From 742696640e0ee93499e04b2713dc3f75bc91a397 Mon Sep 17 00:00:00 2001 From: agent Date: Sun, 8 Nov 2026 03:02:34 -0400 Subject: [PATCH] docs/npu-plan: P4's A2 is tracked on its own page, so this list stops snapshotting it MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit #1380 landed "A2b, part 1" — the M=128 ELF set adopted in XdnaTarget::open_prefill() instead of in the forward — after this page said "A2 is what remains". Nothing was wrong when it was written; the page had turned another lane's in-flight work into a snapshot, and A2 is a sequence of slices (A2b part 1 of N), so any enumeration here goes stale within hours while its own page of record stays right. docs/npu-p4-target-split.md is that page of record (it already carries §3's A1/A2/A3 split), so the P4 entry now points at it and keeps only what stays visible from this page: load_block is still a named no-op that says why, so weights still fill lazily per projection. Checked against main rather than on paper: XdnaTarget::open_prefill exists (xdna_target.cpp:99) and is called from q4nx_forward.cpp:1120; XdnaTarget::load_block is still the named no-op (xdna_target.cpp:219-224); the link resolves (both files are in docs/); §3 does carry A1/A2/A3. Also re-checked the two claims of mine this page carries, since another lane has touched npu/target/ since they landed: check_kv_caps is still called from XdnaTarget::open() (xdna_target.cpp:175) before the device checks, and int8_kv is still only ever written false (target.h:92, xdna_target.cpp:242) — so "today every target reports int8_kv = false, so every int8 plan is refused" is still true and needed no change. Docs only, one file, +4/-3. tools/copyright.py --check rc 0. --- docs/npu-plan.md | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/docs/npu-plan.md b/docs/npu-plan.md index 19c72fbf2..44c1f44a1 100644 --- a/docs/npu-plan.md +++ b/docs/npu-plan.md @@ -175,10 +175,12 @@ layers[0] {"index": 0, "role": "gdn", "qkv": {"K": 5120, "N": 10240, "fused" ## Not yet -- **P4** (#803): **landed** in #808 (`0a2cbb9`) as **A1** — the `NpuTarget` interface (`npu/target/target.h`), - the XDNA target behind it, and `open()` consuming the plan instead of duplicating the derivation. **A2** is what - remains: the design derivation moves out of the forward and into the target, so `open()` becomes the source rather - than the check (and `load_block`, which is a named no-op today). +- **P4** (#803): **A1 landed** in #808 (`0a2cbb9`) — the `NpuTarget` interface (`npu/target/target.h`), the XDNA + target behind it, and `open()` consuming the plan instead of duplicating the derivation. **A2 is in progress and is + tracked on its own page of record**, [npu-p4-target-split.md](npu-p4-target-split.md) §3 (steps A1/A2/A3), so this + list does not try to snapshot it — as of #1380 (`A2b, part 1`) the M=128 ELF set is adopted in + `XdnaTarget::open_prefill()` instead of in the forward. What is still true from *this* page: `XdnaTarget::load_block` + remains a named no-op that says why, so weights still fill lazily per projection. - **P5** (#804): Hexagon target (t2 gemv on HVX/HMX, QEMU v73 oracle) has no engine rows yet. - **P6** (#805): the shared drafter protocol is on main; tau=0.6 and Hexagon acceptance rows are still missing. - **P7** (#806): the `kv` contract above is in the plan and its arithmetic is gated in CI (`ctest -R npu_kv_int8`,