URL https://opencores.org/ocsvn/s6soc/s6soc/trunk
Subversion Repositories s6soc

[/] [s6soc/] [trunk/] [rtl/] [cpu/] [zipcpu.v] - Blame information for rev 7

Go to most recent revision | Details | Compare with Previous | View Log

///////////////////////////////////////////////////////////////////////////////
//
// Filename:    zipcpu.v
//
// Project:     Zip CPU -- a small, lightweight, RISC CPU soft core
//
// Purpose:     This is the top level module holding the core of the Zip CPU
//              together.  The Zip CPU is designed to be as simple as possible.
//      (actual implementation aside ...)  The instruction set is about as
//      RISC as you can get, there are only 16 instruction types supported.
//      Please see the accompanying spec.pdf file for a description of these
//      instructions.
//
//      All instructions are 32-bits wide.  All bus accesses, both address and
//      data, are 32-bits over a wishbone bus.
//
//      The Zip CPU is fully pipelined with the following pipeline stages:
//
//              1. Prefetch, returns the instruction from memory. 
//
//              2. Instruction Decode
//
//              3. Read Operands
//
//              4. Apply Instruction
//
//              4. Write-back Results
//
//      Further information about the inner workings of this CPU may be
//      found in the spec.pdf file.  (The documentation within this file
//      had become out of date and out of sync with the spec.pdf, so look
//      to the spec.pdf for accurate and up to date information.)
//
//
//      In general, the pipelining is controlled by three pieces of logic
//      per stage: _ce, _stall, and _valid.  _valid means that the stage
//      holds a valid instruction.  _ce means that the instruction from the
//      previous stage is to move into this one, and _stall means that the
//      instruction from the previous stage may not move into this one.
//      The difference between these control signals allows individual stages
//      to propagate instructions independently.  In general, the logic works
//      as:
//
//
//      assign  (n)_ce = (n-1)_valid && (~(n)_stall)
//
//
//      always @(posedge i_clk)
//              if ((i_rst)||(clear_pipeline))
//                      (n)_valid = 0
//              else if (n)_ce
//                      (n)_valid = 1
//              else if (n+1)_ce
//                      (n)_valid = 0
//
//      assign (n)_stall = (  (n-1)_valid && ( pipeline hazard detection )  )
//                      || (  (n)_valid && (n+1)_stall );
//
//      and ...
//
//      always @(posedge i_clk)
//              if (n)_ce
//                      (n)_variable = ... whatever logic for this stage
//
//      Note that a stage can stall even if no instruction is loaded into
//      it.
//
//
// Creator:     Dan Gisselquist, Ph.D.
//              Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015, Gisselquist Technology, LLC
//
// This program is free software (firmware): you can redistribute it and/or
// modify it under the terms of  the GNU General Public License as published
// by the Free Software Foundation, either version 3 of the License, or (at
// your option) any later version.
//
// This program is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
// for more details.
//
// License:     GPL, v3, as defined and found on www.gnu.org,
//              http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////////
//
// We can either pipeline our fetches, or issue one fetch at a time.  Pipelined
// fetches are more complicated and therefore use more FPGA resources, while
// single fetches will cause the CPU to stall for about 5 stalls each 
// instruction cycle, effectively reducing the instruction count per clock to
// about 0.2.  However, the area cost may be worth it.  Consider:
//
//      Slice LUTs              ZipSystem       ZipCPU
//      Single Fetching         2521            1734
//      Pipelined fetching      2796            2046
//
//
//
`define CPU_CC_REG      4'he
`define CPU_PC_REG      4'hf
`define CPU_FPUERR_BIT  12      // Floating point error flag, set on error
`define CPU_DIVERR_BIT  11      // Divide error flag, set on divide by zero
`define CPU_BUSERR_BIT  10      // Bus error flag, set on error
`define CPU_TRAP_BIT    9       // User TRAP has taken place
`define CPU_ILL_BIT     8       // Illegal instruction
`define CPU_BREAK_BIT   7
`define CPU_STEP_BIT    6       // Will step one or two (VLIW) instructions
`define CPU_GIE_BIT     5
`define CPU_SLEEP_BIT   4
// Compile time defines
//
`include "cpudefs.v"
//
//
module  zipcpu(i_clk, i_rst, i_interrupt,
                // Debug interface
                i_halt, i_clear_pf_cache, i_dbg_reg, i_dbg_we, i_dbg_data,
                        o_dbg_stall, o_dbg_reg, o_dbg_cc,
                        o_break,
                // CPU interface to the wishbone bus
                o_wb_gbl_cyc, o_wb_gbl_stb,
                        o_wb_lcl_cyc, o_wb_lcl_stb,
                        o_wb_we, o_wb_addr, o_wb_data,
                        i_wb_ack, i_wb_stall, i_wb_data,
                        i_wb_err,
                // Accounting/CPU usage interface
                o_op_stall, o_pf_stall, o_i_count
`ifdef  DEBUG_SCOPE
                , o_debug
`endif
                );
        parameter       RESET_ADDRESS=32'h0100000, ADDRESS_WIDTH=24,
                        LGICACHE=6;
`ifdef  OPT_MULTIPLY
        parameter       IMPLEMENT_MPY = `OPT_MULTIPLY;
`else
        parameter       IMPLEMENT_MPY = 0;
`endif
`ifdef  OPT_DIVIDE
        parameter       IMPLEMENT_DIVIDE = 1;
`else
        parameter       IMPLEMENT_DIVIDE = 0;
`endif
`ifdef  OPT_IMPLEMENT_FPU
        parameter       IMPLEMENT_FPU = 1,
`else
        parameter       IMPLEMENT_FPU = 0,
`endif
                        IMPLEMENT_LOCK=1;
`ifdef  OPT_EARLY_BRANCHING
        parameter       EARLY_BRANCHING = 1;
`else
        parameter       EARLY_BRANCHING = 0;
`endif
        parameter       AW=ADDRESS_WIDTH;
        input                   i_clk, i_rst, i_interrupt;
        // Debug interface -- inputs
        input                   i_halt, i_clear_pf_cache;
        input           [4:0]    i_dbg_reg;
        input                   i_dbg_we;
        input           [31:0]   i_dbg_data;
        // Debug interface -- outputs
        output  reg             o_dbg_stall;
        output  reg     [31:0]   o_dbg_reg;
        output  reg     [3:0]    o_dbg_cc;
        output  wire            o_break;
        // Wishbone interface -- outputs
        output  wire            o_wb_gbl_cyc, o_wb_gbl_stb;
        output  wire            o_wb_lcl_cyc, o_wb_lcl_stb, o_wb_we;
        output  wire    [(AW-1):0]       o_wb_addr;
        output  wire    [31:0]   o_wb_data;
        // Wishbone interface -- inputs
        input                   i_wb_ack, i_wb_stall;
        input           [31:0]   i_wb_data;
        input                   i_wb_err;
        // Accounting outputs ... to help us count stalls and usage
        output  wire            o_op_stall;
        output  wire            o_pf_stall;
        output  wire            o_i_count;
        //
`ifdef  DEBUG_SCOPE
        output  reg     [31:0]   o_debug;
`endif
 
 
        // Registers
        //
        //      The distributed RAM style comment is necessary on the
        // SPARTAN6 with XST to prevent XST from oversimplifying the register
        // set and in the process ruining everything else.  It basically
        // optimizes logic away, to where it no longer works.  The logic
        // as described herein will work, this just makes sure XST implements
        // that logic.
        //
        (* ram_style = "distributed" *)
        reg     [31:0]   regset [0:31];
 
        // Condition codes
        // (BUS, TRAP,ILL,BREAKEN,STEP,GIE,SLEEP ), V, N, C, Z
        reg     [3:0]    flags, iflags;
        wire    [13:0]   w_uflags, w_iflags;
        reg             trap, break_en, step, gie, sleep;
`ifdef  OPT_ILLEGAL_INSTRUCTION
        reg             ill_err_u, ill_err_i;
`else
        wire            ill_err_u, ill_err_i;
`endif
        reg             ibus_err_flag, ubus_err_flag;
        wire            idiv_err_flag, udiv_err_flag;
        wire            ifpu_err_flag, ufpu_err_flag;
        wire            ihalt_phase, uhalt_phase;
 
        // The master chip enable
        wire            master_ce;
 
        //
        //
        //      PIPELINE STAGE #1 :: Prefetch
        //              Variable declarations
        //
        reg     [(AW-1):0]       pf_pc;
        reg     new_pc;
        wire    clear_pipeline;
        assign  clear_pipeline = new_pc || i_clear_pf_cache;
 
        wire            dcd_stalled;
        wire            pf_cyc, pf_stb, pf_we, pf_busy, pf_ack, pf_stall, pf_err;
        wire    [(AW-1):0]       pf_addr;
        wire    [31:0]           pf_data;
        wire    [31:0]           instruction;
        wire    [(AW-1):0]       instruction_pc;
        wire    pf_valid, instruction_gie, pf_illegal;
 
