1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
|
//
// Copyright 2018 Ettus Research, A National Instruments Company
//
// SPDX-License-Identifier: LGPL-3.0-or-later
//
// Module: chdr_crossbar_nxn
// Description:
// This module implements a full-bandwidth NxN crossbar with N input and output ports
// for CHDR traffic. It supports multiple optimization strategies for performance,
// area and timing tradeoffs. It uses AXI-Stream for all of its links. The crossbar
// has a dynamic routing table based on a Content Addressable Memory (CAM). The SID
// is used to determine the destination of a packet and the routing table contains
// a re-programmable SID to crossbar port mapping. The table is programmed using
// special route config packets on the data input ports or using an optional
// management port.
// The topology, routing algorithms and the router architecture is
// described in README.md in this directory.
// Parameters:
// - CHDR_W: Width of the AXI-Stream data bus
// - NPORTS: Number of ports to instantiate
// - DEFAULT_PORT: The failsafe port to forward a packet to is SID mapping is missing
// - MTU: log2 of max packet size (in words)
// - ROUTE_TBL_SIZE: log2 of the number of mappings that the routing table can hold
// at any time. Mapping values are maintained in a FIFO fashion.
// - MUX_ALLOC: Algorithm to allocate the egress MUX
// * PRIO: Priority based. Lower port numbers have a higher priority
// * ROUND-ROBIN: Round robin input port allocation
// - OPTIMIZE: Optimization strategy for performance vs area vs timing tradeoffs
// * AREA: Attempt to minimize area at the cost of performance (throughput) and/or timing
// * PERFORMANCE: Attempt to maximize performance at the cost of area and/or timing
// * TIMING: Attempt to maximize Fmax at the cost of area and/or performance
// - NPORTS_MGMT: Number of ports with management endpoint. The first NPORTS_MGMT ports will
// have the management port instantiated
// - EXT_RTCFG_PORT: Enable a side-channel AXI-Stream management port to configure the
// routing table
// Signals:
// - s_axis_*: Slave port for router (flattened)
// - m_axis_*: Master port for router (flattened)
// - s_axis_mgmt_*: Management slave port
// - device_id: The ID of the device that has instantiated this module
//
module chdr_crossbar_nxn #(
parameter [15:0] PROTOVER = {8'd1, 8'd0},
parameter CHDR_W = 64,
parameter [7:0] NPORTS = 8,
parameter [7:0] DEFAULT_PORT = 0,
parameter MTU = 10,
parameter ROUTE_TBL_SIZE = 6,
parameter MUX_ALLOC = "ROUND-ROBIN",
parameter OPTIMIZE = "AREA",
parameter [7:0] NPORTS_MGMT = NPORTS,
parameter [0:0] EXT_RTCFG_PORT = 0
) (
input wire clk,
input wire reset,
// Device info
input wire [15:0] device_id,
// Inputs
input wire [(CHDR_W*NPORTS)-1:0] s_axis_tdata,
input wire [NPORTS-1:0] s_axis_tlast,
input wire [NPORTS-1:0] s_axis_tvalid,
output wire [NPORTS-1:0] s_axis_tready,
// Output
output wire [(CHDR_W*NPORTS)-1:0] m_axis_tdata,
output wire [NPORTS-1:0] m_axis_tlast,
output wire [NPORTS-1:0] m_axis_tvalid,
input wire [NPORTS-1:0] m_axis_tready,
// Router config management port
input wire ext_rtcfg_stb,
input wire [15:0] ext_rtcfg_addr,
input wire [31:0] ext_rtcfg_data,
output wire ext_rtcfg_ack
);
// ---------------------------------------------------
// RFNoC Includes
// ---------------------------------------------------
`include "../core/rfnoc_chdr_utils.vh"
`include "../core/rfnoc_chdr_internal_utils.vh"
localparam NPORTS_W = $clog2(NPORTS);
localparam EPID_W = 16;
localparam [17:0] EXT_INFO = {1'b0, EXT_RTCFG_PORT, NPORTS_MGMT, NPORTS};
localparam [0:0] PKT_ST_HEAD = 1'b0;
localparam [0:0] PKT_ST_BODY = 1'b1;
// The compute_mux_alloc function is the switch allocation function for the MUX
// i.e. it chooses which input port reserves the output MUX for packet transfer.
function [NPORTS_W-1:0] compute_mux_alloc;
input [NPORTS-1:0] pkt_waiting;
input [NPORTS_W-1:0] last_alloc;
reg signed [NPORTS_W:0] i;
begin
compute_mux_alloc = last_alloc;
for (i = NPORTS-1; i >= 0; i=i-1) begin
if (MUX_ALLOC == "PRIO") begin
// Priority. Lower port index gets a higher priority.
