Line 7... |
Line 7... |
Authors:
|
Authors:
|
Wei Song wsong83@gmail.com
|
Wei Song wsong83@gmail.com
|
|
|
License: LGPL 3.0 or later
|
License: LGPL 3.0 or later
|
|
|
Data crossbar for wormhole and SDM routers.
|
Unidirectional crossbar using the XY routing algorithm.
|
Optimized by removing disabled turn models according to the XY routing algorithm.
|
|
*** SystemVerilog is used ***
|
*** SystemVerilog is used ***
|
|
|
History:
|
History:
|
21/08/2009 Initial version. <wsong83@gmail.com>
|
09/07/2009 Initial version. <wsong83@gmail.com>
|
20/09/2010 Supporting channel slicing and SDM using macro difinitions. <wsong83@gmail.com>
|
|
23/05/2011 Clean up for opensource. <wsong83@gmail.com>
|
|
22/06/2011 Prepare to support buffered Clos. <wsong83@gmail.com>
|
|
|
|
*/
|
*/
|
|
|
// the router structure definitions
|
// the router structure definitions
|
`include "define.v"
|
|
|
|
module dcb_xy (
|
module cb_xy (/*AUTOARG*/
|
// Outputs
|
// Outputs
|
so0, so1, so2, so3, wo0, wo1, wo2, wo3, no0, no1, no2, no3, eo0,
|
so, wo, no, eo, lo,
|
eo1, eo2, eo3, lo0, lo1, lo2, lo3, so4, sia, wo4, wia, no4, nia,
|
|
eo4, eia, lo4, lia,
|
|
// Inputs
|
// Inputs
|
si0, si1, si2, si3, wi0, wi1, wi2, wi3, ni0, ni1, ni2, ni3, ei0,
|
si, wi, ni, ei, li, scfg, ncfg, wcfg, ecfg, lcfg
|
ei1, ei2, ei3, li0, li1, li2, li3, si4, soa, wi4, woa, ni4, noa,
|
|
ei4, eoa, li4, loa, scfg, ncfg, wcfg, ecfg, lcfg
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
, soa4, woa4, noa4, eoa4, loa4
|
|
`endif
|
|
) ;
|
) ;
|
|
|
parameter VCN = 1; // number of virtual circuits per port
|
parameter DW = 8; // the wire count of the crossbar
|
parameter VCW = 8; // the datawidth of a single virtual circuit
|
|
parameter SCN = VCW/2; // number of 1-of-4 sub-channels in one virtual circuit
|
input [DW-1:0] si, wi, ni, ei, li; // data input
|
|
output [DW-1:0] so, wo, no, eo, lo; // data output
|
input [VCN-1:0][SCN-1:0] si0, si1, si2, si3; // south input, X+1
|
input [1:0] scfg, ncfg; // configuration
|
output [VCN-1:0][SCN-1:0] so0, so1, so2, so3; // south output
|
input [3:0] wcfg, ecfg, lcfg; // configuration
|
input [VCN-1:0][SCN-1:0] wi0, wi1, wi2, wi3; // west input, Y-1
|
|
output [VCN-1:0][SCN-1:0] wo0, wo1, wo2, wo3; // west output
|
|
input [VCN-1:0][SCN-1:0] ni0, ni1, ni2, ni3; // north input, X-1
|
|
output [VCN-1:0][SCN-1:0] no0, no1, no2, no3; // north output
|
|
input [VCN-1:0][SCN-1:0] ei0, ei1, ei2, ei3; // east input, Y+1
|
|
output [VCN-1:0][SCN-1:0] eo0, eo1, eo2, eo3; // east output
|
|
input [VCN-1:0][SCN-1:0] li0, li1, li2, li3; // local input
|
|
output [VCN-1:0][SCN-1:0] lo0, lo1, lo2, lo3; // local output
|
|
|
|
// ack and eof bits
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
input [VCN-1:0][SCN-1:0] si4, soa;
|
|
output [VCN-1:0][SCN-1:0] so4, sia;
|
|
input [VCN-1:0][SCN-1:0] wi4, woa;
|
|
output [VCN-1:0][SCN-1:0] wo4, wia;
|
|
input [VCN-1:0][SCN-1:0] ni4, noa;
|
|
output [VCN-1:0][SCN-1:0] no4, nia;
|
|
input [VCN-1:0][SCN-1:0] ei4, eoa;
|
|
output [VCN-1:0][SCN-1:0] eo4, eia;
|
|
input [VCN-1:0][SCN-1:0] li4, loa;
|
|
output [VCN-1:0][SCN-1:0] lo4, lia;
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
input [VCN-1:0][SCN-1:0] soa4, woa4, noa4, eoa4, loa4;
|
|
`endif
|
|
`else // !`ifdef ENABLE_CHANNEL_SLICING
|
|
input [VCN-1:0] si4, soa;
|
|
output [VCN-1:0] so4, sia;
|
|
input [VCN-1:0] wi4, woa;
|
|
output [VCN-1:0] wo4, wia;
|
|
input [VCN-1:0] ni4, noa;
|
|
output [VCN-1:0] no4, nia;
|
|
input [VCN-1:0] ei4, eoa;
|
|
output [VCN-1:0] eo4, eia;
|
|
input [VCN-1:0] li4, loa;
|
|
output [VCN-1:0] lo4, lia;
