菜单
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.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| 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. |
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.
ncu --set full -o profile_output python your_kernel.py
| 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 |
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
rocprof --stats python your_kernel.py
# Or with specific counters:
rocprof -i counters.txt python your_kernel.py
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
| 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 |
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
}
scripts/detect_gpu.py)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.
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.
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.
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.
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.
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.