09fbf01ea523bf98926fc2b58aafaf0ab61d2e0e
29
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  micheleandClaude Sonnet 5
|
09fbf01ea5 |
feat: first genuine N=2 multi-core system, two real bugs found+fixed (EXP-0066)
New sdram_slot_arbiter2.v + tb_np_director_n2_system.v: real
neural_director_packed.v dispatching to 2 real packed_slot.v
instances sharing one real SDRAM controller. Jobs submitted one at a
time through the Director's own producer interface -- the Director's
own scheduling decisions determine slot assignment here, unlike every
prior V3 test.
Bug 1 (real, structural): the arbiter's first design registered its
grant one cycle late; layer_prefetch_ctrl.v's ctrl_req is a one-shot
pulse with no retry (every prior use wired it directly to a
controller, never behind arbitration), so a slot's first request
could be silently lost, hanging it forever. Fixed with a new
S_MEMWAIT state in packed_slot.v (wait for a combinational mem_grant
before ever pulsing layer_prefetch_ctrl's start) and a combinational-
first grant in the arbiter.
Bug 2 (testbench): node_id used a stray bit-slice (li[15:8]) instead
of a real multiply, making every layer produce the same node_ids and
silently checking results against the wrong layer's golden value.
Fixed.
Result: 12/12 PASS, 0 errors, real concurrent execution across both
slots (slot 0: positions {0,1,4,5,8,9}, slot 1: {2,3,6,7,10,11}).
Also noted (user correction): the SDR SDRAM controller used
throughout this memory path is a declared placeholder -- the real
target is DDR3 on a custom XC7A100T board, not yet built.
Full writeup in hardware/v2/logs/experiments.log EXP-0066.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC
|
||
|
micheleandClaude Sonnet 5
|
124a0dbca0 |
feat: packed_slot.v, real per-slot sequencer FSM (EXP-0065)
Promotes EXP-0062's own procedural testbench sequence (prefetch -> buffer swap -> per-tile gather -> operand streaming -> result capture) into real synthesizable RTL, wrapping layer_prefetch_ctrl.v -> layer_weight_buffer.v -> weight_tile_gather.v -> neural_processor_packed.v behind a 9-state FSM matching neural_director_packed.v's own per-slot contract. First run: 4/9 failed, deterministic. Root-caused (not a sequencer bug): the testbench's own w_base computation wrongly treated it as a byte address needing *2 conversion; layer_prefetch_ctrl.v expects a word address directly, and packed_slot.v already passes it through unconverted to match. Fixed the testbench. Re-verified: 9/9 PASS, 0 errors, bit-exact results and correct node_id/result_addr passthrough, entirely self-sequenced (no testbench-side procedural driving of the sub-modules). Full writeup in hardware/v2/logs/experiments.log EXP-0065. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
71600096f9 |
feat: neural_director_packed.v, job-pairing scheduler for packed cores (EXP-0064)
Forked from neural_director.v (M5): dispatches PAIRS of queued jobs (sharing w_base+n_tiles) to packed-core slots instead of one job per slot, matching neural_processor_packed.v's A/B job structure. If the two oldest queue entries don't share w_base/n_tiles, the Director stalls (never mis-pairs) -- a disclosed scope limitation, not hidden. Isolated testbench with behavioral per-slot stubs (same DEC-0007 scope decision as tb_neural_director.v). First run: 3/7 tests failed -- investigated each, root-caused as testbench timing bugs (checking dispatch state before the Director's own FSM had caught up, and a held-too-long job_in_valid making push counts ambiguous), not Director bugs. Fixed the testbench, re-verified: 8/8 PASS, 0 errors. Full writeup in hardware/v2/logs/experiments.log EXP-0064. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
5afa6a7477 |
exp: real P&R of verified weight-reuse memory path + packed core (EXP-0063)
New hardware/v3/rtl/np_packed_weight_reuse_top.v: flat structural synthesis wrapper around the EXP-0062-verified module chain (real SDR SDRAM controller -> layer_prefetch_ctrl.v -> layer_weight_buffer.v -> weight_tile_gather.v -> neural_processor_packed.v). Real Vivado post-route: 8/240 DSP48E1 (unchanged, memory path uses zero DSPs), WNS -2.502ns @ 200MHz -> Fmax ~133.3MHz, only -1.2% vs the isolated single core (134.9MHz, EXP-0059). Real memory-path control logic adds negligible Fmax cost at this scale. Full writeup in hardware/v2/logs/experiments.log EXP-0063. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
9ca180a787 |
test: first real end-to-end weight-reuse -> packed-core integration, bug found+fixed (EXP-0062)
New hardware/v3/sim/tb_np_packed_layer_reuse.v: real SDR SDRAM -> layer_prefetch_ctrl.v -> layer_weight_buffer.v -> weight_tile_gather.v -> neural_processor_packed.v, ALL real synthesizable RTL (unlike EXP-0058, which still had a testbench-only gather step). First run: 15/16 PASS, 1 FAIL. Root-caused (not re-run away): a testbench handshake bug, not a DUT bug -- operand_valid was held one extra clock edge after each accepted tile, double-consuming stale data every tile on every pair. 15 of 16 "passed" only because this test's saturating outputs happened to clamp to the same value whether or not the accumulator was inflated -- disclosed as a real methodology risk, not swept under the rug. Fixed by dropping operand_valid the same delta the handshake is observed. Re-verified after the fix: 16/16 PASS, 0 errors, bit-exact against an independent golden model, through the complete real RTL path. Full writeup in hardware/v2/logs/experiments.log EXP-0062. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
5c127fb069 |
feat: weight_tile_gather.v, real synthesizable byte-to-tile adapter (EXP-0061)
