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

Subversion Repositories zipcpu

Compare Revisions

  • This comparison shows the changes necessary to convert path
    /zipcpu/trunk/rtl/core
    from Rev 196 to Rev 201
    Reverse comparison

Rev 196 → Rev 201

/idecode_deprecated.v File deleted
/cpuops_deprecated.v File deleted
/cpuops.v
1,4 → 1,4
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: cpuops.v
//
12,9 → 12,9
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
26,11 → 26,16
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
`include "cpudefs.v"
//
45,12 → 50,6
output reg o_valid;
output wire o_busy;
 
// Rotate-left pre-logic
wire [63:0] w_rol_tmp;
assign w_rol_tmp = { i_a, i_a } << i_b[4:0];
wire [31:0] w_rol_result;
assign w_rol_result = w_rol_tmp[63:32]; // Won't set flags
 
// Shift register pre-logic
wire [32:0] w_lsr_result, w_asr_result, w_lsl_result;
wire signed [32:0] w_pre_asr_input, w_pre_asr_shifted;
75,28 → 74,20
assign w_brev_result[k] = i_b[31-k];
end endgenerate
 
// Popcount pre-logic
wire [31:0] w_popc_result;
assign w_popc_result[5:0]=
({5'h0,i_b[ 0]}+{5'h0,i_b[ 1]}+{5'h0,i_b[ 2]}+{5'h0,i_b[ 3]})
+({5'h0,i_b[ 4]}+{5'h0,i_b[ 5]}+{5'h0,i_b[ 6]}+{5'h0,i_b[ 7]})
+({5'h0,i_b[ 8]}+{5'h0,i_b[ 9]}+{5'h0,i_b[10]}+{5'h0,i_b[11]})
+({5'h0,i_b[12]}+{5'h0,i_b[13]}+{5'h0,i_b[14]}+{5'h0,i_b[15]})
+({5'h0,i_b[16]}+{5'h0,i_b[17]}+{5'h0,i_b[18]}+{5'h0,i_b[19]})
+({5'h0,i_b[20]}+{5'h0,i_b[21]}+{5'h0,i_b[22]}+{5'h0,i_b[23]})
+({5'h0,i_b[24]}+{5'h0,i_b[25]}+{5'h0,i_b[26]}+{5'h0,i_b[27]})
+({5'h0,i_b[28]}+{5'h0,i_b[29]}+{5'h0,i_b[30]}+{5'h0,i_b[31]});
assign w_popc_result[31:6] = 26'h00;
 
// Prelogic for our flags registers
wire z, n, v;
reg c, pre_sign, set_ovfl;
reg c, pre_sign, set_ovfl, keep_sgn_on_ovfl;
always @(posedge i_clk)
if (i_ce) // 1 LUT
set_ovfl =(((i_op==4'h0)&&(i_a[31] != i_b[31]))//SUB&CMP
set_ovfl<=(((i_op==4'h0)&&(i_a[31] != i_b[31]))//SUB&CMP
||((i_op==4'h2)&&(i_a[31] == i_b[31])) // ADD
||(i_op == 4'h6) // LSL
||(i_op == 4'h5)); // LSR
always @(posedge i_clk)
if (i_ce) // 1 LUT
keep_sgn_on_ovfl<=
(((i_op==4'h0)&&(i_a[31] != i_b[31]))//SUB&CMP
||((i_op==4'h2)&&(i_a[31] == i_b[31]))); // ADD
 
wire [63:0] mpy_result; // Where we dump the multiply result
reg mpyhi; // Return the high half of the multiply
110,7 → 101,7
// this will cost a minimum of 132 6-LUTs.
 
wire this_is_a_multiply_op;
assign this_is_a_multiply_op = (i_ce)&&((i_op[3:1]==3'h5)||(i_op[3:0]==4'h8));
assign this_is_a_multiply_op = (i_ce)&&((i_op[3:1]==3'h5)||(i_op[3:0]==4'hc));
 
generate
if (IMPLEMENT_MPY == 0)
333,13 → 324,11
4'b0101:{o_c,c } <= w_lsr_result[32:0]; // LSR
4'b0110:{c,o_c } <= w_lsl_result[32:0]; // LSL
4'b0111:{o_c,c } <= w_asr_result[32:0]; // ASR
4'b1000: o_c <= mpy_result[31:0]; // MPY
4'b1000: o_c <= w_brev_result; // BREV
4'b1001: o_c <= { i_a[31:16], i_b[15:0] }; // LODILO
4'b1010: o_c <= mpy_result[63:32]; // MPYHU
4'b1011: o_c <= mpy_result[63:32]; // MPYHS
4'b1100: o_c <= w_brev_result; // BREV
4'b1101: o_c <= w_popc_result; // POPC
4'b1110: o_c <= w_rol_result; // ROL
4'b1010: o_c <= mpy_result[63:32]; // MPYHU
4'b1011: o_c <= mpy_result[63:32]; // MPYHS
4'b1100: o_c <= mpy_result[31:0]; // MPY
default: o_c <= i_b; // MOV, LDI
endcase
end else // if (mpydone)
359,8 → 348,9
assign z = (o_c == 32'h0000);
assign n = (o_c[31]);
assign v = (set_ovfl)&&(pre_sign != o_c[31]);
wire vx = (keep_sgn_on_ovfl)&&(pre_sign != o_c[31]);
 
assign o_f = { v, n, c, z };
assign o_f = { v, n^vx, c, z };
 
initial o_valid = 1'b0;
always @(posedge i_clk)
/dcache.v
0,0 → 1,507
////////////////////////////////////////////////////////////////////////////////
//
// Filename: dcache.v
//
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
//
// Purpose: To provide a simple data cache for the ZipCPU. The cache is
// designed to be a drop in replacement for the pipememm memory
// unit currently existing within the ZipCPU. The goal of this unit is
// to achieve single cycle read access to any memory in the last cache line
// used, or two cycle access to any memory currently in the cache.
//
// The cache separates between four types of accesses, one write and three
// read access types. The read accesses are split between those that are
// not cacheable, those that are in the cache, and those that are not.
//
// 1. Write accesses always create writes to the bus. For these reasons,
// these may always be considered cache misses.
//
// Writes to memory locations within the cache must also update
// cache memory immediately, to keep the cache in synch.
//
// It is our goal to be able to maintain single cycle write
// accesses for memory bursts.
//
// 2. Read access to non-cacheable memory locations will also immediately
// go to the bus, just as all write accesses go to the bus.
//
// 3. Read accesses to cacheable memory locations will immediately read
// from the appropriate cache line. However, since thee valid
// line will take a second clock to read, it may take up to two
// clocks to know if the memory was in cache. For this reason,
// we bypass the test for the last validly accessed cache line.
//
// We shall design these read accesses so that reads to the cache
// may take place concurrently with other writes to the bus.
//
// Errors in cache reads will void the entire cache line. For this reason,
// cache lines must always be of a smaller in size than any associated
// virtual page size--lest in the middle of reading a page a TLB miss
// take place referencing only a part of the cacheable page.
//
//
//
//
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2016, Gisselquist Technology, LLC
//
// This program is free software (firmware): you can redistribute it and/or
// modify it under the terms of the GNU General Public License as published
// by the Free Software Foundation, either version 3 of the License, or (at
// your option) any later version.
//
// This program is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
////////////////////////////////////////////////////////////////////////////////
//
//
module dcache(i_clk, i_rst, i_pipe_stb, i_lock,
i_op, i_addr, i_data, i_oreg,
o_busy, o_pipe_stalled, o_valid, o_err, o_wreg,o_data,
o_wb_cyc_gbl, o_wb_cyc_lcl, o_wb_stb_gbl, o_wb_stb_lcl,
o_wb_we, o_wb_addr, o_wb_data,
i_wb_ack, i_wb_stall, i_wb_err, i_wb_data);
parameter LGCACHELEN = 8,
ADDRESS_WIDTH=32,
LGNLINES=5, // Log of the number of separate cache lines
IMPLEMENT_LOCK=0,
NAUX=5; // # of aux d-wires to keep aligned w/memops
localparam SDRAM_BIT = 26;
localparam FLASH_BIT = 22;
localparam BLKRAM_BIT= 15;
localparam AW = ADDRESS_WIDTH; // Just for ease of notation below
localparam CS = LGCACHELEN; // Number of bits in a cache address
localparam LS = CS-LGNLINES; // Bits to spec position w/in cline
localparam LGAUX = 3; // log_2 of the maximum number of piped data
input i_clk, i_rst;
// Interface from the CPU
input i_pipe_stb, i_lock;
input i_op;
input [31:0] i_addr;
input [31:0] i_data;
input [(NAUX-1):0] i_oreg; // Aux data, such as reg to write to
// Outputs, going back to the CPU
output wire o_busy, o_pipe_stalled, o_valid, o_err;
output reg [(NAUX-1):0] o_wreg;
output reg [31:0] o_data;
// Wishbone bus master outputs
output wire o_wb_cyc_gbl, o_wb_cyc_lcl;
output reg o_wb_stb_gbl, o_wb_stb_lcl;
output reg o_wb_we;
output reg [(AW-1):0] o_wb_addr;
output reg [31:0] o_wb_data;
// Wishbone bus slave response inputs
input i_wb_ack, i_wb_stall, i_wb_err;
input [31:0] i_wb_data;
 
 
reg cyc, stb, last_ack, end_of_line, last_line_stb;
 
 
reg [((1<<LGNLINES)-1):0] c_v; // One bit per cache line, is it valid?
reg [(AW-LS-1):0] c_vtags [0:((1<<LGNLINES)-1)];
reg [31:0] c_mem [0:((1<<CS)-1)];
// reg [((1<<LGNLINES)-1):0] c_wr; // Is the cache line writable?
// reg c_wdata;
// reg c_waddr;
 
// To simplify writing to the cache, and the job of the synthesizer to
// recognize that a cache write needs to take place, we'll take an extra
// clock to get there, and use these c_w... registers to capture the
// data in the meantime.
reg c_wr;
reg [31:0] c_wdata;
reg [(CS-1):0] c_waddr;
 
reg [(AW-LS-1):0] last_tag;
 
 
wire [(LGNLINES-1):0] i_cline;
wire [(CS-1):0] i_caddr;
wire [(AW-LS-1):0] i_ctag;
 
assign i_cline = i_addr[(CS-1):LS];
assign i_caddr = i_addr[(CS-1):0];
assign i_ctag = i_addr[(AW-1):LS];
 
wire cache_miss_inow, w_cachable;
assign cache_miss_inow = (last_tag != i_addr[31:LS])||(!c_v[i_cline]);
assign w_cachable = (i_addr[31:30]!=2'b11)&&(!i_lock)&&(
((SDRAM_BIT>0)&&(i_addr[SDRAM_BIT]))
||((FLASH_BIT>0)&&(i_addr[FLASH_BIT]))
||((BLKRAM_BIT>0)&&(i_addr[BLKRAM_BIT])));
 
reg r_cachable, r_svalid, r_dvalid, r_rd, r_cache_miss, r_rvalid;
reg [(AW-1):0] r_addr;
reg [31:0] r_idata, r_ddata, r_rdata;
wire [(LGNLINES-1):0] r_cline;
wire [(CS-1):0] r_caddr;
wire [(AW-LS-1):0] r_ctag;
 
assign r_cline = r_addr[(CS-1):LS];
assign r_caddr = r_addr[(CS-1):0];
assign r_ctag = r_addr[(AW-1):LS];
 
 
reg wr_cstb, r_iv, pipeable_op, non_pipeable_op, in_cache;
reg [(AW-LS-1):0] r_itag;
 
//
// The one-clock delayed read values from the cache.
//
initial r_rd = 1'b0;
initial r_cachable = 1'b0;
initial r_svalid = 1'b0;
initial r_dvalid = 1'b0;
always @(posedge i_clk)
begin
// The single clock path
r_idata <= c_mem[i_addr[(CS-1):0]];
// The valid for the single clock path
// Only ... we need to wait if we are currently writing
// to our cache.
r_svalid<= (!i_op)&&(!cache_miss_inow)&&(w_cachable)
&&(i_pipe_stb)&&(!c_wr)&&(!wr_cstb);
 
//
// The two clock in-cache path
//
// Some preliminaries that needed to be calculated on the first
// clock
if (!o_busy)
begin
r_iv <= c_v[i_cline];
r_itag <= c_vtags[i_cline];
r_addr <= i_addr;
r_cachable <= (!i_op)&&(w_cachable)&&(i_pipe_stb);
end else begin
r_iv <= c_v[r_cline];
r_itag <= c_vtags[r_cline];
end
// r_idata still contains the right answer
r_rd <= (i_pipe_stb)&&(!i_op);
r_ddata <= r_idata;
// r_itag contains the tag we didn't have available to us on the
// last clock, r_ctag is a bit select from r_addr containing a
// one clock delayed address.
r_dvalid <= (r_itag == r_ctag)&&(r_iv)&&(r_cachable);
if ((r_itag == r_ctag)&&(r_iv)&&(r_cachable))
last_tag <= r_ctag;
 
// r_cache miss takes a clock cycle. It is only ever true for
// something that should be cachable, but isn't in the cache.
// A cache miss is only true _if_
// 1. A read was requested
// 2. It is for a cachable address, AND
// 3. It isn't in the cache on the first read
// or the second read
// 4. The read hasn't yet started to get this address
r_cache_miss <= ((!cyc)||(o_wb_we))&&(r_cachable)
// One clock path -- miss
&&(!r_svalid)
// Two clock path -- misses as well
&&(r_rd)&&(!r_svalid)
&&((r_itag != r_ctag)||(!r_iv));
 
r_rdata <= c_mem[r_addr[(CS-1):0]];
r_rvalid<= ((i_wb_ack)&&(last_ack));
end
 
`define DC_IDLE 2'b00
`define DC_WRITE 2'b01
`define DC_READS 2'b10
`define DC_READC 2'b11
reg [1:0] state;
 
reg [(AW-LS-1):0] wr_wtag, wr_vtag;
reg [31:0] wr_data;
reg [(CS-1):0] wr_addr;
always @(posedge i_clk)
begin
// By default, update the cache from the write 1-clock ago
c_wr <= (wr_cstb)&&(wr_wtag == wr_vtag);
c_wdata <= wr_data;
c_waddr <= wr_addr[(CS-1):0];
 
wr_cstb <= 1'b0;
wr_vtag <= c_vtags[o_wb_addr[(CS-LS-1):0]];
wr_wtag <= o_wb_addr[(AW-LS-1):0];
wr_data <= o_wb_data;
wr_addr <= o_wb_addr[(CS-1):0];
 
 
if (LS <= 1)
end_of_line <= 1'b1;
else
end_of_line<=(cyc)&&((c_waddr[(LS-1):1]=={(LS-1){1'b1}})
||((i_wb_ack)
&&(c_waddr[(LS-1):0]=={{(LS-2){1'b1}},2'b01})));
 
if (LS <= 1)
last_line_stb <= 1'b1;
else
last_line_stb <= (stb)&&
((o_wb_addr[(LS-1):1]=={(LS-1){1'b1}})
||((!i_wb_stall)
&&(o_wb_addr[(LS-1):0]
=={{(LS-2){1'b1}},2'b01})));
 
