Skip to content

Ascend/msagent / skills/operator/ops-performance-tuning

ops-performance-tuning

Verified

Compile, profile, diagnose, optimize, and compare Ascend NPU operators across CANN versions and A2/A3/A5 for Ascend C, CATLASS, Triton-Ascend, TileLang-Ascend, PyPTO, and SHMEM/MC2. Use when an operator already has a runnable implementation and the user asks for msOpProf collection, bottleneck analysis, source-level tuning, or a reproducible before/after performance report. Do not use as the primary workflow for operator creation, migration, or unresolved correctness failures.

FileKindSize
SKILL.mdskill md15.4 KB
evals/evals.jsonconfig3.6 KB
references/case-routing.mdmarkdown30.0 KB
references/cases/ascendc/pr_ascendc_allgathermm_scalar_align.mdmarkdown1.7 KB
references/cases/ascendc/pr_ascendc_cann_samples_perf_story.mdmarkdown3.9 KB
references/cases/ascendc/pr_ascendc_cgdr_vf_inverse.mdmarkdown1.6 KB
references/cases/ascendc/pr_ascendc_conv3ddx_tque_tbuf.mdmarkdown1.6 KB
references/cases/ascendc/pr_ascendc_layernorm_areg.mdmarkdown1.6 KB
references/cases/ascendc/pr_ascendc_logit_a5_microapi.mdmarkdown1.9 KB
references/cases/ascendc/pr_ascendc_matmul_case_18x.mdmarkdown1.5 KB
references/cases/ascendc/pr_ascendc_matmul_full_load_wiki.mdmarkdown1.6 KB
references/cases/ascendc/pr_ascendc_matmul_vcv_wiki.mdmarkdown1.8 KB
references/cases/ascendc/pr_ascendc_mc2_fusion.mdmarkdown1.4 KB
references/cases/ascendc/pr_ascendc_mhc_24x.mdmarkdown1.2 KB
references/cases/ascendc/pr_ascendc_msopprof_add_block_imbalance.mdmarkdown1.7 KB
references/cases/ascendc/pr_ascendc_quantize_a2_micro_opt.mdmarkdown2.0 KB
references/cases/ascendc/pr_ascendc_rotate_quant_blaze.mdmarkdown2.8 KB
references/cases/ascendc/pr_ascendc_tiling_l2_official.mdmarkdown1.6 KB
references/cases/catlass/pr_catlass_baseline.mdmarkdown1.3 KB
references/cases/catlass/pr_catlass_fixpipe_dualdst.mdmarkdown1.6 KB
references/cases/catlass/pr_catlass_grouped_matmul_tiling.mdmarkdown1.5 KB
references/cases/catlass/pr_catlass_mstuner_3p8x.mdmarkdown1.9 KB
references/cases/catlass/pr_catlass_swizzle_balance.mdmarkdown1.6 KB
references/cases/pypto/pr_pypto_gqa_decode_vector_merge.mdmarkdown2.4 KB
references/cases/pypto/pr_pypto_l2_hit_layout_s13.mdmarkdown1.9 KB
references/cases/pypto/pr_pypto_per_matmul_tile_shapes.mdmarkdown2.4 KB
references/cases/pypto/pr_pypto_small_shape_matmul.mdmarkdown2.3 KB
references/cases/pypto/pr_pypto_weight_none_cacheable.mdmarkdown2.6 KB
references/cases/shmem/pr_shmem_alltoallv_mr152.mdmarkdown2.4 KB
references/cases/shmem/pr_shmem_reduce_scatter_serial.mdmarkdown2.6 KB
references/cases/tilelang/pr_tilelang_benchmark_vs_ascendc.mdmarkdown1.7 KB
references/cases/tilelang/pr_tilelang_compiler_prs.mdmarkdown2.3 KB
references/cases/tilelang/pr_tilelang_fa_pr698.mdmarkdown1.5 KB
references/cases/tilelang/pr_tilelang_gqa_bwd_pr1494.mdmarkdown1.3 KB
references/cases/tilelang/pr_tilelang_persistent_mqa_decode.mdmarkdown2.2 KB
references/cases/tilelang/pr_tilelang_sfa_pr665.mdmarkdown914 B
references/cases/triton/pr_triton_atomic_simd_24x.mdmarkdown1.8 KB
references/cases/triton/pr_triton_autotune_background.mdmarkdown1.0 KB
references/cases/triton/pr_triton_constexpr_recompile.mdmarkdown1.1 KB
references/cases/triton/pr_triton_lgamma_simt.mdmarkdown1.6 KB
references/compile/compile-ascendc.mdmarkdown9.0 KB
references/compile/compile-catlass.mdmarkdown2.4 KB
references/compile/compile-pypto.mdmarkdown2.0 KB
references/compile/compile-shmem.mdmarkdown2.6 KB
references/compile/compile-tilelang.mdmarkdown2.7 KB
references/compile/compile-triton.mdmarkdown4.4 KB
references/diagnose/diagnose-l2-cache.mdmarkdown2.2 KB
references/diagnose/diagnose-occupancy.mdmarkdown4.0 KB
references/diagnose/diagnose-pipeline.mdmarkdown4.1 KB
references/diagnose/diagnose-roofline.mdmarkdown3.1 KB
references/optimize/optimize-api-usage.mdmarkdown1.7 KB
references/optimize/optimize-ascendc-regbase-vf.mdmarkdown5.9 KB
references/optimize/optimize-ascendc-tiling.mdmarkdown935 B
references/optimize/optimize-catlass.mdmarkdown3.0 KB
references/optimize/optimize-data-copy.mdmarkdown870 B
references/optimize/optimize-index.mdmarkdown3.6 KB
references/optimize/optimize-memory-hierarchy.mdmarkdown599 B
references/optimize/optimize-pipeline.mdmarkdown1.0 KB
references/optimize/optimize-pypto.mdmarkdown1.5 KB
references/optimize/optimize-shmem.mdmarkdown5.1 KB
references/optimize/optimize-tilelang.mdmarkdown1.7 KB
references/optimize/optimize-triton.mdmarkdown3.8 KB
references/profile/benchmark-harness.mdmarkdown2.2 KB
references/profile/profile-msopprof.mdmarkdown7.4 KB
references/troubleshooting.mdmarkdown5.8 KB
scripts/bound_analyzer.pyscript9.0 KB
scripts/delta_report.pyscript4.8 KB
scripts/env_check.shscript5.0 KB
scripts/result_saver.pyscript5.0 KB

