Outstanding transactions and throughput
Keep several writes and reads in flight, drive the address and data channels at the same time, and measure what a serial driver, a late READY, a register slice and a shared ID each cost.
- Lesson
- P2.5
- Time
- 35 min
- Runs on
- Your browser
- Check
- 5 questions
A verification team wrote an AXI driver the obvious way: offer the address, wait until it is taken, send the data beats, wait for the response, start the next write. Nothing in it broke a rule, and the memory controller under test accepted a request every cycle and answered three cycles after the last beat. The team measured 3 cycles per beat on a bus that could do one, filed a performance bug against the controller, and spent a week in the wrong block. The bench on this page reproduces the serial driver, the fix, and the number in between.
| How the four 2-beat writes are issued | Last response | Cycles for 8 beats | Beats per cycle on W |
|---|---|---|---|
| One at a time: AW, W, wait for B, next | edge 26 | 24 | 0.33 |
| AW then W per write, responses collected alongside | edge 17 | 15 | 0.73 |
| AW and W driven in parallel, responses alongside | edge 13 | 11 | 1.0 |
Measured on the bench with the default write latency. The memory is the same in all three rows; only the manager changed.
Latency is the memory's, throughput is yours
P1.5 drew the line between latency and throughput on the generic link. On AXI the line runs between channels. A write's latency is the time from its last beat to its response, and the subordinate owns it: this memory answers three edges after WLAST. The throughput is how many beats cross the W channel per cycle, and the manager owns that. Nothing in the protocol makes one write wait for the previous one's response.
The serial driver pays that latency on every write: two beats, three idle cycles, one more for the next address. The specification allows a manager to issue a request before earlier ones have completed; a transaction whose request has been taken and whose response has not come back is outstanding. Two duties come with it. The manager must be able to supply all the write data of each request without depending on any other transaction, and to accept every response the same way. That is what lets a subordinate hold several writes safely.
Four writes in flight
The example drives the address channel and the data channel from two processes. The memory takes a request on every edge from 3 to 6 and a beat on every edge from 3 to 10. The first response arrives at edge 7, while three writes are still being issued.
All cycles
The slots count writes the memory has accepted and not yet answered.
- Cycle 0
The first request and the first beat move in the same cycle. The address and data channels are independent.
- Cycle 1
Request 2 is taken while beat 1 of write 1 completes it. One write in flight becomes two.
- Cycle 3
All four requests have been taken in four cycles. The memory owes four responses and has the data of two.
- Cycle 4
The first response arrives: three edges after write 1's last beat. Beats keep flowing underneath it.
- Cycle 7
The last beat moves at edge 10. Eight beats in eight cycles: the data channel never idled.
- Cycle 10
The fourth response at edge 13. The serial driver reached this point at edge 26.
Throughput came from issuing, not from the memory: its latency per write is unchanged.
Reads work the same way: several AR requests may be outstanding and their data comes back later. The manager only needs enough storage to accept the data of every read it has issued.
READY in advance, or READY after VALID
A receiver may keep READY high before anything is offered, or raise it only after it sees VALID. Both are legal; they are not equal. Measured on the bench, a four-beat read with RREADY high in advance delivers its beats at edges 14, 15, 16 and 17. The same read with a manager that raises RREADY one cycle after it sees RVALID delivers them at edges 21, 23, 25 and 27. Every beat costs two cycles, and the checker counts three stalled edges the manager itself caused.
That is why the specification recommends a default of high for the READY of a request channel, provided the receiver can then accept whatever is offered. A subordinate whose AWREADY waits for AWVALID adds a cycle to every request; a manager whose BREADY waits for BVALID adds a cycle to every response.
Register slices: a cycle each, in each direction
Each channel moves information one way and no channel depends on another's timing. So a register stage can be inserted anywhere on any channel, at the price of one cycle of latency there. Interconnects use them to close timing on long wires. Measured through the lesson fabric with two stages on every channel of the manager port. The 2-beat write and 2-beat read that complete at edges 10 and 17 without slices complete at edges 14 and 25 with them. Two stages in and two back are four cycles on every round trip.
| Register stages per channel | B of the write | AR taken | Last R beat | AR to last beat |
|---|---|---|---|---|
| 0 | edge 10 | edge 11 | edge 17 | 6 cycles |
| 2 | edge 14 | edge 15 | edge 25 | 10 cycles |
Measured with fabric: true and slices: 0 versus slices: 2. Latency grows by one cycle per stage per direction; the rate of beats does not change.
A shared ID is a queue
Responses with the same ID return in the order of their requests; P2.10 is about that rule. For throughput it means one thing: every transaction issued under an ID queues behind the earlier ones. The bench has a slow region of memory (mem.slow_above, eight cycles) and reordering allowed. A 2-beat read of a fast address is issued one cycle after a 4-beat read of a slow one. Under the same ID its data arrives twelve cycles after its request, at edges 36 and 37, behind the slow read's four beats. Under its own ID it arrives three cycles after its request, at edges 42 and 43, ahead of the slow read.
This is head-of-line blocking, and it is the manager's choice: independent streams get independent IDs, traffic that must stay in order shares one.
Where you will meet it
Every high-bandwidth manager pipelines: a DMA engine keeps a window of requests in flight, a cache issues line fills under several IDs. The UVM driver of P2.17 runs one process per channel for the same reason. APB has nothing to pipeline, which is why P3 keeps it for registers.
Asked in interviews: outstanding transactions
What is an outstanding transaction?
One whose request has been accepted and whose response has not yet returned. AXI lets a manager have several in flight; the number it allows itself is its window.
May the write data be sent before the write address?
Yes. The channels are independent, so W beats may arrive before, with or after the AW. The manager must never make WVALID wait for AWREADY, or hold back data because another transaction is incomplete.
Why does the spec recommend READY high by default on request channels?
Because a READY that waits for VALID costs a cycle on every transfer. High by default lets a request move in its first cycle; it is only allowed if the receiver can then take whatever is offered.
What does a register slice change?
Latency: one cycle per stage, in that direction, on that channel. The handshake rules hold on both sides of it and the throughput is unchanged once the pipeline is full.
How many transactions must be in flight to saturate the bus?
Enough to cover the latency: a subordinate that answers L cycles after a request needs about L requests outstanding to keep a beat on every cycle.
How does the ID affect throughput?
Same-ID responses come back in request order, so a transaction behind a slow one waits even if its own data is ready. Separate IDs let independent streams overtake each other.
Try it
Three processes run at once: one offers the four requests, one streams the eight beats, one collects the four responses as they arrive. Eight beats cross the W channel on eight consecutive edges and the last response lands at edge 13.
Now serialise it: replace the fork with four calls of axi_write(t[3:0], 32'h800 + 16 * t, 1, 2, INCR) in a loop and the last response moves to edge 26. Keep the fork but set mem.wr_latency = 6; first: the responses move later, the beats do not. Add ready_pattern(1, 16'h00ff) to make BREADY low half the time and watch the responses queue without slowing a single beat.