//
if (state == `DC_IDLE)
pipeable_op <= 1'b0;
if (state == `DC_IDLE)
non_pipeable_op <= 1'b0;
 
 
if (state == `DC_IDLE)
begin
o_wb_we <= 1'b0;
o_wb_data <= i_data;
pipeable_op <= 1'b0;
non_pipeable_op <= 1'b1;
 
cyc <= 1'b0;
stb <= 1'b0;
 
r_wb_cyc_gbl <= 1'b0;
r_wb_cyc_lcl <= 1'b0;
o_wb_stb_gbl <= 1'b0;
o_wb_stb_lcl <= 1'b0;
 
in_cache <= (i_op)&&(w_cachable);
if ((i_pipe_stb)&&(i_op))
begin // Write operation
state <= `DC_WRITE;
o_wb_addr <= i_addr;
o_wb_we <= 1'b1;
pipeable_op <= 1'b1;
 
cyc <= 1'b1;
stb <= 1'b1;
 
r_wb_cyc_gbl <= (i_addr[31:30]!=2'b11);
r_wb_cyc_lcl <= (i_addr[31:30]==2'b11);
o_wb_stb_gbl <= (i_addr[31:30]!=2'b11);
o_wb_stb_lcl <= (i_addr[31:30]==2'b11);
 
end else if (r_cache_miss)
begin
state <= `DC_READC;
o_wb_addr <= { i_ctag, {(LS){1'b0}} };
non_pipeable_op <= 1'b1;
 
cyc <= 1'b1;
stb <= 1'b1;
r_wb_cyc_gbl <= 1'b1;
o_wb_stb_gbl <= 1'b1;
end else if ((i_pipe_stb)&&(!w_cachable))
begin // Read non-cachable memory area
state <= `DC_READS;
o_wb_addr <= i_addr;
pipeable_op <= 1'b1;
 
cyc <= 1'b1;
stb <= 1'b1;
r_wb_cyc_gbl <= (i_addr[31:30]!=2'b11);
r_wb_cyc_lcl <= (i_addr[31:30]==2'b11);
o_wb_stb_gbl <= (i_addr[31:30]!=2'b11);
o_wb_stb_lcl <= (i_addr[31:30]==2'b11);
end // else we stay idle
 
end else if (state == `DC_READC)
begin
// We enter here once we have committed to reading
// data into a cache line.
if ((stb)&&(!i_wb_stall))
begin
stb <= (!last_line_stb);
o_wb_stb_gbl <= (!last_line_stb);
o_wb_addr[(LS-1):0] <= o_wb_addr[(LS-1):0]+1'b1;
end
 
if(stb)
c_v[o_wb_addr[(CS-LS-1):0]] <= 1'b0;
 
c_wr <= (i_wb_ack);
c_wdata <= o_wb_data;
c_waddr <= ((c_wr)?(c_waddr+1'b1):c_waddr);
 
c_vtags[o_wb_addr[(CS-LS-1):0]]<= o_wb_addr[(AW-LS-1):0];
 
if (((i_wb_ack)&&(end_of_line))||(i_wb_err))
begin
state <= `DC_IDLE;
non_pipeable_op <= 1'b0;
cyc <= 1'b0;
r_wb_cyc_gbl <= 1'b0;
r_wb_cyc_lcl <= 1'b0;
//
c_v[o_wb_addr[(CS-LS-1):0]] <= i_wb_ack;
end
end else if (state == `DC_READS)
begin
// We enter here once we have committed to reading
// data that cannot go into a cache line
if ((!i_wb_stall)&&(!i_pipe_stb))
begin
stb <= 1'b0;
o_wb_stb_gbl <= 1'b0;
o_wb_stb_lcl <= 1'b0;
pipeable_op <= 1'b0;
end
 
if ((!i_wb_stall)&&(i_pipe_stb))
o_wb_addr <= i_data;
 
c_wr <= 1'b0;
 
if (((i_wb_ack)&&(last_ack))||(i_wb_err))
begin
state <= `DC_IDLE;
cyc <= 1'b0;
r_wb_cyc_gbl <= 1'b0;
r_wb_cyc_lcl <= 1'b0;
end
end else if (state == `DC_WRITE)
begin
// c_wr <= (c_v[])&&(c_tag[])&&(in_cache)&&(stb);
c_wdata <= o_wb_data;
c_waddr <= (state == `DC_IDLE)?i_caddr
: ((c_wr)?(c_waddr+1'b1):c_waddr);
 
if ((!i_wb_stall)&&(!i_pipe_stb))
begin
stb <= 1'b0;
o_wb_stb_gbl <= 1'b0;
o_wb_stb_lcl <= 1'b0;
pipeable_op <= 1'b0;
end
 
wr_cstb <= (stb)&&(!i_wb_stall)&&(in_cache);
 
if ((stb)&&(!i_wb_stall)&&(i_pipe_stb))
o_wb_addr <= i_addr;
if ((stb)&&(!i_wb_stall)&&(i_pipe_stb))
o_wb_data <= i_data;
if (((i_wb_ack)&&(last_ack))||(i_wb_err))
begin
state <= `DC_IDLE;
cyc <= 1'b0;
r_wb_cyc_gbl <= 1'b0;
r_wb_cyc_lcl <= 1'b0;
end
end
end
 
//
// Writes to the cache
//
// These have been made as simple as possible. Note that the c_wr
// line has already been determined, as have the write value and address
// on the last clock. Further, this structure is defined to match the
// block RAM design of as many architectures as possible.
//
always @(posedge i_clk)
if (c_wr)
c_mem[c_waddr] <= c_wdata;
 
//
// Reads from the cache
//
// Some architectures require that all reads be registered. We
// accomplish that here. Whether or not the result of this read is
// going to be our output will need to be determined with combinatorial
// logic on the output.
//
reg [31:0] cached_idata, cached_rdata;
always @(posedge i_clk)
cached_idata <= c_mem[i_caddr];
 
always @(posedge i_clk)
cached_rdata <= c_mem[r_caddr];
 
// o_data can come from one of three places:
// 1. The cache, assuming the data was in the last cache line
// 2. The cache, second clock, assuming the data was in the cache at all
// 3. The cache, after filling the cache
// 4. The wishbone state machine, upon reading the value desired.
always @(posedge i_clk)
if (r_svalid)
o_data <= cached_idata;
else if ((i_wb_ack)&&(pipeable_op))
o_data <= i_wb_data;
else
o_data <= cached_rdata;
always @(posedge i_clk)
o_valid <= (r_svalid)||((i_wb_ack)&&(pipeable_op))
||(r_dvalid)||(r_rvalid);
always @(posedge i_clk)
o_err <= (cyc)&&(i_wb_err);
 
assign o_busy = (state != `DC_IDLE);
 
 
//
// Handle our auxilliary data lines.
//
// These just go into a FIFO upon request, and then get fed back out
// upon completion of an OP.
//
// These are currently designed for handling bursts of writes or
// non-cachable reads.
//
// A very similar structure will be used once we switch to using an
// MMU, in order to make certain memory operations are synchronous
// enough to deal with bus errors.
//
reg [(LGAUX-1):0] aux_head, aux_tail;
reg [(NAUX-1):0] aux_fifo [0:((1<<LGAUX)-1)];
initial aux_head = 0;
initial aux_tail = 0;
always @(posedge i_clk)
begin
if ((i_rst)||(i_wb_err))
aux_head <= 0;
else if ((i_pipe_stb)&&(!o_busy))
aux_head <= aux_head + 1'b1;
aux_fifo[aux_head] <= i_oreg;
end
always @(posedge i_clk)
begin
if ((i_rst)||(i_wb_err))
aux_tail <= 0;
else if (o_valid) // ||(aux_tail[WBIT])&&(no-mmu-error)
aux_tail <= aux_tail + 1'b1;
o_wreg <= aux_fifo[aux_tail];
end
 
//
// We can use our FIFO addresses to pre-calculate when an ACK is going
// to be the last_noncachable_ack.
 
 
assign o_pipe_stalled=((pipeable_op)&&(i_wb_stall))||(non_pipeable_op);
// pipeable_op must become zero when stb goes low
 
always @(posedge i_clk)
begin
lock_gbl <= (i_lock)&&((r_wb_cyc_gbl)||(lock_gbl));
lock_lcl <= (i_lock)&&((r_wb_cyc_lcl)||(lock_lcl));
end
 
assign o_wb_cyc_gbl = (r_wb_cyc_gbl)||(lock_gbl);
assign o_wb_cyc_lcl = (r_wb_cyc_lcl)||(lock_lcl);
endmodule
/div.v
1,18 → 1,77
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: div.v
//
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
//
// Purpose: Provide an Integer divide capability to the Zip CPU.
// Purpose: Provide an Integer divide capability to the Zip CPU. Provides
// for both signed and unsigned divide.
//
// Steps:
// i_rst The DIVide unit starts in idle. It can also be placed into an
// idle by asserting the reset input.
//
// i_wr When i_rst is asserted, a divide begins. On the next clock:
//
// o_busy is set high so everyone else knows we are at work and they can
// wait for us to complete.
//
// pre_sign is set to true if we need to do a signed divide. In this
// case, we take a clock cycle to turn the divide into an unsigned
// divide.
//
// o_quotient, a place to store our result, is initialized to all zeros.
//
// r_dividend is set to the numerator
//
// r_divisor is set to 2^31 * the denominator (shift left by 31, or add
// 31 zeros to the right of the number.
//
// pre_sign When true (clock cycle after i_wr), a clock cycle is used
// to take the absolute value of the various arguments (r_dividend
// and r_divisor), and to calculate what sign the output result
// should be.
//
//
// At this point, the divide is has started. The divide works by walking
// through every shift of the
//
// DIVIDEND over the
// DIVISOR
//
// If the DIVISOR is bigger than the dividend, the divisor is shifted
// right, and nothing is done to the output quotient.
//
// DIVIDEND
// DIVISOR
//
// This repeats, until DIVISOR is less than or equal to the divident, as in
//
// DIVIDEND
// DIVISOR
//
// At this point, if the DIVISOR is less than the dividend, the
// divisor is subtracted from the dividend, and the DIVISOR is again
// shifted to the right. Further, a '1' bit gets set in the output
// quotient.
//
// Once we've done this for 32 clocks, we've accumulated our answer into
// the output quotient, and we can proceed to the next step. If the
// result will be signed, the next step negates the quotient, otherwise
// it returns the result.
//
// On the clock when we are done, o_busy is set to false, and o_valid set
// to true. (It is a violation of the ZipCPU internal protocol for both
// busy and valid to ever be true on the same clock. It is also a
// violation for busy to be false with valid true thereafter.)
//
//
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
24,12 → 83,18
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
//
// `include "cpudefs.v"
//
module div(i_clk, i_rst, i_wr, i_signed, i_numerator, i_denominator,
/idecode.v
1,4 → 1,4
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: idecode.v
//
17,9 → 17,9
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
31,14 → 31,20
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
//
//
`define CPU_SP_REG 4'hd
`define CPU_CC_REG 4'he
`define CPU_PC_REG 4'hf
//
56,7 → 62,8
o_op, o_ALU, o_M, o_DV, o_FP, o_break, o_lock,
o_wR, o_rA, o_rB,
o_early_branch, o_branch_pc, o_ljmp,
o_pipe
o_pipe,
o_sim, o_sim_immv
);
parameter ADDRESS_WIDTH=24, IMPLEMENT_MPY=1, EARLY_BRANCHING=1,
IMPLEMENT_DIVIDE=1, IMPLEMENT_FPU=0, AW = ADDRESS_WIDTH;
67,7 → 74,7
input i_pf_valid, i_illegal;
output wire o_phase;
output reg o_illegal;
output reg [(AW-1):0] o_pc;
output reg [AW:0] o_pc;
output reg o_gie;
output reg [6:0] o_dcdR, o_dcdA, o_dcdB;
output wire [31:0] o_I;
82,6 → 89,8
output wire [(AW-1):0] o_branch_pc;
output wire o_ljmp;
output wire o_pipe;
output reg o_sim /* verilator public_flat */;
output reg [22:0] o_sim_immv /* verilator public_flat */;
 
wire dcdA_stall, dcdB_stall, dcdF_stall;
wire o_dcd_early_branch;
102,18 → 111,18
wire w_dcdA_pc, w_dcdA_cc;
wire w_dcdB_pc, w_dcdB_cc;
wire [3:0] w_cond;
wire w_wF, w_dcdM, w_dcdDV, w_dcdFP;
wire w_wF, w_mem, w_sto, w_lod, w_div, w_fpu;
wire w_wR, w_rA, w_rB, w_wR_n;
wire w_ljmp, w_ljmp_dly;
wire w_ljmp, w_ljmp_dly, w_cis_ljmp;
wire [31:0] iword;
 
 
`ifdef OPT_VLIW
reg [16:0] r_nxt_half;
`ifdef OPT_CIS
reg [15:0] r_nxt_half;
assign iword = (o_phase)
// set second half as a NOOP ... but really
// set second half as a NOOP ... but really
// shouldn't matter
? { r_nxt_half[16:7], 1'b0, r_nxt_half[6:0], 5'b11000, 3'h7, 6'h00 }
? { r_nxt_half[15:0], i_instruction[15:0] }
: i_instruction;
`else
assign iword = { 1'b0, i_instruction[30:0] };
121,27 → 130,87
 
generate
if (EARLY_BRANCHING != 0)
begin
`ifdef OPT_CIS
reg r_pre_ljmp;
always @(posedge i_clk)
if ((i_rst)||(o_early_branch))
r_pre_ljmp <= 1'b0;
else if ((i_ce)&&(i_pf_valid))
r_pre_ljmp <= (!o_phase)&&(i_instruction[31])
&&(i_instruction[14:0] == 15'h7cf8);
else if (i_ce)
r_pre_ljmp <= 1'b0;
 
assign w_cis_ljmp = r_pre_ljmp;
`else
assign w_cis_ljmp = 1'b0;
`endif
// 0.1111.10010.000.1.1111.000000000...
// 0111.1100.1000.0111.11000....
assign w_ljmp = (iword == 32'h7c87c000);
else
end else begin
assign w_cis_ljmp = 1'b0;
assign w_ljmp = 1'b0;
end
endgenerate
 
`ifdef OPT_CIS
`ifdef VERILATOR
wire [4:0] w_cis_op;
always @(iword)
if (!iword[31])
w_cis_op = w_op;
else case(iword[26:24])
3'h0: w_cis_op = 5'h00;
3'h1: w_cis_op = 5'h01;
3'h2: w_cis_op = 5'h02;
3'h3: w_cis_op = 5'h10;
3'h4: w_cis_op = 5'h12;
3'h5: w_cis_op = 5'h13;
3'h6: w_cis_op = 5'h18;
3'h7: w_cis_op = 5'h0d;
endcase
`else
reg [4:0] w_cis_op;
always @(iword,w_op)
if (!iword[31])
w_cis_op <= w_op;
else case(iword[26:24])
3'h0: w_cis_op <= 5'h00;
3'h1: w_cis_op <= 5'h01;
3'h2: w_cis_op <= 5'h02;
3'h3: w_cis_op <= 5'h10;
3'h4: w_cis_op <= 5'h12;
3'h5: w_cis_op <= 5'h13;
3'h6: w_cis_op <= 5'h18;
3'h7: w_cis_op <= 5'h0d;
endcase
`endif
`else
wire [4:0] w_cis_op;
assign w_cis_op = w_op;
`endif
 
assign w_op= iword[26:22];
assign w_mov = (w_op == 5'h0f);
assign w_ldi = (w_op[4:1] == 4'hb);
assign w_brev = (w_op == 5'hc);
assign w_cmptst = (w_op[4:1] == 4'h8);
assign w_ldilo = (w_op[4:0] == 5'h9);
assign w_ALU = (~w_op[4]);
assign w_mov = (w_cis_op == 5'h0d);
assign w_ldi = (w_cis_op[4:1] == 4'hc);
assign w_brev = (w_cis_op == 5'h8);
assign w_cmptst = (w_cis_op[4:1] == 4'h8);
assign w_ldilo = (w_cis_op[4:0] == 5'h9);
assign w_ALU = (!w_cis_op[4]) // anything with [4]==0, but ...
&&(w_cis_op[3:1] != 3'h7); // not the divide
 
// 4 LUTs
 
// w_dcdR (4 LUTs)
//
// What register will we be placing results into (if at all)?
//
// Two parts to the result register: the register set, given for
// moves in i_word[18] but only for the supervisor, and the other
// moves in iword[18] but only for the supervisor, and the other
// four bits encoded in the instruction.
//
assign w_dcdR = { ((~iword[31])&&(w_mov)&&(~i_gie))?iword[18]:i_gie,
assign w_dcdR = { ((!iword[31])&&(w_mov)&&(~i_gie))?iword[18]:i_gie,
iword[30:27] };
// 2 LUTs
//
148,16 → 217,20
// If the result register is either CC or PC, and this would otherwise
// be a floating point instruction with floating point opcode of 0,
// then this is a NOOP.
assign w_noop = (w_op[4:0] == 5'h18)&&(
assign w_noop = (!iword[31])&&(w_op[4:0] == 5'h1f)&&(
((IMPLEMENT_FPU>0)&&(w_dcdR[3:1] == 3'h7))
||(IMPLEMENT_FPU==0));
 
// 4 LUTs
assign w_dcdB = { ((~iword[31])&&(w_mov)&&(~i_gie))?iword[13]:i_gie,
iword[17:14] };
// dcdB - What register is used in the opB?
//
assign w_dcdB[4] = ((!iword[31])&&(w_mov)&&(~i_gie))?iword[13]:i_gie;
assign w_dcdB[3:0]= (iword[31])
? (((!iword[23])&&(iword[26:25]==2'b10))
? `CPU_SP_REG : iword[22:19])
: iword[17:14];
 
// 0 LUTs
assign w_dcdA = w_dcdR;
assign w_dcdA = w_dcdR; // on ZipCPU, A is always result reg
// 2 LUTs, 1 delay each
assign w_dcdR_pc = (w_dcdR == {i_gie, `CPU_PC_REG});
assign w_dcdR_cc = (w_dcdR == {i_gie, `CPU_CC_REG});
165,8 → 238,8
assign w_dcdA_pc = w_dcdR_pc;
assign w_dcdA_cc = w_dcdR_cc;
// 2 LUTs, 1 delays each
assign w_dcdB_pc = (w_dcdB[3:0] == `CPU_PC_REG);
assign w_dcdB_cc = (w_dcdB[3:0] == `CPU_CC_REG);
assign w_dcdB_pc = (w_rB)&&(w_dcdB[3:0] == `CPU_PC_REG);
assign w_dcdB_cc = (w_rB)&&(w_dcdB[3:0] == `CPU_CC_REG);
 
// Under what condition will we execute this
// instruction? Only the load immediate instruction
173,46 → 246,55
// is completely unconditional.
//
// 3+4 LUTs
assign w_cond = (w_ldi) ? 4'h8 :
(iword[31])?{(iword[20:19]==2'b00),
1'b0,iword[20:19]}
: { (iword[21:19]==3'h0), iword[21:19] };
assign w_cond = ((w_ldi)||(iword[31])) ? 4'h8 :
{ (iword[21:19]==3'h0), iword[21:19] };
 
