Files
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
2026-09-05 14:06:53 +02:00

239 lines
7.6 KiB
Verilog

`timescale 1ns/1ps
// ================================================================
// MEM_ARBITER
//
// Arbitrates a single shared byte-level memory master port (feeding
// a shared int8_memory_access -> memory_interface -> psram_controller
// chain) between three byte-level requesters:
//
// Port A: spi_engine.v (WRITE_RAM / READ_RAM opcodes)
// Port B: neuron_memory.v (its own X/W/bias reads during a run)
// Port C: layer_sequencer.v (Phase 5: descriptor reads + output
// buffer writes between layers)
// Port D: flash_copy_engine.v (flash-subsystem F2: flash<->PSRAM
// block DMA, LOWEST priority -- see
// below)
//
// Fixed priority B > C > A > D when more than one requests on the
// same idle cycle (an in-progress inference is treated as more
// time-critical than the sequencer's own bookkeeping, which in turn
// is treated as more time-critical than a newly-arriving manual SPI
// RAM access, which in turn is treated as more time-critical than
// the flash copy engine -- flash operations are ms-scale and never
// meant to compete with inference for memory bandwidth, per the
// flash-subsystem phase-plan's explicit "priorita bassa" requirement:
// a flash load/save simply waits its turn, one byte-transaction at a
// time, behind anything else that wants the shared PSRAM port).
// In normal operation B and C are temporally disjoint anyway --
// neuron_memory only requests while running, and layer_sequencer
// only requests in the gaps between layers -- so priority among
// A/B/C mostly matters for the edge case of a manual
// WRITE_RAM/READ_RAM arriving while a Phase 5 run is in progress.
// Port D is expected to be active only during flash load/save,
// which this design assumes does not overlap real-time inference
// (the same "not the hot path" assumption the flash phase-plan
// states explicitly) -- if it ever did overlap, its lowest-priority
// placement here means it simply gets stretched out, never starves
// or corrupts A/B/C.
// Once a port is granted, the arbiter holds ownership until that
// single transaction's m_ready pulse, then releases -- all four
// masters already issue `req` as a clean one-cycle pulse (matching
// int8_memory_access's own contract), so a simple grant-and-forward
// design is sufficient; no request queuing/pipelining is needed.
// ================================================================
module mem_arbiter #(
parameter ADDR_WIDTH = 23
)(
input wire clk,
input wire rst,
// ------------------------------------------------------------
// Port A - spi_engine
// ------------------------------------------------------------
input wire a_req,
input wire a_wr,
input wire [ADDR_WIDTH-1:0] a_addr,
input wire signed [7:0] a_wdata,
output reg signed [7:0] a_rdata,
output reg a_ready,
// ------------------------------------------------------------
// Port B - neuron_memory
// ------------------------------------------------------------
input wire b_req,
input wire b_wr,
input wire [ADDR_WIDTH-1:0] b_addr,
input wire signed [7:0] b_wdata,
output reg signed [7:0] b_rdata,
output reg b_ready,
// ------------------------------------------------------------
// Port C - layer_sequencer
// ------------------------------------------------------------
input wire c_req,
input wire c_wr,
input wire [ADDR_WIDTH-1:0] c_addr,
input wire signed [7:0] c_wdata,
output reg signed [7:0] c_rdata,
output reg c_ready,
// ------------------------------------------------------------
// Port D - flash_copy_engine (F2, lowest priority)
// ------------------------------------------------------------
input wire d_req,
input wire d_wr,
input wire [ADDR_WIDTH-1:0] d_addr,
input wire signed [7:0] d_wdata,
output reg signed [7:0] d_rdata,
output reg d_ready,
// ------------------------------------------------------------
// Shared master port
// ------------------------------------------------------------
output reg m_req,
output reg m_wr,
output reg [ADDR_WIDTH-1:0] m_addr,
output reg signed [7:0] m_wdata,
input wire signed [7:0] m_rdata,
input wire m_ready
);
localparam SEL_NONE = 3'd0;
localparam SEL_A = 3'd1;
localparam SEL_B = 3'd2;
localparam SEL_C = 3'd3;
localparam SEL_D = 3'd4;
reg [2:0] owner;
always @(posedge clk) begin
if (rst) begin
owner <= SEL_NONE;
m_req <= 1'b0;
m_wr <= 1'b0;
m_addr <= {ADDR_WIDTH{1'b0}};
m_wdata <= 8'sd0;
a_rdata <= 8'sd0;
a_ready <= 1'b0;
b_rdata <= 8'sd0;
b_ready <= 1'b0;
c_rdata <= 8'sd0;
c_ready <= 1'b0;
d_rdata <= 8'sd0;
d_ready <= 1'b0;
end else begin
m_req <= 1'b0;
a_ready <= 1'b0;
b_ready <= 1'b0;
c_ready <= 1'b0;
d_ready <= 1'b0;
case (owner)
SEL_NONE: begin
if (b_req) begin
owner <= SEL_B;
m_req <= 1'b1;
m_wr <= b_wr;
m_addr <= b_addr;
m_wdata <= b_wdata;
end else if (c_req) begin
owner <= SEL_C;
m_req <= 1'b1;
m_wr <= c_wr;
m_addr <= c_addr;
m_wdata <= c_wdata;
end else if (a_req) begin
owner <= SEL_A;
m_req <= 1'b1;
m_wr <= a_wr;
m_addr <= a_addr;
m_wdata <= a_wdata;
end else if (d_req) begin
owner <= SEL_D;
m_req <= 1'b1;
m_wr <= d_wr;
m_addr <= d_addr;
m_wdata <= d_wdata;
end
end
SEL_A: begin
if (m_ready) begin
a_rdata <= m_rdata;
a_ready <= 1'b1;
owner <= SEL_NONE;
end
end
SEL_B: begin
if (m_ready) begin
b_rdata <= m_rdata;
b_ready <= 1'b1;
owner <= SEL_NONE;
end
end
SEL_C: begin
if (m_ready) begin
c_rdata <= m_rdata;
c_ready <= 1'b1;
owner <= SEL_NONE;
end
end
SEL_D: begin
if (m_ready) begin
d_rdata <= m_rdata;
d_ready <= 1'b1;
owner <= SEL_NONE;
end
end
default: begin
owner <= SEL_NONE;
end
endcase
end
end
endmodule