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//	had become out of date and out of sync with the spec.pdf, so look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.)
//
//
// Creator:	Dan Gisselquist, Ph.D.
//		Gisselquist Tecnology, 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_BUSERR_BIT	10
`define	CPU_TRAP_BIT	9
`define	CPU_ILL_BIT	8
`define	CPU_BREAK_BIT	7
`define	CPU_STEP_BIT	6
`define	CPU_GIE_BIT	5
`define	CPU_SLEEP_BIT	4
// Compile time defines
//
`include "cpudefs.v"
//
//
//
// `define	DEBUG_SCOPE
//
//
//
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, AW=ADDRESS_WIDTH;
`ifdef	OPT_MULTIPLY
	parameter	IMPLEMENT_MPY = 1;
`else
	parameter	IMPLEMENT_MPY = 0;
`endif
	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	[10:0]	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;
 
	// The master chip enable
	wire		master_ce;
 
	//
	//
	//	PIPELINE STAGE #1 :: Prefetch
	//		Variable declarations
	//
	reg	[(AW-1):0]	pf_pc;
	reg		new_pc, op_break;
	wire	clear_pipeline;
	assign	clear_pipeline = new_pc || i_clear_pf_cache; //  || op_break;
 
	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;
	wire		op_stall, dcd_ce;
	reg	[3:0]	dcdOp;
	reg	[4:0]	dcdA, dcdB;
	reg		dcdA_cc, dcdB_cc, dcdA_pc, dcdB_pc;
	reg	[3:0]	dcdF;
	reg		dcdA_rd, dcdA_wr, dcdB_rd, dcdvalid,
				dcdM, dcdF_wr, dcd_gie, dcd_break;
	reg	[(AW-1):0]	dcd_pc;
	reg	[23:0]	r_dcdI;
`ifdef	OPT_SINGLE_CYCLE
	reg		dcd_zI;	// true if dcdI == 0
`endif
	wire	dcdA_stall, dcdB_stall, dcdF_stall;
 
`ifdef	OPT_PRECLEAR_BUS
	reg	dcd_clear_bus;
`endif
`ifdef	OPT_ILLEGAL_INSTRUCTION
	reg	dcd_illegal;
`endif
`ifdef	OPT_EARLY_BRANCHING
	reg			dcd_early_branch_stb, dcd_early_branch;
	reg	[(AW-1):0]	dcd_branch_pc;
`else
	wire			dcd_early_branch_stb, dcd_early_branch;
	wire	[(AW-1):0]	dcd_branch_pc;
`endif
 
 
	//
	//
	//	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	[10:0]	opFl;
	reg	[5:0]	r_opF;
	wire	[7:0]	opF;
	reg	[2:0]	opF_cp;
	wire		op_ce;
	// Some pipeline control wires
`ifdef	OPT_SINGLE_CYCLE
	reg	opA_alu, opA_mem;
	reg	opB_alu, opB_mem;
`endif
`ifdef	OPT_PRECLEAR_BUS
	reg	op_clear_bus;
`endif
`ifdef	OPT_ILLEGAL_INSTRUCTION
	reg	op_illegal;
`endif
 
 
	//
	//
	//	PIPELINE STAGE #4 :: ALU / Memory
	//		Variable declarations
	//
	//
	reg	[(AW-1):0]	alu_pc;
	reg		alu_pc_valid;;
	wire		alu_ce, alu_stall;
	wire	[31:0]	alu_result;
	wire	[3:0]	alu_flags;
	wire		alu_valid;
	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;
	reg	[4:0]		mem_last_reg; // Last register result to go in
 
 
 
