| 1 |
2 |
dgisselq |
///////////////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//
|
| 3 |
|
|
// Filename: zipcpu.v
|
| 4 |
|
|
//
|
| 5 |
|
|
// Project: Zip CPU -- a small, lightweight, RISC CPU soft core
|
| 6 |
|
|
//
|
| 7 |
|
|
// Purpose: This is the top level module holding the core of the Zip CPU
|
| 8 |
|
|
// together. The Zip CPU is designed to be as simple as possible.
|
| 9 |
|
|
// (actual implementation aside ...) The instruction set is about as
|
| 10 |
|
|
// RISC as you can get, there are only 16 instruction types supported.
|
| 11 |
|
|
// Please see the accompanying spec.pdf file for a description of these
|
| 12 |
|
|
// instructions.
|
| 13 |
|
|
//
|
| 14 |
|
|
// All instructions are 32-bits wide. All bus accesses, both address and
|
| 15 |
|
|
// data, are 32-bits over a wishbone bus.
|
| 16 |
|
|
//
|
| 17 |
|
|
// The Zip CPU is fully pipelined with the following pipeline stages:
|
| 18 |
|
|
//
|
| 19 |
|
|
// 1. Prefetch, returns the instruction from memory.
|
| 20 |
|
|
//
|
| 21 |
|
|
// 2. Instruction Decode
|
| 22 |
|
|
//
|
| 23 |
|
|
// 3. Read Operands
|
| 24 |
|
|
//
|
| 25 |
|
|
// 4. Apply Instruction
|
| 26 |
|
|
//
|
| 27 |
|
|
// 4. Write-back Results
|
| 28 |
|
|
//
|
| 29 |
|
|
// Further information about the inner workings of this CPU may be
|
| 30 |
|
|
// found in the spec.pdf file. (The documentation within this file
|
| 31 |
|
|
// had become out of date and out of sync with the spec.pdf, so look
|
| 32 |
|
|
// to the spec.pdf for accurate and up to date information.)
|
| 33 |
|
|
//
|
| 34 |
|
|
//
|
| 35 |
|
|
// In general, the pipelining is controlled by three pieces of logic
|
| 36 |
|
|
// per stage: _ce, _stall, and _valid. _valid means that the stage
|
| 37 |
|
|
// holds a valid instruction. _ce means that the instruction from the
|
| 38 |
|
|
// previous stage is to move into this one, and _stall means that the
|
| 39 |
|
|
// instruction from the previous stage may not move into this one.
|
| 40 |
|
|
// The difference between these control signals allows individual stages
|
| 41 |
|
|
// to propagate instructions independently. In general, the logic works
|
| 42 |
|
|
// as:
|
| 43 |
|
|
//
|
| 44 |
|
|
//
|
| 45 |
|
|
// assign (n)_ce = (n-1)_valid && (~(n)_stall)
|
| 46 |
|
|
//
|
| 47 |
|
|
//
|
| 48 |
|
|
// always @(posedge i_clk)
|
| 49 |
|
|
// if ((i_rst)||(clear_pipeline))
|
| 50 |
|
|
// (n)_valid = 0
|
| 51 |
|
|
// else if (n)_ce
|
| 52 |
|
|
// (n)_valid = 1
|
| 53 |
|
|
// else if (n+1)_ce
|
| 54 |
|
|
// (n)_valid = 0
|
| 55 |
|
|
//
|
| 56 |
|
|
// assign (n)_stall = ( (n-1)_valid && ( pipeline hazard detection ) )
|
| 57 |
|
|
// || ( (n)_valid && (n+1)_stall );
|
| 58 |
|
|
//
|
| 59 |
|
|
// and ...
|
| 60 |
|
|
//
|
| 61 |
|
|
// always @(posedge i_clk)
|
| 62 |
|
|
// if (n)_ce
|
| 63 |
|
|
// (n)_variable = ... whatever logic for this stage
|
| 64 |
|
|
//
|
| 65 |
|
|
// Note that a stage can stall even if no instruction is loaded into
|
| 66 |
|
|
// it.
|
| 67 |
|
|
//
|
| 68 |
|
|
//
|
| 69 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
| 70 |
|
|
// Gisselquist Technology, LLC
|
| 71 |
|
|
//
|
| 72 |
|
|
///////////////////////////////////////////////////////////////////////////////
|
| 73 |
|
|
//
|
| 74 |
|
|
// Copyright (C) 2015, Gisselquist Technology, LLC
|
| 75 |
|
|
//
|
| 76 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
| 77 |
|
|
// modify it under the terms of the GNU General Public License as published
|
| 78 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
| 79 |
|
|
// your option) any later version.
|
| 80 |
|
|
//
|
| 81 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
| 82 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
| 83 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 84 |
|
|
// for more details.
|
| 85 |
|
|
//
|
| 86 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
| 87 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
| 88 |
|
|
//
|
| 89 |
|
|
//
|
| 90 |
|
|
///////////////////////////////////////////////////////////////////////////////
|
| 91 |
|
|
//
|
| 92 |
|
|
// We can either pipeline our fetches, or issue one fetch at a time. Pipelined
|
| 93 |
|
|
// fetches are more complicated and therefore use more FPGA resources, while
|
| 94 |
|
|
// single fetches will cause the CPU to stall for about 5 stalls each
|
| 95 |
|
|
// instruction cycle, effectively reducing the instruction count per clock to
|
| 96 |
|
|
// about 0.2. However, the area cost may be worth it. Consider:
|
| 97 |
|
|
//
|
| 98 |
|
|
// Slice LUTs ZipSystem ZipCPU
|
| 99 |
|
|
// Single Fetching 2521 1734
|
| 100 |
|
|
// Pipelined fetching 2796 2046
|
| 101 |
|
|
//
|
| 102 |
|
|
//
|
| 103 |
|
|
//
|
| 104 |
|
|
`define CPU_CC_REG 4'he
|
| 105 |
|
|
`define CPU_PC_REG 4'hf
|
| 106 |
|
|
`define CPU_FPUERR_BIT 12 // Floating point error flag, set on error
|
| 107 |
|
|
`define CPU_DIVERR_BIT 11 // Divide error flag, set on divide by zero
|
| 108 |
|
|
`define CPU_BUSERR_BIT 10 // Bus error flag, set on error
|
| 109 |
|
|
`define CPU_TRAP_BIT 9 // User TRAP has taken place
|
| 110 |
|
|