scripts/delta_report.py

View on GitHub
#!/usr/bin/env python3
"""性能对比报告 — 从 baseline+after json 生成 Δ% 对比表。

用法:
  python3 scripts/delta_report.py \
    --baseline results/baseline_board_xxx.json \
    --after results/after_board_xxx.json \
    --output delta_report.md
"""

from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path


# 硬口径:这些字段不一致则禁止比较(跨 shape/dtype/SoC/设备/计时法无意义)
HARD_KEYS = (
    "op", "soc", "mode", "timing_method", "baseline_kind", "cann_version", "device_id",
    "shape", "dtype", "format", "warmup", "repeat",
)
# 机制变量:分核/tiling 类优化本身就会改变它们,差异降级为 WARN 并写入报告
MECHANISM_KEYS = ("tiling_key", "block_dim")


def main() -> int:
    parser = argparse.ArgumentParser(description="生成优化前后 Δ% 对比报告")
    parser.add_argument("--baseline", required=True, help="基线 result json")
    parser.add_argument("--after", required=True, help="优化后 result json")
    parser.add_argument("--output", required=True, help="输出 markdown 报告路径")
    parser.add_argument("--strict", action="store_true",
                        help="恢复旧行为:tiling_key/block_dim 差异也视为口径不一致")
    args = parser.parse_args()

    baseline = json.loads(Path(args.baseline).read_text(encoding="utf-8"))
    after = json.loads(Path(args.after).read_text(encoding="utf-8"))

    hard_keys = tuple(HARD_KEYS) + (tuple(MECHANISM_KEYS) if args.strict else ())
    differences = []
    for key in hard_keys:
        before_value = baseline.get(key)
        after_value = after.get(key)
        if before_value is not None and after_value is not None and before_value != after_value:
            differences.append(f"{key}: {before_value!r} != {after_value!r}")
    if differences:
        print("ERROR: 基线与优化后结果口径不一致:", file=sys.stderr)
        for item in differences:
            print(f"  - {item}", file=sys.stderr)
        return 2