	//
	//
	//	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
	assign		dcd_ce = (pf_valid)&&(~dcd_stalled)&&(~clear_pipeline);
	assign		dcd_stalled = (dcdvalid)&&(
					(op_stall)
					||((dcdA_stall)||(dcdB_stall)||(dcdF_stall))
					||((opvalid_mem)&&(op_wr_pc))
					||((opvalid_mem)&&(opR_cc)));
	//
	//	PIPELINE STAGE #3 :: Read Operands
	//		Calculate stall conditions
	assign	op_stall = ((opvalid)&&(~master_ce))||(
			// Stall if going into the ALU and the ALU is stalled
			//	i.e. if the memory is busy, or we are single
			//	stepping
			((opvalid_alu)&&(alu_stall))
			//
			// ||((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
`ifdef	OPT_PIPELINED_BUS_ACCESS
			||((opvalid_mem)&&(~op_pipe)&&(mem_busy))
			//
			// Stall if we are going into memory with a pipeable
			//	operation, but the memory unit declares it is
			//	not going to accept any more pipeline operations
			||((opvalid_mem)&&( op_pipe)&&(mem_pipe_stalled))
`else
			||((opvalid_mem)&&(mem_busy))
`endif
			);
	assign	op_ce = (dcdvalid)&&((~opvalid)||(~op_stall));
 
	//
	//	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.
	assign	alu_stall = (((~master_ce)||(mem_rdbusy))&&(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_alu)&&(op_break)); // Case 3
	assign	alu_ce = (master_ce)&&(~mem_rdbusy)&&(opvalid_alu)&&(~alu_stall)&&(~clear_pipeline);
	//
 
	//
	// Note: if you change the conditions for mem_ce, you must also change
	// alu_pc_valid.
	//
	assign	mem_ce = (master_ce)&&(opvalid_mem)&&(~clear_pipeline)
			&&(set_cond)&&(~mem_stalled);
`ifdef	OPT_PIPELINED_BUS_ACCESS
	assign	mem_stalled = (~master_ce)||((opvalid_mem)&&(
				(mem_pipe_stalled)
				||((~op_pipe)&&(mem_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
	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)))));
`endif
 
 
	//
	//
	//	PIPELINE STAGE #1 :: Prefetch
	//
	//
`ifdef	OPT_SINGLE_FETCH
	wire		pf_ce;
 
	assign		pf_ce = (~dcd_stalled);
	prefetch	#(ADDRESS_WIDTH)
			pf(i_clk, i_rst, (pf_ce), 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);
`else // Pipe fetch
	pipefetch	#(RESET_ADDRESS, LGICACHE, ADDRESS_WIDTH)
			pf(i_clk, i_rst, (new_pc)|(dcd_early_branch_stb),
					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);
	assign	instruction_gie = gie;
`endif
 
	initial	dcdvalid = 1'b0;
	always @(posedge i_clk)
		if (i_rst)
			dcdvalid <= 1'b0;
		else if (dcd_ce)
			dcdvalid <= (~clear_pipeline)&&(~dcd_early_branch_stb);
		else if ((~dcd_stalled)||(clear_pipeline)||(dcd_early_branch))
			dcdvalid <= 1'b0;
 
`ifdef	OPT_EARLY_BRANCHING
	always @(posedge i_clk)
		if ((dcd_ce)&&(instruction[27:24]==`CPU_PC_REG)&&(master_ce))
		begin
			dcd_early_branch <= 1'b0;
			// First case, a move to PC instruction
			if ((instruction[31:28] == 4'h2)
				// Offsets of the PC register *only*
				&&(instruction[19:16] == `CPU_PC_REG)
				&&((instruction_gie)
					||((~instruction[20])&&(~instruction[15])))
				&&(instruction[23:21]==3'h0)) // Unconditional
			begin
				dcd_early_branch_stb <= 1'b1;
				dcd_early_branch <= 1'b1;
				// r_dcdI <= { {(17){instruction[14]}}, instruction[14:0] };
 
