|
|
6f44201295
|
v1.0: 修正源码仓地址为gitcode.com/cann/ops-nn
|
2026-08-26 11:30:25 +00:00 |
|
|
|
71590ec422
|
v1.0: 同步HTML
|
2026-08-26 11:05:56 +00:00 |
|
|
|
6b03624de3
|
v1.0: MergeBatch vs IterBatch 专项对比分析
|
2026-08-26 11:05:56 +00:00 |
|
|
|
16b37ebc9e
|
v0.96: 同步HTML
|
2026-08-26 09:28:22 +00:00 |
|
|
|
667241719f
|
v0.96: MergeBatch实现方案重写——b0→kL0→kL1反推→bL1最大化
|
2026-08-26 09:28:21 +00:00 |
|
|
|
5ceaeafc97
|
v1.1: 同步HTML
|
2026-08-26 08:03:11 +00:00 |
|
|
|
994ce3648a
|
v1.1: ASW_Basic重点对比增强——BaseM/N+SingleCoreM/N+swizzle+L2管理
|
2026-08-26 08:03:10 +00:00 |
|
|
|
4ec7435ce3
|
v0.95: 同步HTML
|
2026-08-26 07:36:15 +00:00 |
|
|
|
ffdb7c6f7b
|
v0.95: 修正baseK无dValue约束,dValue约束在kL1(GM→L1)
|
2026-08-26 07:36:13 +00:00 |
|
|
|
2035a569fb
|
v0.95: 同步HTML
|
2026-08-26 07:09:38 +00:00 |
|
|
|
5d6e3197ce
|
v0.95: ASW Step0/1分层+MergeBatch补L1绑定情形分界
|
2026-08-26 07:09:35 +00:00 |
|
|
|
d5a46280df
|
v0.94: 同步HTML
|
2026-08-26 06:50:59 +00:00 |
|
|
|
66e50d2e7e
|
v0.94: ASW_Basic Step0修正——SingleCoreM/N不受L0约束,由GM→L1效率+L2复用决定
|
2026-08-26 06:50:57 +00:00 |
|
|
|
95861e026c
|
v0.94: 同步HTML
|
2026-08-26 06:34:25 +00:00 |
|
|
|
586e8b2d1f
|
v0.94: ASW_Basic补Step0 singleCoreM/N确定+步骤编号统一
|
2026-08-26 06:34:24 +00:00 |
|
|
|
370a56715c
|
v0.94: 同步HTML
|
2026-08-26 03:55:24 +00:00 |
|
|
|
862e816c41
|
v0.94: ASW_Basic L2切分改用singleCoreM/N+执行组概念,去除mL2/nL2
|
2026-08-26 03:55:23 +00:00 |
|
|
|
44178b6175
|
v0.94: 同步HTML
|
2026-08-26 03:20:59 +00:00 |
|
|
|
1688782f14
|
v0.94: ASW_Basic L2切分补mL2/nL2确定方法+L1/L2独立性+修正示例
|
2026-08-26 03:20:58 +00:00 |
|
|
|
402b6390e7
|
v1.0: 同步HTML
|
2026-08-26 01:41:22 +00:00 |
|
|
|
1acde9ced4
|
v1.0: BMM理论最优vs源码实现逐分支对比分析
|
2026-08-26 01:41:21 +00:00 |
|
|
|
ccaab862f3
|
v0.93: 同步HTML
|
2026-08-25 12:40:14 +00:00 |
|
|
|
ca21aaebd4
|
v0.93: MergeBatch核心=GM→L1合并搬移+T_cmd替代T_bd+L0C双缓冲无同步开销
|
2026-08-25 12:40:13 +00:00 |
|
|
|
2795ad2062
|
v0.92: 同步HTML
|
2026-08-25 07:58:03 +00:00 |
|
|
|
cf617a3a4f
|
v0.92: 修复公式渲染——$$独占一行
|