RUNIcarus Verilog · in your browser
Expected:
e3 AW id=0 addr=0x800 len=1 size=2 INCR e3 W beat 0 data=0x700 strb=1111 last=0 addr 0x800 lanes 0-3 e4 W beat 1 data=0x701 strb=1111 last=1 addr 0x804 lanes 0-3 e4 AW id=1 addr=0x810 len=1 size=2 INCR e5 AW id=2 addr=0x820 len=1 size=2 INCR e5 W beat 0 data=0x702 strb=1111 last=0 addr 0x810 lanes 0-3 e6 W beat 1 data=0x703 strb=1111 last=1 addr 0x814 lanes 0-3 e6 AW id=3 addr=0x830 len=1 size=2 INCR e7 W beat 0 data=0x704 strb=1111 last=0 addr 0x820 lanes 0-3 e7 B id=0 OKAY e8 W beat 1 data=0x705 strb=1111 last=1 addr 0x824 lanes 0-3 e9 W beat 0 data=0x706 strb=1111 last=0 addr 0x830 lanes 0-3 e9 B id=1 OKAY e10 W beat 1 data=0x707 strb=1111 last=1 addr 0x834 lanes 0-3 e11 B id=2 OKAY e13 B id=3 OKAY done at edge 13 checker: 4 writes, 0 reads, 0 stalled edges, 0 rules brokenRead included file: axi_pkg.sv
`timescale 1ns/1ns
// axi_pkg.sv: the arithmetic of a burst (read-only).
// Every transfer n of a burst has an address and a set of byte lanes, computed from the start
// address, the burst length, the transfer size and the burst type exactly as the AXI specification
// defines them. The memory, the manager tasks and the checker all use these functions, so they
// never disagree about where a byte lands.
package axi_pkg;
localparam logic [1:0] FIXED = 2'b00, INCR = 2'b01, WRAP = 2'b10;
localparam logic [1:0] OKAY = 2'b00, EXOKAY = 2'b01, SLVERR = 2'b10, DECERR = 2'b11;
function automatic string burst_name(input logic [1:0] b);
case (b) FIXED: return "FIXED"; INCR: return "INCR"; WRAP: return "WRAP"; default: return "RESERVED"; endcase
endfunction
function automatic string resp_name(input logic [1:0] r);
case (r) OKAY: return "OKAY"; EXOKAY: return "EXOKAY"; SLVERR: return "SLVERR"; default: return "DECERR"; endcase
endfunction
// Bytes in one transfer: 1, 2, 4, ... 128.
function automatic int bytes_of(input logic [2:0] size);
return 1 << size;
endfunction
// The address of transfer n (n counts from 0).
// transfer 0 uses the start address as given, aligned or not;
// later transfers use the start address aligned down to the transfer size, plus n transfers;
// a WRAP burst returns to its wrap boundary (the start aligned down to the whole burst) when
// it reaches the end of that window; a FIXED burst repeats the start address.
function automatic int unsigned beat_addr(input int unsigned start, input int len, input logic [2:0] size,
input logic [1:0] burst, input int n);
int unsigned nb = 1 << size;
int unsigned aligned = (start / nb) * nb;
int unsigned window = nb * (len + 1);
int unsigned boundary = (start / window) * window;
int unsigned a;
if (burst == FIXED || n == 0) return start;
a = aligned + n * nb;
if (burst == WRAP && a >= boundary + window) a = a - window;
return a;
endfunction
// The byte lanes transfer n uses on a bus of bus_bytes bytes, packed as {upper, lower}.
// Transfer 0 of an unaligned burst (and every transfer of a FIXED burst) uses only the lanes from
// its address up to the end of its aligned transfer; every other transfer uses all the lanes of
// its size.
function automatic logic [15:0] beat_lanes(input int unsigned start, input int len, input logic [2:0] size,
input logic [1:0] burst, input int n, input int bus_bytes);
int unsigned nb = 1 << size;
int unsigned a = beat_addr(start, len, size, burst, n);
int unsigned aligned = (a / nb) * nb;
int unsigned word = (a / bus_bytes) * bus_bytes;
int lower = a - word;
int upper = (n == 0 || burst == FIXED) ? (aligned + nb - 1 - word) : (lower + nb - 1);
return {upper[7:0], lower[7:0]};
endfunction
// The strobe of transfer n: a 1 for every lane it uses (bit l is lane l).
function automatic logic [63:0] beat_strb(input int unsigned start, input int len, input logic [2:0] size,
input logic [1:0] burst, input int n, input int bus_bytes);
logic [15:0] lanes = beat_lanes(start, len, size, burst, n, bus_bytes);
logic [63:0] s = 0;
for (int l = lanes[7:0]; l <= lanes[15:8]; l++) s[l] = 1'b1;
return s;
endfunction
// True when the burst crosses a 4 KB boundary (only an INCR burst can).
function automatic bit crosses_4k(input int unsigned start, input int len, input logic [2:0] size, input logic [1:0] burst);
if (burst != INCR) return 0;
return (beat_addr(start, len, size, burst, 0) / 4096) != ((beat_addr(start, len, size, burst, len) + (1 << size) - 1) / 4096);
endfunction
endpackage
Read included file: axi_mem.sv
`timescale 1ns/1ns
// axi_mem.sv: an AXI4 subordinate with a byte-addressed memory behind it (read-only in this lesson).
// It accepts requests on AW and AR, write beats on W (before or after their AW), writes the bytes a
// burst addresses using the strobes, and answers on B and R after a latency. The switches below are
// plain variables: the stimulus sets them by name (mem.reorder = 1;) before it starts.
// rd_latency, wr_latency cycles from a request (or its last write beat) to its first response
// slow_latency, slow_above addresses at or above slow_above answer after slow_latency instead
// err_above addresses at or above err_above get SLVERR (and are not written)
// reorder answer transactions with DIFFERENT IDs as soon as each is ready;
// transactions with the same ID still answer in order
// interleave interleave the read data beats of ready transactions with different IDs
// excl_on support exclusive access (one monitor per ID), else answer OKAY
// posted apply a write to the memory only when its response is sent, so a read
// that arrives between the last beat and B still sees the old bytes
// trace print every byte the memory writes
// Broken twins, each off by default: early_b (B before the last write beat and before AW),
// ignore_strobes (write every lane of the bus word), drop_hold (RDATA changes while RVALID waits
// for RREADY), reorder_same_id (a later transaction with the same ID may answer first),
// wrap_bug (a WRAP burst addressed like INCR).
