URL
https://opencores.org/ocsvn/async_sdm_noc/async_sdm_noc/trunk
Subversion Repositories async_sdm_noc
[/] [async_sdm_noc/] [branches/] [init/] [common/] [src/] [dclos.v] - Rev 29
Go to most recent revision | Compare with Previous | Blame | View Log
/* Asynchronous SDM NoC (C)2011 Wei Song Advanced Processor Technologies Group Computer Science, the Univ. of Manchester, UK Authors: Wei Song wsong83@gmail.com License: LGPL 3.0 or later Data Clos network. *** SystemVerilog is used *** History: 17/07/2010 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> */ // the router structure definitions `include "define.v" module dclos (/*AUTOARG*/ // Outputs so0, so1, so2, so3, wo0, wo1, wo2, wo3, no0, no1, no2, no3, eo0, eo1, eo2, eo3, lo0, lo1, lo2, lo3, so4, wo4, no4, eo4, lo4, sia, wia, nia, eia, lia, // Inputs si0, si1, si2, si3, wi0, wi1, wi2, wi3, ni0, ni1, ni2, ni3, ei0, ei1, ei2, ei3, li0, li1, li2, li3, si4, wi4, ni4, ei4, li4, soa, woa, noa, eoa, loa, imcfg, scfg, ncfg, wcfg, ecfg, lcfg ); parameter MN = 2; // number of CMs parameter NN = 2; // number of ports in an IM or OM, equ. to number of virtual circuits parameter DW = 8; // datawidth of a single virtual circuit/port parameter SCN = DW/2; // number of 1-of-4 sub-channels in one port input [NN-1:0][SCN-1:0] si0, si1, si2, si3; // south input [0], X+1 input [NN-1:0][SCN-1:0] wi0, wi1, wi2, wi3; // west input [1], Y-1 input [NN-1:0][SCN-1:0] ni0, ni1, ni2, ni3; // north input [2], X-1 input [NN-1:0][SCN-1:0] ei0, ei1, ei2, ei3; // east input [3], Y+1 input [NN-1:0][SCN-1:0] li0, li1, li2, li3; // local input output [NN-1:0][SCN-1:0] so0, so1, so2, so3; // south output output [NN-1:0][SCN-1:0] wo0, wo1, wo2, wo3; // west output output [NN-1:0][SCN-1:0] no0, no1, no2, no3; // north output output [NN-1:0][SCN-1:0] eo0, eo1, eo2, eo3; // east output output [NN-1:0][SCN-1:0] lo0, lo1, lo2, lo3; // local output // eof bits and ack lines `ifdef ENABLE_CHANNEL_SLICING input [NN-1:0][SCN-1:0] si4, wi4, ni4, ei4, li4; output [NN-1:0][SCN-1:0] so4, wo4, no4, eo4, lo4; output [NN-1:0][SCN-1:0] sia, wia, nia, eia, lia; input [NN-1:0][SCN-1:0] soa, woa, noa, eoa, loa; `else input [NN-1:0] si4, wi4, ni4, ei4, li4; output [NN-1:0] so4, wo4, no4, eo4, lo4; output [NN-1:0] sia, wia, nia, eia, lia; input [NN-1:0] soa, woa, noa, eoa, loa; `endif // !