			end else // Next case, an Add Imm -> PC instruction
			if ((instruction[31:28] == 4'ha) // Add
				&&(~instruction[20]) // Immediate
				&&(instruction[23:21]==3'h0)) // Always
			begin
				dcd_early_branch_stb <= 1'b1;
				dcd_early_branch <= 1'b1;
				// r_dcdI <= { {(4){instruction[19]}}, instruction[19:0] };
			end else // Next case: load Immediate to PC
			if (instruction[31:28] == 4'h3)
			begin
				dcd_early_branch_stb <= 1'b1;
				dcd_early_branch <= 1'b1;
				// r_dcdI <= { instruction[23:0] };
			end
		end else
		begin
			if (dcd_ce) dcd_early_branch <= 1'b0;
			dcd_early_branch_stb <= 1'b0;
		end
	generate
	if (AW == 24)
	begin
		always @(posedge i_clk)
		if (dcd_ce)
		begin
			if (instruction[31]) // Add
			begin
				dcd_branch_pc <= instruction_pc
						+ { {(AW-20){instruction[19]}}, instruction[19:0] }
						+ {{(AW-1){1'b0}},1'b1};
			end else if (~instruction[28]) // 4'h2 = MOV
				dcd_branch_pc <= instruction_pc+{ {(AW-15){instruction[14]}}, instruction[14:0] } + {{(AW-1){1'b0}},1'b1};
			else // if (instruction[28]) // 4'h3 = LDI
				dcd_branch_pc <= instruction_pc+{ instruction[23:0] } + {{(AW-1){1'b0}},1'b1};
		end
	end else begin
		always @(posedge i_clk)
		if (dcd_ce)
		begin
			if (instruction[31]) // Add
			begin
				dcd_branch_pc <= instruction_pc
							+ { {(AW-20){instruction[19]}}, instruction[19:0] }
							+ {{(AW-1){1'b0}},1'b1};
			end else if (~instruction[28]) // 4'h2 = MOV
			begin
					dcd_branch_pc <= instruction_pc+{ {(AW-15){instruction[14]}}, instruction[14:0] } + {{(AW-1){1'b0}},1'b1};
			end else // if (instruction[28]) // 4'h3 = LDI
			begin
				dcd_branch_pc <= instruction_pc+{ {(AW-24){instruction[23]}}, instruction[23:0] } + {{(AW-1){1'b0}},1'b1};
			end
		end
	end endgenerate
`else	//	OPT_EARLY_BRANCHING
	assign	dcd_early_branch_stb = 1'b0;
	assign	dcd_early_branch     = 1'b0;
	assign	dcd_branch_pc        = {(AW){1'b0}};
`endif	//	OPT_EARLY_BRANCHING
 
	always @(posedge i_clk)
		if (dcd_ce)
		begin
			dcd_pc <= instruction_pc
				+{{(AW-1){1'b0}},1'b1}; // i.e. dcd_pc+1
 
			// Record what operation we are doing
			dcdOp <= instruction[31:28];
 
			// Default values
			dcdA[4:0] <= { instruction_gie, instruction[27:24] };
			dcdB[4:0] <= { instruction_gie, instruction[19:16] };
			dcdA_cc <=  (instruction[27:24] == `CPU_CC_REG);
			dcdB_cc <=  (instruction[19:16] == `CPU_CC_REG);
			dcdA_pc <=  (instruction[27:24] == `CPU_PC_REG);
			dcdB_pc <=  (instruction[19:16] == `CPU_PC_REG);
			dcdM    <= 1'b0;
`ifdef	OPT_CONDITIONAL_FLAGS
			// Don't change the flags on conditional instructions,
			// UNLESS: the conditional instruction was a CMP
			// or TST instruction.
			dcdF_wr <= ((instruction[23:21]==3'h0)
					||(instruction[31:29] == 3'h0));
`else
			dcdF_wr <= 1'b1;
`endif
`ifdef	OPT_PRECLEAR_BUS
			dcd_clear_bus <= 1'b0;
`endif
`ifdef	OPT_ILLEGAL_INSTRUCTION
			dcd_illegal <= pf_illegal;
`endif
 
