// Copyright 2006, 2007 Dennis van Weeren // // This file is part of Minimig // // Minimig is free software; 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. // // Minimig is distributed in the hope that it will be useful, // but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTABILITY or 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. If not, see . // // // // This is the Copper (part of the Agnus chip) // // 24-05-2005 -started coding (created all user accessible registers) // 25-05-2005 -added beam counter compare logic // 29-05-2005 -added blitter finished disable logic // -added copper danger/address range check logic // -added controlling state machine // -adapted to use reqdma/ackdma model // -first finished version // 11-09-2005 -added proper reset for copper location registers // 24-09-2005 -fixed bug, when an illegal register is loaded by MOVE, // the copper must halt until the next strobe or vertical blank. // the copper now does this properly // 02-10-2005 -modified skip instruction to only skip MOVE instructions. // 19-10-2005 -replaced vertb (vertical blank) signal by sof (start of frame) // 07-12-2005 -added dummy cycle after copper wakeup, this is needed for copperlists // that wait for vertical beamcounter rollover ($FFDF,FFFE) // The dummy cycle is indicated by making both selins and selreg high. // 26-12-2005 -added exception for last cycle of horizontal line, this cycle is not used by copper // // JB: // 2008-03-03 - ECS copper danger behaviour // 2008-07-08 - clean-up // 2008-07-17 - real Amiga timing behaviour (thanks to Toni Wilen for help) // 2008-11-24 - clean-up // - wait/skip free cycles reimplementation // // Although I spend a lot of time trying to figure out real behaviour of Amiga hardware this solution is not complete. // more cycle-exact // - dma_bpl implementation // - WAIT and SKIP states no longer keep CPU and blitter off the bus // 2009-01-16 - clean-up // 2009-05-24 - clean-up & renaming // 2009-05-29 - dma_bpl replaced with dma_ena // 2009-10-07 - implementation of blocked $E1 cycles // - modified copper restart // 2010-06-16 - ECS/OCS CDANG behaviour implemented module agnus_copper ( input clk, // bus clock input clk7_en, input reset, // system reset (synchronous) input ecs, // enable ECS chipset features output reqdma, // copper requests dma cycle input ackdma, // agnus dma priority logic grants dma cycle input enadma, // current slot is not used by any higher priority DMA channel input sof, // start of frame input input blit_busy, // blitter busy flag input input [7:0] vpos, // vertical beam counter input [8:0] hpos, // horizontal beam counter input [8:0] hpos_slot, // dma slot grid index input [15:0] data_in, // data bus input input [8:1] reg_address_in, // register address input output reg [8:1] reg_address_out, // register address output output reg [20:1] address_out // chip address output ); // register names and adresses parameter COP1LCH = 9'h080; parameter COP1LCL = 9'h082; parameter COP2LCH = 9'h084; parameter COP2LCL = 9'h086; parameter COPCON = 9'h02e; parameter COPINS = 9'h08c; parameter COPJMP1 = 9'h088; parameter COPJMP2 = 9'h08a; // copper states parameter RESET = 3'b000; parameter FETCH1 = 3'b100; parameter FETCH2 = 3'b101; parameter WAITSKIP1 = 3'b111; parameter WAITSKIP2 = 3'b110; // local signals reg [20:16] cop1lch; // copper location register 1 reg [15:1] cop1lcl; // copper location register 1 reg [20:16] cop2lch; // copper location register 2 reg [15:1] cop2lcl; // copper location register 2 reg cdang; // copper danger bit reg [15:1] ir1; // instruction register 1 reg [15:0] ir2; // instruction register 2 reg [2:0] copper_state; // current state of copper state machine reg [2:0] copper_next; // next state of copper state machine reg strobe1; // strobe 1 reg strobe2; // strobe 2 reg strobe; reg illegalreg; // illegal register (MOVE instruction) reg skip_flag; // skip move instruction latch reg