Capture and analyze GPU kernel ATT traces using rocprofv3.
npx skills add https://github.com/ROCm/FlyDSL --skill capture-kernel-trace
Please help me install this Agent Skill. Skill: capture-kernel-trace Source: https://github.com/ROCm/FlyDSL/tree/main/.claude/skills/capture-kernel-trace Command: npx skills add https://github.com/ROCm/FlyDSL --skill capture-kernel-trace
Capture GPU kernel ATT traces from a GPU environment (local or remote Docker), configure input.yaml for the target kernel, execute rocprofv3 -i input.yaml with source mapping enabled, and download the latest ui_output_agent_* data for analysis.
Run the capture-kernel-trace skill with your test script and an optional kernel pattern to start collecting ATT traces.
High-intent search queries and answers about installing and using this skill.
To capture GPU kernel ATT traces remotely, you need SSH or Docker access to the target environment, rocprofv3 installed, and an input.yaml file configured with your kernel patterns to execute the tracing.
Kernel discovery identifies available kernels on the target GPU environment to determine tracing targets, allowing you to generate a tailored input.yaml with specific kernel_include_regex and tracing parameters for rocprofv3.
After executing rocprofv3 -i input.yaml, retrieve the latest ui_output_agent_* directory and associated trace artifacts from the GPU environment, organizing them locally for post-analysis and inspection.
Yes, FLYDSL_DEBUG_ENABLE_DEBUG_INFO is required to enable DWARF source mappings in your ATT traces, ensuring that collected Advanced Thread Trace data includes accurate source-level mapping for analysis.
GPU kernel ATT trace collection works across both local and remote Docker environments, requiring rocprofv3 and proper input.yaml configuration to execute tracing and retrieve ui_output_agent_* data for analysis.
ATT trace collection with rocprofv3 requires proper remote access via SSH or Docker, FLYDSL_DEBUG_ENABLE_DEBUG_INFO for source mappings, and depends on accurate kernel discovery and input.yaml configuration for successful tracing.
fsx950223Build a custom Docker image with ROCm tools on a remote host via SSH.
tharunramagiriCapture and analyze Unreal Engine trace data through automated profiling workflows.
fsx950223Profile GPU kernels with rocprofv3 and generate optimization plans.
alibabaAnalyzes GPU kernel performance using Roofline models and profiling data.
