URL https://opencores.org/ocsvn/zipcpu/zipcpu/trunk

Subversion Repositories zipcpu

[/] [zipcpu/] [trunk/] [rtl/] [core/] [zipcpu.v] - Blame information for rev 187

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, with only 26 instruction types currently supported.
//      (There are still 8-instruction Op-Codes reserved for floating point,
//      and 5 which can be used for transactions not requiring registers.)
//      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, such as
//      what causes pipeline stalls, 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//      documentation within this file had become out of date and out of sync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) 2015-2016, 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_CLRCACHE_BIT 14     // Floating point error flag, set on error
`define CPU_PHASE_BIT   13      // Floating point error flag, set on error
`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=8;
`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  wire            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    [14:0]   w_uflags, w_iflags;
        reg             trap, break_en, step, gie, sleep, r_halted;
        wire            break_pending;
        wire            w_clear_icache;
`ifdef  OPT_ILLEGAL_INSTRUCTION
        reg             ill_err_u, ill_err_i;
`else
        wire            ill_err_u, ill_err_i;
`endif
        reg             ubreak;
        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;
 
        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;
        wire    [3:0]    opn;
        wire    [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, opF_wr;
        wire            op_gie, opR_cc;
        wire    [14:0]   opFl;
        reg     [5:0]    r_opF;
        wire    [7:0]    opF;
        wire            op_ce, op_phase, op_pipe, op_change_data_ce;
        // 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;
`else
        wire    op_illegal;
        assign  op_illegal = 1'b0;
`endif
        wire    op_break;
        wire    op_lock;
 
 
        //
        //
        //      PIPELINE STAGE #4 :: ALU / Memory
        //              Variable declarations
        //
        //
        wire    [(AW-1):0]       alu_pc;
        reg             r_alu_pc_valid, mem_pc_valid;
        wire            alu_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;
        wire            alu_gie, alu_illegal_op, 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);
 
        wire    adf_ce_unconditional;
 
        //
        //
        //      PIPELINE STAGE #5 :: Write-back
        //              Variable declarations
        //
        wire            wr_reg_ce, wr_flags_ce, wr_write_pc, wr_write_cc,
                        wr_write_scc, wr_write_ucc;
        wire    [4:0]    wr_reg_id;
        wire    [31:0]   wr_gpreg_vl, wr_spreg_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
        assign          dcd_ce = ((~dcdvalid)||(~dcd_stalled))&&(~clear_pipeline);
 
`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
        reg     cc_invalid_for_dcd;
        always @(posedge i_clk)
                cc_invalid_for_dcd <= (wr_flags_ce)
                        ||(wr_reg_ce)&&(wr_reg_id[3:0] == `CPU_CC_REG)
                        ||(opvalid)&&((opF_wr)||((opR_wr)&&(opR[3:0] == `CPU_CC_REG)))
                        ||((alF_wr)||((alu_wr)&&(alu_reg[3:0] == `CPU_CC_REG)))
                        ||(mem_busy)||(div_busy)||(fpu_busy);
 
        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
                        ||(opR_cc)
                        )
                        ||(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);
 
 
        // BUT ... op_ce is too complex for many of the data operations.  So
        // let's make their circuit enable code simpler.  In particular, if
        // op_ doesn't need to be preserved, we can change it all we want
        // ... right?  The clear_pipeline code, for example, really only needs
        // to determine whether opvalid is true.
        assign  op_change_data_ce = (~op_stall);
`else
        assign  op_stall = (opvalid)&&(~master_ce);
        assign  op_ce = ((dcdvalid)||(dcd_illegal))&&(~clear_pipeline);
        assign  op_change_data_ce = 1'b1;
`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
                        ||((opvalid)&&(op_lock)&&(op_lock_stall))
                        ||((opvalid)&&(op_break))
                        ||(wr_reg_ce)&&(wr_write_cc)
                        ||(div_busy)||(fpu_busy);
        assign  alu_ce = (master_ce)&&(opvalid_alu)&&(~alu_stall)
                                &&(~clear_pipeline);
`else
        assign  alu_stall = (opvalid_alu)&&((~master_ce)||(op_break));
        assign  alu_ce = (master_ce)&&((opvalid_alu)||(op_illegal))&&(~alu_stall)&&(~clear_pipeline);
`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.
        //
        // However, in hind sight this logic didn't work.  What happens when
        // something gets in the pipeline and then (due to interrupt or some
        // such) needs to be voided?  Thus we avoid simplification and keep
        // what worked here.
        assign  mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)
                        &&(~clear_pipeline);
`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
 
        // ALU, DIV, or FPU CE ... equivalent to the OR of all three of these
        assign  adf_ce_unconditional = (master_ce)&&(~clear_pipeline)&&(opvalid)
                                &&(~opvalid_mem)&&(~mem_rdbusy)
                                &&((~opvalid_alu)||(~alu_stall))&&(~op_break)
                                &&(~div_busy)&&(~fpu_busy)&&(~clear_pipeline);
 
        //
        //
        //      PIPELINE STAGE #1 :: Prefetch
        //
        //
`ifdef  OPT_SINGLE_FETCH
        wire            pf_ce;
 
        assign          pf_ce = (~pf_valid)&&(~dcdvalid)&&(~opvalid)&&(~alu_busy)&&(~mem_busy)&&(~alu_pc_valid)&&(~mem_pc_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)||(clear_pipeline))
                        r_dcdvalid <= 1'b0;
                else if (dcd_ce)
                        r_dcdvalid <= (pf_valid)||(pf_illegal);
                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)),
                                        w_clear_icache,
                                // 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),
                                        w_clear_icache, ~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)||(w_clear_icache))
                        r_dcdvalid <= 1'b0;
                else if (dcd_ce)
                        r_dcdvalid <= (pf_valid)&&(~dcd_ljmp)&&(~dcd_early_branch);
                else if (op_ce)
                        r_dcdvalid <= 1'b0;
        assign  dcdvalid = r_dcdvalid;
`endif
 
