URL
https://opencores.org/ocsvn/async_sdm_noc/async_sdm_noc/trunk
Subversion Repositories async_sdm_noc
[/] [async_sdm_noc/] [branches/] [init/] [common/] [src/] [mrma.v] - Rev 40
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 Multi-resource match arbiter *** SystemVerilog is used *** References Stanislavs Golubcovs, Delong Shang, Fei Xia, Andrey Mokhov and Alex Yakovlev, Multi-resource arbiter decomposition, Tech report NCL-EECE-MSD-TR-2009-143, Microelectronic System Design Group, School of EECE, Newcastle University, 2009. Stanislavs Golubcovs, Delong Shang, Fei Xia, Andrey Mokhov and Alex Yakovlev, Modular approach to multi-resource arbiter design, IEEE Symposium on Asynchronous Circuits and Systems, 2009. History: 05/09/2009 Initial version. <wsong83@gmail.com> 05/11/2009 Speed up the arbiter. <wsong83@gmail.com> 27/05/2011 Clean up for opensource. <wsong83@gmail.com> */ module mrma (/*AUTOARG*/ // Outputs ca, ra, cfg, // Inputs c, r, rst_n ); // parameters parameter N = 2; // the number of requests/clients parameter M = 2; // the number of resources input [N-1:0] c; // requests/clients output [N-1:0] ca; // requests ack input [M-1:0] r; // resources output [M-1:0] ra; // resource ack output [M-1:0][N-1:0] cfg; // the generated configuration wire [N-1:0][M-1:0] scfg; wire [M-1:0][N-1:0] hs; // match results wire [M-1:0][N-1:0] blk; // blockage wire [N-1:0][M-1:0] sblk; // shuffled blockage wire [M-1:0] rbi; // resource blockage wire [N-1:0] cbi; // client blockage wire [N-1:0] cg, cm; // client requests wire [M-1:0] rg, rm; // resource requests input rst_n; // active low reset // generate variables genvar i, j; // input arbiters tree_arb #(N) CIArb ( .req ( cm ), .gnt ( cg ) ); tree_arb #(M) RIArb ( .req ( rm ), .gnt ( rg ) ); generate // tile matrix for (i=0; i<M; i++) begin: Row for(j=0; j<N; j++) begin: Clm cr_blk E ( .bo ( blk[i][j] ), .hs ( hs[i][j] ), .cbi ( cbi[j] ), .rbi ( rbi[i] ), .rg ( rg[i] ), .cg ( cg[j] ) ); // shuffle the blockage assign sblk[j][i] = blk[i][j]; // shuffle the configuration assign scfg[j][i] = cfg[i][j]; // store the match results c2p C (.q(cfg[i][j]), .a(c[j]), .b(hs[i][j])); end // block: Clm end // block: Row // combine the row blockage and generate input requests for(i=0; i<M; i++) begin: RB assign rbi[i] = (|blk[i]) & rst_n; and AND_RG (rm[i], r[i], ~ra[i], rst_n); assign ra[i] = |cfg[i]; end // combine the column blockage and generate input requests for(j=0; j<N; j++) begin: CB assign cbi[j] = (|sblk[j]) & rst_n; and AND_CG (cm[j], c[j], ~ca[j], rst_n); assign ca[j] = |scfg[j]; end endgenerate endmodule // mrma
Go to most recent revision | Compare with Previous | Blame | View Log