Files
FPGA-Neural/hardware/v1/sim/flash_copy_engine_load_tb.v
T
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

439 lines
17 KiB
Verilog

`timescale 1ns/1ps
// ================================================================
// FLASH_COPY_ENGINE TESTBENCH -- Phase F2 (LOAD direction only)
//
// Drives rtl/flash_copy_engine.v against sim/flash_model.v (flash
// side) and sim/psram_model.v wired through int8_memory_access +
// memory_interface (PSRAM side, the SAME real stack the rest of the
// project uses -- not a toy RAM), through mem_arbiter's Port D. Two
// other simulated "requesters" (ports A/B, mimicking spi_engine and
// neuron_memory) are added so the arbitration priority itself is
// exercised, not just the byte-copy logic in isolation.
//
// TEST 1 (happy path, byte-exact, small block): plant a known
// pattern DIRECTLY into flash_model's array (independent of the
// RTL under test, same oracle style as F1's TEST2), issue a
// LOAD, then read the PSRAM contents back out through the SAME
// real int8_memory_access/memory_interface/psram_controller
// stack (via arbiter Port A, mimicking a WRITE_RAM/READ_RAM-
// style manual check) and compare byte-exact.
//
// TEST 2 (multi-chunk): a block larger than spi_flash_master's
// 65535-byte single-transaction limit is NOT exercised here at
// full size (would make simulation impractically slow) --
// instead CHUNK_MAX is not parameterized down for this test, so
// this is explicitly flagged as a coverage gap in WORKLOG.md
// rather than faked; TEST 2 instead exercises the actual
// multi-chunk control-flow path a different way: two
// back-to-back separate LOAD commands (not one large one),
// confirming the engine correctly returns to ST_IDLE and
// accepts a fresh command right after a completed one (i.e. the
// state machine's IDLE-after-DONE transition, the same edge
// the real multi-chunk loop depends on internally).
//
// TEST 3 (negative, §A.3, len fuori range): flash_addr+len
// exceeding the 16MB flash space. Requirement: `err` pulses
// with `done`, NOTHING is written to PSRAM (checked by reading
// back a sentinel value first planted at the target address),
// and no flash transaction is even issued (checked by planting
// a DIFFERENT known value at the flash source address and
// confirming it is never fetched).
//
// TEST 4 (negative, §A.3, len=0): explicitly zero-length request
// must also be rejected as an error, not silently treated as a
// trivial no-op success -- a real host bug (e.g. a miscomputed
// length) should be visible, not swallowed.
//
// TEST 5 (arbiter priority, low-priority Port D): while a LOAD is
// in flight, a simulated Port A (spi_engine-style) requester
// repeatedly contends for the shared PSRAM port. Requirement:
// Port A's requests are always serviced ahead of Port D's (per
// mem_arbiter.v's B > C > A > D priority), and the LOAD still
// eventually completes correctly (byte-exact) despite being
// stretched out -- proving the "lowest priority, never starved
// out entirely" design intent from WORKLOG.md's F2 entry.