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
input [VCN-1:0] soa4, woa4, noa4, eoa4, loa4;
|
|
`endif
|
|
`endif
|
|
|
|
// configurations
|
|
input [VCN-1:0][1:0][VCN-1:0] scfg, ncfg;
|
|
input [VCN-1:0][3:0][VCN-1:0] wcfg, ecfg, lcfg;
|
|
|
|
// ANDed wires
|
// ANDed wires
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] tos0, tos1, tos2, tos3; // the wires to the south output port
|
wire [DW-1:0][1:0] tos, ton;
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] tow0, tow1, tow2, tow3; // the wires to the west output port
|
wire [DW-1:0][3:0] tow, toe, tol;
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] ton0, ton1, ton2, ton3; // the wires to the north output port
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] toe0, toe1, toe2, toe3; // the wires to the east output port
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] tol0, tol1, tol2, tol3; // the wires to the local output port
|
|
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] tos4, tosa; // the wires to the south output port
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] tow4, towa; // the wires to the west output port
|
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] ton4, tona; // the wires to the north output port
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] toe4, toea; // the wires to the east output port
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] tol4, tola; // the wires to the local output port
|
|
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] isa; // ack back to south
|
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] iwa; // ack back to west
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] ina; // ack back to north
|
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] iea; // ack back to east
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] ila; // ack back to local
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
wire [VCN-1:0][SCN-1:0][1:0][VCN-1:0] tosad, tosa4, tonad, tona4;
|
|
wire [VCN-1:0][SCN-1:0][3:0][VCN-1:0] towad, towa4, toead, toea4, tolad, tola4;
|
|
`endif
|
|
`else // !`ifdef ENABLE_CHANNEL_SLICING
|
|
wire [VCN-1:0][1:0][VCN-1:0] tos4, tosa; // the wires to the south output port
|
|
wire [VCN-1:0][3:0][VCN-1:0] tow4, towa; // the wires to the west output port
|
|
wire [VCN-1:0][1:0][VCN-1:0] ton4, tona; // the wires to the north output port
|
|
wire [VCN-1:0][3:0][VCN-1:0] toe4, toea; // the wires to the east output port
|
|
wire [VCN-1:0][3:0][VCN-1:0] tol4, tola; // the wires to the local output port
|
|
|
|
wire [VCN-1:0][3:0][VCN-1:0] isa; // ack back to south
|
|
wire [VCN-1:0][1:0][VCN-1:0] iwa; // ack back to west
|
|
wire [VCN-1:0][3:0][VCN-1:0] ina; // ack back to north
|
|
wire [VCN-1:0][1:0][VCN-1:0] iea; // ack back to east
|
|
wire [VCN-1:0][3:0][VCN-1:0] ila; // ack back to local
|
|
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
wire [VCN-1:0][1:0][VCN-1:0] tosad, tosa4, tonad, tona4;
|
|
wire [VCN-1:0][3:0][VCN-1:0] towad, towa4, toead, toea4, tolad, tola4;
|
|
`endif
|
|
`endif // !`ifdef ENABLE_CHANNEL_SLICING
|
|
|
|
// generate
|
// generate
|
genvar i, j, k;
|
genvar i, j;
|
|
|
|
|
/*---------------------------- SOUTH OUPUT -------------------------------------*/
|
|
generate for (i=0; i<VCN; i=i+1)
|
|
begin:SOP
|
|
for(j=0; j<VCN; j++) begin: V
|
|
for(k=0; k<SCN; k++) begin: SC
|
|
and AN0 (tos0[i][k][0][j], ni0[j][k], scfg[i][0][j]);
|
|
and AN1 (tos1[i][k][0][j], ni1[j][k], scfg[i][0][j]);
|
|
and AN2 (tos2[i][k][0][j], ni2[j][k], scfg[i][0][j]);
|
|
and AN3 (tos3[i][k][0][j], ni3[j][k], scfg[i][0][j]);
|
|
