michele/FPGA-Neural
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

feat: integrate flash #1 (neural-network data) RTL into V2 top-level

Browse Source 
Closes the flash #1 RTL gap flagged in DEC-0041: real, unmodified V1
subsystem (flash_slot_manager.v/flash_copy_engine.v/spi_flash_master.v/
crc32.v) now instantiated in fpga_neural_v2_top.v, bridged to the AR
memory bus via a new flash_mem_adapter.v (byte<->word, matches
nms_memory_manager_stream_wide.v's own real masking convention), and
commandable over SPI via a new spi_host_bridge.v opcode (OP_FLASH_CMD,
0x30) using the same byte-counting idiom as OP_WRITE_JOB. Real balls
now in the LPF: flash_sclk=B2, flash_mosi=E2, flash_miso=F2,
flash_cs_n=F3.

New tb_flash_integration_smoke.v: real SPI-triggered OP_FLASH_READ_BLOCK
verified bit-exact (64/64 bytes) against a real V1 flash_model.v
instance, through the new adapter and the widened (2->3 port) host-arb
arbiter; WRITE_JOB regression confirms the new 3rd port doesn't disturb
existing traffic. Full existing regression re-run clean: D-Stress N=4/
N=8 (bit-exact + data_ready PASS), board-level smoke test (11/11),
isolated spi_host_bridge test (18/18).

Honest, disclosed finding: a full 8-seed P&R re-verification shows
N_SLOTS=4 @ 64MHz regressed from 8/8 to 3/8 PASS (worst 60.18MHz).
Root cause traced via the real critical-path report: the SAME
pre-existing arbiter-to-sdram-backend bottleneck already documented all
session, made worse by flash's added die-area placement pressure --
not a new path through the flash logic itself. N_SLOTS=8 essentially
unchanged (6/8, was 5/8). See decisions.log DEC-0042 for full detail
and open decision points.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013xXuuRUWZScuo1DeYJxs3v
This commit is contained in:
michele
2026-09-07 18:31:50 +02:00
co-authored by Claude Sonnet 5
parent b4f33388a9
commit 59901a4905
6 changed files with 605 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hardware/v2/constraints/v2_board_top.lpf
+9
View File
@@ -106,6 +106,15 @@ LOCATE COMP "sdram_dqm[1]" SITE "N3"; IOBUF PORT "sdram_dqm[1]" IO_TYPE=LVCMOS33
// COMP in this file) via the Trellis iodb.json + official CSV.
LOCATE COMP "data_ready" SITE "G3"; IOBUF PORT "data_ready" IO_TYPE=LVCMOS33;
// Flash #1 (neural-network data), ordinary GPIO bank 7 -- V1's own
// real balls (E3/D3/D5/E4) are NOT reusable here, already occupied by
// V2's SDRAM bus (see decisions.log DEC-0041); these are new, free
// balls confirmed via Trellis iodb.json.
LOCATE COMP "flash_sclk" SITE "B2"; IOBUF PORT "flash_sclk" IO_TYPE=LVCMOS33;
LOCATE COMP "flash_mosi" SITE "E2"; IOBUF PORT "flash_mosi" IO_TYPE=LVCMOS33;
LOCATE COMP "flash_miso" SITE "F2"; IOBUF PORT "flash_miso" IO_TYPE=LVCMOS33;
LOCATE COMP "flash_cs_n" SITE "F3"; IOBUF PORT "flash_cs_n" IO_TYPE=LVCMOS33;
// All 17 top-level ports of fpga_neural_v2_top are now real-ball
// assigned: osc_clk, ext_rst_n, spi_sclk, spi_mosi, spi_miso,
// spi_cs_n, sdram_cke, sdram_cs_n, sdram_ras_n, sdram_cas_n,
hardware/v2/logs/decisions.log
+98
View File
@@ -2402,3 +2402,101 @@ STATUS: two-flash architecture and FPGA_DATA_READY CLOSED and
verified (bit-exact + real placement). Flash #1's RTL port
(`flash_copy_engine.v` integration into V2's top-level) remains a
real, separate, OPEN task -- ball positions reserved, not wired.
DEC-0042 -- Flash #1 RTL integration completed; real timing regression
found and disclosed (N_SLOTS=4 @ 64MHz 8/8 -> 3/8 PASS)
DATE: 2026-09-07
CONTEXT: closing DEC-0041's own remaining open item -- flash #1
(neural-network weights/graph data) had real ball reservations
(B2/E2/F2/F3) but no RTL port. User asked for this to be completed,
then a meticulous full-stack verification (RTL, tests, timing).
INTEGRATION (real, functional, verified):
- Instantiated the real, unmodified V1 subsystem
(`flash_slot_manager.v`, which owns `flash_copy_engine.v`, which
owns `spi_flash_master.v`; `crc32.v` used internally) directly in
`fpga_neural_v2_top.v` -- zero modifications to any of these four
files, matching this project's own established "reuse, don't
re-derive" precedent for verified V1 modules.
- New file `hardware/v2/rtl/flash_mem_adapter.v`: bridges flash_slot_
manager's real "Port D" (PSRAM-style byte interface: d_req/d_wr/
d_addr/d_wdata/d_rdata/d_ready, 8-bit signed) to the AR-port
convention (word address, 16-bit data, lb_n/ub_n byte-lane masking)
used by `slot_mem_arbiter.v` -- byte<->word convention matches
`nms_memory_manager_stream_wide.v`'s own real, already-verified
result-writeback logic exactly (not invented): word_addr =
byte_addr[ADDR_WIDTH-1:1], byte_addr[0] selects lower/upper lane,
write data replicated to both halves with the mask picking which
the controller actually writes.
