Skip to content

Add: DeepSeek-V4 FLASH full 43-layer decode network as an EP2 scene test - #1838

Merged
ChaoZheng109 merged 2 commits into
hw-native-sys:mainfrom
lwDavid:add-dsv4-flash-decode-ep2-example
Aug 14, 2026
Merged

Add: DeepSeek-V4 FLASH full 43-layer decode network as an EP2 scene test#1838
ChaoZheng109 merged 2 commits into
hw-native-sys:mainfrom
lwDavid:add-dsv4-flash-decode-ep2-example

Conversation

@lwDavid

@lwDavid lwDavid commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

What

Add the DeepSeek-V4 FLASH full 43-layer decode network (EP2/TP2, 2 dies) as an a2a3 tensormap_and_ringbuffer scene test — the first pypto-harvested distributed network example in the repo. The single-chip counterpart is qwen3_14b_decode/; the raw-Worker multi-chip counterpart is examples/workers/l3/ep_dispatch_combine/.

  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/: harvested chip orchestration (7.8k lines) + 367 incore kernels from pypto-lib models/deepseek_v4_flash_mtp/decode_fwd.py (pypto 289290e6, pypto-lib ef88d34, ptoas v0.57, pto-isa 83d01313). Codegen output is verbatim apart from three mechanical transforms: license headers, repo clang-format, and the whole-word renames L2TaskArgs->ChipTaskArgs / L0TaskArgs->CoreTaskArgs / Tensor->ChipTensor (pypto's pinned runtime predates the role-based naming contract).
  • The level-3 scene orch fn transcribes the generated host_orch.py: one comm domain (12 window buffers, ~6.6 MiB — MoE meta/payload/route, arrival counters, LM-head hidden/logits windows), 92 tensors + 13 scalars per rank, per-rank submit_next_level.
  • simpler_setup/goldens/deepseek_v4_flash_decode.py: deterministic ~100 GiB fixture for the 80 network parameters. Stacked per-rank tensors become independent _r0/_r1 args (the rehost passes whole buffers); metadata (block tables, routing tid2eid, positions, freqs) is bitwise-equal to the upstream builder; pypto's device-resident weight upload becomes plain host tensor args (verified equivalent upstream: the network passes with residency stripped).
  • Case is manual (the 367-kernel compile takes minutes — it stays out of the default sweep) and skip_golden (upstream drives the same fixture with golden_fn=None; numeric coverage lives with pypto-lib's standalone kernel harnesses). What it pins down: the harvested distributed program compiles, both ranks dispatch, the cross-die window protocol drains, and the run terminates cleanly.

Status

Blocked on the #1644 pto-isa pin bump — see issue #1839. The case PASSES on every runtime from #1729 (wire-ABI cutover) through #1771, all pinning pto-isa 83d01313; from 7a1b9b11 (pto-isa -> 0cefc9a5) onward, including main, both ranks stall mid-network in the comm-window polling path (full bisect matrix and candidate pto-isa commits in the issue). Landing the example gives that regression an in-tree repro; the manual marker keeps CI green meanwhile.

Testing

  • TestDeepseekV4FlashDecode::DecodeFwdEP2TP2 PASSED standalone on a2a3 (2 dies) at aa1d7c7d, cf0fbc06, and 4b13b69c with these exact sources.
  • On main the case reproduces the pto-isa regression deterministically (S1 running-stalled at ~layer 37 with the checked-in seed).
  • ruff check / ruff format clean on the two Python files; all 368 .cpp files pass the repo clang-format check.

@lwDavid lwDavid mentioned this pull request Aug 14, 2026
Open
@coderabbitai

coderabbitai Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown

Important

Review skipped

Too many files!

This PR contains 373 files, which is 273 over the limit of 100.

To get a review, reduce the PR to 100 files or fewer by splitting it into smaller PRs or changing its base branch.

Upgrade to a paid plan to raise the limit.

Usage-priced reviews support at most 300 files.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3265b2db-e1cb-4a7a-ad67-810f5e64b39a

📥 Commits

Reviewing files that changed from the base of the PR and between 3f4686c and 3358d0e.