module axi_mem #(parameter int DATA_W = 32, parameter int ID_W = 4, parameter int DEPTH = 65536) (
input logic aclk,
input logic aresetn,
input logic [ID_W-1:0] awid,
input logic [31:0] awaddr,
input logic [7:0] awlen,
input logic [2:0] awsize,
input logic [1:0] awburst,
input logic awlock,
input logic [3:0] awcache,
input logic [2:0] awprot,
input logic awvalid,
output logic awready,
input logic [DATA_W-1:0] wdata,
input logic [DATA_W/8-1:0] wstrb,
input logic wlast,
input logic wvalid,
output logic wready,
output logic [ID_W-1:0] bid,
output logic [1:0] bresp,
output logic bvalid,
input logic bready,
input logic [ID_W-1:0] arid,
input logic [31:0] araddr,
input logic [7:0] arlen,
input logic [2:0] arsize,
input logic [1:0] arburst,
input logic arlock,
input logic [3:0] arcache,
input logic [2:0] arprot,
input logic arvalid,
output logic arready,
output logic [ID_W-1:0] rid,
output logic [DATA_W-1:0] rdata,
output logic [1:0] rresp,
output logic rlast,
output logic rvalid,
input logic rready
);
import axi_pkg::*;
localparam int BUS = DATA_W / 8;
localparam int N = 16; // transactions in flight per direction
// ---- switches ----
int rd_latency = 2, wr_latency = 2, slow_latency = 8;
int unsigned slow_above = 32'hffff_ffff, err_above = 32'hffff_ffff;
bit reorder = 0, interleave = 0, excl_on = 0, posted = 0, trace = 0;
bit early_b = 0, ignore_strobes = 0, drop_hold = 0, reorder_same_id = 0, wrap_bug = 0;
// ---- the memory ----
logic [7:0] mem [0:DEPTH-1];
initial for (int i = 0; i < DEPTH; i++) mem[i] = 8'h00;
int cycle = -1;
int seq = 0; // every request gets a sequence number: issue order
// ---- write side: accepted AW requests (a ring) and write beats (a ring) ----
logic [ID_W-1:0] aw_id [0:N-1];
int unsigned aw_addr [0:N-1];
int aw_len [0:N-1];
logic [2:0] aw_size [0:N-1];
logic [1:0] aw_burst [0:N-1];
logic aw_lock [0:N-1];
int aw_head = 0, aw_tail = 0, aw_count = 0;
localparam int WQ = 1024;
logic [DATA_W-1:0] w_data [0:WQ-1];
logic [BUS-1:0] w_strb [0:WQ-1];
logic w_last [0:WQ-1];
int w_head = 0, w_tail = 0, w_count = 0, w_bursts = 0; // w_bursts: complete bursts in the ring
// ---- write responses waiting for their latency ----
logic b_use [0:N-1];
logic [ID_W-1:0] b_id [0:N-1];
logic [1:0] b_rsp [0:N-1];
int b_at [0:N-1], b_seq [0:N-1];
logic b_driving = 0;
logic early_done = 0; // the early_b twin answers once per burst
// posted writes: the beats of a write kept until its response goes out
logic [DATA_W-1:0] p_data [0:N-1][0:255];
logic [BUS-1:0] p_strb [0:N-1][0:255];
int unsigned p_addr [0:N-1];
int p_len [0:N-1];
logic [2:0] p_size [0:N-1];
logic [1:0] p_burst [0:N-1];
logic p_pending [0:N-1];
// ---- read side: accepted AR requests with their progress ----
logic r_use [0:N-1];
logic [ID_W-1:0] r_id [0:N-1];
int unsigned r_addr [0:N-1];
int r_len [0:N-1];
logic [2:0] r_size [0:N-1];
logic [1:0] r_burst [0:N-1];
logic r_lock [0:N-1];
int r_at [0:N-1], r_seq [0:N-1], r_beat [0:N-1], r_served [0:N-1];
int r_cur = -1; // the transaction whose beat is on the wires
logic r_driving = 0;
// ---- exclusive monitors, one per ID ----
logic mon_use [0:(1<<ID_W)-1];
int unsigned mon_addr [0:(1<<ID_W)-1];
int unsigned mon_bytes [0:(1<<ID_W)-1];
initial begin
for (int i = 0; i < N; i++) begin b_use[i] = 0; r_use[i] = 0; p_pending[i] = 0; end
for (int i = 0; i < (1 << ID_W); i++) mon_use[i] = 0;
awready = 0; wready = 0; arready = 0; bvalid = 0; rvalid = 0; bid = 0; bresp = 0; rid = 0; rresp = 0; rlast = 0; rdata = 0;
end
function automatic int latency_of(input int unsigned a, input int base);
return (a >= slow_above) ? slow_latency : base;
endfunction
function automatic logic [1:0] burst_used(input logic [1:0] b);
return (wrap_bug && b == WRAP) ? INCR : b;
endfunction
// ---- selection policies ----
// The oldest of all pending entries, or the oldest of its ID: the in-order rule of the protocol.
function automatic bit b_oldest(input int i, input bit same_id_only);
for (int j = 0; j < N; j++)
if (b_use[j] && j != i && b_seq[j] < b_seq[i] && (!same_id_only || b_id[j] == b_id[i])) return 0;
return 1;
endfunction
function automatic bit r_oldest(input int i, input bit same_id_only);
for (int j = 0; j < N; j++)
if (r_use[j] && j != i && r_seq[j] < r_seq[i] && (!same_id_only || r_id[j] == r_id[i])) return 0;
return 1;
endfunction
function automatic bit b_eligible(input int i);
if (!b_use[i] || b_at[i] > cycle) return 0;
if (!reorder) return b_oldest(i, 0);
return reorder_same_id ? 1 : b_oldest(i, 1);
endfunction
function automatic bit r_eligible(input int i);
if (!r_use[i] || r_at[i] > cycle) return 0;
if (!reorder) return r_oldest(i, 0);
return reorder_same_id ? 1 : r_oldest(i, 1);
endfunction
function automatic int free_b();
for (int i = 0; i < N; i++) if (!b_use[i]) return i;
return -1;
endfunction
function automatic int free_r();
for (int i = 0; i < N; i++) if (!r_use[i]) return i;
return -1;
endfunction
function automatic int pick_b();
int best = -1;
for (int i = 0; i < N; i++) if (b_eligible(i) && (best < 0 || b_seq[i] < b_seq[best])) best = i;
return best;
endfunction
function automatic int pick_r();
int best = -1;
for (int i = 0; i < N; i++) if (r_eligible(i)) begin
if (best < 0) best = i;
else if (interleave ? (r_served[i] < r_served[best]) : (r_seq[i] < r_seq[best])) best = i;
end
return best;
endfunction
// ---- the bytes of one read beat, as the bus word that holds them ----
function automatic logic [DATA_W-1:0] read_word(input int unsigned a);
logic [DATA_W-1:0] d = 0;
int unsigned word = (a / BUS) * BUS;
for (int l = 0; l < BUS; l++) d[8*l +: 8] = (word + l < DEPTH) ? mem[word + l] : 8'h00;
return d;
endfunction
// ---- the write of one complete burst into the memory ----
task automatic write_burst(input int e);
logic [1:0] rsp = OKAY;
logic do_write = 1, ended = 0;
int unsigned a, word;
logic [15:0] lanes;
int n = 0, slot;
int unsigned total = (1 << aw_size[e]) * (aw_len[e] + 1);
slot = free_b();
if (aw_lock[e] && excl_on) begin
if (mon_use[aw_id[e]] && mon_addr[aw_id[e]] == aw_addr[e]) begin
rsp = EXOKAY;
$display("mem e%0d: exclusive write id=%0d at 0x%0h: the monitor matched, EXOKAY", cycle, aw_id[e], aw_addr[e]);
end else begin
do_write = 0;
$display("mem e%0d: exclusive write id=%0d at 0x%0h: no monitor, OKAY and nothing written", cycle, aw_id[e], aw_addr[e]);
end
end
while (!ended && n <= aw_len[e]) begin
a = beat_addr(aw_addr[e], aw_len[e], aw_size[e], burst_used(aw_burst[e]), n);
lanes = beat_lanes(aw_addr[e], aw_len[e], aw_size[e], burst_used(aw_burst[e]), n, BUS);
word = (a / BUS) * BUS;
if (a >= err_above) rsp = SLVERR;
else if (do_write && posted && slot >= 0) begin
p_data[slot][n] = w_data[w_head]; p_strb[slot][n] = w_strb[w_head];
end else if (do_write) begin
for (int l = 0; l < BUS; l++)
if ((w_strb[w_head][l] || ignore_strobes) && word + l < DEPTH) begin
mem[word + l] = w_data[w_head][8*l +: 8];
if (trace) $display("mem e%0d: beat %0d writes 0x%02h to 0x%0h (lane %0d)", cycle, n, w_data[w_head][8*l +: 8], word + l, l);
end
end
ended = w_last[w_head]; // a burst ends at its WLAST, whatever AWLEN said
w_head = (w_head + 1) % WQ; w_count = w_count - 1;
n = n + 1;
end
while (!ended) begin // beats past AWLEN (a broken manager) are dropped
ended = w_last[w_head];
w_head = (w_head + 1) % WQ; w_count = w_count - 1;
end
// A write clears every monitor that watches a byte it touched.