and AL0 (tos0[i][k][1][j], li0[j][k], scfg[i][1][j]);
|
|
and AL1 (tos1[i][k][1][j], li1[j][k], scfg[i][1][j]);
|
|
and AL2 (tos2[i][k][1][j], li2[j][k], scfg[i][1][j]);
|
|
and AL3 (tos3[i][k][1][j], li3[j][k], scfg[i][1][j]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
and AN4 (tos4[i][k][0][j], ni4[j][k], scfg[i][0][j]);
|
|
and AL4 (tos4[i][k][1][j], li4[j][k], scfg[i][1][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ALAd (tosad[i][k][1][j], soa[i][k], scfg[i][1][j]);
|
|
and ANAd (tosad[i][k][0][j], soa[i][k], scfg[i][0][j]);
|
|
c2 ALA4 (.q(tosa4[i][k][1][j]), .a0(soa4[i][k]), .a1(scfg[i][1][j]));
|
|
c2 ANA4 (.q(tosa4[i][k][0][j]), .a0(soa4[i][k]), .a1(scfg[i][0][j]));
|
|
assign tosa[i][k][1][j] = tosad[i][k][1][j] | tosa4[i][k][1][j];
|
|
assign tosa[i][k][0][j] = tosad[i][k][0][j] | tosa4[i][k][0][j];
|
|
`else
|
|
and ALA (tosa[i][k][1][j], soa[i][k], scfg[i][1][j]);
|
|
and ANA (tosa[i][k][0][j], soa[i][k], scfg[i][0][j]);
|
|
`endif
|
|
`endif
|
|
end // block: SC
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
and AN4 (tos4[i][0][j], ni4[j], scfg[i][0][j]);
|
|
and AL4 (tos4[i][1][j], li4[j], scfg[i][1][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ANAd (tosad[i][0][j], soa[i], scfg[i][0][j]);
|
|
and ALAd (tosad[i][1][j], soa[i], scfg[i][1][j]);
|
|
c2 ANA4 (.q(tosa4[i][0][j]), .a0(soa4[i]), .a1(scfg[i][0][j]));
|
|
c2 ALA4 (.q(tosa4[i][1][j]), .a0(soa4[i]), .a1(scfg[i][1][j]));
|
|
assign tosa[i][0][j] = tosad[i][0][j] | tosa4[i][0][j];
|
|
assign tosa[i][1][j] = tosad[i][0][j] | tosa4[i][1][j];
|
|
`else
|
|
and ANA (tosa[i][0][j], soa[i], scfg[i][0][j]);
|
|
and ALA (tosa[i][1][j], soa[i], scfg[i][1][j]);
|
|
`endif
|
|
`endif
|
|
end // block: V
|
|
|
|
for(k=0; k<SCN; k++) begin: SCOR
|
|
assign so0[i][k] = |(tos0[i][k][0]|tos0[i][k][1]);
|
|
assign so1[i][k] = |(tos1[i][k][0]|tos1[i][k][1]);
|
|
assign so2[i][k] = |(tos2[i][k][0]|tos2[i][k][1]);
|
|
assign so3[i][k] = |(tos3[i][k][0]|tos3[i][k][1]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
assign so4[i][k] = |(tos4[i][k][0]|tos4[i][k][1]);
|
|
`endif
|
|
end
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
assign so4[i] = |(tos4[i][0]|tos4[i][1]);
|
|
`endif
|
|
end
|
|
endgenerate
|
|
|
|
/*---------------------------- WEST OUPUT -------------------------------------*/
|
|
generate for (i=0; i<VCN; i=i+1)
|
|
begin:WOP
|
|
for(j=0; j<VCN; j++) begin: V
|
|
for(k=0; k<SCN; k++) begin: SC
|
|
and AS0 (tow0[i][k][0][j], si0[j][k], wcfg[i][0][j]);
|
|
and AS1 (tow1[i][k][0][j], si1[j][k], wcfg[i][0][j]);
|
|
and AS2 (tow2[i][k][0][j], si2[j][k], wcfg[i][0][j]);
|
|
and AS3 (tow3[i][k][0][j], si3[j][k], wcfg[i][0][j]);
|
|
and AN0 (tow0[i][k][1][j], ni0[j][k], wcfg[i][1][j]);
|
|
and AN1 (tow1[i][k][1][j], ni1[j][k], wcfg[i][1][j]);
|
|
and AN2 (tow2[i][k][1][j], ni2[j][k], wcfg[i][1][j]);
|
|
and AN3 (tow3[i][k][1][j], ni3[j][k], wcfg[i][1][j]);
|
|
and AE0 (tow0[i][k][2][j], ei0[j][k], wcfg[i][2][j]);
|
|
and AE1 (tow1[i][k][2][j], ei1[j][k], wcfg[i][2][j]);
|
|
and AE2 (tow2[i][k][2][j], ei2[j][k], wcfg[i][2][j]);
|
|
and AE3 (tow3[i][k][2][j], ei3[j][k], wcfg[i][2][j]);
|
|
and AL0 (tow0[i][k][3][j], li0[j][k], wcfg[i][3][j]);
|
|
and AL1 (tow1[i][k][3][j], li1[j][k], wcfg[i][3][j]);
|
|
and AL2 (tow2[i][k][3][j], li2[j][k], wcfg[i][3][j]);
|
|
and AL3 (tow3[i][k][3][j], li3[j][k], wcfg[i][3][j]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
and AS4 (tow4[i][k][0][j], si4[j][k], wcfg[i][0][j]);
|
|
and AN4 (tow4[i][k][1][j], ni4[j][k], wcfg[i][1][j]);
|
|
and AE4 (tow4[i][k][2][j], ei4[j][k], wcfg[i][2][j]);
|
|