📒 Files selected for processing (373)
  • .pre-commit-config.yaml
  • examples/a2a3/tensormap_and_ringbuffer/README.md
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/README.md
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_gate_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_gate_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_gate_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_gate_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_gate_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_up_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_up_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_up_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_up_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_up_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_w2_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_w2_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_w2_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_w2_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/exp_w2_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/gate_0_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/gate_1_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/gate_2_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/gate_3_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/gate_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_head_linear.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/hc_pre_linear_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/idx_qr_proj_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/idx_qr_proj_matmul_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_hadamard.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_hadamard_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_proj_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_proj_matmul_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_proj_matmul_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_proj_matmul_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_proj_matmul_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_score_proj.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_score_proj_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_score_proj_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_score_proj_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/kv_score_proj_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/lm_head_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_a_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_a_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_a_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_a_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_a_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_b_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_b_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_b_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_b_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/proj_b_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qk_pv_0_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qk_pv_1_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qk_pv_2_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qk_pv_3_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qk_pv_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qproj_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qproj_matmul_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qproj_matmul_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qproj_matmul_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qproj_matmul_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_hadamard_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_hadamard_matmul_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_proj_matmul.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_proj_matmul_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_proj_matmul_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_proj_matmul_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/qr_proj_matmul_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/score_0_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/score_aic.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_gate_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_gate_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_gate_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_gate_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_gate_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_up_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_up_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_up_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_up_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_up_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_w2_mm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_w2_mm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_w2_mm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_w2_mm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/sh_w2_mm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/weights_proj.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aic/weights_proj_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/build_valid.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/comb_sinkhorn_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_wait.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_wait_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_wait_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_wait_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/combine_wait_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_cache_writeback.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_cache_writeback_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_cmp_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_cmp_rope_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_rope_step.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_rope_step_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_slots_build_valid_qk_plan.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/csa_slots_build_valid_qk_plan_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/decode_build_cache_metadata.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/decode_build_swa_metadata.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/decode_build_swa_scalar_metadata.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_gather.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_gather_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_gather_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_gather_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_gather_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_meta.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_meta_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_meta_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_meta_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_meta_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_push.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_push_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_push_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_push_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_push_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_wait.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_wait_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_wait_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_wait_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/dispatch_wait_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_gate_up_act.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_gate_up_act_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_gate_up_act_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_gate_up_act_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_gate_up_act_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_h_q.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_h_q_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_h_q_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_h_q_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_h_q_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_w2_act.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_w2_act_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_w2_act_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_w2_act_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/exp_w2_act_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/ffn_norm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/ffn_norm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/ffn_norm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/ffn_norm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/ffn_norm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_0_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_1_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_2_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_3_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_pre_route.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_pre_route_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_pre_route_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_pre_route_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/gate_pre_route_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_head_reduce.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_head_rms.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_post_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_linear_reduce_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hc_pre_rms_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hca_cache_topk.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hca_cache_writeback.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hca_gather_kv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/hca_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/idx_qr_proj_dequant.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/idx_qr_proj_dequant_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_and_cache_write.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_and_cache_write_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_proj_reduce.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_proj_reduce_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_proj_reduce_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_proj_reduce_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_proj_reduce_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_rms_norm_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_rms_norm_rope_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_rms_norm_rope_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_rms_norm_rope_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_rms_norm_rope_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_touch.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/kv_touch_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_combine_gather.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_combine_push.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_combine_wait.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_dispatch_gather.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_dispatch_push.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_dispatch_wait.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_greedy_sample.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/lm_head_signal_clear.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/merge_norm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/merge_norm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/merge_norm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/merge_norm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/merge_norm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/mix_x_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/moe_signal_clear.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/pack_x_hc.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/proj_b_act.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/proj_b_act_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/proj_b_act_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/proj_b_act_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/proj_b_act_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/q_rope_prepare.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/q_rope_prepare_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/q_rope_prepare_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/q_rope_prepare_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/q_rope_prepare_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qk_pv_0_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qk_pv_1_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qk_pv_2_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qk_pv_3_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qk_pv_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qproj_dequant_rms_nope_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qproj_dequant_rms_nope_rope_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qproj_dequant_rms_nope_rope_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qproj_dequant_rms_nope_rope_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qproj_dequant_rms_nope_rope_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_hadamard_quant.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_hadamard_quant_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_proj_reduce.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_proj_reduce_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_proj_reduce_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_proj_reduce_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_proj_reduce_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rms_norm_quant.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rms_norm_quant_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rms_norm_quant_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rms_norm_quant_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rms_norm_quant_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rope_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rope_swap_idx.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/qr_rope_swap_idx_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/quant.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/quant_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/quant_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/quant_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/quant_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rms_norm_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rmsnorm_rope.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rmsnorm_rope_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rmsnorm_rope_cache_write.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rmsnorm_rope_cache_write_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rmsnorm_rope_cache_write_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_cs.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_cs_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_cs_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_cs_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_cs_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_interleave.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_interleave_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_interleave_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_interleave_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_interleave_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/rope_swap.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_hash.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_hash_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_hash_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_sort.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_sort_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/route_sort_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/scatter_softmax_pool.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/scatter_softmax_pool_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/scatter_softmax_pool_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/scatter_softmax_pool_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/scatter_softmax_pool_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/score_0_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/score_aiv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_gate_up_act_q.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_gate_up_act_q_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_gate_up_act_q_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_gate_up_act_q_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_gate_up_act_q_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_w2_act.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_w2_act_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_w2_act_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_w2_act_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/sh_w2_act_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/shared_routed.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/shared_routed_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/shared_routed_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/shared_routed_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/shared_routed_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_4.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_5.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_6.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_7.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/split_pre_post_8.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_cache_insert_valid_bias.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_cache_insert_valid_bias_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_gather_kv.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_gather_kv_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_rope_step.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/swa_rope_step_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/topk.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/topk_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/weights_proj_reduce.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/weights_proj_reduce_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/x_norm_quant.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/x_norm_quant_0.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/x_norm_quant_1.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/x_norm_quant_2.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/aiv/x_norm_quant_3.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/kernels/orchestration/decode_fwd.cpp
  • examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/test_deepseek_v4_flash_decode.py
  • simpler_setup/goldens/deepseek_v4_flash_decode.py

