Research-Stack/4-Infrastructure/hardware/tangnano9k_rrc_q16_accel.v
2026-05-11 22:18:31 -05:00

344 lines
12 KiB
Verilog

// Tang Nano 9K Rainbow Raccoon Q16.16 accelerator surface.
//
// UART frame in:
// A5 02 seq opcode len payload[len] crc8
//
// UART frame out:
// A6 02 seq status len payload[len] crc8
//
// Opcodes:
// 0x20 = Q16.16 arithmetic shift/divide by 65536 witness
// payload: int32 x
// output: opcode, int32 (x >>> 16), pass
// 0x21 = Q16.16 weighted-term bounded witness
// payload: int32 E, int32 alpha
// output: opcode, int32 ((E * alpha) >>> 16), pass
// 0x22 = Q16.16 shift monotonicity witness
// payload: int32 a, int32 b
// output: opcode, int32 (a >>> 16), int32 (b >>> 16), pass
//
// This is a receipt surface for Lean-proved fixed-point lowering lemmas. The
// host owns route selection and proof admission; the FPGA only accelerates the
// deterministic arithmetic witness lane.
`timescale 1ns / 1ps
module tangnano9k_rrc_q16_accel (
input clk,
input rst_n,
input uart_rx_pin,
output uart_tx_pin,
output [5:0] led
);
localparam MAGIC_IN = 8'hA5;
localparam MAGIC_OUT = 8'hA6;
localparam VERSION = 8'h02;
localparam OP_SHIFT_DIV = 8'h20;
localparam OP_WEIGHTED = 8'h21;
localparam OP_MONOTONE = 8'h22;
localparam RX_WAIT_MAGIC = 4'd0;
localparam RX_VERSION = 4'd1;
localparam RX_SEQ = 4'd2;
localparam RX_OPCODE = 4'd3;
localparam RX_LEN = 4'd4;
localparam RX_PAYLOAD = 4'd5;
localparam RX_CRC = 4'd6;
localparam RX_PROCESS = 4'd7;
localparam TX_WAIT = 4'd8;
localparam RX_TIMEOUT_CYCLES = 19'd270000; // about 10 ms at 27 MHz
wire [7:0] rx_data;
wire rx_done;
reg [7:0] tx_data;
reg tx_start;
wire tx_busy;
reg [3:0] state;
reg [7:0] seq;
reg [7:0] opcode;
reg [4:0] payload_len;
reg [4:0] payload_index;
reg [7:0] payload [0:15];
reg [7:0] calc_crc;
reg [7:0] frame_crc;
reg [7:0] status;
reg [7:0] resp_len;
reg [7:0] resp_crc;
reg [4:0] tx_index;
reg [18:0] rx_idle_cycles;
reg signed [31:0] result0;
reg signed [31:0] result1;
reg pass_bit;
wire signed [31:0] payload_i32_0;
wire signed [31:0] payload_i32_1;
wire signed [63:0] weighted_product;
wire signed [31:0] shifted0;
wire signed [31:0] shifted1;
wire signed [31:0] weighted_shifted;
assign payload_i32_0 = {payload[0], payload[1], payload[2], payload[3]};
assign payload_i32_1 = {payload[4], payload[5], payload[6], payload[7]};
assign weighted_product = $signed(payload_i32_0) * $signed(payload_i32_1);
assign shifted0 = payload_i32_0 >>> 16;
assign shifted1 = payload_i32_1 >>> 16;
assign weighted_shifted = weighted_product >>> 16;
uart_rx rx_inst (
.clk(clk),
.rst_n(rst_n),
.rx_pin(uart_rx_pin),
.rx_data(rx_data),
.rx_done(rx_done)
);
uart_tx tx_inst (
.clk(clk),
.rst_n(rst_n),
.tx_start(tx_start),
.tx_data(tx_data),
.uart_tx(uart_tx_pin),
.tx_busy(tx_busy)