`define CPU_ILL_BIT 8 // Illegal instruction
|
| 111 |
|
|
`define CPU_BREAK_BIT 7
|
| 112 |
|
|
`define CPU_STEP_BIT 6 // Will step one or two (VLIW) instructions
|
| 113 |
|
|
`define CPU_GIE_BIT 5
|
| 114 |
|
|
`define CPU_SLEEP_BIT 4
|
| 115 |
|
|
// Compile time defines
|
| 116 |
|
|
//
|
| 117 |
|
|
`include "cpudefs.v"
|
| 118 |
|
|
//
|
| 119 |
|
|
//
|
| 120 |
|
|
module zipcpu(i_clk, i_rst, i_interrupt,
|
| 121 |
|
|
// Debug interface
|
| 122 |
|
|
i_halt, i_clear_pf_cache, i_dbg_reg, i_dbg_we, i_dbg_data,
|
| 123 |
|
|
o_dbg_stall, o_dbg_reg, o_dbg_cc,
|
| 124 |
|
|
o_break,
|
| 125 |
|
|
// CPU interface to the wishbone bus
|
| 126 |
|
|
o_wb_gbl_cyc, o_wb_gbl_stb,
|
| 127 |
|
|
o_wb_lcl_cyc, o_wb_lcl_stb,
|
| 128 |
|
|
o_wb_we, o_wb_addr, o_wb_data,
|
| 129 |
|
|
i_wb_ack, i_wb_stall, i_wb_data,
|
| 130 |
|
|
i_wb_err,
|
| 131 |
|
|
// Accounting/CPU usage interface
|
| 132 |
|
|
o_op_stall, o_pf_stall, o_i_count
|
| 133 |
|
|
`ifdef DEBUG_SCOPE
|
| 134 |
|
|
, o_debug
|
| 135 |
|
|
`endif
|
| 136 |
|
|
);
|
| 137 |
|
|
parameter RESET_ADDRESS=32'h0100000, ADDRESS_WIDTH=24,
|
| 138 |
|
|
LGICACHE=6;
|
| 139 |
|
|
`ifdef OPT_MULTIPLY
|
| 140 |
|
|
parameter IMPLEMENT_MPY = 1;
|
| 141 |
|
|
`else
|
| 142 |
|
|
parameter IMPLEMENT_MPY = 0;
|
| 143 |
|
|
`endif
|
| 144 |
|
|
`ifdef OPT_DIVIDE
|
| 145 |
|
|
parameter IMPLEMENT_DIVIDE = 1;
|
| 146 |
|
|
`else
|
| 147 |
|
|
parameter IMPLEMENT_DIVIDE = 0;
|
| 148 |
|
|
`endif
|
| 149 |
|
|
`ifdef OPT_IMPLEMENT_FPU
|
| 150 |
|
|
parameter IMPLEMENT_FPU = 1,
|
| 151 |
|
|
`else
|
| 152 |
|
|
parameter IMPLEMENT_FPU = 0,
|
| 153 |
|
|
`endif
|
| 154 |
|
|
IMPLEMENT_LOCK=1;
|
| 155 |
|
|
`ifdef OPT_EARLY_BRANCHING
|
| 156 |
|
|
parameter EARLY_BRANCHING = 1;
|
| 157 |
|
|
`else
|
| 158 |
|
|
parameter EARLY_BRANCHING = 0;
|
| 159 |
|
|
`endif
|
| 160 |
|
|
parameter AW=ADDRESS_WIDTH;
|
| 161 |
|
|
input i_clk, i_rst, i_interrupt;
|
| 162 |
|
|
// Debug interface -- inputs
|
| 163 |
|
|
input i_halt, i_clear_pf_cache;
|
| 164 |
|
|
input [4:0] i_dbg_reg;
|
| 165 |
|
|
input i_dbg_we;
|
| 166 |
|
|
input [31:0] i_dbg_data;
|
| 167 |
|
|
// Debug interface -- outputs
|
| 168 |
|
|
output reg o_dbg_stall;
|
| 169 |
|
|
output reg [31:0] o_dbg_reg;
|
| 170 |
|
|
output reg [3:0] o_dbg_cc;
|
| 171 |
|
|
output wire o_break;
|
| 172 |
|
|
// Wishbone interface -- outputs
|
| 173 |
|
|
output wire o_wb_gbl_cyc, o_wb_gbl_stb;
|
| 174 |
|
|
output wire o_wb_lcl_cyc, o_wb_lcl_stb, o_wb_we;
|
| 175 |
|
|
output wire [(AW-1):0] o_wb_addr;
|
| 176 |
|
|
output wire [31:0] o_wb_data;
|
| 177 |
|
|
// Wishbone interface -- inputs
|
| 178 |
|
|
input i_wb_ack, i_wb_stall;
|
| 179 |
|
|
input [31:0] i_wb_data;
|
| 180 |
|
|
input i_wb_err;
|
| 181 |
|
|
// Accounting outputs ... to help us count stalls and usage
|
| 182 |
|
|
output wire o_op_stall;
|
| 183 |
|
|
output wire o_pf_stall;
|
| 184 |
|
|
output wire o_i_count;
|
| 185 |
|
|
//
|
| 186 |
|
|
`ifdef DEBUG_SCOPE
|
| 187 |
|
|
output reg [31:0] o_debug;
|
| 188 |
|
|
`endif
|
| 189 |
|
|
|
| 190 |
|
|
|
| 191 |
|
|
// Registers
|
| 192 |
|
|
//
|
| 193 |
|
|
// The distributed RAM style comment is necessary on the
|
| 194 |
|
|
// SPARTAN6 with XST to prevent XST from oversimplifying the register
|
| 195 |
|
|
// set and in the process ruining everything else. It basically
|
| 196 |
|
|
// optimizes logic away, to where it no longer works. The logic
|
| 197 |
|
|
// as described herein will work, this just makes sure XST implements
|
| 198 |
|
|
// that logic.
|
| 199 |
|
|
//
|
| 200 |
|
|
(* ram_style = "distributed" *)
|
| 201 |
|
|
reg [31:0] regset [0:31];
|
| 202 |
|
|
|
| 203 |
|
|
// Condition codes
|
| 204 |
|
|
// (BUS, TRAP,ILL,BREAKEN,STEP,GIE,SLEEP ), V, N, C, Z
|
| 205 |
|
|
reg [3:0] flags, iflags;
|
| 206 |
|
|
wire [13:0] w_uflags, w_iflags;
|
| 207 |
|
|
reg trap, break_en, step, gie, sleep;
|
| 208 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 209 |
|
|
reg ill_err_u, ill_err_i;
|
| 210 |
|
|
`else
|
| 211 |
|
|
wire ill_err_u, ill_err_i;
|
| 212 |
|
|
`endif
|
| 213 |
|
|
reg ibus_err_flag, ubus_err_flag;
|
| 214 |
|
|
wire idiv_err_flag, udiv_err_flag;
|
| 215 |
|
|
wire ifpu_err_flag, ufpu_err_flag;
|
| 216 |
|
|
wire ihalt_phase, uhalt_phase;
|
| 217 |
|
|
|
| 218 |
|
|
// The master chip enable
|
| 219 |
|
|
wire master_ce;
|
| 220 |
|
|
|
| 221 |
|
|
//
|
| 222 |
|
|
//
|
| 223 |
|
|
// PIPELINE STAGE #1 :: Prefetch
|
| 224 |
|
|
// Variable declarations
|
| 225 |
|
|
//
|
| 226 |
|
|
reg [(AW-1):0] pf_pc;
|
| 227 |
|
|
reg new_pc;
|
| 228 |
|
|
wire clear_pipeline;
|
| 229 |
|
|
assign clear_pipeline = new_pc || i_clear_pf_cache;
|
| 230 |
|
|
|
| 231 |
|
|
wire dcd_stalled;
|
| 232 |
|
|
wire pf_cyc, pf_stb, pf_we, pf_busy, pf_ack, pf_stall, pf_err;
|
| 233 |
|
|
wire [(AW-1):0] pf_addr;
|
| 234 |
|
|
wire [31:0] pf_data;
|
| 235 |
|
|
wire [31:0] instruction;
|
| 236 |
|
|
wire [(AW-1):0] instruction_pc;
|
| 237 |
|
|
wire pf_valid, instruction_gie, pf_illegal;
|
| 238 |
|
|
|
| 239 |
|
|
//
|
| 240 |
|
|
//
|
| 241 |
|
|
// PIPELINE STAGE #2 :: Instruction Decode
|
| 242 |
|
|
// Variable declarations
|
| 243 |
|
|
//
|