// 1 LUT
assign w_dcdM = (w_op[4:1] == 4'h9);
assign w_mem = (w_cis_op[4:3] == 2'b10)&&(w_cis_op[2:1] !=2'b00);
assign w_sto = (w_mem)&&( w_cis_op[0]);
assign w_lod = (w_mem)&&(!w_cis_op[0]);
// 1 LUT
assign w_dcdDV = (w_op[4:1] == 4'ha);
// 1 LUT
assign w_dcdFP = (w_op[4:3] == 2'b11)&&(w_dcdR[3:1] != 3'h7);
// 4 LUT's--since it depends upon FP/NOOP condition (vs 1 before)
// Everything reads A but ... NOOP/BREAK/LOCK, LDI, LOD, MOV
assign w_rA = (w_dcdFP)
// Divide's read A
||(w_dcdDV)
// ALU read's A, unless it's a MOV to A
// This includes LDIHI/LDILO
||((~w_op[4])&&(w_op[3:0]!=4'hf))
// STO's read A
||((w_dcdM)&&(w_op[0]))
// Test/compares
||(w_op[4:1]== 4'h8);
// 1 LUTs -- do we read a register for operand B? Specifically, do
// we need to stall if the register is not (yet) ready?
assign w_rB = (w_mov)||((iword[18])&&(~w_ldi));
assign w_div = (!iword[31])&&(w_op[4:1] == 4'h7);
// 2 LUTs
assign w_fpu = (!iword[31])&&(w_op[4:3] == 2'b11)
&&(w_dcdR[3:1] != 3'h7)&&(w_op[2:1] != 2'b00);
//
// rA - do we need to read register A?
assign w_rA = // Floating point reads reg A
((w_fpu)&&(w_cis_op[4:1] != 4'hf))
// Divide's read A
||(w_div)
// ALU ops read A,
// except for MOV's and BREV's which don't
||((w_ALU)&&(!w_brev)&&(!w_mov))
// STO's read A
||(w_sto)
// Test/compares
||(w_cmptst);
// rB -- do we read a register for operand B? Specifically, do we
// add the registers value to the immediate to create opB?
assign w_rB = (w_mov)
||((!iword[31])&&(iword[18])&&(!w_ldi))
||(( iword[31])&&(iword[23])&&(!w_ldi))
// If using compressed instruction sets,
// we *always* read on memory operands.
||(( iword[31])&&(w_mem));
// wR -- will we be writing our result back?
// wR_n = !wR
// 1 LUT: All but STO, NOOP/BREAK/LOCK, and CMP/TST write back to w_dcdR
assign w_wR_n = ((w_dcdM)&&(w_op[0]))
||((w_op[4:3]==2'b11)&&(w_dcdR[3:1]==3'h7))
assign w_wR_n = (w_sto)
||((!iword[31])&&(w_cis_op[4:3]==2'b11)
&&(w_cis_op[2:1]!=2'b00)
&&(w_dcdR[3:1]==3'h7))
||(w_cmptst);
assign w_wR = ~w_wR_n;
//
// 1-output bit (5 Opcode bits, 4 out-reg bits, 3 condition bits)
//
// This'd be 4 LUTs, save that we have the carve out for NOOPs
// and writes to the PC/CC register(s).
// wF -- do we write flags when we are done?
//
assign w_wF = (w_cmptst)
||((w_cond[3])&&((w_dcdFP)||(w_dcdDV)
||((w_ALU)&&(~w_mov)&&(~w_ldilo)&&(~w_brev)
&&(iword[30:28] != 3'h7))));
||((w_cond[3])&&((w_fpu)||(w_div)
||((w_ALU)&&(!w_mov)&&(!w_ldilo)&&(!w_brev)
&&(w_dcdR[3:1] != 3'h7))));
 
// Bottom 13 bits: no LUT's
// w_dcd[12: 0] -- no LUTs
224,16 → 306,23
wire w_Iz;
 
assign w_fullI = (w_ldi) ? { iword[22:0] } // LDI
:((w_mov) ?{ {(23-13){iword[12]}}, iword[12:0] } // Move
// MOVE immediates have one less bit
:((w_mov) ?{ {(23-13){iword[12]}}, iword[12:0] }
// Normal Op-B immediate ... 18 or 14 bits
:((~iword[18]) ? { {(23-18){iword[17]}}, iword[17:0] }
: { {(23-14){iword[13]}}, iword[13:0] }
));
 
`ifdef OPT_VLIW
wire [5:0] w_halfI;
assign w_halfI = (w_ldi) ? iword[5:0]
:((iword[5]) ? 6'h00 : {iword[4],iword[4:0]});
assign w_I = (iword[31])? {{(23-6){w_halfI[5]}}, w_halfI }:w_fullI;
`ifdef OPT_CIS
wire [7:0] w_halfbits;
assign w_halfbits = iword[23:16];
 
wire [7:0] w_halfI;
assign w_halfI = (iword[26:24]==3'h6) ? w_halfbits[7:0]
:(w_halfbits[7])?
{ {(6){w_halfbits[2]}}, w_halfbits[1:0]}
:{ w_halfbits[6], w_halfbits[6:0] };
assign w_I = (iword[31])?{{(23-8){w_halfI[7]}}, w_halfI }:w_fullI;
`else
assign w_I = w_fullI;
`endif
240,12 → 329,12
assign w_Iz = (w_I == 0);
 
 
`ifdef OPT_VLIW
`ifdef OPT_CIS
//
// The o_phase parameter is special. It needs to let the software
// following know that it cannot break/interrupt on an o_phase asserted
// instruction, lest the break take place between the first and second
// half of a VLIW instruction. To do this, o_phase must be asserted
// half of a CIS instruction. To do this, o_phase must be asserted
// when the first instruction half is valid, but not asserted on either
// a 32-bit instruction or the second half of a 2x16-bit instruction.
reg r_phase;
255,7 → 344,10
||(o_early_branch)||(w_ljmp_dly))
r_phase <= 1'b0;
else if ((i_ce)&&(i_pf_valid))
r_phase <= (o_phase)? 1'b0:(i_instruction[31]);
r_phase <= (o_phase)? 1'b0
: ((i_instruction[31])&&(i_pf_valid));
else if (i_ce)
r_phase <= 1'b0;
// Phase is '1' on the first instruction of a two-part set
// But, due to the delay in processing, it's '1' when our output is
// valid for that first part, but that'll be the same time we
274,32 → 366,32
o_illegal <= 1'b0;
else if (i_ce)
begin
`ifdef OPT_VLIW
`ifdef OPT_CIS
o_illegal <= (i_illegal);
`else
o_illegal <= ((i_illegal) || (i_instruction[31]));
`endif
if ((IMPLEMENT_MPY==0)&&((w_op[4:1]==4'h5)||(w_op[4:0]==5'h08)))
if ((IMPLEMENT_MPY==0)&&((w_cis_op[4:1]==4'h5)||(w_cis_op[4:0]==5'h0c)))
o_illegal <= 1'b1;
 
if ((IMPLEMENT_DIVIDE==0)&&(w_dcdDV))
if ((IMPLEMENT_DIVIDE==0)&&(w_div))
o_illegal <= 1'b1;
else if ((IMPLEMENT_DIVIDE!=0)&&(w_dcdDV)&&(w_dcdR[3:1]==3'h7))
else if ((IMPLEMENT_DIVIDE!=0)&&(w_div)&&(w_dcdR[3:1]==3'h7))
o_illegal <= 1'b1;
 
 
if ((IMPLEMENT_FPU!=0)&&(w_dcdFP)&&(w_dcdR[3:1]==3'h7))
if ((IMPLEMENT_FPU==0)&&(w_fpu))
o_illegal <= 1'b1;
else if ((IMPLEMENT_FPU==0)&&(w_dcdFP))
o_illegal <= 1'b1;
 
if ((w_op[4:3]==2'b11)&&(w_dcdR[3:1]==3'h7)
if ((w_cis_op[4:3]==2'b11)&&(w_cis_op[2:1]!=2'b00)
&&(w_dcdR[3:1]==3'h7)
&&(
(w_op[2:0] != 3'h1) // BREAK
(w_cis_op[2:0] != 3'h4) // BREAK
`ifdef OPT_PIPELINED
&&(w_op[2:0] != 3'h2) // LOCK
&&(w_cis_op[2:0] != 3'h5) // LOCK
`endif
&&(w_op[2:0] != 3'h0))) // NOOP
// SIM instructions are always illegal
&&(w_cis_op[2:0] != 3'h7))) // NOOP
o_illegal <= 1'b1;
end
 
307,16 → 399,19
always @(posedge i_clk)
if (i_ce)
begin
`ifdef OPT_VLIW
if (~o_phase)
begin
`ifdef OPT_CIS
if (!o_phase)
o_gie<= i_gie;
// i.e. dcd_pc+1
o_pc <= i_pc+{{(AW-1){1'b0}},1'b1};
end
 
if ((iword[31])&&(!o_phase))
o_pc <= { i_pc, 1'b1 };
else if ((iword[31])&&(i_pf_valid))
o_pc <= { i_pc, 1'b0 };
else
o_pc <= { i_pc + 1'b1, 1'b0 };
`else
o_gie<= i_gie;
o_pc <= i_pc+{{(AW-1){1'b0}},1'b1};
o_pc <= { i_pc + 1'b1, 1'b0 };
`endif
 
// Under what condition will we execute this
334,12 → 429,12
// the ALU. Likewise, the two compare instructions
// CMP and TST becomes SUB and AND here as well.
// We keep only the bottom four bits, since we've
// already done the rest of the decode necessary to
// already done the rest of the decode necessary to
// settle between the other instructions. For example,
// o_FP plus these four bits uniquely defines the FP
// instruction, o_DV plus the bottom of these defines
// the divide, etc.
o_op <= (w_ldi)||(w_noop)? 4'hf:w_op[3:0];
o_op <= ((w_ldi)||(w_noop))? 4'hd : w_cis_op[3:0];
 
// Default values
o_dcdR <= { w_dcdR_cc, w_dcdR_pc, w_dcdR};
351,29 → 446,36
r_I <= w_I;
o_zI <= w_Iz;
 
// Turn a NOOP into an ALU operation--subtract in
// Turn a NOOP into an ALU operation--subtract in
// particular, although it doesn't really matter as long
// as it doesn't take longer than one clock. Note
// also that this depends upon not setting any registers
// or flags, which should already be true.
o_ALU <= (w_ALU)||(w_ldi)||(w_cmptst)||(w_noop); // 2 LUT
o_M <= w_dcdM;
o_DV <= w_dcdDV;
o_FP <= w_dcdFP;
o_break <= (w_op[4:0]==5'b11001)&&(
o_ALU <= (w_ALU)||(w_ldi)||(w_cmptst)||(w_noop);
o_M <= w_mem;
o_DV <= w_div;
o_FP <= w_fpu;
 
o_break <= (!iword[31])&&(w_op[4:0]==5'h1c)&&(
((IMPLEMENT_FPU>0)&&(w_dcdR[3:1]==3'h7))
||(IMPLEMENT_FPU==0));
`ifdef OPT_PIPELINED
r_lock <= (w_op[4:0]==5'b11010)&&(
r_lock <= (!iword[31])&&(w_op[4:0]==5'h1d)&&(
((IMPLEMENT_FPU>0)&&(w_dcdR[3:1]==3'h7))
||(IMPLEMENT_FPU==0));
`endif
`ifdef OPT_VLIW
r_nxt_half <= { iword[31], iword[13:5],
((iword[21])? iword[20:19] : 2'h0),
iword[4:0] };
`ifdef OPT_CIS
r_nxt_half <= { iword[31], iword[14:0] };
`endif
 
`ifdef VERILATOR
// Support the SIM instruction(s)
o_sim <= (!iword[31])&&(w_op[4:1] == 4'hf)
&&(w_dcdR[3:1] == 3'h7);
`else
o_sim <= 1'b0;
`endif
o_sim_immv <= iword[22:0];
end
 
`ifdef OPT_PIPELINED
392,6 → 494,10
always @(posedge i_clk)
if (i_rst)
r_ljmp <= 1'b0;
`ifdef OPT_CIS
else if ((i_ce)&&(o_phase))
r_ljmp <= w_cis_ljmp;
`endif
else if ((i_ce)&&(i_pf_valid))
r_ljmp <= (w_ljmp);
assign o_ljmp = r_ljmp;
404,17 → 510,14
if (r_ljmp)
// LOD (PC),PC
r_early_branch <= 1'b1;
else if ((~iword[31])&&(iword[30:27]==`CPU_PC_REG)&&(w_cond[3]))
else if ((!iword[31])&&(iword[30:27]==`CPU_PC_REG)
&&(w_cond[3]))
begin
if (w_op[4:1] == 4'hb) // LDI to PC
// LDI x,PC
r_early_branch <= 1'b1;
else if ((w_op[4:0]==5'h02)&&(~iword[18]))
if ((w_op[4:0]==5'h02)&&(!iword[18]))
// Add x,PC
r_early_branch <= 1'b1;
else begin
else
r_early_branch <= 1'b0;
end
end else
r_early_branch <= 1'b0;
end else if (i_ce)
424,12 → 527,10
if (i_ce)
begin
if (r_ljmp)
r_branch_pc <= iword[(AW-1):0];
else if (w_op[4:1] == 4'hb) // LDI
r_branch_pc <= {{(AW-23){iword[22]}},iword[22:0]};
r_branch_pc <= iword[(AW+1):2];
else // Add x,PC
r_branch_pc <= i_pc
+ {{(AW-17){iword[17]}},iword[16:0]}
+ {{(AW-15){iword[17]}},iword[16:2]}
+ {{(AW-1){1'b0}},1'b1};
end
 
459,15 → 560,26
initial r_pipe = 1'b0;
always @(posedge i_clk)
if (i_ce)
r_pipe <= (r_valid)&&(i_pf_valid)&&(~i_instruction[31])
&&(w_dcdM)&&(o_M)&&(o_op[0] ==i_instruction[22])
&&(i_instruction[17:14] == o_dcdB[3:0])
&&(i_instruction[17:14] != o_dcdA[3:0])
r_pipe <= (r_valid)&&((i_pf_valid)||(o_phase))
// Both must be memory operations
&&(w_mem)&&(o_M)
// Both must be writes, or both stores
&&(o_op[0] == w_cis_op[0])
// Both must be register ops
&&(w_rB)
// Both must use the same register for B
&&(w_dcdB[3:0] == o_dcdB[3:0])
// But ... the result can never be B
&&((o_op[0])
||(w_dcdB[3:0] != o_dcdA[3:0]))
// Needs to be to the mode, supervisor or user
&&(i_gie == o_gie)
// Same condition, or no condition before
&&((i_instruction[21:19]==o_cond[2:0])
||(o_cond[2:0] == 3'h0))
&&((i_instruction[13:0]==r_I[13:0])
||({1'b0, i_instruction[13:0]}==(r_I[13:0]+14'h1)));
// Same immediate
&&((w_I[13:2]==r_I[13:2])
||({1'b0, w_I[13:2]}==(r_I[13:2]+12'h1)));
assign o_pipe = r_pipe;
`else
assign o_pipe = 1'b0;
482,8 → 594,8
r_valid <= 1'b1;
else if (~i_stalled)
r_valid <= 1'b0;
 
 
assign o_I = { {(32-22){r_I[22]}}, r_I[21:0] };
 
endmodule
/memops.v
1,4 → 1,4
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: memops.v
//
17,9 → 17,9
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015, Gisselquist Technology, LLC
// Copyright (C) 2015,2017, 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
31,24 → 31,31
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
//
module memops(i_clk, i_rst, i_stb, i_lock,
i_op, i_addr, i_data, i_oreg,
o_busy, o_valid, o_err, o_wreg, o_result,
o_wb_cyc_gbl, o_wb_cyc_lcl,
o_wb_stb_gbl, o_wb_stb_lcl,
o_wb_we, o_wb_addr, o_wb_data,
o_wb_we, o_wb_addr, o_wb_data, o_wb_sel,
i_wb_ack, i_wb_stall, i_wb_err, i_wb_data);
parameter ADDRESS_WIDTH=24, IMPLEMENT_LOCK=0, AW=ADDRESS_WIDTH;
parameter ADDRESS_WIDTH=30, IMPLEMENT_LOCK=0, WITH_LOCAL_BUS=0;
localparam AW=ADDRESS_WIDTH;
input i_clk, i_rst;
input i_stb, i_lock;
// CPU interface
input i_op;
input [2:0] i_op;
input [31:0] i_addr;
input [31:0] i_data;
input [4:0] i_oreg;
66,6 → 73,7
output reg o_wb_we;
output reg [(AW-1):0] o_wb_addr;
output reg [31:0] o_wb_data;
output reg [3:0] o_wb_sel;
// Wishbone inputs
input i_wb_ack, i_wb_stall, i_wb_err;
input [31:0] i_wb_data;
72,8 → 80,8
 
reg r_wb_cyc_gbl, r_wb_cyc_lcl;
wire gbl_stb, lcl_stb;
assign lcl_stb = (i_stb)&&(i_addr[31:8]==24'hc00000)&&(i_addr[7:5]==3'h0);
assign gbl_stb = (i_stb)&&((i_addr[31:8]!=24'hc00000)||(i_addr[7:5]!=3'h0));
assign lcl_stb = (i_stb)&&(WITH_LOCAL_BUS!=0)&&(i_addr[31:24]==8'hff);
assign gbl_stb = (i_stb)&&((WITH_LOCAL_BUS==0)||(i_addr[31:24]!=8'hff));
 
initial r_wb_cyc_gbl = 1'b0;
initial r_wb_cyc_lcl = 1'b0;
104,13 → 112,54
o_wb_stb_lcl <= (o_wb_stb_lcl)&&(i_wb_stall);
else
o_wb_stb_lcl <= lcl_stb; // Grab wishbone on new operation
 