        //
        //
        //      PIPELINE STAGE #2 :: Instruction Decode
        //              Variable declarations
        //
        //
        reg             opvalid, opvalid_mem, opvalid_alu;
        reg             opvalid_div, opvalid_fpu;
        wire            op_stall, dcd_ce, dcd_phase;
        wire    [3:0]    dcdOp;
        wire    [4:0]    dcdA, dcdB, dcdR;
        wire            dcdA_cc, dcdB_cc, dcdA_pc, dcdB_pc, dcdR_cc, dcdR_pc;
        wire    [3:0]    dcdF;
        wire            dcdR_wr, dcdA_rd, dcdB_rd,
                                dcdALU, dcdM, dcdDV, dcdFP,
                                dcdF_wr, dcd_gie, dcd_break, dcd_lock,
                                dcd_pipe, dcd_ljmp;
        reg             r_dcdvalid;
        wire            dcdvalid;
        wire    [(AW-1):0]       dcd_pc;
        wire    [31:0]   dcdI;
        wire            dcd_zI; // true if dcdI == 0
        wire    dcdA_stall, dcdB_stall, dcdF_stall;
 
        wire    dcd_illegal;
        wire                    dcd_early_branch;
        wire    [(AW-1):0]       dcd_branch_pc;
 
 
        //
        //
        //      PIPELINE STAGE #3 :: Read Operands
        //              Variable declarations
        //
        //
        //
        // Now, let's read our operands
        reg     [4:0]    alu_reg;
        reg     [3:0]    opn;
        reg     [4:0]    opR;
        reg     [31:0]   r_opA, r_opB;
        reg     [(AW-1):0]       op_pc;
        wire    [31:0]   w_opA, w_opB;
        wire    [31:0]   opA_nowait, opB_nowait, opA, opB;
        reg             opR_wr, opR_cc, opF_wr, op_gie;
        wire    [13:0]   opFl;
        reg     [5:0]    r_opF;
        wire    [7:0]    opF;
        wire            op_ce, op_phase, op_pipe;
        // Some pipeline control wires
`ifdef  OPT_PIPELINED
        reg     opA_alu, opA_mem;
        reg     opB_alu, opB_mem;
`endif
`ifdef  OPT_ILLEGAL_INSTRUCTION
        reg     op_illegal;
`endif
        reg     op_break;
        wire    op_lock;
 
 
        //
        //
        //      PIPELINE STAGE #4 :: ALU / Memory
        //              Variable declarations
        //
        //
        reg     [(AW-1):0]       alu_pc;
        reg             alu_pc_valid, mem_pc_valid;
        wire            alu_phase;
        wire            alu_ce, alu_stall;
        wire    [31:0]   alu_result;
        wire    [3:0]    alu_flags;
        wire            alu_valid, alu_busy;
        wire            set_cond;
        reg             alu_wr, alF_wr, alu_gie;
        wire            alu_illegal_op;
        wire            alu_illegal;
 
 
 
        wire    mem_ce, mem_stalled;
`ifdef  OPT_PIPELINED_BUS_ACCESS
        wire    mem_pipe_stalled;
`endif
        wire    mem_valid, mem_ack, mem_stall, mem_err, bus_err,
                mem_cyc_gbl, mem_cyc_lcl, mem_stb_gbl, mem_stb_lcl, mem_we;
        wire    [4:0]            mem_wreg;
 
        wire                    mem_busy, mem_rdbusy;
        wire    [(AW-1):0]       mem_addr;
        wire    [31:0]           mem_data, mem_result;
 
        wire    div_ce, div_error, div_busy, div_valid;
        wire    [31:0]   div_result;
        wire    [3:0]    div_flags;
 
        assign  div_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_div)
                                &&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
                                &&(set_cond);
 
        wire    fpu_ce, fpu_error, fpu_busy, fpu_valid;
        wire    [31:0]   fpu_result;
        wire    [3:0]    fpu_flags;
 
        assign  fpu_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_fpu)
                                &&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
                                &&(set_cond);
 
 
        //
        //
        //      PIPELINE STAGE #5 :: Write-back
        //              Variable declarations
        //
        wire            wr_reg_ce, wr_flags_ce, wr_write_pc, wr_write_cc;
        wire    [4:0]    wr_reg_id;
        wire    [31:0]   wr_reg_vl;
        wire    w_switch_to_interrupt, w_release_from_interrupt;
        reg     [(AW-1):0]       upc, ipc;
 
 
 
        //
        //      MASTER: clock enable.
        //
        assign  master_ce = (~i_halt)&&(~o_break)&&(~sleep);
 
 
        //
        //      PIPELINE STAGE #1 :: Prefetch
        //              Calculate stall conditions
        //
        //      These are calculated externally, within the prefetch module.
        //
 
        //
        //      PIPELINE STAGE #2 :: Instruction Decode
        //              Calculate stall conditions
`ifdef  OPT_PIPELINED
        assign          dcd_ce = ((~dcdvalid)||(~dcd_stalled))&&(~clear_pipeline);
`else
        assign          dcd_ce = 1'b1;
`endif
`ifdef  OPT_PIPELINED
        assign          dcd_stalled = (dcdvalid)&&(op_stall);
`else
        // If not pipelined, there will be no opvalid_ anything, and the
        // op_stall will be false, dcdX_stall will be false, thus we can simply
        // do a ...
        assign          dcd_stalled = 1'b0;
`endif
        //
        //      PIPELINE STAGE #3 :: Read Operands
        //              Calculate stall conditions
        wire    op_lock_stall;
`ifdef  OPT_PIPELINED
        assign  op_stall = (opvalid)&&( // Only stall if we're loaded w/validins
                        // Stall if we're stopped, and not allowed to execute
                        // an instruction
                        // (~master_ce)         // Already captured in alu_stall
                        //
                        // Stall if going into the ALU and the ALU is stalled
                        //      i.e. if the memory is busy, or we are single
                        //      stepping.  This also includes our stalls for
                        //      op_break and op_lock, so we don't need to
                        //      include those as well here.
                        // This also includes whether or not the divide or
                        // floating point units are busy.
                        (alu_stall)
                        //
                        // Stall if we are going into memory with an operation
                        //      that cannot be pipelined, and the memory is
                        //      already busy
                        ||(mem_stalled) // &&(opvalid_mem) part of mem_stalled
                        )
                        ||(dcdvalid)&&(
                                // Stall if we need to wait for an operand A
                                // to be ready to read
                                (dcdA_stall)
                                // Likewise for B, also includes logic
                                // regarding immediate offset (register must
                                // be in register file if we need to add to
                                // an immediate)
                                ||(dcdB_stall)
                                // Or if we need to wait on flags to work on the
                                // CC register
                                ||(dcdF_stall)
                        );
        assign  op_ce = ((dcdvalid)||(dcd_illegal))&&(~op_stall)&&(~clear_pipeline);
`else
        assign  op_stall = (opvalid)&&(~master_ce);
        assign  op_ce = ((dcdvalid)||(dcd_illegal));
`endif
 
        //
        //      PIPELINE STAGE #4 :: ALU / Memory
        //              Calculate stall conditions
        //
        // 1. Basic stall is if the previous stage is valid and the next is
        //      busy.  
        // 2. Also stall if the prior stage is valid and the master clock enable
        //      is de-selected
        // 3. Stall if someone on the other end is writing the CC register,
        //      since we don't know if it'll put us to sleep or not.
        // 4. Last case: Stall if we would otherwise move a break instruction
        //      through the ALU.  Break instructions are not allowed through
        //      the ALU.
`ifdef  OPT_PIPELINED
        assign  alu_stall = (((~master_ce)||(mem_rdbusy)||(alu_busy))&&(opvalid_alu)) //Case 1&2
                        // Old case #3--this isn't an ALU stall though ...
                        ||((opvalid_alu)&&(wr_reg_ce)&&(wr_reg_id[4] == op_gie)
                                &&(wr_write_cc)) // Case 3
                        ||((opvalid)&&(op_lock)&&(op_lock_stall))
                        ||((opvalid)&&(op_break))
                        ||(div_busy)||(fpu_busy);
        assign  alu_ce = (master_ce)&&((opvalid_alu)||(op_illegal))
                                &&(~alu_stall)
                                &&(~clear_pipeline);
`else
        assign  alu_stall = ((~master_ce)&&(opvalid_alu))
                                ||((opvalid_alu)&&(op_break));
        assign  alu_ce = (master_ce)&&((opvalid_alu)||(op_illegal))&&(~alu_stall);
`endif
        //
 