if (pkt_waiting[i])
compute_mux_alloc = i;
end else begin
// Round-robin
if (pkt_waiting[(last_alloc + i + 1) % NPORTS])
compute_mux_alloc = (last_alloc + i + 1) % NPORTS;
end
end
end
endfunction
wire [NPORTS-1:0] rtcfg_req_wr;
wire [(16*NPORTS)-1:0] rtcfg_req_addr;
wire [(32*NPORTS)-1:0] rtcfg_req_data;
wire [NPORTS-1:0] rtcfg_resp_ack;
wire [(EPID_W*NPORTS)-1:0] find_tdata;
wire [NPORTS-1:0] find_tvalid;
wire [NPORTS-1:0] find_tready;
wire [(NPORTS_W*NPORTS)-1:0] result_tdata;
wire [NPORTS-1:0] result_tkeep;
wire [NPORTS-1:0] result_tvalid;
wire [NPORTS-1:0] result_tready;
// Instantiate a single CAM-based routing table that will be shared between all
// input ports. Configuration and lookup is performed using an AXI-Stream iface.
// If multiple packets arrive simultaneously, only the headers of those packets will
// be serialized in order to arbitrate this map. Selection is done round-robin.
chdr_xb_routing_table #(
.SIZE(ROUTE_TBL_SIZE), .NPORTS(NPORTS),
.EXT_INS_PORT_EN(EXT_RTCFG_PORT)
) routing_tbl_i (
.clk (clk ),
.reset (reset ),
.port_req_wr (rtcfg_req_wr ),
.port_req_addr (rtcfg_req_addr),
.port_req_data (rtcfg_req_data),
.port_resp_ack (rtcfg_resp_ack),
.ext_req_wr (ext_rtcfg_stb ),
.ext_req_addr (ext_rtcfg_addr),
.ext_req_data (ext_rtcfg_data),
.ext_resp_ack (ext_rtcfg_ack ),
.axis_find_tdata (find_tdata ),
.axis_find_tvalid (find_tvalid ),
.axis_find_tready (find_tready ),
.axis_result_tdata (result_tdata ),
.axis_result_tkeep (result_tkeep ),
.axis_result_tvalid(result_tvalid ),
.axis_result_tready(result_tready )
);
wire [CHDR_W-1:0] i_tdata [0:NPORTS-1];
wire [9:0] i_tdest [0:NPORTS-1];
wire [1:0] i_tid [0:NPORTS-1];
wire i_tlast [0:NPORTS-1];
wire i_tvalid [0:NPORTS-1];
wire i_tready [0:NPORTS-1];
wire [CHDR_W-1:0] buf_tdata [0:NPORTS-1];
wire [NPORTS_W-1:0] buf_tdest [0:NPORTS-1], buf_tdest_tmp[0:NPORTS-1];
wire buf_tkeep [0:NPORTS-1];
wire buf_tlast [0:NPORTS-1];
wire buf_tvalid[0:NPORTS-1];
wire buf_tready[0:NPORTS-1];
wire [CHDR_W-1:0] swi_tdata [0:NPORTS-1];
wire [NPORTS_W-1:0] swi_tdest [0:NPORTS-1];
wire swi_tlast [0:NPORTS-1];
wire swi_tvalid[0:NPORTS-1];
wire swi_tready[0:NPORTS-1];
wire [(CHDR_W*NPORTS)-1:0] swo_tdata [0:NPORTS-1], muxi_tdata [0:NPORTS-1];
wire [NPORTS-1:0] swo_tlast [0:NPORTS-1], muxi_tlast [0:NPORTS-1];
wire [NPORTS-1:0] swo_tvalid[0:NPORTS-1], muxi_tvalid[0:NPORTS-1];
wire [NPORTS-1:0] swo_tready[0:NPORTS-1], muxi_tready[0:NPORTS-1];
genvar n, i, j;
generate
for (n = 0; n < NPORTS; n = n + 1) begin: i_ports
// For each input port, first check if we have a management packet
// arriving. If it arrives, the top config commands are extrated, sent to the
// routing table for configuration, and the rest of the packet is forwarded
// down to the router.