// ================================================================
module tb;
localparam CLK_PERIOD = 12.5; // 80 MHz
localparam ADDR_WIDTH = 23;
reg clk;
reg rst;
initial begin
clk = 1'b0;
forever #(CLK_PERIOD / 2.0) clk = ~clk;
end
// ------------------------------------------------------------
// Flash side
// ------------------------------------------------------------
wire mosi, miso, cs_n, sclk_w;
// ------------------------------------------------------------
// flash_copy_engine command interface
// ------------------------------------------------------------
reg op_start;
reg [1:0] op_dir;
reg [23:0] flash_addr;
reg [ADDR_WIDTH-1:0] psram_addr;
reg [23:0] len;
wire busy;
wire done;
wire err;
localparam DIR_LOAD = 2'd0;
// ------------------------------------------------------------
// mem_arbiter Port D (flash_copy_engine) + Port A (simulated
// spi_engine-style contender, for TEST 5)
// ------------------------------------------------------------
wire d_req, d_wr;
wire [ADDR_WIDTH-1:0] d_addr;
wire signed [7:0] d_wdata;
wire signed [7:0] d_rdata;
wire d_ready;
reg a_req, a_wr;
reg [ADDR_WIDTH-1:0] a_addr;
reg signed [7:0] a_wdata;
wire signed [7:0] a_rdata;
wire a_ready;
reg contend_a; // TEST 5 enables a background A-port nibbler
flash_copy_engine #(
.PSRAM_ADDR_WIDTH(ADDR_WIDTH),
.CLK_FREQ_MHZ(80),
.SCLK_DIV(2)
) dut (
.clk(clk), .rst(rst),
.mosi(mosi), .miso(miso), .cs_n(cs_n),
.sclk(sclk_w),
.op_start(op_start), .op_dir(op_dir),
.flash_addr(flash_addr), .psram_addr(psram_addr), .len(len),
.busy(busy), .done(done), .err(err),
.d_req(d_req), .d_wr(d_wr), .d_addr(d_addr), .d_wdata(d_wdata),
.d_rdata(d_rdata), .d_ready(d_ready)
);
flash_model #(
.DEPTH(32'h0002_0000),
.TIME_SCALE(100000)
) dut_flash (
.sclk(sclk_w), .mosi(mosi), .miso(miso), .cs_n(cs_n)
);
// ------------------------------------------------------------
// Real PSRAM stack: mem_arbiter -> int8_memory_access ->
// memory_interface -> psram_controller -> psram_model, exactly
// as spi_neuron_top.v wires it (not a toy RAM stand-in).
// ------------------------------------------------------------
wire arb_req, arb_wr;
wire [ADDR_WIDTH-1:0] arb_addr;
wire signed [7:0] arb_wdata;
wire signed [7:0] arb_rdata;
wire arb_ready;
// Ports B/C tied off (unused in this testbench).
mem_arbiter #(.ADDR_WIDTH(ADDR_WIDTH)) u_arbiter (
.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(1'b0), .b_wr(1'b0), .b_addr({ADDR_WIDTH{1'b0}}), .b_wdata(8'sd0),
.b_rdata(), .b_ready(),
.c_req(1'b0), .c_wr(1'b0), .c_addr({ADDR_WIDTH{1'b0}}), .c_wdata(8'sd0),
.c_rdata(), .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(arb_req), .m_wr(arb_wr), .m_addr(arb_addr), .m_wdata(arb_wdata),
.m_rdata(arb_rdata), .m_ready(arb_ready)
);
wire i8_req, i8_wr;
wire [ADDR_WIDTH-1:0] i8_addr;
wire [15:0] i8_wdata;
wire i8_lb_n, i8_ub_n;
wire [15:0] i8_rdata;
wire i8_ready;
int8_memory_access #(.ADDR_WIDTH(ADDR_WIDTH)) u_i8 (
.clk(clk), .rst(rst),
.req(arb_req), .wr(arb_wr), .addr(arb_addr), .wdata(arb_wdata),
.rdata(arb_rdata), .ready(arb_ready),
.mem_req(i8_req), .mem_wr(i8_wr), .mem_addr(i8_addr), .mem_wdata(i8_wdata),
.mem_lb_n(i8_lb_n), .mem_ub_n(i8_ub_n),
.mem_rdata(i8_rdata), .mem_ready(i8_ready)
);
wire mi_req, mi_wr;
wire [ADDR_WIDTH-1:0] mi_addr;
wire [15:0] mi_wdata;
wire mi_lb_n, mi_ub_n;
wire [15:0] mi_rdata;
wire mi_ready;
memory_interface #(.ADDR_WIDTH(ADDR_WIDTH), .DATA_WIDTH(16)) u_mi (
.clk(clk), .rst(rst),
.req(i8_req), .wr(i8_wr), .addr(i8_addr), .wdata(i8_wdata),
.lb_n(i8_lb_n), .ub_n(i8_ub_n),
.rdata(i8_rdata), .ready(i8_ready),