`ifdef  OPT_NEW_INSTRUCTION_SET
 
        // If not pipelined, there will be no opvalid_ anything, and the
        idecode #(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
                        IMPLEMENT_FPU)
                instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
                        (~dcdvalid)||(~op_stall), 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;
                else if (mem_ce) // Clear us any time an op_ is clocked in
                        r_op_pipe <= 1'b0;
        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)
`ifdef  OPT_PIPELINED
                if (op_change_data_ce)
`endif
                begin
`ifdef  OPT_PIPELINED
                        if ((wr_reg_ce)&&(wr_reg_id == dcdA))
                                r_opA <= wr_gpreg_vl;
                        else
`endif
                        if (dcdA_pc)
                                r_opA <= w_pcA_v;
                        else if (dcdA_cc)
                                r_opA <= { w_cpu_info, w_opA[22:16], 1'b0, (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_gpreg_vl;
                        if ((wr_reg_ce)&&(wr_reg_id == opA_id)&&(opA_rd))
                                r_opA <= wr_gpreg_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
`ifdef  OPT_PIPELINED
                : ((wr_reg_ce)&&(wr_reg_id == dcdB)) ? wr_gpreg_vl
`endif
                : ((dcdB_pc) ? w_pcB_v
                : ((dcdB_cc) ? { w_cpu_info, w_opB[22:16], // w_opB[31:14],
                        1'b0, (dcdB[4])?w_uflags:w_iflags}
                : w_opB));
 
        always @(posedge i_clk)
`ifdef  OPT_PIPELINED
                if (op_change_data_ce)
                        r_opB <= w_opBnI + dcdI;
                else if ((wr_reg_ce)&&(opB_id == wr_reg_id)&&(opB_rd))
                        r_opB <= wr_gpreg_vl;
`else
                r_opB <= w_opBnI + dcdI;
`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)
`ifdef  OPT_PIPELINED
                if (op_ce) // Cannot do op_change_data_ce here since opF depends
                        // upon being either correct for a valid op, or correct
                        // for the last valid op
`endif
                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;
                        opvalid_div <= 1'b0;
                        opvalid_fpu <= 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)||(dcd_illegal)&&(dcdvalid);
`ifdef  OPT_ILLEGAL_INSTRUCTION
                        opvalid_alu <= (w_opvalid)&&((dcdALU)||(dcd_illegal));
                        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)||(adf_ce_unconditional)||(mem_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);
`ifdef  OPT_PIPELINED
        reg     r_op_break;
 
        initial r_op_break = 1'b0;
        always @(posedge i_clk)
                if (i_rst)      r_op_break <= 1'b0;
                else if (op_ce) r_op_break <= (dcd_break); // &&(dcdvalid)
                else if ((clear_pipeline)||(~opvalid))
                                r_op_break <= 1'b0;
        assign  op_break = r_op_break;
`else
        assign  op_break = dcd_break;
`endif
 
`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)||(clear_pipeline))
                                r_op_lock <= 1'b0;
                        else if (op_ce)
                                r_op_lock <= (dcd_lock)&&(~clear_pipeline);
                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 <= (dcdvalid)&&((dcd_illegal)||((dcd_lock)&&(IMPLEMENT_LOCK == 0)));
`else
                        op_illegal <= (dcdvalid)&&((dcd_illegal)||(dcd_lock));
`endif
                else if(alu_ce)
                        op_illegal <= 1'b0;
`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.
`ifdef  OPT_PIPELINED
        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
`else
        always @(posedge i_clk)
        begin
                opF_wr <= (dcdF_wr)&&((~dcdR_cc)||(~dcdR_wr))
                        &&(~dcd_early_branch)&&(~dcd_illegal);
                opR_wr <= (dcdR_wr)&&(~dcd_early_branch)&&(~dcd_illegal);
        end
`endif
 
`ifdef  OPT_PIPELINED
        reg     [3:0]    r_opn;
        reg     [4:0]    r_opR;
        reg             r_opR_cc;
        reg             r_op_gie;
        always @(posedge i_clk)
                if (op_change_data_ce)
                begin
                        r_opn    <= dcdOp;      // Which ALU operation?
                        // opM  <= dcdM;        // Is this a memory operation?
                        // What register will these results be written into?
                        r_opR    <= dcdR;
                        r_opR_cc <= (dcdR_cc)&&(dcdR_wr)&&(dcdR[4]==dcd_gie);
                        // User level (1), vs supervisor (0)/interrupts disabled
                        r_op_gie <= dcd_gie;
 
 
                        //
                        op_pc  <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
                end
        assign  opn = r_opn;
        assign  opR = r_opR;
        assign  op_gie = r_op_gie;
        assign  opR_cc = r_opR_cc;
`else
        assign  opn = dcdOp;
        assign  opR = dcdR;
        assign  op_gie = dcd_gie;
        // With no pipelining, there is no early branching.  We keep it
        always @(posedge i_clk)
                op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
`endif
        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_change_data_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_gpreg_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)||(cc_invalid_for_dcd))&&(dcdA_cc))
                        ||((dcdA_rd)&&(dcdA_cc)&&(cc_invalid_for_dcd));
`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_gpreg_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))
                                &&(
                                // Okay, what happens if the result register
                                // from instruction 1 becomes the input for
                                // instruction two, *and* there's an immediate
                                // offset in instruction two?  In that case, we
                                // need an extra clock between the two 
                                // instructions to calculate the base plus 
                                // offset.
                                //
                                // What if instruction 1 (or before) is in a
                                // memory pipeline?  We may no longer know what
                                // the register was!  We will then need  to 
                                // blindly wait.  We'll temper this only waiting
                                // if we're not piping this new instruction.
                                // If we were piping, the pipe logic in the
                                // decode circuit has told us that the hazard
                                // is clear, so we're okay then.
                                //
                                ((~dcd_zI)&&(
                                        ((opR == dcdB)&&(opR_wr))
                                        ||((mem_rdbusy)&&(~dcd_pipe))
                                        ))
                                // Stall following any instruction that will
                                // set the flags, if we're going to need the
                                // flags (CC) register for opB.
                                ||(((opF_wr)||(cc_invalid_for_dcd))&&(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))
                                )
                        ||((dcdB_rd)&&(dcdB_cc)&&(cc_invalid_for_dcd));
        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'b0; // Can't be high unless div_valid
                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'b0; // Must only be true if fpu_valid
                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'b0;
                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 ((adf_ce_unconditional)||(mem_ce))
                        r_alu_phase <= op_phase;
        assign  alu_phase = r_alu_phase;
`else
        assign  alu_phase = 1'b0;
`endif
 
`ifdef  OPT_PIPELINED
        always @(posedge i_clk)
                if (adf_ce_unconditional)
                        alu_reg <= opR;
                else if ((i_halt)&&(i_dbg_we))
                        alu_reg <= i_dbg_reg;
`else
        always @(posedge i_clk)
                if ((i_halt)&&(i_dbg_we))
                        alu_reg <= i_dbg_reg;
                else
                        alu_reg <= opR;
`endif
 