`ifdef ENABLE_CHANNEL_SLICING input [4:0][MN-1:0][NN-1:0] imcfg; // configuration for IMs // configuration for CMs input [MN-1:0][1:0] scfg, ncfg; input [MN-1:0][3:0] wcfg, ecfg, lcfg; // no OMs // output of IMs wire [MN-1:0][SCN-1:0] imos0, imos1, imos2, imos3; wire [MN-1:0][SCN-1:0] imow0, imow1, imow2, imow3; wire [MN-1:0][SCN-1:0] imon0, imon1, imon2, imon3; wire [MN-1:0][SCN-1:0] imoe0, imoe1, imoe2, imoe3; wire [MN-1:0][SCN-1:0] imol0, imol1, imol2, imol3; `ifdef ENABLE_CHANNEL_SLICING wire [MN-1:0][SCN-1:0] imos4, imow4, imon4, imoe4, imol4; wire [MN-1:0][SCN-1:0] imosa, imowa, imona, imoea, imola; `else wire [MN-1:0] imos4, imow4, imon4, imoe4, imol4; wire [MN-1:0] imosa, imowa, imona, imoea, imola; `endif // input of CMs wire [MN-1:0][4:0][SCN-1:0] cmi0, cmi1, cmi2, cmi3; `ifdef ENABLE_CHANNEL_SLICING wire [MN-1:0][4:0][SCN-1:0] cmi4, cmia; `else wire [MN-1:0][4:0] cmi4, cmia; `endif // output of CMs wire [MN-1:0][4:0][SCN-1:0] cmo0, cmo1, cmo2, cmo3; `ifdef ENABLE_CHANNEL_SLICING wire [MN-1:0][4:0][SCN-1:0] cmo4, cmoa; `else wire [MN-1:0][4:0] cmo4, cmoa; `endif genvar i,j,k; dcb #(.NN(NN), .MN(MN), .DW(DW)) SIM ( .o0 ( imos0 ), .o1 ( imos1 ), .o2 ( imos2 ), .o3 ( imos3 ), .o4 ( imos4 ), .ia ( sia ), .i0 ( si0 ), .i1 ( si1 ), .i2 ( si2 ), .i3 ( si3 ), .i4 ( si4 ), .oa ( imosa ), .cfg ( imcfg[0] ) ); dcb #(.NN(NN), .MN(MN), .DW(DW)) WIM ( .o0 ( imow0 ), .o1 ( imow1 ), .o2 ( imow2 ), .o3 ( imow3 ), .o4 ( imow4 ), .ia ( wia ), .i0 ( wi0 ), .i1 ( wi1 ), .i2 ( wi2 ), .i3 ( wi3 ), .i4 ( wi4 ), .oa ( imowa ), .cfg ( imcfg[1] ) ); dcb #(.NN(NN), .MN(MN), .DW(DW)) NIM ( .o0 ( imon0 ), .o1 ( imon1 ), .o2 ( imon2 ), .o3 ( imon3 ), .o4 ( imon4 ), .ia ( nia ), .i0 ( ni0 ), .i1 ( ni1 ), .i2 ( ni2 ), .i3 ( ni3 ), .i4 ( ni4 ), .oa ( imona ), .cfg ( imcfg[2] ) ); dcb #(.NN(NN), .MN(MN), .DW(DW)) EIM ( .o0 ( imoe0 ), .o1 ( imoe1 ), .o2 ( imoe2 ), .o3 ( imoe3 ), .o4 ( imoe4 ), .ia ( eia ), .i0 ( ei0 ), .i1 ( ei1 ), .i2 ( ei2 ), .i3 ( ei3 ), .i4 ( ei4 ), .oa ( imoea ), .cfg ( imcfg[3] ) ); dcb #(.NN(NN), .MN(MN), .DW(DW)) LIM ( .o0 ( imol0 ), .o1 ( imol1 ), .o2 ( imol2 ), .o3 ( imol3 ), .o4 ( imol4 ), .ia ( lia ), .i0 ( li0 ), .i1 ( li1 ), .i2 ( li2 ), .i3 ( li3 ), .i4 ( li4 ), .oa ( imola ), .cfg ( imcfg[4] ) ); generate for(i=0; i<MN; i++) begin: IMSHF // shuffle the interconnects between IMs and CMs assign cmi0[i][0] = imos0[i]; assign cmi1[i][0] = imos1[i]; assign cmi2[i][0] = imos2[i]; assign cmi3[i][0] = imos3[i]; assign cmi4[i][0] = imos4[i]; assign imosa[i] = cmia[i][0]; assign cmi0[i][1] = imow0[i]; assign cmi1[i][1] = imow1[i]; assign cmi2[i][1] = imow2[i]; assign cmi3[i][1] = imow3[i]; assign cmi4[i][1] = imow4[i]; assign imowa[i] = cmia[i][1]; assign cmi0[i][2] = imon0[i]; assign cmi1[i][2] = imon1[i]; assign cmi2[i][2] = imon2[i]; assign cmi3[i][2] = imon3[i]; assign cmi4[i][2] = imon4[i]; assign imona[i] = cmia[i][2]; assign cmi0[i][3] = imoe0[i]; assign cmi1[i][3] = imoe1[i]; assign cmi2[i][3] = imoe2[i]; assign