if (do_write)
for (int i = 0; i < (1 << ID_W); i++)
if (mon_use[i] && mon_addr[i] < aw_addr[e] + total && aw_addr[e] < mon_addr[i] + mon_bytes[i]) mon_use[i] = 0;
if (slot >= 0) begin
b_use[slot] = 1; b_id[slot] = aw_id[e]; b_rsp[slot] = rsp; b_seq[slot] = seq; seq = seq + 1;
b_at[slot] = cycle + latency_of(aw_addr[e], wr_latency);
if (posted && do_write) begin
p_pending[slot] = 1; p_addr[slot] = aw_addr[e]; p_len[slot] = n - 1; p_size[slot] = aw_size[e]; p_burst[slot] = aw_burst[e];
end
end
endtask
// A posted write reaches the memory when its response goes out.
task automatic apply_posted(input int i);
int unsigned a, word;
for (int n = 0; n <= p_len[i]; n++) begin
a = beat_addr(p_addr[i], p_len[i], p_size[i], burst_used(p_burst[i]), n);
word = (a / BUS) * BUS;
if (a < err_above)
for (int l = 0; l < BUS; l++)
if ((p_strb[i][n][l] || ignore_strobes) && word + l < DEPTH) begin
mem[word + l] = p_data[i][n][8*l +: 8];
if (trace) $display("mem e%0d: posted beat %0d writes 0x%02h to 0x%0h (lane %0d)", cycle, n, p_data[i][n][8*l +: 8], word + l, l);
end
end
p_pending[i] = 0;
endtask
int pick, extra;
always @(posedge aclk) begin
cycle = cycle + 1;
if (!aresetn) begin
aw_head = 0; aw_tail = 0; aw_count = 0; w_head = 0; w_tail = 0; w_count = 0; w_bursts = 0;
for (int i = 0; i < N; i++) begin b_use[i] = 0; r_use[i] = 0; end
b_driving = 0; r_driving = 0; r_cur = -1;
awready <= 0; wready <= 0; arready <= 0; bvalid <= 0; rvalid <= 0;
end else begin
// -- AW: take a request while there is room --
if (awvalid && awready) begin
aw_id[aw_tail] = awid; aw_addr[aw_tail] = awaddr; aw_len[aw_tail] = awlen; aw_size[aw_tail] = awsize;
aw_burst[aw_tail] = awburst; aw_lock[aw_tail] = awlock;
aw_tail = (aw_tail + 1) % N; aw_count = aw_count + 1;
end
// -- W: take a beat while there is room --
if (wvalid && wready) begin
w_data[w_tail] = wdata; w_strb[w_tail] = wstrb; w_last[w_tail] = wlast;
w_tail = (w_tail + 1) % WQ; w_count = w_count + 1;
if (wlast) w_bursts = w_bursts + 1;
end
// -- a complete burst with its request: write it and schedule B --
if (aw_count > 0 && w_bursts > 0) begin
write_burst(aw_head);
aw_head = (aw_head + 1) % N; aw_count = aw_count - 1; w_bursts = w_bursts - 1;
early_done = 0;
end else if (early_b && !early_done && w_count > 0 && w_bursts == 0) begin
// The broken twin: a response as soon as the first write beat arrived, before WLAST and,
// when the data came first, before the address.
pick = free_b();
if (pick >= 0) begin
b_use[pick] = 1; b_id[pick] = (aw_count > 0) ? aw_id[aw_head] : awid; b_rsp[pick] = OKAY; b_seq[pick] = seq; seq = seq + 1; b_at[pick] = cycle;
early_done = 1;
end
end
// -- B: finish a handshake, then put the next response on the wires --
if (b_driving && bready) b_driving = 0;
if (!b_driving) begin
pick = pick_b();
if (pick >= 0) begin
if (p_pending[pick]) apply_posted(pick);
b_driving = 1; bid <= b_id[pick]; bresp <= b_rsp[pick]; b_use[pick] = 0;
end
end
bvalid <= b_driving;
// -- AR: take a request while there is room --
if (arvalid && arready) begin
pick = free_r();
if (pick >= 0) begin
r_use[pick] = 1; r_id[pick] = arid; r_addr[pick] = araddr; r_len[pick] = arlen; r_size[pick] = arsize; r_burst[pick] = arburst;
r_lock[pick] = arlock; r_beat[pick] = 0; r_seq[pick] = seq; seq = seq + 1; r_served[pick] = -1;
r_at[pick] = cycle + latency_of(araddr, rd_latency);
if (arlock && excl_on) begin
mon_use[arid] = 1; mon_addr[arid] = araddr; mon_bytes[arid] = (1 << arsize) * (arlen + 1);
$display("mem e%0d: exclusive read id=%0d at 0x%0h: monitor set", cycle, arid, araddr);
end
end
end
// -- R: finish a handshake, then put the next beat on the wires --
if (r_driving && !rready && drop_hold) rdata <= rdata + 1; // the broken twin: data not held
if (r_driving && rready) begin
r_driving = 0;
if (r_beat[r_cur] > r_len[r_cur]) begin r_use[r_cur] = 0; r_cur = -1; end
else if (interleave) r_cur = -1;
end
if (!r_driving) begin
pick = (r_cur >= 0) ? r_cur : pick_r();
if (pick >= 0) begin
r_cur = pick; r_driving = 1; r_served[pick] = cycle;
extra = r_beat[pick];
rid <= r_id[pick];
rdata <= read_word(beat_addr(r_addr[pick], r_len[pick], r_size[pick], burst_used(r_burst[pick]), extra));
rresp <= (beat_addr(r_addr[pick], r_len[pick], r_size[pick], burst_used(r_burst[pick]), extra) >= err_above) ? SLVERR
: ((r_lock[pick] && excl_on) ? EXOKAY : OKAY);
rlast <= (extra == r_len[pick]);
r_beat[pick] = extra + 1;
end
end
rvalid <= r_driving;
awready <= (aw_count < N - 1);
wready <= (w_count < WQ - 2);
arready <= 1;
end
end
endmodule
Read included file: axi_checker.sv
`timescale 1ns/1ns
// axi_checker.sv: the rules of an AXI4 interface, checked at every rising edge (read-only).
