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

references/optimize/optimize-ascendc-regbase-vf.md

View on GitHub
# Ascend C · RegBase / SIMD VF 优化(`__simd_vf__` + `AscendC::Reg::*`)

> 适用:Ascend 950(A5, dav-3510)的寄存器级向量计算范式。MemBase 层机制(分核、合并搬运、删冗余同步)仍然有效且应先做;本文档覆盖 VF 语义层的迁移判断与优化技术。
> 来源:[昇腾 950 Simd-VF 编程系列](https://cann.csdn.net/6a5f15b6662f9a54cb92a013.html)([高性能技术篇](https://hwcomputing.csdn.net/6a6002df662f9a54cb92fc5b.html),CANN 开发者社区)+ cann-samples VF 样例实测。

## 1. 迁移判断:什么时候值得 MemBase→RegBase

| 维度 | MemBase(`LocalTensor`) | RegBase(`RegTensor`) |
|---|---|---|
| 数据载体 | UB 内存 | VF 寄存器(VL=256B/寄存器) |
| 处理粒度 | 任意长(硬件内部 tiling) | 一次一个 VL,软件显式分块循环 |
| 搬运 | DataCopy(MTE2/MTE3) | LoadAlign/StoreAlign(UB↔Reg) |
| 调用层级 | `__aicore__` 直接调 | 必须 `__simd_vf__` + `asc_vf_call` |
| Mask | count 参数自动 | 显式 `MaskReg`(32B=256bit) |

**值得迁移**(按收益排序):

1. **多步融合计算链**(Sigmoid/LayerNorm/Softmax/GELU):中间结果留在寄存器,UB 流量从 2×步数×VL 降到 2×VL(4 步算子约 -75% UB 往返)——这是 RegBase 的主收益来源
2. 计算指令周期长的算子(Exp/Ln/Sqrt/Div),UB 往返占比高
3. 非连续访问(Gather/Scatter/Squeeze/Block-Strided)MemBase 难高效表达
4. MemBase 优化已尽(vec_ratio 高、无冗余搬运可删)后的最后一档

**不值得迁移**:简单单步算子(纯 Add/Sub,MemBase 已足够);瓶颈在 MTE2 带宽(mte2_ratio>90%)的算子——RegBase 不解决带宽问题。实测 Add 级单步算子 MemBase→RegBase 收益不明显(计算链太短)。

## 2. 调用层级(违反即编译报错)

```
__global__ __aicore__ kernel → __aicore__ 函数
  → asc_vf_call<VFFunc>(args...)   // 唯一进入 VF 上下文的方式
    → __simd_vf__ 函数              // Reg API 的唯一执行环境
      → __simd_callee__(所有 AscendC::Reg::*)
```

```cpp
template <typename T>
__simd_vf__ inline void AddVF(__ubuf__ T* dst, __ubuf__ T* src0, __ubuf__ T* src1,
                              uint16_t repeatTimes, uint32_t oneRepeatSize) {
    AscendC::Reg::RegTensor<T> r0, r1, r2;
    AscendC::Reg::MaskReg mask = AscendC::Reg::CreateMask<T>();
    for (uint16_t i = 0; i < repeatTimes; ++i) {
        AscendC::Reg::LoadAlign(r0, src0 + i * oneRepeatSize);
        AscendC::Reg::LoadAlign(r1, src1 + i * oneRepeatSize);
        AscendC::Reg::Add(r2, r0, r1, mask);
        AscendC::Reg::StoreAlign(dst + i * oneRepeatSize, r2, mask);
    }
}
// __aicore__ 侧:asc_vf_call<AddVF<T>>(dst, s0, s1, repeats, oneRepeatSize);
```