        //
        // DEBUG Register write access starts here
        //
        reg             dbgv;
        initial dbgv = 1'b0;
        always @(posedge i_clk)
                dbgv <= (~i_rst)&&(i_halt)&&(i_dbg_we)&&(r_halted);
        reg     [31:0]   dbg_val;
        always @(posedge i_clk)
                dbg_val <= i_dbg_data;
`ifdef  OPT_PIPELINED
        reg     r_alu_gie;
 
        always @(posedge i_clk)
                if ((adf_ce_unconditional)||(mem_ce))
                        r_alu_gie  <= op_gie;
        assign  alu_gie = r_alu_gie;
 
        reg     [(AW-1):0]       r_alu_pc;
        always @(posedge i_clk)
                if ((adf_ce_unconditional)
                        ||((master_ce)&&(opvalid_mem)&&(~clear_pipeline)
                                &&(~mem_stalled)))
                        r_alu_pc  <= op_pc;
        assign  alu_pc = r_alu_pc;
`else
        assign  alu_gie = op_gie;
        assign  alu_pc = op_pc;
`endif
 
`ifdef  OPT_ILLEGAL_INSTRUCTION
        reg     r_alu_illegal;
        initial r_alu_illegal = 0;
        always @(posedge i_clk)
                if ((i_rst)||(clear_pipeline))
                        r_alu_illegal <= 1'b0;
                else if (alu_ce)
                        r_alu_illegal <= op_illegal;
                else
                        r_alu_illegal <= 1'b0;
        assign  alu_illegal = (alu_illegal_op)||(r_alu_illegal);
`else
        assign  alu_illegal = 1'b0;
`endif
 
        initial r_alu_pc_valid = 1'b0;
        initial mem_pc_valid = 1'b0;
        always @(posedge i_clk)
                if (i_rst)
                        r_alu_pc_valid <= 1'b0;
                else if (adf_ce_unconditional)//Includes&&(~alu_clear_pipeline)
                        r_alu_pc_valid <= 1'b1;
                else if (((~alu_busy)&&(~div_busy)&&(~fpu_busy))||(clear_pipeline))
                        r_alu_pc_valid <= 1'b0;
        assign  alu_pc_valid = (r_alu_pc_valid)&&((~alu_busy)&&(~div_busy)&&(~fpu_busy));
        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     [1:0]    r_bus_lock;
                initial r_bus_lock = 2'b00;
                always @(posedge i_clk)
                        if (i_rst)
                                r_bus_lock <= 2'b00;
                        else if ((op_ce)&&(op_lock))
                                r_bus_lock <= 2'b11;
                        else if ((|r_bus_lock)&&((~opvalid_mem)||(~op_ce)))
                                r_bus_lock <= r_bus_lock + 2'b11;
                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)||(mem_valid)
                                ||((~clear_pipeline)&&(~alu_illegal)
                                        &&(((alu_wr)&&(alu_valid))
                                                ||(div_valid)||(fpu_valid)));
`else
        assign  wr_reg_ce = (dbgv)||(mem_valid)
                                ||((~clear_pipeline)
                                        &&(((alu_wr)&&(alu_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|div_valid|fpu_valid)?alu_reg:mem_wreg;
        // Are we writing to the CC register?
        assign  wr_write_cc = (wr_reg_id[3:0] == `CPU_CC_REG);
        assign  wr_write_scc = (wr_reg_id[4:0] == {1'b0, `CPU_CC_REG});
        assign  wr_write_ucc = (wr_reg_id[4:0] == {1'b1, `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_gpreg_vl = ((mem_valid) ? mem_result
                                :((div_valid|fpu_valid))
                                        ? ((div_valid) ? div_result:fpu_result)
                                :((dbgv) ? dbg_val : alu_result));
        assign  wr_spreg_vl = ((mem_valid) ? mem_result
                                :((dbgv) ? dbg_val : alu_result));
        always @(posedge i_clk)
                if (wr_reg_ce)
                        regset[wr_reg_id] <= wr_gpreg_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 = { 1'b0, uhalt_phase, ufpu_err_flag,
                        udiv_err_flag, ubus_err_flag, trap, ill_err_u,
                        ubreak, step, 1'b1, sleep,
                        ((wr_flags_ce)&&(alu_gie))?alu_flags:flags };
        assign  w_iflags = { 1'b0, 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_write_ucc))
                        flags <= wr_gpreg_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_write_scc))
                        iflags <= wr_gpreg_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, set the user break bit,
        //              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_write_scc))
                        break_en <= wr_spreg_vl[`CPU_BREAK_BIT];
 
`ifdef  OPT_PIPELINED
        reg     r_break_pending;
 
        initial r_break_pending = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(clear_pipeline)||(~opvalid))
                        r_break_pending <= 1'b0;
                else if (op_break)
                        r_break_pending <= (~alu_busy)&&(~div_busy)&&(~fpu_busy)&&(~mem_busy);
                else
                        r_break_pending <= 1'b0;
        assign  break_pending = r_break_pending;
`else
        assign  break_pending = op_break;
`endif
 
 
        assign  o_break = ((break_en)||(~op_gie))&&(break_pending)
                                &&(~clear_pipeline)
                        ||((~alu_gie)&&(bus_err))
                        ||((~alu_gie)&&(div_error))
                        ||((~alu_gie)&&(fpu_error))
                        ||((~alu_gie)&&(alu_illegal));
 