reg [3:0] r_op;
always @(posedge i_clk)
if (i_stb)
begin
o_wb_we <= i_op;
o_wb_data <= i_data;
o_wb_addr <= i_addr[(AW-1):0];
o_wb_we <= i_op[0];
casez({ i_op[2:1], i_addr[1:0] })
`ifdef ZERO_ON_IDLE
4'b100?: o_wb_data <= { i_data[15:0], 16'h00 };
4'b101?: o_wb_data <= { 16'h00, i_data[15:0] };
4'b1100: o_wb_data <= { i_data[7:0], 24'h00 };
4'b1101: o_wb_data <= { 8'h00, i_data[7:0], 16'h00 };
4'b1110: o_wb_data <= { 16'h00, i_data[7:0], 8'h00 };
4'b1111: o_wb_data <= { 24'h00, i_data[7:0] };
`else
4'b10??: o_wb_data <= { (2){ i_data[15:0] } };
4'b11??: o_wb_data <= { (4){ i_data[7:0] } };
`endif
default: o_wb_data <= i_data;
endcase
 
o_wb_addr <= i_addr[(AW+1):2];
`ifdef SET_SEL_ON_READ
if (i_op[0] == 1'b0)
o_wb_sel <= 4'hf;
else
`endif
casez({ i_op[2:1], i_addr[1:0] })
4'b01??: o_wb_sel <= 4'b1111;
4'b100?: o_wb_sel <= 4'b1100;
4'b101?: o_wb_sel <= 4'b0011;
4'b1100: o_wb_sel <= 4'b1000;
4'b1101: o_wb_sel <= 4'b0100;
4'b1110: o_wb_sel <= 4'b0010;
4'b1111: o_wb_sel <= 4'b0001;
default: o_wb_sel <= 4'b1111;
endcase
r_op <= { i_op[2:1] , i_addr[1:0] };
end
`ifdef ZERO_ON_IDLE
else if ((!o_wb_cyc_gbl)&&(!o_wb_cyc_lcl))
begin
o_wb_we <= 1'b0;
o_wb_addr <= 0;
o_wb_data <= 32'h0;
o_wb_sel <= 4'h0;
end
`endif
 
initial o_valid = 1'b0;
always @(posedge i_clk)
124,8 → 173,21
if (i_stb)
o_wreg <= i_oreg;
always @(posedge i_clk)
if (i_wb_ack)
o_result <= i_wb_data;
`ifdef ZERO_ON_IDLE
if (!i_wb_ack)
o_result <= 32'h0;
else
`endif
casez(r_op)
4'b01??: o_result <= i_wb_data;
4'b100?: o_result <= { 16'h00, i_wb_data[31:16] };
4'b101?: o_result <= { 16'h00, i_wb_data[15: 0] };
4'b1100: o_result <= { 24'h00, i_wb_data[31:24] };
4'b1101: o_result <= { 24'h00, i_wb_data[23:16] };
4'b1110: o_result <= { 24'h00, i_wb_data[15: 8] };
4'b1111: o_result <= { 24'h00, i_wb_data[ 7: 0] };
default: o_result <= i_wb_data;
endcase
 
generate
if (IMPLEMENT_LOCK != 0)
/pfcache.v
13,7 → 13,7
//
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
25,6 → 25,11
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
31,6 → 36,7
//
////////////////////////////////////////////////////////////////////////////////
//
//
module pfcache(i_clk, i_rst, i_new_pc, i_clear_cache,
// i_early_branch, i_from_addr,
i_stall_n, i_pc, o_i, o_pc, o_v,
/pipefetch.v
30,7 → 30,7
//
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015, Gisselquist Technology, LLC
// Copyright (C) 2015,2017, 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
42,6 → 42,11
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
48,6 → 53,7
//
////////////////////////////////////////////////////////////////////////////////
//
//
module pipefetch(i_clk, i_rst, i_new_pc, i_clear_cache, i_stall_n, i_pc,
o_i, o_pc, o_v,
o_wb_cyc, o_wb_stb, o_wb_we, o_wb_addr, o_wb_data,
/pipemem.v
1,4 → 1,4
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: pipemem.v
//
15,9 → 15,9
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
29,24 → 29,31
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory, run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
//
module pipemem(i_clk, i_rst, i_pipe_stb, i_lock,
i_op, i_addr, i_data, i_oreg,
o_busy, o_pipe_stalled, o_valid, o_err, o_wreg, o_result,
o_wb_cyc_gbl, o_wb_cyc_lcl,
o_wb_stb_gbl, o_wb_stb_lcl,
o_wb_we, o_wb_addr, o_wb_data,
o_wb_we, o_wb_addr, o_wb_data, o_wb_sel,
i_wb_ack, i_wb_stall, i_wb_err, i_wb_data);
parameter ADDRESS_WIDTH=32, IMPLEMENT_LOCK=0, AW=ADDRESS_WIDTH;
parameter ADDRESS_WIDTH=30, IMPLEMENT_LOCK=0;
localparam AW=ADDRESS_WIDTH;
input i_clk, i_rst;
input i_pipe_stb, i_lock;
// CPU interface
input i_op;
input [2:0] i_op;
input [31:0] i_addr;
input [31:0] i_data;
input [4:0] i_oreg;
64,6 → 71,7
output reg o_wb_stb_lcl, o_wb_we;
output reg [(AW-1):0] o_wb_addr;
output reg [31:0] o_wb_data;
output reg [3:0] o_wb_sel;
// Wishbone inputs
input i_wb_ack, i_wb_stall, i_wb_err;
input [31:0] i_wb_data;
72,25 → 80,27
reg r_wb_cyc_gbl, r_wb_cyc_lcl;
reg [3:0] rdaddr, wraddr;
wire [3:0] nxt_rdaddr;
reg [(5-1):0] fifo_oreg [0:15];
reg [(4+5-1):0] fifo_oreg [0:15];
initial rdaddr = 0;
initial wraddr = 0;
 
always @(posedge i_clk)
fifo_oreg[wraddr] <= i_oreg;
fifo_oreg[wraddr] <= { i_oreg, i_op[2:1], i_addr[1:0] };
 
always @(posedge i_clk)
if ((i_rst)||(i_wb_err))
wraddr <= 0;
else if (i_pipe_stb)
wraddr <= wraddr + 4'h1;
wraddr <= wraddr + 1'b1;
always @(posedge i_clk)
if ((i_rst)||(i_wb_err))
rdaddr <= 0;
else if ((i_wb_ack)&&(cyc))
rdaddr <= rdaddr + 4'h1;
assign nxt_rdaddr = rdaddr + 4'h1;
rdaddr <= rdaddr + 1'b1;
assign nxt_rdaddr = rdaddr + 1'b1;
 
wire gbl_stb, lcl_stb;
assign lcl_stb = (i_addr[31:8]==24'hc00000)&&(i_addr[7:5]==3'h0);
assign lcl_stb = (i_addr[31:24]==8'hff);
assign gbl_stb = (~lcl_stb);
//= ((i_addr[31:8]!=24'hc00000)||(i_addr[7:5]!=3'h0));
 
135,19 → 145,36
// o_wb_we <= i_op
end
always @(posedge i_clk)
if ((cyc)&&(i_pipe_stb)&&(~i_wb_stall))
if ((!cyc)||(!i_wb_stall))
begin
o_wb_addr <= i_addr[(AW-1):0];
o_wb_data <= i_data;
end else if ((~cyc)&&(i_pipe_stb))
begin
o_wb_addr <= i_addr[(AW-1):0];
o_wb_data <= i_data;
o_wb_addr <= i_addr[(AW+1):2];
if (!i_op[0]) // Always select everything on reads
o_wb_sel <= 4'b1111; // Op is even
else casez({ i_op[2:1], i_addr[1:0] })
4'b100?: o_wb_sel <= 4'b1100; // Op = 5
4'b101?: o_wb_sel <= 4'b0011; // Op = 5
4'b1100: o_wb_sel <= 4'b1000; // Op = 5
4'b1101: o_wb_sel <= 4'b0100; // Op = 7
4'b1110: o_wb_sel <= 4'b0010; // Op = 7
4'b1111: o_wb_sel <= 4'b0001; // Op = 7
default: o_wb_sel <= 4'b1111; // Op = 7
endcase
 
casez({ i_op[2:1], i_addr[1:0] })
4'b100?: o_wb_data <= { i_data[15:0], 16'h00 };
4'b101?: o_wb_data <= { 16'h00, i_data[15:0] };
4'b1100: o_wb_data <= { i_data[7:0], 24'h00 };
4'b1101: o_wb_data <= { 8'h00, i_data[7:0], 16'h00 };
4'b1110: o_wb_data <= { 16'h00, i_data[7:0], 8'h00 };
4'b1111: o_wb_data <= { 24'h00, i_data[7:0] };
default: o_wb_data <= i_data;
endcase
 
end
 
always @(posedge i_clk)
if ((i_pipe_stb)&&(~cyc))
o_wb_we <= i_op;
o_wb_we <= i_op[0];
 
initial o_valid = 1'b0;
always @(posedge i_clk)
157,12 → 184,20
o_err <= (cyc)&&(i_wb_err);
assign o_busy = cyc;
 
wire [8:0] w_wreg;
assign w_wreg = fifo_oreg[rdaddr];
always @(posedge i_clk)
o_wreg <= fifo_oreg[rdaddr];
o_wreg <= w_wreg[8:4];
always @(posedge i_clk)
// if (i_wb_ack) isn't necessary, since o_valid won't be true
// then either.
o_result <= i_wb_data;
casez(w_wreg[3:0])
4'b1100: o_result = { 24'h00, i_wb_data[31:24] };
4'b1101: o_result = { 24'h00, i_wb_data[23:16] };
4'b1110: o_result = { 24'h00, i_wb_data[15: 8] };
4'b1111: o_result = { 24'h00, i_wb_data[ 7: 0] };
4'b100?: o_result = { 16'h00, i_wb_data[31:16] };
4'b101?: o_result = { 16'h00, i_wb_data[15: 0] };
default: o_result = i_wb_data[31:0];
endcase
 
assign o_pipe_stalled = (cyc)
&&((i_wb_stall)||((~o_wb_stb_lcl)&&(~o_wb_stb_gbl)));
177,7 → 212,7
always @(posedge i_clk)
begin
lock_gbl <= (i_lock)&&((r_wb_cyc_gbl)||(lock_gbl));
lock_lcl <= (i_lock)&&((r_wb_cyc_lcl)||(lock_gbl));
lock_lcl <= (i_lock)&&((r_wb_cyc_lcl)||(lock_lcl));
end
 
assign o_wb_cyc_gbl = (r_wb_cyc_gbl)||(lock_gbl);
/prefetch.v
24,7 → 24,7
//
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015, Gisselquist Technology, LLC
// Copyright (C) 2015,2017, 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
36,6 → 36,11
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
42,6 → 47,7
//
////////////////////////////////////////////////////////////////////////////////
//
//
// Flash requires a minimum of 4 clocks per byte to read, so that would be
// 4*(4bytes/32bit word) = 16 clocks per word read---and that's in pipeline
// mode which this prefetch does not support. In non--pipelined mode, the
114,7 → 120,7
if ((o_wb_cyc)&&(i_wb_ack))
begin
o_valid <= (i_pc == o_wb_addr)&&(~i_wb_err);
o_illegal <= i_wb_err;
o_illegal <= (i_wb_err)&&(i_pc == o_wb_addr);
end else if (i_stalled_n)
begin
o_valid <= 1'b0;
/zipcpu.v
1,4 → 1,4
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Filename: zipcpu.v
//
18,7 → 18,7
//
// The Zip CPU is fully pipelined with the following pipeline stages:
//
// 1. Prefetch, returns the instruction from memory.
// 1. Prefetch, returns the instruction from memory.
//
// 2. Instruction Decode
//
72,9 → 72,9
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015-2016, Gisselquist Technology, LLC
// Copyright (C) 2015-2017, 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
86,28 → 86,23
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. (It's in the $(ROOT)/doc directory. Run make with no
// target there if the PDF file isn't present.) If not, see
// <http://www.gnu.org/licenses/> for a copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
///////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
//
// We can either pipeline our fetches, or issue one fetch at a time. Pipelined
// fetches are more complicated and therefore use more FPGA resources, while
// single fetches will cause the CPU to stall for about 5 stalls each
// instruction cycle, effectively reducing the instruction count per clock to
// about 0.2. However, the area cost may be worth it. Consider:
//
// Slice LUTs ZipSystem ZipCPU
// Single Fetching 2521 1734
// Pipelined fetching 2796 2046
//
//
//
`define CPU_CC_REG 4'he
`define CPU_PC_REG 4'hf
`define CPU_CLRCACHE_BIT 14 // Set to clear the I-cache, automatically clears
`define CPU_PHASE_BIT 13 // Set if we are executing the latter half of a VLIW
`define CPU_PHASE_BIT 13 // Set if we are executing the latter half of a CIS
`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
114,7 → 109,7
`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_STEP_BIT 6 // Will step one (or two CIS) instructions
`define CPU_GIE_BIT 5
`define CPU_SLEEP_BIT 4
// Compile time defines
130,7 → 125,7
// 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,
o_wb_we, o_wb_addr, o_wb_data, o_wb_sel,
i_wb_ack, i_wb_stall, i_wb_data,
i_wb_err,
// Accounting/CPU usage interface
139,7 → 134,8
, o_debug
`endif
);
parameter RESET_ADDRESS=32'h0100000, ADDRESS_WIDTH=32,
parameter [31:0] RESET_ADDRESS=32'h0100000;
parameter ADDRESS_WIDTH=30,
LGICACHE=8;
`ifdef OPT_MULTIPLY
parameter IMPLEMENT_MPY = `OPT_MULTIPLY;
162,7 → 158,9
`else
parameter EARLY_BRANCHING = 0;
`endif
parameter WITH_LOCAL_BUS = 1;
localparam AW=ADDRESS_WIDTH;
localparam [(AW-1):0] RESET_BUS_ADDRESS = RESET_ADDRESS[(AW+1):2];
input i_clk, i_rst, i_interrupt;
// Debug interface -- inputs
input i_halt, i_clear_pf_cache;
179,6 → 177,7
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;
output wire [3:0] o_wb_sel;
// Wishbone interface -- inputs
input i_wb_ack, i_wb_stall;
input [31:0] i_wb_data;
203,22 → 202,22
// that logic.
//
(* ram_style = "distributed" *)
`ifdef OPT_NO_USERMODE
reg [31:0] regset [0:15];
`else
reg [31:0] regset [0:31];
`endif
 
// Condition codes
// (BUS, TRAP,ILL,BREAKEN,STEP,GIE,SLEEP ), V, N, C, Z
reg [3:0] flags, iflags;
wire [14:0] w_uflags, w_iflags;
reg trap, break_en, step, gie, sleep, r_halted;
wire break_pending;
wire w_clear_icache;
`ifdef OPT_ILLEGAL_INSTRUCTION
reg ill_err_u, ill_err_i;
`else
wire ill_err_u, ill_err_i;
`endif
reg ubreak;
reg ibus_err_flag, ubus_err_flag;
reg break_en, step, sleep, r_halted;
wire break_pending, trap, gie, ubreak;
wire w_clear_icache, ill_err_u;
reg ill_err_i;
reg ibus_err_flag;
wire ubus_err_flag;
wire idiv_err_flag, udiv_err_flag;
wire ifpu_err_flag, ufpu_err_flag;
wire ihalt_phase, uhalt_phase;
231,7 → 230,7
// PIPELINE STAGE #1 :: Prefetch
// Variable declarations
//
reg [(AW-1):0] pf_pc;
reg [(AW+1):0] pf_pc;
reg new_pc;
wire clear_pipeline;
assign clear_pipeline = new_pc;
240,9 → 239,9
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;
wire [31:0] pf_instruction;
wire [(AW-1):0] pf_instruction_pc;
wire pf_valid, pf_gie, pf_illegal;
 
//
//
250,29 → 249,33
// Variable declarations
//
//
reg opvalid, opvalid_mem, opvalid_alu;
reg opvalid_div, opvalid_fpu;
reg op_valid /* verilator public_flat */,
op_valid_mem, op_valid_alu;
reg op_valid_div, op_valid_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,
wire [3:0] dcd_opn;
wire [4:0] dcd_A, dcd_B, dcd_R;
wire dcd_Acc, dcd_Bcc, dcd_Apc, dcd_Bpc, dcd_Rcc, dcd_Rpc;
wire [3:0] dcd_F;
wire dcd_wR, dcd_rA, dcd_rB,
dcd_ALU, dcd_M, dcd_DIV, dcd_FP,
dcd_wF, dcd_gie, dcd_break, dcd_lock,
dcd_pipe, dcd_ljmp;
reg r_dcdvalid;
wire dcdvalid;
wire [(AW-1):0] dcd_pc;
wire [31:0] dcdI;
wire dcd_zI; // true if dcdI == 0
wire dcdA_stall, dcdB_stall, dcdF_stall;
reg r_dcd_valid;
wire dcd_valid;
wire [AW:0] dcd_pc /* verilator public_flat */;
wire [31:0] dcd_I;
wire dcd_zI; // true if dcd_I == 0
wire dcd_A_stall, dcd_B_stall, dcd_F_stall;
 
wire dcd_illegal;
wire dcd_early_branch;
wire [(AW-1):0] dcd_branch_pc;
 
wire dcd_sim;
wire [22:0] dcd_sim_immv;
 