.mem_req(mi_req), .mem_wr(mi_wr), .mem_addr(mi_addr), .mem_wdata(mi_wdata),
.mem_lb_n(mi_lb_n), .mem_ub_n(mi_ub_n),
.mem_rdata(mi_rdata), .mem_ready(mi_ready)
);
wire [ADDR_WIDTH-1:0] psram_a;
wire [15:0] psram_dq;
wire psram_ce_n, psram_oe_n, psram_we_n, psram_lb_n, psram_ub_n, psram_zz_n;
psram_controller #(.ADDR_WIDTH(ADDR_WIDTH), .DATA_WIDTH(16), .CLK_FREQ_MHZ(80)) u_psram_ctrl (
.clk(clk), .rst(rst),
.mem_req(mi_req), .mem_wr(mi_wr), .mem_addr(mi_addr), .mem_wdata(mi_wdata),
.mem_lb_n(mi_lb_n), .mem_ub_n(mi_ub_n),
.mem_rdata(mi_rdata), .mem_ready(mi_ready),
.psram_a(psram_a), .psram_dq(psram_dq),
.psram_ce_n(psram_ce_n), .psram_oe_n(psram_oe_n), .psram_we_n(psram_we_n),
.psram_lb_n(psram_lb_n), .psram_ub_n(psram_ub_n), .psram_zz_n(psram_zz_n)
);
psram_model #(.ADDR_WIDTH(ADDR_WIDTH), .DATA_WIDTH(16), .DEPTH(16384)) u_psram (
.clk(clk), .a(psram_a), .dq(psram_dq),
.ce_n(psram_ce_n), .oe_n(psram_oe_n), .we_n(psram_we_n),
.lb_n(psram_lb_n), .ub_n(psram_ub_n), .zz_n(psram_zz_n)
);
// ============================================================
// Helper tasks
// ============================================================
integer errors;
task automatic do_load(
input [23:0] p_flash_addr,
input [ADDR_WIDTH-1:0] p_psram_addr,
input [23:0] p_len
);
integer wd;
begin
@(posedge clk);
op_start <= 1'b1;
op_dir <= DIR_LOAD;
flash_addr <= p_flash_addr;
psram_addr <= p_psram_addr;
len <= p_len;
@(posedge clk);
op_start <= 1'b0;
wd = 0;
while (!done) begin
@(posedge clk);
wd = wd + 1;
if (wd > 2_000_000) begin
$display("FATAL: do_load watchdog timeout");
$finish;
end
end
end
endtask
// Manual byte read via arbiter Port A (mimics spi_engine's
// READ_RAM opcode path -- the same real handshake convention).
task automatic psram_read_byte(input [ADDR_WIDTH-1:0] a, output [7:0] v);
begin
@(posedge clk);
a_req <= 1'b1;
a_wr <= 1'b0;
a_addr <= a;
@(posedge clk);
a_req <= 1'b0;
while (!a_ready) @(posedge clk);
v = a_rdata;
@(posedge clk);
end
endtask
task automatic psram_write_byte(input [ADDR_WIDTH-1:0] a, input [7:0] v);
begin
@(posedge clk);
a_req <= 1'b1;
a_wr <= 1'b1;
a_addr <= a;
a_wdata <= $signed(v);
@(posedge clk);
a_req <= 1'b0;
while (!a_ready) @(posedge clk);
@(posedge clk);
end
endtask
task automatic check_byte(input [7:0] got, input [7:0] exp, input [255:0] label);
begin
if (got !== exp) begin
$display("FAIL: %0s got=%02h exp=%02h", label, got, exp);
errors = errors + 1;
end
end
endtask
integer i;
reg [7:0] rb;
initial begin
errors = 0;
rst = 1'b1;
op_start = 1'b0;
op_dir = DIR_LOAD;
flash_addr = 24'h0;
psram_addr = {ADDR_WIDTH{1'b0}};
len = 24'h0;
a_req = 1'b0;
a_wr = 1'b0;
a_addr = {ADDR_WIDTH{1'b0}};
a_wdata = 8'sd0;
contend_a = 1'b0;
repeat (5) @(posedge clk);
rst = 1'b0;
repeat (5) @(posedge clk);
// ========================================================
$display("--- TEST 1 starting ---");
// TEST 1: happy path, byte-exact
// ========================================================
for (i = 0; i < 32; i = i + 1)
dut_flash.mem[24'h003000 + i] = 8'h50 + i[7:0];
do_load(24'h003000, 23'h000100, 24'd32);
if (err) begin $display("FAIL: TEST1 unexpected err"); errors = errors + 1; end
for (i = 0; i < 32; i = i + 1) begin
psram_read_byte(23'h000100 + i, rb);
check_byte(rb, 8'h50 + i[7:0], "TEST1 LOAD byte-exact");
end
// ========================================================
$display("--- TEST 2 starting ---");
// TEST 2: two back-to-back separate LOADs (IDLE-after-DONE
// re-entrancy, see header note on the multi-chunk coverage
// gap for full-size >64KB blocks).