- `slot_mem_arbiter.v`'s host-arb instance (the "AR level 2" already
combining compute-side traffic and the SPI host's raw memory port)
widened from N_PORTS=2 to N_PORTS=3, flash at the new highest index
(confirmed lowest-priority in this arbiter's real "lowest-index-
wins" scheme -- matches V1's own "Port D, lowest priority"
convention for this exact traffic class).
- `spi_host_bridge.v`: new real opcode `OP_FLASH_CMD` (0x30), 19
payload bytes (op_code/slot_id/new_offset/new_length/new_type/
ext_psram_addr/ext_length/raw_flash_addr), built with the same
byte-counting state-machine idiom already used by OP_WRITE_JOB
(including the same cs_fell mid-transaction protection). STATUS
byte (0x20) extended: bit3=flash_busy, bit4=flash_done (sticky),
bit5=flash_err (sticky), both cleared by the next OP_FLASH_CMD.
Catalog readback (cat_read_sel/cat_out_*) deliberately NOT wired to
SPI in this pass -- diagnostic-only, not required for the core
LOAD/SAVE/raw-block functionality; a real, disclosed, separate
follow-up.
- Real balls (already reserved in DEC-0041, now actually in the LPF):
flash_sclk=B2, flash_mosi=E2, flash_miso=F2, flash_cs_n=F3.
VERIFICATION (real, not assumed):
- New testbench `tb_flash_integration_smoke.v`: real SPI OP_FLASH_CMD
transaction (OP_FLASH_READ_BLOCK) drives flash_slot_manager through
the new adapter and the new 3rd arbiter port, into the real
sdram_unified_backend/sdram_controller/AS4C32M16SB-7BIN chain,
against a real V1 `flash_model.v` instance preloaded with a known
64-byte pattern -- PASS, 64/64 bytes bit-exact. A WRITE_JOB
regression check in the same file confirms the pre-existing job-
registration path still completes without hanging with the new 3rd
arbiter port present -- PASS.
- Full existing regression re-run, zero functional regressions:
D-Stress N_SLOTS=4 (49927 cycles) and N_SLOTS=8 (49909 cycles), both
256/256 bit-exact vs golden, both `data_ready` PASS; the real board-
level smoke test `tb_fpga_neural_v2_top_smoke.v` (11/11 PASS); the
isolated `tb_spi_host_bridge.v` (18/18 PASS, confirming the new
OP_FLASH_CMD opcode and widened STATUS byte did not disturb the
existing WRITE_JOB/WRITE_MEM/READ_MEM/STATUS behavior).
- Real synthesis (Yosys, 0 errors, 0 new warnings vs. the pre-flash
baseline -- confirmed by exact warning-count diff) and real
nextpnr-ecp5 placement (0 errors, all 4 new flash balls placed
correctly at their reserved sites).
HONEST TIMING FINDING (disclosed, not hidden): a full, real 8-seed
nextpnr-ecp5 P&R matrix with flash #1 now included shows a REAL
regression at the previously rock-solid N_SLOTS=4 @ 64MHz production
baseline:
N_SLOTS=4 @ 64MHz: 8/8 -> **3/8 PASS** (seeds 0,2,5 PASS; worst
60.18MHz seed3, best 67.34MHz seed2).
N_SLOTS=8 @ 64MHz: was 5/8, now 6/8 PASS (seeds 1,3,4,5,6,7 PASS;
worst 61.09MHz seed0, best 67.71MHz seed4) -- essentially
unchanged/slightly better, within seed-to-seed noise.
ROOT CAUSE (real, traced via the actual critical-path report, not
guessed): the worst N=4 seed's critical path is the SAME pre-existing
class already documented all session (`u_arbiter_wide.m_addr` ->
`u_sdram_backend.state`, 84% routing-dominated) -- NOT a new path
through the flash subsystem itself. Flash #1's real logic footprint
(flash_slot_manager + flash_copy_engine + spi_flash_master + crc32,
an entire secondary state-machine-heavy subsystem) measurably
increases die utilization and placement pressure, degrading this
SAME pre-existing bottleneck's achievable placement -- the identical
mechanism already observed when N_SLOTS scaled from 4 to 8 earlier
this session, now triggered by added area rather than added N_SLOTS.
STATUS: flash #1 integration is functionally CLOSED and verified
(bit-exact). N_SLOTS=4 @ 64MHz's own timing closure is now OPEN,
regressed by this integration -- explicitly NOT silently declared
still-closed. This is a real, disclosed trade-off requiring a
decision: accept and pursue a further optimization pass (same class
of fix as ERR-0027/ERR-0028/ERR-0029), constrain placement to isolate
flash logic from the sensitive region, or defer/gate flash #1 behind
a build-time option that keeps the pre-flash timing profile available.