| 244 |
|
|
//
|
| 245 |
|
|
reg opvalid, opvalid_mem, opvalid_alu;
|
| 246 |
|
|
reg opvalid_div, opvalid_fpu;
|
| 247 |
|
|
wire op_stall, dcd_ce, dcd_phase;
|
| 248 |
|
|
wire [3:0] dcdOp;
|
| 249 |
|
|
wire [4:0] dcdA, dcdB, dcdR;
|
| 250 |
|
|
wire dcdA_cc, dcdB_cc, dcdA_pc, dcdB_pc, dcdR_cc, dcdR_pc;
|
| 251 |
|
|
wire [3:0] dcdF;
|
| 252 |
|
|
wire dcdR_wr, dcdA_rd, dcdB_rd,
|
| 253 |
|
|
dcdALU, dcdM, dcdDV, dcdFP,
|
| 254 |
|
|
dcdF_wr, dcd_gie, dcd_break, dcd_lock,
|
| 255 |
|
|
dcd_pipe, dcd_ljmp;
|
| 256 |
|
|
reg r_dcdvalid;
|
| 257 |
|
|
wire dcdvalid;
|
| 258 |
|
|
wire [(AW-1):0] dcd_pc;
|
| 259 |
|
|
wire [31:0] dcdI;
|
| 260 |
|
|
wire dcd_zI; // true if dcdI == 0
|
| 261 |
|
|
wire dcdA_stall, dcdB_stall, dcdF_stall;
|
| 262 |
|
|
|
| 263 |
|
|
wire dcd_illegal;
|
| 264 |
|
|
wire dcd_early_branch;
|
| 265 |
|
|
wire [(AW-1):0] dcd_branch_pc;
|
| 266 |
|
|
|
| 267 |
|
|
|
| 268 |
|
|
//
|
| 269 |
|
|
//
|
| 270 |
|
|
// PIPELINE STAGE #3 :: Read Operands
|
| 271 |
|
|
// Variable declarations
|
| 272 |
|
|
//
|
| 273 |
|
|
//
|
| 274 |
|
|
//
|
| 275 |
|
|
// Now, let's read our operands
|
| 276 |
|
|
reg [4:0] alu_reg;
|
| 277 |
|
|
reg [3:0] opn;
|
| 278 |
|
|
reg [4:0] opR;
|
| 279 |
|
|
reg [31:0] r_opA, r_opB;
|
| 280 |
|
|
reg [(AW-1):0] op_pc;
|
| 281 |
|
|
wire [31:0] w_opA, w_opB;
|
| 282 |
|
|
wire [31:0] opA_nowait, opB_nowait, opA, opB;
|
| 283 |
|
|
reg opR_wr, opR_cc, opF_wr, op_gie;
|
| 284 |
|
|
wire [13:0] opFl;
|
| 285 |
|
|
reg [5:0] r_opF;
|
| 286 |
|
|
wire [7:0] opF;
|
| 287 |
|
|
wire op_ce, op_phase;
|
| 288 |
|
|
// Some pipeline control wires
|
| 289 |
|
|
`ifdef OPT_PIPELINED
|
| 290 |
|
|
reg opA_alu, opA_mem;
|
| 291 |
|
|
reg opB_alu, opB_mem;
|
| 292 |
|
|
`endif
|
| 293 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 294 |
|
|
reg op_illegal;
|
| 295 |
|
|
`endif
|
| 296 |
|
|
reg op_break;
|
| 297 |
|
|
wire op_lock;
|
| 298 |
|
|
|
| 299 |
|
|
|
| 300 |
|
|
//
|
| 301 |
|
|
//
|
| 302 |
|
|
// PIPELINE STAGE #4 :: ALU / Memory
|
| 303 |
|
|
// Variable declarations
|
| 304 |
|
|
//
|
| 305 |
|
|
//
|
| 306 |
|
|
reg [(AW-1):0] alu_pc;
|
| 307 |
|
|
reg alu_pc_valid;
|
| 308 |
|
|
wire alu_phase;
|
| 309 |
|
|
wire alu_ce, alu_stall;
|
| 310 |
|
|
wire [31:0] alu_result;
|
| 311 |
|
|
wire [3:0] alu_flags;
|
| 312 |
|
|
wire alu_valid, alu_busy;
|
| 313 |
|
|
wire set_cond;
|
| 314 |
|
|
reg alu_wr, alF_wr, alu_gie;
|
| 315 |
|
|
wire alu_illegal_op;
|
| 316 |
|
|
wire alu_illegal;
|
| 317 |
|
|
|
| 318 |
|
|
|
| 319 |
|
|
|
| 320 |
|
|
wire mem_ce, mem_stalled;
|
| 321 |
|
|
`ifdef OPT_PIPELINED_BUS_ACCESS
|
| 322 |
|
|
wire mem_pipe_stalled;
|
| 323 |
|
|
`endif
|
| 324 |
|
|
wire mem_valid, mem_ack, mem_stall, mem_err, bus_err,
|
| 325 |
|
|
mem_cyc_gbl, mem_cyc_lcl, mem_stb_gbl, mem_stb_lcl, mem_we;
|
| 326 |
|
|
wire [4:0] mem_wreg;
|
| 327 |
|
|
|
| 328 |
|
|
wire mem_busy, mem_rdbusy;
|
| 329 |
|
|
wire [(AW-1):0] mem_addr;
|
| 330 |
|
|
wire [31:0] mem_data, mem_result;
|
| 331 |
|
|
|
| 332 |
|
|
wire div_ce, div_error, div_busy, div_valid;
|
| 333 |
|
|
wire [31:0] div_result;
|
| 334 |
|
|
wire [3:0] div_flags;
|
| 335 |
|
|
|
| 336 |
|
|
assign div_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_div)
|
| 337 |
|
|
&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
|
| 338 |
|
|
&&(set_cond);
|
| 339 |
|
|
|
| 340 |
|
|
wire fpu_ce, fpu_error, fpu_busy, fpu_valid;
|
| 341 |
|
|
wire [31:0] fpu_result;
|
| 342 |
|
|
wire [3:0] fpu_flags;
|
| 343 |
|
|
|
| 344 |
|
|
assign fpu_ce = (master_ce)&&(~clear_pipeline)&&(opvalid_fpu)
|
| 345 |
|
|
&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
|
| 346 |
|
|
&&(set_cond);
|
| 347 |
|
|
|
| 348 |
|
|
|
| 349 |
|
|
//
|
| 350 |
|
|
//
|
| 351 |
|
|
// PIPELINE STAGE #5 :: Write-back
|
| 352 |
|
|
// Variable declarations
|
| 353 |
|
|
//
|
| 354 |
|
|
wire wr_reg_ce, wr_flags_ce, wr_write_pc, wr_write_cc;
|
| 355 |
|
|
wire [4:0] wr_reg_id;
|
| 356 |
|
|
wire [31:0] wr_reg_vl;
|
| 357 |
|
|
wire w_switch_to_interrupt, w_release_from_interrupt;
|
| 358 |
|
|
reg [(AW-1):0] upc, ipc;
|
| 359 |
|
|
|
| 360 |
|
|
|
| 361 |
|
|
|
| 362 |
|
|
//
|
| 363 |
|
|
// MASTER: clock enable.
|
| 364 |
|
|
//
|
| 365 |
|
|
assign master_ce = (~i_halt)&&(~o_break)&&(~sleep);
|
| 366 |
|
|
|
| 367 |
|
|
|
| 368 |
|
|
//
|
| 369 |
|
|
// PIPELINE STAGE #1 :: Prefetch
|
| 370 |
|
|
// Calculate stall conditions
|
| 371 |
|
|
//
|
| 372 |
|
|
// These are calculated externally, within the prefetch module.
|
| 373 |
|
|
//
|
| 374 |
|
|
|
| 375 |
|
|
//
|
| 376 |
|
|
// PIPELINE STAGE #2 :: Instruction Decode
|
| 377 |
|
|
// Calculate stall conditions
|
| 378 |
|
|
`ifdef OPT_PIPELINED
|
| 379 |
|
|
assign dcd_ce = ((~dcdvalid)||(~dcd_stalled))&&(~clear_pipeline);
|
| 380 |
|
|
`else
|
| 381 |
|
|
assign dcd_ce = 1'b1;
|
| 382 |
|
|
`endif
|
| 383 |
|
|
`ifdef OPT_PIPELINED
|
| 384 |
|
|
assign dcd_stalled = (dcdvalid)&&(op_stall);
|
| 385 |
|
|
`else
|
| 386 |
|
|
// If not pipelined, there will be no opvalid_ anything, and the
|
| 387 |
|
|
// op_stall will be false, dcdX_stall will be false, thus we can simply
|
| 388 |
|
|
// do a ...