and AL4 (tow4[i][k][3][j], li4[j][k], wcfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (towad[i][k][0][j], woa[i][k], wcfg[i][0][j]);
|
|
and ANAd (towad[i][k][1][j], woa[i][k], wcfg[i][1][j]);
|
|
and AEAd (towad[i][k][2][j], woa[i][k], wcfg[i][2][j]);
|
|
and ALAd (towad[i][k][3][j], woa[i][k], wcfg[i][3][j]);
|
|
c2 ASA4 (.q(towa4[i][k][0][j]), .a0(woa4[i][k]), .a1(wcfg[i][0][j]));
|
|
c2 ANA4 (.q(towa4[i][k][1][j]), .a0(woa4[i][k]), .a1(wcfg[i][1][j]));
|
|
c2 AEA4 (.q(towa4[i][k][2][j]), .a0(woa4[i][k]), .a1(wcfg[i][2][j]));
|
|
c2 ALA4 (.q(towa4[i][k][3][j]), .a0(woa4[i][k]), .a1(wcfg[i][3][j]));
|
|
assign towa[i][k][0][j] = towad[i][k][0][j] | towa4[i][k][0][j];
|
|
assign towa[i][k][1][j] = towad[i][k][1][j] | towa4[i][k][1][j];
|
|
assign towa[i][k][2][j] = towad[i][k][2][j] | towa4[i][k][2][j];
|
|
assign towa[i][k][3][j] = towad[i][k][3][j] | towa4[i][k][3][j];
|
|
`else
|
|
and ASA (towa[i][k][0][j], woa[i][k], wcfg[i][0][j]);
|
|
and ANA (towa[i][k][1][j], woa[i][k], wcfg[i][1][j]);
|
|
and AEA (towa[i][k][2][j], woa[i][k], wcfg[i][2][j]);
|
|
and ALA (towa[i][k][3][j], woa[i][k], wcfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: SC
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
and AS4 (tow4[i][0][j], si4[j], wcfg[i][0][j]);
|
|
and AN4 (tow4[i][1][j], ni4[j], wcfg[i][1][j]);
|
|
and AE4 (tow4[i][2][j], ei4[j], wcfg[i][2][j]);
|
|
and AL4 (tow4[i][3][j], li4[j], wcfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (towad[i][0][j], woa[i], wcfg[i][0][j]);
|
|
and ANAd (towad[i][1][j], woa[i], wcfg[i][1][j]);
|
|
and AEAd (towad[i][2][j], woa[i], wcfg[i][2][j]);
|
|
and ALAd (towad[i][3][j], woa[i], wcfg[i][3][j]);
|
|
c2 ASA4 (.q(towa4[i][0][j]), .a0(woa4[i]), .a1(wcfg[i][0][j]));
|
|
c2 ANA4 (.q(towa4[i][1][j]), .a0(woa4[i]), .a1(wcfg[i][1][j]));
|
|
c2 AEA4 (.q(towa4[i][2][j]), .a0(woa4[i]), .a1(wcfg[i][2][j]));
|
|
c2 ALA4 (.q(towa4[i][3][j]), .a0(woa4[i]), .a1(wcfg[i][3][j]));
|
|
assign towa[i][0][j] = towad[i][0][j] | towa4[i][0][j];
|
|
assign towa[i][1][j] = towad[i][1][j] | towa4[i][1][j];
|
|
assign towa[i][2][j] = towad[i][2][j] | towa4[i][2][j];
|
|
assign towa[i][3][j] = towad[i][3][j] | towa4[i][3][j];
|
|
`else
|
|
and ASA (towa[i][0][j], woa[i], wcfg[i][0][j]);
|
|
and ANA (towa[i][1][j], woa[i], wcfg[i][1][j]);
|
|
and AEA (towa[i][2][j], woa[i], wcfg[i][2][j]);
|
|
and ALA (towa[i][3][j], woa[i], wcfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: V
|
|
|
|
for(k=0; k<SCN; k++) begin: SCOR
|
|
assign wo0[i][k] = |(tow0[i][k][0]|tow0[i][k][1]|tow0[i][k][2]|tow0[i][k][3]);
|
|
assign wo1[i][k] = |(tow1[i][k][0]|tow1[i][k][1]|tow1[i][k][2]|tow1[i][k][3]);
|
|
assign wo2[i][k] = |(tow2[i][k][0]|tow2[i][k][1]|tow2[i][k][2]|tow2[i][k][3]);
|
|
assign wo3[i][k] = |(tow3[i][k][0]|tow3[i][k][1]|tow3[i][k][2]|tow3[i][k][3]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
assign wo4[i][k] = |(tow4[i][k][0]|tow4[i][k][1]|tow4[i][k][2]|tow4[i][k][3]);
|
|
`endif
|
|
end
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
assign wo4[i] = |(tow4[i][0]|tow4[i][1]|tow4[i][2]|tow4[i][3]);
|
|
`endif
|
|
end
|
|
endgenerate
|
|
|
|
/*---------------------------- NORTH OUPUT -------------------------------------*/
|
|
generate for (i=0; i<VCN; i=i+1)
|
|
begin:NOP
|
|
for(j=0; j<VCN; j++) begin: V
|
|
for(k=0; k<SCN; k++) begin: SC
|
|
and AS0 (ton0[i][k][0][j], si0[j][k], ncfg[i][0][j]);
|
|
and AS1 (ton1[i][k][0][j], si1[j][k], ncfg[i][0][j]);
|
|
and AS2 (ton2[i][k][0][j], si2[j][k], ncfg[i][0][j]);
|
|
and AS3 (ton3[i][k][0][j], si3[j][k], ncfg[i][0][j]);
|
|
and AL0 (ton0[i][k][1][j], li0[j][k], ncfg[i][1][j]);
|
|