    mechanism_notes = []
    if not args.strict:
        for key in MECHANISM_KEYS:
            before_value = baseline.get(key)
            after_value = after.get(key)
            if before_value is not None and after_value is not None and before_value != after_value:
                mechanism_notes.append(f"{key}: {before_value!r} -> {after_value!r}")
        for item in mechanism_notes:
            print(f"WARN: 机制变量变化(分核/tiling 类优化属预期): {item}", file=sys.stderr)

    if baseline.get("precision") != "pass" or after.get("precision") != "pass":
        print("ERROR: 只有精度均为 pass 的结果才能生成性能结论", file=sys.stderr)
        return 2

    b_us = baseline.get("kernel_avg_us", 0)
    a_us = after.get("kernel_avg_us", 0)
    speedup = b_us / a_us if a_us > 0 else 0
    delta_pct = ((a_us - b_us) / b_us * 100) if b_us > 0 else 0

    op = baseline.get("op", after.get("op", "unknown"))
    soc = baseline.get("soc_full", baseline.get("soc", "?"))

    lines = [
        f"# 性能对比报告:{op}",
        "",
        f"| 项目 | 基线 (before) | 优化后 (after) | Δ% | 加速比 |",
        f"|---|---|---|---|---|",
        f"| kernel 耗时 | {b_us:.2f} µs | {a_us:.2f} µs | {delta_pct:+.1f}% | {speedup:.2f}x |",
        f"| 精度 | {baseline.get('precision','?')} | {after.get('precision','?')} | — | — |",
        f"| 芯片 | {baseline.get('soc','?')} | {after.get('soc','?')} | — | — |",
        f"| 模式 | {baseline.get('mode','?')} | {after.get('mode','?')} | — | — |",
        f"| shape | {baseline.get('shape','?')} | {after.get('shape','?')} | — | — |",
        f"| dtype | {baseline.get('dtype','?')} | {after.get('dtype','?')} | — | — |",
        "",
        f"## 结论",
        "",
    ]
    if mechanism_notes:
        lines.append("机制变量变化(分核/tiling 类优化的预期变量,非口径问题):")
        for item in mechanism_notes:
            lines.append(f"- {item}")
        lines.append("")
    if speedup > 1.05:
        lines.append(f"性能提升 {speedup:.2f}x({delta_pct:+.1f}%),可进入稳定性复测。")
    elif speedup >= 0.95:
        lines.append(
            f"性能变化 {speedup:.2f}x({delta_pct:+.1f}%)落在 ±5% 噪声带内:"
            "按 SKILL Step 7 规则应视为无稳定收益,增加采样轮次确认或回滚,"
            "不要把噪声内变化当作优化收益。")
    else:
        lines.append(f"性能劣化 {speedup:.2f}x({delta_pct:+.1f}%),应回滚该轮修改。")

    Path(args.output).write_text("\n".join(lines) + "\n", encoding="utf-8")
    print(f"报告已生成: {args.output}")
    print(f"加速比: {speedup:.2f}x (Δ={delta_pct:+.1f}%)")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())
planning-and-task-breakdownaddyosmani/agent-skillsBreaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
screenpipethedotmack/claude-memSearch screen and audio history recorded by Screenpipe and optionally save selected context into claude-mem. Use when the user asks to recall desktop activity, a meeting, or something seen in another app through Screenpipe.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
contact-widgetnexu-io/open-designSelf-contained floating chat widget with welcome screen, social links, meeting button, and message input. Single HTML file, zero dependencies.Stars: 108kUpdated Claude CodeCodexVerifiedInstall
meeting-notesnexu-io/open-designMeeting notes page — title bar with attendees, agenda checklist, decisions block, action items table with owners + dates, and a "next meeting" footer. Use when the brief mentions "meeting notes", "minutes", "1:1 notes", "all-hands recap", or "会议纪要".Stars: 108kUpdated Claude CodeCodexVerifiedInstall