|
| 389 |
|
|
assign dcd_stalled = 1'b0;
|
| 390 |
|
|
`endif
|
| 391 |
|
|
//
|
| 392 |
|
|
// PIPELINE STAGE #3 :: Read Operands
|
| 393 |
|
|
// Calculate stall conditions
|
| 394 |
|
|
wire op_lock_stall;
|
| 395 |
|
|
`ifdef OPT_PIPELINED
|
| 396 |
|
|
assign op_stall = (opvalid)&&( // Only stall if we're loaded w/validins
|
| 397 |
|
|
// Stall if we're stopped, and not allowed to execute
|
| 398 |
|
|
// an instruction
|
| 399 |
|
|
// (~master_ce) // Already captured in alu_stall
|
| 400 |
|
|
//
|
| 401 |
|
|
// Stall if going into the ALU and the ALU is stalled
|
| 402 |
|
|
// i.e. if the memory is busy, or we are single
|
| 403 |
|
|
// stepping. This also includes our stalls for
|
| 404 |
|
|
// op_break and op_lock, so we don't need to
|
| 405 |
|
|
// include those as well here.
|
| 406 |
|
|
// This also includes whether or not the divide or
|
| 407 |
|
|
// floating point units are busy.
|
| 408 |
|
|
(alu_stall)
|
| 409 |
|
|
//
|
| 410 |
|
|
// Stall if we are going into memory with an operation
|
| 411 |
|
|
// that cannot be pipelined, and the memory is
|
| 412 |
|
|
// already busy
|
| 413 |
|
|
||(mem_stalled) // &&(opvalid_mem) part of mem_stalled
|
| 414 |
|
|
)
|
| 415 |
|
|
||(dcdvalid)&&(
|
| 416 |
|
|
// Stall if we need to wait for an operand A
|
| 417 |
|
|
// to be ready to read
|
| 418 |
|
|
(dcdA_stall)
|
| 419 |
|
|
// Likewise for B, also includes logic
|
| 420 |
|
|
// regarding immediate offset (register must
|
| 421 |
|
|
// be in register file if we need to add to
|
| 422 |
|
|
// an immediate)
|
| 423 |
|
|
||(dcdB_stall)
|
| 424 |
|
|
// Or if we need to wait on flags to work on the
|
| 425 |
|
|
// CC register
|
| 426 |
|
|
||(dcdF_stall)
|
| 427 |
|
|
);
|
| 428 |
|
|
assign op_ce = ((dcdvalid)||(dcd_illegal))&&(~op_stall)&&(~clear_pipeline);
|
| 429 |
|
|
`else
|
| 430 |
|
|
assign op_stall = (opvalid)&&(~master_ce);
|
| 431 |
|
|
assign op_ce = ((dcdvalid)||(dcd_illegal));
|
| 432 |
|
|
`endif
|
| 433 |
|
|
|
| 434 |
|
|
//
|
| 435 |
|
|
// PIPELINE STAGE #4 :: ALU / Memory
|
| 436 |
|
|
// Calculate stall conditions
|
| 437 |
|
|
//
|
| 438 |
|
|
// 1. Basic stall is if the previous stage is valid and the next is
|
| 439 |
|
|
// busy.
|
| 440 |
|
|
// 2. Also stall if the prior stage is valid and the master clock enable
|
| 441 |
|
|
// is de-selected
|
| 442 |
|
|
// 3. Stall if someone on the other end is writing the CC register,
|
| 443 |
|
|
// since we don't know if it'll put us to sleep or not.
|
| 444 |
|
|
// 4. Last case: Stall if we would otherwise move a break instruction
|
| 445 |
|
|
// through the ALU. Break instructions are not allowed through
|
| 446 |
|
|
// the ALU.
|
| 447 |
|
|
`ifdef OPT_PIPELINED
|
| 448 |
|
|
assign alu_stall = (((~master_ce)||(mem_rdbusy)||(alu_busy))&&(opvalid_alu)) //Case 1&2
|
| 449 |
|
|
// Old case #3--this isn't an ALU stall though ...
|
| 450 |
|
|
||((opvalid_alu)&&(wr_reg_ce)&&(wr_reg_id[4] == op_gie)
|
| 451 |
|
|
&&(wr_write_cc)) // Case 3
|
| 452 |
|
|
||((opvalid)&&(op_lock)&&(op_lock_stall))
|
| 453 |
|
|
||((opvalid)&&(op_break))
|
| 454 |
|
|
||(div_busy)||(fpu_busy);
|
| 455 |
|
|
assign alu_ce = (master_ce)&&((opvalid_alu)||(op_illegal))
|
| 456 |
|
|
&&(~alu_stall)
|
| 457 |
|
|
&&(~clear_pipeline);
|
| 458 |
|
|
`else
|
| 459 |
|
|
assign alu_stall = ((~master_ce)&&(opvalid_alu))
|
| 460 |
|
|
||((opvalid_alu)&&(op_break));
|
| 461 |
|
|
assign alu_ce = (master_ce)&&((opvalid_alu)||(op_illegal))&&(~alu_stall);
|
| 462 |
|
|
`endif
|
| 463 |
|
|
//
|
| 464 |
|
|
|
| 465 |
|
|
//
|
| 466 |
|
|
// Note: if you change the conditions for mem_ce, you must also change
|
| 467 |
|
|
// alu_pc_valid.
|
| 468 |
|
|
//
|
| 469 |
|
|
`ifdef OPT_PIPELINED
|
| 470 |
|
|
assign mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled)
|
| 471 |
|
|
&&(~clear_pipeline);
|
| 472 |
|
|
`else
|
| 473 |
|
|
// If we aren't pipelined, then no one will be changing what's in the
|
| 474 |
|
|
// pipeline (i.e. clear_pipeline), while our only instruction goes
|
| 475 |
|
|
// through the ... pipeline.