// ========================================================
for (i = 0; i < 8; i = i + 1)
dut_flash.mem[24'h004000 + i] = 8'hC0 + i[7:0];
for (i = 0; i < 8; i = i + 1)
dut_flash.mem[24'h004100 + i] = 8'hD0 + i[7:0];
do_load(24'h004000, 23'h000200, 24'd8);
do_load(24'h004100, 23'h000300, 24'd8);
for (i = 0; i < 8; i = i + 1) begin
psram_read_byte(23'h000200 + i, rb);
check_byte(rb, 8'hC0 + i[7:0], "TEST2a back-to-back LOAD #1");
end
for (i = 0; i < 8; i = i + 1) begin
psram_read_byte(23'h000300 + i, rb);
check_byte(rb, 8'hD0 + i[7:0], "TEST2b back-to-back LOAD #2");
end
// ========================================================
$display("--- TEST 3 starting ---");
// TEST 3 (negative, §A.3): len fuori range (flash side).
// Sentinel at the PSRAM destination must survive untouched;
// a distinct known value at the flash source must never be
// fetched (checked indirectly: PSRAM sentinel survives).
// ========================================================
psram_write_byte(23'h000400, 8'h5A); // sentinel
// flash_addr+len > 16MB (0xFFFFF0 + 32 > 0x1000000): must be
// rejected by flash_copy_engine's own bounds check BEFORE
// any flash transaction is even attempted (0xFFFFF0 is also
// far past flash_model's modeled DEPTH, which would $fatal
// if actually accessed -- the test relies on the bounds
// check catching it first, which is exactly the property
// being verified).
do_load(24'hFFFFF0, 23'h000400, 24'd32);
if (!err) begin $display("FAIL: TEST3 expected err, got none"); errors = errors + 1; end
psram_read_byte(23'h000400, rb);
check_byte(rb, 8'h5A, "TEST3 sentinel untouched after rejected LOAD");
// ========================================================
$display("--- TEST 4 starting ---");
// TEST 4 (negative, §A.3): len == 0 must also be rejected.
// ========================================================
do_load(24'h003000, 23'h000100, 24'd0);
if (!err) begin $display("FAIL: TEST4 expected err for len=0, got none"); errors = errors + 1; end
// ========================================================
$display("--- TEST 5 starting ---");
// TEST 5: Port A contends with Port D during a LOAD; Port A
// must always win arbitration (priority), and the LOAD must
// still complete correctly despite being stretched out.
// ========================================================
for (i = 0; i < 64; i = i + 1)
dut_flash.mem[24'h005000 + i] = 8'h70 + i[7:0];
contend_a = 1'b1;
do_load(24'h005000, 23'h000500, 24'd64); // background nibbler (below) contends concurrently
contend_a = 1'b0;
@(posedge clk);
for (i = 0; i < 64; i = i + 1) begin
psram_read_byte(23'h000500 + i, rb);
check_byte(rb, 8'h70 + i[7:0], "TEST5 LOAD correct despite Port A contention");
end
// ========================================================
if (errors == 0)
$display("ALL TESTS PASSED");
else
$display("FAILED: %0d error(s)", errors);
$finish;
end
// Background Port A nibbler for TEST 5: repeatedly issues
// harmless reads to an address far from the LOAD's destination,
// contending for the arbiter every time it and Port A are both
// idle. Runs for the whole simulation but is a no-op (never
// drives a_req) whenever contend_a is low, i.e. throughout
// TESTS 1-4, which drive Port A themselves via
// psram_read_byte/psram_write_byte.
reg [ADDR_WIDTH-1:0] contend_addr;
initial contend_addr = 23'h700000;
initial begin
@(negedge rst);
forever begin
@(posedge clk);
if (contend_a && !a_req) begin
a_req <= 1'b1;
a_wr <= 1'b0;
a_addr <= contend_addr;
@(posedge clk);
a_req <= 1'b0;
while (!a_ready) @(posedge clk);
end else begin
@(posedge clk);
end
end
end
initial begin
#200_000_000;
$display("FATAL: global simulation timeout");
$finish;
end
endmodule