selins; // load instruction register (register address out = COPINS) reg selreg; // load chip register address, when both selins and selreg are active // a dummy cycle is executed reg skip; // skip next move instruction (input to skip_flag register) wire enable; // enables copper fsm and dma slot reg dma_req; wire dma_ack; wire dma_ena; reg beam_match; // delayed beam match signal wire beam_match_skip; // beam match signal for SKIP condition check reg beam_match_wait; // beam match signal for WAIT condition chaeck wire clk_ena; // enables every other clock cycle for chipset use reg bus_ena; // enables CCK cycle for copper use reg bus_blk; // bus blocked by attempting an access in the "unusable" cycle //-------------------------------------------------------------------------------------- // since Minimig's memory bus runs twice as fast as its real Amiga counterpart // the chipset is required to use every other memory cycle to run virtually at the same speed assign clk_ena = hpos[0]; // horizontal counter in Agnus is advanced 4 lores pixels in comparision with the one in Denise // if the horizontal line contains odd number of CCK cycles (short lines of NTSC mode and all lines of PAL mode) // there is a place where two cycles usable by the copper are located back to back // in such a situation the first cycle is not used (but locks the bus if it has a chance) //write copper location register 1 high and low word always @(posedge clk) if (clk7_en) begin if (reset) cop1lch[20:16] <= 0; else if (reg_address_in[8:1]==COP1LCH[8:1]) cop1lch[20:16] <= data_in[4:0]; end always @(posedge clk) if (clk7_en) begin if (reset) cop1lcl[15:1] <= 0; else if (reg_address_in[8:1]==COP1LCL[8:1]) cop1lcl[15:1] <= data_in[15:1]; end //write copper location register 2 high and low word always @(posedge clk) if (clk7_en) begin if (reset) cop2lch[20:16]<=0; else if (reg_address_in[8:1]==COP2LCH[8:1]) cop2lch[20:16] <= data_in[4:0]; end always @(posedge clk) if (clk7_en) begin if (reset) cop2lcl[15:1] <= 0; else if (reg_address_in[8:1]==COP2LCL[8:1]) cop2lcl[15:1] <= data_in[15:1]; end //write copcon register (copper danger bit) always @(posedge clk) if (clk7_en) begin if (reset) cdang <= 0; else if (reg_address_in[8:1]==COPCON[8:1]) cdang <= data_in[1]; end //copper instruction registers ir1 and ir2 always @(posedge clk) if (clk7_en) begin if (reg_address_in[8:1]==COPINS[8:1]) begin ir1[15:1] <= ir2[15:1]; ir2[15:0] <= data_in[15:0]; end end //-------------------------------------------------------------------------------------- //chip address pointer (or copper program counter) controller always @(posedge clk) if (clk7_en) begin if (dma_ack && strobe1 && copper_state==RESET)//load pointer with location register 1 address_out[20:1] <= {cop1lch[20:16],cop1lcl[15:1]}; else if (dma_ack && strobe2 && copper_state==RESET)//load pointer with location register 2 address_out[20:1] <= {cop2lch[20:16],cop2lcl[15:1]}; else if (dma_ack && (selins || selreg))//increment address pointer (when not dummy cycle) address_out[20:1] <= address_out[20:1] + 1'b1; end //-------------------------------------------------------------------------------------- // regaddress output select // if selins=1 the address of the copper instruction register // is sent out (not strictly necessary as we can load copins directly. However, this is // more according to what happens in a real amiga... I think), else the contents of // ir2[8:1] is selected // (if you ask yourself: IR2? is this a bug? then check how ir1/ir2 are loaded in this design) always @(*) if (enable & selins) //load our instruction register reg_address_out[8:1] = COPINS[8:1]; else if (enable & selreg)//load register in move instruction reg_address_out[8:1] = ir2[8:1]; else reg_address_out[8:1] = 8'hFF;//during dummy cycle null register address is present // detect illegal register access // CDANG = 0 (OCS/ECS) : $080-$1FE allowed // CDANG = 1 (OCS) : $040-$1FE allowed // CDANG = 1 (ECS) : $000-$1FE allowed always @(*) if (ir2[8:7]==2'b00 && !cdang || ir2[8:6]==3'b000 && !ecs) // illegal access illegalreg = 1'b1; else // $080 -> $1FE always allowed illegalreg = 1'b0; //-------------------------------------------------------------------------------------- reg copjmp1, copjmp2; always @(posedge clk) if (clk7_en) begin if (reg_address_in[8:1]==COPJMP1[8:1] || sof) copjmp1 = 1; else if (clk_ena) copjmp1 = 0; end always @(posedge clk) if (clk7_en) begin if (reg_address_in[8:1]==COPJMP2[8:1]) copjmp2 = 1; else if (clk_ena) copjmp2 = 0; end //strobe1 (also triggered by sof, start of frame) always @(posedge clk) if (clk7_en) begin if (copjmp1 && clk_ena) strobe1 = 1; else if (copper_state==RESET && dma_ack) strobe1 = 0; end //strobe2 always @(posedge clk) if (clk7_en) begin if (copjmp2 && clk_ena) strobe2 = 1; else if (copper_state==RESET && dma_ack) strobe2 = 0; end always @(posedge clk) if (clk7_en) begin if (clk_ena) strobe = copjmp1 | copjmp2; end //-------------------------------------------------------------------------------------- //beam compare circuitry //when the mask for a compare bit is 1, the beamcounter is compared with that bit, //when the mask is 0, the compare bit is replaced with the corresponding beamcounter bit //itself, thus the compare is always true. //the blitter busy flag is also checked if blitter finished disable is false wire [8:2] horcmp; wire [7:0] vercmp; //construct compare value for horizontal beam counter (4 lores pixels resolution) assign horcmp[2] = (ir2[1]) ? ir1[1] : hpos[2]; assign horcmp[3] = (ir2[2]) ? ir1[2] : hpos[3]; assign horcmp[4] = (ir2[3]) ? ir1[3] : hpos[4]; assign horcmp[5] = (ir2[4]) ? ir1[4] : hpos[5]; assign horcmp[6] = (ir2[5]) ? ir1[5] : hpos[6]; assign horcmp[7] = (ir2[6]) ? ir1[6] : hpos[7]; assign horcmp[8] = (ir2[7]) ? ir1[7] : hpos[8]; //construct compare value for vertical beam counter (1 line resolution) assign vercmp[0] = (ir2[8]) ? ir1[8] : vpos[0]; assign vercmp[1] = (ir2[9]) ? ir1[9] : vpos[1]; assign vercmp[2] = (ir2[10]) ? ir1[10] : vpos[2]; assign vercmp[3] = (ir2[11]) ? ir1[11] : vpos[3]; assign vercmp[4] = (ir2[12]) ? ir1[12] : vpos[4]; assign vercmp[5] = (ir2[13]) ? ir1[13] : vpos[5]; assign vercmp[6] = (ir2[14]) ? ir1[14] : vpos[6]; assign vercmp[7] = ir1[15]; // actual beam position comparator always @(posedge clk) if (clk7_en) begin if (clk_ena) if ({vpos[7:0],hpos[8:2]} >= {vercmp[7:0],horcmp[8:2]}) beam_match <= 1'b1; else beam_match <= 1'b0; end assign beam_match_skip = beam_match & (ir2[15] | ~blit_busy); always @(posedge clk) if (clk7_en) begin if (clk_ena) beam_match_wait <= beam_match_skip; end //-------------------------------------------------------------------------------------- /* WAIT: first cycle after fetch of second instruction word is a cycle when comparision with beam counter takes place this comparision is beeing done all the time regardless of the available DMA slot when the comparision condition is safisfied the FSM goes to wait_wake_up state, it stays in this state as long as display DMA takes the DMA slots when display DMA doesn't use even bus cycle the FSM advances to the fetch state (the slot isn't used by the copper, DBR is deasserted) such a behaviour is caused by dma request pipelining in real Agnus */ //-------------------------------------------------------------------------------------- //generate dma request signal (reqdma) //copper only uses even cycles: hpos[1:0]==2'b01) //the last cycle of the short line is not usable by the copper //in PAL mode when the copper wants to access memory bus in cycle $E1 the DBR is activated //(blocks the blitter and CPU) but actual transfer takes place in the next cycle (DBR still asserted) always @(posedge clk) if (clk7_en) begin if (clk_ena) if (hpos_slot[8:1]==8'h01) bus_blk <= 1; //cycle $E1 is blocked else bus_blk <= 0; end always @(posedge clk) if (clk7_en) begin if (clk_ena) if (bus_blk) bus_ena <= 1; //cycle $E2 is usable else bus_ena <= ~bus_ena; end assign enable = ~bus_blk & bus_ena & clk_ena; assign reqdma = dma_req & bus_ena & clk_ena; //dma is request also during $E1 but output register address is idle assign dma_ack = ackdma & enable; //dma ack is masked during $E1 assign dma_ena = enadma; //dma slot is empty and can be used by copper //hint: during vblank copper instruction pointer is reloaded just after the first refresh slot //there is at least 2 CCK delay between writing COPJMPx register and pointer reload //copper state machine and skip_flag latch always @(posedge clk) if (clk7_en) begin if (reset || clk_ena && strobe) // on strobe or reset fetch first instruction word copper_state <= RESET; else if (enable) // go to next state copper_state <= copper_next; end always @(posedge clk) if (clk7_en) begin if (enable) skip_flag <= skip; end always @(*)//(copper_state or ir2 or beam_match_wait or beam_match_skip or illegalreg or skip_flag or dma_ack or dma_ena) begin case (copper_state) //when COPJMPx is written there is 2 cycle delay before data from new location is read to COPINS //usually first cycle is a read of the next instruction to COPINS or bitplane DMA, //the second is dma free cycle (it's a dummy cycle requested by copper but not used to transfer data) //after reset or strobe write an allocated DMA cycle is required to reload instruction pointer from location registers RESET: begin skip = 0; selins = 0; selreg = 0; dma_req = 1; //a DMA access is requested to reload instuction pointer if (dma_ack) copper_next = FETCH1; else copper_next = RESET; end //fetch first instruction word FETCH1: begin skip = skip_flag; selins = 1; selreg = 0; dma_req = 1; if (dma_ack) copper_next = FETCH2; else copper_next = FETCH1; end //fetch second instruction word, skip or do MOVE instruction or halt copper FETCH2: begin if (!ir2[0] && illegalreg) // illegal MOVE instruction, halt copper begin skip = 0; selins = 0; selreg = 0; dma_req = 0; copper_next = FETCH2; end else if (!ir2[0] && skip_flag) // skip this MOVE instruction begin selins = 1; selreg = 0; dma_req = 1; if (dma_ack) begin skip = 0; copper_next = FETCH1; end else begin skip = 1; copper_next = FETCH2; end end else if (!ir2[0]) // MOVE instruction begin skip = 0; selins = 0; selreg = 1; dma_req = 1; if (dma_ack) copper_next = FETCH1; else copper_next = FETCH2; end else//fetch second instruction word of WAIT or SKIP instruction begin skip = 0; selins = 1; selreg = 0; dma_req = 1; if (dma_ack) copper_next = WAITSKIP1; else copper_next = FETCH2; end end //both SKIP and WAIT have the same timing when WAIT is immediatelly complete //both these instructions complete in 4 cycles and these cycles must be allocated dma cycles //first cycle seems to be dummy WAITSKIP1: begin skip = 0; selins = 0; selreg = 0; dma_req = 0; if (dma_ena) copper_next = WAITSKIP2; else copper_next = WAITSKIP1; end //second cycle of WAIT or SKIP (allocated dma) //WAIT or SKIP instruction WAITSKIP2: begin if (!ir2[0]) // WAIT instruction begin if (beam_match_wait) // wait is over, fetch next instruction begin skip = 0; selins = 0; selreg = 0; dma_req = 0; if (dma_ena) copper_next = FETCH1; else copper_next = WAITSKIP2; end else//still waiting begin skip = 0; selins = 0; selreg = 0; dma_req = 0; copper_next = WAITSKIP2; end end else // SKIP instruction begin if (beam_match_skip) // compare is true, fetch next instruction and skip it if it's MOVE begin skip = 1; selins = 0; selreg = 0; dma_req = 0; if (dma_ena) copper_next = FETCH1; else copper_next = WAITSKIP2; end else//do not skip, fetch next instruction begin skip = 0; selins = 0; selreg = 0; dma_req = 0; if (dma_ena) copper_next = FETCH1; else copper_next = WAITSKIP2; end end end //default, go back to reset state default: begin skip = 0; selins = 0; selreg = 0; dma_req = 0; copper_next = FETCH1; end endcase end //-------------------------------------------------------------------------------------- endmodule