        //
        // Note: if you change the conditions for mem_ce, you must also change
        // alu_pc_valid.
        //
`ifdef  OPT_PIPELINED
        assign  mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)
                        &&(~clear_pipeline);
`else
        // If we aren't pipelined, then no one will be changing what's in the
        // pipeline (i.e. clear_pipeline), while our only instruction goes
        // through the ... pipeline.
        assign  mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled);
`endif
`ifdef  OPT_PIPELINED_BUS_ACCESS
        assign  mem_stalled = (~master_ce)||(alu_busy)||((opvalid_mem)&&(
                                (mem_pipe_stalled)
                                ||((~op_pipe)&&(mem_busy))
                                ||(div_busy)
                                ||(fpu_busy)
                                // Stall waiting for flags to be valid
                                // Or waiting for a write to the PC register
                                // Or CC register, since that can change the
                                //  PC as well
                                ||((wr_reg_ce)&&(wr_reg_id[4] == op_gie)
                                        &&((wr_write_pc)||(wr_write_cc)))));
`else
`ifdef  OPT_PIPELINED
        assign  mem_stalled = (mem_busy)||((opvalid_mem)&&(
                                (~master_ce)
                                // Stall waiting for flags to be valid
                                // Or waiting for a write to the PC register
                                // Or CC register, since that can change the
                                //  PC as well
                                ||((wr_reg_ce)&&(wr_reg_id[4] == op_gie)&&((wr_write_pc)||(wr_write_cc)))));
`else
        assign  mem_stalled = (opvalid_mem)&&(~master_ce);
`endif
`endif
 
 
        //
        //
        //      PIPELINE STAGE #1 :: Prefetch
        //
        //
`ifdef  OPT_SINGLE_FETCH
        wire            pf_ce;
 
        assign          pf_ce = (~pf_valid)&&(~dcdvalid)&&(~opvalid)&&(~alu_valid);
        prefetch        #(ADDRESS_WIDTH)
                        pf(i_clk, i_rst, (pf_ce), (~dcd_stalled), pf_pc, gie,
                                instruction, instruction_pc, instruction_gie,
                                        pf_valid, pf_illegal,
                                pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
                                pf_ack, pf_stall, pf_err, i_wb_data);
 
        initial r_dcdvalid = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        r_dcdvalid <= 1'b0;
                else if (dcd_ce)
                        r_dcdvalid <= (pf_valid);
                else if (op_ce)
                        r_dcdvalid <= 1'b0;
        assign  dcdvalid = r_dcdvalid;
 
`else // Pipe fetch
 
`ifdef  OPT_TRADITIONAL_PFCACHE
        pfcache #(LGICACHE, ADDRESS_WIDTH)
                pf(i_clk, i_rst, (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
                                        i_clear_pf_cache,
                                // dcd_pc,
                                ~dcd_stalled,
                                ((dcd_early_branch)&&(~clear_pipeline))
                                        ? dcd_branch_pc:pf_pc,
                                instruction, instruction_pc, pf_valid,
                                pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
                                        pf_ack, pf_stall, pf_err, i_wb_data,
                                pf_illegal);
`else
        pipefetch       #(RESET_ADDRESS, LGICACHE, ADDRESS_WIDTH)
                        pf(i_clk, i_rst, (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
                                        i_clear_pf_cache, ~dcd_stalled,
                                        (new_pc)?pf_pc:dcd_branch_pc,
                                        instruction, instruction_pc, pf_valid,
                                pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
                                        pf_ack, pf_stall, pf_err, i_wb_data,
//`ifdef        OPT_PRECLEAR_BUS
                                //((dcd_clear_bus)&&(dcdvalid))
                                //||((op_clear_bus)&&(opvalid))
                                //||
//`endif
                                (mem_cyc_lcl)||(mem_cyc_gbl),
                                pf_illegal);
`endif
        assign  instruction_gie = gie;
 
        initial r_dcdvalid = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(clear_pipeline))
                        r_dcdvalid <= 1'b0;
                else if (dcd_ce)
                        r_dcdvalid <= (pf_valid)&&(~clear_pipeline)&&(~dcd_ljmp)&&((~r_dcdvalid)||(~dcd_early_branch));
                else if (op_ce)
                        r_dcdvalid <= 1'b0;
        assign  dcdvalid = r_dcdvalid;
`endif
 
`ifdef  OPT_NEW_INSTRUCTION_SET
        idecode #(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
                        IMPLEMENT_FPU)
                instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
                        dcd_ce, dcd_stalled, instruction, instruction_gie,
                        instruction_pc, pf_valid, pf_illegal, dcd_phase,
                        dcd_illegal, dcd_pc, dcd_gie,
                        { dcdR_cc, dcdR_pc, dcdR },
                        { dcdA_cc, dcdA_pc, dcdA },
                        { dcdB_cc, dcdB_pc, dcdB },
                        dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
                        dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
                        dcdR_wr,dcdA_rd, dcdB_rd,
                        dcd_early_branch,
                        dcd_branch_pc, dcd_ljmp,
                        dcd_pipe);
`else
        idecode_deprecated
                #(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
                        IMPLEMENT_FPU)
                instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
                        dcd_ce, dcd_stalled, instruction, instruction_gie,
                        instruction_pc, pf_valid, pf_illegal, dcd_phase,
                        dcd_illegal, dcd_pc, dcd_gie,
                        { dcdR_cc, dcdR_pc, dcdR },
                        { dcdA_cc, dcdA_pc, dcdA },
                        { dcdB_cc, dcdB_pc, dcdB },
                        dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
                        dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
                        dcdR_wr,dcdA_rd, dcdB_rd,
                        dcd_early_branch,
                        dcd_branch_pc,
                        dcd_pipe);
        assign  dcd_ljmp = 1'b0;
`endif
 
`ifdef  OPT_PIPELINED_BUS_ACCESS
        reg             r_op_pipe;
 
        initial r_op_pipe = 1'b0;
        // To be a pipeable operation, there must be 
        //      two valid adjacent instructions
        //      Both must be memory instructions
        //      Both must be writes, or both must be reads
        //      Both operations must be to the same identical address,
        //              or at least a single (one) increment above that address
        //
        // However ... we need to know this before this clock, hence this is
        // calculated in the instruction decoder.
        always @(posedge i_clk)
                if (op_ce)
                        r_op_pipe <= dcd_pipe;
        assign  op_pipe = r_op_pipe;
`else
        assign  op_pipe = 1'b0;
`endif
 
        //
        //
        //      PIPELINE STAGE #3 :: Read Operands (Registers)
        //
        //
        assign  w_opA = regset[dcdA];
        assign  w_opB = regset[dcdB];
 
        wire    [8:0]    w_cpu_info;
        assign  w_cpu_info = {
`ifdef  OPT_ILLEGAL_INSTRUCTION
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_MULTIPLY
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_DIVIDE
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_IMPLEMENT_FPU
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_PIPELINED
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_TRADITIONAL_CACHE
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_EARLY_BRANCHING
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_PIPELINED_BUS_ACCESS
        1'b1,
`else
        1'b0,
`endif
`ifdef  OPT_VLIW
        1'b1
`else
        1'b0
`endif
        };
 
        wire    [31:0]   w_pcA_v;
        generate
        if (AW < 32)
                assign  w_pcA_v = {{(32-AW){1'b0}}, (dcdA[4] == dcd_gie)?dcd_pc:upc };
        else
                assign  w_pcA_v = (dcdA[4] == dcd_gie)?dcd_pc:upc;
        endgenerate
 
`ifdef  OPT_PIPELINED
        reg     [4:0]    opA_id, opB_id;
        reg             opA_rd, opB_rd;
        always @(posedge i_clk)
                if (op_ce)
                begin
                        opA_id <= dcdA;
                        opB_id <= dcdB;
                        opA_rd <= dcdA_rd;
                        opB_rd <= dcdB_rd;
                end
`endif
 
        always @(posedge i_clk)
                if (op_ce) // &&(dcdvalid))
                begin
                        if ((wr_reg_ce)&&(wr_reg_id == dcdA))
                                r_opA <= wr_reg_vl;
                        else if (dcdA_pc)
                                r_opA <= w_pcA_v;
                        else if (dcdA_cc)
                                r_opA <= { w_cpu_info, w_opA[22:14], (dcdA[4])?w_uflags:w_iflags };
                        else
                                r_opA <= w_opA;
`ifdef  OPT_PIPELINED
                end else
                begin // We were going to pick these up when they became valid,
                        // but for some reason we're stuck here as they became
                        // valid.  Pick them up now anyway
                        // if (((opA_alu)&&(alu_wr))||((opA_mem)&&(mem_valid)))
                                // r_opA <= wr_reg_vl;
                        if ((wr_reg_ce)&&(wr_reg_id == opA_id)&&(opA_rd))
                                r_opA <= wr_reg_vl;
`endif
                end
 
        wire    [31:0]   w_opBnI, w_pcB_v;
        generate
        if (AW < 32)
                assign  w_pcB_v = {{(32-AW){1'b0}}, (dcdB[4] == dcd_gie)?dcd_pc:upc };
        else
                assign  w_pcB_v = (dcdB[4] == dcd_gie)?dcd_pc:upc;
        endgenerate
 
        assign  w_opBnI = (~dcdB_rd) ? 32'h00
                : (((wr_reg_ce)&&(wr_reg_id == dcdB)) ? wr_reg_vl
                : ((dcdB_pc) ? w_pcB_v
                : ((dcdB_cc) ? { w_cpu_info, w_opB[22:14], // w_opB[31:14],
                        (dcdB[4])?w_uflags:w_iflags}
                : w_opB)));
 
        always @(posedge i_clk)
                if (op_ce) // &&(dcdvalid))
                        r_opB <= w_opBnI + dcdI;
`ifdef  OPT_PIPELINED
                else if ((wr_reg_ce)&&(opB_id == wr_reg_id)&&(opB_rd))
                        r_opB <= wr_reg_vl;
`endif
 