Closes the gap EXP-0058 left testbench-only: assembling P_IN sequential layer_weight_buffer.v byte reads into one weight_data tile bus, as real RTL instead of a testbench driver task. Avoids the runtime-indexed-part-select anti-pattern already found and fixed once in neural_director.v (ERR-0027-class Fmax collapse) by using a fixed shift-concat instead. Verified in isolation against a real, unmodified layer_weight_buffer.v: 37/37 tests, 0 errors, bit-exact across sequential, back-to-back, and non-sequential/repeated (real reuse-position-style) access patterns. Full writeup in hardware/v2/logs/experiments.log EXP-0061. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
94b63705be |
exp: N=8 packed-core array real P&R, isolates placement density from interconnect (EXP-0060)
New hardware/v3/rtl/np_packed_array.v: flat array of 8 unmodified neural_processor_packed.v instances, fully independent I/O, zero shared arbiter/Director logic -- isolates exactly one variable (DSP/placement density) from EXP-0059's single-core baseline. Real Vivado post-route: 64/240 DSP48E1 (26.67%), WNS -2.592ns @ 200MHz -> Fmax ~131.7MHz, only -2.4% vs the single-core 134.9MHz. Placement density alone is NOT the main driver of the ECP5-era gap between isolated-core and full-system Fmax -- narrows the question for the still-unbuilt real Director/arbiter/memory integration. Full writeup in hardware/v2/logs/experiments.log EXP-0060. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
34b15cc0a7 |
exp: first real Vivado synthesis+P&R on XC7A100T for V3 DSP-packed core (EXP-0059)
Real Vivado 2026.1 run (not estimated) confirms the DSP48 packing survives actual Xilinx synthesis: mac2_dsp_packed.v uses exactly 1 DSP48E1, and the full neural_processor_packed.v pipeline uses 8 DSP48E1/240 for 2 jobs -- half the DSP of two separate V2 cores for the same work. Post-route (real place_design+route_design, not synthesis-only): WNS -2.414ns @ 200MHz -> Fmax ~134.9MHz, within 0.5% of the post-synthesis-only estimate. This is the isolated compute core, out-of-context -- not yet a real N-core system number, flagged as such in the log entry. Full writeup, including the two real toolchain fixes needed to get Vivado running on this machine (CRLF line endings in installLibs.sh, missing libncurses.so.5 on Ubuntu 26.04), in hardware/v2/logs/experiments.log EXP-0059. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
80c89fa10d |
exp: real end-to-end weight-reuse integration with neural_processor.v (EXP-0058)
New tb_neural_processor_layer_reuse.v wires the real SDRAM controller, layer_prefetch_ctrl.v and layer_weight_buffer.v into a real neural_processor.v compute engine: one resident filter is fetched once and reused across 8 independent jobs per layer, verified bit-exact against an independent golden dot-product model (32/32 PASS). Also found and fixed a real testbench-vs-DUT scheduling race present in tb_layer_prefetch_ctrl.v (and hardened in the new file): clearing a one-cycle control pulse on the very next clock edge lands the clear in the same active-region pass as the edge a receiving module's own synchronous logic reads it at, so the pulse can be silently missed depending on implementation-defined process ordering. This had been silently preventing tb_layer_prefetch_ctrl.v's own claimed 8192/8192 result from ever actually being observed; fixed by holding the pulse past the edge with a real time delay before clearing, and the 8192/8192 result is now genuinely reproducible (5/5 consecutive runs). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
49b25f6eb6 |
exp: layer_prefetch_ctrl.v, real synthesizable RTL for layer-weight-reuse prefetch, fixes a real address-truncation bug (EXP-0057b)
Built the real FSM version of EXP-0057's own task-based prefetch pattern (bulk-sequential layer fetch via sdram_controller_openrow.v into layer_weight_buffer.v), so it's an actual instantiable module, not just a simulation convenience. Found and fixed a real bug in the process: cur_fill_addr's own address arithmetic bit-sliced BYTES_PER_BURST down to too few bits (BYTES_PER_BURST[BIDXW-1:0]), silently truncating 16 to 0 -- every burst's bytes landed at fill offset 0-15 instead of their real position, overwriting each other (only each layer's last burst survived). Root cause: misapplied a widening idiom used safely elsewhere in this codebase to a case where the target width was actually too small. Found via a standalone control-flow debug test first, then tracing data once control-flow was cleared. Verified: 8192/8192 bit-exact, 0 errors (was 512/8192 before the fix) through the real controller + SDRAM model, 16 layers. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
1ce78dff6e |
exp: N=16 timing closure fixed (EXP-0056), weight-reuse gives real 7.16x memory speedup without DDR3 (EXP-0057)
EXP-0056: N_SLOTS=16 failed timing on LFE5U-85F (23-24MHz vs 64MHz
target). First hypothesis (dependency_manager.v's serial ready-scan)
was wrong but real -- built and verified priority_encoder_lsb.v (a
generic recursive tree encoder) and dependency_manager_fast.v, bit-
exact equivalent to the original, but integrated it made no real
difference (24.26MHz). The real cause, found from nextpnr's own
critical-path report: nms_activation_fill_ctrl_v3.v's balanced max-
tree was only ever extended to N_SLOTS in {1,2,4,8}, silently falling
back to the original slow scan for 16. Added the missing case
(nms_activation_fill_ctrl_v3_n16.v), verified isolated (10017/10017)
and functionally (D-Stress N=16 still 256/256 bit-exact). Real result:
71.01MHz, PASS at 64MHz (single seed so far).