and AL1 (ton1[i][k][1][j], li1[j][k], ncfg[i][1][j]);
|
|
and AL2 (ton2[i][k][1][j], li2[j][k], ncfg[i][1][j]);
|
|
and AL3 (ton3[i][k][1][j], li3[j][k], ncfg[i][1][j]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
and AS4 (ton4[i][k][0][j], si4[j][k], ncfg[i][0][j]);
|
|
and AL4 (ton4[i][k][1][j], li4[j][k], ncfg[i][1][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (tonad[i][k][0][j], noa[i][k], ncfg[i][0][j]);
|
|
and ALAd (tonad[i][k][1][j], noa[i][k], ncfg[i][1][j]);
|
|
c2 ASA4 (.q(tona4[i][k][0][j]), .a0(noa4[i][k]), .a1(ncfg[i][0][j]));
|
|
c2 ALA4 (.q(tona4[i][k][1][j]), .a0(noa4[i][k]), .a1(ncfg[i][1][j]));
|
|
assign tona[i][k][0][j] = tonad[i][k][0][j] | tona4[i][k][0][j];
|
|
assign tona[i][k][1][j] = tonad[i][k][1][j] | tona4[i][k][1][j];
|
|
`else
|
|
and ASA (tona[i][k][0][j], noa[i][k], ncfg[i][0][j]);
|
|
and ALA (tona[i][k][1][j], noa[i][k], ncfg[i][1][j]);
|
|
`endif
|
|
`endif
|
|
end // block: SC
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
and AS4 (ton4[i][0][j], si4[j], ncfg[i][0][j]);
|
|
and AL4 (ton4[i][1][j], li4[j], ncfg[i][1][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (tonad[i][0][j], noa[i], ncfg[i][0][j]);
|
|
and ALAd (tonad[i][1][j], noa[i], ncfg[i][1][j]);
|
|
c2 ASA4 (.q(tona4[i][0][j]), .a0(noa4[i]), .a1(ncfg[i][0][j]));
|
|
c2 ALA4 (.q(tona4[i][1][j]), .a0(noa4[i]), .a1(ncfg[i][1][j]));
|
|
assign tona[i][0][j] = tonad[i][0][j] | tona4[i][0][j];
|
|
assign tona[i][1][j] = tonad[i][1][j] | tona4[i][1][j];
|
|
`else
|
|
and ASA (tona[i][0][j], noa[i], ncfg[i][0][j]);
|
|
and ALA (tona[i][1][j], noa[i], ncfg[i][1][j]);
|
|
`endif
|
|
`endif
|
|
end // block: V
|
|
|
|
for(k=0; k<SCN; k++) begin: SCOR
|
|
assign no0[i][k] = |(ton0[i][k][0]|ton0[i][k][1]);
|
|
assign no1[i][k] = |(ton1[i][k][0]|ton1[i][k][1]);
|
|
assign no2[i][k] = |(ton2[i][k][0]|ton2[i][k][1]);
|
|
assign no3[i][k] = |(ton3[i][k][0]|ton3[i][k][1]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
assign no4[i][k] = |(ton4[i][k][0]|ton4[i][k][1]);
|
|
`endif
|
|
end
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
assign no4[i] = |(ton4[i][0]|ton4[i][1]);
|
|
`endif
|
|
end
|
|
endgenerate
|
|
|
|
/*---------------------------- EAST OUPUT -------------------------------------*/
|
|
generate for (i=0; i<VCN; i=i+1)
|
|
begin:EOP
|
|
for(j=0; j<VCN; j++) begin: V
|
|
for(k=0; k<SCN; k++) begin: SC
|
|
and AS0 (toe0[i][k][0][j], si0[j][k], ecfg[i][0][j]);
|
|
and AS1 (toe1[i][k][0][j], si1[j][k], ecfg[i][0][j]);
|
|
and AS2 (toe2[i][k][0][j], si2[j][k], ecfg[i][0][j]);
|
|
and AS3 (toe3[i][k][0][j], si3[j][k], ecfg[i][0][j]);
|
|
and AW0 (toe0[i][k][1][j], wi0[j][k], ecfg[i][1][j]);
|
|
and AW1 (toe1[i][k][1][j], wi1[j][k], ecfg[i][1][j]);
|
|
and AW2 (toe2[i][k][1][j], wi2[j][k], ecfg[i][1][j]);
|
|
and AW3 (toe3[i][k][1][j], wi3[j][k], ecfg[i][1][j]);
|
|
and AN0 (toe0[i][k][2][j], ni0[j][k], ecfg[i][2][j]);
|
|
and AN1 (toe1[i][k][2][j], ni1[j][k], ecfg[i][2][j]);
|
|
and AN2 (toe2[i][k][2][j], ni2[j][k], ecfg[i][2][j]);
|
|
and AN3 (toe3[i][k][2][j], ni3[j][k], ecfg[i][2][j]);
|
|
and AL0 (toe0[i][k][3][j], li0[j][k], ecfg[i][3][j]);
|
|
and AL1 (toe1[i][k][3][j], li1[j][k], ecfg[i][3][j]);
|
|
and AL2 (toe2[i][k][3][j], li2[j][k], ecfg[i][3][j]);
|
|
and AL3 (toe3[i][k][3][j], li3[j][k], ecfg[i][3][j]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
and AS4 (toe4[i][k][0][j], si4[j][k], ecfg[i][0][j]);
|
|
and AW4 (toe4[i][k][1][j], wi4[j][k], ecfg[i][1][j]);
|
|
and AN4 (toe4[i][k][2][j], ni4[j][k], ecfg[i][2][j]);
|
|
and AL4 (toe4[i][k][3][j], li4[j][k], ecfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (toead[i][k][0][j], eoa[i][k], ecfg[i][0][j]);