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@lwDavid lwDavid self-assigned this Aug 14, 2026
@lwDavid lwDavid added the enhancement New feature or request label Aug 14, 2026
@lwDavid lwDavid moved this to Done in pto project Aug 14, 2026
@lwDavid
lwDavid force-pushed the add-dsv4-flash-decode-ep2-example branch 3 times, most recently from 767c6b1 to d0e9c92 Compare August 14, 2026 09:08
@lwDavid
Add: DeepSeek-V4 FLASH full 43-layer decode network as an EP2 scene test
d0e9c92 
Port the complete DeepSeek-V4 FLASH decode forward from pypto-lib
(models/deepseek_v4_flash_mtp/decode_fwd.py) as the first pypto-harvested
distributed example: a level-3 scene test on 2 dies with expert-parallel MoE
dispatch/combine and a TP2 LM head through a comm domain.

- examples/a2a3/tensormap_and_ringbuffer/deepseek_v4_flash_decode/: harvested
  chip orchestration (7.8k lines) + 367 incore kernels, verbatim pypto codegen
  plus license headers, clang-format, and the whole-word renames
  L2TaskArgs->ChipTaskArgs / L0TaskArgs->CoreTaskArgs / Tensor->ChipTensor
  (pypto's pinned runtime predates the role-based naming contract).
- The scene orch fn transcribes the generated host_orch.py: one comm domain
  (12 window buffers, ~6.6 MiB), 92 tensors + 13 scalars per rank, per-rank
  submits. Stacked per-rank slices become independent _r0/_r1 host tensors
  (the rehost passes whole buffers), and pypto's device-resident weight
  upload becomes plain host tensor args.
- simpler_setup/goldens/deepseek_v4_flash_decode.py: deterministic fixture for
  the 80 network parameters (~100 GiB at full size), metadata transcribed
  exactly, weight distributions matching the upstream spec builders.
- The case is manual (367-kernel compile takes minutes) and skip_golden
  (upstream runs the same fixture with golden_fn=None; numeric coverage lives
  with the standalone kernel harnesses in pypto-lib).

- .pre-commit-config.yaml: exclude the harvested orchestration translation unit
  from check-added-large-files (708 KiB is inherent to the 43-layer program).
ChaoZheng109
ChaoZheng109 previously approved these changes Aug 14, 2026
View reviewed changes
@ChaoZheng109
ChaoZheng109 dismissed their stale review August 14, 2026 11:31

.pre-commit-config.yaml

@lwDavid
Raise check-added-large-files cap to 750 kB instead of excluding the …
3358d0e 
…orchestration

The harvested whole-network chip orchestration decode_fwd.cpp is 725 350
bytes (~708 kB) — one generated translation unit whose size is inherent
to the 43-layer program. Keep it under the hook's scrutiny by raising the
global cap from the 500 kB default to 750 kB rather than excluding its
directory.
@ChaoZheng109
ChaoZheng109 merged commit 8a6a28f into hw-native-sys:main Aug 14, 2026
34 of 35 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@ChaoZheng109 ChaoZheng109 ChaoZheng109 approved these changes
@ChaoWao ChaoWao Awaiting requested review from ChaoWao

Assignees

@lwDavid lwDavid

Labels

enhancement New feature or request

Projects

pto project
Status: Done

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@lwDavid @ChaoZheng109