// Each broken rule is named with the channel and the edge where it was seen:
// reset a VALID is 1 while reset is active;
// hold a VALID dropped before its READY: the offer was withdrawn;
// stable a channel's payload changed while its offer waited;
// known a payload is x while its VALID is 1;
// len a FIXED or WRAP burst longer than 16 beats;
// wrap a WRAP burst whose length is not 2, 4, 8 or 16 beats or whose address is not aligned;
// boundary a burst that crosses a 4 KB boundary;
// size a transfer wider than the data bus;
// strobe a write strobe on a lane the beat does not address;
// wlast WLAST on the wrong beat (not AWLEN), or missing after it;
// rlast RLAST on the wrong beat of the oldest read with that ID, or missing after it;
// border BVALID before any outstanding write has its last beat, or before its AW;
// bid, rid a response ID with no outstanding request of that ID;
// exokay EXOKAY for a request that was not exclusive;
// exclusive an exclusive request whose bytes are not a power of two up to 128, or not aligned to them.
// The rules no checker can see on the wires (VALID must not wait for READY; same-ID responses
// in the order of their requests) are checked by the data the tasks compare (lessons P2.2, P2.10).
module axi_checker #(parameter int DATA_W = 32, parameter int ID_W = 4) (
input logic aclk,
input logic aresetn,
input logic [ID_W-1:0] awid,
input logic [31:0] awaddr,
input logic [7:0] awlen,
input logic [2:0] awsize,
input logic [1:0] awburst,
input logic awlock,
input logic awvalid,
input logic awready,
input logic [DATA_W-1:0] wdata,
input logic [DATA_W/8-1:0] wstrb,
input logic wlast,
input logic wvalid,
input logic wready,
input logic [ID_W-1:0] bid,
input logic [1:0] bresp,
input logic bvalid,
input logic bready,
input logic [ID_W-1:0] arid,
input logic [31:0] araddr,
input logic [7:0] arlen,
input logic [2:0] arsize,
input logic [1:0] arburst,
input logic arlock,
input logic arvalid,
input logic arready,
input logic [ID_W-1:0] rid,
input logic [DATA_W-1:0] rdata,
input logic [1:0] rresp,
input logic rlast,
input logic rvalid,
input logic rready
);
import axi_pkg::*;
localparam int BUS = DATA_W / 8;
localparam int Q = 64;
int cycle = -1, writes = 0, reads = 0, stalls = 0, broken = 0;
logic [DATA_W-1:0] used;
function void report(input string rule, input string what);
broken++;
$display("RULE %s broken at edge %0d: %s", rule, cycle, what);
endfunction
// Previous values of every channel, for hold and stable.
logic p_awv = 0, p_awr = 0, p_wv = 0, p_wr = 0, p_bv = 0, p_br = 0, p_arv = 0, p_arr = 0, p_rv = 0, p_rr = 0;
logic [31:0] p_awaddr, p_araddr;
logic [7:0] p_awlen, p_arlen;
logic [2:0] p_awsize, p_arsize;
logic [1:0] p_awburst, p_arburst, p_bresp, p_rresp;
logic [ID_W-1:0] p_awid, p_arid, p_bid, p_rid;
logic [DATA_W-1:0] p_wdata, p_rdata;
logic [BUS-1:0] p_wstrb;
logic p_wlast, p_rlast, p_awlock, p_arlock;
// Write requests in order (AXI4 write data follows the order of the addresses).
int unsigned q_addr [0:Q-1];
int q_len [0:Q-1];
logic [2:0] q_size [0:Q-1];
logic [1:0] q_burst [0:Q-1];
int aw_n = 0; // requests accepted
int wb_n = 0; // write bursts started
int wb_done = 0; // write bursts ended (WLAST seen)
int w_beat = 0; // beats so far in the current burst
int b_n = 0; // responses given
// Beats recorded for bursts that ran ahead of their AW: strobes and count, checked when AW arrives.
logic [BUS-1:0] rec_strb [0:Q-1][0:255];
int rec_beats [0:Q-1];
logic rec_open [0:Q-1];
// Outstanding requests per ID, and the read requests per ID in order.
int aw_out [0:(1<<ID_W)-1], ar_out [0:(1<<ID_W)-1], aw_excl [0:(1<<ID_W)-1], ar_excl [0:(1<<ID_W)-1];
int rq_len [0:(1<<ID_W)-1][0:15];
int rq_head [0:(1<<ID_W)-1], rq_tail [0:(1<<ID_W)-1], r_beat [0:(1<<ID_W)-1];
initial for (int i = 0; i < (1 << ID_W); i++) begin aw_out[i] = 0; ar_out[i] = 0; aw_excl[i] = 0; ar_excl[i] = 0; rq_head[i] = 0; rq_tail[i] = 0; r_beat[i] = 0; end
initial for (int i = 0; i < Q; i++) begin rec_beats[i] = 0; rec_open[i] = 0; end
function automatic bit pow2_le128(input int unsigned n);
return (n == 1 || n == 2 || n == 4 || n == 8 || n == 16 || n == 32 || n == 64 || n == 128);
endfunction
task automatic check_request(input string ch, input int unsigned addr, input int len, input logic [2:0] size, input logic [1:0] burst, input logic lock);
int unsigned bytes = (1 << size) * (len + 1);
if ((burst == FIXED || burst == WRAP) && len > 15) report("len", $sformatf("%s: a %s burst of %0d beats (16 at most)", ch, burst_name(burst), len + 1));
if (burst == WRAP && !(len == 1 || len == 3 || len == 7 || len == 15)) report("wrap", $sformatf("%s: a WRAP burst of %0d beats (2, 4, 8 or 16)", ch, len + 1));
if (burst == WRAP && (addr % (1 << size)) != 0) report("wrap", $sformatf("%s: WRAP burst at 0x%0h is not aligned to its %0d-byte transfers", ch, addr, 1 << size));
if (burst == 2'b11) report("len", $sformatf("%s: burst type 2'b11 is reserved", ch));
if ((1 << size) > BUS) report("size", $sformatf("%s: %0d-byte transfers on a %0d-byte bus", ch, 1 << size, BUS));
if (crosses_4k(addr, len, size, burst)) report("boundary", $sformatf("%s: burst at 0x%0h, %0d beats of %0d bytes, crosses a 4 KB boundary", ch, addr, len + 1, 1 << size));
if (lock && (!pow2_le128(bytes) || (addr % bytes) != 0)) report("exclusive", $sformatf("%s: exclusive access of %0d bytes at 0x%0h (a power of two up to 128, aligned)", ch, bytes, addr));
endtask
task automatic check_strobe(input int k, input int n, input logic [BUS-1:0] strb);
logic [63:0] allowed = beat_strb(q_addr[k % Q], q_len[k % Q], q_size[k % Q], q_burst[k % Q], n, BUS);
if ((strb & ~allowed[BUS-1:0]) != 0)
report("strobe", $sformatf("W beat %0d of write %0d: strobe %b on lanes outside %b", n, k, strb, allowed[BUS-1:0]));
endtask
always @(posedge aclk) begin
cycle++;
if (!aresetn) begin
if (awvalid === 1 || wvalid === 1 || arvalid === 1) report("reset", "a manager VALID is 1 during reset");
if (bvalid === 1 || rvalid === 1) report("reset", "a subordinate VALID is 1 during reset");
end else begin
// hold, stable and known, per channel
if (p_awv && !p_awr && awvalid !== 1) report("hold", "AWVALID dropped before AWREADY");
if (p_awv && !p_awr && awvalid === 1 && (awaddr !== p_awaddr || awlen !== p_awlen || awsize !== p_awsize || awburst !== p_awburst || awid !== p_awid || awlock !== p_awlock)) report("stable", "AW payload changed while the request waited");
