daedalus-fourier/docs/k3_mc_phase7.md

cycle, phase, status, date_opened, date_closed, parent, host, verdict

cycle	phase	status	date_opened	date_closed	parent	host	verdict
3	7	closed 2026-05-18 — RED engineering / PASS 30fps-floor / M4 NEGATIVE	2026-05-18	2026-05-18	k3_mc_phase4.md (revised per phase5''')	hertz	cycle 3 closes; MC stays on CPU for higgs deployment; engineering negative documented

Cycle 3, Phase 7 — Verification (v1 + M4''')

v1 first-light

=== v3d MC 8h bench ===
  n_blocks: 65536  iters: 100

=== M1''': QPU vs C reference bit-exact ===
  blocks bit-exact: 65536 / 65536 (100.0000 %)

=== M2''': QPU throughput ===
  M2''' = 1.413 Mblock/s
  per-block = 707.9 ns
  per-dispatch = 46390.5 us
  R''' = 0.067 → RED band
  30fps@1080p floor: 1.5x margin (isolation)

shaderdb (v1 MC):

SHADER-DB-ffcca249...: 488 inst, 2 threads, 0 loops, 197 uniforms,
  25 max-temps, 0:0 spills:fills, 0 sfu-stalls, 488 inst-and-stalls, 7 nops

Phase 5''' finding 2 prediction confirmed: filter LUT inflated uniforms to 197 (gate was at ~144). Compiler also forced to 2 threads (from cycle-2's 4) due to register pressure (25 max-temps vs cycle-2's 21). The "no DP4A" structural deficit shows up directly here — 8 SMUL24 + 7 ADD per output pixel × 64 pixels per block × 8-lane geometry = 488 instructions, 30× heavier than the LPF kernel.

M4''' concurrent matrix (8s windows)

Config	Mblock/s	per-core (NEON)	vs NEON-4	30fps
NEON 1-core	14.479	—	—	14.9×
NEON 4-core	15.248	3.24 – 4.48	baseline	15.7×
QPU only	1.380	—	—	1.4×
Mixed NEON-3 + QPU	12.277	3.78 – 4.16	−19.5 %	12.6×
Mixed NEON-4 + QPU	12.158	2.49 – 3.35	−20.3 %	12.5×

M4 gate: FAIL. Mixed (12.28) < pure NEON-4 (15.25) by 2.97 Mblock/s. The QPU's 0.45 Mblock/s contribution under contention doesn't compensate for losing one NEON core that delivers ~3.8.

Cross-cycle comparison

	Cycle 1 IDCT	Cycle 2 LPF	Cycle 3 MC
R isolation	0.92	0.41	0.067
30fps floor margin (isolation)	7.9×	10×	1.5×
M4 mixed vs pure NEON-4	+7.2 %	+6.9 %	−19.5 %
30fps floor margin (mixed)	7.2×	7.2×	12.6×
Verdict for higgs	GO QPU	GO QPU	STAY CPU
NEON 4-core scaling vs 1-core	0.56× (bw-bound)	0.82× (bw-bound)	1.05× (compute-bound)

The MC result is structurally consistent with the V3D substrate profile from phase0.md:

• No DP4A → 8-wide convolution doesn't pack as it does on NEON SDOT
• Filter coefficients drive uniform count high → register pressure → 2 threads
• High per-output-pixel multiply count → compiled instruction count 3× cycle 1, 6× cycle 2

NEON 4-core is compute-bound for MC (not bandwidth-bound like the other two kernels). So 4-core scales nearly linearly with cores — the NEON CPU has plenty of headroom and the QPU has nothing to add even in concurrent mode.

Deployment recipe (for higgs / libva-v4l2-request-fourier)

Per project_consumer_target.md, the eventual integration target is V4L2 stateless → libva-v4l2-request-fourier → firefox-fourier. The back-end-on-QPU/CPU split for the consumed decoder pipeline:

• IDCT (cycle 1) → QPU. R = 0.92, +7 % mixed, frees a CPU core.
• LPF (cycle 2) → QPU. R = 0.41, +7 % mixed, frees a CPU core.
• MC (cycle 3) → CPU NEON baseline; QPU offload viable as opportunistic helper, not yet measured. R = 0.067 in isolation was discouraging; M4 same-kernel mixed was −19.5 % which looks conclusive but isn't — see M4 methodology caveat below.
• Entropy (VP9 Bool / AV1 ANS) → CPU. Structurally serial.

This is a mixed-substrate deployment, not a "QPU does everything" plan. Realistic for higgs: entropy + MC on 2-3 ARM cores; IDCT + LPF dispatched to QPU concurrently; 1-2 ARM cores left for vscode / etc.

M4 methodology caveat (added 2026-05-18 after cycle 5)

The M4 mixed bench (bench_concurrent_mc.c) tests NEON-3 + QPU running the SAME kernel concurrently. This is the worst case for memory-bandwidth contention — both substrates competing for the same bus with the same access pattern.