EXP-0057: built layer_weight_buffer.v, a double-buffered per-layer
weight scratchpad (fill one buffer in the background from SDRAM while
compute reads many times from the other -- weight-stationary reuse,
as opposed to D-Stress's own deliberately zero-reuse pattern). Wired
to the real sdram_controller_openrow.v + sdram_model.v, no new
hardware. For the same 32768 bytes of useful data: zero-reuse costs
27048 real cycles, reuse costs 3777 -- 7.16x real measured speedup on
the SAME SDR SDRAM, no DDR3, no clock change. This is the answer to
whether DDR3 is necessary for a workload class that actually has
reuse (e.g. conv-style face recognition, unlike D-Stress) -- it isn't,
at least not for this reason.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC
|
||
|
micheleandClaude Sonnet 5
|
ee5a68f6e6 |
exp: SDRAM CDC bridge + open-row policy (EXP-0053/54/55) -- open-row is a real ~5% D-Stress win, CDC bridge measured net-negative once integrated
EXP-0053: sdram_cdc_bridge.v decouples the SDRAM clock (115.2MHz, real value derived from the board's own existing PLL VCO=576MHz, verified via ecppll) from the 64MHz compute domain. Isolated: 137/137 tests, 0 errors, but real measured speedup is only 1.095x (not the naive 1.8x clock-ratio estimate) -- the CDC handshake's own synchronizer round-trip is a fixed per-transaction tax. EXP-0054: sdram_controller_openrow.v implements the page-hit/ keep-row-open optimization sdram_controller.v's own header had always deferred. weight_prefetch_engine_wide.v's real production traffic is strictly sequential per job and mostly stays within one SDRAM row -- closing/reopening it every tile (today's fixed auto-precharge policy) wastes tRP+tRCD for no reason. Isolated: 154/154 tests, 0 errors, 0 protocol violations (including the new refresh-while-row-open hazard, fixed via an explicit precharge-before-refresh path). Real measured speedup on the actual sequential access pattern: 1.141x. EXP-0055: composed both, then integrated into the real D-Stress benchmark (N=4/N=8, 256/256 bit-exact in every config). Result: open-row ALONE gives a real, consistent ~5% cycle-count improvement (47445/47468 vs baseline 49927/49909). CDC alone is a real ~8% REGRESSION. Combined is still a ~4% regression -- the CDC's fixed tax is paid on every transaction regardless of row-hit, and real D-Stress traffic interleaves weight-fetch/activation-result access far more than the isolated same-row test exercised, so open-row's real saving doesn't offset it. Decision: do not adopt the CDC approach; open-row alone is the disclosed, real win worth considering for production next, pending an explicit go-ahead (not applied to the real board top in this commit -- all additive, existing production RTL untouched). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MUG92aM9m68TRc4rG55BcC |
||
|
micheleandClaude Sonnet 5
|
03b5cbc25b |
exp: bank-interleaved SDRAM pipelining works in isolation, ~0.3% gain integrated (EXP-0052)
Follow-up to EXP-0051: built sdram_controller_pipelined.v, remapping addr->bank to low-order bits (today's weight region always maps to bank 0) and adding a shadow-slot ACTIVATE lookahead so a different-bank request can start its tRCD wait during the current transaction's tail. Phase A (isolated tb_sdram_controller_pipelined.v, 38/38 bit-exact, independently re-verified this session): mechanism works, saves exactly 2 cycles (tRCD) per different-bank back-to-back pair, matching the theoretical ceiling derived before measuring (CAS_LATENCY+BURST_LEN are serial on the shared data bus regardless of bank, so more than tRCD/tRP was never on the table). Phase B (integration, tb_nms_dstress_sdram_pipelined.v, independently rebuilt/rerun): N=4 49760 cycles (-0.33% vs baseline), N=8 49755 (-0.31%) -- both 256/256 bit-exact. Root cause of the gap: the W port's request/ready protocol is one-at-a-time, so a second, different-bank request is essentially never already pending while the first is still in flight, so the mechanism rarely triggers in the real system even though it's correct when directly stimulated. Not integrated into production; kept as additive reference for a possible future arbiter/backend pipelined-dispatch rewrite (out of scope here, larger and riskier). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YHENedK76onD2Vtc2CMjej |
||
|
micheleandClaude Sonnet 5
|
cc5db09f61 |
exp: dual-bank SDRAM sim shows W/AR split gives only ~9% cycle gain, not thrashing removal (EXP-0051)
Forked nms_neural_multiprocessor_sdram_unified.v + its D-Stress testbench into a dual-bank variant (two independent sdram_unified_backend.v instances, one for weight-fetch, one for activation+result) to test the Fase-3 memory-bound hypothesis ahead of Phase 2. Simulation-only: the real board (v2_board_top.lpf) still wires exactly one physical chip, per STEP19's governing single-SDRAM mandate. Result is honest but not the hoped-for one: splitting by traffic class only cuts D-Stress cycles ~8-10% (N=4: 49927->45724, N=8: 49909->44980), because the AR (activation+result) path was already lightly loaded (~12% busy) even alone. The real ceiling is the weight-fetch channel itself, which stays ~77-78% busy even with its own dedicated bank and zero cross-traffic. Full writeup in experiments.log EXP-0051, including the refined next-step options this suggests instead of a straight 2-bank board revision. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YHENedK76onD2Vtc2CMjej |
||
|
micheleandClaude Sonnet 5
|
685a4d6cfe |
docs+synth: Phase 0 baseline (N_SLOTS=4, real board top), fix stale LPF freq
