URL
https://opencores.org/ocsvn/async_sdm_noc/async_sdm_noc/trunk
Subversion Repositories async_sdm_noc
[/] [async_sdm_noc/] [branches/] [init/] [vc/] [src/] [rtu.v] - Rev 79
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 The routing calculation unit in every input buffer. p0 -> L -> p1 -> RTC -> p2 -> L -> p3 -> DEMUX -> p[0] -> L -> p[1] -> L -> p[2] -> L -> p[3] -> L -> p[4] History: 02/04/2010 Initial version. <wsong83@gmail.com> 02/06/2011 Clean up for opensource. <wsong83@gmail.com> 09/06/2011 The selection pin of the demux must be considered in the ack process. <wsong83@gmail.com> */ module rtu (/*AUTOARG*/ // Outputs dia, dort, // Inputs rst_n, di0, di1, di2, di3, dit, divc, addrx, addry, doa ); parameter VCN = 2; parameter DIR = 0; parameter SN = 4; parameter PD = 2; input rst_n; input [3:0] di0, di1, di2, di3; input [2:0] dit; input [VCN-1:0] divc; output dia; input [7:0] addrx, addry; output [VCN-1:0][SN-1:0] dort; input [VCN-1:0] doa; //------------------------------------------ wire p0an, p0ap, p0den, p0ad, p0avc; wire [3:0] p1d0, p1d1, p1d2, p1d3; wire [VCN-1:0] p1vc; wire p1a, p1ad, p1avc, p1an; wire [SN-1:0] p2d; wire [VCN-1:0] p2vc; wire p2a, p2ad, p2avc, p2an; wire [SN-1:0] p3d; wire [VCN-1:0] p3vc; wire p3a, p3an; wire [PD*2:0][VCN-1:0][SN-1:0] pd; wire [PD*2:0][VCN-1:0] pda, pdan; wire [2:0] x_cmp [1:0]; wire [2:0] y_cmp [1:0]; wire [5:0] dec; wire ds, dw, dn, de, dl; genvar gvc, gbd, i; //------------------------------------------ // p0 -> L -> p1 c2 CP0DE ( .a0(dit[0]), .a1(p1an), .q(p0den)); c2 CP0A ( .a0(p0ad), .a1(p0avc), .q(p0ap)); assign p0an = |dit[2:1]; assign dia = p0an | p0ap; pipe4 #(.DW(8)) L0D ( .ia ( p0ad ), .o0 ( p1d0 ), .o1 ( p1d1 ), .o2 ( p1d2 ), .o3 ( p1d3 ), .i0 ( di0 ), .i1 ( di1 ), .i2 ( di2 ), .i3 ( di3 ), .oa ( p0den ) ); pipen #(.DW(VCN)) L0V ( .d_in ( divc ), .d_in_a ( p0avc ), .d_out ( p1vc ), .d_out_a ( p0den ) ); // p1 -> RTC -> p2 comp4 X0 ( .a({p1d3[0], p1d2[0], p1d1[0], p1d0[0]}), .b(addrx[3:0]), .q(x_cmp[0])); comp4 X1 ( .a({p1d3[1], p1d2[1], p1d1[1], p1d0[1]}), .b(addrx[7:4]), .q(x_cmp[1])); comp4 Y0 ( .a({p1d3[2], p1d2[2], p1d1[2], p1d0[2]}), .b(addry[3:0]), .q(y_cmp[0])); comp4 Y1 ( .a({p1d3[3], p1d2[3], p1d1[3], p1d0[3]}), .b(addry[7:4]), .q(y_cmp[1])); assign dec[0] = x_cmp[1][0] | (x_cmp[1][2]&x_cmp[0][0]); // frame x > addr x assign dec[1] = x_cmp[1][1] | (x_cmp[1][2]&x_cmp[0][1]); // frame x < addr x assign dec[2] = x_cmp[1][2] & x_cmp[0][2]; // frame x = addr x assign dec[3] = y_cmp[1][0] | (y_cmp[1][2]&y_cmp[0][0]); // frame y > addr y assign dec[4] = y_cmp[1][1] | (y_cmp[1][2]&y_cmp[0][1]); // frame y < addr y assign dec[5] = y_cmp[1][2] & y_cmp[0][2]; // frame y = addr y assign ds = dec[0]; assign dw = dec[2]&dec[4]; assign dn = dec[1]; assign de = dec[2]&dec[3]; assign dl = dec[2]&dec[5]; assign p2d = DIR == 0 ? {dl, de, dn, dw} : // south port DIR == 1 ? {dl, de} : // west port DIR == 2 ? {dl, de, dw, ds} : // north port DIR == 3 ? {dl, dw} : // east port {de, dn, dw, ds} ; // local port assign p2vc = p1vc; assign p1an = p2an; // p2 -> L -> p3 c2 CP2A ( .a0(p2ad), .a1(p2avc), .q(p2a)); assign p2an = (~p2a) & rst_n; pipen #(.DW(SN)) L2R ( .d_in ( p2d ), .d_in_a ( p2ad ), .d_out ( p3d ), .d_out_a ( p3an ) ); pipen #(.DW(VCN)) L2V ( .d_in ( p2vc ), .d_in_a ( p2avc ), .d_out ( p3vc ), .d_out_a ( p3an ) ); // p3 -> DEMUX -> pd[0] ddmux #(.DW(SN), .VCN(VCN)) RTDM ( .d_in_a ( p3a ), .d_out ( pd[0] ), .d_in ( p3d ), .d_sel ( p3vc ), .d_out_a ( pda[0]) ); //c2 CP3A (.q(p3a), .a0(p3ad), .a1(|p3vc)); assign p3an = (~p3a) & rst_n; // pd pipeline generate for(gbd=0; gbd<PD*2; gbd++) begin:RT for(gvc=0; gvc<VCN; gvc++) begin:V pipen #(.DW(SN)) P ( .d_in ( pd[gbd][gvc] ), .d_in_a ( pda[gbd][gvc] ), .d_out ( pd[gbd+1][gvc] ), .d_out_a ( pdan[gbd+1][gvc] ) ); assign pdan[gbd+1][gvc] = (~pda[gbd+1][gvc])&rst_n; end end // block: RT endgenerate // output generate for(gvc=0; gvc<VCN; gvc++) begin:OV assign dort[gvc] = pd[PD*2][gvc]; assign pda[PD*2][gvc] = doa[gvc]; end endgenerate endmodule // rtu
Go to most recent revision | Compare with Previous | Blame | View Log