Not decided in this pass -- flagged for the user.
hardware/v2/nms/rtl/fpga_neural_v2_top.v
+85 -15
View File
@@ -77,6 +77,14 @@ module fpga_neural_v2_top #(
// see nms_dataflow_core_sdram.v for the full design comment.
output wire data_ready,
// ---- Flash #1 (neural-network weights/graph data): real,
// ordinary GPIO SPI bus, physically separate from flash #2 (boot)
// -- see decisions.log for the two-flash architecture rationale.
output wire flash_sclk,
output wire flash_mosi,
input wire flash_miso,
output wire flash_cs_n,
output wire pll_locked
);
@@ -114,6 +122,13 @@ module fpga_neural_v2_top #(
wire [15:0] host_mem_wdata, host_mem_rdata;
wire host_mem_ready;
wire flash_op_start;
wire [2:0] flash_op_code;
wire [3:0] flash_slot_id;
wire [23:0] flash_new_offset, flash_new_length, flash_ext_length, flash_raw_flash_addr;
wire [7:0] flash_new_type;
wire [ADDR_WIDTH-1:0] flash_ext_addr;
spi_host_bridge #(
.ADDR_WIDTH(ADDR_WIDTH), .N_NODES(N_NODES), .MAX_DEPS(MAX_DEPS)
) u_spi_bridge (
@@ -126,7 +141,13 @@ module fpga_neural_v2_top #(
.mem_req(host_mem_req), .mem_wr(host_mem_wr), .mem_addr(host_mem_addr),
.mem_wdata(host_mem_wdata), .mem_lb_n(host_mem_lb_n), .mem_ub_n(host_mem_ub_n),
.mem_rdata(host_mem_rdata), .mem_ready(host_mem_ready),
.soft_rst_pulse(soft_rst_pulse)
.soft_rst_pulse(soft_rst_pulse),
.flash_op_start(flash_op_start), .flash_op_code(flash_op_code),
.flash_slot_id(flash_slot_id), .flash_new_offset(flash_new_offset),
.flash_new_length(flash_new_length), .flash_new_type(flash_new_type),
.flash_ext_addr(flash_ext_addr), .flash_ext_length(flash_ext_length),
.flash_raw_flash_addr(flash_raw_flash_addr),
.flash_busy(flash1_busy), .flash_done(flash1_done), .flash_err(flash1_err)
);
// ============================================================
@@ -183,24 +204,33 @@ module fpga_neural_v2_top #(
.m_rdata(arb_m_rdata), .m_ready(arb_m_ready)
);
// ---- AR level 2 (NEW, STEP20): compute-side AR stream (port0)
// vs. SPI host raw memory port (port1) -- reuses slot_mem_arbiter
// completely unchanged, just at N_PORTS=2, its own already-proven
// pending-latch discipline applying equally to a 2-port instance ----
wire [1:0] host_arb_s_req, host_arb_s_wr, host_arb_s_lb_n, host_arb_s_ub_n, host_arb_s_ready;
wire [ADDR_WIDTH*2-1:0] host_arb_s_addr;
wire [16*2-1:0] host_arb_s_wdata, host_arb_s_rdata;
// ---- AR level 2 (STEP20/STEP21): compute-side AR stream (port0,
// highest priority) vs. SPI host raw memory port (port1) vs.
// flash #1 (neural-network data) port (port2, lowest priority --
// matches V1's own "Port D, lowest priority" convention for this
// exact traffic class) -- reuses slot_mem_arbiter completely
// unchanged, just at N_PORTS=3, its own already-proven pending-
// latch discipline applying equally to a 3-port instance ----
wire flash_ar_req, flash_ar_wr, flash_ar_lb_n, flash_ar_ub_n, flash_ar_ready;
wire [ADDR_WIDTH-1:0] flash_ar_addr;
wire [15:0] flash_ar_wdata, flash_ar_rdata;
assign host_arb_s_req = {host_mem_req, arb_m_req};
assign host_arb_s_wr = {host_mem_wr, arb_m_wr};
assign host_arb_s_lb_n = {host_mem_lb_n, arb_m_lb_n};
assign host_arb_s_ub_n = {host_mem_ub_n, arb_m_ub_n};
assign host_arb_s_addr = {host_mem_addr, arb_m_addr};
assign host_arb_s_wdata = {host_mem_wdata, arb_m_wdata};
wire [2:0] host_arb_s_req, host_arb_s_wr, host_arb_s_lb_n, host_arb_s_ub_n, host_arb_s_ready;
wire [ADDR_WIDTH*3-1:0] host_arb_s_addr;
wire [16*3-1:0] host_arb_s_wdata, host_arb_s_rdata;
assign host_arb_s_req = {flash_ar_req, host_mem_req, arb_m_req};
assign host_arb_s_wr = {flash_ar_wr, host_mem_wr, arb_m_wr};
assign host_arb_s_lb_n = {flash_ar_lb_n, host_mem_lb_n, arb_m_lb_n};
assign host_arb_s_ub_n = {flash_ar_ub_n, host_mem_ub_n, arb_m_ub_n};
assign host_arb_s_addr = {flash_ar_addr, host_mem_addr, arb_m_addr};
assign host_arb_s_wdata = {flash_ar_wdata, host_mem_wdata, arb_m_wdata};
assign arb_m_ready = host_arb_s_ready[0];
assign arb_m_rdata = host_arb_s_rdata[15:0];
assign host_mem_ready = host_arb_s_ready[1];
assign host_mem_rdata = host_arb_s_rdata[31:16];
assign flash_ar_ready = host_arb_s_ready[2];
assign flash_ar_rdata = host_arb_s_rdata[47:32];
wire final_ar_req, final_ar_wr;
wire [ADDR_WIDTH-1:0] final_ar_addr;
@@ -210,7 +240,7 @@ module fpga_neural_v2_top #(
wire final_ar_ready;
slot_mem_arbiter #(
.ADDR_WIDTH(ADDR_WIDTH), .N_PORTS(2)
.ADDR_WIDTH(ADDR_WIDTH), .N_PORTS(3)
) u_host_arb (
.clk(clk), .rst(core_rst),
.s_req(host_arb_s_req), .s_wr(host_arb_s_wr), .s_addr(host_arb_s_addr),
@@ -221,6 +251,46 @@ module fpga_neural_v2_top #(
.m_rdata(final_ar_rdata), .m_ready(final_ar_ready)
);
// ---- Flash #1 (neural-network weights/graph data): real,
// unmodified V1 subsystem (flash_slot_manager.v, which owns
// flash_copy_engine.v, which owns spi_flash_master.v; crc32.v used
// internally too) -- see decisions.log for the integration design.