        // The logic here has become more complex than it should be, no thanks
        // to Xilinx's Vivado trying to help.  The conditions are supposed to
        // be two sets of four bits: the top bits specify what bits matter, the
        // bottom specify what those top bits must equal.  However, two of
        // conditions check whether bits are on, and those are the only two
        // conditions checking those bits.  Therefore, Vivado complains that
        // these two bits are redundant.  Hence the convoluted expression
        // below, arriving at what we finally want in the (now wire net)
        // opF.
        always @(posedge i_clk)
                if (op_ce)
                begin // Set the flag condition codes, bit order is [3:0]=VNCZ
                        case(dcdF[2:0])
                        3'h0:   r_opF <= 6'h00; // Always
`ifdef  OPT_NEW_INSTRUCTION_SET
                        // These were remapped as part of the new instruction
                        // set in order to make certain that the low order
                        // two bits contained the most commonly used 
                        // conditions: Always, LT, Z, and NZ.
                        3'h1:   r_opF <= 6'h24; // LT
                        3'h2:   r_opF <= 6'h11; // Z
                        3'h3:   r_opF <= 6'h10; // NE
                        3'h4:   r_opF <= 6'h30; // GT (!N&!Z)
                        3'h5:   r_opF <= 6'h20; // GE (!N)
`else
                        3'h1:   r_opF <= 6'h11; // Z
                        3'h2:   r_opF <= 6'h10; // NE
                        3'h3:   r_opF <= 6'h20; // GE (!N)
                        3'h4:   r_opF <= 6'h30; // GT (!N&!Z)
                        3'h5:   r_opF <= 6'h24; // LT
`endif
                        3'h6:   r_opF <= 6'h02; // C
                        3'h7:   r_opF <= 6'h08; // V
                        endcase
                end // Bit order is { (flags_not_used), VNCZ mask, VNCZ value }
        assign  opF = { r_opF[3], r_opF[5], r_opF[1], r_opF[4:0] };
 
        wire    w_opvalid;
        assign  w_opvalid = (~clear_pipeline)&&(dcdvalid)&&(~dcd_ljmp);
        initial opvalid     = 1'b0;
        initial opvalid_alu = 1'b0;
        initial opvalid_mem = 1'b0;
        initial opvalid_div = 1'b0;
        initial opvalid_fpu = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                begin
                        opvalid     <= 1'b0;
                        opvalid_alu <= 1'b0;
                        opvalid_mem <= 1'b0;
                end else if (op_ce)
                begin
                        // Do we have a valid instruction?
                        //   The decoder may vote to stall one of its
                        //   instructions based upon something we currently
                        //   have in our queue.  This instruction must then
                        //   move forward, and get a stall cycle inserted.
                        //   Hence, the test on dcd_stalled here.  If we must
                        //   wait until our operands are valid, then we aren't
                        //   valid yet until then.
                        opvalid<= w_opvalid;
`ifdef  OPT_ILLEGAL_INSTRUCTION
                        opvalid_alu <= ((dcdALU)||(dcd_illegal))&&(w_opvalid);
                        opvalid_mem <= (dcdM)&&(~dcd_illegal)&&(w_opvalid);
                        opvalid_div <= (dcdDV)&&(~dcd_illegal)&&(w_opvalid);
                        opvalid_fpu <= (dcdFP)&&(~dcd_illegal)&&(w_opvalid);
`else
                        opvalid_alu <= (dcdALU)&&(w_opvalid);
                        opvalid_mem <= (dcdM)&&(w_opvalid);
                        opvalid_div <= (dcdDV)&&(w_opvalid);
                        opvalid_fpu <= (dcdFP)&&(w_opvalid);
`endif
                end else if ((clear_pipeline)||(alu_ce)||(mem_ce)||(div_ce)||(fpu_ce))
                begin
                        opvalid     <= 1'b0;
                        opvalid_alu <= 1'b0;
                        opvalid_mem <= 1'b0;
                        opvalid_div <= 1'b0;
                        opvalid_fpu <= 1'b0;
                end
 
        // Here's part of our debug interface.  When we recognize a break
        // instruction, we set the op_break flag.  That'll prevent this
        // instruction from entering the ALU, and cause an interrupt before
        // this instruction.  Thus, returning to this code will cause the
        // break to repeat and continue upon return.  To get out of this
        // condition, replace the break instruction with what it is supposed
        // to be, step through it, and then replace it back.  In this fashion,
        // a debugger can step through code.
        // assign w_op_break = (dcd_break)&&(r_dcdI[15:0] == 16'h0001);
        initial op_break = 1'b0;
        always @(posedge i_clk)
                if (i_rst)      op_break <= 1'b0;
                else if (op_ce) op_break <= (dcd_break);
                else if ((clear_pipeline)||(~opvalid))
                                op_break <= 1'b0;
 
`ifdef  OPT_PIPELINED
        generate
        if (IMPLEMENT_LOCK != 0)
        begin
                reg     r_op_lock, r_op_lock_stall;
 
                initial r_op_lock_stall = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_op_lock_stall <= 1'b0;
                        else
                                r_op_lock_stall <= (~opvalid)||(~op_lock)
                                                ||(~dcdvalid)||(~pf_valid);
 
                assign  op_lock_stall = r_op_lock_stall;
 
                initial r_op_lock = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_op_lock <= 1'b0;
                        else if ((op_ce)&&(dcd_lock))
                                r_op_lock <= 1'b1;
                        else if ((op_ce)||(clear_pipeline))
                                r_op_lock <= 1'b0;
                assign  op_lock = r_op_lock;
 
        end else begin
                assign  op_lock_stall = 1'b0;
                assign  op_lock = 1'b0;
        end endgenerate
 
`else
        assign op_lock_stall = 1'b0;
        assign op_lock       = 1'b0;
`endif
 
`ifdef  OPT_ILLEGAL_INSTRUCTION
        initial op_illegal = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(clear_pipeline))
                        op_illegal <= 1'b0;
                else if(op_ce)
`ifdef  OPT_PIPELINED
                        op_illegal <=(dcd_illegal)||((dcd_lock)&&(IMPLEMENT_LOCK == 0));
`else
                        op_illegal <= (dcd_illegal)||(dcd_lock);
`endif
`endif
 
        // No generate on EARLY_BRANCHING here, since if EARLY_BRANCHING is not
        // set, dcd_early_branch will simply be a wire connected to zero and
        // this logic should just optimize.
        always @(posedge i_clk)
                if (op_ce)
                begin
                        opF_wr <= (dcdF_wr)&&((~dcdR_cc)||(~dcdR_wr))
                                &&(~dcd_early_branch)&&(~dcd_illegal);
                        opR_wr <= (dcdR_wr)&&(~dcd_early_branch)&&(~dcd_illegal);
                end
 
        always @(posedge i_clk)
                if (op_ce)
                begin
                        opn    <= dcdOp;        // Which ALU operation?
                        // opM  <= dcdM;        // Is this a memory operation?
                        // What register will these results be written into?
                        opR    <= dcdR;
                        opR_cc <= (dcdR_cc)&&(dcdR_wr)&&(dcdR[4]==dcd_gie);
                        // User level (1), vs supervisor (0)/interrupts disabled
                        op_gie <= dcd_gie;
 
 
                        //
                        op_pc  <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
                end
        assign  opFl = (op_gie)?(w_uflags):(w_iflags);
 
`ifdef  OPT_VLIW
        reg     r_op_phase;
        initial r_op_phase = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(clear_pipeline))
                        r_op_phase <= 1'b0;
                else if (op_ce)
                        r_op_phase <= dcd_phase;
        assign  op_phase = r_op_phase;
`else
        assign  op_phase = 1'b0;
`endif
 
        // This is tricky.  First, the PC and Flags registers aren't kept in
        // register set but in special registers of their own.  So step one
        // is to select the right register.  Step to is to replace that
        // register with the results of an ALU or memory operation, if such
        // results are now available.  Otherwise, we'd need to insert a wait
        // state of some type.
        //
        // The alternative approach would be to define some sort of
        // op_stall wire, which would stall any upstream stage.
        // We'll create a flag here to start our coordination.  Once we
        // define this flag to something other than just plain zero, then
        // the stalls will already be in place.
`ifdef  OPT_PIPELINED
        assign  opA = ((wr_reg_ce)&&(wr_reg_id == opA_id)) // &&(opA_rd))
                        ?  wr_reg_vl : r_opA;
`else
        assign  opA = r_opA;
`endif
 