New brief (N=8 timing closure, LFE5U-85F retarget, 4/8/16 x 1/2-bank SDRAM sweep). Phase 0: no RTL changes, only measure the current baseline. ERR-0030: constraints/v2_unified.lpf's FREQUENCY PORT "clk" was still 80MHz, a leftover from the STEP19 freeze, never updated to the project's real 64MHz target -- fixed (LPF only, zero RTL/datapath effect). ERR-0031 (bigger one): the first two synthesis attempts targeted nms_neural_multiprocessor_sdram_unified.v, which is NOT the real board-level top -- it's an obsolete wrapper only exercised by one testbench now. The real target is fpga_neural_v2_top.v (adds the real PLL, reset_sync, spi_host_bridge, and a second arbitration level), which is what actually goes through synthesis+P&R for hardware. Re-targeted correctly, matched against constraints/v2_board_top.lpf (all 17 ports real-ball-assigned). An N_SLOTS=8 P&R attempt against the WRONG (wrapper) target ran for 2h42m without converging on a single seed; discarded rather than trusted. N_SLOTS=8 baseline deferred by explicit user request until N_SLOTS=4 is fully understood -- re-attempt against the correct fpga_neural_v2_top target with an agreed time budget. Result (EXP-0049, fresh 8-seed nextpnr-ecp5 P&R, real pins): N_SLOTS=4 8/8 PASS at 64MHz, worst-seed 81.20MHz, mean 91.05MHz. Higher than DEC-0042's historical worst/best (64.55/77.21MHz) despite identical RTL -- disclosed, unresolved (leading hypothesis: nextpnr-ecp5 build/version difference), adopted as the operative baseline for this session's toolchain going forward. Full writeup in errors.log/timing.log/ experiments.log (EXP-0049). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
||
|
micheleandClaude Sonnet 5
|
8e014d8d49 |
V2.0.0 hardware freeze - single SDRAM
FASE #1 hardware freeze for FPGA-Neural V2, N4/P8, single external SDRAM (Alliance Memory AS4C4M16SA-6TIN) serving weights, activations, and results through one physical sdram_controller.v instance. Removes the PSRAM dependency (hardware/v1/rtl/psram_controller.v + memory_interface.v) from the V2 physical path entirely -- V1 itself remains fully unmodified, the golden reference. New RTL: sdram_unified_backend.v (2-way W/AR arbitration over one SDRAM controller, real per-byte DQM write masking added to sdram_controller.v for correct single-byte result writes with no read-modify-write), nms_neural_multiprocessor_sdram_unified.v (the frozen top-level). Two real bugs found and fixed via full-system testing before being accepted (ERR-0023): a deadlock and an off-by-one data-shift bug in the new arbitration logic. Real results: N=4 and N=2 D-Stress bit-exact (256/256 neurons), 40 real AUTO REFRESH events interleaved with zero corruption, real Yosys+nextpnr-ecp5 synthesis/P&R for LFE5U-45F-8CABGA381 (149/245 TRELLIS_IO, a real 45-pin reduction from the prior dual-memory design). Timing is MARGINAL (1/8 P&R seeds >=80MHz), reported honestly rather than masked by the best seed. Real, sourced ball-level pinout for the SDRAM bus + clk/rst (39/149 signals, P&R-verified) using the official Lattice ECP5U-45 pinout CSV found on disk during this step's own pre-commit review -- corrects an earlier draft that wrongly assumed no real pinout data was available. Chip readiness: NO. Real, disclosed blockers remain (no physical host interface exists yet -- the RTL's own reg_* ports are a 110-pin raw test-harness bus; clock source/PLL decision; power/configuration component selection) -- see hardware/v2/docs/{HARDWARE_FREEZE, CHIP_READINESS,OPEN_ITEMS}.md for the complete, itemized status. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
63cac6a7e5 |
perf(v2): shared activation cache - further 1.66-2.00x real speedup (DEC-0016)
Implements optimization #2 from the final benchmark campaign's own recommendation, on top of DEC-0015's word-level burst rewrite: a new shared activation_cache.v module fetches a given activation (X) vector from PSRAM once instead of once per neuron sharing it - the exact redundant traffic pattern the dense-layer workloads in this project's benchmark suite exhibit. Each memory_manager's own prefetch_engine now fetches WEIGHTS only; the activation half is requested from the shared cache instead (single-tag, tile-granular, N_SLOTS request ports, its own real word-level PSRAM backend via a new dedicated arbiter port). dataflow_core.v/slot_mem_arbiter.v/neural_multiprocessor.v widened to N_SLOTS+1 ports to arbitrate the cache's traffic alongside each slot's weight traffic. Two real bugs found and fixed during implementation (ERR-0010): a target-bank/pending-bank race in memory_manager.v's activation-cache wiring (the same bug class ERR-0006 already fixed once for pf_target_bank - a later handoff's queued request can overwrite which bank an earlier, still-in-flight request's ack applies to), and a repeat of ERR-0009's N_SLOTS=1 zero-width replication bug in activation_cache.v itself. Real, measured results: the full final-benchmark campaign (24/24 workload/config combinations) re-verified bit-exact. D-Stress cycles fall a further 1.66-2.00x on top of DEC-0015 (~4x combined vs the original byte-level baseline). But the cache's real Fmax cost is much steeper than DEC-0015's own: N_SLOTS=2 (the recommended default, DEC-0014) drops from 133.58 to 87.72 MHz (-34%, margin over 80MHz shrinks from +67% to +9.7%), and N_SLOTS=4 drops to 65.01 MHz - now FAILING the 80MHz target it previously passed. Combined real wall-clock speedup vs the original baseline: N=1 3.86x, N=2 2.45x (both real net wins); N=4 is a real regression once its own now-failing Fmax is honestly used, though N=4 was never the recommended configuration. N_SLOTS=2 remains the recommended default (DEC-0014 unaffected) with a thinner but still real Fmax margin. Cache hit-detection pipelining is flagged as concrete follow-up work if N_SLOTS>2 is ever needed with the cache active - not attempted this round. Logged: simulation/synthesis/timing/benchmark/decisions (DEC-0016)/ experiments (EXP-0016)/errors (ERR-0010)/development.log, ROADMAP.md updated. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