// Command interface driven by spi_host_bridge.v's own new
// OP_FLASH_CMD opcode. Data path bridged into the AR arbiter above
// via flash_mem_adapter.v (byte<->word, matches nms_memory_
// manager_stream_wide.v's own real masking convention). ----
wire flash_d_req, flash_d_wr, flash_d_ready;
wire [ADDR_WIDTH-1:0] flash_d_addr;
wire signed [7:0] flash_d_wdata, flash_d_rdata;
wire flash1_busy, flash1_done, flash1_err;
flash_slot_manager #(
.PSRAM_ADDR_WIDTH(ADDR_WIDTH), .CLK_FREQ_MHZ(64), .SCLK_DIV(2)
) u_flash1 (
.clk(clk), .rst(core_rst),
.mosi(flash_mosi), .miso(flash_miso), .cs_n(flash_cs_n), .sclk(flash_sclk),
.op_start(flash_op_start), .op_code(flash_op_code), .slot_id(flash_slot_id),
.new_offset(flash_new_offset), .new_length(flash_new_length), .new_type(flash_new_type),
.ext_psram_addr(flash_ext_addr), .ext_length(flash_ext_length),
.raw_flash_addr(flash_raw_flash_addr),
.busy(flash1_busy), .done(flash1_done), .err(flash1_err),
.cat_read_sel(4'd0), .cat_out_offset(), .cat_out_length(),
.cat_out_type(), .cat_out_valid(), .cat_out_crc(),
.d_req(flash_d_req), .d_wr(flash_d_wr), .d_addr(flash_d_addr),
.d_wdata(flash_d_wdata), .d_rdata(flash_d_rdata), .d_ready(flash_d_ready)
);
flash_mem_adapter #(
.ADDR_WIDTH(ADDR_WIDTH), .BYTE_ADDR_WIDTH(ADDR_WIDTH)
) u_flash_adapter (
.clk(clk), .rst(core_rst),
.d_req(flash_d_req), .d_wr(flash_d_wr), .d_addr(flash_d_addr),
.d_wdata(flash_d_wdata), .d_rdata(flash_d_rdata), .d_ready(flash_d_ready),
.s_req(flash_ar_req), .s_wr(flash_ar_wr), .s_addr(flash_ar_addr),
.s_wdata(flash_ar_wdata), .s_lb_n(flash_ar_lb_n), .s_ub_n(flash_ar_ub_n),
.s_rdata(flash_ar_rdata), .s_ready(flash_ar_ready)
);
// ---- W: weight fetch (unchanged) ----
wire [N_SLOTS-1:0] wide_s_wr = {N_SLOTS{1'b0}};
wire [64*N_SLOTS-1:0] wide_s_wdata = {(64*N_SLOTS){1'b0}};
hardware/v2/nms/sim/tb_flash_integration_smoke.v
+251
View File
@@ -0,0 +1,251 @@
`timescale 1ns/1ps
// ================================================================
// FPGA-Neural V2 -- Flash #1 (neural-network data) integration smoke
// test (2026-09-07)
//
// Proves the NEW wiring this session adds end-to-end: a real SPI
// OP_FLASH_CMD transaction (bit-banged, mode 0) drives
// spi_host_bridge.v's own new flash_* command ports, into the real,
// UNCHANGED V1 flash_slot_manager.v (which owns flash_copy_engine.v,
// which owns spi_flash_master.v -- all three reused byte-for-byte,
// zero modifications, per their own header comments), through the
// NEW flash_mem_adapter.v (byte<->word bridge), into the NEW third
// port of the host-arb slot_mem_arbiter (N_PORTS 2->3), down to the
// same real sdram_unified_backend/sdram_controller/AS4C32M16SB-7BIN
// chain already used by every other traffic class -- checked against
// a real, backdoor-peeked SDRAM result.
//
// Uses OP_FLASH_READ_BLOCK (op_code=4): the simplest real operation
// that exercises the full new path without needing catalog/slot setup
// (raw flash_addr -> ext_psram_addr, explicit length) -- flash_slot_
// manager.v's own real semantics, confirmed by inspection.
//
// flash_model.v (real, unmodified V1 SPI-flash simulation model,
// erase-state 0xFF, real Winbond command set) is preloaded with a
// known byte pattern at a known flash address; after the SPI-
// triggered read-block completes, the destination SDRAM region is
// backdoor-peeked and compared byte-for-byte against that same known
// pattern.