			// Set the condition under which we do this operation
			// The top four bits are a mask, the bottom four the
			// value the flags must equal once anded with the mask
			dcdF <= { (instruction[23:21]==3'h0), instruction[23:21] };
			casez(instruction[31:28])
			4'h2: begin // Move instruction
				if (~instruction_gie)
				begin
					dcdA[4] <= instruction[20];
					dcdB[4] <= instruction[15];
				end
				dcdA_wr <= 1'b1;
				dcdA_rd <= 1'b0;
				dcdB_rd <= 1'b1;
				r_dcdI <= { {(9){instruction[14]}}, instruction[14:0] };
`ifdef	OPT_SINGLE_CYCLE
				dcd_zI <= (instruction[14:0] == 0);
`endif
				dcdF_wr <= 1'b0; // Don't write flags
				end
			4'h3: begin // Load immediate
				dcdA_wr <= 1'b1;
				dcdA_rd <= 1'b0;
				dcdB_rd <= 1'b0;
				r_dcdI <= { instruction[23:0] };
`ifdef	OPT_SINGLE_CYCLE
				dcd_zI <= (instruction[23:0] == 0);
`endif
				dcdF_wr <= 1'b0; // Don't write flags
				dcdF    <= 4'h8; // This is unconditional
				dcdOp <= 4'h2;
				end
			4'h4: begin // Multiply, LDI[HI|LO], or NOOP/BREAK
`ifdef	OPT_CONDITIONAL_FLAGS
				// Don't write flags except for multiplies
				//   and then only if they are unconditional
				dcdF_wr <= ((instruction[27:25] != 3'h7)
					&&(instruction[23:21]==3'h0));
`else
				// Don't write flags except for multiplies
				dcdF_wr <= (instruction[27:25] != 3'h7);
`endif
				r_dcdI <= { 8'h00, instruction[15:0] };
`ifdef	OPT_SINGLE_CYCLE
				dcd_zI <= (instruction[15:0] == 0);
`endif
				if (instruction[27:24] == 4'he)
				begin
					// NOOP instruction
					dcdA_wr <= 1'b0;
					dcdA_rd <= 1'b0;
					dcdB_rd <= 1'b0;
					dcdOp <= 4'h2;
					// Might also be a break.  Big
					// instruction set hole here.
`ifdef	OPT_ILLEGAL_INSTRUCTION
					dcd_illegal <= (pf_illegal)||(instruction[23:1] != 0);
`endif
				end else if (instruction[27:24] == 4'hf)
				begin // Load partial immediate(s)
					dcdA_wr <= 1'b1;
					dcdA_rd <= 1'b1;
					dcdB_rd <= 1'b0;
					dcdA[4:0] <= { instruction_gie, instruction[19:16] };
					dcdA_cc <= (instruction[19:16] == `CPU_CC_REG);
					dcdA_pc <= (instruction[19:16] == `CPU_PC_REG);
					dcdOp <= { 3'h3, instruction[20] };
				end else begin
					// Actual multiply instruction
					r_dcdI <= { 8'h00, instruction[15:0] };
`ifdef	OPT_SINGLE_CYCLE
					dcd_zI <= (instruction[15:0] == 0);
`endif
					dcdA_rd <= 1'b1;
					dcdB_rd <= (instruction[19:16] != 4'hf);
					dcdOp[3:0] <= (instruction[20])? 4'h4:4'h3;
				end end
			4'b011?: begin // LOD/STO or Load/Store
				dcdF_wr <= 1'b0; // Don't write flags
				dcdA_wr <= (~instruction[28]); // Write on loads
				dcdA_rd <= (instruction[28]); // Read on stores
				dcdB_rd <= instruction[20];
				if (instruction[20])
				begin
					r_dcdI <= { {(8){instruction[15]}}, instruction[15:0] };
`ifdef	OPT_SINGLE_CYCLE
					dcd_zI <= (instruction[15:0] == 0);
`endif
				end else begin
					r_dcdI <= { {(4){instruction[19]}}, instruction[19:0] };
`ifdef	OPT_SINGLE_CYCLE
					dcd_zI <= (instruction[19:0] == 0);
`endif
				end
				dcdM <= 1'b1; // Memory operation
`ifdef	OPT_PRECLEAR_BUS
				dcd_clear_bus <= (instruction[23:21]==3'h0);
`endif
				end
			default: begin
				dcdA_wr <= (instruction[31])||(instruction[31:28]==4'h5);
				dcdA_rd <= 1'b1;
				dcdB_rd <= instruction[20];
				if (instruction[20])
				begin
					r_dcdI <= { {(8){instruction[15]}}, instruction[15:0] };
`ifdef	OPT_SINGLE_CYCLE
					dcd_zI <= (instruction[15:0] == 0);
`endif
				end else begin
					r_dcdI <= { {(4){instruction[19]}}, instruction[19:0] };
`ifdef	OPT_SINGLE_CYCLE
					dcd_zI <= (instruction[19:0] == 0);
`endif
				end end
			endcase
 
