Run any Skill in Manus with one click

kernel-profiling

Stars8

Forks1

UpdatedMay 15, 2026 at 23:39

Profile GPU kernels using NCU (NVIDIA) or rocprof (AMD) to collect performance metrics. Produces structured metrics.json with throughput, bandwidth, occupancy, and bottleneck classification. Use when profiling kernels, collecting NCU/rocprof data, or diagnosing performance issues.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

Jerry2423

Jerry2423/MoE-Kernel-Agent

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	kernel-profiling
description	Profile GPU kernels using NCU (NVIDIA) or rocprof (AMD) to collect performance metrics. Produces structured metrics.json with throughput, bandwidth, occupancy, and bottleneck classification. Use when profiling kernels, collecting NCU/rocprof data, or diagnosing performance issues.

Kernel Profiling Skill

Overview

Profiling is the first step in any kernel optimization loop. This skill covers collecting hardware performance counters and deriving actionable metrics on both NVIDIA and AMD GPUs.

NVIDIA: Nsight Compute (NCU)

Quick Command

ncu --set full -o profile_output python your_kernel.py

Key Metrics to Collect

Metric	What It Tells You
`sm__throughput.avg.pct_of_peak_sustained_elapsed`	SM utilization
`dram__throughput.avg.pct_of_peak_sustained_elapsed`	HBM bandwidth utilization
`l1tex__throughput.avg.pct_of_peak_sustained_elapsed`	L1/Tex cache pressure
`sm__warps_active.avg.pct_of_peak_sustained_active`	Occupancy
`smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.pct`	Global load efficiency

Roofline Derivation

arithmetic_intensity = flops / bytes_transferred
peak_flops = gpu_spec["fp16_tflops"] * 1e12
peak_bw = gpu_spec["hbm_bw_bytes"]
roofline_bound = min(peak_flops, arithmetic_intensity * peak_bw)
achieved_pct = actual_flops / roofline_bound

AMD: rocprof

Quick Command

rocprof --stats python your_kernel.py
# Or with specific counters:
rocprof -i counters.txt python your_kernel.py

Key Counters

pmc : SQ_INSTS_VALU SQ_INSTS_MFMA_MOPS_FP16 SQ_INSTS_MFMA_MOPS_BF16
pmc : SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_MFMA SQ_BUSY_CYCLES
pmc : TCC_HIT TCC_MISS TCC_REQ
pmc : TCC_EA_RDREQ TCC_EA_WRREQ

Derived Metrics

Metric	Formula	Target
GPU Utilization	`GRBM_GUI_ACTIVE / GRBM_COUNT`	>90%
MFMA Efficiency	`SQ_ACTIVE_INST_MFMA / SQ_BUSY_CYCLES`	>50% for matmul
L2 Hit Rate	`TCC_HIT / TCC_REQ`	>80% for tiled kernels
Memory BW Util	`(reads + writes) * line_size / time / peak_bw`	>70% for mem-bound

Output Format

Produce a metrics.json:

{
  "kernel": "matmul",
  "gpu": "RTX 4090",
  "latency_us": 101.9,
  "tflops": 168.7,
  "gbps": null,
  "occupancy_pct": 85.2,
  "sm_throughput_pct": 72.1,
  "dram_throughput_pct": 45.3,
  "bottleneck": "compute",
  "roofline_pct": 68.5
}

Agent Instructions

Detect GPU vendor (scripts/detect_gpu.py)
Run appropriate profiler (NCU or rocprof)
Parse output into structured metrics
Classify bottleneck (memory / compute / latency)
Feed into bottleneck-diagnosis skill for optimization recommendations

More from this repository

same repository

cross-entropy-kernel

Jerry2423/MoE-Kernel-Agent

Optimize fused cross-entropy loss kernels in Triton for NVIDIA and AMD GPUs. Covers fused log-softmax + NLL, online log-sum-exp, and large vocabulary handling. Use when writing or optimizing cross-entropy, label-smoothed CE, or similar loss kernels.

2026-05-158

flash-attention-kernel

Jerry2423/MoE-Kernel-Agent

Optimize FlashAttention-style fused attention kernels in Triton for NVIDIA and AMD GPUs. Covers online softmax, tiled QK/AV GEMM, causal masking, and memory-efficient attention. Use when writing or optimizing self-attention, cross-attention, or any QKV attention kernel.