`ifdef  OPT_PIPELINED
        // Stall if we have decoded an instruction that will read register A
        //      AND ... something that may write a register is running
        //      AND (series of conditions here ...)
        //              The operation might set flags, and we wish to read the
        //                      CC register
        //              OR ... (No other conditions)
        assign  dcdA_stall = (dcdA_rd) // &&(dcdvalid) is checked for elsewhere
                                &&((opvalid)||(mem_rdbusy)
                                        ||(div_busy)||(fpu_busy))
                                &&((opF_wr)&&(dcdA_cc));
`else
        // There are no pipeline hazards, if we aren't pipelined
        assign  dcdA_stall = 1'b0;
`endif
 
`ifdef  OPT_PIPELINED
        assign  opB = ((wr_reg_ce)&&(wr_reg_id == opB_id)&&(opB_rd))
                        ? wr_reg_vl: r_opB;
`else
        assign  opB = r_opB;
`endif
 
`ifdef  OPT_PIPELINED
        // Stall if we have decoded an instruction that will read register B
        //      AND ... something that may write a (unknown) register is running
        //      AND (series of conditions here ...)
        //              The operation might set flags, and we wish to read the
        //                      CC register
        //              OR the operation might set register B, and we still need
        //                      a clock to add the offset to it
        assign  dcdB_stall = (dcdB_rd) // &&(dcdvalid) is checked for elsewhere
                                // If the op stage isn't valid, yet something
                                // is running, then it must have been valid.
                                // We'll use the last values from that stage
                                // (opR_wr, opF_wr, opR) in our logic below.
                                &&((opvalid)||(mem_rdbusy)
                                        ||(div_busy)||(fpu_busy)||(alu_busy))
                                &&(
                                // Stall on memory ops writing to my register
                                //      (i.e. loads), or on any write to my
                                //      register if I have an immediate offset
                                //      Actually, this is worse.  I can't tell
                                //      whether or not my register is going to
                                //      be written to, so 
                                // Note the exception for writing to the PC:
                                //      if I write to the PC, the whole next
                                //      instruction is invalid, not just the
                                //      operand.  That'll get wiped in the
                                //      next operation anyway, so don't stall
                                //      here.  This keeps a BC X, BNZ Y from
                                //      stalling between the two branches.
                                //      BC X, BRA Y is still clear, since BRA Y
                                //      is an early branch instruction.
                                //      (This exception is commented out in
                                //      order to help keep our logic simple, and
                                //      because multiple conditional branches
                                //      following each other constitutes a
                                //      fairly unusualy code structure.)
                                //      
                                ((~dcd_zI)&&(
                                        ((opR == dcdB)&&(opR_wr))
                                        ||(((opvalid_mem)||(mem_rdbusy))
                                        &&(op_pipe))))
                                // Stall following any instruction that will
                                // set the flags, if we're going to need the
                                // flags (CC) register for opB.
                                ||((opF_wr)&&(dcdB_cc))
                                // Stall on any ongoing memory operation that
                                // will write to opB -- captured above
                                // ||((mem_busy)&&(~mem_we)&&(mem_last_reg==dcdB)&&(~dcd_zI))
                                );
        assign  dcdF_stall = ((~dcdF[3])
                                        ||((dcdA_rd)&&(dcdA_cc))
                                        ||((dcdB_rd)&&(dcdB_cc)))
                                        &&(opvalid)&&(opR_cc);
                                // &&(dcdvalid) is checked for elsewhere
`else
        // No stalls without pipelining, 'cause how can you have a pipeline
        // hazard without the pipeline?
        assign  dcdB_stall = 1'b0;
        assign  dcdF_stall = 1'b0;
`endif
        //
        //
        //      PIPELINE STAGE #4 :: Apply Instruction
        //
        //
`ifdef  OPT_NEW_INSTRUCTION_SET
        cpuops  #(IMPLEMENT_MPY) doalu(i_clk, i_rst, alu_ce,
                        (opvalid_alu), opn, opA, opB,
                        alu_result, alu_flags, alu_valid, alu_illegal_op,
                        alu_busy);
`else
        cpuops_deprecated       #(IMPLEMENT_MPY) doalu(i_clk, i_rst, alu_ce,
                        (opvalid_alu), opn, opA, opB,
                        alu_result, alu_flags, alu_valid, alu_illegal_op);
        assign  alu_busy = 1'b0;
`endif
 
        generate
        if (IMPLEMENT_DIVIDE != 0)
        begin
                div thedivide(i_clk, (i_rst)||(clear_pipeline), div_ce, opn[0],
                        opA, opB, div_busy, div_valid, div_error, div_result,
                        div_flags);
        end else begin
                assign  div_error = 1'b1;
                assign  div_busy  = 1'b0;
                assign  div_valid = 1'b0;
                assign  div_result= 32'h00;
                assign  div_flags = 4'h0;
        end endgenerate
 
        generate
        if (IMPLEMENT_FPU != 0)
        begin
                //
                // sfpu thefpu(i_clk, i_rst, fpu_ce,
                //      opA, opB, fpu_busy, fpu_valid, fpu_err, fpu_result,
                //      fpu_flags);
                //
                assign  fpu_error = 1'b1;
                assign  fpu_busy  = 1'b0;
                assign  fpu_valid = 1'b0;
                assign  fpu_result= 32'h00;
                assign  fpu_flags = 4'h0;
        end else begin
                assign  fpu_error = 1'b1;
                assign  fpu_busy  = 1'b0;
                assign  fpu_valid = 1'b0;
                assign  fpu_result= 32'h00;
                assign  fpu_flags = 4'h0;
        end endgenerate
 
 
        assign  set_cond = ((opF[7:4]&opFl[3:0])==opF[3:0]);
        initial alF_wr   = 1'b0;
        initial alu_wr   = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                begin
                        alu_wr   <= 1'b0;
                        alF_wr   <= 1'b0;
                end else if (alu_ce)
                begin
                        // alu_reg <= opR;
                        alu_wr  <= (opR_wr)&&(set_cond);
                        alF_wr  <= (opF_wr)&&(set_cond);
                end else if (~alu_busy) begin
                        // These are strobe signals, so clear them if not
                        // set for any particular clock
                        alu_wr <= (i_halt)&&(i_dbg_we);
                        alF_wr <= 1'b0;
                end
 
`ifdef  OPT_VLIW
        reg     r_alu_phase;
        initial r_alu_phase = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        r_alu_phase <= 1'b0;
                else if ((alu_ce)||(mem_ce)||(div_ce)||(fpu_ce))
                        r_alu_phase <= op_phase;
        assign  alu_phase = r_alu_phase;
`else
        assign  alu_phase = 1'b0;
`endif
 
        always @(posedge i_clk)
                if ((alu_ce)||(div_ce)||(fpu_ce))
                        alu_reg <= opR;
                else if ((i_halt)&&(i_dbg_we))
                        alu_reg <= i_dbg_reg;
 
        //
        // DEBUG Register write access starts here
        //
        reg             dbgv;
        initial dbgv = 1'b0;
        always @(posedge i_clk)
                dbgv <= (~i_rst)&&(~alu_ce)&&((i_halt)&&(i_dbg_we));
        reg     [31:0]   dbg_val;
        always @(posedge i_clk)
                dbg_val <= i_dbg_data;
        always @(posedge i_clk)
                if ((alu_ce)||(mem_ce))
                        alu_gie  <= op_gie;
        always @(posedge i_clk)
                if ((alu_ce)||((master_ce)&&(opvalid_mem)&&(~clear_pipeline)
                                &&(~mem_stalled)))
                        alu_pc  <= op_pc;
 
`ifdef  OPT_ILLEGAL_INSTRUCTION
        reg     r_alu_illegal;
        initial r_alu_illegal = 0;
        always @(posedge i_clk)
                if (clear_pipeline)
                        r_alu_illegal <= 1'b0;
                else if ((alu_ce)||(mem_ce))
                        r_alu_illegal <= op_illegal;
        assign  alu_illegal = (alu_illegal_op)||(r_alu_illegal);
`endif
 
        initial alu_pc_valid = 1'b0;
        initial mem_pc_valid = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        alu_pc_valid <= 1'b0;
                else
                        alu_pc_valid <= (alu_ce);
        always @(posedge i_clk)
                if (i_rst)
                        mem_pc_valid <= 1'b0;
                else
                        mem_pc_valid <= (mem_ce);
 
        wire    bus_lock;
`ifdef  OPT_PIPELINED
        generate
        if (IMPLEMENT_LOCK != 0)
        begin
                reg     r_bus_lock;
                initial r_bus_lock = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_bus_lock <= 1'b0;
                        else if ((op_ce)&&(op_lock))
                                r_bus_lock <= 1'b1;
                        else if (~opvalid_mem)
                                r_bus_lock <= 1'b0;
                assign  bus_lock = r_bus_lock;
        end else begin
                assign  bus_lock = 1'b0;
        end endgenerate
`else
        assign  bus_lock = 1'b0;
`endif
 
