters for each of the plurality of performance counters at defined workload boundaries;
• means for transmitting an indication of the read value of each of the multiple clusters for all of the plurality of performance counters;
• means for processing a second workload of the plurality of workloads at each of the multiple clusters when, for the first workload, the reading of the plurality of performance counters is completed for each of the multiple clusters;
• and
• means for transmitting, upon finishing processing the first workload at each of the multiple clusters, an indication that the first workload is finished processing from each of the multiple clusters to a debug controller (DBGC),
• wherein information corresponding to the incremented plurality of performance counters is stored at a register backbone management (RBBM) component of the GPU pipeline, and
• wherein the means for scoping the GPU pipeline by reading the value of each of the multiple clusters for each of the plurality of performance counters provides accurate capturing of the performance counters without any mixing or modifying of the workloads while allowing multiple workloads to be present in the GPU pipeline at a time and capturing the performance counters at the defined workload boundaries.

A non-transitory computer-readable medium storing computer executable code for graphics processing, the code when executed by a processor causes the processor to:

• capture a plurality of performance counters at each stage of a graphics processing unit (GPU) pipeline;
• process a first workload of a plurality of workloads at each of multiple clusters in the GPU pipeline;
• increment the plurality of performance counters during the processing of the first workload at each of the multiple clusters;
• determine, at each of the multiple clusters, whether the first workload is finished processing;
• scope the GPU pipeline by reading, upon determining that the first workload is finished processing, a value of each of the multiple clusters for each of the plurality of performance counters at defined workload boundaries;
• transmit an indication of the read value of each of the multiple clusters for all of the plurality of performance counters;
• process a second workload of the plurality of workloads at each of the multiple clusters when, for the first workload, the reading of the plurality of performance counters is completed for each of the multiple clusters;
• and
• transmit, upon finishing processing the first workload at each of the multiple clusters, an indication that the first workload is finished processing from each of the multiple clusters to a debug controller (DBGC),
• wherein information corresponding to the incremented plurality of performance counters is stored at a register backbone management (RBBM) component of the GPU pipeline, and
• wherein the scoping the GPU pipeline by reading the value of each of the multiple clusters for each of the plurality of performance counters provides accurate capturing of the performance counters without any mixing or modifying of the workloads while allowing multiple workloads to be present in the GPU pipeline at a time and capturing the performance counters at the defined workload boundaries.