 
			dcd_gie <= instruction_gie;
		end
	always @(posedge i_clk)
		if (dcd_ce)
			dcd_break <= (instruction[31:0] == 32'h4e000001);
		else if ((clear_pipeline)||(~dcdvalid)) // SHOULDNT THIS BE ||op_ce?
			dcd_break <= 1'b0;
 
`ifdef	OPT_PIPELINED_BUS_ACCESS
	reg	[23:0]	r_opI;
	reg	[4:0]	op_B;
	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
	always @(posedge i_clk)
		if (op_ce)
			op_pipe <= (dcdvalid)&&(opvalid_mem)&&(dcdM) // Both mem
				&&(dcdOp[0]==opn[0]) // Both Rd, or both Wr
				&&(dcdB == op_B) // Same address register
				&&((dcdF[2:0] == opF_cp) // Same condition
					||(opF_cp == 3'h0)) // or no prev condition
				&&((r_dcdI == r_opI)||(r_dcdI==r_opI+24'h1));
	always @(posedge i_clk)
		if (op_ce) // &&(dcdvalid))
			r_opI <= r_dcdI;
	always @(posedge i_clk)
		if (op_ce) // &&(dcdvalid))
			op_B <= dcdB;
`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
	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[31:11], (dcd_gie)?w_uflags:w_iflags };
			else
				r_opA <= w_opA;
`ifdef	OPT_SINGLE_CYCLE
		end else if (opvalid)
		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;
`endif
		end
 
	wire	[31:0]	dcdI, w_opBnI, w_pcB_v;
	assign	dcdI = { {(8){r_dcdI[23]}}, r_dcdI };
	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[31:11], (dcd_gie)?w_uflags:w_iflags}
		: w_opB)));
 
	always @(posedge i_clk)
		if (op_ce) // &&(dcdvalid))
			r_opB <= w_opBnI + dcdI;
`ifdef	OPT_SINGLE_CYCLE
		else if ((opvalid)&&(
				((opB_alu)&&(alu_wr))
				||((opB_mem)&&(mem_valid))))
			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
			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
			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];
 
	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<= (~clear_pipeline)&&(dcdvalid)&&(~dcd_stalled);
`ifdef	OPT_ILLEGAL_INSTRUCTION
			opvalid_mem <= (dcdM)&&(~dcd_illegal)&&(~clear_pipeline)&&(dcdvalid)&&(~dcd_stalled);
			opvalid_alu <= ((~dcdM)||(dcd_illegal))&&(~clear_pipeline)&&(dcdvalid)&&(~dcd_stalled);
`else
			opvalid_alu <= (~dcdM)&&(~clear_pipeline)&&(dcdvalid)&&(~dcd_stalled);
			opvalid_mem <= (dcdM)&&(~clear_pipeline)&&(dcdvalid)&&(~dcd_stalled);
`endif
		end else if ((~op_stall)||(clear_pipeline))
		begin
			opvalid     <= 1'b0;
			opvalid_alu <= 1'b0;
			opvalid_mem <= 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_ILLEGAL_INSTRUCTION
	always @(posedge i_clk)
		if(op_ce)
			op_illegal <= dcd_illegal;
`endif
 
	always @(posedge i_clk)
		if (op_ce)
		begin
			opn    <= dcdOp;	// Which ALU operation?
			// opM  <= dcdM;	// Is this a memory operation?
`ifdef	OPT_EARLY_BRANCHING
			opF_wr <= (dcdF_wr)&&((~dcdA_cc)||(~dcdA_wr))&&(~dcd_early_branch);
			opR_wr <= (dcdA_wr)&&(~dcd_early_branch);
`else
			// Will we write the flags/CC Register with our result?
			opF_wr <= (dcdF_wr)&&((~dcdA_cc)||(~dcdA_wr));
			// Will we be writing our results into a register?
			opR_wr <= dcdA_wr;
`endif
			// What register will these results be written into?
			opR    <= dcdA;
			opR_cc <= (dcdA_wr)&&(dcdA_cc)&&(dcdA[4]==dcd_gie);
			// User level (1), vs supervisor (0)/interrupts disabled
			op_gie <= dcd_gie;
 