e4a5540b6e |
perf(v2): word-level burst reads - 2.24-2.37x real wall-clock speedup (DEC-0015)
Implements optimization #1 from the final benchmark campaign's own recommendation: exploit psram_controller.v's already-implemented page-mode support (confirmed present by direct inspection) by fetching multiple bytes per real backend transaction instead of one at a time. Root cause addressed: int8_memory_access.v (the byte-level backend prefetch_engine.v originally sat on) already converts every 8-bit logical request into a full 16-bit PSRAM word access internally (mem_addr <= addr >> 1), discarding half of every word it already paid for. prefetch_engine.v/memory_manager.v now speak memory_interface.v's own 16-bit word protocol directly, bypassing int8_memory_access.v entirely - which remains untouched, still frozen V1 (§1/§34); V2 simply reuses the lower layer of the same frozen chain instead of the byte-splitting layer on top of it, the same "reuse what fits" precedent slot_mem_arbiter.v already set. slot_mem_arbiter.v and neural_multiprocessor.v widened to match (lb_n/ub_n added, master port wired directly to memory_interface.v). Real, measured results: M4's own single-job testbench shows 49-56% fewer cycles (166->84, 446->204, 728->322, all still bit-exact). The full final-benchmark campaign (24/24 workload/config combinations) re-verified bit-exact with D-Stress's real wall-clock time (cycles / real POST-P&R Fmax) improving 2.24-2.37x across every N_SLOTS tested, against a small real Fmax cost (unchanged at N=1, -6.2% at N=2, -1.2% at N=4). tb_neural_multiprocessor.v (M8) and tb_benchmark_suite.v (final campaign) needed zero changes - both treat neural_multiprocessor.v as a black box. Only tb_memory_manager.v (M4, rewired to skip int8_memory_access.v) and tb_dataflow_core.v (M7, behavioral model widened to word-level) needed updates. The "real parallel scaling is flat beyond N_SLOTS=2" finding (DEC-0014) still holds - this optimization made the shared PSRAM port more efficient per transaction, not multi-ported - so N_SLOTS=2 remains the recommended default. Logged: simulation/synthesis/timing/benchmark/decisions (DEC-0015)/ experiments (EXP-0015)/development.log. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
3cdaeaee35 |
test(v2): final benchmark campaign - real end-to-end characterization (EXP-0014)
Post-M10, user-requested final benchmark campaign: 6 realistic workloads (16-256 independent neurons in a shared-input dense-layer shape, a random-seeded 2-layer network with real cross-node PSRAM forwarding, and a 6-node 2-hop dependency diamond) x 4 concurrency levels (N_SLOTS=1/2/4/8) through the real, full neural_multiprocessor system (real V1 PSRAM chain, real slot_mem_arbiter). 24/24 runs PASS bit-exact against a software golden model (11,520 individual neuron/ node checks, zero mismatches). Three real bugs found and fixed during the campaign itself (ERR-0009): 1. neural_director.v (M5) had a real RTL bug at N_SLOTS=1 ($clog2(1)=0 makes a replication expression illegal) - never caught because M5-M10 only ever tested N_SLOTS=2/4/8. Fixed with a width-agnostic '0 literal; M5's own testbench re-verified unaffected. 2/3. Two testbench sizing bugs in tb_benchmark_suite.v itself (psram_model DEPTH too small for the Large workload's address range; N_NODES too small for the Stress workload's node-id range, causing a real deadlock via node-id wraparound colliding with an already-DISPATCHED node - a real, honest consequence of DEC-0008's own "no node-slot reclamation" design choice). Headline finding: real parallel scaling is essentially flat beyond N_SLOTS=2 - the single shared PSRAM port saturates at ~91% utilization regardless of slot count, so memory-bound workloads gain only 1.05-1.06x real speedup from N=1 to N=8. Once real POST-P&R Fmax degradation is also factored in, N_SLOTS=4 is measurably 21% SLOWER in real wall-clock time than N_SLOTS=1 for the largest workload tested. N_SLOTS=2 is recommended as the default (DEC-0014, superseding DEC-0012's resource-only "N_SLOTS=8 ceiling" framing for general use). Full 21-section report (every number classified THEORETICAL/ SIMULATED/POST-P&R MEASURED/DERIVED, per the user's own methodology requirements): hardware/v2/docs/benchmarks/ final-benchmark.md Logged: simulation/synthesis/timing/benchmark/decisions (DEC-0014)/ experiments (EXP-0014)/errors (ERR-0009)/development.log, ROADMAP.md updated. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
91bbbe2fe5 |
feat(v2): M10 Optimization - data-driven findings, completes the V2 roadmap
Final milestone of docs/v2-description.md's §33 roadmap, scoped exactly to its own mandate: optimize only on data already gathered in M1-M9, across the pipeline/P_IN/processor-count/scheduling/memory axes - no speculative new features. Three concrete, data-driven results: 1. N_SLOTS=8 (numero processor axis): real synthesis + nextpnr-ecp5 P&R for dataflow_core at N_SLOTS=8, extending M7's N_SLOTS=2/4 sweep to the real DSP ceiling DEC-0005 predicted. 92.63 MHz POST-P&R, PASS at 80MHz, DSP 64/72 (88.9%). DEC-0012 recommends N_SLOTS=8 as the practical ceiling for P_IN=8 on the LFE5U-45F-8BG381. 2. ACC_WIDTH 24 vs 32 (pipeline axis): a real 6-seed nextpnr-ecp5 placement sweep (reusing already-synthesized netlists, no new synthesis needed) resolves EXP-0002's single-seed inconclusiveness. ACC_WIDTH=24 wins on both mean Fmax (+6.2%, 180.71 vs 170.12 MHz) and seed-to-seed variance (~3.4x tighter), on top of its already-known resource advantage. DEC-0013 recommends ACC_WIDTH=24 as the new default. 3. Stall %/utilization (scheduling/memory axes): testbench-only cycle counters added to tb_neural_multiprocessor.v (no RTL touched) close DEC-0011's deferred measurement gap with real data - shared PSRAM port 81.7% utilized, slot 0 95.2%, slot 1 65.2%, no conclusive evidence of harmful fixed-priority starvation at this scale. The 10-milestone V2 roadmap (docs/v2-description.md §33) is now complete end-to-end: real Verilator simulation, real Yosys synthesis, real nextpnr-ecp5 place & route for every milestone, fully logged (EXP-0001..EXP-0013, DEC-0001..DEC-0013, ERR-0001..ERR-0008) with no invented results (§30) and V1 kept frozen and untouched throughout (§1/§34). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