|
| 476 |
|
|
assign mem_ce = (master_ce)&&(opvalid_mem)&&(~mem_stalled);
|
| 477 |
|
|
`endif
|
| 478 |
|
|
`ifdef OPT_PIPELINED_BUS_ACCESS
|
| 479 |
|
|
assign mem_stalled = (~master_ce)||(alu_busy)||((opvalid_mem)&&(
|
| 480 |
|
|
(mem_pipe_stalled)
|
| 481 |
|
|
||((~op_pipe)&&(mem_busy))
|
| 482 |
|
|
||(div_busy)
|
| 483 |
|
|
||(fpu_busy)
|
| 484 |
|
|
// Stall waiting for flags to be valid
|
| 485 |
|
|
// Or waiting for a write to the PC register
|
| 486 |
|
|
// Or CC register, since that can change the
|
| 487 |
|
|
// PC as well
|
| 488 |
|
|
||((wr_reg_ce)&&(wr_reg_id[4] == op_gie)
|
| 489 |
|
|
&&((wr_write_pc)||(wr_write_cc)))));
|
| 490 |
|
|
`else
|
| 491 |
|
|
`ifdef OPT_PIPELINED
|
| 492 |
|
|
assign mem_stalled = (mem_busy)||((opvalid_mem)&&(
|
| 493 |
|
|
(~master_ce)
|
| 494 |
|
|
// Stall waiting for flags to be valid
|
| 495 |
|
|
// Or waiting for a write to the PC register
|
| 496 |
|
|
// Or CC register, since that can change the
|
| 497 |
|
|
// PC as well
|
| 498 |
|
|
||((wr_reg_ce)&&(wr_reg_id[4] == op_gie)&&((wr_write_pc)||(wr_write_cc)))));
|
| 499 |
|
|
`else
|
| 500 |
|
|
assign mem_stalled = (opvalid_mem)&&(~master_ce);
|
| 501 |
|
|
`endif
|
| 502 |
|
|
`endif
|
| 503 |
|
|
|
| 504 |
|
|
|
| 505 |
|
|
//
|
| 506 |
|
|
//
|
| 507 |
|
|
// PIPELINE STAGE #1 :: Prefetch
|
| 508 |
|
|
//
|
| 509 |
|
|
//
|
| 510 |
|
|
`ifdef OPT_SINGLE_FETCH
|
| 511 |
|
|
wire pf_ce;
|
| 512 |
|
|
|
| 513 |
|
|
assign pf_ce = (~pf_valid)&&(~dcdvalid)&&(~opvalid)&&(~alu_valid);
|
| 514 |
|
|
prefetch #(ADDRESS_WIDTH)
|
| 515 |
|
|
pf(i_clk, i_rst, (pf_ce), (~dcd_stalled), pf_pc, gie,
|
| 516 |
|
|
instruction, instruction_pc, instruction_gie,
|
| 517 |
|
|
pf_valid, pf_illegal,
|
| 518 |
|
|
pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
|
| 519 |
|
|
pf_ack, pf_stall, pf_err, i_wb_data);
|
| 520 |
|
|
|
| 521 |
|
|
initial r_dcdvalid = 1'b0;
|
| 522 |
|
|
always @(posedge i_clk)
|
| 523 |
|
|
if (i_rst)
|
| 524 |
|
|
r_dcdvalid <= 1'b0;
|
| 525 |
|
|
else if (dcd_ce)
|
| 526 |
|
|
r_dcdvalid <= (pf_valid);
|
| 527 |
|
|
else if (op_ce)
|
| 528 |
|
|
r_dcdvalid <= 1'b0;
|
| 529 |
|
|
assign dcdvalid = r_dcdvalid;
|
| 530 |
|
|
|
| 531 |
|
|
`else // Pipe fetch
|
| 532 |
|
|
|
| 533 |
|
|
`ifdef OPT_TRADITIONAL_PFCACHE
|
| 534 |
|
|
pfcache #(LGICACHE, ADDRESS_WIDTH)
|
| 535 |
|
|
pf(i_clk, i_rst, (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
|
| 536 |
|
|
i_clear_pf_cache,
|
| 537 |
|
|
// dcd_pc,
|
| 538 |
|
|
~dcd_stalled,
|
| 539 |
|
|
((dcd_early_branch)&&(~clear_pipeline))
|
| 540 |
|
|
? dcd_branch_pc:pf_pc,
|
| 541 |
|
|
instruction, instruction_pc, pf_valid,
|
| 542 |
|
|
pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
|
| 543 |
|
|
pf_ack, pf_stall, pf_err, i_wb_data,
|
| 544 |
|
|
pf_illegal);
|
| 545 |
|
|
`else
|
| 546 |
|
|
pipefetch #(RESET_ADDRESS, LGICACHE, ADDRESS_WIDTH)
|
| 547 |
|
|
pf(i_clk, i_rst, (new_pc)||((dcd_early_branch)&&(~clear_pipeline)),
|
| 548 |
|
|
i_clear_pf_cache, ~dcd_stalled,
|
| 549 |
|
|
(new_pc)?pf_pc:dcd_branch_pc,
|
| 550 |
|
|
instruction, instruction_pc, pf_valid,
|
| 551 |
|
|
pf_cyc, pf_stb, pf_we, pf_addr, pf_data,
|
| 552 |
|
|
pf_ack, pf_stall, pf_err, i_wb_data,
|
| 553 |
|
|
//`ifdef OPT_PRECLEAR_BUS
|
| 554 |
|
|
//((dcd_clear_bus)&&(dcdvalid))
|
| 555 |
|
|
//||((op_clear_bus)&&(opvalid))
|
| 556 |
|
|
//||
|
| 557 |
|
|
//`endif
|
| 558 |
|
|
(mem_cyc_lcl)||(mem_cyc_gbl),
|
| 559 |
|
|
pf_illegal);
|
| 560 |
|
|
`endif
|
| 561 |
|
|
assign instruction_gie = gie;
|
| 562 |
|
|
|
| 563 |
|
|
initial r_dcdvalid = 1'b0;
|
| 564 |
|
|
always @(posedge i_clk)
|
| 565 |
|
|
if ((i_rst)||(clear_pipeline))
|
| 566 |
|
|
r_dcdvalid <= 1'b0;
|
| 567 |
|
|
else if (dcd_ce)
|
| 568 |
|
|
r_dcdvalid <= (pf_valid)&&(~clear_pipeline)&&(~dcd_ljmp)&&((~r_dcdvalid)||(~dcd_early_branch));
|
| 569 |
|
|
else if (op_ce)
|
| 570 |
|
|
r_dcdvalid <= 1'b0;
|
| 571 |
|
|
assign dcdvalid = r_dcdvalid;
|
| 572 |
|
|
`endif
|
| 573 |
|
|
|
| 574 |
|
|
`ifdef OPT_NEW_INSTRUCTION_SET
|
| 575 |
|
|
idecode #(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
|
| 576 |
|
|
IMPLEMENT_FPU)
|
| 577 |
|
|
instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
|
| 578 |
|
|
dcd_ce, dcd_stalled, instruction, instruction_gie,
|
| 579 |
|
|
instruction_pc, pf_valid, pf_illegal, dcd_phase,
|
| 580 |
|
|
dcd_illegal, dcd_pc, dcd_gie,
|
| 581 |
|
|
{ dcdR_cc, dcdR_pc, dcdR },
|
| 582 |
|
|
{ dcdA_cc, dcdA_pc, dcdA },
|
| 583 |
|
|
{ dcdB_cc, dcdB_pc, dcdB },
|
| 584 |
|
|
dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
|
| 585 |
|
|
dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
|
| 586 |
|
|
dcdR_wr,dcdA_rd, dcdB_rd,
|
| 587 |
|
|
dcd_early_branch,
|
| 588 |
|
|
dcd_branch_pc, dcd_ljmp,
|
| 589 |
|
|
dcd_pipe);
|
| 590 |
|
|
`else
|
| 591 |
|
|
idecode_deprecated
|
| 592 |
|
|
#(AW, IMPLEMENT_MPY, EARLY_BRANCHING, IMPLEMENT_DIVIDE,
|
| 593 |
|
|
IMPLEMENT_FPU)
|
| 594 |
|
|
instruction_decoder(i_clk, (i_rst)||(clear_pipeline),
|
| 595 |
|
|
dcd_ce, dcd_stalled, instruction, instruction_gie,
|
| 596 |
|
|
instruction_pc, pf_valid, pf_illegal, dcd_phase,
|
| 597 |
|
|
dcd_illegal, dcd_pc, dcd_gie,
|
| 598 |
|
|
{ dcdR_cc, dcdR_pc, dcdR },
|
| 599 |
|
|
{ dcdA_cc, dcdA_pc, dcdA },
|
| 600 |
|
|
{ dcdB_cc, dcdB_pc, dcdB },
|
| 601 |
|
|
dcdI, dcd_zI, dcdF, dcdF_wr, dcdOp,
|
| 602 |
|
|
dcdALU, dcdM, dcdDV, dcdFP, dcd_break, dcd_lock,
|
| 603 |
|
|
dcdR_wr,dcdA_rd, dcdB_rd,
|
| 604 |
|
|
dcd_early_branch,
|
| 605 |
|
|
dcd_branch_pc,
|
| 606 |
|
|
dcd_pipe);
|
| 607 |
|
|
assign dcd_ljmp = 1'b0;
|
| 608 |
|
|
`endif
|
| 609 |
|
|
|
| 610 |
|
|
`ifdef OPT_PIPELINED_BUS_ACCESS
|
| 611 |
|
|
reg op_pipe;
|
| 612 |
|
|
|
| 613 |
|
|
initial op_pipe = 1'b0;
|
| 614 |
|
|
// To be a pipeable operation, there must be
|
| 615 |
|
|
// two valid adjacent instructions
|
| 616 |
|
|
// Both must be memory instructions
|
| 617 |
|
|
// Both must be writes, or both must be reads
|
| 618 |
|
|
// Both operations must be to the same identical address,
|
| 619 |
|
|
// or at least a single (one) increment above that address
|
| 620 |
|
|
//
|
| 621 |
|
|
// However ... we need to know this before this clock, hence this is
|
| 622 |
|
|
// calculated in the instruction decoder.