			//
`ifdef	OPT_EARLY_BRANCHING
			op_wr_pc <= ((dcdA_wr)&&(dcdA_pc)&&(dcdA[4] == dcd_gie))&&(~dcd_early_branch);
`else
			op_wr_pc <= ((dcdA_wr)&&(dcdA_pc)&&(dcdA[4] == dcd_gie));
`endif
			op_pc  <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
			// op_pc  <= dcd_pc;
 
`ifdef	OPT_PRECLEAR_BUS
			op_clear_bus <= dcd_clear_bus;
`endif
		end
	assign	opFl = (op_gie)?(w_uflags):(w_iflags);
 
	// 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_SINGLE_CYCLE
	initial	opA_alu = 1'b0;
	always @(posedge i_clk)
		if (op_ce)
			opA_alu <= (opvalid_alu)&&(opR == dcdA)&&(opR_wr)&&(dcdA_rd);
		else if ((opvalid)&&(opA_alu)&&(alu_valid))
			opA_alu <= 1'b0;
	initial	opA_mem = 1'b0;
	always @(posedge i_clk)
		if (op_ce)
			opA_mem <= ((opvalid_mem)&&(opR == dcdA)&&(dcdA_rd)&&(~opn[0]))
				||((~opvalid)&&(mem_busy)&&(~mem_we)
					&&(mem_last_reg == dcdA)&&(dcdA_rd));
		else if ((opvalid)&&(opA_mem)&&(mem_valid))
			opA_mem <= 1'b0;
`endif
 
	always @(posedge i_clk)
		if (mem_ce)
			mem_last_reg <= opR;
`ifdef	OPT_SINGLE_CYCLE
	assign	opA = ((opA_alu)&&(alu_wr)) ? alu_result
			: ( ((opA_mem)&&(mem_valid))?mem_result
			: r_opA );
`else
	assign	opA = r_opA;
`endif
 
	assign	dcdA_stall = (dcdvalid)&&(dcdA_rd)&&(
`ifdef	OPT_SINGLE_CYCLE
		// Skip the requirement on writing back opA
		// Stall on memory, since we'll always need to stall for a 
		// memory access anyway
				((opvalid_alu)&&(opF_wr)&&(dcdA_cc)));
`else
				((opvalid)&&(opR_wr)&&(opR == dcdA))
				||((opvalid_alu)&&(opF_wr)&&(dcdA_cc))
				||((mem_rdbusy)&&(mem_last_reg == dcdA))
				);
`endif
 
`ifdef	OPT_SINGLE_CYCLE
	always @(posedge i_clk)
		if (op_ce)
			opB_alu <= (opvalid_alu)&&(opR == dcdB)&&(opR_wr)&&(dcdB_rd)&&(dcd_zI);
	always @(posedge i_clk)
		if (op_ce)
			opB_mem <= (dcd_zI)&&(dcdB_rd)&&(
				((opvalid_mem)&&(opR == dcdB)&&(~opn[0]))
				||((~opvalid)&&(mem_busy)&&(~mem_we)
					&&(mem_last_reg == dcdB)));
		else if ((opvalid)&&(opB_mem)&&(mem_valid))
			opB_mem <= 1'b0;
	assign	opB = ((opB_alu)&&(alu_wr)) ? alu_result
			: ( ((opB_mem)&&(mem_valid))?mem_result
			: r_opB );
`else
	assign	opB = r_opB;
`endif
 