84794a3d25 |
docs(v2): M9 full benchmark - V1 vs V2 comparison table (§32)
Consolidates real, already-measured data from hardware/v1/ (frozen, pre-certified) and V2's own M1-M8 logs into the §32-mandated comparison table, on an apples-to-apples basis: both full systems (V1's spi_neuron_top post_fix_verify vs V2's neural_multiprocessor N_SLOTS=2), both PARALLEL=8/P_IN=8, both using the real unmodified V1 PSRAM backend. Headline, all real measurements: V2 full-system Fmax 142.45 MHz POST-P&R (PASS at 80MHz) vs V1's 68.65 MHz (FAIL at 80MHz); 166 vs 209 real simulated cycles for one neuron's 8-input dot product through the same real PSRAM chain (2.6x wall-clock speedup); peak MAC/cycle 16 (N_SLOTS=2 concurrent slots, real contention already demonstrated in EXP-0009) vs V1's 8 (single sequential core); lower LUT/FF despite V2 already including full dependency-graph scheduling that V1 has none of. 9 of the table's 12 rows carry real sourced numbers; stall %/memory utilization/processor utilization are reported as NOT MEASURED rather than approximated (DEC-0011) - a real number needs dedicated cycle-accounting instrumentation neither system has had built for it yet, and approximating from partial data would violate §30's "no invented results" rule. Deferred to M10, which needs exactly this data to decide what to optimize. No new RTL this milestone - pure data consolidation, logged as EXP-0010. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
6cff2c8a7c |
feat(v2): M8 PSRAM integration - real V1 backend shared across concurrent slots
neural_multiprocessor.v wraps dataflow_core.v (M7, unmodified) around the real, unmodified V1 PSRAM backend chain (int8_memory_access -> memory_interface -> psram_controller), funneling N_SLOTS independent Memory Backend Interface ports through a new generic N-port arbiter (slot_mem_arbiter.v) inspired by (not copied from) V1's own mem_arbiter.v. Real concurrent-slot simulation immediately surfaced a genuine bug (ERR-0008): memory_manager/prefetch_engine's byte-level backend protocol is fire-and-forget (a single-cycle mem_req pulse with no accept handshake) - correct for M4's direct 1:1 connection, but a naive arbiter silently drops a pulse arriving while the shared bus is owned by another slot, hanging that slot forever. Fixed with a per-port pending-request latch, the same "queue, don't drop" idiom already used by memory_manager's own pf_pending register (ERR-0006). Verified (Verilator): 4/4 PASS with 2 slots genuinely contending for one real PSRAM port (444 cycles). No regression on M4's own testbench. Real synthesis + nextpnr-ecp5 P&R (no harness needed - real PSRAM pins keep the top-level at 157 pins): 0 problems, Fmax 142.45 MHz, PASS at 80MHz. Arbitration policy is fixed lowest-index priority, not fairness- balanced (DEC-0010) - consistent with every other "simplest correct policy first" scheduling choice in this roadmap, revisited only if M9's real measurement shows starvation matters. Logged: simulation/synthesis/timing/benchmark/decisions (DEC-0010)/ experiments (EXP-0009)/errors (ERR-0008)/development.log, ROADMAP.md updated. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
77baa8fc16 |
feat(v2): M7 Dataflow Core - full M1-M6 integration, wake-up loop closed end-to-end
dataflow_core.v integrates dependency_manager (M6) -> neural_director (M5) -> N_SLOTS x (memory_manager (M4) + neural_processor (M1)) for the first time. A slot's completion (via neural_director's new slot_node_id tracking, an additive port) feeds back as a producer_done event to dependency_manager, waking up any node that depended on it - closing the dataflow loop without external glue. Verified end-to-end (Verilator) on a 3-node DAG: two independent nodes plus a third depending on both, confirmed to dispatch only after both genuinely complete via real neural_processor computation. 4/4 PASS. Real synthesis + nextpnr-ecp5 P&R via a synthesis-only timing harness (bare per-slot backend ports exceed the LFE5U-45F's TRELLIS_IO budget, same pattern as ERR-0005): N_SLOTS=2 -> 165.15 MHz, N_SLOTS=4 -> 133.19 MHz, both PASS at 80MHz, 0 synthesis problems. Scope explicitly deferred to M8 (DEC-0009): M3's BRAM buffers not wired in yet, per-slot Memory Backend Interface ports not arbitrated to one shared PSRAM master yet - both need real measured data before committing to a design, not guessed at here. Logged: simulation/synthesis/timing/benchmark/decisions (DEC-0009)/ experiments (EXP-0008)/errors (ERR-0007, a Yosys chparam-ordering build quirk, not an RTL bug)/development.log, ROADMAP.md updated. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
8af16d3a12 |
feat(v2): M6 Dependency Manager, multi-dependency wake-up
Implements M6: dependency_manager.v tracks a table of node
descriptors (node_id/state/required_dependencies/resolved_
dependencies/producer_ids -- §10's exact field list), incrementing a
waiting node's resolved count whenever one of its listed producers
completes, transitioning it to READY once resolved==required, and
dispatching ready nodes to the Neural Director (M5) one at a time via
a backpressure-safe valid/ready interface.
