Repository navigation

[ASM] [HIP] [MLA v4 nm] Split plan helper and persistent decode kernel with in-kernel split and merge - #6126

Merged

yhl-amd merged 15 commits into

ROCm:mainfrom

yhl-amd:feat/mla-v4-nm-persistent-hca

Oct 4, 2026

Commits on Sep 27, 2026
- [MLA v4 nm] Cost-model num_kv_splits auto pick and kv_len_hint
  
  yhl-amd
  and
  claude
  committed
  1c835dc
Commits on Sep 28, 2026
- [MLA v4 nm] Expose the split plan to callers instead of kv_len_hint
  
  yhl-amd
  and
  claude
  committed
  3102016
- [MLA v4 nm] Return the split plan as a NamedTuple, per-arch cost table
  
  yhl-amd
  and
  claude
  committed
  1f0176a
Commits on Sep 29, 2026
- [MLA v4 nm] Add a persistent decode kernel with in-kernel split and merge
  
  yhl-amd
  and
  claude
  committed
  67d7422
- [MLA v4 nm] Add the generator for the persistent decode code object
  
  yhl-amd
  and
  claude
  committed
  e447b8b
- [MLA v4 nm] Drop the unused span buffer from the persistent decode API
  
  yhl-amd
  and
  claude
  committed
  c7cf2b0
- [MLA v4 nm] Add a guard-page out-of-bounds test for the persistent kernel
  
  yhl-amd
  and
  claude
  committed
  fe8dd38
- [MLA v4 nm] Rename research identifiers in the persistent kernel generator
  
  yhl-amd
  and
  claude
  committed
  4363a56
- [MLA v4 nm] Mark kernarg slot 0x68 of the persistent kernel as unused
  
  yhl-amd
  and
  claude
  committed
  1b6ca77
Commits on Oct 3, 2026
- [MLA v4 nm] Ship the persistent decode kernel as a code object only
  
  yhl-amd
  and
  claude
  committed
  5e6d258
Commits on Oct 4, 2026
- Merge main into feat/mla-v4-nm-persistent-hca
  
  yhl-amd
  and
  claude
  committed
  122d675
- [MLA v4 nm] Satisfy ruff in the persistent decode changes
  
  yhl-amd
  and
  claude
  committed
  5166ffb
- [MLA v4 nm] Expose the split count as a host-only helper
  
  yhl-amd
  and
  claude
  committed
  ba4551d
- [MLA v4 nm] Drop get_mla_v4_nm_split_plan; the split count is the API
  
  yhl-amd
  and
  claude
  committed
  7c96c03
- Merge branch 'main' into feat/mla-v4-nm-persistent-hca
  yhl-amd
  authored
  be371cc

Navigation Menu

Repository navigation

[ASM] [HIP] [MLA v4 nm] Split plan helper and persistent decode kernel with in-kernel split and merge - #6126

[ASM] [HIP] [MLA v4 nm] Split plan helper and persistent decode kernel with in-kernel split and merge#6126
yhl-amd merged 15 commits into
ROCm:mainfrom
yhl-amd:feat/mla-v4-nm-persistent-hca

Commits on Sep 27, 2026

[MLA v4 nm] Cost-model num_kv_splits auto pick and kv_len_hint

Commits on Sep 28, 2026

[MLA v4 nm] Expose the split plan to callers instead of kv_len_hint

[MLA v4 nm] Return the split plan as a NamedTuple, per-arch cost table

Commits on Sep 29, 2026

[MLA v4 nm] Add a persistent decode kernel with in-kernel split and merge

[MLA v4 nm] Add the generator for the persistent decode code object

[MLA v4 nm] Drop the unused span buffer from the persistent decode API

[MLA v4 nm] Add a guard-page out-of-bounds test for the persistent kernel

[MLA v4 nm] Rename research identifiers in the persistent kernel generator

[MLA v4 nm] Mark kernarg slot 0x68 of the persistent kernel as unused

Commits on Oct 3, 2026

[MLA v4 nm] Ship the persistent decode kernel as a code object only

Commits on Oct 4, 2026

Merge main into feat/mla-v4-nm-persistent-hca

[MLA v4 nm] Satisfy ruff in the persistent decode changes

[MLA v4 nm] Expose the split count as a host-only helper

[MLA v4 nm] Drop get_mla_v4_nm_split_plan; the split count is the API

Merge branch 'main' into feat/mla-v4-nm-persistent-hca