|
|
and AWAd (toead[i][k][1][j], eoa[i][k], ecfg[i][1][j]);
|
|
and ANAd (toead[i][k][2][j], eoa[i][k], ecfg[i][2][j]);
|
|
and ALAd (toead[i][k][3][j], eoa[i][k], ecfg[i][3][j]);
|
|
c2 ASA4 (.q(toea4[i][k][0][j]), .a0(eoa4[i][k]), .a1(ecfg[i][0][j]));
|
|
c2 AWA4 (.q(toea4[i][k][1][j]), .a0(eoa4[i][k]), .a1(ecfg[i][1][j]));
|
|
c2 ANA4 (.q(toea4[i][k][2][j]), .a0(eoa4[i][k]), .a1(ecfg[i][2][j]));
|
|
c2 ALA4 (.q(toea4[i][k][3][j]), .a0(eoa4[i][k]), .a1(ecfg[i][3][j]));
|
|
assign toea[i][k][0][j] = toead[i][k][0][j] | toea4[i][k][0][j];
|
|
assign toea[i][k][1][j] = toead[i][k][1][j] | toea4[i][k][1][j];
|
|
assign toea[i][k][2][j] = toead[i][k][2][j] | toea4[i][k][2][j];
|
|
assign toea[i][k][3][j] = toead[i][k][3][j] | toea4[i][k][3][j];
|
|
`else
|
|
and ASA (toea[i][k][0][j], eoa[i][k], ecfg[i][0][j]);
|
|
and AWA (toea[i][k][1][j], eoa[i][k], ecfg[i][1][j]);
|
|
and ANA (toea[i][k][2][j], eoa[i][k], ecfg[i][2][j]);
|
|
and ALA (toea[i][k][3][j], eoa[i][k], ecfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: SC
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
and AS4 (toe4[i][0][j], si4[j], ecfg[i][0][j]);
|
|
and AW4 (toe4[i][1][j], wi4[j], ecfg[i][1][j]);
|
|
and AN4 (toe4[i][2][j], ni4[j], ecfg[i][2][j]);
|
|
and AL4 (toe4[i][3][j], li4[j], ecfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (toead[i][0][j], eoa[i], ecfg[i][0][j]);
|
|
and AWAd (toead[i][1][j], eoa[i], ecfg[i][1][j]);
|
|
and ANAd (toead[i][2][j], eoa[i], ecfg[i][2][j]);
|
|
and ALAd (toead[i][3][j], eoa[i], ecfg[i][3][j]);
|
|
c2 ASA4 (.q(toea4[i][0][j]), .a0(eoa4[i]), .a1(ecfg[i][0][j]));
|
|
c2 AWA4 (.q(toea4[i][1][j]), .a0(eoa4[i]), .a1(ecfg[i][1][j]));
|
|
c2 ANA4 (.q(toea4[i][2][j]), .a0(eoa4[i]), .a1(ecfg[i][2][j]));
|
|
c2 ALA4 (.q(toea4[i][3][j]), .a0(eoa4[i]), .a1(ecfg[i][3][j]));
|
|
assign toea[i][0][j] = toead[i][0][j] | toea4[i][0][j];
|
|
assign toea[i][1][j] = toead[i][1][j] | toea4[i][1][j];
|
|
assign toea[i][2][j] = toead[i][2][j] | toea4[i][2][j];
|
|
assign toea[i][3][j] = toead[i][3][j] | toea4[i][3][j];
|
|
`else
|
|
and ASA (toea[i][0][j], eoa[i], ecfg[i][0][j]);
|
|
and AWA (toea[i][1][j], eoa[i], ecfg[i][1][j]);
|
|
and ANA (toea[i][2][j], eoa[i], ecfg[i][2][j]);
|
|
and ALA (toea[i][3][j], eoa[i], ecfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: V
|
|
|
|
for(k=0; k<SCN; k++) begin: SCOR
|
|
assign eo0[i][k] = |(toe0[i][k][0]|toe0[i][k][1]|toe0[i][k][2]|toe0[i][k][3]);
|
|
assign eo1[i][k] = |(toe1[i][k][0]|toe1[i][k][1]|toe1[i][k][2]|toe1[i][k][3]);
|
|
assign eo2[i][k] = |(toe2[i][k][0]|toe2[i][k][1]|toe2[i][k][2]|toe2[i][k][3]);
|
|
assign eo3[i][k] = |(toe3[i][k][0]|toe3[i][k][1]|toe3[i][k][2]|toe3[i][k][3]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
assign eo4[i][k] = |(toe4[i][k][0]|toe4[i][k][1]|toe4[i][k][2]|toe4[i][k][3]);
|
|
`endif
|
|
end
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
assign eo4[i] = |(toe4[i][0]|toe4[i][1]|toe4[i][2]|toe4[i][3]);
|
|
`endif
|
|
end
|
|
endgenerate
|
|
|
|
|
|
/*---------------------------- LOCAL OUPUT -------------------------------------*/
|
|
generate for (i=0; i<VCN; i=i+1)
|
|
begin:LOP
|
|
for(j=0; j<VCN; j++) begin: V
|
|
for(k=0; k<SCN; k++) begin: SC
|
|
and AS0 (tol0[i][k][0][j], si0[j][k], lcfg[i][0][j]);
|
|
and AS1 (tol1[i][k][0][j], si1[j][k], lcfg[i][0][j]);
|
|
and AS2 (tol2[i][k][0][j], si2[j][k], lcfg[i][0][j]);
|
|
and AS3 (tol3[i][k][0][j], si3[j][k], lcfg[i][0][j]);
|
|
and AW0 (tol0[i][k][1][j], wi0[j][k], lcfg[i][1][j]);
|
|
and AW1 (tol1[i][k][1][j], wi1[j][k], lcfg[i][1][j]);
|
|