Verified with Verilator on a small hand-built DAG: node0/node1 have no
dependencies (dispatch immediately); node2 depends on BOTH node0 AND
node1 ("dipendenze multiple") and stays WAITING until both complete,
confirmed via an explicit negative check after only one resolves;
node3 depends on node0 ALONE, demonstrating a single producer
("node0") satisfying two different consumers' dependencies
("risultati condivisi... piu' consumer") -- node3 fully, node2
partially. 4/4 tests pass.
Scope for this milestone (decisions.log DEC-0008): dependency
COUNTING/readiness only, no direct producer-to-consumer value
forwarding (§11 frames that as a "quando possibile" optimization, not
a correctness requirement -- deferred until real bandwidth
measurements justify it) and no node-slot reclamation after dispatch
(not exercised by any scenario built so far).
Real synthesis: 0 CHECK problems, 763 LUT4/474 FF/0 DSP/0 CCU2C. Real
place&route (module fits the TRELLIS_IO budget as a bare top-level
this time, no harness needed): Fmax 155.30 MHz, PASS at 80MHz.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v
|
||
|
micheleandClaude Sonnet 5
|
2e4cedc761 |
feat(v2): M5 Neural Director, first-free job scheduling
Implements M5: neural_director.v dispatches job descriptors to whichever of N_SLOTS (memory_manager, neural_processor) pairs is currently free (first-free scheduling per §9's initial policy), with a parametric-depth ready-queue FIFO for jobs arriving faster than slots can absorb them. Scope for this milestone (see decisions.log DEC-0007): a reduced 4-state FSM (DIR_IDLE/SCAN_READY/ALLOCATE/ERROR) rather than §9's full 8-state baseline -- dependency tracking, the waiting queue, and wake-up are §10's explicit responsibility (Dependency Manager, M6, not yet built), and slot-completion detection runs as an always-active per-slot tracker rather than a dedicated FSM state, for the same reason DEC-0002 already gave for the Neural Processor's own FSM (gating concurrent per-unit progress behind one shared state kills throughput). Verified with Verilator (N_SLOTS=2, each slot backed by its own independent behavioral memory rather than sharing V1's real PSRAM -- M4 already proved that path for one slot; this milestone's own concern is scheduling across multiple slots): 4/4 tests pass -- 3 jobs submitted to 2 slots (first two dispatch immediately, third correctly queues until a slot frees), and a deliberate burst that forces the ready queue to genuinely fill and recover. Real synthesis: 0 CHECK problems, 382 LUT4/366 FF/4 CCU2C/0 DSP. Real place&route (via a synthesis-only timing harness, same TRELLIS_IO pin-budget reason as M2/M4): Fmax 250.50 MHz, PASS at 80MHz. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
175f697ae1 |
feat(v2): M4 Memory Manager + Prefetch Engine, real V1 PSRAM backend
Implements M4: memory_manager.v (arbitration/buffering/forwarding/ latency hiding/double buffering, §12) + prefetch_engine.v (double-buffered tile fetch, §13), sitting on the REAL, UNMODIFIED V1 PSRAM backend chain (int8_memory_access.v -> memory_interface.v -> psram_controller.v, per §15's explicit mandate not to touch the controller). Verified fully end-to-end with Verilator: real neural_processor (M1) fed entirely by memory_manager, computing against PSRAM-resident X/W tiles (double-buffered prefetch across up to 5 tiles) and writing its result back to PSRAM -- checked via an independent PSRAM read-back, with poison bytes around the operand regions to catch addressing errors. 3/3 jobs pass (1/3/5-tile configurations). Three real RTL bugs found and fixed during integration (full diagnostic trail in errors.log ERR-0006): prefetch_engine had no single-in-flight-request discipline, letting a queued request corrupt the bank bookkeeping of a fetch already running; the fix's own !pf_busy guard had a one-cycle blind spot (pf_busy lags pf_start by a clock) that needed an explicit !pf_start term; and a state-based mux for the shared backend port was off by one cycle, silently dropping the PSRAM result write entirely. Real synthesis: 0 CHECK problems, 851 LUT4/789 FF/108 CCU2C/0 DSP (expected, no multiplication in this module). Real place&route (via a synthesis-only timing harness, needed for the same TRELLIS_IO pin- budget reason as M2's array): Fmax 165.86 MHz, PASS at 80MHz. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
5f0d7f101c |
feat(v2): M3 activation/weight/result buffers, real BRAM mapping
Implements M3: three parametric dual-port buffers for the §12 data-plane (Input/Weight/Result), reusing the proven BRAM-inference idiom from the frozen hardware/v1/rtl/act_buffer.v (synchronous write, synchronous REGISTERED read, no reset on the read register -- keeps Yosys off the LUT-RAM path). Verified with Verilator: 10/10 tests pass (write-then-read correctness, extreme INT8 round-tripping, weight_buffer's full 64-bit tile width round-tripping, undisturbed re-reads). Real synthesis at two depths per module (6 configs total): 0 CHECK problems, every configuration correctly infers DP16KD (never LUT-RAM). Non-obvious real finding: weight_buffer's BRAM cost is driven by its P_IN*DATA_WIDTH tile width, not its DEPTH -- an 8x depth reduction (512->64) left DP16KD usage unchanged at 2, while activation_buffer/result_buffer (byte-wide) scale as naively expected (2->1). All default-depth configs PASS at 80MHz with large margin (287-367 MHz) via real nextpnr-ecp5 place&route. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |
||
|
micheleandClaude Sonnet 5
|
3026dcd997 |
feat(v2): M2 Neural Processor Array, N_PROCESSORS resource sweep
Implements M2 of the V2 roadmap: neural_processor_array.v instantiates
N_PROCESSORS independent neural_processor (M1) units, each with its
own dedicated point-to-point job/operand/result interface -- no shared
bus or mux at this level (arbitration is explicitly the Neural
Director's job, M5).