if (p_wv && !p_wr && wvalid !== 1) report("hold", "WVALID dropped before WREADY");
if (p_wv && !p_wr && wvalid === 1 && (wdata !== p_wdata || wstrb !== p_wstrb || wlast !== p_wlast)) report("stable", "W payload changed while the beat waited");
if (p_bv && !p_br && bvalid !== 1) report("hold", "BVALID dropped before BREADY");
if (p_bv && !p_br && bvalid === 1 && (bresp !== p_bresp || bid !== p_bid)) report("stable", "B payload changed while the response waited");
if (p_arv && !p_arr && arvalid !== 1) report("hold", "ARVALID dropped before ARREADY");
if (p_arv && !p_arr && arvalid === 1 && (araddr !== p_araddr || arlen !== p_arlen || arsize !== p_arsize || arburst !== p_arburst || arid !== p_arid || arlock !== p_arlock)) report("stable", "AR payload changed while the request waited");
if (p_rv && !p_rr && rvalid !== 1) report("hold", "RVALID dropped before RREADY");
if (p_rv && !p_rr && rvalid === 1 && (rdata !== p_rdata || rresp !== p_rresp || rid !== p_rid || rlast !== p_rlast)) report("stable", "R payload changed while the beat waited");
if (awvalid === 1 && ($isunknown(awaddr) || $isunknown(awlen) || $isunknown(awsize) || $isunknown(awburst) || $isunknown(awid))) report("known", "AW payload is x while AWVALID is 1");
if (wvalid === 1 && ($isunknown(wstrb) || $isunknown(wlast))) report("known", "W strobe or last is x while WVALID is 1");
else if (wvalid === 1) begin
used = '0;
for (int l = 0; l < BUS; l++) if (wstrb[l] === 1) used[8*l +: 8] = 8'hff;
if ($isunknown(wdata & used)) report("known", "W data is x on a strobed lane while WVALID is 1");
end
if (arvalid === 1 && ($isunknown(araddr) || $isunknown(arlen) || $isunknown(arsize) || $isunknown(arburst) || $isunknown(arid))) report("known", "AR payload is x while ARVALID is 1");
if (bvalid === 1 && ($isunknown(bresp) || $isunknown(bid))) report("known", "B payload is x while BVALID is 1");
if (rvalid === 1 && ($isunknown(rresp) || $isunknown(rid) || $isunknown(rlast))) report("known", "R payload is x while RVALID is 1");
if ((awvalid === 1 && awready !== 1) || (wvalid === 1 && wready !== 1) || (arvalid === 1 && arready !== 1) ||
(bvalid === 1 && bready !== 1) || (rvalid === 1 && rready !== 1)) stalls++;
// AW accepted
if (awvalid === 1 && awready === 1) begin
check_request("AW", awaddr, awlen, awsize, awburst, awlock);
q_addr[aw_n % Q] = awaddr; q_len[aw_n % Q] = awlen; q_size[aw_n % Q] = awsize; q_burst[aw_n % Q] = awburst;
aw_out[awid]++; if (awlock) aw_excl[awid]++;
// beats of this write that arrived before its address: check them now
if (rec_beats[aw_n % Q] > 0) begin
for (int n = 0; n < rec_beats[aw_n % Q]; n++) check_strobe(aw_n, n, rec_strb[aw_n % Q][n]);
if (!rec_open[aw_n % Q] && rec_beats[aw_n % Q] != awlen + 1)
report("wlast", $sformatf("write %0d sent %0d beats before its AW said AWLEN=%0d", aw_n, rec_beats[aw_n % Q], awlen));
if (!rec_open[aw_n % Q]) rec_beats[aw_n % Q] = 0;
end
aw_n++;
end
// W beat accepted
if (wvalid === 1 && wready === 1) begin
if (w_beat == 0) begin wb_n++; rec_beats[(wb_n - 1) % Q] = 0; rec_open[(wb_n - 1) % Q] = 1; end
if (wb_n <= aw_n) begin
check_strobe(wb_n - 1, w_beat, wstrb);
if (wlast === 1 && w_beat != q_len[(wb_n - 1) % Q]) report("wlast", $sformatf("WLAST on beat %0d of write %0d, AWLEN=%0d", w_beat, wb_n - 1, q_len[(wb_n - 1) % Q]));
if (wlast !== 1 && w_beat == q_len[(wb_n - 1) % Q]) report("wlast", $sformatf("beat %0d of write %0d is AWLEN=%0d but WLAST is 0", w_beat, wb_n - 1, q_len[(wb_n - 1) % Q]));
end else begin
rec_strb[(wb_n - 1) % Q][w_beat] = wstrb;
end
rec_beats[(wb_n - 1) % Q] = w_beat + 1;
w_beat++;
if (wlast === 1) begin w_beat = 0; wb_done++; rec_open[(wb_n - 1) % Q] = 0; if (wb_n <= aw_n) rec_beats[(wb_n - 1) % Q] = 0; end
end
// B offered or accepted
if (bvalid === 1 && !p_bv) begin
if (b_n >= aw_n && b_n >= wb_done) report("border", "BVALID with no write to answer: neither an AW nor a WLAST is outstanding");
else if (b_n >= wb_done) report("border", "BVALID before the last beat of the write it answers");
else if (b_n >= aw_n) report("border", "BVALID before the AW of the write it answers was accepted");
end
if (bvalid === 1 && bready === 1) begin
if (aw_out[bid] == 0) report("bid", $sformatf("B with id %0d, which has no outstanding write", bid));
else begin
aw_out[bid]--; writes++;
if (bresp === EXOKAY && aw_excl[bid] == 0) report("exokay", $sformatf("EXOKAY for write id %0d, which was not exclusive", bid));
if (aw_excl[bid] > 0 && aw_excl[bid] > aw_out[bid]) aw_excl[bid] = aw_out[bid];
end
b_n++;
end
// AR accepted
if (arvalid === 1 && arready === 1) begin
check_request("AR", araddr, arlen, arsize, arburst, arlock);
rq_len[arid][rq_tail[arid]] = arlen; rq_tail[arid] = (rq_tail[arid] + 1) % 16;
ar_out[arid]++; if (arlock) ar_excl[arid]++;
end
// R beat accepted
if (rvalid === 1 && rready === 1) begin
if (ar_out[rid] == 0) report("rid", $sformatf("R with id %0d, which has no outstanding read", rid));
else begin
if (rlast === 1 && r_beat[rid] != rq_len[rid][rq_head[rid]])
report("rlast", $sformatf("RLAST on beat %0d of the oldest read with id %0d, whose ARLEN is %0d", r_beat[rid], rid, rq_len[rid][rq_head[rid]]));
if (rlast !== 1 && r_beat[rid] == rq_len[rid][rq_head[rid]])
report("rlast", $sformatf("beat %0d of the oldest read with id %0d is its ARLEN but RLAST is 0", r_beat[rid], rid));
if (rresp === EXOKAY && ar_excl[rid] == 0) report("exokay", $sformatf("EXOKAY for read id %0d, which was not exclusive", rid));
r_beat[rid]++;
if (rlast === 1) begin
r_beat[rid] = 0; rq_head[rid] = (rq_head[rid] + 1) % 16; ar_out[rid]--; reads++;
if (ar_excl[rid] > ar_out[rid]) ar_excl[rid] = ar_out[rid];
end
end
end
end
p_awv = (awvalid === 1); p_awr = (awready === 1); p_wv = (wvalid === 1); p_wr = (wready === 1);
p_bv = (bvalid === 1); p_br = (bready === 1); p_arv = (arvalid === 1); p_arr = (arready === 1);
p_rv = (rvalid === 1); p_rr = (rready === 1);
p_awaddr = awaddr; p_awlen = awlen; p_awsize = awsize; p_awburst = awburst; p_awid = awid; p_awlock = awlock;
p_wdata = wdata; p_wstrb = wstrb; p_wlast = wlast; p_bresp = bresp; p_bid = bid;
p_araddr = araddr; p_arlen = arlen; p_arsize = arsize; p_arburst = arburst; p_arid = arid; p_arlock = arlock;
p_rdata = rdata; p_rresp = rresp; p_rid = rid; p_rlast = rlast;
end
final $display("checker: %0d writes, %0d reads, %0d stalled edges, %0d rules broken", writes, reads, stalls, broken);
endmodule
Read included file: axi_tasks.svh
// axi_tasks.svh: the manager side of the bench, as tasks for the stimulus (read-only).