);
function [7:0] crc8_next;
input [7:0] crc;
input [7:0] data;
begin
crc8_next = crc ^ data;
end
endfunction
function [7:0] result0_byte;
input [1:0] idx;
begin
case (idx)
2'd0: result0_byte = result0[31:24];
2'd1: result0_byte = result0[23:16];
2'd2: result0_byte = result0[15:8];
default: result0_byte = result0[7:0];
endcase
end
endfunction
function [7:0] result1_byte;
input [1:0] idx;
begin
case (idx)
2'd0: result1_byte = result1[31:24];
2'd1: result1_byte = result1[23:16];
2'd2: result1_byte = result1[15:8];
default: result1_byte = result1[7:0];
endcase
end
endfunction
function [7:0] tx_byte_at;
input [4:0] idx;
begin
case (idx)
5'd0: tx_byte_at = MAGIC_OUT;
5'd1: tx_byte_at = VERSION;
5'd2: tx_byte_at = seq;
5'd3: tx_byte_at = status;
5'd4: tx_byte_at = resp_len;
5'd5: tx_byte_at = opcode;
5'd6: tx_byte_at = result0_byte(2'd0);
5'd7: tx_byte_at = result0_byte(2'd1);
5'd8: tx_byte_at = result0_byte(2'd2);
5'd9: tx_byte_at = result0_byte(2'd3);
5'd10: tx_byte_at = (resp_len == 8'd10) ? result1_byte(2'd0) : {7'd0, pass_bit};
5'd11: tx_byte_at = (resp_len == 8'd10) ? result1_byte(2'd1) : resp_crc;
5'd12: tx_byte_at = result1_byte(2'd2);
5'd13: tx_byte_at = result1_byte(2'd3);
5'd14: tx_byte_at = {7'd0, pass_bit};
5'd15: tx_byte_at = resp_crc;
default: tx_byte_at = 8'h00;
endcase
end
endfunction
integer i;
always @(posedge clk or negedge rst_n) begin
if (!rst_n) begin
state <= RX_WAIT_MAGIC;
seq <= 8'h00;
opcode <= 8'h00;
payload_len <= 5'd0;
payload_index <= 5'd0;
calc_crc <= 8'h00;
frame_crc <= 8'h00;
status <= 8'h00;
resp_len <= 8'd0;
resp_crc <= 8'h00;
tx_index <= 5'd0;
tx_data <= 8'h00;
tx_start <= 1'b0;
rx_idle_cycles <= 19'd0;
result0 <= 32'sd0;
result1 <= 32'sd0;
pass_bit <= 1'b0;
for (i = 0; i < 16; i = i + 1) begin
payload[i] <= 8'h00;
end
end else begin
tx_start <= 1'b0;
if (rx_done || state == RX_WAIT_MAGIC || state == RX_PROCESS || state == TX_WAIT) begin
rx_idle_cycles <= 19'd0;
end else if (rx_idle_cycles >= RX_TIMEOUT_CYCLES) begin
state <= RX_WAIT_MAGIC;
calc_crc <= 8'h00;
payload_index <= 5'd0;
rx_idle_cycles <= 19'd0;
end else begin
rx_idle_cycles <= rx_idle_cycles + 19'd1;
end
if (rx_done && rx_data == MAGIC_IN && state != TX_WAIT) begin
calc_crc <= rx_data;
payload_index <= 5'd0;
state <= RX_VERSION;
end else case (state)
RX_WAIT_MAGIC: begin
if (rx_done && rx_data == MAGIC_IN) begin
calc_crc <= rx_data;
state <= RX_VERSION;
end
end
RX_VERSION: begin
if (rx_done) begin
calc_crc <= crc8_next(calc_crc, rx_data);
state <= (rx_data == VERSION) ? RX_SEQ : RX_WAIT_MAGIC;
end
end
RX_SEQ: begin
if (rx_done) begin
seq <= rx_data;
calc_crc <= crc8_next(calc_crc, rx_data);
state <= RX_OPCODE;
end
end
RX_OPCODE: begin
if (rx_done) begin
opcode <= rx_data;
calc_crc <= crc8_next(calc_crc, rx_data);