`ifdef  OPT_PIPELINED_BUS_ACCESS
        pipemem #(AW,IMPLEMENT_LOCK) domem(i_clk, i_rst,(mem_ce)&&(set_cond), bus_lock,
                                (opn[0]), opB, opA, opR,
                                mem_busy, mem_pipe_stalled,
                                mem_valid, bus_err, mem_wreg, mem_result,
                        mem_cyc_gbl, mem_cyc_lcl,
                                mem_stb_gbl, mem_stb_lcl,
                                mem_we, mem_addr, mem_data,
                                mem_ack, mem_stall, mem_err, i_wb_data);
 
`else // PIPELINED_BUS_ACCESS
        memops  #(AW,IMPLEMENT_LOCK) domem(i_clk, i_rst,(mem_ce)&&(set_cond), bus_lock,
                                (opn[0]), opB, opA, opR,
                                mem_busy,
                                mem_valid, bus_err, mem_wreg, mem_result,
                        mem_cyc_gbl, mem_cyc_lcl,
                                mem_stb_gbl, mem_stb_lcl,
                                mem_we, mem_addr, mem_data,
                                mem_ack, mem_stall, mem_err, i_wb_data);
`endif // PIPELINED_BUS_ACCESS
        assign  mem_rdbusy = ((mem_busy)&&(~mem_we));
 
        // Either the prefetch or the instruction gets the memory bus, but 
        // never both.
        wbdblpriarb     #(32,AW) pformem(i_clk, i_rst,
                // Memory access to the arbiter, priority position
                mem_cyc_gbl, mem_cyc_lcl, mem_stb_gbl, mem_stb_lcl,
                        mem_we, mem_addr, mem_data, mem_ack, mem_stall, mem_err,
                // Prefetch access to the arbiter
                pf_cyc, 1'b0, pf_stb, 1'b0, pf_we, pf_addr, pf_data,
                        pf_ack, pf_stall, pf_err,
                // Common wires, in and out, of the arbiter
                o_wb_gbl_cyc, o_wb_lcl_cyc, o_wb_gbl_stb, o_wb_lcl_stb,
                        o_wb_we, o_wb_addr, o_wb_data,
                        i_wb_ack, i_wb_stall, i_wb_err);
 
 
 
        //
        //
        //
        //
        //
        //
        //
        //
        //      PIPELINE STAGE #5 :: Write-back results
        //
        //
        // This stage is not allowed to stall.  If results are ready to be
        // written back, they are written back at all cost.  Sleepy CPU's
        // won't prevent write back, nor debug modes, halting the CPU, nor
        // anything else.  Indeed, the (master_ce) bit is only as relevant
        // as knowinig something is available for writeback.
 
        //
        // Write back to our generic register set ...
        // When shall we write back?  On one of two conditions
        //      Note that the flags needed to be checked before issuing the
        //      bus instruction, so they don't need to be checked here.
        //      Further, alu_wr includes (set_cond), so we don't need to
        //      check for that here either.
`ifdef  OPT_ILLEGAL_INSTRUCTION
        assign  wr_reg_ce = (dbgv)||(~alu_illegal)&&
                        (((alu_wr)&&(~clear_pipeline)
                                &&((alu_valid)||(div_valid)||(fpu_valid)))
                        ||(mem_valid));
`else
        assign  wr_reg_ce = (dbgv)||((alu_wr)&&(~clear_pipeline))||(mem_valid)||(div_valid)||(fpu_valid);
`endif
        // Which register shall be written?
        //      COULD SIMPLIFY THIS: by adding three bits to these registers,
        //              One or PC, one for CC, and one for GIE match
        //      Note that the alu_reg is the register to write on a divide or
        //      FPU operation.
        assign  wr_reg_id = (alu_wr)?alu_reg:mem_wreg;
        // Are we writing to the CC register?
        assign  wr_write_cc = (wr_reg_id[3:0] == `CPU_CC_REG);
        // Are we writing to the PC?
        assign  wr_write_pc = (wr_reg_id[3:0] == `CPU_PC_REG);
        // What value to write?
        assign  wr_reg_vl = ((mem_valid) ? mem_result
                                :((div_valid|fpu_valid))
                                        ? ((div_valid) ? div_result:fpu_result)
                                :((dbgv) ? dbg_val : alu_result));
        always @(posedge i_clk)
                if (wr_reg_ce)
                        regset[wr_reg_id] <= wr_reg_vl;
 
        //
        // Write back to the condition codes/flags register ...
        // When shall we write to our flags register?  alF_wr already
        // includes the set condition ...
        assign  wr_flags_ce = ((alF_wr)||(div_valid)||(fpu_valid))&&(~clear_pipeline)&&(~alu_illegal);
        assign  w_uflags = { uhalt_phase, ufpu_err_flag,
                        udiv_err_flag, ubus_err_flag, trap, ill_err_u,
                        1'b0, step, 1'b1, sleep,
                        ((wr_flags_ce)&&(alu_gie))?alu_flags:flags };
        assign  w_iflags = { ihalt_phase, ifpu_err_flag,
                        idiv_err_flag, ibus_err_flag, trap, ill_err_i,
                        break_en, 1'b0, 1'b0, sleep,
                        ((wr_flags_ce)&&(~alu_gie))?alu_flags:iflags };
 
 
        // What value to write?
        always @(posedge i_clk)
                // If explicitly writing the register itself
                if ((wr_reg_ce)&&(wr_reg_id[4])&&(wr_write_cc))
                        flags <= wr_reg_vl[3:0];
                // Otherwise if we're setting the flags from an ALU operation
                else if ((wr_flags_ce)&&(alu_gie))
                        flags <= (div_valid)?div_flags:((fpu_valid)?fpu_flags
                                : alu_flags);
 
        always @(posedge i_clk)
                if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_cc))
                        iflags <= wr_reg_vl[3:0];
                else if ((wr_flags_ce)&&(~alu_gie))
                        iflags <= (div_valid)?div_flags:((fpu_valid)?fpu_flags
                                : alu_flags);
 
        // The 'break' enable  bit.  This bit can only be set from supervisor
        // mode.  It control what the CPU does upon encountering a break
        // instruction.
        //
        // The goal, upon encountering a break is that the CPU should stop and
        // not execute the break instruction, choosing instead to enter into
        // either interrupt mode or halt first.  
        //      if ((break_en) AND (break_instruction)) // user mode or not
        //              HALT CPU
        //      else if (break_instruction) // only in user mode
        //              set an interrupt flag, go to supervisor mode
        //              allow supervisor to step the CPU.
        //      Upon a CPU halt, any break condition will be reset.  The
        //      external debugger will then need to deal with whatever
        //      condition has taken place.
        initial break_en = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(i_halt))
                        break_en <= 1'b0;
                else if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_cc))
                        break_en <= wr_reg_vl[`CPU_BREAK_BIT];
`ifdef  OPT_ILLEGAL_INSTRUCTION
        assign  o_break = ((break_en)||(~op_gie))&&(op_break)
                                &&(~alu_valid)&&(~mem_valid)&&(~mem_busy)
                                &&(~div_busy)&&(~fpu_busy)
                                &&(~clear_pipeline)
                        ||((~alu_gie)&&(bus_err))
                        ||((~alu_gie)&&(div_valid)&&(div_error))
                        ||((~alu_gie)&&(fpu_valid)&&(fpu_error))
                        ||((~alu_gie)&&(alu_pc_valid)&&(alu_illegal));
`else
        assign  o_break = (((break_en)||(~op_gie))&&(op_break)
                                &&(~alu_valid)&&(~mem_valid)&&(~mem_busy)
                                &&(~clear_pipeline))
                        ||((~alu_gie)&&(bus_err))
                        ||((~alu_gie)&&(div_valid)&&(div_error))
                        ||((~alu_gie)&&(fpu_valid)&&(fpu_error));
`endif
 
