菜单
Unified kernel benchmarking protocol producing JSON results with latency, TFLOPS, GBps, and comparison against PyTorch baselines. Covers warmup, timing, and cross-platform reporting. Use when benchmarking kernels, measuring performance, or comparing implementations.
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-benchmark |
| description | Unified kernel benchmarking protocol producing JSON results with latency, TFLOPS, GBps, and comparison against PyTorch baselines. Covers warmup, timing, and cross-platform reporting. Use when benchmarking kernels, measuring performance, or comparing implementations. |
Consistent benchmarking methodology across all kernels and platforms. Every benchmark produces a JSON result for automated tracking.
Run the kernel n_warmup times (default: 25) to:
torch.cuda.synchronize() after warmuptorch.cuda.synchronize()
start = time.perf_counter()
for _ in range(n_iter): # default: 100
kernel(inputs)
torch.cuda.synchronize()
elapsed = (time.perf_counter() - start) / n_iter
| Kernel Type | Primary Metric | Formula |
|---|---|---|
| Compute-bound (GEMM, attention) | TFLOPS | flops / elapsed / 1e12 |
| Memory-bound (norm, softmax, RoPE) | GBps | total_bytes / elapsed / 1e9 |
FLOPS formulas:
2 * M * N * K4 * B * H * N * N * D (2 GEMMs)Bytes formulas:
rows * cols * elem_size * 2 (read + write)rows * cols * elem_size * 3 (read x, read w, write out)Always compare against PyTorch:
vs_pytorch = triton_metric / pytorch_metric
{
"kernel": "gemm",
"gpu": "NVIDIA GeForce RTX 4090",
"vendor": "nvidia",
"correctness": true,
"latency_us": 101.9,
"tflops": 168.68,
"gbps": null,
"vs_pytorch": 1.014,
"dtype": "float16",
"shape": "2048x2048x2048",
"status": "pass"
}
When benchmarking:
torch.cuda.synchronize() before and after timingexperiments/ for trackingOptimize 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.