	assign	dcdB_stall = (dcdvalid)&&(dcdB_rd)&&(
`ifdef	OPT_SINGLE_CYCLE
				// 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.
				((opvalid)&&(opR_wr)&&(opR == dcdB)
					&&(opR != { op_gie, `CPU_PC_REG} )
					&&(~dcd_zI))
				// 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
				||((mem_busy)&&(~mem_we)&&(mem_last_reg==dcdB)));
`else
				((opvalid)&&(opR_wr)&&(opR == dcdB))
				||((opvalid_alu)&&(opF_wr)&&(dcdB_cc))
				||((mem_rdbusy)&&(mem_last_reg == dcdB))
				);
`endif
	assign	dcdF_stall = (dcdvalid)&&((~dcdF[3])||(dcdA_cc)||(dcdB_cc))
					&&(opvalid)&&(opR_cc);
	//
	//
	//	PIPELINE STAGE #4 :: Apply Instruction
	//
	//
	cpuops	#(IMPLEMENT_MPY) doalu(i_clk, i_rst, alu_ce,
			(opvalid_alu), opn, opA, opB,
			alu_result, alu_flags, alu_valid, alu_illegal_op);
 
	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 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
	always @(posedge i_clk)
		if (alu_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 ((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)));
 
`ifdef	OPT_PIPELINED_BUS_ACCESS
	pipemem	#(AW) domem(i_clk, i_rst, mem_ce,
				(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) domem(i_clk, i_rst, mem_ce,
				(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)&&(~clear_pipeline))||(mem_valid);
`else
	assign	wr_reg_ce = ((alu_wr)&&(~clear_pipeline))||(mem_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
	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 = (alu_wr)?((dbgv)?dbg_val: alu_result) :mem_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)&&(~clear_pipeline)&&(~alu_illegal);
`ifdef	OPT_ILLEGAL_INSTRUCTION
	assign	w_uflags = { ubus_err_flag, trap, ill_err_u,    1'b0, step, 1'b1, sleep, ((wr_flags_ce)&&(alu_gie))?alu_flags:flags };
	assign	w_iflags = { ibus_err_flag, trap, ill_err_i,break_en, 1'b0, 1'b0, sleep, ((wr_flags_ce)&&(~alu_gie))?alu_flags:iflags };
`else
	assign	w_uflags = { ubus_err_flag, trap, ill_err_u,     1'b0, step, 1'b1, sleep, ((wr_flags_ce)&&(alu_gie))?alu_flags:flags };
	assign	w_iflags = { ibus_err_flag, trap, ill_err_i, break_en, 1'b0, 1'b0, sleep, ((wr_flags_ce)&&(~alu_gie))?alu_flags:iflags };
`endif
	// 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 <= 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 <= 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)
				&&(~clear_pipeline)
			||((~alu_gie)&&(bus_err))
			||((~alu_gie)&&(alu_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)||((i_interrupt)&&(gie)))
			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.
			sleep <= wr_reg_vl[`CPU_SLEEP_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)
			// If we are stepping the CPU
			||((alu_pc_valid)&&(step))
			// If we encounter a break instruction, if the break
			//	enable isn't set.
			||((master_ce)&&(~mem_rdbusy)&&(op_break)&&(~break_en))
`ifdef	OPT_ILLEGAL_INSTRUCTION
			// On an illegal instruction
			||((alu_valid)&&(alu_illegal))
`endif
			// 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 ((gie)&&(wr_reg_ce)&&(~wr_reg_vl[`CPU_GIE_BIT])
				&&(wr_reg_id[4])&&(wr_write_cc))
			trap <= 1'b1;
		// else if ((i_halt)&&(i_dbg_we)&&(i_dbg_reg[3:0] == `CPU_CC_REG)
				// &&(~i_dbg_data[`CPU_GIE_BIT]))
			// trap <= i_dbg_data[`CPU_TRAP_BIT];
		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_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_valid)&&(alu_illegal)&&(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;
 
	//
	// 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];
		else if (dcd_ce)
			pf_pc <= pf_pc + {{(AW-1){1'b0}},1'b1};
 
	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[10:0] <= (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[10:0] <= (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)&&((~opvalid)||(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[7:0],
			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_vl[5:0]
			};
`endif
 
endmodule