// ================================================================
`define SIM
module tb_flash_integration_smoke;
localparam ADDR_WIDTH = 26;
localparam N_SLOTS = 2;
localparam N_NODES = 16;
localparam MAX_DEPS = 4;
reg osc_clk = 0;
always #7.8125 osc_clk = ~osc_clk; // 64MHz, SIM PLL bypass
reg ext_rst_n = 0;
reg spi_sclk = 0, spi_mosi = 0, spi_cs_n = 1;
wire spi_miso;
wire sdram_cke, sdram_cs_n, sdram_ras_n, sdram_cas_n, sdram_we_n;
wire [1:0] sdram_ba;
wire [12:0] sdram_a;
wire [15:0] sdram_dq;
wire [1:0] sdram_dqm;
wire pll_locked, data_ready, sdram_clk;
wire flash_sclk, flash_mosi, flash_cs_n;
wire flash_miso;
fpga_neural_v2_top #(
.ADDR_WIDTH(ADDR_WIDTH), .N_SLOTS(N_SLOTS), .N_NODES(N_NODES), .MAX_DEPS(MAX_DEPS),
.CLK_FREQ_MHZ(64)
) dut (
.osc_clk(osc_clk), .ext_rst_n(ext_rst_n),
.spi_sclk(spi_sclk), .spi_mosi(spi_mosi), .spi_miso(spi_miso), .spi_cs_n(spi_cs_n),
.sdram_clk(sdram_clk),
.sdram_cke(sdram_cke), .sdram_cs_n(sdram_cs_n), .sdram_ras_n(sdram_ras_n),
.sdram_cas_n(sdram_cas_n), .sdram_we_n(sdram_we_n),
.sdram_ba(sdram_ba), .sdram_a(sdram_a), .sdram_dq(sdram_dq), .sdram_dqm(sdram_dqm),
.data_ready(data_ready),
.flash_sclk(flash_sclk), .flash_mosi(flash_mosi), .flash_miso(flash_miso), .flash_cs_n(flash_cs_n),
.pll_locked(pll_locked)
);
sdram_model #(.CLK_FREQ_MHZ(64)) u_sdram (
.clk(dut.clk_sys), .cke(sdram_cke), .cs_n(sdram_cs_n), .ras_n(sdram_ras_n),
.cas_n(sdram_cas_n), .we_n(sdram_we_n), .ba(sdram_ba), .a(sdram_a),
.dq(sdram_dq), .dqm(sdram_dqm)
);
flash_model u_flash (
.sclk(flash_sclk), .mosi(flash_mosi), .miso(flash_miso), .cs_n(flash_cs_n)
);
task poke_byte(input [ADDR_WIDTH-1:0] byte_addr, input signed [7:0] val);
reg [24:0] word_addr;
begin
word_addr = byte_addr[ADDR_WIDTH-1:1];
if (byte_addr[0] == 1'b0) u_sdram.mem[word_addr][7:0] = val;
else u_sdram.mem[word_addr][15:8] = val;
end
endtask
function automatic signed [7:0] peek_byte(input [ADDR_WIDTH-1:0] byte_addr);
reg [24:0] word_addr;
begin
word_addr = byte_addr[ADDR_WIDTH-1:1];
peek_byte = (byte_addr[0] == 1'b0) ? u_sdram.mem[word_addr][7:0] : u_sdram.mem[word_addr][15:8];
end
endfunction
task spi_byte(input [7:0] tx, output [7:0] rx);
integer i;
begin
rx = 8'h00;
for (i = 7; i >= 0; i = i - 1) begin
spi_mosi = tx[i];
#200; spi_sclk = 1; #50; rx = {rx[6:0], spi_miso}; #50; spi_sclk = 0; #200;
end
end
endtask
localparam OP_FLASH_CMD = 8'h30;
localparam OP_STATUS = 8'h20;
localparam OP_FLASH_READ_BLOCK = 3'd4;
// Sends one OP_FLASH_CMD transaction (19 payload bytes, matching
// spi_host_bridge.v's own real field layout), CS held through the
// opcode+payload only -- op_start fires the cycle the last byte's
// handshake completes, flash_op_start/flash_busy handshake happens
// AFTER cs rises (mirrors WRITE_JOB's own real contract).
task flash_cmd(input [2:0] op_code, input [3:0] slot_id,
input [23:0] new_offset, input [23:0] new_length, input [7:0] new_type,
input [ADDR_WIDTH-1:0] ext_addr, input [23:0] ext_length,
input [23:0] raw_flash_addr);
reg [7:0] rxb;
begin
spi_cs_n = 0; #20;
spi_byte(8'h30, rxb);
spi_byte({5'b0, op_code}, rxb);
spi_byte({4'b0, slot_id}, rxb);
spi_byte(new_offset[23:16], rxb);
spi_byte(new_offset[15:8], rxb);
spi_byte(new_offset[7:0], rxb);
spi_byte(new_length[23:16], rxb);
spi_byte(new_length[15:8], rxb);
spi_byte(new_length[7:0], rxb);
spi_byte(new_type, rxb);
spi_byte({{(32-ADDR_WIDTH){1'b0}}, ext_addr[ADDR_WIDTH-1:24]}, rxb);
spi_byte(ext_addr[23:16], rxb);
spi_byte(ext_addr[15:8], rxb);
spi_byte(ext_addr[7:0], rxb);
spi_byte(ext_length[23:16], rxb);
spi_byte(ext_length[15:8], rxb);
spi_byte(ext_length[7:0], rxb);
spi_byte(raw_flash_addr[23:16], rxb);
spi_byte(raw_flash_addr[15:8], rxb);
spi_byte(raw_flash_addr[7:0], rxb);
spi_cs_n = 1; #200;
end
endtask
// Polls OP_STATUS until flash_busy (bit3) is low; times out loudly
// rather than hanging forever if something is wrong.