2026-08-25 07:58:01 +00:00 |
|
|
|
33ea441a3f
|
v0.92: 同步HTML
|
2026-08-25 07:47:16 +00:00 |
|
|
|
756c406cb0
|
v0.92: StreamK P≤C/2+grid_K=⌊C/⌈P⌉⌋+O(100)推导+152→304修正
|
2026-08-25 07:47:15 +00:00 |
|
|
|
e2caca7cee
|
v0.91: 同步HTML
|
2026-08-25 07:18:28 +00:00 |
|
|
|
a534c280b4
|
v0.91: 一二阶合并建模+L1绑定/K截断解释+删SIMT引用+AIV不参与说明
|
2026-08-25 07:18:26 +00:00 |
|
|
|
ea28956f8f
|
v0.91: 同步HTML
|
2026-08-25 06:45:46 +00:00 |
|
|
|
5ddd8519f3
|
v0.91: 补T_bd证据(源码Set/Wait+知识库文献引用)
|
2026-08-25 06:45:45 +00:00 |
|
|
|
88ae1a84cb
|
v0.91: 同步HTML
|
2026-08-25 03:47:14 +00:00 |
|
|
|
8cf7dcf659
|
v0.91: 补T_bd物理成因+MergeBatch优势总结表
|
2026-08-25 03:47:12 +00:00 |
|
|
|
5e0ae53d40
|
v0.91: 同步HTML(修复math段<转义)
|
2026-08-25 03:26:28 +00:00 |
|
|
|
128076e8ae
|
v0.91: MergeBatch分界建模完善(执行模型+符号表+K截断区分)
|
2026-08-25 03:26:27 +00:00 |
|
|
|
5f4cbfa7b3
|
v0.91: 同步HTML
|
2026-08-25 03:23:02 +00:00 |
|
|
|
01ae8d70b3
|
v0.91: 修复cases环境中<转义
|
2026-08-25 03:23:01 +00:00 |
|
|
|
5e27f1840e
|
v0.91: 同步HTML
|
2026-08-25 03:21:20 +00:00 |
|
|
|
aad0a0e359
|
v0.91: MergeBatch分界建模补充执行模型/符号定义/K截断区分
|
2026-08-25 03:21:19 +00:00 |
|
|
|
826b662a96
|
v0.91: 同步HTML
|
2026-08-25 02:57:24 +00:00 |
|
|
|
8640a5c80d
|
v0.91: MergeBatch vs IterBatch 二阶分界模型
|
2026-08-25 02:57:22 +00:00 |
|
|
|
cfef8265ea
|
v0.9: 同步HTML
|
2026-08-25 02:20:02 +00:00 |
|
|
|
83d905fe7c
|
v0.9: MergeBatch vs IterBatch 一阶时延模型,条件1定位为二阶效应启发式
|
2026-08-25 02:20:01 +00:00 |
|
|
|
4af6eee0e3
|
v0.9: 同步HTML
|
2026-08-24 12:16:24 +00:00 |
|
|
|
26fa687bad
|
v0.9: StreamK推导用tile维度L0C/4B替代全矩阵MN,补芯片级→核级映射
|
2026-08-24 12:16:23 +00:00 |
|
|
|
5852556398
|
v0.9: 同步HTML
|
2026-08-24 03:43:00 +00:00 |
|
|
|
c8fabc39db
|
v0.9: MergeBatch b_core=2 drain暴露解释澄清
|
2026-08-24 03:42:59 +00:00 |
|
|
|
73b9cb2b06
|
PPT: MergeBatch b_core=2 说明
|
2026-08-24 03:24:09 +00:00 |
|
|
|
c04a850b02
|
v0.9: 同步HTML
|
2026-08-24 03:24:08 +00:00 |
|
|
|
3e26ece74c
|
v0.9: StreamK条件3改闭式K阈值+完整推导(含访存Bound分析),MergeBatch补b_core=2分析
|
2026-08-24 03:24:07 +00:00 |
|