2026-05-158

gemm-kernel-optimization

Jerry2423/MoE-Kernel-Agent

Optimize dense matrix multiplication (GEMM) kernels in Triton for NVIDIA and AMD GPUs. Covers tiled blocking, L2 cache grouping, tensor core utilization, and dual-platform autotune. Use when writing or optimizing matmul, linear layers, or any GEMM-based kernel.

2026-05-158

rmsnorm-kernel-optimization

Jerry2423/MoE-Kernel-Agent

Optimize RMS Normalization kernels in Triton for NVIDIA and AMD GPUs. Covers row-parallel reduction, vectorized loads, warp shuffle patterns, and rsqrt usage. Use when writing or optimizing RMSNorm, LayerNorm, or similar normalization kernels.

2026-05-158

rotary-embedding-kernel

Jerry2423/MoE-Kernel-Agent

Optimize Rotary Position Embedding (RoPE) kernels in Triton for NVIDIA and AMD GPUs. Covers interleaved/sequential layouts, sincos precomputation, and multi-row processing. Use when writing or optimizing RoPE, position embeddings, or rotary transformations.

2026-05-158

softmax-kernel-optimization

Jerry2423/MoE-Kernel-Agent

Optimize fused softmax kernels in Triton for NVIDIA and AMD GPUs. Covers online max/sum reduction, multi-row processing, numerical stability, and warp-level patterns. Use when writing or optimizing softmax, log-softmax, or similar row-wise normalization kernels.

2026-05-158

name	kernel-profiling
description	Profile GPU kernels using NCU (NVIDIA) or rocprof (AMD) to collect performance metrics. Produces structured metrics.json with throughput, bandwidth, occupancy, and bottleneck classification. Use when profiling kernels, collecting NCU/rocprof data, or diagnosing performance issues.

Kernel Profiling Skill

Overview

Profiling is the first step in any kernel optimization loop. This skill covers collecting hardware performance counters and deriving actionable metrics on both NVIDIA and AMD GPUs.

NVIDIA: Nsight Compute (NCU)

Quick Command

ncu --set full -o profile_output python your_kernel.py

Key Metrics to Collect

Metric	What It Tells You
`sm__throughput.avg.pct_of_peak_sustained_elapsed`	SM utilization
`dram__throughput.avg.pct_of_peak_sustained_elapsed`	HBM bandwidth utilization
`l1tex__throughput.avg.pct_of_peak_sustained_elapsed`	L1/Tex cache pressure
`sm__warps_active.avg.pct_of_peak_sustained_active`	Occupancy
`smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.pct`	Global load efficiency

Roofline Derivation

arithmetic_intensity = flops / bytes_transferred
peak_flops = gpu_spec["fp16_tflops"] * 1e12
peak_bw = gpu_spec["hbm_bw_bytes"]
roofline_bound = min(peak_flops, arithmetic_intensity * peak_bw)
achieved_pct = actual_flops / roofline_bound

AMD: rocprof

Quick Command

rocprof --stats python your_kernel.py
# Or with specific counters:
rocprof -i counters.txt python your_kernel.py

Key Counters

pmc : SQ_INSTS_VALU SQ_INSTS_MFMA_MOPS_FP16 SQ_INSTS_MFMA_MOPS_BF16
pmc : SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_MFMA SQ_BUSY_CYCLES
pmc : TCC_HIT TCC_MISS TCC_REQ
pmc : TCC_EA_RDREQ TCC_EA_WRREQ

Derived Metrics

Metric	Formula	Target
GPU Utilization	`GRBM_GUI_ACTIVE / GRBM_COUNT`	>90%
MFMA Efficiency	`SQ_ACTIVE_INST_MFMA / SQ_BUSY_CYCLES`	>50% for matmul
L2 Hit Rate	`TCC_HIT / TCC_REQ`	>80% for tiled kernels
Memory BW Util	`(reads + writes) * line_size / time / peak_bw`	>70% for mem-bound

Output Format

Produce a metrics.json:

{
  "kernel": "matmul",
  "gpu": "RTX 4090",
  "latency_us": 101.9,
  "tflops": 168.7,
  "gbps": null,
  "occupancy_pct": 85.2,
  "sm_throughput_pct": 72.1,
  "dram_throughput_pct": 45.3,
  "bottleneck": "compute",
  "roofline_pct": 68.5
}

Agent Instructions

Detect GPU vendor (scripts/detect_gpu.py)
Run appropriate profiler (NCU or rocprof)
Parse output into structured metrics
Classify bottleneck (memory / compute / latency)
Feed into bottleneck-diagnosis skill for optimization recommendations