A real decoder pipeline has different shape: CPU runs entropy + MC

• other CPU-bound work; QPU runs IDCT + LPF + (potentially) MC as opportunistic helper. Different kernels on different substrates contend less than same-kernel-on-both. Our M4-same-kernel result is a pessimistic lower bound, not the actual deployment number.

Empirically supporting this: cycle 3 M4 showed per-core NEON throughput in 3-core mode (3.78-4.16 Mblock/s) was higher than in 4-core mode (3.24-4.48), confirming bandwidth saturation at ≥4 cores. So freeing 1 core via QPU offload costs ~25 % of total NEON MC throughput, but the QPU contributes 0.45 (-MC) or 1.4 (in CDEF isolation) on top.

To rigorously test the helper hypothesis: see docs/issues/003-mixed-kernel-m4-bench.md. A bench that runs NEON-3 on kernel-A + QPU on kernel-B concurrently would close the question. ~½ day of additional bench work; would update the deployment recipe for cycles 3 + 5 if the result is positive.

Issue 003 results (2026-05-18, closed)

bench_concurrent_mixed matrix in docs/issues/003-mixed-kernel-m4-bench.md confirms the methodology critique:

QPU side	CPU MC agg	per-core MC	QPU contribution
MC (V3 control, same kernel)	22.64 Mblock/s	7.5 avg	0.39 Mblock/s MC
LPF4 real QPU (V4)	27.87 Mblock/s	9.3 avg	12.74 Medge/s LPF4

Switching QPU off MC (same kernel) onto LPF4 (a different bandwidth-bound kernel) gave CPU MC +23 % per-core uplift. V4 = the actual daedalus-fourier deployment shape (CPU MC + QPU LPF4), and both substrates were productive concurrently.

Cycle 3 MC verdict unchanged: QPU MC contributes ~0.4 Mblock/s under any contention scenario (V3, V5). The 4 NEON cores do MC dramatically better. MC stays on CPU. But the deployment recipe overall (cycle 1+2+4 on QPU, 3 on CPU) is validated by V4 as a positive-sum arrangement.

Decision per Phase 1 rules + 30fps-floor calibration

Rule	Result	Status
M1''' bit-exact	100.0000 %	✓ PASS
R''' = M2'''/M3'''	0.067 (RED)	structural mismatch
M4''' > pure-CPU 4-core	−19.5 %	✗ FAIL gate
30fps@1080p floor (isolation)	1.5×	✓ PASS (user-facing)
30fps@1080p floor (mixed)	12.6×	✓ PASS (user-facing)

Engineering cycle verdict: do not deploy MC on QPU; deploy on CPU. User-facing cycle verdict: 30fps floor easily met in any configuration; either path works for daily YouTube.

For the deployment recipe above, MC stays on CPU. The Phase 1 ORANGE/RED "honest close" rule applies here: cycle 3 closes as a documented negative for this kernel without affecting the project-level "continue" verdict (cycles 1+2 GO results stand).

Phase 9 lessons (added to project memory)

1. Multiply-heavy workloads expose V3D's no-DP4A deficit in a way that cycle 1+2 didn't. CPU SDOT/UDOT pack 4 INT8 MACs in one instruction; V3D's SMUL24 is one scalar mult at a time. The 4× gap shows up directly as a 6-15× per-block slowdown.

2. Compute-bound CPU workloads make the QPU offload story collapse. When NEON 4-core scales near-linearly (not bandwidth-saturated), the "freed-core" argument from cycle 1+2 doesn't apply — there are no free cycles to free. Mixed mode is strictly worse.

3. The 30fps@1080p user-facing test (project_30fps_floor_is_fine.md) passes regardless of engineering verdict. All three cycles pass it in isolation. This is a project-level win to communicate separately from per-cycle engineering R numbers.

4. The shaderdb filter-LUT gate from phase5''' finding 2 fired exactly as predicted (197 uniforms > 144 threshold; 2 threads instead of 4). This validates the cycle-discipline of running V3D_DEBUG=shaderdb early and using the result as an actionable gate. Cycle 4 (if any) should bake this in from Phase 4 §design.

Leaves open

• Cycle 3 v2 with filter LUT escalated to SSBO (per phase5''' finding 2 trigger). Would reduce uniforms to ~30, potentially restore 4 threads. Expected upside: ~2× → R''' = 0.13. Still RED, still M4- negative. Skipped — even doubling doesn't change the deployment recipe.
• Vertical / hv / 4-tap / wider variants — all of cycle 3 same multiply-shape, same structural verdict expected. Not worth Phase 1+ for those.
• Cycle 4 candidates (per phase7_M4.md §"Cycle 3 candidates"): CDEF (AV1-only directional filter), Loop Restoration (AV1-only), or higgs deployment plumbing.