 
        // The sleep register.  Setting the sleep register causes the CPU to
        // sleep until the next interrupt.  Setting the sleep register within
        // interrupt mode causes the processor to halt until a reset.  This is
        // a panic/fault halt.  The trick is that you cannot be allowed to
        // set the sleep bit and switch to supervisor mode in the same 
        // instruction: users are not allowed to halt the CPU.
        always @(posedge i_clk)
                if ((i_rst)||(w_switch_to_interrupt))
                        sleep <= 1'b0;
                else if ((wr_reg_ce)&&(wr_write_cc)&&(~alu_gie))
                        // In supervisor mode, we have no protections.  The
                        // supervisor can set the sleep bit however he wants.
                        // Well ... not quite.  Switching to user mode and
                        // sleep mode shouold only be possible if the interrupt
                        // flag isn't set.
                        //      Thus: if (i_interrupt)&&(wr_reg_vl[GIE])
                        //              don't set the sleep bit
                        //      otherwise however it would o.w. be set
                        sleep <= (wr_reg_vl[`CPU_SLEEP_BIT])
                                &&((~i_interrupt)||(~wr_reg_vl[`CPU_GIE_BIT]));
                else if ((wr_reg_ce)&&(wr_write_cc)&&(wr_reg_vl[`CPU_GIE_BIT]))
                        // In user mode, however, you can only set the sleep
                        // mode while remaining in user mode.  You can't switch
                        // to sleep mode *and* supervisor mode at the same
                        // time, lest you halt the CPU.
                        sleep <= wr_reg_vl[`CPU_SLEEP_BIT];
 
        always @(posedge i_clk)
                if ((i_rst)||(w_switch_to_interrupt))
                        step <= 1'b0;
                else if ((wr_reg_ce)&&(~alu_gie)&&(wr_reg_id[4])&&(wr_write_cc))
                        step <= wr_reg_vl[`CPU_STEP_BIT];
                else if (((alu_pc_valid)||(mem_pc_valid))&&(step)&&(gie))
                        step <= 1'b0;
 
        // The GIE register.  Only interrupts can disable the interrupt register
        assign  w_switch_to_interrupt = (gie)&&(
                        // On interrupt (obviously)
                        ((i_interrupt)&&(~alu_phase)&&(~bus_lock))
                        // If we are stepping the CPU
                        ||(((alu_pc_valid)||(mem_pc_valid))&&(step)&&(~alu_phase)&&(~bus_lock))
                        // If we encounter a break instruction, if the break
                        //      enable isn't set.
                        ||((master_ce)&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
                                &&(op_break)&&(~break_en))
`ifdef  OPT_ILLEGAL_INSTRUCTION
                        // On an illegal instruction
                        ||((alu_pc_valid)&&(alu_illegal))
`endif
                        // On division by zero.  If the divide isn't
                        // implemented, div_valid and div_error will be short
                        // circuited and that logic will be bypassed
                        ||((div_valid)&&(div_error))
                        // Same thing on a floating point error.
                        ||((fpu_valid)&&(fpu_error))
                        //      
                        ||(bus_err)
                        // If we write to the CC register
                        ||((wr_reg_ce)&&(~wr_reg_vl[`CPU_GIE_BIT])
                                &&(wr_reg_id[4])&&(wr_write_cc))
                        );
        assign  w_release_from_interrupt = (~gie)&&(~i_interrupt)
                        // Then if we write the CC register
                        &&(((wr_reg_ce)&&(wr_reg_vl[`CPU_GIE_BIT])
                                &&(~wr_reg_id[4])&&(wr_write_cc))
                        );
        always @(posedge i_clk)
                if (i_rst)
                        gie <= 1'b0;
                else if (w_switch_to_interrupt)
                        gie <= 1'b0;
                else if (w_release_from_interrupt)
                        gie <= 1'b1;
 
        initial trap = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        trap <= 1'b0;
                else if (w_release_from_interrupt)
                        trap <= 1'b0;
                else if ((alu_gie)&&(wr_reg_ce)&&(~wr_reg_vl[`CPU_GIE_BIT])
                                &&(wr_write_cc)) // &&(wr_reg_id[4]) implied
                        trap <= 1'b1;
                else if ((wr_reg_ce)&&(wr_write_cc)&&(wr_reg_id[4]))
                        trap <= wr_reg_vl[`CPU_TRAP_BIT];
 
`ifdef  OPT_ILLEGAL_INSTRUCTION
        initial ill_err_i = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        ill_err_i <= 1'b0;
                // Only the debug interface can clear this bit
                else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
                                &&(~wr_reg_vl[`CPU_ILL_BIT]))
                        ill_err_i <= 1'b0;
                else if ((alu_pc_valid)&&(alu_illegal)&&(~alu_gie))
                        ill_err_i <= 1'b1;
        initial ill_err_u = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        ill_err_u <= 1'b0;
                // The bit is automatically cleared on release from interrupt
                else if (w_release_from_interrupt)
                        ill_err_u <= 1'b0;
                // If the supervisor writes to this register, clearing the
                // bit, then clear it
                else if (((~alu_gie)||(dbgv))
                                &&(wr_reg_ce)&&(~wr_reg_vl[`CPU_ILL_BIT])
                                &&(wr_reg_id[4])&&(wr_write_cc))
                        ill_err_u <= 1'b0;
                else if ((alu_pc_valid)&&(alu_illegal)&&(alu_gie))
                        ill_err_u <= 1'b1;
`else
        assign ill_err_u = 1'b0;
        assign ill_err_i = 1'b0;
`endif
        // Supervisor/interrupt bus error flag -- this will crash the CPU if
        // ever set.
        initial ibus_err_flag = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        ibus_err_flag <= 1'b0;
                else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
                                &&(~wr_reg_vl[`CPU_BUSERR_BIT]))
                        ibus_err_flag <= 1'b0;
                else if ((bus_err)&&(~alu_gie))
                        ibus_err_flag <= 1'b1;
        // User bus error flag -- if ever set, it will cause an interrupt to
        // supervisor mode.  
        initial ubus_err_flag = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        ubus_err_flag <= 1'b0;
                else if (w_release_from_interrupt)
                        ubus_err_flag <= 1'b0;
                else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
                                &&(~wr_reg_vl[`CPU_BUSERR_BIT])
                                &&(wr_reg_id[4])&&(wr_write_cc))
                        ubus_err_flag <= 1'b0;
                else if ((bus_err)&&(alu_gie))
                        ubus_err_flag <= 1'b1;
 
        generate
        if (IMPLEMENT_DIVIDE != 0)
        begin
                reg     r_idiv_err_flag, r_udiv_err_flag;
 
                // Supervisor/interrupt divide (by zero) error flag -- this will
                // crash the CPU if ever set.  This bit is thus available for us
                // to be able to tell if/why the CPU crashed.
                initial r_idiv_err_flag = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_idiv_err_flag <= 1'b0;
                        else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
                                        &&(~wr_reg_vl[`CPU_DIVERR_BIT]))
                                r_idiv_err_flag <= 1'b0;
                        else if ((div_error)&&(div_valid)&&(~alu_gie))
                                r_idiv_err_flag <= 1'b1;
                // User divide (by zero) error flag -- if ever set, it will
                // cause a sudden switch interrupt to supervisor mode.  
                initial r_udiv_err_flag = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_udiv_err_flag <= 1'b0;
                        else if (w_release_from_interrupt)
                                r_udiv_err_flag <= 1'b0;
                        else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
                                        &&(~wr_reg_vl[`CPU_DIVERR_BIT])
                                        &&(wr_reg_id[4])&&(wr_write_cc))
                                r_udiv_err_flag <= 1'b0;
                        else if ((div_error)&&(alu_gie)&&(div_valid))
                                r_udiv_err_flag <= 1'b1;
 
                assign  idiv_err_flag = r_idiv_err_flag;
                assign  udiv_err_flag = r_udiv_err_flag;
        end else begin
                assign  idiv_err_flag = 1'b0;
                assign  udiv_err_flag = 1'b0;
        end endgenerate
 
        generate
        if (IMPLEMENT_FPU !=0)
        begin
                // Supervisor/interrupt floating point error flag -- this will
                // crash the CPU if ever set.
                reg             r_ifpu_err_flag, r_ufpu_err_flag;
                initial r_ifpu_err_flag = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_ifpu_err_flag <= 1'b0;
                        else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
                                        &&(~wr_reg_vl[`CPU_FPUERR_BIT]))
                                r_ifpu_err_flag <= 1'b0;
                        else if ((fpu_error)&&(fpu_valid)&&(~alu_gie))
                                r_ifpu_err_flag <= 1'b1;
                // User floating point error flag -- if ever set, it will cause
                // a sudden switch interrupt to supervisor mode.  
                initial r_ufpu_err_flag = 1'b0;
                always @(posedge i_clk)
                        if (i_rst)
                                r_ufpu_err_flag <= 1'b0;
                        else if (w_release_from_interrupt)
                                r_ufpu_err_flag <= 1'b0;
                        else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
                                        &&(~wr_reg_vl[`CPU_FPUERR_BIT])
                                        &&(wr_reg_id[4])&&(wr_write_cc))
                                r_ufpu_err_flag <= 1'b0;
                        else if ((fpu_error)&&(alu_gie)&&(fpu_valid))
                                r_ufpu_err_flag <= 1'b1;
 
                assign  ifpu_err_flag = r_ifpu_err_flag;
                assign  ufpu_err_flag = r_ufpu_err_flag;
        end else begin
                assign  ifpu_err_flag = 1'b0;
                assign  ufpu_err_flag = 1'b0;
        end endgenerate
 
