micheleandClaude Sonnet 5
e4a5540b6e
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
129 lines
6.9 KiB
Plaintext
129 lines
6.9 KiB
Plaintext
# V2 timing log -- solo append, mai troncato/sovrascritto (vedi README.md)
|
|
# Nessuna entry ancora -- popolato incrementalmente man mano che avanza lo sviluppo V2.
|
|
|
|
[2026-09-05] EXP-0001 -- neural_processor (P_IN=8, ACC_WIDTH=32)
|
|
nextpnr-ecp5 --45k --package CABGA381 --speed 8 --freq 80
|
|
--lpf-allow-unconstrained
|
|
Fmax: 183.12 MHz -- PASS at 80 MHz (real place&route measurement)
|
|
Critical path: job_ready -> TRELLIS_IO output pad (i.e. the design's
|
|
own logic is NOT the bottleneck at this Fmax -- an unconstrained
|
|
output pin dominates; a real system-level Fmax will differ once
|
|
this signal is consumed on-chip instead of driven to a pad in
|
|
isolation). Log: hardware/v2/synthesis/neural_processor_p8/nextpnr.log
|
|
WNS/TNS: not reported by this nextpnr version in this invocation.
|
|
|
|
[2026-09-05] EXP-0002 -- neural_processor (P_IN=8, ACC_WIDTH=24)
|
|
Same command as above, ACC_WIDTH=24 build.
|
|
Fmax: 176.21 MHz -- PASS at 80 MHz (real place&route measurement)
|
|
Delta vs ACC_WIDTH=32: -6.91 MHz (-3.8%) despite FEWER resources --
|
|
attributed to placement noise, not a real architectural effect
|
|
(no seed sweep run to confirm either way -- see experiments.log
|
|
EXP-0002 and benchmark.log). Log: hardware/v2/synthesis/
|
|
neural_processor_p8_acc24/nextpnr.log
|
|
|
|
[2026-09-05] EXP-0003 -- N_PROCESSORS sweep (P_IN=8), real nextpnr-ecp5
|
|
--45k --package CABGA381 --speed 8 --freq 80 --lpf-allow-unconstrained
|
|
N=1: Fmax=159.11 MHz PASS MULT18X18D 8/72 (11%)
|
|
N=2: Fmax=149.59 MHz PASS MULT18X18D 16/72 (22%)
|
|
N=4: Fmax=151.01 MHz PASS MULT18X18D 32/72 (44%)
|
|
N=8: Fmax=134.70 MHz PASS MULT18X18D 64/72 (88%)
|
|
TRELLIS_IO stays at 18/245 (7%) throughout (harness has only
|
|
clk/rst/seed/checksum as real pins, by design -- see ERR-0005).
|
|
See decisions.log DEC-0005: DSP, not Fmax or LUT/FF, is the resource
|
|
that will first prevent scaling N_PROCESSORS further at P_IN=8.
|
|
|
|
[2026-09-05] EXP-0004 -- default-depth buffer configs, real nextpnr-ecp5
|
|
activation_buffer (D=4096): Fmax=325.20 MHz PASS
|
|
weight_buffer (D=512): Fmax=339.67 MHz PASS
|
|
result_buffer (D=4096): Fmax=325.20 MHz PASS
|
|
All far above the 80MHz target -- buffers are not a timing concern in
|
|
isolation at these depths.
|
|
|
|
[2026-09-05] EXP-0005 -- memory_manager + prefetch_engine (via
|
|
harness_memory_manager.v, see errors.log ERR-0005 for why a harness
|
|
was needed), real nextpnr-ecp5 --45k --package CABGA381 --speed 8
|
|
--freq 80 --lpf-allow-unconstrained
|
|
Fmax: 165.86 MHz -- PASS at 80MHz (real place&route measurement)
|
|
|
|
[2026-09-05] EXP-0006 -- neural_director (via harness_neural_director.v,
|
|
see errors.log ERR-0005 for why), real nextpnr-ecp5 --45k --package
|
|
CABGA381 --speed 8 --freq 80 --lpf-allow-unconstrained
|
|
Fmax: 250.50 MHz -- PASS at 80MHz (real place&route measurement)
|
|
|
|
[2026-09-05] EXP-0007 -- dependency_manager (N_NODES=16, standalone,
|
|
no harness needed), real nextpnr-ecp5 --45k --package CABGA381
|
|
--speed 8 --freq 80 --lpf-allow-unconstrained
|
|
Fmax: 155.30 MHz -- PASS at 80MHz (real place&route measurement)
|
|
|
|
[2026-09-05] EXP-0008 -- dataflow_core (via harness_dataflow_core.v,
|
|
see errors.log ERR-0005 for why a harness was needed), real
|
|
nextpnr-ecp5 --45k --package CABGA381 --speed 8 --freq 80
|
|
--lpf-allow-unconstrained
|
|
N_SLOTS=2: Fmax = 165.15 MHz -- PASS at 80MHz (real place&route)
|
|
N_SLOTS=4: Fmax = 133.19 MHz -- PASS at 80MHz (real place&route)
|
|
Fmax drops as N_SLOTS grows (more concurrent memory_manager+
|
|
neural_processor instances competing for the same routing fabric
|
|
around the shared neural_director/dependency_manager hub) -- both
|
|
configs still clear the 80MHz target with real margin.