// the router.
if (n < NPORTS_MGMT) begin
chdr_mgmt_pkt_handler #(
.PROTOVER(PROTOVER), .CHDR_W(CHDR_W), .MGMT_ONLY(0)
) mgmt_ep_i (
.clk (clk ),
.rst (reset ),
.node_info (chdr_mgmt_build_node_info(EXT_INFO, n, NODE_TYPE_XBAR, device_id)),
.s_axis_chdr_tdata (s_axis_tdata [(n*CHDR_W)+:CHDR_W] ),
.s_axis_chdr_tlast (s_axis_tlast [n] ),
.s_axis_chdr_tvalid (s_axis_tvalid[n] ),
.s_axis_chdr_tready (s_axis_tready[n] ),
.s_axis_chdr_tuser ('d0 ),
.m_axis_chdr_tdata (i_tdata [n] ),
.m_axis_chdr_tdest (i_tdest [n] ),
.m_axis_chdr_tid (i_tid [n] ),
.m_axis_chdr_tlast (i_tlast [n] ),
.m_axis_chdr_tvalid (i_tvalid [n] ),
.m_axis_chdr_tready (i_tready [n] ),
.ctrlport_req_wr (rtcfg_req_wr [n] ),
.ctrlport_req_rd (/* unused */ ),
.ctrlport_req_addr (rtcfg_req_addr[(n*16)+:16] ),
.ctrlport_req_data (rtcfg_req_data[(n*32)+:32] ),
.ctrlport_resp_ack (rtcfg_resp_ack[n] ),
.ctrlport_resp_data (32'h0 /* unused */ ),
.op_stb (/* unused */ ),
.op_dst_epid (/* unused */ ),
.op_src_epid (/* unused */ ),
.op_data (/* unused */ )
);
end else begin
assign i_tdata [n] = s_axis_tdata [(n*CHDR_W)+:CHDR_W];
assign i_tid [n] = CHDR_MGMT_ROUTE_EPID;
assign i_tdest [n] = 10'd0; // Unused
assign i_tlast [n] = s_axis_tlast [n];
assign i_tvalid [n] = s_axis_tvalid[n];
assign s_axis_tready[n] = i_tready [n];
assign rtcfg_req_wr [n] = 1'b0;
assign rtcfg_req_addr[(n*16)+:16] = 16'h0;
assign rtcfg_req_data[(n*32)+:32] = 32'h0;
end
// Ingress buffer module that does the following:
// - Stores and gates an incoming packet
// - Looks up destination in routing table and attaches a tdest for the packet
chdr_xb_ingress_buff #(
.WIDTH(CHDR_W), .MTU(MTU), .DEST_W(NPORTS_W), .NODE_ID(n)
) buf_i (
.clk (clk ),
.reset (reset ),
.s_axis_chdr_tdata (i_tdata [n] ),
.s_axis_chdr_tdest (i_tdest [n][NPORTS_W-1:0] ),
.s_axis_chdr_tid (i_tid [n] ),
.s_axis_chdr_tlast (i_tlast [n] ),
.s_axis_chdr_tvalid (i_tvalid [n] ),
.s_axis_chdr_tready (i_tready [n] ),
.m_axis_chdr_tdata (buf_tdata [n] ),
.m_axis_chdr_tdest (buf_tdest_tmp[n] ),
.m_axis_chdr_tkeep (buf_tkeep [n] ),
.m_axis_chdr_tlast (buf_tlast [n] ),
.m_axis_chdr_tvalid (buf_tvalid [n] ),
.m_axis_chdr_tready (buf_tready [n] ),
.m_axis_find_tdata (find_tdata [(n*EPID_W)+:EPID_W] ),
.m_axis_find_tvalid (find_tvalid [n] ),
.m_axis_find_tready (find_tready [n] ),
.s_axis_result_tdata (result_tdata [(n*NPORTS_W)+:NPORTS_W]),
.s_axis_result_tkeep (result_tkeep [n] ),
.s_axis_result_tvalid(result_tvalid[n] ),
.s_axis_result_tready(result_tready[n] )
);
assign buf_tdest[n] = buf_tkeep[n] ? buf_tdest_tmp[n] : DEFAULT_PORT[NPORTS_W-1:0];
// Pipeline state
axi_fifo #(
.WIDTH(CHDR_W+1+NPORTS_W), .SIZE(1)
) pipe_i (
.clk (clk ),
.reset (reset ),
.clear (1'b0 ),
.i_tdata ({buf_tlast[n], buf_tdest[n], buf_tdata[n]}),
.i_tvalid (buf_tvalid[n] ),
.i_tready (buf_tready[n] ),
.o_tdata ({swi_tlast[n], swi_tdest[n], swi_tdata[n]}),
.o_tvalid (swi_tvalid[n] ),
.o_tready (swi_tready[n] ),
.space (/* Unused */ ),
.occupied (/* Unused */ )
);
// Ingress demux. Use the tdest field to determine packet destination
axis_switch #(
.DATA_W(CHDR_W), .DEST_W(1), .IN_PORTS(1), .OUT_PORTS(NPORTS), .PIPELINE(1)
) demux_i (
.clk (clk ),
.reset (reset ),
.s_axis_tdata (swi_tdata [n] ),
.s_axis_tdest ({1'b0, swi_tdest [n]}),
.s_axis_tlast (swi_tlast [n] ),
.s_axis_tvalid (swi_tvalid[n] ),
.s_axis_tready (swi_tready[n] ),
.s_axis_alloc (1'b0 ),
.m_axis_tdata (swo_tdata [n] ),
.m_axis_tdest (/* Unused */ ),
.m_axis_tlast (swo_tlast [n] ),
.m_axis_tvalid (swo_tvalid[n] ),
.m_axis_tready (swo_tready[n] )
);
end
for (i = 0; i < NPORTS; i = i + 1) begin
for (j = 0; j < NPORTS; j = j + 1) begin
assign muxi_tdata [i][j*CHDR_W+:CHDR_W] = swo_tdata [j][i*CHDR_W+:CHDR_W];
assign muxi_tlast [i][j] = swo_tlast [j][i];
assign muxi_tvalid[i][j] = swo_tvalid [j][i];
assign swo_tready [i][j] = muxi_tready[j][i];
end
end
for (n = 0; n < NPORTS; n = n + 1) begin: o_ports
if (OPTIMIZE == "PERFORMANCE") begin
// Use the axis_switch module when optimizing for performance
// This logic has some extra levels of logic to ensure
// that the switch allocation happens in 0 clock cycles which
// means that Fmax for this implementation will be lower.