cmi3[i][3] = imoe3[i]; assign cmi4[i][3] = imoe4[i]; assign imoea[i] = cmia[i][3]; assign cmi0[i][4] = imol0[i]; assign cmi1[i][4] = imol1[i]; assign cmi2[i][4] = imol2[i]; assign cmi3[i][4] = imol3[i]; assign cmi4[i][4] = imol4[i]; assign imola[i] = cmia[i][4]; // CM modules dcb_xy #(.VCN(1), .VCW(DW)) CM ( .sia ( cmia[i][0] ), .wia ( cmia[i][1] ), .nia ( cmia[i][2] ), .eia ( cmia[i][3] ), .lia ( cmia[i][4] ), .so0 ( cmo0[i][0] ), .so1 ( cmo1[i][0] ), .so2 ( cmo2[i][0] ), .so3 ( cmo3[i][0] ), .so4 ( cmo4[i][0] ), .wo0 ( cmo0[i][1] ), .wo1 ( cmo1[i][1] ), .wo2 ( cmo2[i][1] ), .wo3 ( cmo3[i][1] ), .wo4 ( cmo4[i][1] ) , .no0 ( cmo0[i][2] ), .no1 ( cmo1[i][2] ), .no2 ( cmo2[i][2] ), .no3 ( cmo3[i][2] ), .no4 ( cmo4[i][2] ), .eo0 ( cmo0[i][3] ), .eo1 ( cmo1[i][3] ), .eo2 ( cmo2[i][3] ), .eo3 ( cmo3[i][3] ), .eo4 ( cmo4[i][3] ), .lo0 ( cmo0[i][4] ), .lo1 ( cmo1[i][4] ), .lo2 ( cmo2[i][4] ), .lo3 ( cmo3[i][4] ), .lo4 ( cmo4[i][4] ), .si0 ( cmi0[i][0] ), .si1 ( cmi1[i][0] ), .si2 ( cmi2[i][0] ), .si3 ( cmi3[i][0] ), .si4 ( cmi4[i][0] ), .wi0 ( cmi0[i][1] ), .wi1 ( cmi1[i][1] ), .wi2 ( cmi2[i][1] ), .wi3 ( cmi3[i][1] ), .wi4 ( cmi4[i][1] ), .ni0 ( cmi0[i][2] ), .ni1 ( cmi1[i][2] ), .ni2 ( cmi2[i][2] ), .ni3 ( cmi3[i][2] ), .ni4 ( cmi4[i][2] ), .ei0 ( cmi0[i][3] ), .ei1 ( cmi1[i][3] ), .ei2 ( cmi2[i][3] ), .ei3 ( cmi3[i][3] ), .ei4 ( cmi4[i][3] ), .li0 ( cmi0[i][4] ), .li1 ( cmi1[i][4] ), .li2 ( cmi2[i][4] ), .li3 ( cmi3[i][4] ), .li4 ( cmi4[i][4] ), .soa ( cmoa[i][0] ), .woa ( cmoa[i][1] ), .noa ( cmoa[i][2] ), .eoa ( cmoa[i][3] ), .loa ( cmoa[i][4] ), .wcfg ( wcfg[i] ), .ecfg ( ecfg[i] ), .lcfg ( lcfg[i] ), .scfg ( scfg[i] ), .ncfg ( ncfg[i] ) ); // shuffle between CMs and OMs(OPs) assign so0[i] = cmo0[i][0]; assign so1[i] = cmo1[i][0]; assign so2[i] = cmo2[i][0]; assign so3[i] = cmo3[i][0]; assign so4[i] = cmo4[i][0]; assign cmoa[i][0] = soa[i]; assign wo0[i] = cmo0[i][1]; assign wo1[i] = cmo1[i][1]; assign wo2[i] = cmo2[i][1]; assign wo3[i] = cmo3[i][1]; assign wo4[i] = cmo4[i][1]; assign cmoa[i][1] = woa[i]; assign no0[i] = cmo0[i][2]; assign no1[i] = cmo1[i][2]; assign no2[i] = cmo2[i][2]; assign no3[i] = cmo3[i][2]; assign no4[i] = cmo4[i][2]; assign cmoa[i][2] = noa[i]; assign eo0[i] = cmo0[i][3]; assign eo1[i] = cmo1[i][3]; assign eo2[i] = cmo2[i][3]; assign eo3[i] = cmo3[i][3]; assign eo4[i] = cmo4[i][3]; assign cmoa[i][3] = eoa[i]; assign lo0[i] = cmo0[i][4]; assign lo1[i] = cmo1[i][4]; assign lo2[i] = cmo2[i][4]; assign lo3[i] = cmo3[i][4]; assign lo4[i] = cmo4[i][4]; assign cmoa[i][4] = loa[i]; end // block: IMSHF endgenerate endmodule // dclos
Go to most recent revision | Compare with Previous | Blame | View Log