|
| 623 |
|
|
always @(posedge i_clk)
|
| 624 |
|
|
if (op_ce)
|
| 625 |
|
|
op_pipe <= dcd_pipe;
|
| 626 |
|
|
`endif
|
| 627 |
|
|
|
| 628 |
|
|
//
|
| 629 |
|
|
//
|
| 630 |
|
|
// PIPELINE STAGE #3 :: Read Operands (Registers)
|
| 631 |
|
|
//
|
| 632 |
|
|
//
|
| 633 |
|
|
assign w_opA = regset[dcdA];
|
| 634 |
|
|
assign w_opB = regset[dcdB];
|
| 635 |
|
|
|
| 636 |
|
|
wire [31:0] w_pcA_v;
|
| 637 |
|
|
generate
|
| 638 |
|
|
if (AW < 32)
|
| 639 |
|
|
assign w_pcA_v = {{(32-AW){1'b0}}, (dcdA[4] == dcd_gie)?dcd_pc:upc };
|
| 640 |
|
|
else
|
| 641 |
|
|
assign w_pcA_v = (dcdA[4] == dcd_gie)?dcd_pc:upc;
|
| 642 |
|
|
endgenerate
|
| 643 |
|
|
|
| 644 |
|
|
`ifdef OPT_PIPELINED
|
| 645 |
|
|
reg [4:0] opA_id, opB_id;
|
| 646 |
|
|
reg opA_rd, opB_rd;
|
| 647 |
|
|
always @(posedge i_clk)
|
| 648 |
|
|
if (op_ce)
|
| 649 |
|
|
begin
|
| 650 |
|
|
opA_id <= dcdA;
|
| 651 |
|
|
opB_id <= dcdB;
|
| 652 |
|
|
opA_rd <= dcdA_rd;
|
| 653 |
|
|
opB_rd <= dcdB_rd;
|
| 654 |
|
|
end
|
| 655 |
|
|
`endif
|
| 656 |
|
|
|
| 657 |
|
|
always @(posedge i_clk)
|
| 658 |
|
|
if (op_ce) // &&(dcdvalid))
|
| 659 |
|
|
begin
|
| 660 |
|
|
if ((wr_reg_ce)&&(wr_reg_id == dcdA))
|
| 661 |
|
|
r_opA <= wr_reg_vl;
|
| 662 |
|
|
else if (dcdA_pc)
|
| 663 |
|
|
r_opA <= w_pcA_v;
|
| 664 |
|
|
else if (dcdA_cc)
|
| 665 |
|
|
r_opA <= { w_opA[31:14], (dcdA[4])?w_uflags:w_iflags };
|
| 666 |
|
|
else
|
| 667 |
|
|
r_opA <= w_opA;
|
| 668 |
|
|
`ifdef OPT_PIPELINED
|
| 669 |
|
|
end else
|
| 670 |
|
|
begin // We were going to pick these up when they became valid,
|
| 671 |
|
|
// but for some reason we're stuck here as they became
|
| 672 |
|
|
// valid. Pick them up now anyway
|
| 673 |
|
|
// if (((opA_alu)&&(alu_wr))||((opA_mem)&&(mem_valid)))
|
| 674 |
|
|
// r_opA <= wr_reg_vl;
|
| 675 |
|
|
if ((wr_reg_ce)&&(wr_reg_id == opA_id)&&(opA_rd))
|
| 676 |
|
|
r_opA <= wr_reg_vl;
|
| 677 |
|
|
`endif
|
| 678 |
|
|
end
|
| 679 |
|
|
|
| 680 |
|
|
wire [31:0] w_opBnI, w_pcB_v;
|
| 681 |
|
|
generate
|
| 682 |
|
|
if (AW < 32)
|
| 683 |
|
|
assign w_pcB_v = {{(32-AW){1'b0}}, (dcdB[4] == dcd_gie)?dcd_pc:upc };
|
| 684 |
|
|
else
|
| 685 |
|
|
assign w_pcB_v = (dcdB[4] == dcd_gie)?dcd_pc:upc;
|
| 686 |
|
|
endgenerate
|
| 687 |
|
|
|
| 688 |
|
|
assign w_opBnI = (~dcdB_rd) ? 32'h00
|
| 689 |
|
|
: (((wr_reg_ce)&&(wr_reg_id == dcdB)) ? wr_reg_vl
|
| 690 |
|
|
: ((dcdB_pc) ? w_pcB_v
|
| 691 |
|
|
: ((dcdB_cc) ? { w_opB[31:14], (dcdB[4])?w_uflags:w_iflags}
|
| 692 |
|
|
: w_opB)));
|
| 693 |
|
|
|
| 694 |
|
|
always @(posedge i_clk)
|
| 695 |
|
|
if (op_ce) // &&(dcdvalid))
|
| 696 |
|
|
r_opB <= w_opBnI + dcdI;
|
| 697 |
|
|
`ifdef OPT_PIPELINED
|
| 698 |
|
|
else if ((wr_reg_ce)&&(opB_id == wr_reg_id)&&(opB_rd))
|
| 699 |
|
|
r_opB <= wr_reg_vl;
|
| 700 |
|
|
`endif
|
| 701 |
|
|
|
| 702 |
|
|
// The logic here has become more complex than it should be, no thanks
|
| 703 |
|
|
// to Xilinx's Vivado trying to help. The conditions are supposed to
|
| 704 |
|
|
// be two sets of four bits: the top bits specify what bits matter, the
|
| 705 |
|
|
// bottom specify what those top bits must equal. However, two of
|
| 706 |
|
|
// conditions check whether bits are on, and those are the only two
|
| 707 |
|
|
// conditions checking those bits. Therefore, Vivado complains that
|
| 708 |
|
|
// these two bits are redundant. Hence the convoluted expression
|
| 709 |
|
|
// below, arriving at what we finally want in the (now wire net)
|
| 710 |
|
|
// opF.
|
| 711 |
|
|
always @(posedge i_clk)
|
| 712 |
|
|
if (op_ce)
|
| 713 |
|
|
begin // Set the flag condition codes, bit order is [3:0]=VNCZ
|
| 714 |
|
|
case(dcdF[2:0])
|
| 715 |
|
|
3'h0: r_opF <= 6'h00; // Always
|
| 716 |
|
|
`ifdef OPT_NEW_INSTRUCTION_SET
|
| 717 |
|
|
// These were remapped as part of the new instruction
|
| 718 |
|
|
// set in order to make certain that the low order
|
| 719 |
|
|
// two bits contained the most commonly used
|
| 720 |
|
|
// conditions: Always, LT, Z, and NZ.