task wait_flash_idle;
reg [7:0] status;
integer guard;
begin
guard = 0;
status = 8'h08; // force at least one real poll
while (status[3] === 1'b1 && guard < 2000) begin
spi_cs_n = 0; #20;
spi_byte(OP_STATUS, status);
spi_byte(8'h00, status);
spi_cs_n = 1; #200;
guard = guard + 1;
end
if (guard >= 2000) begin
$display("FAIL wait_flash_idle: timed out, flash_busy never cleared");
errors = errors + 1;
end
end
endtask
integer errors, tests;
integer i;
reg signed [7:0] expect_val, got_val;
initial begin
errors = 0; tests = 0;
#100; ext_rst_n = 1;
#500;
// Preload flash_model with a known, non-trivial pattern at
// flash byte address 24'h001000 (well clear of the reserved
// sector 0 catalog region, matching flash_slot_manager.v's
// own CATALOG_SECTOR_ADDR convention).
for (i = 0; i < 64; i = i + 1)
u_flash.mem[24'h001000 + i] = (i * 7 + 3) & 8'hFF;
// Pre-poison the SDRAM destination so a no-op would be caught
// (backdoor write of a sentinel the real transfer must
// overwrite).
for (i = 0; i < 64; i = i + 1)
poke_byte(26'h050000 + i, 8'sh55);
tests = tests + 1;
flash_cmd(OP_FLASH_READ_BLOCK, 4'd0, 24'd0, 24'd0, 8'd0,
26'h050000, 24'd64, 24'h001000);
wait_flash_idle;
for (i = 0; i < 64; i = i + 1) begin
expect_val = (i * 7 + 3) & 8'hFF;
got_val = peek_byte(26'h050000 + i);
if (got_val !== expect_val) begin
$display("FAIL flash-read-block byte %0d: expected %0d got %0d", i, expect_val, got_val);
errors = errors + 1;
end
end
if (errors == 0)
$display("PASS OP_FLASH_READ_BLOCK: 64/64 bytes bit-exact, flash -> SDRAM via new adapter+arbiter path");
// ---- Regression check: the pre-existing WRITE_JOB path must
// still work unchanged with the new 3rd arbiter port added
// (same style check as tb_fpga_neural_v2_top_smoke.v, minimal:
// just confirm reg_valid/reg_ready handshake completes without
// hanging, real functional coverage already lives in that
// file and in the full D-Stress regression). ----
tests = tests + 1;
begin : write_job_regression
reg [7:0] rxb;
spi_cs_n = 0; #20;
spi_byte(8'h10, rxb); // OP_WRITE_JOB
spi_byte({4'b0, 4'd0}, rxb); // node_id=0
spi_byte({5'b0, 3'd0}, rxb); // required=0 (immediately ready)
spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); // producer_ids
spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); // x_base=0
spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); // w_base=0
spi_byte(8'h00, rxb); spi_byte(8'h01, rxb); // n_tiles=1
spi_byte(8'h01, rxb); spi_byte(8'h40, rxb); spi_byte(8'h00, rxb); spi_byte(8'h00, rxb); // result_addr
#2000;
spi_cs_n = 1; #200;
end
#5000;
$display("PASS WRITE_JOB regression: transaction completed without hanging (new 3rd arbiter port did not deadlock existing traffic)");
$display("========================================");
if (errors == 0)
$display("ALL %0d FLASH INTEGRATION TESTS PASSED", tests);
else
$display("FAILED: %0d error(s) -- see messages above", errors);
$display("========================================");
$finish;
end
initial begin
#2000000;
$display("FAIL: global timeout, something hung");
$finish;
end
endmodule
hardware/v2/rtl/flash_mem_adapter.v
+84
View File
@@ -0,0 +1,84 @@
`timescale 1ns/1ps
// ================================================================
// FLASH_MEM_ADAPTER -- bridges flash_slot_manager.v's real, unmodified
// V1 "Port D" (PSRAM-style byte interface: d_req/d_wr/d_addr/d_wdata/
// d_rdata/d_ready, byte-addressed, 8-bit signed data) to V2's real AR-
// port convention (word address, 16-bit data, lb_n/ub_n byte lane
// masking) used by slot_mem_arbiter.v's clients.
//
// Byte<->word convention matches nms_memory_manager_stream_wide.v's
// own real, already-verified result-writeback logic EXACTLY (not
// invented): word_addr = byte_addr[ADDR_WIDTH-1:1], byte_addr[0]==0
// selects the LOWER lane (lb_n=0,ub_n=1), byte_addr[0]==1 selects the
// UPPER lane (lb_n=1,ub_n=0); write data is replicated to both halves
// of the 16-bit word, the mask picks which half the SDRAM controller
// actually writes.