 
//
//
// PIPELINE STAGE #3 :: Read Operands
282,33 → 285,33
//
// Now, let's read our operands
reg [4:0] alu_reg;
wire [3:0] opn;
wire [4:0] opR;
reg [31:0] r_opA, r_opB;
wire [3:0] op_opn;
wire [4:0] op_R;
reg [31:0] r_op_Av, r_op_Bv;
reg [(AW-1):0] op_pc;
wire [31:0] w_opA, w_opB;
wire [31:0] opA_nowait, opB_nowait, opA, opB;
reg opR_wr, opF_wr;
wire op_gie, opR_cc;
wire [14:0] opFl;
reg [5:0] r_opF;
wire [7:0] opF;
wire [31:0] w_op_Av, w_op_Bv;
wire [31:0] op_A_nowait, op_B_nowait, op_Av, op_Bv;
reg op_wR, op_wF;
wire op_gie, op_Rcc;
wire [14:0] op_Fl;
reg [6:0] r_op_F;
wire [7:0] op_F;
wire op_ce, op_phase, op_pipe, op_change_data_ce;
// Some pipeline control wires
`ifdef OPT_PIPELINED
reg opA_alu, opA_mem;
reg opB_alu, opB_mem;
reg op_A_alu, op_A_mem;
reg op_B_alu, op_B_mem;
`endif
`ifdef OPT_ILLEGAL_INSTRUCTION
reg op_illegal;
`else
wire op_illegal;
assign op_illegal = 1'b0;
`endif
wire op_break;
wire op_lock;
 
`ifdef VERILATOR
reg op_sim /* verilator public_flat */;
reg [22:0] op_sim_immv /* verilator public_flat */;
`endif
 
 
//
//
// PIPELINE STAGE #4 :: ALU / Memory
319,20 → 322,18
reg r_alu_pc_valid, mem_pc_valid;
wire alu_pc_valid;
wire alu_phase;
wire alu_ce, alu_stall;
wire alu_ce /* verilator public_flat */, 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;
reg alu_wR, alu_wF;
wire alu_gie, 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;
340,12 → 341,13
wire mem_busy, mem_rdbusy;
wire [(AW-1):0] mem_addr;
wire [31:0] mem_data, mem_result;
wire [3:0] mem_sel;
 
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)
assign div_ce = (master_ce)&&(~clear_pipeline)&&(op_valid_div)
&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
&&(set_cond);
 
353,7 → 355,7
wire [31:0] fpu_result;
wire [3:0] fpu_flags;
 
assign fpu_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_fpu)
assign fpu_ce = (master_ce)&&(~clear_pipeline)&&(op_valid_fpu)
&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
&&(set_cond);
 
369,7 → 371,8
wire [4:0] wr_reg_id;
wire [31:0] wr_gpreg_vl, wr_spreg_vl;
wire w_switch_to_interrupt, w_release_from_interrupt;
reg [(AW-1):0] upc, ipc;
reg [(AW+1):0] ipc;
wire [(AW+1):0] upc;
 
 
 
376,7 → 379,7
//
// MASTER: clock enable.
//
assign master_ce = (~i_halt)&&(~o_break)&&(~sleep);
assign master_ce = ((~i_halt)||(alu_phase))&&(~o_break)&&(~sleep);
 
 
//
389,13 → 392,13
//
// PIPELINE STAGE #2 :: Instruction Decode
// Calculate stall conditions
assign dcd_ce = ((~dcdvalid)||(~dcd_stalled))&&(~clear_pipeline);
assign dcd_ce = ((~dcd_valid)||(~dcd_stalled))&&(~clear_pipeline);
 
`ifdef OPT_PIPELINED
assign dcd_stalled = (dcdvalid)&&(op_stall);
assign dcd_stalled = (dcd_valid)&&(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
// If not pipelined, there will be no op_valid_ anything, and the
// op_stall will be false, dcd_X_stall will be false, thus we can simply
// do a ...
assign dcd_stalled = 1'b0;
`endif
402,17 → 405,17
//
// PIPELINE STAGE #3 :: Read Operands
// Calculate stall conditions
wire op_lock_stall;
wire prelock_stall;
`ifdef OPT_PIPELINED
reg cc_invalid_for_dcd;
always @(posedge i_clk)
cc_invalid_for_dcd <= (wr_flags_ce)
||(wr_reg_ce)&&(wr_reg_id[3:0] == `CPU_CC_REG)
||(opvalid)&&((opF_wr)||((opR_wr)&&(opR[3:0] == `CPU_CC_REG)))
||((alF_wr)||((alu_wr)&&(alu_reg[3:0] == `CPU_CC_REG)))
||(op_valid)&&((op_wF)||((op_wR)&&(op_R[3:0] == `CPU_CC_REG)))
||((alu_wF)||((alu_wR)&&(alu_reg[3:0] == `CPU_CC_REG)))
||(mem_busy)||(div_busy)||(fpu_busy);
 
assign op_stall = (opvalid)&&( // Only stall if we're loaded w/validins
assign op_stall = (op_valid)&&( // 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
425,27 → 428,29
// This also includes whether or not the divide or
// floating point units are busy.
(alu_stall)
||(((op_valid_div)||(op_valid_fpu))
&&(!adf_ce_unconditional))
//
// Stall if we are going into memory with an operation
// that cannot be pipelined, and the memory is
// already busy
||(mem_stalled) // &&(opvalid_mem) part of mem_stalled
||(opR_cc)
||(mem_stalled) // &&(op_valid_mem) part of mem_stalled
||(op_Rcc)
)
||(dcdvalid)&&(
||(dcd_valid)&&(
// Stall if we need to wait for an operand A
// to be ready to read
(dcdA_stall)
(dcd_A_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)
||(dcd_B_stall)
// Or if we need to wait on flags to work on the
// CC register
||(dcdF_stall)
||(dcd_F_stall)
);
assign op_ce = ((dcdvalid)||(dcd_illegal))&&(~op_stall)&&(~clear_pipeline);
assign op_ce = ((dcd_valid)||(dcd_illegal)||(dcd_early_branch))&&(!op_stall);
 
 
// BUT ... op_ce is too complex for many of the data operations. So
452,11 → 457,11
// let's make their circuit enable code simpler. In particular, if
// op_ doesn't need to be preserved, we can change it all we want
// ... right? The clear_pipeline code, for example, really only needs
// to determine whether opvalid is true.
// to determine whether op_valid is true.
assign op_change_data_ce = (~op_stall);
`else
assign op_stall = (opvalid)&&(~master_ce);
assign op_ce = ((dcdvalid)||(dcd_illegal))&&(~clear_pipeline);
assign op_stall = (op_valid)&&(~master_ce);
assign op_ce = ((dcd_valid)||(dcd_illegal)||(dcd_early_branch))&&(~clear_pipeline);
assign op_change_data_ce = 1'b1;
`endif
 
465,7 → 470,7
// Calculate stall conditions
//
// 1. Basic stall is if the previous stage is valid and the next is
// busy.
// 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,
474,16 → 479,16
// through the ALU. Break instructions are not allowed through
// the ALU.
`ifdef OPT_PIPELINED
assign alu_stall = (((~master_ce)||(mem_rdbusy)||(alu_busy))&&(opvalid_alu)) //Case 1&2
||((opvalid)&&(op_lock)&&(op_lock_stall))
||((opvalid)&&(op_break)) // || op_illegal
assign alu_stall = (((~master_ce)||(mem_rdbusy)||(alu_busy))&&(op_valid_alu)) //Case 1&2
||(prelock_stall)
||((op_valid)&&(op_break))
||(wr_reg_ce)&&(wr_write_cc)
||(div_busy)||(fpu_busy);
assign alu_ce = (master_ce)&&(opvalid_alu)&&(~alu_stall)
assign alu_ce = (master_ce)&&(op_valid_alu)&&(~alu_stall)
&&(~clear_pipeline);
`else
assign alu_stall = (opvalid_alu)&&((~master_ce)||(op_break));
assign alu_ce = (master_ce)&&(opvalid_alu)&&(~alu_stall)&&(~clear_pipeline);
assign alu_stall = (op_valid_alu)&&((~master_ce)||(op_break));
assign alu_ce = (master_ce)&&(op_valid_alu)&&(~alu_stall)&&(~clear_pipeline);
`endif
//
 
492,7 → 497,7
// alu_pc_valid.
//
`ifdef OPT_PIPELINED
assign mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)
assign mem_ce = (master_ce)&&(op_valid_mem)&&(~mem_stalled)
&&(~clear_pipeline);
`else
// If we aren't pipelined, then no one will be changing what's in the
503,12 → 508,13
// something gets in the pipeline and then (due to interrupt or some
// such) needs to be voided? Thus we avoid simplification and keep
// what worked here.
assign mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)
assign mem_ce = (master_ce)&&(op_valid_mem)&&(~mem_stalled)
&&(~clear_pipeline);
`endif
`ifdef OPT_PIPELINED_BUS_ACCESS
assign mem_stalled = (~master_ce)||(alu_busy)||((opvalid_mem)&&(
assign mem_stalled = (~master_ce)||(alu_busy)||((op_valid_mem)&&(
(mem_pipe_stalled)
||(prelock_stall)
||((~op_pipe)&&(mem_busy))
||(div_busy)
||(fpu_busy)
520,7 → 526,7
&&((wr_write_pc)||(wr_write_cc)))));
`else
`ifdef OPT_PIPELINED
assign mem_stalled = (mem_busy)||((opvalid_mem)&&(
assign mem_stalled = (mem_busy)||((op_valid_mem)&&(
(~master_ce)
// Stall waiting for flags to be valid
// Or waiting for a write to the PC register
528,14 → 534,14
// 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);
assign mem_stalled = (op_valid_mem)&&(~master_ce);
`endif
`endif
 
// ALU, DIV, or FPU CE ... equivalent to the OR of all three of these
assign adf_ce_unconditional = (master_ce)&&(~clear_pipeline)&&(opvalid)
&&(~opvalid_mem)&&(~mem_rdbusy)
&&((~opvalid_alu)||(~alu_stall))&&(~op_break)
assign adf_ce_unconditional = (master_ce)&&(~clear_pipeline)&&(op_valid)
&&(~op_valid_mem)&&(~mem_rdbusy)
&&((~op_valid_alu)||(~alu_stall))&&(~op_break)
&&(~div_busy)&&(~fpu_busy)&&(~clear_pipeline);
 
//
546,110 → 552,94
`ifdef OPT_SINGLE_FETCH
wire pf_ce;
 
assign pf_ce = (~pf_valid)&&(~dcdvalid)&&(~opvalid)&&(~alu_busy)&&(~mem_busy)&&(~alu_pc_valid)&&(~mem_pc_valid);
assign pf_ce = (~pf_valid)&&(~dcd_valid)&&(~op_valid)&&(~alu_busy)&&(~mem_busy)&&(~alu_pc_valid)&&(~mem_pc_valid);
prefetch #(ADDRESS_WIDTH)
pf(i_clk, (i_rst), (pf_ce), (~dcd_stalled), pf_pc, gie,
instruction, instruction_pc, instruction_gie,
pf(i_clk, (i_rst), (pf_ce), (~dcd_stalled), pf_pc[(AW+1):2], gie,
pf_instruction, pf_instruction_pc, pf_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;
initial r_dcd_valid = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
r_dcdvalid <= 1'b0;
if (clear_pipeline)
r_dcd_valid <= 1'b0;
else if (dcd_ce)
r_dcdvalid <= (pf_valid)||(pf_illegal);
r_dcd_valid <= (pf_valid)||(pf_illegal);
else if (op_ce)
r_dcdvalid <= 1'b0;
assign dcdvalid = r_dcdvalid;
r_dcd_valid <= 1'b0;
assign dcd_valid = r_dcd_valid;
 
`else // Pipe fetch
 
wire pf_stalled;
assign pf_stalled = (dcd_stalled)||(dcd_phase);
`ifdef OPT_TRADITIONAL_PFCACHE
wire [(AW-1):0] pf_request_address;
assign pf_request_address = ((dcd_early_branch)&&(!clear_pipeline))
? dcd_branch_pc:pf_pc[(AW+1):2];
pfcache #(LGICACHE, ADDRESS_WIDTH)
pf(i_clk, i_rst, (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
w_clear_icache,
// dcd_pc,
~dcd_stalled,
((dcd_early_branch)&&(~clear_pipeline))
? dcd_branch_pc:pf_pc,
instruction, instruction_pc, pf_valid,
(!pf_stalled),
pf_request_address,
pf_instruction, pf_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)
pipefetch #(RESET_BUS_ADDRESS, LGICACHE, ADDRESS_WIDTH)
pf(i_clk, i_rst, (new_pc)||(dcd_early_branch),
w_clear_icache, ~dcd_stalled,
(new_pc)?pf_pc:dcd_branch_pc,
instruction, instruction_pc, pf_valid,
w_clear_icache, (!pf_stalled),
(new_pc)?pf_pc[(AW+1):2]:dcd_branch_pc,
pf_instruction, pf_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;
`ifdef OPT_NO_USERMODE
assign pf_gie = 1'b0;
`else
assign pf_gie = gie;
`endif
 
initial r_dcdvalid = 1'b0;
initial r_dcd_valid = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline)||(w_clear_icache))
r_dcdvalid <= 1'b0;
if ((clear_pipeline)||(w_clear_icache))
r_dcd_valid <= 1'b0;
else if (dcd_ce)
r_dcdvalid <= (pf_valid)&&(~dcd_ljmp)&&(~dcd_early_branch);
r_dcd_valid <= ((dcd_phase)||(pf_valid))
&&(~dcd_ljmp)&&(~dcd_early_branch);
else if (op_ce)
r_dcdvalid <= 1'b0;
assign dcdvalid = r_dcdvalid;
r_dcd_valid <= 1'b0;
assign dcd_valid = r_dcd_valid;
`endif
 
`ifdef OPT_NEW_INSTRUCTION_SET
 
// If not pipelined, there will be no opvalid_ anything, and the
// If not pipelined, there will be no op_valid_ anything, and the
idecode #(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
IMPLEMENT_FPU)
instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
(~dcdvalid)||(~op_stall), dcd_stalled, instruction, instruction_gie,
instruction_pc, pf_valid, pf_illegal, dcd_phase,
dcd_illegal, dcd_pc, dcd_gie,
{ dcdR_cc, dcdR_pc, dcdR },
{ dcdA_cc, dcdA_pc, dcdA },
{ dcdB_cc, dcdB_pc, dcdB },
dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
dcdR_wr,dcdA_rd, dcdB_rd,
instruction_decoder(i_clk, (clear_pipeline),
(~dcd_valid)||(~op_stall), dcd_stalled, pf_instruction, pf_gie,
pf_instruction_pc, pf_valid, pf_illegal, dcd_phase,
dcd_illegal, dcd_pc, dcd_gie,
{ dcd_Rcc, dcd_Rpc, dcd_R },
{ dcd_Acc, dcd_Apc, dcd_A },
{ dcd_Bcc, dcd_Bpc, dcd_B },
dcd_I, dcd_zI, dcd_F, dcd_wF, dcd_opn,
dcd_ALU, dcd_M, dcd_DIV, dcd_FP, dcd_break, dcd_lock,
dcd_wR,dcd_rA, dcd_rB,
dcd_early_branch,
dcd_branch_pc, dcd_ljmp,
dcd_pipe);
`else
idecode_deprecated
#(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
IMPLEMENT_FPU)
instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
dcd_ce, dcd_stalled, instruction, instruction_gie,
instruction_pc, pf_valid, pf_illegal, dcd_phase,
dcd_illegal, dcd_pc, dcd_gie,
{ dcdR_cc, dcdR_pc, dcdR },
{ dcdA_cc, dcdA_pc, dcdA },
{ dcdB_cc, dcdB_pc, dcdB },
dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
dcdR_wr,dcdA_rd, dcdB_rd,
dcd_early_branch,
dcd_branch_pc,
dcd_pipe);
assign dcd_ljmp = 1'b0;
`endif
dcd_pipe,
dcd_sim, dcd_sim_immv);
 
`ifdef OPT_PIPELINED_BUS_ACCESS
reg r_op_pipe;
 
initial r_op_pipe = 1'b0;
// To be a pipeable operation, there must be
// 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
659,7 → 649,9
// However ... we need to know this before this clock, hence this is
// calculated in the instruction decoder.
always @(posedge i_clk)
if (op_ce)
if (clear_pipeline)
r_op_pipe <= 1'b0;
else if (op_ce)
r_op_pipe <= dcd_pipe;
else if (mem_ce) // Clear us any time an op_ is clocked in
r_op_pipe <= 1'b0;
673,31 → 665,20
// PIPELINE STAGE #3 :: Read Operands (Registers)
//
//
assign w_opA = regset[dcdA];
assign w_opB = regset[dcdB];
`ifdef OPT_NO_USERMODE
assign w_op_Av = regset[dcd_A[3:0]];
assign w_op_Bv = regset[dcd_B[3:0]];
`else
assign w_op_Av = regset[dcd_A];
assign w_op_Bv = regset[dcd_B];
`endif
 
wire [8:0] w_cpu_info;
assign w_cpu_info = {
`ifdef OPT_ILLEGAL_INSTRUCTION
1'b1,
`else
1'b0,
`endif
`ifdef OPT_MULTIPLY
1'b1,
`else
1'b0,
`endif
`ifdef OPT_DIVIDE
1'b1,
`else
1'b0,
`endif
`ifdef OPT_IMPLEMENT_FPU
1'b1,
`else
1'b0,
`endif
(IMPLEMENT_MPY >0)? 1'b1:1'b0,
(IMPLEMENT_DIVIDE >0)? 1'b1:1'b0,
(IMPLEMENT_FPU >0)? 1'b1:1'b0,
`ifdef OPT_PIPELINED
1'b1,
`else
718,7 → 699,7
`else
1'b0,
`endif
`ifdef OPT_VLIW
`ifdef OPT_CIS
1'b1
`else
1'b0
726,79 → 707,85
};
 
wire [31:0] w_pcA_v;
assign w_pcA_v[(AW+1):0] = { (dcd_A[4] == dcd_gie)
? { dcd_pc[AW:1], 2'b00 }
: { upc[(AW+1):2], uhalt_phase, 1'b0 } };
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;
if (AW < 30)
assign w_pcA_v[31:(AW+2)] = 0;
endgenerate
 