wire mux_ready = |muxi_tready[n]; // Max 1 bit should be high
wire mux_valid = |muxi_tvalid[n];
wire mux_last = |(muxi_tvalid[n] & muxi_tlast[n]);
// Track the input packet state
reg [0:0] pkt_state = PKT_ST_HEAD;
always @(posedge clk) begin
if (reset) begin
pkt_state <= PKT_ST_HEAD;
end else if (mux_valid & mux_ready) begin
pkt_state <= mux_last ? PKT_ST_HEAD : PKT_ST_BODY;
end
end
// The switch requires the allocation to stay valid until the
// end of the packet. We also might need to keep the previous
// packet's allocation to compute the current one
reg [NPORTS_W-1:0] prev_sw_alloc = {NPORTS_W{1'b0}};
reg [NPORTS_W-1:0] pkt_sw_alloc = {NPORTS_W{1'b0}};
wire [NPORTS_W-1:0] muxi_sw_alloc = (mux_valid && pkt_state == PKT_ST_HEAD) ?
compute_mux_alloc(muxi_tvalid[n], prev_sw_alloc) : pkt_sw_alloc;
always @(posedge clk) begin
if (reset) begin
prev_sw_alloc <= {NPORTS_W{1'b0}};
pkt_sw_alloc <= {NPORTS_W{1'b0}};
end else if (mux_valid & mux_ready) begin
if (pkt_state == PKT_ST_HEAD)
pkt_sw_alloc <= muxi_sw_alloc;
if (mux_last)
prev_sw_alloc <= muxi_sw_alloc;
end
end
axis_switch #(
.DATA_W(CHDR_W), .DEST_W(1), .IN_PORTS(NPORTS), .OUT_PORTS(1),
.PIPELINE(0)
) mux_i (
.clk (clk ),
.reset (reset ),
.s_axis_tdata (muxi_tdata [n] ),
.s_axis_tdest ({NPORTS{1'b0}} /* Unused */ ),
.s_axis_tlast (muxi_tlast [n] ),
.s_axis_tvalid (muxi_tvalid[n] ),
.s_axis_tready (muxi_tready[n] ),
.s_axis_alloc (muxi_sw_alloc ),
.m_axis_tdata (m_axis_tdata [(n*CHDR_W)+:CHDR_W]),
.m_axis_tdest (/* Unused */ ),
.m_axis_tlast (m_axis_tlast [n] ),
.m_axis_tvalid (m_axis_tvalid[n] ),
.m_axis_tready (m_axis_tready[n] )
);
end else begin
// axi_mux has an additional bubble cycle but the logic
// to allocate an input port has fewer levels and takes
// up fewer resources.
axi_mux #(
.PRIO(MUX_ALLOC == "PRIO"), .WIDTH(CHDR_W), .SIZE(NPORTS),
.PRE_FIFO_SIZE(OPTIMIZE == "TIMING" ? 1 : 0), .POST_FIFO_SIZE(1)
) mux_i (
.clk (clk ),
.reset (reset ),
.clear (1'b0 ),
.i_tdata (muxi_tdata [n] ),
.i_tlast (muxi_tlast [n] ),
.i_tvalid (muxi_tvalid [n] ),
.i_tready (muxi_tready [n] ),
.o_tdata (m_axis_tdata [(n*CHDR_W)+:CHDR_W]),
.o_tlast (m_axis_tlast [n] ),
.o_tvalid (m_axis_tvalid[n] ),
.o_tready (m_axis_tready[n] )
);
end
end
endgenerate
endmodule
|