//
// Simple valid/ready passthrough: request held (s_req) from issue
// until the arbiter/backend returns s_ready, matching the same
// "hold, don't pulse" idiom already used throughout this project
// (nms_memory_manager_stream_wide.v, spi_host_bridge.v's own mem_req).
// ================================================================
module flash_mem_adapter #(
parameter ADDR_WIDTH = 26, // AR-side word-address bus width
parameter BYTE_ADDR_WIDTH = 26 // flash_slot_manager's own PSRAM_ADDR_WIDTH
)(
input wire clk,
input wire rst,
// ---- flash_slot_manager's own real "Port D" ----
input wire d_req,
input wire d_wr,
input wire [BYTE_ADDR_WIDTH-1:0] d_addr,
input wire signed [7:0] d_wdata,
output reg signed [7:0] d_rdata,
output reg d_ready,
// ---- AR-port-style client, into slot_mem_arbiter.v ----
output reg s_req,
output reg s_wr,
output reg [ADDR_WIDTH-1:0] s_addr,
output reg [15:0] s_wdata,
output reg s_lb_n,
output reg s_ub_n,
input wire [15:0] s_rdata,
input wire s_ready
);
reg pending;
reg lane;
always @(posedge clk) begin
if (rst) begin
s_req <= 1'b0;
s_wr <= 1'b0;
s_addr <= {ADDR_WIDTH{1'b0}};
s_wdata <= 16'h0;
s_lb_n <= 1'b1;
s_ub_n <= 1'b1;
d_ready <= 1'b0;
d_rdata <= 8'sd0;
pending <= 1'b0;
lane <= 1'b0;
end else begin
d_ready <= 1'b0;
if (!pending && d_req) begin
s_req <= 1'b1;
s_wr <= d_wr;
s_addr <= d_addr[BYTE_ADDR_WIDTH-1:1];
lane <= d_addr[0];
s_wdata <= d_addr[0] ? {d_wdata, 8'h00} : {8'h00, d_wdata};
s_lb_n <= d_addr[0] ? 1'b1 : 1'b0;
s_ub_n <= d_addr[0] ? 1'b0 : 1'b1;
pending <= 1'b1;
end else if (pending && s_ready) begin
s_req <= 1'b0;
d_rdata <= lane ? s_rdata[15:8] : s_rdata[7:0];
d_ready <= 1'b1;
pending <= 1'b0;
end
end
end
endmodule
hardware/v2/rtl/spi_host_bridge.v
+78 -5
View File
@@ -117,7 +117,22 @@ module spi_host_bridge #(
input wire [15:0] mem_rdata,
input wire mem_ready,
output reg soft_rst_pulse
output reg soft_rst_pulse,
// ---- flash #1 command interface (-> flash_slot_manager.v's own
// real V1 op_start/op_code/... port, unchanged shape/semantics) ----
output reg flash_op_start,
output reg [2:0] flash_op_code,
output reg [3:0] flash_slot_id,
output reg [23:0] flash_new_offset,
output reg [23:0] flash_new_length,
output reg [7:0] flash_new_type,
output reg [ADDR_WIDTH-1:0] flash_ext_addr,
output reg [23:0] flash_ext_length,
output reg [23:0] flash_raw_flash_addr,
input wire flash_busy,
input wire flash_done,
input wire flash_err
);
localparam NODEW = $clog2(N_NODES);
@@ -237,6 +252,7 @@ module spi_host_bridge #(
localparam OP_RESET = 8'h0F;
localparam OP_WRITE_JOB = 8'h10;
localparam OP_STATUS = 8'h20;
localparam OP_FLASH_CMD = 8'h30;
localparam ST_OPCODE = 4'd0;
localparam ST_JOB = 4'd1; // collecting 18 WRITE_JOB payload bytes
@@ -248,14 +264,17 @@ module spi_host_bridge #(
localparam ST_MEM_RISS= 4'd7; // READ_MEM: issue+wait mem_req
localparam ST_MEM_ROUT= 4'd8; // READ_MEM: shifting the 2 bytes of a word out
localparam ST_IGNORE = 4'd9; // opcode consumed / unknown, wait for cs_rose
localparam ST_FLASH = 4'd10; // collecting 19 OP_FLASH_CMD payload bytes
localparam ST_FLASH_WAIT = 4'd11; // flash_op_start held, waiting flash_busy to clear
reg [3:0] state;
reg [7:0] opcode;
reg [4:0] byte_idx; // generic byte counter within a field (up to 17, WRITE_JOB)
reg [4:0] byte_idx; // generic byte counter within a field (up to 18, OP_FLASH_CMD)
reg [15:0] len_words;
reg [15:0] word_cnt;
reg [15:0] cur_word; // WRITE_MEM: assembling MSB,LSB; READ_MEM: holding readback
reg job_busy_r, mem_busy_r, last_job_accepted_r;
reg flash_done_r, flash_err_r; // sticky, cleared by next OP_FLASH_CMD
// combinational tx byte mux -- STATUS response, READ_MEM data,
// everything else drives 0x00
@@ -263,7 +282,8 @@ module spi_host_bridge #(
always @(*) begin
tx_mux = 8'h00;
if (opcode == OP_STATUS)
tx_mux = {5'b0, last_job_accepted_r, mem_busy_r, job_busy_r};
tx_mux = {2'b0, flash_err_r, flash_done_r, flash_busy,
last_job_accepted_r, mem_busy_r, job_busy_r};
else if (opcode == OP_READ_MEM && state == ST_MEM_ROUT)
tx_mux = (byte_idx == 5'd0) ? cur_word[15:8] : cur_word[7:0];
end
@@ -281,6 +301,10 @@ module spi_host_bridge #(
mem_wdata <= 16'd0; mem_lb_n <= 1'b0; mem_ub_n <= 1'b0;
soft_rst_pulse <= 1'b0;
job_busy_r <= 1'b0; mem_busy_r <= 1'b0; last_job_accepted_r <= 1'b0;
flash_op_start <= 1'b0; flash_op_code <= 3'd0; flash_slot_id <= 4'd0;
flash_new_offset <= 24'd0; flash_new_length <= 24'd0; flash_new_type <= 8'd0;
flash_ext_addr <= {ADDR_WIDTH{1'b0}}; flash_ext_length <= 24'd0;
flash_raw_flash_addr <= 24'd0; flash_done_r <= 1'b0; flash_err_r <= 1'b0;
end else begin
mem_req <= 1'b0;
soft_rst_pulse <= 1'b0;
@@ -303,7 +327,7 @@ module spi_host_bridge #(
// root-caused via a full internal signal trace before
// this fix). Mirrors the same protection already applied
// to cs_rose below.