// Data for a write burst goes in wd[0], wd[1], ... (one entry per beat, the lowest byte at the
// lowest address); the data of a read comes back in rd[0], rd[1], ... with its responses in rr[].
// Every task prints the beat log: one line per handshake, with the edge where it happened.
// `cycle` counts rising edges from 0.
int cycle = -1;
always @(posedge aclk) cycle = cycle + 1;
logic [63:0] wd [0:255];
logic [63:0] rd [0:255];
logic [1:0] rr [0:255];
logic [63:0] rdq [0:15][0:255]; // read data per ID, for reads collected out of order
logic [1:0] rrq [0:15][0:255];
int rdn [0:15]; // beats received per ID
logic [1:0] last_bresp = 0;
int last_beats = 0;
// What the task is driving right now, for the waveform scenes: the address and lanes of the beat.
int unsigned beat_a = 0;
logic [7:0] beat_m = 0;
// quiet = 1 silences the beat log (a predict item can then ask for a dump line instead).
bit quiet = 0;
// Strobe override: when strb_force is 1, every write beat uses strb_val as it is.
bit strb_force = 0;
logic [7:0] strb_val = 8'hff;
// Outstanding read requests per ID, so the tasks know the lanes of the beats that come back.
int unsigned arq_addr [0:15][0:7];
int arq_len [0:15][0:7];
logic [2:0] arq_size [0:15][0:7];
logic [1:0] arq_burst [0:15][0:7];
int arq_head [0:15], arq_tail [0:15];
initial for (int i = 0; i < 16; i++) begin arq_head[i] = 0; arq_tail[i] = 0; rdn[i] = 0; end
// READY patterns for the response channels: bit (cycle % 16) of the pattern drives the signal.
bit rpat_on = 0, bpat_on = 0;
logic [15:0] rpat = 16'hffff, bpat = 16'hffff;
always @(posedge aclk) begin
#1;
if (rpat_on) rready = rpat[cycle % 16];
if (bpat_on) bready = bpat[cycle % 16];
end
// A text waveform for the lesson's trace blocks: with watch = 1, one line per rising edge with the
// values sampled at that edge (cycle k of a trace is what edge k sees).
bit watch = 0;
int wave_edge = -1;
always @(posedge aclk) begin
wave_edge = wave_edge + 1;
if (watch) $display("wave e%0d: aw=%b%b w=%b%b b=%b%b ar=%b%b r=%b%b awaddr=%h wdata=%h wstrb=%b wlast=%b bid=%h bresp=%b araddr=%h rid=%h rdata=%h rresp=%b rlast=%b",
wave_edge, awvalid, awready, wvalid, wready, bvalid, bready, arvalid, arready, rvalid, rready,
awaddr, wdata, wstrb, wlast, bid, bresp, araddr, rid, rdata, rresp, rlast);
end
// Drive RREADY (ch = 0) or BREADY (ch = 1) from a 16-bit pattern; 16'hffff restores always ready.
task automatic ready_pattern(input int ch, input logic [15:0] pat);
if (ch == 0) begin rpat = pat; rpat_on = 1; rready = pat[cycle % 16]; end
else begin bpat = pat; bpat_on = 1; bready = pat[cycle % 16]; end
endtask
// The write address channel: offer the request and hold it until it is taken.
task automatic aw_issue(input logic [ID_W-1:0] id, input int unsigned addr, input int len, input logic [2:0] size,
input logic [1:0] burst, input logic lock = 0);
awid = id; awaddr = addr; awlen = len[7:0]; awsize = size; awburst = burst; awlock = lock; awvalid = 1;
do @(posedge aclk); while (!awready);
#1 if (!quiet) $display("e%0d AW id=%0d addr=0x%0h len=%0d size=%0d %0s%0s", cycle, id, addr, len, size, burst_name(burst), lock ? " exclusive" : "");
awvalid = 0; awaddr = 'x; awlen = 'x; awsize = 'x; awburst = 'x; awid = 'x; awlock = 0;
endtask
// The write data channel: the beats of one burst from wd[first..], each placed in its lanes.
task automatic w_burst(input int unsigned addr, input int len, input logic [2:0] size, input logic [1:0] burst, input int first = 0);
logic [15:0] lanes;
for (int n = 0; n <= len; n++) begin
lanes = beat_lanes(addr, len, size, burst, n, BUS_BYTES);
beat_a = beat_addr(addr, len, size, burst, n);
beat_m = strb_force ? strb_val : beat_strb(addr, len, size, burst, n, BUS_BYTES);
wdata = wd[first + n] << (8 * lanes[7:0]);
wstrb = beat_m[BUS_BYTES-1:0];
wlast = (n == len);
wvalid = 1;
do @(posedge aclk); while (!wready);
#1 if (!quiet) $display("e%0d W beat %0d data=0x%0h strb=%b last=%0d addr 0x%0h lanes %0d-%0d", cycle, n, wdata, wstrb, wlast, beat_a, lanes[7:0], lanes[15:8]);
end
wvalid = 0; wdata = 'x; wstrb = 'x; wlast = 0; beat_m = 0;
endtask
// Wait for the write response of one transaction with this ID.
task automatic b_wait(input logic [ID_W-1:0] id);
do @(posedge aclk); while (!(bvalid && bready && bid == id));
last_bresp = bresp; // sampled at the edge, before the subordinate moves on
#1 if (!quiet) $display("e%0d B id=%0d %0s", cycle, id, resp_name(last_bresp));
endtask
// A complete write. order 0 offers the address, then the first beat one cycle later whether or
// not the address has been taken yet (a manager must never make WVALID wait for AWREADY); 1 offers
// the beats first and the address after beat 0; 2 offers the address and beat 0 in the same cycle.