`ifdef OPT_PIPELINED
reg [4:0] opA_id, opB_id;
reg opA_rd, opB_rd;
reg [4:0] op_Aid, op_Bid;
reg op_rA, op_rB;
always @(posedge i_clk)
if (op_ce)
begin
opA_id <= dcdA;
opB_id <= dcdB;
opA_rd <= dcdA_rd;
opB_rd <= dcdB_rd;
op_Aid <= dcd_A;
op_Bid <= dcd_B;
op_rA <= dcd_rA;
op_rB <= dcd_rB;
end
`endif
 
always @(posedge i_clk)
`ifdef OPT_PIPELINED
if (op_change_data_ce)
if (op_ce)
`endif
begin
`ifdef OPT_PIPELINED
if ((wr_reg_ce)&&(wr_reg_id == dcdA))
r_opA <= wr_gpreg_vl;
if ((wr_reg_ce)&&(wr_reg_id == dcd_A))
r_op_Av <= wr_gpreg_vl;
else
`endif
if (dcdA_pc)
r_opA <= w_pcA_v;
else if (dcdA_cc)
r_opA <= { w_cpu_info, w_opA[22:16], 1'b0, (dcdA[4])?w_uflags:w_iflags };
if (dcd_Apc)
r_op_Av <= w_pcA_v;
else if (dcd_Acc)
r_op_Av <= { w_cpu_info, w_op_Av[22:16], 1'b0, (dcd_A[4])?w_uflags:w_iflags };
else
r_opA <= w_opA;
r_op_Av <= w_op_Av;
`ifdef OPT_PIPELINED
end else
begin // We were going to pick these up when they became valid,
// but for some reason we're stuck here as they became
// valid. Pick them up now anyway
// if (((opA_alu)&&(alu_wr))||((opA_mem)&&(mem_valid)))
// r_opA <= wr_gpreg_vl;
if ((wr_reg_ce)&&(wr_reg_id == opA_id)&&(opA_rd))
r_opA <= wr_gpreg_vl;
// if (((op_A_alu)&&(alu_wR))||((op_A_mem)&&(mem_valid)))
// r_op_Av <= wr_gpreg_vl;
if ((wr_reg_ce)&&(wr_reg_id == op_Aid)&&(op_rA))
r_op_Av <= wr_gpreg_vl;
`endif
end
 
wire [31:0] w_opBnI, w_pcB_v;
wire [31:0] w_op_BnI, w_pcB_v;
assign w_pcB_v[(AW+1):0] = { (dcd_B[4] == dcd_gie)
? { dcd_pc[AW:1], 2'b00 }
: { upc[(AW+1):2], uhalt_phase, 1'b0 } };
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;
if (AW < 30)
assign w_pcB_v[31:(AW+2)] = 0;
endgenerate
 
assign w_opBnI = (~dcdB_rd) ? 32'h00
assign w_op_BnI = (!dcd_rB) ? 32'h00
`ifdef OPT_PIPELINED
: ((wr_reg_ce)&&(wr_reg_id == dcdB)) ? wr_gpreg_vl
: ((wr_reg_ce)&&(wr_reg_id == dcd_B)) ? wr_gpreg_vl
`endif
: ((dcdB_pc) ? w_pcB_v
: ((dcdB_cc) ? { w_cpu_info, w_opB[22:16], // w_opB[31:14],
1'b0, (dcdB[4])?w_uflags:w_iflags}
: w_opB));
: ((dcd_Bcc) ? { w_cpu_info, w_op_Bv[22:16], // w_op_B[31:14],
1'b0, (dcd_B[4])?w_uflags:w_iflags}
: w_op_Bv);
 
always @(posedge i_clk)
`ifdef OPT_PIPELINED
if (op_change_data_ce)
r_opB <= w_opBnI + dcdI;
else if ((wr_reg_ce)&&(opB_id == wr_reg_id)&&(opB_rd))
r_opB <= wr_gpreg_vl;
if ((op_ce)&&(dcd_Bpc)&&(dcd_rB))
r_op_Bv <= w_pcB_v + { dcd_I[29:0], 2'b00 };
else if (op_ce)
r_op_Bv <= w_op_BnI + dcd_I;
else if ((wr_reg_ce)&&(op_Bid == wr_reg_id)&&(op_rB))
r_op_Bv <= wr_gpreg_vl;
`else
r_opB <= w_opBnI + dcdI;
if ((dcd_Bpc)&&(dcd_rB))
r_op_Bv <= w_pcB_v + { dcd_I[29:0], 2'b00 };
else
r_op_Bv <= w_op_BnI + dcd_I;
`endif
 
// The logic here has become more complex than it should be, no thanks
809,54 → 796,42
// 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.
// op_F.
always @(posedge i_clk)
`ifdef OPT_PIPELINED
if (op_ce) // Cannot do op_change_data_ce here since opF depends
if (op_ce) // Cannot do op_change_data_ce here since op_F depends
// upon being either correct for a valid op, or correct
// for the last valid op
`endif
begin // Set the flag condition codes, bit order is [3:0]=VNCZ
case(dcdF[2:0])
3'h0: r_opF <= 6'h00; // Always
`ifdef OPT_NEW_INSTRUCTION_SET
// These were remapped as part of the new instruction
// set in order to make certain that the low order
// two bits contained the most commonly used
// conditions: Always, LT, Z, and NZ.
3'h1: r_opF <= 6'h24; // LT
3'h2: r_opF <= 6'h11; // Z
3'h3: r_opF <= 6'h10; // NE
3'h4: r_opF <= 6'h30; // GT (!N&!Z)
3'h5: r_opF <= 6'h20; // GE (!N)
`else
3'h1: r_opF <= 6'h11; // Z
3'h2: r_opF <= 6'h10; // NE
3'h3: r_opF <= 6'h20; // GE (!N)
3'h4: r_opF <= 6'h30; // GT (!N&!Z)
3'h5: r_opF <= 6'h24; // LT
`endif
3'h6: r_opF <= 6'h02; // C
3'h7: r_opF <= 6'h08; // V
case(dcd_F[2:0])
3'h0: r_op_F <= 7'h00; // Always
3'h1: r_op_F <= 7'h11; // Z
3'h2: r_op_F <= 7'h44; // LT
3'h3: r_op_F <= 7'h22; // C
3'h4: r_op_F <= 7'h08; // V
3'h5: r_op_F <= 7'h10; // NE
3'h6: r_op_F <= 7'h40; // GE (!N)
3'h7: r_op_F <= 7'h20; // NC
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] };
assign op_F = { r_op_F[3], r_op_F[6:0] };
 
wire w_opvalid;
assign w_opvalid = (~clear_pipeline)&&(dcdvalid)&&(~dcd_ljmp);
initial opvalid = 1'b0;
initial opvalid_alu = 1'b0;
initial opvalid_mem = 1'b0;
initial opvalid_div = 1'b0;
initial opvalid_fpu = 1'b0;
wire w_op_valid;
assign w_op_valid = (~clear_pipeline)&&(dcd_valid)&&(~dcd_ljmp)&&(!dcd_early_branch);
initial op_valid = 1'b0;
initial op_valid_alu = 1'b0;
initial op_valid_mem = 1'b0;
initial op_valid_div = 1'b0;
initial op_valid_fpu = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
if (clear_pipeline)
begin
opvalid <= 1'b0;
opvalid_alu <= 1'b0;
opvalid_mem <= 1'b0;
opvalid_div <= 1'b0;
opvalid_fpu <= 1'b0;
op_valid <= 1'b0;
op_valid_alu <= 1'b0;
op_valid_mem <= 1'b0;
op_valid_div <= 1'b0;
op_valid_fpu <= 1'b0;
end else if (op_ce)
begin
// Do we have a valid instruction?
867,25 → 842,19
// Hence, the test on dcd_stalled here. If we must
// wait until our operands are valid, then we aren't
// valid yet until then.
opvalid<= (w_opvalid)||(dcd_illegal)&&(dcdvalid);
`ifdef OPT_ILLEGAL_INSTRUCTION
opvalid_alu <= (w_opvalid)&&((dcdALU)||(dcd_illegal));
opvalid_mem <= (dcdM)&&(~dcd_illegal)&&(w_opvalid);
opvalid_div <= (dcdDV)&&(~dcd_illegal)&&(w_opvalid);
opvalid_fpu <= (dcdFP)&&(~dcd_illegal)&&(w_opvalid);
`else
opvalid_alu <= (dcdALU)&&(w_opvalid);
opvalid_mem <= (dcdM)&&(w_opvalid);
opvalid_div <= (dcdDV)&&(w_opvalid);
opvalid_fpu <= (dcdFP)&&(w_opvalid);
`endif
op_valid<= (w_op_valid)||(dcd_illegal)&&(dcd_valid)||(dcd_early_branch);
op_valid_alu <= (w_op_valid)&&((dcd_ALU)||(dcd_illegal)
||(dcd_early_branch));
op_valid_mem <= (dcd_M)&&(~dcd_illegal)&&(w_op_valid);
op_valid_div <= (dcd_DIV)&&(~dcd_illegal)&&(w_op_valid);
op_valid_fpu <= (dcd_FP)&&(~dcd_illegal)&&(w_op_valid);
end else if ((adf_ce_unconditional)||(mem_ce))
begin
opvalid <= 1'b0;
opvalid_alu <= 1'b0;
opvalid_mem <= 1'b0;
opvalid_div <= 1'b0;
opvalid_fpu <= 1'b0;
op_valid <= 1'b0;
op_valid_alu <= 1'b0;
op_valid_mem <= 1'b0;
op_valid_div <= 1'b0;
op_valid_fpu <= 1'b0;
end
 
// Here's part of our debug interface. When we recognize a break
896,16 → 865,17
// 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);
// assign w_op_break = (dcd_break)&&(r_dcd_I[15:0] == 16'h0001);
`ifdef OPT_PIPELINED
reg r_op_break;
 
initial r_op_break = 1'b0;
always @(posedge i_clk)
if (i_rst) r_op_break <= 1'b0;
else if (op_ce) r_op_break <= (dcd_break); //||dcd_illegal &&(dcdvalid)
else if ((clear_pipeline)||(~opvalid))
r_op_break <= 1'b0;
if ((i_rst)||(clear_pipeline)) r_op_break <= 1'b0;
else if (op_ce)
r_op_break <= (dcd_break);
else if (!op_valid)
r_op_break <= 1'b0;
assign op_break = r_op_break;
`else
assign op_break = dcd_break;
915,33 → 885,21
generate
if (IMPLEMENT_LOCK != 0)
begin
reg r_op_lock, r_op_lock_stall;
reg r_op_lock;
 
initial r_op_lock_stall = 1'b0;
always @(posedge i_clk)
if (i_rst)
r_op_lock_stall <= 1'b0;
else
r_op_lock_stall <= (~opvalid)||(~op_lock)
||(~dcdvalid)||(~pf_valid);
 
assign op_lock_stall = r_op_lock_stall;
 
initial r_op_lock = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
if (clear_pipeline)
r_op_lock <= 1'b0;
else if (op_ce)
r_op_lock <= (dcd_lock)&&(~clear_pipeline);
r_op_lock <= (dcd_valid)&&(dcd_lock)&&(~clear_pipeline);
assign op_lock = r_op_lock;
 
end else begin
assign op_lock_stall = 1'b0;
assign op_lock = 1'b0;
end endgenerate
 
`else
assign op_lock_stall = 1'b0;
assign op_lock = 1'b0;
`endif
 
948,13 → 906,13
`ifdef OPT_ILLEGAL_INSTRUCTION
initial op_illegal = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
if (clear_pipeline)
op_illegal <= 1'b0;
else if(op_ce)
`ifdef OPT_PIPELINED
op_illegal <= (dcdvalid)&&((dcd_illegal)||((dcd_lock)&&(IMPLEMENT_LOCK == 0)));
op_illegal <= (dcd_valid)&&((dcd_illegal)||((dcd_lock)&&(IMPLEMENT_LOCK == 0)));
`else
op_illegal <= (dcdvalid)&&((dcd_illegal)||(dcd_lock));
op_illegal <= (dcd_valid)&&((dcd_illegal)||(dcd_lock));
`endif
else if(alu_ce)
op_illegal <= 1'b0;
967,61 → 925,87
always @(posedge i_clk)
if (op_ce)
begin
opF_wr <= (dcdF_wr)&&((~dcdR_cc)||(~dcdR_wr))
op_wF <= (dcd_wF)&&((~dcd_Rcc)||(~dcd_wR))
&&(~dcd_early_branch)&&(~dcd_illegal);
opR_wr <= (dcdR_wr)&&(~dcd_early_branch)&&(~dcd_illegal);
op_wR <= (dcd_wR)&&(~dcd_early_branch)&&(~dcd_illegal);
end
`else
always @(posedge i_clk)
begin
opF_wr <= (dcdF_wr)&&((~dcdR_cc)||(~dcdR_wr))
op_wF <= (dcd_wF)&&((~dcd_Rcc)||(~dcd_wR))
&&(~dcd_early_branch)&&(~dcd_illegal);
opR_wr <= (dcdR_wr)&&(~dcd_early_branch)&&(~dcd_illegal);
op_wR <= (dcd_wR)&&(~dcd_early_branch)&&(~dcd_illegal);
end
`endif
 
`ifdef VERILATOR
`ifdef OPT_PIPELINED
reg [3:0] r_opn;
reg [4:0] r_opR;
reg r_opR_cc;
always @(posedge i_clk)
if (op_change_data_ce)
begin
op_sim <= dcd_sim;
op_sim_immv <= dcd_sim_immv;
end
`else
always @(*)
begin
op_sim = dcd_sim;
op_sim_immv = dcd_sim_immv;
end
`endif
`endif
 
`ifdef OPT_PIPELINED
reg [3:0] r_op_opn;
reg [4:0] r_op_R;
reg r_op_Rcc;
reg r_op_gie;
always @(posedge i_clk)
if (op_change_data_ce)
begin
r_opn <= dcdOp; // Which ALU operation?
// opM <= dcdM; // Is this a memory operation?
// Which ALU operation? Early branches are
// unimplemented moves
r_op_opn <= (dcd_early_branch) ? 4'hf : dcd_opn;
// opM <= dcd_M; // Is this a memory operation?
// What register will these results be written into?
r_opR <= dcdR;
r_opR_cc <= (dcdR_cc)&&(dcdR_wr)&&(dcdR[4]==dcd_gie);
r_op_R <= dcd_R;
r_op_Rcc <= (dcd_Rcc)&&(dcd_wR)&&(dcd_R[4]==dcd_gie);
// User level (1), vs supervisor (0)/interrupts disabled
r_op_gie <= dcd_gie;
 
 
//
op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc[AW:1];
end
assign opn = r_opn;
assign opR = r_opR;
assign op_opn = r_op_opn;
assign op_R = r_op_R;
`ifdef OPT_NO_USERMODE
assign op_gie = 1'b0;
`else
assign op_gie = r_op_gie;
assign opR_cc = r_opR_cc;
`endif
assign op_Rcc = r_op_Rcc;
`else
assign opn = dcdOp;
assign opR = dcdR;
assign op_opn = dcd_opn;
assign op_R = dcd_R;
`ifdef OPT_NO_USERMODE
assign op_gie = 1'b0;
`else
assign op_gie = dcd_gie;
`endif
// With no pipelining, there is no early branching. We keep it
always @(posedge i_clk)
op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc[AW:1];
`endif
assign opFl = (op_gie)?(w_uflags):(w_iflags);
assign op_Fl = (op_gie)?(w_uflags):(w_iflags);
 
`ifdef OPT_VLIW
`ifdef OPT_CIS
reg r_op_phase;
initial r_op_phase = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
if (clear_pipeline)
r_op_phase <= 1'b0;
else if (op_change_data_ce)
r_op_phase <= dcd_phase;
r_op_phase <= (dcd_phase)&&((!dcd_wR)||(!dcd_Rpc));
assign op_phase = r_op_phase;
`else
assign op_phase = 1'b0;
1040,10 → 1024,10
// define this flag to something other than just plain zero, then
// the stalls will already be in place.
`ifdef OPT_PIPELINED
assign opA = ((wr_reg_ce)&&(wr_reg_id == opA_id)) // &&(opA_rd))
? wr_gpreg_vl : r_opA;
assign op_Av = ((wr_reg_ce)&&(wr_reg_id == op_Aid)) // &&(op_rA))
? wr_gpreg_vl : r_op_Av;
`else
assign opA = r_opA;
assign op_Av = r_op_Av;
`endif
 
