Identify correctness issues in parallel CUDA, OpenMP, and OpenCL code.
npx skills add https://github.com/SamyakJhaveri/loam --skill hpc-code-reviewer
Please help me install this Agent Skill. Skill: hpc-code-reviewer Source: https://github.com/SamyakJhaveri/loam/tree/main/seed/_research/skills/hpc-code-reviewer Command: npx skills add https://github.com/SamyakJhaveri/loam --skill hpc-code-reviewer
Provides a structured checklist to identify correctness issues in parallel CUDA, OpenMP, and OpenCL code, including data races, memory-model violations, and synchronization bugs, especially when reviewing translated code, auditing benchmark sources, or documenting failure modes.
Run the HPC Code Reviewer on your parallel kernels to systematically identify and root-cause correctness issues.
High-intent search queries and answers about installing and using this skill.
To review parallel CUDA code for data races, apply a structured reviewer checklist that systematically identifies correctness issues like memory model violations, synchronization bugs, numerical precision issues, and API-specific pitfalls to ensure root-cause analysis.
Common failure modes in OpenMP and OpenCL translations include data races, memory model violations, synchronization bugs, and numerical precision issues. A structured review checklist identifies these API-specific pitfalls to document failure modes and ensure reproducible results.
To audit benchmark sources for correctness in high performance computing, evaluate parallel kernels using a structured checklist covering data races, memory model violations, and synchronization bugs. This ensures root-cause analysis and reproducible results across CUDA, OpenMP, and OpenCL code.
Yes, you can use a structured reviewer checklist to find numerical precision issues in parallel kernels. The checklist systematically covers data races, memory model violations, synchronization bugs, and API-specific pitfalls across CUDA, OpenMP, and OpenCL code.
The best way to identify memory model violations in LLM-generated parallel code is to apply a structured reviewer checklist triggered when evaluating parallel kernels. This systematically detects data races, synchronization bugs, and numerical precision issues to ensure reproducible results.
igamenovoerConfigure CUDA build environments for Pixi projects with CMake and nvcc.
shafibabarOptimize Go application performance using pprof profiling and benchstat benchmark evidence.
dandacompanyDetect security and compliance issues in code repositories using pattern matching and rule sets.
harperaaDetect race conditions and business-logic vulnerabilities in AI-generated code.