        // 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_spreg_vl[GIE])
                        //              don't set the sleep bit
                        //      otherwise however it would o.w. be set
                        sleep <= (wr_spreg_vl[`CPU_SLEEP_BIT])
                                &&((~i_interrupt)||(~wr_spreg_vl[`CPU_GIE_BIT]));
                else if ((wr_reg_ce)&&(wr_write_cc)&&(wr_spreg_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_spreg_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_write_ucc))
                        step <= wr_spreg_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)&&(break_pending)&&(~break_en))
`ifdef  OPT_ILLEGAL_INSTRUCTION
                        // On an illegal instruction
                        ||(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_error)
                        // Same thing on a floating point error.  Note that
                        // fpu_error must *never* be set unless fpu_valid is
                        // also set as well, else this will fail.
                        ||(fpu_error)
                        //      
                        ||(bus_err)
                        // If we write to the CC register
                        ||((wr_reg_ce)&&(~wr_spreg_vl[`CPU_GIE_BIT])
                                &&(wr_reg_id[4])&&(wr_write_cc))
                        );
        assign  w_release_from_interrupt = (~gie)&&(~i_interrupt)
                        // Then if we write the sCC register
                        &&(((wr_reg_ce)&&(wr_spreg_vl[`CPU_GIE_BIT])
                                &&(wr_write_scc))
                        );
        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)||(w_release_from_interrupt))
                        trap <= 1'b0;
                else if ((alu_gie)&&(wr_reg_ce)&&(~wr_spreg_vl[`CPU_GIE_BIT])
                                &&(wr_write_ucc)) // &&(wr_reg_id[4]) implied
                        trap <= 1'b1;
                else if ((wr_reg_ce)&&(wr_write_ucc)&&(~alu_gie))
                        trap <= (trap)&&(wr_spreg_vl[`CPU_TRAP_BIT]);
 
        initial ubreak = 1'b0;
        always @(posedge i_clk)
                if ((i_rst)||(w_release_from_interrupt))
                        ubreak <= 1'b0;
                else if ((op_gie)&&(break_pending)&&(w_switch_to_interrupt))
                        ubreak <= 1'b1;
                else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
                        ubreak <= (ubreak)&&(wr_spreg_vl[`CPU_BREAK_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_write_scc))
                        ill_err_i <= (ill_err_i)&&(wr_spreg_vl[`CPU_ILL_BIT]);
                else if ((alu_illegal)&&(~alu_gie))
                        ill_err_i <= 1'b1;
        initial ill_err_u = 1'b0;
        always @(posedge i_clk)
                // The bit is automatically cleared on release from interrupt
                // or reset
                if ((i_rst)||(w_release_from_interrupt))
                        ill_err_u <= 1'b0;
                // If the supervisor (or debugger) writes to this register,
                // clearing the bit, then clear it
                else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
                        ill_err_u <=((ill_err_u)&&(wr_spreg_vl[`CPU_ILL_BIT]));
                else if ((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_write_scc))
                        ibus_err_flag <= (ibus_err_flag)&&(wr_spreg_vl[`CPU_BUSERR_BIT]);
                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)||(w_release_from_interrupt))
                        ubus_err_flag <= 1'b0;
                else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
                        ubus_err_flag <= (ubus_err_flag)&&(wr_spreg_vl[`CPU_BUSERR_BIT]);
                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_write_scc))
                                r_idiv_err_flag <= (r_idiv_err_flag)&&(wr_spreg_vl[`CPU_DIVERR_BIT]);
                        else if ((div_error)&&(~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)||(w_release_from_interrupt))
                                r_udiv_err_flag <= 1'b0;
                        else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
                                        &&(wr_write_ucc))
                                r_udiv_err_flag <= (r_udiv_err_flag)&&(wr_spreg_vl[`CPU_DIVERR_BIT]);
                        else if ((div_error)&&(alu_gie))
                                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_write_scc))
                                r_ifpu_err_flag <= (r_ifpu_err_flag)&&(wr_spreg_vl[`CPU_FPUERR_BIT]);
                        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)&&(w_release_from_interrupt))
                                r_ufpu_err_flag <= 1'b0;
                        else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
                                        &&(wr_write_ucc))
                                r_ufpu_err_flag <= (r_ufpu_err_flag)&&(wr_spreg_vl[`CPU_FPUERR_BIT]);
                        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 (i_rst)
                        r_ihalt_phase <= 1'b0;
                else if ((~alu_gie)&&(alu_pc_valid)&&(~clear_pipeline))
                        r_ihalt_phase <= alu_phase;
        always @(posedge i_clk)
                if ((i_rst)||(w_release_from_interrupt))
                        r_uhalt_phase <= 1'b0;
                else if ((alu_gie)&&(alu_pc_valid))
                        r_uhalt_phase <= alu_phase;
                else if ((~alu_gie)&&(wr_reg_ce)&&(wr_write_ucc))
                        r_uhalt_phase <= wr_spreg_vl[`CPU_PHASE_BIT];
 
        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_spreg_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_spreg_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)||((~gie)&&(w_clear_icache)))
                        pf_pc <= ipc;
                else if ((w_release_from_interrupt)||((gie)&&(w_clear_icache)))
                        pf_pc <= upc;
                else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
                        pf_pc <= wr_spreg_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_gie==gie)&&(
                                ((alu_pc_valid)&&(~clear_pipeline))
                                ||(mem_pc_valid)))
                        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;
 
`ifdef  OPT_PIPELINED
        reg     r_clear_icache;
        initial r_clear_icache = 1'b1;
        always @(posedge i_clk)
                if ((i_rst)||(i_clear_pf_cache))
                        r_clear_icache <= 1'b1;
                else if ((wr_reg_ce)&&(wr_write_scc))
                        r_clear_icache <=  wr_spreg_vl[`CPU_CLRCACHE_BIT];
                else
                        r_clear_icache <= 1'b0;
        assign  w_clear_icache = r_clear_icache;
`else
        assign  w_clear_icache = 1'b0;
`endif
 
        //
        // 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[14:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
                                o_dbg_reg[15] <= 1'b0;
                                o_dbg_reg[31:23] <= w_cpu_info;
                                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[14:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
                                o_dbg_reg[15] <= 1'b0;
                                o_dbg_reg[31:23] <= w_cpu_info;
                                o_dbg_reg[`CPU_GIE_BIT] <= gie;
                        end
                end
        end endgenerate
 
        always @(posedge i_clk)
                o_dbg_cc <= { o_break, bus_err, gie, sleep };
 
`ifdef  OPT_PIPELINED
        always @(posedge i_clk)
                r_halted <= (i_halt)&&(
                        // To be halted, any long lasting instruction must
                        // be completed.
                        (~pf_cyc)&&(~mem_busy)&&(~alu_busy)
                                &&(~div_busy)&&(~fpu_busy)
                        // Operations must either be valid, or illegal
                        &&((opvalid)||(i_rst)||(dcd_illegal))
                        // Decode stage must be either valid, in reset, or ill
                        &&((dcdvalid)||(i_rst)||(pf_illegal)));
`else
        always @(posedge i_clk)
                r_halted <= (i_halt)&&((opvalid)||(i_rst));
`endif
        assign  o_dbg_stall = ~r_halted;
 
        //
        //
        // 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_gpreg_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