|
| 721 |
|
|
3'h1: r_opF <= 6'h24; // LT
|
| 722 |
|
|
3'h2: r_opF <= 6'h11; // Z
|
| 723 |
|
|
3'h3: r_opF <= 6'h10; // NE
|
| 724 |
|
|
3'h4: r_opF <= 6'h30; // GT (!N&!Z)
|
| 725 |
|
|
3'h5: r_opF <= 6'h20; // GE (!N)
|
| 726 |
|
|
`else
|
| 727 |
|
|
3'h1: r_opF <= 6'h11; // Z
|
| 728 |
|
|
3'h2: r_opF <= 6'h10; // NE
|
| 729 |
|
|
3'h3: r_opF <= 6'h20; // GE (!N)
|
| 730 |
|
|
3'h4: r_opF <= 6'h30; // GT (!N&!Z)
|
| 731 |
|
|
3'h5: r_opF <= 6'h24; // LT
|
| 732 |
|
|
`endif
|
| 733 |
|
|
3'h6: r_opF <= 6'h02; // C
|
| 734 |
|
|
3'h7: r_opF <= 6'h08; // V
|
| 735 |
|
|
endcase
|
| 736 |
|
|
end // Bit order is { (flags_not_used), VNCZ mask, VNCZ value }
|
| 737 |
|
|
assign opF = { r_opF[3], r_opF[5], r_opF[1], r_opF[4:0] };
|
| 738 |
|
|
|
| 739 |
|
|
wire w_opvalid;
|
| 740 |
|
|
assign w_opvalid = (~clear_pipeline)&&(dcdvalid)&&(~dcd_ljmp);
|
| 741 |
|
|
initial opvalid = 1'b0;
|
| 742 |
|
|
initial opvalid_alu = 1'b0;
|
| 743 |
|
|
initial opvalid_mem = 1'b0;
|
| 744 |
|
|
initial opvalid_div = 1'b0;
|
| 745 |
|
|
initial opvalid_fpu = 1'b0;
|
| 746 |
|
|
always @(posedge i_clk)
|
| 747 |
|
|
if (i_rst)
|
| 748 |
|
|
begin
|
| 749 |
|
|
opvalid <= 1'b0;
|
| 750 |
|
|
opvalid_alu <= 1'b0;
|
| 751 |
|
|
opvalid_mem <= 1'b0;
|
| 752 |
|
|
end else if (op_ce)
|
| 753 |
|
|
begin
|
| 754 |
|
|
// Do we have a valid instruction?
|
| 755 |
|
|
// The decoder may vote to stall one of its
|
| 756 |
|
|
// instructions based upon something we currently
|
| 757 |
|
|
// have in our queue. This instruction must then
|
| 758 |
|
|
// move forward, and get a stall cycle inserted.
|
| 759 |
|
|
// Hence, the test on dcd_stalled here. If we must
|
| 760 |
|
|
// wait until our operands are valid, then we aren't
|
| 761 |
|
|
// valid yet until then.
|
| 762 |
|
|
opvalid<= w_opvalid;
|
| 763 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 764 |
|
|
opvalid_alu <= ((dcdALU)||(dcd_illegal))&&(w_opvalid);
|
| 765 |
|
|
opvalid_mem <= (dcdM)&&(~dcd_illegal)&&(w_opvalid);
|
| 766 |
|
|
opvalid_div <= (dcdDV)&&(~dcd_illegal)&&(w_opvalid);
|
| 767 |
|
|
opvalid_fpu <= (dcdFP)&&(~dcd_illegal)&&(w_opvalid);
|
| 768 |
|
|
`else
|
| 769 |
|
|
opvalid_alu <= (dcdALU)&&(w_opvalid);
|
| 770 |
|
|
opvalid_mem <= (dcdM)&&(w_opvalid);
|
| 771 |
|
|
opvalid_div <= (dcdDV)&&(w_opvalid);
|
| 772 |
|
|
opvalid_fpu <= (dcdFP)&&(w_opvalid);
|
| 773 |
|
|
`endif
|
| 774 |
|
|
end else if ((clear_pipeline)||(alu_ce)||(mem_ce)||(div_ce)||(fpu_ce))
|
| 775 |
|
|
begin
|
| 776 |
|
|
opvalid <= 1'b0;
|
| 777 |
|
|
opvalid_alu <= 1'b0;
|
| 778 |
|
|
opvalid_mem <= 1'b0;
|
| 779 |
|
|
opvalid_div <= 1'b0;
|
| 780 |
|
|
opvalid_fpu <= 1'b0;
|
| 781 |
|
|
end
|
| 782 |
|
|
|
| 783 |
|
|
// Here's part of our debug interface. When we recognize a break
|
| 784 |
|
|
// instruction, we set the op_break flag. That'll prevent this
|
| 785 |
|
|
// instruction from entering the ALU, and cause an interrupt before
|
| 786 |
|
|
// this instruction. Thus, returning to this code will cause the
|
| 787 |
|
|
// break to repeat and continue upon return. To get out of this
|
| 788 |
|
|
// condition, replace the break instruction with what it is supposed
|
| 789 |
|
|
// to be, step through it, and then replace it back. In this fashion,
|
| 790 |
|
|
// a debugger can step through code.
|
| 791 |
|
|
// assign w_op_break = (dcd_break)&&(r_dcdI[15:0] == 16'h0001);
|
| 792 |
|
|
initial op_break = 1'b0;
|
| 793 |
|
|
always @(posedge i_clk)
|
| 794 |
|
|
if (i_rst) op_break <= 1'b0;
|
| 795 |
|
|
else if (op_ce) op_break <= (dcd_break);
|
| 796 |
|
|
else if ((clear_pipeline)||(~opvalid))
|
| 797 |
|
|
op_break <= 1'b0;
|
| 798 |
|
|
|
| 799 |
|
|
`ifdef OPT_PIPELINED
|
| 800 |
|
|
generate
|
| 801 |
|
|
if (IMPLEMENT_LOCK != 0)
|
| 802 |
|
|
begin
|
| 803 |
|
|
reg r_op_lock, r_op_lock_stall;
|
| 804 |
|
|
|
| 805 |
|
|
initial r_op_lock_stall = 1'b0;
|
| 806 |
|
|
always @(posedge i_clk)
|
| 807 |
|
|
if (i_rst)
|
| 808 |
|
|
r_op_lock_stall <= 1'b0;
|
| 809 |
|
|
else
|
| 810 |
|
|
r_op_lock_stall <= (~opvalid)||(~op_lock)
|
| 811 |
|
|
||(~dcdvalid)||(~pf_valid);
|
| 812 |
|
|
|
| 813 |
|
|
assign op_lock_stall = r_op_lock_stall;
|
| 814 |
|
|
|
| 815 |
|
|
initial r_op_lock = 1'b0;
|
| 816 |
|
|
always @(posedge i_clk)
|
| 817 |
|
|
if (i_rst)
|
| 818 |
|
|
r_op_lock <= 1'b0;
|
| 819 |
|
|
else if ((op_ce)&&(dcd_lock))
|
| 820 |
|
|
r_op_lock <= 1'b1;
|
| 821 |
|
|
else if ((op_ce)||(clear_pipeline))
|
| 822 |
|
|
r_op_lock <= 1'b0;
|
| 823 |
|
|
assign op_lock = r_op_lock;
|
| 824 |
|
|
|
| 825 |
|
|
end else begin
|
| 826 |
|
|
assign op_lock_stall = 1'b0;
|
| 827 |
|
|
assign op_lock = 1'b0;
|
| 828 |
|
|
end endgenerate
|
| 829 |
|
|
|
| 830 |
|
|
`else
|
| 831 |
|
|
assign op_lock_stall = 1'b0;
|
| 832 |
|
|
assign op_lock = 1'b0;
|
| 833 |
|
|
`endif
|
| 834 |
|
|
|
| 835 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 836 |
|
|
initial op_illegal = 1'b0;
|
| 837 |
|
|
always @(posedge i_clk)
|
| 838 |
|
|
if ((i_rst)||(clear_pipeline))
|
| 839 |
|
|
op_illegal <= 1'b0;
|
| 840 |
|
|
else if(op_ce)
|
| 841 |
|
|
`ifdef OPT_PIPELINED
|
| 842 |
|
|
op_illegal <=(dcd_illegal)||((dcd_lock)&&(IMPLEMENT_LOCK == 0));
|
| 843 |
|
|
`else
|
| 844 |
|
|
op_illegal <= (dcd_illegal)||(dcd_lock);
|
| 845 |
|
|
`endif
|
| 846 |
|
|
`endif
|
| 847 |
|
|
|
| 848 |
|
|
// No generate on EARLY_BRANCHING here, since if EARLY_BRANCHING is not
|
| 849 |
|
|
// set, dcd_early_branch will simply be a wire connected to zero and
|
| 850 |
|
|
// this logic should just optimize.
