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
170 lines
8.0 KiB
Verilog
170 lines
8.0 KiB
Verilog
`timescale 1ns/1ps
|
|
|
|
// ================================================================
|
|
// C.4 certification: rtl/mem_arbiter.v priority (B>C>A>D), grant
|
|
// correctness, and starvation behavior.
|
|
//
|
|
// Oracle: the priority order is a design DECISION stated in the
|
|
// module's own header (not derived from behavior) -- these tests
|
|
// check the RTL actually implements that stated order, and probe the
|
|
// starvation claim's exact wording ("D simply gets stretched out,
|
|
// never starves OR CORRUPTS A/B/C" -- a claim about protecting A/B/C,
|
|
// NOT a claim that D itself can never starve under sustained
|
|
// contention. This test checks both readings explicitly instead of
|
|
// assuming one).
|
|
//
|
|
// Testbench note: request signals are driven with NON-BLOCKING
|
|
// assignments throughout. An earlier version used blocking
|
|
// assignments to clear a request in the same `@(posedge clk)` step
|
|
// that was meant to grant it -- a real race with the DUT's own
|
|
// synchronous always block sampling the same edge (whichever process
|
|
// happens to run first in that Active-region step wins; Icarus
|
|
// resolved it in the DUT's disfavor here, silently losing every
|
|
// grant, `owner` never leaving SEL_NONE, every `wait(x_ready)`
|
|
// blocking forever). Caught via a hierarchical trace of `dut.owner`
|
|
// showing it never changed from 0 despite requests being driven --
|
|
// not an RTL defect, a testbench race, same class as the one found in
|
|
// C.1/C.2 (see docs/validation/04-arbiter.md).
|
|
// ================================================================
|
|
|
|
module tb;
|
|
|
|
localparam ADDR_WIDTH = 23;
|
|
|
|
reg clk, rst;
|
|
reg a_req, b_req, c_req, d_req;
|
|
reg a_wr, b_wr, c_wr, d_wr;
|
|
reg [ADDR_WIDTH-1:0] a_addr, b_addr, c_addr, d_addr;
|
|
reg signed [7:0] a_wdata, b_wdata, c_wdata, d_wdata;
|
|
wire signed [7:0] a_rdata, b_rdata, c_rdata, d_rdata;
|
|
wire a_ready, b_ready, c_ready, d_ready;
|
|
|
|
wire m_req, m_wr;
|
|
wire [ADDR_WIDTH-1:0] m_addr;
|
|
wire signed [7:0] m_wdata;
|
|
reg signed [7:0] m_rdata;
|
|
reg m_ready;
|
|
|
|
mem_arbiter #(.ADDR_WIDTH(ADDR_WIDTH)) dut (
|
|
.clk(clk), .rst(rst),
|
|
.a_req(a_req), .a_wr(a_wr), .a_addr(a_addr), .a_wdata(a_wdata), .a_rdata(a_rdata), .a_ready(a_ready),
|
|
.b_req(b_req), .b_wr(b_wr), .b_addr(b_addr), .b_wdata(b_wdata), .b_rdata(b_rdata), .b_ready(b_ready),
|
|
.c_req(c_req), .c_wr(c_wr), .c_addr(c_addr), .c_wdata(c_wdata), .c_rdata(c_rdata), .c_ready(c_ready),
|
|
.d_req(d_req), .d_wr(d_wr), .d_addr(d_addr), .d_wdata(d_wdata), .d_rdata(d_rdata), .d_ready(d_ready),
|
|
.m_req(m_req), .m_wr(m_wr), .m_addr(m_addr), .m_wdata(m_wdata), .m_rdata(m_rdata), .m_ready(m_ready)
|
|
);
|
|
|
|
initial begin clk = 0; forever #5 clk = ~clk; end
|
|
|
|
initial begin
|
|
#200000;
|
|
$display("WATCHDOG TIMEOUT -- simulation did not finish in time");
|
|
$finish;
|
|
end
|
|
|
|
// Single-cycle-latency downstream memory stub: grants m_ready one
|
|
// cycle after m_req, echoes m_addr as the "data" (distinguishable
|
|
// per-port response, used to confirm rdata routes to the RIGHT
|
|
// requester and not a sibling).
|
|
always @(posedge clk) begin
|
|
m_ready <= m_req;
|
|
m_rdata <= m_addr[7:0];
|
|
end
|
|
|
|
integer errors;
|
|
|
|
initial begin
|
|
errors = 0;
|
|
rst <= 1;
|
|
a_req <= 0; b_req <= 0; c_req <= 0; d_req <= 0;
|
|
a_wr <= 0; b_wr <= 0; c_wr <= 0; d_wr <= 0;
|
|
a_wdata <= 0; b_wdata <= 0; c_wdata <= 0; d_wdata <= 0;
|
|
a_addr <= 23'h001; b_addr <= 23'h002; c_addr <= 23'h003; d_addr <= 23'h004;
|
|
repeat(3) @(posedge clk);
|
|
rst <= 0;
|
|
@(posedge clk);
|
|
|
|
// ------------------------------------------------------------
|
|
// TEST 1: all four request simultaneously -- B must win first.