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	56	dgisselq	`// (actual implementation aside ...) The instruction set is about as`
10	179	dgisselq	`// RISC as you can get, with only 26 instruction types currently supported.`
11			`// (There are still 8-instruction Op-Codes reserved for floating point,`
12			`// and 5 which can be used for transactions not requiring registers.)`
13	56	dgisselq	`// Please see the accompanying spec.pdf file for a description of these`
14			`// instructions.`
15	2	dgisselq	`//`
16	56	dgisselq	`// All instructions are 32-bits wide. All bus accesses, both address and`
17			`// data, are 32-bits over a wishbone bus.`
18	2	dgisselq	`//`
19			`// The Zip CPU is fully pipelined with the following pipeline stages:`
20			`//`
21	56	dgisselq	`// 1. Prefetch, returns the instruction from memory.`
22	2	dgisselq	`//`
23			`// 2. Instruction Decode`
24			`//`
25			`// 3. Read Operands`
26			`//`
27			`// 4. Apply Instruction`
28			`//`
29			`// 4. Write-back Results`
30			`//`
31	179	dgisselq	`// Further information about the inner workings of this CPU, such as`
32			`// what causes pipeline stalls, may be found in the spec.pdf file. (The`
33			`// documentation within this file had become out of date and out of sync`
34			`// with the spec.pdf, so look to the spec.pdf for accurate and up to date`
35			`// information.)`
36	2	dgisselq	`//`
37			`//`
38	69	dgisselq	`// In general, the pipelining is controlled by three pieces of logic`
39			`// per stage: _ce, _stall, and _valid. _valid means that the stage`
40			`// holds a valid instruction. _ce means that the instruction from the`
41			`// previous stage is to move into this one, and _stall means that the`
42			`// instruction from the previous stage may not move into this one.`
43			`// The difference between these control signals allows individual stages`
44			`// to propagate instructions independently. In general, the logic works`
45			`// as:`
46			`//`
47			`//`
48			`// assign (n)_ce = (n-1)_valid && (~(n)_stall)`
49			`//`
50			`//`
51			`// always @(posedge i_clk)`
52			`// if ((i_rst)\|\|(clear_pipeline))`
53			`// (n)_valid = 0`
54			`// else if (n)_ce`
55			`// (n)_valid = 1`
56			`// else if (n+1)_ce`
57			`// (n)_valid = 0`
58			`//`
59			`// assign (n)_stall = ( (n-1)_valid && ( pipeline hazard detection ) )`
60			`// \|\| ( (n)_valid && (n+1)_stall );`
61			`//`
62			`// and ...`
63			`//`
64			`// always @(posedge i_clk)`
65			`// if (n)_ce`
66			`// (n)_variable = ... whatever logic for this stage`
67			`//`
68			`// Note that a stage can stall even if no instruction is loaded into`
69			`// it.`
70			`//`
71			`//`
72	2	dgisselq	`// Creator: Dan Gisselquist, Ph.D.`
73	69	dgisselq	`// Gisselquist Technology, LLC`
74	2	dgisselq	`//`
75			`///////////////////////////////////////////////////////////////////////////////`
76			`//`
77	160	dgisselq	`// Copyright (C) 2015-2016, Gisselquist Technology, LLC`
78	2	dgisselq	`//`
79			`// This program is free software (firmware): you can redistribute it and/or`
80			`// modify it under the terms of the GNU General Public License as published`
81			`// by the Free Software Foundation, either version 3 of the License, or (at`
82			`// your option) any later version.`
83			`//`
84			`// This program is distributed in the hope that it will be useful, but WITHOUT`
85			`// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or`
86			`// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License`
87			`// for more details.`
88			`//`
89			`// License: GPL, v3, as defined and found on www.gnu.org,`
90			`// http://www.gnu.org/licenses/gpl.html`
91			`//`
92			`//`
93			`///////////////////////////////////////////////////////////////////////////////`
94			`//`
95	36	dgisselq	`// We can either pipeline our fetches, or issue one fetch at a time. Pipelined`
96			`// fetches are more complicated and therefore use more FPGA resources, while`
97			`// single fetches will cause the CPU to stall for about 5 stalls each`
98			`// instruction cycle, effectively reducing the instruction count per clock to`
99			`// about 0.2. However, the area cost may be worth it. Consider:`
100			`//`
101			`// Slice LUTs ZipSystem ZipCPU`
102			`// Single Fetching 2521 1734`
103			`// Pipelined fetching 2796 2046`
104			`//`
105			`//`
106			`//`
107	25	dgisselq	`define CPU_CC_REG 4'he
108	2	dgisselq	`define CPU_PC_REG 4'hf
109	179	dgisselq	`define CPU_CLRCACHE_BIT 14 // Floating point error flag, set on error
110			`define CPU_PHASE_BIT 13 // Floating point error flag, set on error
111	69	dgisselq	`define CPU_FPUERR_BIT 12 // Floating point error flag, set on error
112			`define CPU_DIVERR_BIT 11 // Divide error flag, set on divide by zero
113			`define CPU_BUSERR_BIT 10 // Bus error flag, set on error
114			`define CPU_TRAP_BIT 9 // User TRAP has taken place
115			`define CPU_ILL_BIT 8 // Illegal instruction