// Returns after B, with the response in last_bresp.
task automatic axi_write(input logic [ID_W-1:0] id, input int unsigned addr, input int len, input logic [2:0] size,
input logic [1:0] burst, input int order = 0, input logic lock = 0);
case (order)
1: fork
w_burst(addr, len, size, burst);
begin @(posedge aclk); #1 aw_issue(id, addr, len, size, burst, lock); end
join
2: fork
aw_issue(id, addr, len, size, burst, lock);
w_burst(addr, len, size, burst);
join
default: fork
aw_issue(id, addr, len, size, burst, lock);
begin @(posedge aclk); #1 w_burst(addr, len, size, burst); end
join
endcase
b_wait(id);
endtask
// The read address channel: offer the request and remember it for the beats that come back.
task automatic ar_issue(input logic [ID_W-1:0] id, input int unsigned addr, input int len, input logic [2:0] size,
input logic [1:0] burst, input logic lock = 0);
arq_addr[id][arq_tail[id]] = addr; arq_len[id][arq_tail[id]] = len; arq_size[id][arq_tail[id]] = size;
arq_burst[id][arq_tail[id]] = burst; arq_tail[id] = (arq_tail[id] + 1) % 8;
arid = id; araddr = addr; arlen = len[7:0]; arsize = size; arburst = burst; arlock = lock; arvalid = 1;
do @(posedge aclk); while (!arready);
#1 if (!quiet) $display("e%0d AR id=%0d addr=0x%0h len=%0d size=%0d %0s%0s", cycle, id, addr, len, size, burst_name(burst), lock ? " exclusive" : "");
arvalid = 0; araddr = 'x; arlen = 'x; arsize = 'x; arburst = 'x; arid = 'x; arlock = 0;
endtask
// Collect read beats until n transactions have ended (their RLAST). Each beat lands in rdq[rid][k]
// with only the bytes of the transfer (lane-shifted down), and rd[] holds the last transaction.
task automatic r_collect(input int n);
int done = 0, h;
logic [15:0] lanes;
logic [63:0] bytes;
logic [ID_W-1:0] id;
logic [DATA_W-1:0] data;
logic [1:0] resp;
logic last;
while (done < n) begin
do @(posedge aclk); while (!(rvalid && rready));
id = rid; data = rdata; resp = rresp; last = rlast; // sampled at the edge, before the subordinate moves on
h = arq_head[id];
lanes = beat_lanes(arq_addr[id][h], arq_len[id][h], arq_size[id][h], arq_burst[id][h], rdn[id], BUS_BYTES);
beat_a = beat_addr(arq_addr[id][h], arq_len[id][h], arq_size[id][h], arq_burst[id][h], rdn[id]);
beat_m = beat_strb(arq_addr[id][h], arq_len[id][h], arq_size[id][h], arq_burst[id][h], rdn[id], BUS_BYTES);
bytes = (data >> (8 * lanes[7:0])) & ((64'h1 << (8 * (lanes[15:8] - lanes[7:0] + 1))) - 1);
rdq[id][rdn[id]] = bytes; rrq[id][rdn[id]] = resp;
#1 if (!quiet) $display("e%0d R id=%0d beat %0d data=0x%0h %0s%0s addr 0x%0h lanes %0d-%0d", cycle, id, rdn[id], bytes, resp_name(resp),
last ? " last" : "", beat_a, lanes[7:0], lanes[15:8]);
rdn[id] = rdn[id] + 1;
if (last) begin
for (int k = 0; k < rdn[id]; k++) begin rd[k] = rdq[id][k]; rr[k] = rrq[id][k]; end
last_beats = rdn[id]; rdn[id] = 0; arq_head[id] = (h + 1) % 8; done = done + 1;
end
end
beat_m = 0;
endtask
// A complete read: AR, then every beat, into rd[] and rr[].
task automatic axi_read(input logic [ID_W-1:0] id, input int unsigned addr, input int len, input logic [2:0] size,
input logic [1:0] burst, input logic lock = 0);
ar_issue(id, addr, len, size, burst, lock);
r_collect(1);
endtask
// The table of a burst before it runs: the address, lanes and strobe of every beat.
task automatic beat_table(input int unsigned addr, input int len, input logic [2:0] size, input logic [1:0] burst);
logic [15:0] lanes;
logic [63:0] s;
int unsigned a;
$display("burst at 0x%0h: %0d beat%0s of %0d byte%0s, %0s, bus %0d bytes", addr, len + 1, len ? "s" : "", 1 << size, size ? "s" : "",
burst_name(burst), BUS_BYTES);
$display(" beat address lanes strobe");
for (int n = 0; n <= len; n++) begin
lanes = beat_lanes(addr, len, size, burst, n, BUS_BYTES);
s = beat_strb(addr, len, size, burst, n, BUS_BYTES);
a = beat_addr(addr, len, size, burst, n);
$display(" %4d 0x%05h %0d-%0d %b", n, a[19:0], lanes[7:0], lanes[15:8], s[BUS_BYTES-1:0]);
end
if (crosses_4k(addr, len, size, burst)) $display(" this burst crosses a 4 KB boundary");
endtask
// One byte of the memory that owns an address (the fabric's second memory above 0x8000).
function automatic logic [7:0] mem_byte(input int unsigned a);
`ifdef AXI_MEM1
if (a >= 32'h8000) return `AXI_MEM1.mem[a];
`endif
return `AXI_MEM.mem[a];
endfunction
task automatic set_byte(input int unsigned a, input logic [7:0] v);
`ifdef AXI_MEM1
if (a >= 32'h8000) `AXI_MEM1.mem[a] = v;
else `AXI_MEM.mem[a] = v;
`else
`AXI_MEM.mem[a] = v;
`endif
endtask
// The memory, n bytes from addr, one bus word per group.
task automatic dump(input int unsigned addr, input int n);
string line;
int unsigned a = (addr / 16) * 16;
while (a < addr + n) begin
line = $sformatf("0x%05h ", a[19:0]);
for (int i = 0; i < 16; i++) begin
if (i % BUS_BYTES == 0) line = {line, " "};
line = {line, $sformatf(" %02h", mem_byte(a + i))};
end
$display("%s", line);
a = a + 16;
end
endtask
// Fill n bytes from addr with first, first+1, ... so untouched bytes stay recognisable.
task automatic mem_fill(input int unsigned addr, input int n, input logic [7:0] first = 8'h00);
for (int i = 0; i < n; i++) set_byte(addr + i, first + i[7:0]);
endtask
// Drive nothing for n cycles.
task automatic idle(input int n);
repeat (n) @(posedge aclk);
#1;
endtask
Check yourself
Remember
- A transaction is outstanding from the handshake of its request to the handshake of its response. AXI lets a manager keep many in flight.
- Throughput belongs to the manager: drive AW and W in parallel and collect responses as they come, and the data channel never idles. Latency belongs to the subordinate and does not change.
- Once a write is issued, all its data must come and its response must be taken without depending on any other transaction.
- READY high in advance moves a transfer in its first cycle; READY that waits for VALID costs a cycle every time. A register stage costs one cycle per direction and no bandwidth.
- Same-ID responses queue in request order, so a transaction behind a slow one waits. Give independent streams their own IDs.
Practice
- An issuer with N outstanding Implement · Medium