and AW2 (tol2[i][k][1][j], wi2[j][k], lcfg[i][1][j]);
|
|
and AW3 (tol3[i][k][1][j], wi3[j][k], lcfg[i][1][j]);
|
|
and AN0 (tol0[i][k][2][j], ni0[j][k], lcfg[i][2][j]);
|
|
and AN1 (tol1[i][k][2][j], ni1[j][k], lcfg[i][2][j]);
|
|
and AN2 (tol2[i][k][2][j], ni2[j][k], lcfg[i][2][j]);
|
|
and AN3 (tol3[i][k][2][j], ni3[j][k], lcfg[i][2][j]);
|
|
and AE0 (tol0[i][k][3][j], ei0[j][k], lcfg[i][3][j]);
|
|
and AE1 (tol1[i][k][3][j], ei1[j][k], lcfg[i][3][j]);
|
|
and AE2 (tol2[i][k][3][j], ei2[j][k], lcfg[i][3][j]);
|
|
and AE3 (tol3[i][k][3][j], ei3[j][k], lcfg[i][3][j]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
and AS4 (tol4[i][k][0][j], si4[j][k], lcfg[i][0][j]);
|
|
and AW4 (tol4[i][k][1][j], wi4[j][k], lcfg[i][1][j]);
|
|
and AN4 (tol4[i][k][2][j], ni4[j][k], lcfg[i][2][j]);
|
|
and AE4 (tol4[i][k][3][j], ei4[j][k], lcfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (tolad[i][k][0][j], loa[i][k], lcfg[i][0][j]);
|
|
and AWAd (tolad[i][k][1][j], loa[i][k], lcfg[i][1][j]);
|
|
and ANAd (tolad[i][k][2][j], loa[i][k], lcfg[i][2][j]);
|
|
and AEAd (tolad[i][k][3][j], loa[i][k], lcfg[i][3][j]);
|
|
c2 ASA4 (.q(tola4[i][k][0][j]), .a0(loa4[i][k]), .a1(lcfg[i][0][j]));
|
|
c2 AWA4 (.q(tola4[i][k][1][j]), .a0(loa4[i][k]), .a1(lcfg[i][1][j]));
|
|
c2 ANA4 (.q(tola4[i][k][2][j]), .a0(loa4[i][k]), .a1(lcfg[i][2][j]));
|
|
c2 AEA4 (.q(tola4[i][k][3][j]), .a0(loa4[i][k]), .a1(lcfg[i][3][j]));
|
|
assign tola[i][k][0][j] = tolad[i][k][0][j] | tola4[i][k][0][j];
|
|
assign tola[i][k][1][j] = tolad[i][k][1][j] | tola4[i][k][1][j];
|
|
assign tola[i][k][2][j] = tolad[i][k][2][j] | tola4[i][k][2][j];
|
|
assign tola[i][k][3][j] = tolad[i][k][3][j] | tola4[i][k][3][j];
|
|
`else
|
|
and ASA (tola[i][k][0][j], loa[i][k], lcfg[i][0][j]);
|
|
and AWA (tola[i][k][1][j], loa[i][k], lcfg[i][1][j]);
|
|
and ANA (tola[i][k][2][j], loa[i][k], lcfg[i][2][j]);
|
|
and AEA (tola[i][k][3][j], loa[i][k], lcfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: SC
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
and AS4 (tol4[i][0][j], si4[j], lcfg[i][0][j]);
|
|
and AW4 (tol4[i][1][j], wi4[j], lcfg[i][1][j]);
|
|
and AN4 (tol4[i][2][j], ni4[j], lcfg[i][2][j]);
|
|
and AE4 (tol4[i][3][j], ei4[j], lcfg[i][3][j]);
|
|
`ifdef ENABLE_BUFFERED_CLOS
|
|
and ASAd (tolad[i][0][j], loa[i], lcfg[i][0][j]);
|
|
and AWAd (tolad[i][1][j], loa[i], lcfg[i][1][j]);
|
|
and ANAd (tolad[i][2][j], loa[i], lcfg[i][2][j]);
|
|
and AEAd (tolad[i][3][j], loa[i], lcfg[i][3][j]);
|
|
c2 ASA4 (.q(tola4[i][0][j]), .a0(loa4[i]), .a1(lcfg[i][0][j]));
|
|
c2 AWA4 (.q(tola4[i][1][j]), .a0(loa4[i]), .a1(lcfg[i][1][j]));
|
|
c2 ANA4 (.q(tola4[i][2][j]), .a0(loa4[i]), .a1(lcfg[i][2][j]));
|
|
c2 AEA4 (.q(tola4[i][3][j]), .a0(loa4[i]), .a1(lcfg[i][3][j]));
|
|
assign tola[i][0][j] = tolad[i][0][j] | tola4[i][0][j];
|
|
assign tola[i][1][j] = tolad[i][1][j] | tola4[i][1][j];
|
|
assign tola[i][2][j] = tolad[i][2][j] | tola4[i][2][j];
|
|
assign tola[i][3][j] = tolad[i][3][j] | tola4[i][3][j];
|
|
`else
|
|
and ASA (tola[i][0][j], loa[i], lcfg[i][0][j]);
|
|
and AWA (tola[i][1][j], loa[i], lcfg[i][1][j]);
|
|
and ANA (tola[i][2][j], loa[i], lcfg[i][2][j]);
|
|
and AEA (tola[i][3][j], loa[i], lcfg[i][3][j]);
|
|
`endif
|
|
`endif
|
|
end // block: V
|
|
|
|
for(k=0; k<SCN; k++) begin: SCOR
|
|
assign lo0[i][k] = |(tol0[i][k][0]|tol0[i][k][1]|tol0[i][k][2]|tol0[i][k][3]);
|
|
assign lo1[i][k] = |(tol1[i][k][0]|tol1[i][k][1]|tol1[i][k][2]|tol1[i][k][3]);
|
|
assign lo2[i][k] = |(tol2[i][k][0]|tol2[i][k][1]|tol2[i][k][2]|tol2[i][k][3]);
|
|
assign lo3[i][k] = |(tol3[i][k][0]|tol3[i][k][1]|tol3[i][k][2]|tol3[i][k][3]);