116	2	dgisselq	`define CPU_BREAK_BIT 7
117	69	dgisselq	`define CPU_STEP_BIT 6 // Will step one or two (VLIW) instructions
118	2	dgisselq	`define CPU_GIE_BIT 5
119			`define CPU_SLEEP_BIT 4
120	36	dgisselq	`// Compile time defines`
121	56	dgisselq	`//`
122			`include "cpudefs.v"
123			`//`
124	65	dgisselq	`//`
125	2	dgisselq	`module zipcpu(i_clk, i_rst, i_interrupt,`
126			`// Debug interface`
127	18	dgisselq	`i_halt, i_clear_pf_cache, i_dbg_reg, i_dbg_we, i_dbg_data,`
128			`o_dbg_stall, o_dbg_reg, o_dbg_cc,`
129	2	dgisselq	`o_break,`
130			`// CPU interface to the wishbone bus`
131	36	dgisselq	`o_wb_gbl_cyc, o_wb_gbl_stb,`
132			`o_wb_lcl_cyc, o_wb_lcl_stb,`
133			`o_wb_we, o_wb_addr, o_wb_data,`
134	2	dgisselq	`i_wb_ack, i_wb_stall, i_wb_data,`
135	36	dgisselq	`i_wb_err,`
136	2	dgisselq	`// Accounting/CPU usage interface`
137	65	dgisselq	`o_op_stall, o_pf_stall, o_i_count`
138			`ifdef DEBUG_SCOPE
139			`, o_debug`
140			`endif
141			`);`
142	48	dgisselq	`parameter RESET_ADDRESS=32'h0100000, ADDRESS_WIDTH=24,`
143	179	dgisselq	`LGICACHE=8;`
144	56	dgisselq	`ifdef OPT_MULTIPLY
145	132	dgisselq	parameter IMPLEMENT_MPY = `OPT_MULTIPLY;
146	56	dgisselq	`else
147			`parameter IMPLEMENT_MPY = 0;`
148			`endif
149	71	dgisselq	`ifdef OPT_DIVIDE
150			`parameter IMPLEMENT_DIVIDE = 1;`
151			`else
152			`parameter IMPLEMENT_DIVIDE = 0;`
153			`endif
154			`ifdef OPT_IMPLEMENT_FPU
155			`parameter IMPLEMENT_FPU = 1,`
156			`else
157			`parameter IMPLEMENT_FPU = 0,`
158			`endif
159	69	dgisselq	`IMPLEMENT_LOCK=1;`
160			`ifdef OPT_EARLY_BRANCHING
161			`parameter EARLY_BRANCHING = 1;`
162			`else
163			`parameter EARLY_BRANCHING = 0;`
164			`endif
165			`parameter AW=ADDRESS_WIDTH;`
166	2	dgisselq	`input i_clk, i_rst, i_interrupt;`
167			`// Debug interface -- inputs`
168	18	dgisselq	`input i_halt, i_clear_pf_cache;`
169	2	dgisselq	`input [4:0] i_dbg_reg;`
170			`input i_dbg_we;`
171			`input [31:0] i_dbg_data;`
172			`// Debug interface -- outputs`
173	160	dgisselq	`output wire o_dbg_stall;`
174	2	dgisselq	`output reg [31:0] o_dbg_reg;`
175	56	dgisselq	`output reg [3:0] o_dbg_cc;`
176	2	dgisselq	`output wire o_break;`
177			`// Wishbone interface -- outputs`
178	36	dgisselq	`output wire o_wb_gbl_cyc, o_wb_gbl_stb;`
179			`output wire o_wb_lcl_cyc, o_wb_lcl_stb, o_wb_we;`
180	48	dgisselq	`output wire [(AW-1):0] o_wb_addr;`
181			`output wire [31:0] o_wb_data;`
182	2	dgisselq	`// Wishbone interface -- inputs`
183			`input i_wb_ack, i_wb_stall;`
184			`input [31:0] i_wb_data;`
185	36	dgisselq	`input i_wb_err;`
186	2	dgisselq	`// Accounting outputs ... to help us count stalls and usage`
187	9	dgisselq	`output wire o_op_stall;`
188	2	dgisselq	`output wire o_pf_stall;`
189	9	dgisselq	`output wire o_i_count;`
190	56	dgisselq	`//`
191	65	dgisselq	`ifdef DEBUG_SCOPE
192	56	dgisselq	`output reg [31:0] o_debug;`
193	65	dgisselq	`endif
194	2	dgisselq
195	25	dgisselq
196	2	dgisselq	`// Registers`
197	56	dgisselq	`//`
198			`// The distributed RAM style comment is necessary on the`
199			`// SPARTAN6 with XST to prevent XST from oversimplifying the register`
200			`// set and in the process ruining everything else. It basically`
201			`// optimizes logic away, to where it no longer works. The logic`
202			`// as described herein will work, this just makes sure XST implements`
203			`// that logic.`
204			`//`
205			`(* ram_style = "distributed" *)`
206	2	dgisselq	`reg [31:0] regset [0:31];`
207	9	dgisselq
208			`// Condition codes`
209	56	dgisselq	`// (BUS, TRAP,ILL,BREAKEN,STEP,GIE,SLEEP ), V, N, C, Z`
210			`reg [3:0] flags, iflags;`
211	179	dgisselq	`wire [14:0] w_uflags, w_iflags;`
212	160	dgisselq	`reg trap, break_en, step, gie, sleep, r_halted;`
213	179	dgisselq	`wire break_pending;`
214			`wire w_clear_icache;`
215	38	dgisselq	`ifdef OPT_ILLEGAL_INSTRUCTION
216	65	dgisselq	`reg ill_err_u, ill_err_i;`
217	38	dgisselq	`else
218	65	dgisselq	`wire ill_err_u, ill_err_i;`
219	36	dgisselq	`endif
220	179	dgisselq	`reg ubreak;`
221	65	dgisselq	`reg ibus_err_flag, ubus_err_flag;`
222	69	dgisselq	`wire idiv_err_flag, udiv_err_flag;`
223			`wire ifpu_err_flag, ufpu_err_flag;`
224			`wire ihalt_phase, uhalt_phase;`
225	2	dgisselq
226	9	dgisselq	`// The master chip enable`
227			`wire master_ce;`
228	2	dgisselq
229			`//`
230			`//`
231			`// PIPELINE STAGE #1 :: Prefetch`
232			`// Variable declarations`
233			`//`
234	48	dgisselq	`reg [(AW-1):0] pf_pc;`
235	69	dgisselq	`reg new_pc;`
236	18	dgisselq	`wire clear_pipeline;`
237	179	dgisselq	`assign clear_pipeline = new_pc;`
238	9	dgisselq
239			`wire dcd_stalled;`
240	36	dgisselq	`wire pf_cyc, pf_stb, pf_we, pf_busy, pf_ack, pf_stall, pf_err;`
241	48	dgisselq	`wire [(AW-1):0] pf_addr;`
242			`wire [31:0] pf_data;`
243			`wire [31:0] instruction;`
244			`wire [(AW-1):0] instruction_pc;`
245	36	dgisselq	`wire pf_valid, instruction_gie, pf_illegal;`
246	2	dgisselq
247			`//`
248			`//`