state <= RX_LEN;
end
end
RX_LEN: begin
if (rx_done) begin
payload_len <= (rx_data[4:0] > 5'd16) ? 5'd16 : rx_data[4:0];
payload_index <= 5'd0;
calc_crc <= crc8_next(calc_crc, rx_data);
state <= (rx_data == 8'd0) ? RX_CRC : RX_PAYLOAD;
end
end
RX_PAYLOAD: begin
if (rx_done) begin
payload[payload_index] <= rx_data;
calc_crc <= crc8_next(calc_crc, rx_data);
if (payload_index + 5'd1 >= payload_len) begin
state <= RX_CRC;
end
payload_index <= payload_index + 5'd1;
end
end
RX_CRC: begin
if (rx_done) begin
frame_crc <= rx_data;
if (rx_data != calc_crc) begin
status <= 8'hE1;
end else begin
status <= 8'h00;
end
state <= RX_PROCESS;
end
end
RX_PROCESS: begin
result0 <= 32'sd0;
result1 <= 32'sd0;
pass_bit <= 1'b0;
resp_len <= 8'd1;
if (status == 8'hE1) begin
resp_len <= 8'd1;
resp_crc <= MAGIC_OUT ^ VERSION ^ seq ^ status ^ 8'd1 ^ opcode;
end else if (opcode == OP_SHIFT_DIV && payload_len == 5'd4) begin
result0 <= shifted0;
pass_bit <= 1'b1;
resp_len <= 8'd6;
resp_crc <= MAGIC_OUT ^ VERSION ^ seq ^ 8'h00 ^ 8'd6 ^ opcode ^
shifted0[31:24] ^ shifted0[23:16] ^
shifted0[15:8] ^ shifted0[7:0] ^ 8'h01;
end else if (opcode == OP_WEIGHTED && payload_len == 5'd8) begin
result0 <= weighted_shifted;
pass_bit <= (payload_i32_0 >= 32'sd0) &&
(payload_i32_1 >= 32'sd0) &&
(payload_i32_1 <= 32'sd65536) &&
(weighted_shifted <= payload_i32_0);
resp_len <= 8'd6;
resp_crc <= MAGIC_OUT ^ VERSION ^ seq ^ 8'h00 ^ 8'd6 ^ opcode ^
weighted_shifted[31:24] ^ weighted_shifted[23:16] ^
weighted_shifted[15:8] ^ weighted_shifted[7:0] ^
{7'd0, ((payload_i32_0 >= 32'sd0) &&
(payload_i32_1 >= 32'sd0) &&
(payload_i32_1 <= 32'sd65536) &&
(weighted_shifted <= payload_i32_0))};
end else if (opcode == OP_MONOTONE && payload_len == 5'd8) begin
result0 <= shifted0;
result1 <= shifted1;
pass_bit <= (payload_i32_0 <= payload_i32_1) && (shifted0 <= shifted1);
resp_len <= 8'd10;
resp_crc <= MAGIC_OUT ^ VERSION ^ seq ^ 8'h00 ^ 8'd10 ^ opcode ^
shifted0[31:24] ^ shifted0[23:16] ^
shifted0[15:8] ^ shifted0[7:0] ^
shifted1[31:24] ^ shifted1[23:16] ^
shifted1[15:8] ^ shifted1[7:0] ^
{7'd0, ((payload_i32_0 <= payload_i32_1) && (shifted0 <= shifted1))};
end else begin
status <= 8'hE2;
resp_len <= 8'd1;
resp_crc <= MAGIC_OUT ^ VERSION ^ seq ^ 8'hE2 ^ 8'd1 ^ opcode;
end
tx_index <= 5'd0;
state <= TX_WAIT;
end
TX_WAIT: begin
if (!tx_busy && !tx_start) begin
tx_data <= tx_byte_at(tx_index);
tx_start <= 1'b1;
if ((resp_len == 8'd10 && tx_index == 5'd15) ||
(resp_len == 8'd6 && tx_index == 5'd11) ||
(resp_len == 8'd1 && tx_index == 5'd6)) begin
state <= RX_WAIT_MAGIC;
end else begin
tx_index <= tx_index + 5'd1;
end
end
end
default: state <= RX_WAIT_MAGIC;
endcase
end
end
wire [5:0] led_state;
assign led_state = {status[1:0], opcode[3:0]};
assign led = ~led_state;
endmodule