|
|
|
|
[2026-09-05] EXP-0009 -- neural_multiprocessor (M8, N_SLOTS=2, real
|
|
standalone synthesis, no harness needed), real nextpnr-ecp5 --45k
|
|
--package CABGA381 --speed 8 --freq 80 --lpf-allow-unconstrained
|
|
Fmax: 142.45 MHz -- PASS at 80MHz (real place&route measurement)
|
|
|
|
[2026-09-05] EXP-0011 -- dataflow_core N_SLOTS=8 (via
|
|
harness_dataflow_core.v), real nextpnr-ecp5 --45k --package CABGA381
|
|
--speed 8 --freq 80 --lpf-allow-unconstrained
|
|
Fmax: 92.63 MHz -- PASS at 80MHz (real place&route measurement), but
|
|
with a much thinner margin than N_SLOTS=2 (165.15 MHz) or N_SLOTS=4
|
|
(133.19 MHz) -- Fmax falls monotonically with N_SLOTS (165.15 ->
|
|
133.19 -> 92.63 MHz for N_SLOTS=2/4/8) as routing congestion around
|
|
the shared neural_director/dependency_manager hub grows, while DSP
|
|
usage (88.9% at N_SLOTS=8) approaches its hard ceiling at the same
|
|
time -- see decisions.log DEC-0012.
|
|
|
|
[2026-09-05] EXP-0012 -- ACC_WIDTH 24 vs 32 real placement-seed sweep
|
|
(M10, "pipeline" axis, resolving EXP-0002's single-seed
|
|
inconclusiveness), 5 additional real nextpnr-ecp5 seeds (1-5) plus the
|
|
original default-seed run (EXP-0001/EXP-0002), same synthesized
|
|
netlists (hardware/v2/synthesis/neural_processor_p8/top.json and
|
|
neural_processor_p8_acc24/top.json), --45k --package CABGA381 --speed
|
|
8 --freq 80 --lpf-allow-unconstrained
|
|
ACC_WIDTH=32: seeds [default,1,2,3,4,5] -> [183.12, 164.85, 174.25,
|
|
168.83, 183.96, 145.73] MHz. n=6, mean=170.12 MHz, min=145.73 MHz,
|
|
max=183.96 MHz, stdev=14.16 MHz.
|
|
ACC_WIDTH=24: seeds [default,1,2,3,4,5] -> [176.21, 183.72, 180.57,
|
|
183.08, 185.49, 175.16] MHz. n=6, mean=180.71 MHz, min=175.16 MHz,
|
|
max=185.49 MHz, stdev=4.21 MHz.
|
|
Conclusion: over 6 real placement seeds, ACC_WIDTH=24 has a HIGHER
|
|
mean Fmax (+10.6 MHz, +6.2%) AND a much lower seed-to-seed variance
|
|
(stdev 4.21 vs 14.16 MHz, ~3.4x tighter) than ACC_WIDTH=32 --
|
|
EXP-0002's single-seed result (176.21 < 183.12, i.e. ACC=24 looked
|
|
WORSE) was placement-seed noise, not a real trend. See decisions.log
|
|
DEC-0013.
|
|
|
|
[2026-09-05] EXP-0014 -- neural_multiprocessor N_SLOTS=1 and N_SLOTS=4
|
|
(final benchmark campaign), real nextpnr-ecp5 --45k --package CABGA381
|
|
--speed 8 --freq 80 --lpf-allow-unconstrained
|
|
N_SLOTS=1: Fmax = 152.46 MHz -- PASS at 80MHz
|
|
N_SLOTS=4: Fmax = 113.38 MHz -- PASS at 80MHz
|
|
(N_SLOTS=2 reference, EXP-0009: 142.45 MHz)
|
|
Full real N_SLOTS sweep for the complete system (dataflow_core + real
|
|
V1 PSRAM chain + slot_mem_arbiter): 1 -> 152.46 MHz, 2 -> 142.45 MHz,
|
|
4 -> 113.38 MHz. Fmax falls monotonically as N_SLOTS grows (more
|
|
routing congestion around the shared director/dependency_manager/
|
|
arbiter hub), exactly the same trend already observed for
|
|
dataflow_core alone (M7/M10, EXP-0008/EXP-0011) but now measured for
|
|
the REAL FULL SYSTEM including the real PSRAM backend.
|
|
|
|
[2026-09-05] EXP-0015 -- neural_multiprocessor N_SLOTS=1/2/4 after the
|
|
word-level burst-read rewrite (DEC-0015), real nextpnr-ecp5 --45k
|
|
--package CABGA381 --speed 8 --freq 80 --lpf-allow-unconstrained
|
|
N_SLOTS=1: Fmax = 152.44 MHz -- PASS at 80MHz (was 152.46 MHz, unchanged)
|
|
N_SLOTS=2: Fmax = 133.58 MHz -- PASS at 80MHz (was 142.45 MHz, -6.2%)
|
|
N_SLOTS=4: Fmax = 112.07 MHz -- PASS at 80MHz (was 113.38 MHz, -1.2%)
|
|
Small, real Fmax cost from widening the shared arbiter/backend to
|
|
16-bit + lb_n/ub_n (extra routing), overwhelmingly outweighed by the
|
|
real cycle-count reduction (EXP-0015 in experiments.log/benchmark.log):
|
|
D-Stress real wall-clock improves 2.24-2.37x at every N_SLOTS.
|