Diagnose incorrect FlyDSL GPU kernel outputs, NaNs, infs, and crashes.
npx skills add https://github.com/ROCm/FlyDSL --skill debug-flydsl-kernel
Please help me install this Agent Skill. Skill: debug-flydsl-kernel Source: https://github.com/ROCm/FlyDSL/tree/main/.claude/skills/debug-flydsl-kernel Command: npx skills add https://github.com/ROCm/FlyDSL --skill debug-flydsl-kernel
Debug FlyDSL GPU kernels that produce NaN, inf, or incorrect results, or crash, by providing structured diagnostics and reproducible workflows.
Run the FlyDSL kernel debugger on the target kernel to reproduce the issue and collect diagnostics.
High-intent search queries and answers about installing and using this skill.
To debug FlyDSL GPU kernels producing NaN or inf outputs, run the debugger to collect structured diagnostics and reproduce the issue. It identifies incorrect results through cache invalidation checks and host-side tensor inspection.
FlyDSL GPU kernel crashes during scf.for loops often stem from incorrect state packing. The debugger provides targeted diagnostics for scf.for state packing to isolate the defect and ensure robust kernel behavior.
In FlyDSL tracing, range and range_constexpr differ in how runtime conditionals are handled. The debugger traces these pitfalls to identify incorrect kernel outputs and validate partition handling.
Yes, you can diagnose LDS bank conflicts and verify MFMA operand layout in FlyDSL. The debugger pinpoints addressing and layout issues to resolve incorrect GPU kernel computations.
Yes, the FlyDSL debugger supports multi-partition GPU kernel isolation. It provides diagnostic workflows to isolate defects in both single-partition and multi-partition kernels and validate their outputs.
ROCmGenerate and validate FlyDSL GPU kernels from kernel specifications using tile programming.
fsx950223Build and install FlyDSL with LLVM/MLIR on remote hosts or Docker containers.
ROCmOptimizes GEMM kernels on AMD CDNA GPUs using FlyDSL tiling, memory layout, MFMA scheduling.
scottgl9Record CUDA kernel calls and tensor states to diagnose GPU errors.
