Subversion Repositories zipcpu

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//	had become out of date and out of sync with the spec.pdf, so look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