`ifdef  OPT_VLIW
        reg             r_ihalt_phase, r_uhalt_phase;
 
        initial r_ihalt_phase = 0;
        initial r_uhalt_phase = 0;
        always @(posedge i_clk)
                if (~alu_gie)
                        r_ihalt_phase <= alu_phase;
        always @(posedge i_clk)
                if (alu_gie)
                        r_uhalt_phase <= alu_phase;
                else if (w_release_from_interrupt)
                        r_uhalt_phase <= 1'b0;
 
        assign  ihalt_phase = r_ihalt_phase;
        assign  uhalt_phase = r_uhalt_phase;
`else
        assign  ihalt_phase = 1'b0;
        assign  uhalt_phase = 1'b0;
`endif
 
        //
        // Write backs to the PC register, and general increments of it
        //      We support two: upc and ipc.  If the instruction is normal,
        // we increment upc, if interrupt level we increment ipc.  If
        // the instruction writes the PC, we write whichever PC is appropriate.
        //
        // Do we need to all our partial results from the pipeline?
        // What happens when the pipeline has gie and ~gie instructions within
        // it?  Do we clear both?  What if a gie instruction tries to clear
        // a non-gie instruction?
        always @(posedge i_clk)
                if ((wr_reg_ce)&&(wr_reg_id[4])&&(wr_write_pc))
                        upc <= wr_reg_vl[(AW-1):0];
                else if ((alu_gie)&&
                                (((alu_pc_valid)&&(~clear_pipeline))
                                ||(mem_pc_valid)))
                        upc <= alu_pc;
 
        always @(posedge i_clk)
                if (i_rst)
                        ipc <= RESET_ADDRESS;
                else if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_pc))
                        ipc <= wr_reg_vl[(AW-1):0];
                else if ((~alu_gie)&&
                                (((alu_pc_valid)&&(~clear_pipeline))
                                ||(mem_pc_valid)))
                        ipc <= alu_pc;
 
        always @(posedge i_clk)
                if (i_rst)
                        pf_pc <= RESET_ADDRESS;
                else if (w_switch_to_interrupt)
                        pf_pc <= ipc;
                else if (w_release_from_interrupt)
                        pf_pc <= upc;
                else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
                        pf_pc <= wr_reg_vl[(AW-1):0];
`ifdef  OPT_PIPELINED
                else if ((dcd_early_branch)&&(~clear_pipeline))
                        pf_pc <= dcd_branch_pc + 1;
                else if ((new_pc)||((~dcd_stalled)&&(pf_valid)))
                        pf_pc <= pf_pc + {{(AW-1){1'b0}},1'b1};
`else
                else if ((alu_pc_valid)&&(~clear_pipeline))
                        pf_pc <= alu_pc;
`endif
 
        initial new_pc = 1'b1;
        always @(posedge i_clk)
                if ((i_rst)||(i_clear_pf_cache))
                        new_pc <= 1'b1;
                else if (w_switch_to_interrupt)
                        new_pc <= 1'b1;
                else if (w_release_from_interrupt)
                        new_pc <= 1'b1;
                else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
                        new_pc <= 1'b1;
                else
                        new_pc <= 1'b0;
 
        //
        // The debug interface
        generate
        if (AW<32)
        begin
                always @(posedge i_clk)
                begin
                        o_dbg_reg <= regset[i_dbg_reg];
                        if (i_dbg_reg[3:0] == `CPU_PC_REG)
                                o_dbg_reg <= {{(32-AW){1'b0}},(i_dbg_reg[4])?upc:ipc};
                        else if (i_dbg_reg[3:0] == `CPU_CC_REG)
                        begin
                                o_dbg_reg[13:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
                                o_dbg_reg[`CPU_GIE_BIT] <= gie;
                        end
                end
        end else begin
                always @(posedge i_clk)
                begin
                        o_dbg_reg <= regset[i_dbg_reg];
                        if (i_dbg_reg[3:0] == `CPU_PC_REG)
                                o_dbg_reg <= (i_dbg_reg[4])?upc:ipc;
                        else if (i_dbg_reg[3:0] == `CPU_CC_REG)
                        begin
                                o_dbg_reg[13:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
                                o_dbg_reg[`CPU_GIE_BIT] <= gie;
                        end
                end
        end endgenerate
 
        always @(posedge i_clk)
                o_dbg_cc <= { o_break, bus_err, gie, sleep };
 
        always @(posedge i_clk)
                o_dbg_stall <= (i_halt)&&(
                        (pf_cyc)||(mem_cyc_gbl)||(mem_cyc_lcl)||(mem_busy)
                        ||((~opvalid)&&(~i_rst)&&(~dcd_illegal))
                        ||((~dcdvalid)&&(~i_rst)&&(~pf_illegal)));
 
        //
        //
        // Produce accounting outputs: Account for any CPU stalls, so we can
        // later evaluate how well we are doing.
        //
        //
        assign  o_op_stall = (master_ce)&&(op_stall);
        assign  o_pf_stall = (master_ce)&&(~pf_valid);
        assign  o_i_count  = (alu_pc_valid)&&(~clear_pipeline);
 
`ifdef  DEBUG_SCOPE
        always @(posedge i_clk)
                o_debug <= {
                        o_break, i_wb_err, pf_pc[1:0],
                        flags,
                        pf_valid, dcdvalid, opvalid, alu_valid, mem_valid,
                        op_ce, alu_ce, mem_ce,
                        //
                        master_ce, opvalid_alu, opvalid_mem,
                        //
                        alu_stall, mem_busy, op_pipe, mem_pipe_stalled,
                        mem_we,
                        // ((opvalid_alu)&&(alu_stall))
                        // ||((opvalid_mem)&&(~op_pipe)&&(mem_busy))
                        // ||((opvalid_mem)&&( op_pipe)&&(mem_pipe_stalled)));
                        // opA[23:20], opA[3:0],
                        gie, sleep, wr_reg_ce, wr_reg_vl[4:0]
                /*
                        i_rst, master_ce, (new_pc),
                        ((dcd_early_branch)&&(dcdvalid)),
                        pf_valid, pf_illegal,
                        op_ce, dcd_ce, dcdvalid, dcd_stalled,
                        pf_cyc, pf_stb, pf_we, pf_ack, pf_stall, pf_err,
                        pf_pc[7:0], pf_addr[7:0]
                */
                /*
                        i_wb_err, gie, alu_illegal,
                              (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
                        mem_busy,
                                (mem_busy)?{ (o_wb_gbl_stb|o_wb_lcl_stb), o_wb_we,
                                        o_wb_addr[8:0] }
                                        : { instruction[31:21] },
                        pf_valid, (pf_valid) ? alu_pc[14:0]
                                :{ pf_cyc, pf_stb, pf_pc[12:0] }
                */
                /*
                        i_wb_err, gie, new_pc, dcd_early_branch,        // 4
                        pf_valid, pf_cyc, pf_stb, instruction_pc[0],    // 4
                        instruction[30:27],                             // 4
                        dcd_gie, mem_busy, o_wb_gbl_cyc, o_wb_gbl_stb,  // 4
                        dcdvalid,
                        ((dcd_early_branch)&&(~clear_pipeline))         // 15
                                        ? dcd_branch_pc[14:0]:pf_pc[14:0]
                */
                        };
`endif
 
endmodule
Browse

Tools

Subversion Repositories s6soc

[/] [s6soc/] [trunk/] [rtl/] [cpu/] [zipcpu.v] - Blame information for rev 7

Line No.	Rev	Author	Line
1	2	dgisselq	`///////////////////////////////////////////////////////////////////////////////`
2			`//`
3			`// Filename: zipcpu.v`
4			`//`
5			`// Project: Zip CPU -- a small, lightweight, RISC CPU soft core`
6			`//`
7			`// Purpose: This is the top level module holding the core of the Zip CPU`
8			`// together. The Zip CPU is designed to be as simple as possible.`
9			`// (actual implementation aside ...) The instruction set is about as`
10			`// RISC as you can get, there are only 16 instruction types supported.`
11			`// Please see the accompanying spec.pdf file for a description of these`
12			`// instructions.`
13			`//`
14			`// All instructions are 32-bits wide. All bus accesses, both address and`
15			`// data, are 32-bits over a wishbone bus.`
16			`//`
17			`// The Zip CPU is fully pipelined with the following pipeline stages:`
18			`//`
19			`// 1. Prefetch, returns the instruction from memory.`
20			`//`
21			`// 2. Instruction Decode`
22			`//`
23			`// 3. Read Operands`
24			`//`
25			`// 4. Apply Instruction`
26			`//`
27			`// 4. Write-back Results`
28			`//`
29			`// Further information about the inner workings of this CPU may be`
30			`// found in the spec.pdf file. (The documentation within this file`
31			`// had become out of date and out of sync with the spec.pdf, so look`
32			`// to the spec.pdf for accurate and up to date information.)`
33			`//`
34			`//`
35			`// In general, the pipelining is controlled by three pieces of logic`
36			`// per stage: _ce, _stall, and _valid. _valid means that the stage`
37			`// holds a valid instruction. _ce means that the instruction from the`
38			`// previous stage is to move into this one, and _stall means that the`
39			`// instruction from the previous stage may not move into this one.`
40			`// The difference between these control signals allows individual stages`