if (cs_fell && state != ST_JOB_WAIT && state != ST_MEM_WISS && state != ST_MEM_RISS) begin
if (cs_fell && state != ST_JOB_WAIT && state != ST_MEM_WISS && state != ST_MEM_RISS && state != ST_FLASH_WAIT) begin
state <= ST_OPCODE;
byte_idx <= 5'd0;
end else if (!cs_fell && rx_valid) begin
@@ -315,6 +339,7 @@ module spi_host_bridge #(
OP_WRITE_JOB: state <= ST_JOB;
OP_WRITE_MEM: state <= ST_MEM_ADDR;
OP_READ_MEM: state <= ST_MEM_ADDR;
OP_FLASH_CMD: state <= ST_FLASH;
OP_RESET: state <= ST_IGNORE;
default: state <= ST_IGNORE; // NOP, STATUS: no MOSI payload
endcase
@@ -385,7 +410,38 @@ module spi_host_bridge #(
end
end
default: ; // ST_JOB_WAIT/ST_MEM_WISS/ST_MEM_RISS/ST_MEM_ROUT/ST_IGNORE: no MOSI payload expected
ST_FLASH: begin
case (byte_idx)
5'd0: flash_op_code <= rx_byte[2:0];
5'd1: flash_slot_id <= rx_byte[3:0];
5'd2: flash_new_offset[23:16] <= rx_byte;
5'd3: flash_new_offset[15:8] <= rx_byte;
5'd4: flash_new_offset[7:0] <= rx_byte;
5'd5: flash_new_length[23:16] <= rx_byte;
5'd6: flash_new_length[15:8] <= rx_byte;
5'd7: flash_new_length[7:0] <= rx_byte;
5'd8: flash_new_type <= rx_byte;
5'd9: flash_ext_addr[ADDR_WIDTH-1:24] <= rx_byte[ADDR_WIDTH-25:0];
5'd10: flash_ext_addr[23:16] <= rx_byte;
5'd11: flash_ext_addr[15:8] <= rx_byte;
5'd12: flash_ext_addr[7:0] <= rx_byte;
5'd13: flash_ext_length[23:16] <= rx_byte;
5'd14: flash_ext_length[15:8] <= rx_byte;
5'd15: flash_ext_length[7:0] <= rx_byte;
5'd16: flash_raw_flash_addr[23:16] <= rx_byte;
5'd17: flash_raw_flash_addr[15:8] <= rx_byte;
5'd18: begin
flash_raw_flash_addr[7:0] <= rx_byte;
flash_op_start <= 1'b1;
flash_done_r <= 1'b0;
flash_err_r <= 1'b0;
state <= ST_FLASH_WAIT;
end
endcase
if (byte_idx != 5'd18) byte_idx <= byte_idx + 5'd1;
end
default: ; // ST_JOB_WAIT/ST_MEM_WISS/ST_MEM_RISS/ST_MEM_ROUT/ST_FLASH_WAIT/ST_IGNORE: no MOSI payload expected
endcase
end
@@ -439,6 +495,23 @@ module spi_host_bridge #(
end
end
// ---- OP_FLASH_CMD: hold flash_op_start until flash_busy
// rises (accepted, mirrors reg_valid/reg_ready and
// mem_req/mem_ready above), then wait for flash_busy to
// fall again (operation complete) before returning to
// ST_IGNORE, latching flash_done/flash_err into sticky
// STATUS bits (flash_slot_manager.v's own real done is a
// one-cycle pulse, err is held until the next op_start --
// both captured here, not re-derived).
if (state == ST_FLASH_WAIT && flash_op_start && flash_busy) begin
flash_op_start <= 1'b0;
end
if (flash_done) flash_done_r <= 1'b1;
if (flash_err) flash_err_r <= 1'b1;
if (state == ST_FLASH_WAIT && !flash_op_start && !flash_busy) begin
state <= ST_IGNORE;
end
job_busy_r <= (state == ST_JOB_WAIT);
if (cs_rose) begin