Verified with Verilator (tb_neural_processor_array.v, N_PROCESSORS=4):
7/7 tests pass, including a same-cycle 4-way concurrent launch with
different tile counts and a staggered-start test where a
later-launched, shorter job completes before an earlier-launched,
longer one -- confirming genuine independent concurrent execution
(§18/§34: a blocked/busy processor must not block the others).
Real resource/timing sweep for N_PROCESSORS in {1,2,4,8} (Yosys +
nextpnr-ecp5, real place&route): Fmax stays above the 80MHz target
throughout (159.11 -> 134.70 MHz), but MULT18X18D usage scales
linearly and reaches 88% of the LFE5U-45F's 72 DSPs at N=8 while
LUT/FF stay under 6% -- DSP, not LUT/FF/routing, is the first hard
ceiling on N_PROCESSORS at P_IN=8 (decisions.log DEC-0005). Measured
via a dedicated synthesis-only timing harness after the array's wide
per-processor buses were found to exhaust the device's TRELLIS_IO pin
budget as a bare top-level module beyond N=1 (errors.log ERR-0005) --
not a logic limit, an artifact of testing the array in isolation
before the Memory Manager/Director (M4/M5) exist to consume those
ports on-chip.
Full log trail (development/experiments/errors/decisions/simulation/
synthesis/timing/benchmark.log) in hardware/v2/logs/ per the project's
logging mandate.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v
|
||
|
micheleandClaude Sonnet 5
|
dc0b331d3e |
feat(v2): scaffold hardware/v1 frozen baseline + M1 Neural Processor
Begins the V2 Neural Multiprocessor / Dataflow architecture per docs/v2-description.md, per explicit user request to freeze V1 and start V2 development, copying from V1 what's needed. Scaffold: - hardware/v1/: byte-exact, read-only copy of the current V1 codebase (rtl, testbenches, tools, constraints, a representative subset of synthesis results, and reference docs) -- verified identical via diff/cmp against the live top-level tree before being made filesystem-read-only. The live top-level tree is untouched and remains the project's "production" V1 (see hardware/v1/README.md and hardware/v2/logs/decisions.log DEC-0001 for why copy-not-move). - hardware/v2/: mandatory structure (rtl/sim/constraints/synthesis/ reports/scripts/logs/docs) plus the full logging system required by the spec (development/architecture/simulation/synthesis/timing/ benchmark/decisions/experiments/errors.log). M1 -- Neural Processor (hardware/v2/rtl/neural_processor.v): - 8-stage pipelined perceptron unit (P_IN=8): input align, 8 multipliers, 3-level adder tree, accumulator, bias+activation, INT8 saturation. Genuine 1-tile/cycle throughput, not just a wider combinational datapath. - 7-state FSM (NP_IDLE..NP_ERROR per docs/v2-description.md §6, with 4 baseline states merged into NP_WAIT_OPERANDS -- see decisions.log DEC-0002); valid/ready/data/last stream interfaces per §7. - Bit-exact vs the frozen hardware/v1/rtl/neuron_parallel.v + mac8.v + mac_unit.v: 7/7 tests pass (hardware/v2/sim/tb_neural_processor.v), covering regular/mixed-sign/extreme-INT8 vectors, both activations, a zero-idle-gap back-to-back-tiles throughput check, and an 8-tile job -- verified with Verilator (see below for why). - Real synthesis + place&route (Yosys + nextpnr-ecp5): 0 CHECK problems, Fmax 183.12 MHz at ACC_WIDTH=32 (PASS at 80MHz, ~3x V1's isolated PARALLEL=8 Fmax of 61.71 MHz) and 176.21 MHz at ACC_WIDTH=24 (a user-requested comparison experiment, also bit-exact-verified; see experiments.log EXP-0001/EXP-0002 and benchmark.log). Three real bugs found and resolved during M1 development (full diagnostic record in errors.log): - Two independent, reproducible Icarus Verilog v13.0 scheduling defects (ERR-0001, ERR-0002) that silently produced wrong simulation results for standard sequential Verilog -- confirmed via Verilator 5.050 giving correct results on the same minimal repros. Verilator is now the trusted simulator for hardware/v2/ (decisions.log DEC-0004); Icarus's affected protocol-violation check was removed from the RTL and deferred architecturally to the Neural Director (DEC-0003) rather than chased further. - One real RTL bug (ERR-0003): last0 wasn't gated like valid0, letting a "last tile" tag leak into the pipeline ahead of its actual valid tile on back-to-back jobs. Fixed and verified. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v |