|
|
`ifdef ENABLE_CHANNEL_SLICING
|
|
assign lo4[i][k] = |(tol4[i][k][0]|tol4[i][k][1]|tol4[i][k][2]|tol4[i][k][3]);
|
|
`endif
|
|
end
|
|
`ifndef ENABLE_CHANNEL_SLICING
|
|
assign lo4[i] = |(tol4[i][0]|tol4[i][1]|tol4[i][2]|tol4[i][3]);
|
|
`endif
|
|
end
|
|
endgenerate
|
|
|
|
generate for(i=0; i<VCN; i++) begin: IACK
|
generate for (i=0; i<DW; i=i+1)
|
`ifdef ENABLE_CHANNEL_SLICING
|
begin:OPA
|
for(k=0; k<SCN; k++) begin: SC
|
and AN2S (tos[i][0], ni[i], scfg[0]);
|
for(j=0; j<VCN; j++) begin: SHUFFLE
|
and AL2S (tos[i][1], li[i], scfg[1]);
|
assign isa[i][k][0][j] = towa[j][k][0][i];
|
assign so[i] = |tos[i];
|
assign isa[i][k][1][j] = tona[j][k][0][i];
|
|
assign isa[i][k][2][j] = toea[j][k][0][i];
|
and AS2W (tow[i][0], si[i], wcfg[0]);
|
assign isa[i][k][3][j] = tola[j][k][0][i];
|
and AN2W (tow[i][1], ni[i], wcfg[1]);
|
assign iwa[i][k][0][j] = toea[j][k][1][i];
|
and AE2W (tow[i][2], ei[i], wcfg[2]);
|
assign iwa[i][k][1][j] = tola[j][k][1][i];
|
and AL2W (tow[i][3], li[i], wcfg[3]);
|
assign ina[i][k][0][j] = tosa[j][k][0][i];
|
assign wo[i] = |tow[i];
|
assign ina[i][k][1][j] = towa[j][k][1][i];
|
|
assign ina[i][k][2][j] = toea[j][k][2][i];
|
and AS2N (ton[i][0], si[i], ncfg[0]);
|
assign ina[i][k][3][j] = tola[j][k][2][i];
|
and AL2N (ton[i][1], li[i], ncfg[1]);
|
assign iea[i][k][0][j] = towa[j][k][2][i];
|
assign no[i] = |ton[i];
|
assign iea[i][k][1][j] = tola[j][k][3][i];
|
|
assign ila[i][k][0][j] = tosa[j][k][1][i];
|
and AS2E (toe[i][0], si[i], ecfg[0]);
|
assign ila[i][k][1][j] = towa[j][k][3][i];
|
and AW2E (toe[i][1], wi[i], ecfg[1]);
|
assign ila[i][k][2][j] = tona[j][k][1][i];
|
and AN2E (toe[i][2], ni[i], ecfg[2]);
|
assign ila[i][k][3][j] = toea[j][k][3][i];
|
and AL2E (toe[i][3], li[i], ecfg[3]);
|
end // block: SHUFFLE
|
assign eo[i] = |toe[i];
|
assign sia[i][k] = |{isa[i][k][0]|isa[i][k][1]|isa[i][k][2]|isa[i][k][3]};
|
|
assign wia[i][k] = |{iwa[i][k][0]|iwa[i][k][1]};
|
and AS2L (tol[i][0], si[i], lcfg[0]);
|
assign nia[i][k] = |{ina[i][k][0]|ina[i][k][1]|ina[i][k][2]|ina[i][k][3]};
|
and AW2L (tol[i][1], wi[i], lcfg[1]);
|
assign eia[i][k] = |{iea[i][k][0]|iea[i][k][1]};
|
and AN2L (tol[i][2], ni[i], lcfg[2]);
|
assign lia[i][k] = |{ila[i][k][0]|ila[i][k][1]|ila[i][k][2]|ila[i][k][3]};
|
and AE2L (tol[i][3], ei[i], lcfg[3]);
|
end // block: SC
|
assign lo[i] = |tol[i];
|
`else // !`ifdef ENABLE_CHANNEL_SLICING
|
end // block: OPA
|
for(j=0; j<VCN; j++) begin: SHUFFLE
|
|
assign isa[i][0][j] = towa[j][0][i];
|
|
assign isa[i][1][j] = tona[j][0][i];
|
|
assign isa[i][2][j] = toea[j][0][i];
|
|
assign isa[i][3][j] = tola[j][0][i];
|
|
assign iwa[i][0][j] = toea[j][1][i];
|
|
assign iwa[i][1][j] = tola[j][1][i];
|
|
assign ina[i][0][j] = tosa[j][0][i];
|
|
assign ina[i][1][j] = towa[j][1][i];
|
|
assign ina[i][2][j] = toea[j][2][i];
|
|
assign ina[i][3][j] = tola[j][2][i];
|
|
assign iea[i][0][j] = towa[j][2][i];
|
|
assign iea[i][1][j] = tola[j][3][i];
|
|
assign ila[i][0][j] = tosa[j][1][i];
|
|
assign ila[i][1][j] = towa[j][3][i];
|
|
assign ila[i][2][j] = tona[j][1][i];
|
|
assign ila[i][3][j] = toea[j][3][i];
|
|
end // block: SHUFFLE
|
|
assign sia[i] = |{isa[i][0]|isa[i][1]|isa[i][2]|isa[i][3]};
|
|
assign wia[i] = |{iwa[i][0]|iwa[i][1]};
|
|
assign nia[i] = |{ina[i][0]|ina[i][1]|ina[i][2]|ina[i][3]};
|
|
assign eia[i] = |{iea[i][0]|iea[i][1]};
|
|
assign lia[i] = |{ila[i][0]|ila[i][1]|ila[i][2]|ila[i][3]};
|
|
`endif
|
|
end // block: IACK
|
|
endgenerate
|
endgenerate
|
|
|
endmodule // dcb_xy
|
endmodule // cb_xy
|
|
|
|
|
|
|
|
|
No newline at end of file
|
No newline at end of file
|