249			`// PIPELINE STAGE #2 :: Instruction Decode`
250			`// Variable declarations`
251			`//`
252			`//`
253	83	dgisselq	`reg opvalid, opvalid_mem, opvalid_alu;`
254	69	dgisselq	`reg opvalid_div, opvalid_fpu;`
255			`wire op_stall, dcd_ce, dcd_phase;`
256			`wire [3:0] dcdOp;`
257			`wire [4:0] dcdA, dcdB, dcdR;`
258			`wire dcdA_cc, dcdB_cc, dcdA_pc, dcdB_pc, dcdR_cc, dcdR_pc;`
259			`wire [3:0] dcdF;`
260			`wire dcdR_wr, dcdA_rd, dcdB_rd,`
261			`dcdALU, dcdM, dcdDV, dcdFP,`
262	71	dgisselq	`dcdF_wr, dcd_gie, dcd_break, dcd_lock,`
263	105	dgisselq	`dcd_pipe, dcd_ljmp;`
264	69	dgisselq	`reg r_dcdvalid;`
265			`wire dcdvalid;`
266			`wire [(AW-1):0] dcd_pc;`
267			`wire [31:0] dcdI;`
268			`wire dcd_zI; // true if dcdI == 0`
269	2	dgisselq	`wire dcdA_stall, dcdB_stall, dcdF_stall;`
270
271	69	dgisselq	`wire dcd_illegal;`
272			`wire dcd_early_branch;`
273	48	dgisselq	`wire [(AW-1):0] dcd_branch_pc;`
274	2	dgisselq
275
276			`//`
277			`//`
278			`// PIPELINE STAGE #3 :: Read Operands`
279			`// Variable declarations`
280			`//`
281			`//`
282			`//`
283			`// Now, let's read our operands`
284			`reg [4:0] alu_reg;`
285	179	dgisselq	`wire [3:0] opn;`
286			`wire [4:0] opR;`
287	48	dgisselq	`reg [31:0] r_opA, r_opB;`
288			`reg [(AW-1):0] op_pc;`
289	25	dgisselq	`wire [31:0] w_opA, w_opB;`
290	2	dgisselq	`wire [31:0] opA_nowait, opB_nowait, opA, opB;`
291	179	dgisselq	`reg opR_wr, opF_wr;`
292			`wire op_gie, opR_cc;`
293			`wire [14:0] opFl;`
294	56	dgisselq	`reg [5:0] r_opF;`
295			`wire [7:0] opF;`
296	160	dgisselq	`wire op_ce, op_phase, op_pipe, op_change_data_ce;`
297	56	dgisselq	`// Some pipeline control wires`
298	69	dgisselq	`ifdef OPT_PIPELINED
299	56	dgisselq	`reg opA_alu, opA_mem;`
300			`reg opB_alu, opB_mem;`
301			`endif
302	38	dgisselq	`ifdef OPT_ILLEGAL_INSTRUCTION
303	36	dgisselq	`reg op_illegal;`
304	160	dgisselq	`else
305			`wire op_illegal;`
306			`assign op_illegal = 1'b0;`
307	36	dgisselq	`endif
308	179	dgisselq	`wire op_break;`
309	69	dgisselq	`wire op_lock;`
310	2	dgisselq
311
312			`//`
313			`//`
314			`// PIPELINE STAGE #4 :: ALU / Memory`
315			`// Variable declarations`
316			`//`
317			`//`
318	179	dgisselq	`wire [(AW-1):0] alu_pc;`
319	145	dgisselq	`reg r_alu_pc_valid, mem_pc_valid;`
320			`wire alu_pc_valid;`
321	69	dgisselq	`wire alu_phase;`
322	2	dgisselq	`wire alu_ce, alu_stall;`
323			`wire [31:0] alu_result;`
324			`wire [3:0] alu_flags;`
325	71	dgisselq	`wire alu_valid, alu_busy;`
326	2	dgisselq	`wire set_cond;`
327	179	dgisselq	`reg alu_wr, alF_wr;`
328			`wire alu_gie, alu_illegal_op, alu_illegal;`
329	2	dgisselq
330
331
332			`wire mem_ce, mem_stalled;`
333	38	dgisselq	`ifdef OPT_PIPELINED_BUS_ACCESS
334			`wire mem_pipe_stalled;`
335			`endif
336	36	dgisselq	`wire mem_valid, mem_ack, mem_stall, mem_err, bus_err,`
337			`mem_cyc_gbl, mem_cyc_lcl, mem_stb_gbl, mem_stb_lcl, mem_we;`
338	48	dgisselq	`wire [4:0] mem_wreg;`
339	9	dgisselq
340	48	dgisselq	`wire mem_busy, mem_rdbusy;`
341			`wire [(AW-1):0] mem_addr;`
342			`wire [31:0] mem_data, mem_result;`
343	2	dgisselq
344	69	dgisselq	`wire div_ce, div_error, div_busy, div_valid;`
345			`wire [31:0] div_result;`
346			`wire [3:0] div_flags;`
347	2	dgisselq
348	69	dgisselq	`assign div_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_div)`
349			`&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)`
350			`&&(set_cond);`
351	2	dgisselq
352	69	dgisselq	`wire fpu_ce, fpu_error, fpu_busy, fpu_valid;`
353			`wire [31:0] fpu_result;`
354			`wire [3:0] fpu_flags;`
355
356			`assign fpu_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_fpu)`
357			`&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)`
358			`&&(set_cond);`
359
360	179	dgisselq	`wire adf_ce_unconditional;`
361	69	dgisselq
362	2	dgisselq	`//`
363			`//`
364			`// PIPELINE STAGE #5 :: Write-back`
365			`// Variable declarations`
366			`//`
367	179	dgisselq	`wire wr_reg_ce, wr_flags_ce, wr_write_pc, wr_write_cc,`
368			`wr_write_scc, wr_write_ucc;`
369	2	dgisselq	`wire [4:0] wr_reg_id;`
370	160	dgisselq	`wire [31:0] wr_gpreg_vl, wr_spreg_vl;`
371	2	dgisselq	`wire w_switch_to_interrupt, w_release_from_interrupt;`
372	48	dgisselq	`reg [(AW-1):0] upc, ipc;`
373	2	dgisselq
374
375
376			`//`
377			`// MASTER: clock enable.`
378			`//`
379	38	dgisselq	`assign master_ce = (~i_halt)&&(~o_break)&&(~sleep);`
380	2	dgisselq
381
382			`//`
383			`// PIPELINE STAGE #1 :: Prefetch`
384			`// Calculate stall conditions`
385	65	dgisselq	`//`
386			`// These are calculated externally, within the prefetch module.`
387			`//`
388	2	dgisselq
389			`//`
390			`// PIPELINE STAGE #2 :: Instruction Decode`
391			`// Calculate stall conditions`
392	69	dgisselq	`assign dcd_ce = ((~dcdvalid)\|\|(~dcd_stalled))&&(~clear_pipeline);`
393	145	dgisselq
394	69	dgisselq	`ifdef OPT_PIPELINED
395			`assign dcd_stalled = (dcdvalid)&&(op_stall);`
396			`else
397			`// If not pipelined, there will be no opvalid_ anything, and the`