|
|
// ------------------------------------------------------------
|
|
$display("--- TEST 1: simultaneous A+B+C+D request -- B must win ---");
|
|
a_req <= 1; b_req <= 1; c_req <= 1; d_req <= 1;
|
|
@(posedge clk); // this edge: DUT samples all 4 (still their pre-edge values), grants B
|
|
a_req <= 0; b_req <= 0; c_req <= 0; d_req <= 0; // withdraw next cycle (one-cycle-pulse contract)
|
|
wait(b_ready);
|
|
if (b_rdata !== b_addr[7:0]) begin errors=errors+1; $display("FAIL: b_rdata=%0d expected=%0d", b_rdata, b_addr[7:0]); end
|
|
if (a_ready || c_ready || d_ready) begin errors=errors+1; $display("FAIL: a/c/d_ready asserted when B should have been the sole grantee"); end
|
|
else $display("PASS: B granted alone, correct data routed back");
|
|
@(posedge clk);
|
|
|
|
// ------------------------------------------------------------
|
|
// TEST 2: A+C+D request (no B) -- C must win.
|
|
// ------------------------------------------------------------
|
|
$display("--- TEST 2: A+C+D request (no B) -- C must win ---");
|
|
a_req <= 1; c_req <= 1; d_req <= 1;
|
|
@(posedge clk);
|
|
a_req <= 0; c_req <= 0; d_req <= 0;
|
|
wait(c_ready);
|
|
if (c_rdata !== c_addr[7:0]) begin errors=errors+1; $display("FAIL: c_rdata=%0d expected=%0d", c_rdata, c_addr[7:0]); end
|
|
else $display("PASS: C granted (B absent), correct data");
|
|
@(posedge clk);
|
|
|
|
// ------------------------------------------------------------
|
|
// TEST 3: A+D request (no B, no C) -- A must win.
|
|
// ------------------------------------------------------------
|
|
$display("--- TEST 3: A+D request (no B, no C) -- A must win ---");
|
|
a_req <= 1; d_req <= 1;
|
|
@(posedge clk);
|
|
a_req <= 0; d_req <= 0;
|
|
wait(a_ready);
|
|
if (a_rdata !== a_addr[7:0]) begin errors=errors+1; $display("FAIL: a_rdata=%0d expected=%0d", a_rdata, a_addr[7:0]); end
|
|
else $display("PASS: A granted (B,C absent), correct data");
|
|
@(posedge clk);
|
|
|
|
// ------------------------------------------------------------
|
|
// TEST 4: D alone -- must be granted (lowest priority does not
|
|
// mean "never granted").
|
|
// ------------------------------------------------------------
|
|
$display("--- TEST 4: D alone -- must still be granted ---");
|
|
d_req <= 1;
|
|
@(posedge clk);
|
|
d_req <= 0;
|
|
wait(d_ready);
|
|
if (d_rdata !== d_addr[7:0]) begin errors=errors+1; $display("FAIL: d_rdata=%0d expected=%0d", d_rdata, d_addr[7:0]); end
|
|
else $display("PASS: D granted alone, correct data");
|
|
@(posedge clk);
|
|
|
|
// ------------------------------------------------------------
|
|
// TEST 5: sustained back-to-back B requests (held continuously)
|
|
// vs. continuous D requests -- does D ever get a turn? This
|
|
// checks the exact wording of the header's starvation claim,
|
|
// not an assumption.
|
|
// ------------------------------------------------------------
|
|
$display("--- TEST 5: continuous B contention vs continuous D -- does D starve? ---");
|
|
begin : starvation_check
|
|
integer cyc;
|
|
reg d_ever_granted;
|
|
d_ever_granted = 0;
|
|
d_req <= 1;
|
|
b_req <= 1;
|
|
for (cyc = 0; cyc < 500; cyc = cyc + 1) begin
|
|
@(posedge clk);
|
|
if (d_ready) d_ever_granted = 1;
|
|
end
|
|
if (d_ever_granted)
|
|
$display("OBSERVED: D was eventually granted within %0d cycles despite continuous B contention -- D does not starve under this exact pattern", cyc);
|
|
else
|
|
$display("OBSERVED: D was NEVER granted in %0d cycles of continuous B contention -- D CAN starve indefinitely under sustained higher-priority load (a literal 'gets stretched out' reading; does not contradict the header if read as only promising A/B/C's protection, not D's own)", cyc);
|
|
end
|
|
d_req <= 0; b_req <= 0;
|
|
@(posedge clk);
|
|
|
|
if (errors == 0)
|
|
$display("ALL TESTS PASSED (priority order B>C>A>D confirmed; TEST 5 is an observation, not a pass/fail claim about starvation, since the header's wording is ambiguous about D's own guarantee)");
|
|
else
|
|
$display("FAILED: %0d error(s)", errors);
|
|
$finish;
|
|
end
|
|
|
|
endmodule
|