## 3. 性能技术清单(单变量迭代候选)

| # | 技术 | 收益原理 |
|---|---|---|
| 1 | 寄存器复用减少 UB 往返 | 多步算子中间结果留 RegTensor,只在循环头尾访问 UB(Sigmoid 例 UB 流量 -75%) |
| 2 | VALU 三发射友好编排 | 相邻指令用不同 dstReg 制造独立性;Div/Sqrt/Exp(10+ 拍)后跟独立短指令填充流水;多 VL 块双路软件流水 |
| 3 | 标量二元 `*s` 后缀 + Duplicate | `Muls(dst,src,scalar,mask)` 省一次 LoadAlign+UB 存放+发射槽;常量直接 `Duplicate` 不落地 UB |
| 4 | 融合指令 | `Axpy`/`MulAddDst`/`MulsCast`/`AbsSub`/`ExpSub`:1 拍顶 2 拍 + 省 tmp UB |
| 5 | Post-Update / AddrReg 地址外提 | `LoadAlign<T, POST_MODE_UPDATE>(reg, ptr, stride)` 硬件自增地址,省循环内地址计算指令 |
| 6 | Hardware Loop 友好 | 最内层循环 trip count 编译期可知、循环体无分支、循环变量 uint16_t |
| 7 | L2Cache Hint | 流式数据 `SetL2CacheHint(CACHE_MODE_DISABLE)` 避免污染 L2 |
| 8 | 非对齐尾段 | `UpdateMask(remainder)`+`MaskMergeMode::MERGING`(防越界写 0 污染),或 `StoreUnAlign`+`StoreUnAlignPost` |
| 9 | 64 位整型 | `RegTensor<T, RegTraitNumTwo>` 拼 2×VL 逻辑寄存器(int64 必须) |
| 10 | 同地址 RAW | VF 内连续写后读同一 UB 地址:`LocalMemBar<VEC_STORE, VEC_LOAD>` |
| 11 | constexpr 最大化 | MaskPattern/LoadDist/StoreDist/MergeMode/RoundMode、oneRepeatSize、trip count 全部编译期决定 |

## 4. 瓶颈诊断表(VF 语境)

| 现象 | 含义 | 方向 |
|---|---|---|
| aiv_mte2_ratio > 90% | 带宽 bound | 增大 dataCopyLen、多核切分、L2Cache Hint(RegBase 无效) |
| aiv_vec_ratio > 50% 且 mte2 低 | 计算 bound | Bank Conflict、融合指令、寄存器复用 |
| aiv_scalar_ratio > 20% | 标量开销 | 地址计算外提(post-update/AddrReg)、减少循环内条件 |
| RegBase 收益不明显 | 链太短 | 回到 §1 重新评估是否值得;考虑把更多步融进同一 VF 函数 |
| vec_time 异常高 | UB Bank Conflict | dataCopyLen 错开 256B、调整 UB 布局 |

## 5. 常见编译/运行错误

| 错误 | 原因 | 解决 |
|---|---|---|
| Reg API called outside `__simd_vf__` | `__aicore__` 里直接调 Reg API | `asc_vf_call<VFFunc>(args...)` 包裹 |
| `__simd_vf__` called directly | 直接调用 VF 函数 | 必须经 `asc_vf_call` |
| MaskReg type mismatch | Mask 与 RegTensor 的 T 不一致 | `CreateMask<T>` 显式指定 |
| REG_NUM mismatch | int64 用了 RegTraitNumOne | 改 RegTraitNumTwo |
| LoadAlign address not aligned | 首地址非 32B 对齐 | LoadUnAlign 或调整 UB 偏移 |
| StoreAlign dst overflow | dst 长度不足 VL | MaskReg 或 StoreUnAlign |

## 6. 与其他文档的关系

- UB 容量/搬运粒度、DataCopy 合并:[optimize-data-copy.md](optimize-data-copy.md)
- Double Buffer / 流水重叠(MemBase 层先行):[optimize-pipeline.md](optimize-pipeline.md)
- 标量削减通用原则:[optimize-api-usage.md](optimize-api-usage.md)
- 案例路由:cann-samples `vector_function_add`(最小 VF 样例)、`simd_vf_story`(Broadcast/Elemwise/Reduce 三族 VF 调优,含 cannsim VF cycles 口径)
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