398			`// op_stall will be false, dcdX_stall will be false, thus we can simply`
399			`// do a ...`
400			`assign dcd_stalled = 1'b0;`
401			`endif
402	2	dgisselq	`//`
403			`// PIPELINE STAGE #3 :: Read Operands`
404			`// Calculate stall conditions`
405	69	dgisselq	`wire op_lock_stall;`
406			`ifdef OPT_PIPELINED
407	179	dgisselq	`reg cc_invalid_for_dcd;`
408			`always @(posedge i_clk)`
409			`cc_invalid_for_dcd <= (wr_flags_ce)`
410			\|\|(wr_reg_ce)&&(wr_reg_id[3:0] == `CPU_CC_REG)
411			\|\|(opvalid)&&((opF_wr)\|\|((opR_wr)&&(opR[3:0] == `CPU_CC_REG)))
412			\|\|((alF_wr)\|\|((alu_wr)&&(alu_reg[3:0] == `CPU_CC_REG)))
413			`\|\|(mem_busy)\|\|(div_busy)\|\|(fpu_busy);`
414
415	69	dgisselq	`assign op_stall = (opvalid)&&( // Only stall if we're loaded w/validins`
416			`// Stall if we're stopped, and not allowed to execute`
417			`// an instruction`
418			`// (~master_ce) // Already captured in alu_stall`
419			`//`
420	56	dgisselq	`// Stall if going into the ALU and the ALU is stalled`
421			`// i.e. if the memory is busy, or we are single`
422	69	dgisselq	`// stepping. This also includes our stalls for`
423			`// op_break and op_lock, so we don't need to`
424			`// include those as well here.`
425	83	dgisselq	`// This also includes whether or not the divide or`
426			`// floating point units are busy.`
427			`(alu_stall)`
428	56	dgisselq	`//`
429			`// Stall if we are going into memory with an operation`
430			`// that cannot be pipelined, and the memory is`
431			`// already busy`
432	83	dgisselq	`\|\|(mem_stalled) // &&(opvalid_mem) part of mem_stalled`
433	179	dgisselq	`\|\|(opR_cc)`
434	69	dgisselq	`)`
435			`\|\|(dcdvalid)&&(`
436	71	dgisselq	`// Stall if we need to wait for an operand A`
437	69	dgisselq	`// to be ready to read`
438	71	dgisselq	`(dcdA_stall)`
439	69	dgisselq	`// Likewise for B, also includes logic`
440			`// regarding immediate offset (register must`
441			`// be in register file if we need to add to`
442			`// an immediate)`
443			`\|\|(dcdB_stall)`
444			`// Or if we need to wait on flags to work on the`
445			`// CC register`
446			`\|\|(dcdF_stall)`
447			`);`
448	71	dgisselq	`assign op_ce = ((dcdvalid)\|\|(dcd_illegal))&&(~op_stall)&&(~clear_pipeline);`
449	179	dgisselq
450
451	160	dgisselq	`// BUT ... op_ce is too complex for many of the data operations. So`
452			`// let's make their circuit enable code simpler. In particular, if`
453			`// op_ doesn't need to be preserved, we can change it all we want`
454			`// ... right? The clear_pipeline code, for example, really only needs`
455			`// to determine whether opvalid is true.`
456			`assign op_change_data_ce = (~op_stall);`
457	65	dgisselq	`else
458	69	dgisselq	`assign op_stall = (opvalid)&&(~master_ce);`
459	145	dgisselq	`assign op_ce = ((dcdvalid)\|\|(dcd_illegal))&&(~clear_pipeline);`
460	160	dgisselq	`assign op_change_data_ce = 1'b1;`
461	65	dgisselq	`endif
462	2	dgisselq
463			`//`
464			`// PIPELINE STAGE #4 :: ALU / Memory`
465			`// Calculate stall conditions`
466	36	dgisselq	`//`
467			`// 1. Basic stall is if the previous stage is valid and the next is`
468			`// busy.`
469			`// 2. Also stall if the prior stage is valid and the master clock enable`
470			`// is de-selected`
471	56	dgisselq	`// 3. Stall if someone on the other end is writing the CC register,`
472			`// since we don't know if it'll put us to sleep or not.`
473	36	dgisselq	`// 4. Last case: Stall if we would otherwise move a break instruction`
474			`// through the ALU. Break instructions are not allowed through`
475			`// the ALU.`
476	69	dgisselq	`ifdef OPT_PIPELINED
477	71	dgisselq	`assign alu_stall = (((~master_ce)\|\|(mem_rdbusy)\|\|(alu_busy))&&(opvalid_alu)) //Case 1&2`
478	69	dgisselq	`\|\|((opvalid)&&(op_lock)&&(op_lock_stall))`
479			`\|\|((opvalid)&&(op_break))`
480	179	dgisselq	`\|\|(wr_reg_ce)&&(wr_write_cc)`
481	69	dgisselq	`\|\|(div_busy)\|\|(fpu_busy);`
482	160	dgisselq	`assign alu_ce = (master_ce)&&(opvalid_alu)&&(~alu_stall)`
483	69	dgisselq	`&&(~clear_pipeline);`
484			`else
485	179	dgisselq	`assign alu_stall = (opvalid_alu)&&((~master_ce)\|\|(op_break));`
486	145	dgisselq	`assign alu_ce = (master_ce)&&((opvalid_alu)\|\|(op_illegal))&&(~alu_stall)&&(~clear_pipeline);`
487	69	dgisselq	`endif
488	2	dgisselq	`//`
489	65	dgisselq
490			`//`
491			`// Note: if you change the conditions for mem_ce, you must also change`
492			`// alu_pc_valid.`
493			`//`
494	69	dgisselq	`ifdef OPT_PIPELINED
495			`assign mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)`
496	71	dgisselq	`&&(~clear_pipeline);`
497	69	dgisselq	`else
498			`// If we aren't pipelined, then no one will be changing what's in the`
499			`// pipeline (i.e. clear_pipeline), while our only instruction goes`
500			`// through the ... pipeline.`

Browse

Tools

Subversion Repositories zipcpu

[/] [zipcpu/] [trunk/] [rtl/] [core/] [zipcpu.v] - Blame information for rev 187