`ifdef OPT_PIPELINED
1053,21 → 1037,21
// The operation might set flags, and we wish to read the
// CC register
// OR ... (No other conditions)
assign dcdA_stall = (dcdA_rd) // &&(dcdvalid) is checked for elsewhere
&&((opvalid)||(mem_rdbusy)
assign dcd_A_stall = (dcd_rA) // &&(dcd_valid) is checked for elsewhere
&&((op_valid)||(mem_rdbusy)
||(div_busy)||(fpu_busy))
&&(((opF_wr)||(cc_invalid_for_dcd))&&(dcdA_cc))
||((dcdA_rd)&&(dcdA_cc)&&(cc_invalid_for_dcd));
&&(((op_wF)||(cc_invalid_for_dcd))&&(dcd_Acc))
||((dcd_rA)&&(dcd_Acc)&&(cc_invalid_for_dcd));
`else
// There are no pipeline hazards, if we aren't pipelined
assign dcdA_stall = 1'b0;
assign dcd_A_stall = 1'b0;
`endif
 
`ifdef OPT_PIPELINED
assign opB = ((wr_reg_ce)&&(wr_reg_id == opB_id)&&(opB_rd))
? wr_gpreg_vl: r_opB;
assign op_Bv = ((wr_reg_ce)&&(wr_reg_id == op_Bid)&&(op_rB))
? wr_gpreg_vl: r_op_Bv;
`else
assign opB = r_opB;
assign op_Bv = r_op_Bv;
`endif
 
`ifdef OPT_PIPELINED
1078,12 → 1062,12
// CC register
// OR the operation might set register B, and we still need
// a clock to add the offset to it
assign dcdB_stall = (dcdB_rd) // &&(dcdvalid) is checked for elsewhere
assign dcd_B_stall = (dcd_rB) // &&(dcd_valid) is checked for elsewhere
// If the op stage isn't valid, yet something
// is running, then it must have been valid.
// We'll use the last values from that stage
// (opR_wr, opF_wr, opR) in our logic below.
&&((opvalid)||(mem_rdbusy)
// (op_wR, op_wF, op_R) in our logic below.
&&((op_valid)||(mem_rdbusy)
||(div_busy)||(fpu_busy)||(alu_busy))
&&(
// Okay, what happens if the result register
1090,13 → 1074,13
// from instruction 1 becomes the input for
// instruction two, *and* there's an immediate
// offset in instruction two? In that case, we
// need an extra clock between the two
// instructions to calculate the base plus
// need an extra clock between the two
// instructions to calculate the base plus
// offset.
//
// What if instruction 1 (or before) is in a
// memory pipeline? We may no longer know what
// the register was! We will then need to
// the register was! We will then need to
// blindly wait. We'll temper this only waiting
// if we're not piping this new instruction.
// If we were piping, the pipe logic in the
1104,28 → 1088,28
// is clear, so we're okay then.
//
((~dcd_zI)&&(
((opR == dcdB)&&(opR_wr))
((op_R == dcd_B)&&(op_wR))
||((mem_rdbusy)&&(~dcd_pipe))
))
// Stall following any instruction that will
// set the flags, if we're going to need the
// flags (CC) register for opB.
||(((opF_wr)||(cc_invalid_for_dcd))&&(dcdB_cc))
// flags (CC) register for op_B.
||(((op_wF)||(cc_invalid_for_dcd))&&(dcd_Bcc))
// Stall on any ongoing memory operation that
// will write to opB -- captured above
// ||((mem_busy)&&(~mem_we)&&(mem_last_reg==dcdB)&&(~dcd_zI))
// will write to op_B -- captured above
// ||((mem_busy)&&(~mem_we)&&(mem_last_reg==dcd_B)&&(~dcd_zI))
)
||((dcdB_rd)&&(dcdB_cc)&&(cc_invalid_for_dcd));
assign dcdF_stall = ((~dcdF[3])
||((dcdA_rd)&&(dcdA_cc))
||((dcdB_rd)&&(dcdB_cc)))
&&(opvalid)&&(opR_cc);
// &&(dcdvalid) is checked for elsewhere
||((dcd_rB)&&(dcd_Bcc)&&(cc_invalid_for_dcd));
assign dcd_F_stall = ((~dcd_F[3])
||((dcd_rA)&&(dcd_Acc))
||((dcd_rB)&&(dcd_Bcc)))
&&(op_valid)&&(op_Rcc);
// &&(dcd_valid) is checked for elsewhere
`else
// No stalls without pipelining, 'cause how can you have a pipeline
// hazard without the pipeline?
assign dcdB_stall = 1'b0;
assign dcdF_stall = 1'b0;
assign dcd_B_stall = 1'b0;
assign dcd_F_stall = 1'b0;
`endif
//
//
1132,15 → 1116,15
// PIPELINE STAGE #4 :: Apply Instruction
//
//
cpuops #(IMPLEMENT_MPY) doalu(i_clk, (i_rst)||(clear_pipeline),
alu_ce, opn, opA, opB,
cpuops #(IMPLEMENT_MPY) doalu(i_clk, (clear_pipeline),
alu_ce, op_opn, op_Av, op_Bv,
alu_result, alu_flags, alu_valid, alu_busy);
 
generate
if (IMPLEMENT_DIVIDE != 0)
begin
div thedivide(i_clk, (i_rst)||(clear_pipeline), div_ce, opn[0],
opA, opB, div_busy, div_valid, div_error, div_result,
div thedivide(i_clk, (clear_pipeline), div_ce, op_opn[0],
op_Av, op_Bv, div_busy, div_valid, div_error, div_result,
div_flags);
end else begin
assign div_error = 1'b0; // Can't be high unless div_valid
1155,7 → 1139,7
begin
//
// sfpu thefpu(i_clk, i_rst, fpu_ce,
// opA, opB, fpu_busy, fpu_valid, fpu_err, fpu_result,
// op_Av, op_Bv, fpu_busy, fpu_valid, fpu_err, fpu_result,
// fpu_flags);
//
assign fpu_error = 1'b0; // Must only be true if fpu_valid
1172,27 → 1156,27
end endgenerate
 
 
assign set_cond = ((opF[7:4]&opFl[3:0])==opF[3:0]);
initial alF_wr = 1'b0;
initial alu_wr = 1'b0;
assign set_cond = ((op_F[7:4]&op_Fl[3:0])==op_F[3:0]);
initial alu_wF = 1'b0;
initial alu_wR = 1'b0;
always @(posedge i_clk)
if (i_rst)
begin
alu_wr <= 1'b0;
alF_wr <= 1'b0;
alu_wR <= 1'b0;
alu_wF <= 1'b0;
end else if (alu_ce)
begin
// alu_reg <= opR;
alu_wr <= (opR_wr)&&(set_cond);
alF_wr <= (opF_wr)&&(set_cond);
// alu_reg <= op_R;
alu_wR <= (op_wR)&&(set_cond);
alu_wF <= (op_wF)&&(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;
alu_wR <= (i_halt)&&(i_dbg_we);
alu_wF <= 1'b0;
end
 
`ifdef OPT_VLIW
`ifdef OPT_CIS
reg r_alu_phase;
initial r_alu_phase = 1'b0;
always @(posedge i_clk)
1208,7 → 1192,7
`ifdef OPT_PIPELINED
always @(posedge i_clk)
if (adf_ce_unconditional)
alu_reg <= opR;
alu_reg <= op_R;
else if ((i_halt)&&(i_dbg_we))
alu_reg <= i_dbg_reg;
`else
1216,7 → 1200,7
if ((i_halt)&&(i_dbg_we))
alu_reg <= i_dbg_reg;
else
alu_reg <= opR;
alu_reg <= op_R;
`endif
 
//
1229,6 → 1213,9
reg [31:0] dbg_val;
always @(posedge i_clk)
dbg_val <= i_dbg_data;
`ifdef OPT_NO_USERMODE
assign alu_gie = 1'b0;
`else
`ifdef OPT_PIPELINED
reg r_alu_gie;
 
1236,24 → 1223,27
if ((adf_ce_unconditional)||(mem_ce))
r_alu_gie <= op_gie;
assign alu_gie = r_alu_gie;
`else
assign alu_gie = op_gie;
`endif
`endif
 
`ifdef OPT_PIPELINED
reg [(AW-1):0] r_alu_pc;
always @(posedge i_clk)
if ((adf_ce_unconditional)
||((master_ce)&&(opvalid_mem)&&(~clear_pipeline)
||((master_ce)&&(op_valid_mem)&&(~clear_pipeline)
&&(~mem_stalled)))
r_alu_pc <= op_pc;
assign alu_pc = r_alu_pc;
`else
assign alu_gie = op_gie;
assign alu_pc = op_pc;
`endif
 
`ifdef OPT_ILLEGAL_INSTRUCTION
reg r_alu_illegal;
initial r_alu_illegal = 0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline))
if (clear_pipeline)
r_alu_illegal <= 1'b0;
else if (alu_ce)
r_alu_illegal <= op_illegal;
1260,16 → 1250,13
else
r_alu_illegal <= 1'b0;
assign alu_illegal = (r_alu_illegal);
`else
assign alu_illegal = 1'b0;
`endif
 
initial r_alu_pc_valid = 1'b0;
initial mem_pc_valid = 1'b0;
always @(posedge i_clk)
if (i_rst)
if (clear_pipeline)
r_alu_pc_valid <= 1'b0;
else if (adf_ce_unconditional)//Includes&&(~alu_clear_pipeline)
else if ((adf_ce_unconditional)&&(!op_phase)) //Includes&&(~alu_clear_pipeline)
r_alu_pc_valid <= 1'b1;
else if (((~alu_busy)&&(~div_busy)&&(~fpu_busy))||(clear_pipeline))
r_alu_pc_valid <= 1'b0;
1285,17 → 1272,43
generate
if (IMPLEMENT_LOCK != 0)
begin
reg r_prelock_stall;
 
initial r_prelock_stall = 1'b0;
always @(posedge i_clk)
if (clear_pipeline)
r_prelock_stall <= 1'b0;
else if ((op_valid)&&(op_lock)&&(op_ce))
r_prelock_stall <= 1'b1;
else if ((op_valid)&&(dcd_valid)&&(pf_valid))
r_prelock_stall <= 1'b0;
 
assign prelock_stall = r_prelock_stall;
 
reg r_prelock_primed;
always @(posedge i_clk)
if (clear_pipeline)
r_prelock_primed <= 1'b0;
else if (r_prelock_stall)
r_prelock_primed <= 1'b1;
else if ((adf_ce_unconditional)||(mem_ce))
r_prelock_primed <= 1'b0;
 
reg [1:0] r_bus_lock;
initial r_bus_lock = 2'b00;
always @(posedge i_clk)
if (i_rst)
if (clear_pipeline)
r_bus_lock <= 2'b00;
else if ((op_ce)&&(op_lock))
r_bus_lock <= 2'b11;
else if ((|r_bus_lock)&&((~opvalid_mem)||(~op_ce)))
r_bus_lock <= r_bus_lock + 2'b11;
else if ((op_valid)&&((adf_ce_unconditional)||(mem_ce)))
begin
if (r_prelock_primed)
r_bus_lock <= 2'b10;
else if (r_bus_lock != 2'h0)
r_bus_lock <= r_bus_lock + 2'b11;
end
assign bus_lock = |r_bus_lock;
end else begin
assign prelock_stall = 1'b0;
assign bus_lock = 1'b0;
end endgenerate
`else
1304,38 → 1317,49
 
`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,
(op_opn[2:0]), op_Bv, op_Av, op_R,
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_we, mem_addr, mem_data, mem_sel,
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,
memops #(AW,IMPLEMENT_LOCK,WITH_LOCAL_BUS) domem(i_clk, i_rst,(mem_ce)&&(set_cond), bus_lock,
(op_opn[2:0]), op_Bv, op_Av, op_R,
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_we, mem_addr, mem_data, mem_sel,
mem_ack, mem_stall, mem_err, i_wb_data);
assign mem_pipe_stalled = 1'b0;
`endif // PIPELINED_BUS_ACCESS
assign mem_rdbusy = ((mem_busy)&&(~mem_we));
 
// Either the prefetch or the instruction gets the memory bus, but
// 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,
mem_we, mem_addr, mem_data, mem_sel,
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,
//
// At a first glance, we might want something like:
//
// pf_cyc, 1'b0, pf_stb, 1'b0, pf_we, pf_addr, pf_data, 4'hf,
//
// However, we know that the prefetch will not generate any
// writes. Therefore, the write specific lines (mem_data and
// mem_sel) can be shared with the memory in order to ease
// timing and LUT usage.
pf_cyc,1'b0,pf_stb, 1'b0, pf_we, pf_addr, mem_data, mem_sel,
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,
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, o_wb_sel,
i_wb_ack, i_wb_stall, i_wb_err);
 
 
1362,25 → 1386,25
// 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
// Further, alu_wR includes (set_cond), so we don't need to
// check for that here either.
`ifdef OPT_ILLEGAL_INSTRUCTION
assign wr_reg_ce = (dbgv)||(mem_valid)
||((~clear_pipeline)&&(~alu_illegal)
&&(((alu_wr)&&(alu_valid))
&&(((alu_wR)&&(alu_valid))
||(div_valid)||(fpu_valid)));
`else
assign wr_reg_ce = (dbgv)||(mem_valid)
||((~clear_pipeline)
&&(((alu_wr)&&(alu_valid))
||(div_valid)||(fpu_valid)));
`endif
// Which register shall be written?
// COULD SIMPLIFY THIS: by adding three bits to these registers,
// One or PC, one for CC, and one for GIE match
// Note that the alu_reg is the register to write on a divide or
// FPU operation.
assign wr_reg_id = (alu_wr|div_valid|fpu_valid)?alu_reg:mem_wreg;
`ifdef OPT_NO_USERMODE
assign wr_reg_id[3:0] = (alu_wR|div_valid|fpu_valid)
? alu_reg[3:0]:mem_wreg[3:0];
assign wr_reg_id[4] = 1'b0;
`else
assign wr_reg_id = (alu_wR|div_valid|fpu_valid)?alu_reg:mem_wreg;
`endif
 
// Are we writing to the CC register?
assign wr_write_cc = (wr_reg_id[3:0] == `CPU_CC_REG);
assign wr_write_scc = (wr_reg_id[4:0] == {1'b0, `CPU_CC_REG});
1397,13 → 1421,17
:((dbgv) ? dbg_val : alu_result));
always @(posedge i_clk)
if (wr_reg_ce)
regset[wr_reg_id] <= wr_gpreg_vl;
`ifdef OPT_NO_USERMODE
regset[wr_reg_id[3:0]] <= wr_gpreg_vl;
`else
regset[wr_reg_id] <= wr_gpreg_vl;
`endif
 
//
// Write back to the condition codes/flags register ...
// When shall we write to our flags register? alF_wr already
// When shall we write to our flags register? alu_wF already
// includes the set condition ...
assign wr_flags_ce = ((alF_wr)||(div_valid)||(fpu_valid))&&(~clear_pipeline)&&(~alu_illegal);
assign wr_flags_ce = ((alu_wF)||(div_valid)||(fpu_valid))&&(~clear_pipeline)&&(~alu_illegal);
assign w_uflags = { 1'b0, uhalt_phase, ufpu_err_flag,
udiv_err_flag, ubus_err_flag, trap, ill_err_u,
ubreak, step, 1'b1, sleep,
1437,7 → 1465,7
//
// 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.
// 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
1458,7 → 1486,7
 
initial r_break_pending = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(clear_pipeline)||(~opvalid))
if ((clear_pipeline)||(~op_valid))
r_break_pending <= 1'b0;
else if (op_break)
r_break_pending <= (~alu_busy)&&(~div_busy)&&(~fpu_busy)&&(~mem_busy)&&(!wr_reg_ce);
1481,9 → 1509,31
// 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
// set the sleep bit and switch to supervisor mode in the same
// instruction: users are not allowed to halt the CPU.
initial sleep = 1'b0;
`ifdef OPT_NO_USERMODE
reg r_sleep_is_halt;
initial r_sleep_is_halt = 1'b0;
always @(posedge i_clk)
if (i_rst)
r_sleep_is_halt <= 1'b0;
else if ((wr_reg_ce)&&(wr_write_cc)
&&(wr_spreg_vl[`CPU_SLEEP_BIT])
&&(~wr_spreg_vl[`CPU_GIE_BIT]))
r_sleep_is_halt <= 1'b1;
 
// Trying to switch to user mode, either via a WAIT or an RTU
// instruction will cause the CPU to sleep until an interrupt, in
// the NO-USERMODE build.
always @(posedge i_clk)
if ((i_rst)||((i_interrupt)&&(!r_sleep_is_halt)))
sleep <= 1'b0;
else if ((wr_reg_ce)&&(wr_write_cc)
&&(wr_spreg_vl[`CPU_GIE_BIT]))
sleep <= 1'b1;
`else
always @(posedge i_clk)
if ((i_rst)||(w_switch_to_interrupt))
sleep <= 1'b0;
else if ((wr_reg_ce)&&(wr_write_cc)&&(~alu_gie))
1503,6 → 1553,7
// to sleep mode *and* supervisor mode at the same
// time, lest you halt the CPU.
sleep <= wr_spreg_vl[`CPU_SLEEP_BIT];
`endif
 
always @(posedge i_clk)
if (i_rst)
1511,6 → 1562,10
step <= wr_spreg_vl[`CPU_STEP_BIT];
 
// The GIE register. Only interrupts can disable the interrupt register
`ifdef OPT_NO_USERMODE
assign w_switch_to_interrupt = 1'b0;
assign w_release_from_interrupt = 1'b0;
`else
assign w_switch_to_interrupt = (gie)&&(
// On interrupt (obviously)
((i_interrupt)&&(~alu_phase)&&(~bus_lock))
1519,10 → 1574,8
// If we encounter a break instruction, if the break
// enable isn't set.
||((master_ce)&&(break_pending)&&(~break_en))
`ifdef OPT_ILLEGAL_INSTRUCTION
// On an illegal instruction
||((alu_illegal)&&(!clear_pipeline))
`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
1531,7 → 1584,7
// fpu_error must *never* be set unless fpu_valid is
// also set as well, else this will fail.
||(fpu_error)
//
//
||(bus_err)
// If we write to the CC register
||((wr_reg_ce)&&(~wr_spreg_vl[`CPU_GIE_BIT])
1542,34 → 1595,56
&&(((wr_reg_ce)&&(wr_spreg_vl[`CPU_GIE_BIT])
&&(wr_write_scc))
);
`endif
 
`ifdef OPT_NO_USERMODE
assign gie = 1'b0;
`else
reg r_gie;
 
initial r_gie = 1'b0;
always @(posedge i_clk)
if (i_rst)
gie <= 1'b0;
r_gie <= 1'b0;
else if (w_switch_to_interrupt)
gie <= 1'b0;
r_gie <= 1'b0;
else if (w_release_from_interrupt)
gie <= 1'b1;
r_gie <= 1'b1;
assign gie = r_gie;
`endif
 
initial trap = 1'b0;
`ifdef OPT_NO_USERMODE
assign trap = 1'b0;
assign ubreak = 1'b0;
`else
reg r_trap;
 
initial r_trap = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(w_release_from_interrupt))
trap <= 1'b0;
r_trap <= 1'b0;
else if ((alu_gie)&&(wr_reg_ce)&&(~wr_spreg_vl[`CPU_GIE_BIT])
&&(wr_write_ucc)) // &&(wr_reg_id[4]) implied
trap <= 1'b1;
r_trap <= 1'b1;
else if ((wr_reg_ce)&&(wr_write_ucc)&&(~alu_gie))
trap <= (trap)&&(wr_spreg_vl[`CPU_TRAP_BIT]);
r_trap <= (r_trap)&&(wr_spreg_vl[`CPU_TRAP_BIT]);
 
initial ubreak = 1'b0;
reg r_ubreak;
 
initial r_ubreak = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(w_release_from_interrupt))
ubreak <= 1'b0;
r_ubreak <= 1'b0;
else if ((op_gie)&&(break_pending)&&(w_switch_to_interrupt))
ubreak <= 1'b1;
r_ubreak <= 1'b1;
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
ubreak <= (ubreak)&&(wr_spreg_vl[`CPU_BREAK_BIT]);
r_ubreak <= (ubreak)&&(wr_spreg_vl[`CPU_BREAK_BIT]);
 
assign trap = r_trap;
assign ubreak = r_ubreak;
`endif
 
 
`ifdef OPT_ILLEGAL_INSTRUCTION
initial ill_err_i = 1'b0;
always @(posedge i_clk)
1580,18 → 1655,27
ill_err_i <= (ill_err_i)&&(wr_spreg_vl[`CPU_ILL_BIT]);
else if ((alu_illegal)&&(~alu_gie)&&(!clear_pipeline))
ill_err_i <= 1'b1;
initial ill_err_u = 1'b0;
 
`ifdef OPT_NO_USERMODE
assign ill_err_u = 1'b0;
`else
reg r_ill_err_u;
 
initial r_ill_err_u = 1'b0;
always @(posedge i_clk)
// The bit is automatically cleared on release from interrupt
// or reset
if ((i_rst)||(w_release_from_interrupt))
ill_err_u <= 1'b0;
r_ill_err_u <= 1'b0;
// If the supervisor (or debugger) writes to this register,
// clearing the bit, then clear it
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
ill_err_u <=((ill_err_u)&&(wr_spreg_vl[`CPU_ILL_BIT]));
r_ill_err_u <=((ill_err_u)&&(wr_spreg_vl[`CPU_ILL_BIT]));
else if ((alu_illegal)&&(alu_gie)&&(!clear_pipeline))
ill_err_u <= 1'b1;
r_ill_err_u <= 1'b1;
 
assign ill_err_u = r_ill_err_u;
`endif
`else
assign ill_err_u = 1'b0;
assign ill_err_i = 1'b0;
1607,16 → 1691,24
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;
// supervisor mode.
`ifdef OPT_NO_USERMODE
assign ubus_err_flag = 1'b0;
`else
reg r_ubus_err_flag;
 
initial r_ubus_err_flag = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(w_release_from_interrupt))
ubus_err_flag <= 1'b0;
r_ubus_err_flag <= 1'b0;
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)&&(wr_write_ucc))
ubus_err_flag <= (ubus_err_flag)&&(wr_spreg_vl[`CPU_BUSERR_BIT]);
r_ubus_err_flag <= (ubus_err_flag)&&(wr_spreg_vl[`CPU_BUSERR_BIT]);
else if ((bus_err)&&(alu_gie))
ubus_err_flag <= 1'b1;
r_ubus_err_flag <= 1'b1;
 
assign ubus_err_flag = r_ubus_err_flag;
`endif
 
generate
if (IMPLEMENT_DIVIDE != 0)
begin
1633,8 → 1725,13
r_idiv_err_flag <= (r_idiv_err_flag)&&(wr_spreg_vl[`CPU_DIVERR_BIT]);
else if ((div_error)&&(~alu_gie))
r_idiv_err_flag <= 1'b1;
 
assign idiv_err_flag = r_idiv_err_flag;
`ifdef OPT_NO_USERMODE
assign udiv_err_flag = 1'b0;
`else
// User divide (by zero) error flag -- if ever set, it will
// cause a sudden switch interrupt to supervisor mode.
// cause a sudden switch interrupt to supervisor mode.
initial r_udiv_err_flag = 1'b0;
always @(posedge i_clk)
if ((i_rst)||(w_release_from_interrupt))
1645,8 → 1742,8
else if ((div_error)&&(alu_gie))
r_udiv_err_flag <= 1'b1;
 
assign idiv_err_flag = r_idiv_err_flag;
assign udiv_err_flag = r_udiv_err_flag;
`endif
end else begin
assign idiv_err_flag = 1'b0;
assign udiv_err_flag = 1'b0;
1667,7 → 1764,7
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.
// a sudden switch interrupt to supervisor mode.
initial r_ufpu_err_flag = 1'b0;
always @(posedge i_clk)
if ((i_rst)&&(w_release_from_interrupt))
1685,16 → 1782,24
assign ufpu_err_flag = 1'b0;
end endgenerate
 
`ifdef OPT_VLIW
reg r_ihalt_phase, r_uhalt_phase;
`ifdef OPT_CIS
reg r_ihalt_phase;
 
initial r_ihalt_phase = 0;
initial r_uhalt_phase = 0;
always @(posedge i_clk)
if (i_rst)
r_ihalt_phase <= 1'b0;
else if ((~alu_gie)&&(alu_pc_valid)&&(~clear_pipeline))
r_ihalt_phase <= alu_phase;
 
assign ihalt_phase = r_ihalt_phase;
 
`ifdef OPT_NO_USERMODE
assign uhalt_phase = 1'b0;
`else
reg r_uhalt_phase;
 
initial r_uhalt_phase = 0;
always @(posedge i_clk)
if ((i_rst)||(w_release_from_interrupt))
r_uhalt_phase <= 1'b0;
1703,8 → 1808,8
else if ((~alu_gie)&&(wr_reg_ce)&&(wr_write_ucc))
r_uhalt_phase <= wr_spreg_vl[`CPU_PHASE_BIT];
 
assign ihalt_phase = r_ihalt_phase;
assign uhalt_phase = r_uhalt_phase;
`endif
`else
assign ihalt_phase = 1'b0;
assign uhalt_phase = 1'b0;
1720,58 → 1825,52
// 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?
`ifdef OPT_NO_USERMODE
assign upc = {(AW+2){1'b0}};
`else
reg [(AW+1):0] r_upc;
 
always @(posedge i_clk)
if ((wr_reg_ce)&&(wr_reg_id[4])&&(wr_write_pc))
upc <= wr_spreg_vl[(AW-1):0];
r_upc <= { wr_spreg_vl[(AW+1):2], 2'b00 };
else if ((alu_gie)&&
(((alu_pc_valid)&&(~clear_pipeline)&&(!alu_illegal))
||(mem_pc_valid)))
upc <= alu_pc;
r_upc <= { alu_pc, 2'b00 };
assign upc = r_upc;
`endif
 
always @(posedge i_clk)
if (i_rst)
ipc <= RESET_ADDRESS;
ipc <= { RESET_BUS_ADDRESS, 2'b00 };
else if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_pc))
ipc <= wr_spreg_vl[(AW-1):0];
else if ((~alu_gie)&&
(((alu_pc_valid)&&(~clear_pipeline))
ipc <= { wr_spreg_vl[(AW+1):2], 2'b00 };
else if ((!alu_gie)&&(!alu_phase)&&
(((alu_pc_valid)&&(~clear_pipeline)&&(!alu_illegal))
||(mem_pc_valid)))
ipc <= alu_pc;
ipc <= { alu_pc, 2'b00 };
 
always @(posedge i_clk)
if (i_rst)
pf_pc <= RESET_ADDRESS;
pf_pc <= { RESET_BUS_ADDRESS, 2'b00 };
else if ((w_switch_to_interrupt)||((~gie)&&(w_clear_icache)))
pf_pc <= ipc;
pf_pc <= { ipc[(AW+1):2], 2'b00 };
else if ((w_release_from_interrupt)||((gie)&&(w_clear_icache)))
pf_pc <= upc;
pf_pc <= { upc[(AW+1):2], 2'b00 };
else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
pf_pc <= wr_spreg_vl[(AW-1):0];
pf_pc <= { wr_spreg_vl[(AW+1):2], 2'b00 };
`ifdef OPT_PIPELINED
else if ((dcd_early_branch)&&(~clear_pipeline))
pf_pc <= dcd_branch_pc + 1;
else if ((new_pc)||((~dcd_stalled)&&(pf_valid)))
pf_pc <= pf_pc + {{(AW-1){1'b0}},1'b1};
pf_pc <= { dcd_branch_pc + 1'b1, 2'b00 };
else if ((new_pc)||((!pf_stalled)&&(pf_valid)))
pf_pc <= { pf_pc[(AW+1):2] + {{(AW-1){1'b0}},1'b1}, 2'b00 };
`else
else if ((alu_gie==gie)&&(
((alu_pc_valid)&&(~clear_pipeline))
||(mem_pc_valid)))
pf_pc <= alu_pc;
pf_pc <= { alu_pc[(AW-1):0], 2'b00 };
`endif
 
initial new_pc = 1'b1;
always @(posedge i_clk)
if ((i_rst)||(i_clear_pf_cache))
new_pc <= 1'b1;
else if (w_switch_to_interrupt)
new_pc <= 1'b1;
else if (w_release_from_interrupt)
new_pc <= 1'b1;
else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
new_pc <= 1'b1;
else
new_pc <= 1'b0;
 
`ifdef OPT_PIPELINED
reg r_clear_icache;
initial r_clear_icache = 1'b1;
1784,42 → 1883,63
r_clear_icache <= 1'b0;
assign w_clear_icache = r_clear_icache;
`else
assign w_clear_icache = 1'b0;
assign w_clear_icache = i_clear_pf_cache;
`endif
 
initial new_pc = 1'b1;
always @(posedge i_clk)
if ((i_rst)||(w_clear_icache))
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
wire [31:0] w_debug_pc;
`ifdef OPT_NO_USERMODE
assign w_debug_pc[(AW+1):0] = { ipc, 2'b00 };
`else
assign w_debug_pc[(AW+1):0] = { (i_dbg_reg[4])
? { upc[(AW+1):2], uhalt_phase, 1'b0 }
: { ipc[(AW+1):2], ihalt_phase, 1'b0 } };
`endif
generate
if (AW<32)
if (AW<30)
assign w_debug_pc[31:(AW+2)] = 0;
endgenerate
 
always @(posedge i_clk)
begin
always @(posedge i_clk)
`ifdef OPT_NO_USERMODE
o_dbg_reg <= regset[i_dbg_reg[3:0]];
if (i_dbg_reg[3:0] == `CPU_PC_REG)
o_dbg_reg <= w_debug_pc;
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
begin
o_dbg_reg <= regset[i_dbg_reg];
if (i_dbg_reg[3:0] == `CPU_PC_REG)
o_dbg_reg <= {{(32-AW){1'b0}},(i_dbg_reg[4])?upc:ipc};
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
begin
o_dbg_reg[14:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
o_dbg_reg[15] <= 1'b0;
o_dbg_reg[31:23] <= w_cpu_info;
o_dbg_reg[`CPU_GIE_BIT] <= gie;
end
o_dbg_reg[14:0] <= w_iflags;
o_dbg_reg[15] <= 1'b0;
o_dbg_reg[31:23] <= w_cpu_info;
o_dbg_reg[`CPU_GIE_BIT] <= gie;
end
end else begin
always @(posedge i_clk)
`else
o_dbg_reg <= regset[i_dbg_reg];
if (i_dbg_reg[3:0] == `CPU_PC_REG)
o_dbg_reg <= w_debug_pc;
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
begin
o_dbg_reg <= regset[i_dbg_reg];
if (i_dbg_reg[3:0] == `CPU_PC_REG)
o_dbg_reg <= (i_dbg_reg[4])?upc:ipc;
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
begin
o_dbg_reg[14:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
o_dbg_reg[15] <= 1'b0;
o_dbg_reg[31:23] <= w_cpu_info;
o_dbg_reg[`CPU_GIE_BIT] <= gie;
end
o_dbg_reg[14:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
o_dbg_reg[15] <= 1'b0;
o_dbg_reg[31:23] <= w_cpu_info;
o_dbg_reg[`CPU_GIE_BIT] <= gie;
end
end endgenerate
`endif
end
 
always @(posedge i_clk)
o_dbg_cc <= { o_break, bus_err, gie, sleep };
1832,12 → 1952,12
(~pf_cyc)&&(~mem_busy)&&(~alu_busy)
&&(~div_busy)&&(~fpu_busy)
// Operations must either be valid, or illegal
&&((opvalid)||(i_rst)||(dcd_illegal))
&&((op_valid)||(i_rst)||(dcd_illegal))
// Decode stage must be either valid, in reset, or ill
&&((dcdvalid)||(i_rst)||(pf_illegal)));
&&((dcd_valid)||(i_rst)||(pf_illegal)));
`else
always @(posedge i_clk)
r_halted <= (i_halt)&&((opvalid)||(i_rst));
r_halted <= (i_halt)&&((op_valid)||(i_rst));
`endif
assign o_dbg_stall = ~r_halted;
 
1857,24 → 1977,24
/*
o_break, i_wb_err, pf_pc[1:0],
flags,
pf_valid, dcdvalid, opvalid, alu_valid, mem_valid,
pf_valid, dcd_valid, op_valid, alu_valid, mem_valid,
op_ce, alu_ce, mem_ce,
//
master_ce, opvalid_alu, opvalid_mem,
master_ce, op_valid_alu, op_valid_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],
// ((op_valid_alu)&&(alu_stall))
// ||((op_valid_mem)&&(~op_pipe)&&(mem_busy))
// ||((op_valid_mem)&&( op_pipe)&&(mem_pipe_stalled)));
// op_Av[23:20], op_Av[3:0],
gie, sleep, wr_reg_ce, wr_gpreg_vl[4:0]
*/
/*
i_rst, master_ce, (new_pc),
((dcd_early_branch)&&(dcdvalid)),
((dcd_early_branch)&&(dcd_valid)),
pf_valid, pf_illegal,
op_ce, dcd_ce, dcdvalid, dcd_stalled,
op_ce, dcd_ce, dcd_valid, dcd_stalled,
pf_cyc, pf_stb, pf_we, pf_ack, pf_stall, pf_err,
pf_pc[7:0], pf_addr[7:0]
*/
1884,20 → 2004,20
mem_busy,
(mem_busy)?{ (o_wb_gbl_stb|o_wb_lcl_stb), o_wb_we,
o_wb_addr[8:0] }
: { instruction[31:21] },
: { pf_instruction[31:21] },
pf_valid, (pf_valid) ? alu_pc[14:0]
:{ pf_cyc, pf_stb, pf_pc[12:0] }
:{ pf_cyc, pf_stb, pf_pc[14:2] }
 
/*
i_wb_err, gie, new_pc, dcd_early_branch, // 4
pf_valid, pf_cyc, pf_stb, instruction_pc[0], // 4
instruction[30:27], // 4
pf_valid, pf_cyc, pf_stb, pf_instruction_pc[0], // 4
pf_instruction[30:27], // 4
dcd_gie, mem_busy, o_wb_gbl_cyc, o_wb_gbl_stb, // 4
dcdvalid,
dcd_valid,
((dcd_early_branch)&&(~clear_pipeline)) // 15
? dcd_branch_pc[14:0]:pf_pc[14:0]
*/
};
`endif
 
endmodule

powered by: WebSVN 2.1.0

© copyright 1999-2022 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.