|
| 851 |
|
|
always @(posedge i_clk)
|
| 852 |
|
|
if (op_ce)
|
| 853 |
|
|
begin
|
| 854 |
|
|
opF_wr <= (dcdF_wr)&&((~dcdR_cc)||(~dcdR_wr))
|
| 855 |
|
|
&&(~dcd_early_branch)&&(~dcd_illegal);
|
| 856 |
|
|
opR_wr <= (dcdR_wr)&&(~dcd_early_branch)&&(~dcd_illegal);
|
| 857 |
|
|
end
|
| 858 |
|
|
|
| 859 |
|
|
always @(posedge i_clk)
|
| 860 |
|
|
if (op_ce)
|
| 861 |
|
|
begin
|
| 862 |
|
|
opn <= dcdOp; // Which ALU operation?
|
| 863 |
|
|
// opM <= dcdM; // Is this a memory operation?
|
| 864 |
|
|
// What register will these results be written into?
|
| 865 |
|
|
opR <= dcdR;
|
| 866 |
|
|
opR_cc <= (dcdR_cc)&&(dcdR_wr)&&(dcdR[4]==dcd_gie);
|
| 867 |
|
|
// User level (1), vs supervisor (0)/interrupts disabled
|
| 868 |
|
|
op_gie <= dcd_gie;
|
| 869 |
|
|
|
| 870 |
|
|
|
| 871 |
|
|
//
|
| 872 |
|
|
op_pc <= (dcd_early_branch)?dcd_branch_pc:dcd_pc;
|
| 873 |
|
|
end
|
| 874 |
|
|
assign opFl = (op_gie)?(w_uflags):(w_iflags);
|
| 875 |
|
|
|
| 876 |
|
|
`ifdef OPT_VLIW
|
| 877 |
|
|
reg r_op_phase;
|
| 878 |
|
|
initial r_op_phase = 1'b0;
|
| 879 |
|
|
always @(posedge i_clk)
|
| 880 |
|
|
if ((i_rst)||(clear_pipeline))
|
| 881 |
|
|
r_op_phase <= 1'b0;
|
| 882 |
|
|
else if (op_ce)
|
| 883 |
|
|
r_op_phase <= dcd_phase;
|
| 884 |
|
|
assign op_phase = r_op_phase;
|
| 885 |
|
|
`else
|
| 886 |
|
|
assign op_phase = 1'b0;
|
| 887 |
|
|
`endif
|
| 888 |
|
|
|
| 889 |
|
|
// This is tricky. First, the PC and Flags registers aren't kept in
|
| 890 |
|
|
// register set but in special registers of their own. So step one
|
| 891 |
|
|
// is to select the right register. Step to is to replace that
|
| 892 |
|
|
// register with the results of an ALU or memory operation, if such
|
| 893 |
|
|
// results are now available. Otherwise, we'd need to insert a wait
|
| 894 |
|
|
// state of some type.
|
| 895 |
|
|
//
|
| 896 |
|
|
// The alternative approach would be to define some sort of
|
| 897 |
|
|
// op_stall wire, which would stall any upstream stage.
|
| 898 |
|
|
// We'll create a flag here to start our coordination. Once we
|
| 899 |
|
|
// define this flag to something other than just plain zero, then
|
| 900 |
|
|
// the stalls will already be in place.
|
| 901 |
|
|
`ifdef OPT_PIPELINED
|
| 902 |
|
|
assign opA = ((wr_reg_ce)&&(wr_reg_id == opA_id)) // &&(opA_rd))
|
| 903 |
|
|
? wr_reg_vl : r_opA;
|
| 904 |
|
|
`else
|
| 905 |
|
|
assign opA = r_opA;
|
| 906 |
|
|
`endif
|
| 907 |
|
|
|
| 908 |
|
|
`ifdef OPT_PIPELINED
|
| 909 |
|
|
// Stall if we have decoded an instruction that will read register A
|
| 910 |
|
|
// AND ... something that may write a register is running
|
| 911 |
|
|
// AND (series of conditions here ...)
|
| 912 |
|
|
// The operation might set flags, and we wish to read the
|
| 913 |
|
|
// CC register
|
| 914 |
|
|
// OR ... (No other conditions)
|
| 915 |
|
|
assign dcdA_stall = (dcdA_rd) // &&(dcdvalid) is checked for elsewhere
|
| 916 |
|
|
&&((opvalid)||(mem_rdbusy)
|
| 917 |
|
|
||(div_busy)||(fpu_busy))
|
| 918 |
|
|
&&((opF_wr)&&(dcdA_cc));
|
| 919 |
|
|
`else
|
| 920 |
|
|
// There are no pipeline hazards, if we aren't pipelined
|
| 921 |
|
|
assign dcdA_stall = 1'b0;
|
| 922 |
|
|
`endif
|
| 923 |
|
|
|
| 924 |
|
|
`ifdef OPT_PIPELINED
|
| 925 |
|
|
assign opB = ((wr_reg_ce)&&(wr_reg_id == opB_id)&&(opB_rd))
|
| 926 |
|
|
? wr_reg_vl: r_opB;
|
| 927 |
|
|
`else
|
| 928 |
|
|
assign opB = r_opB;
|
| 929 |
|
|
`endif
|
| 930 |
|
|
|
| 931 |
|
|
`ifdef OPT_PIPELINED
|
| 932 |
|
|
// Stall if we have decoded an instruction that will read register B
|
| 933 |
|
|
// AND ... something that may write a (unknown) register is running
|
| 934 |
|
|
// AND (series of conditions here ...)
|
| 935 |
|
|
// The operation might set flags, and we wish to read the
|
| 936 |
|
|
// CC register
|
| 937 |
|
|
// OR the operation might set register B, and we still need
|
| 938 |
|
|
// a clock to add the offset to it
|
| 939 |
|
|
assign dcdB_stall = (dcdB_rd) // &&(dcdvalid) is checked for elsewhere
|
| 940 |
|
|
// If the op stage isn't valid, yet something
|
| 941 |
|
|
// is running, then it must have been valid.
|
| 942 |
|
|
// We'll use the last values from that stage
|
| 943 |
|
|
// (opR_wr, opF_wr, opR) in our logic below.
|
| 944 |
|
|
&&((opvalid)||(mem_rdbusy)
|
| 945 |
|
|
||(div_busy)||(fpu_busy))
|
| 946 |
|
|
&&(
|
| 947 |
|
|
// Stall on memory ops writing to my register
|
| 948 |
|
|
// (i.e. loads), or on any write to my
|
| 949 |
|
|
// register if I have an immediate offset
|
| 950 |
|
|
// Note the exception for writing to the PC:
|
| 951 |
|
|
// if I write to the PC, the whole next
|
| 952 |
|
|
// instruction is invalid, not just the
|
| 953 |
|
|
// operand. That'll get wiped in the
|
| 954 |
|
|
// next operation anyway, so don't stall
|
| 955 |
|
|
// here. This keeps a BC X, BNZ Y from
|
| 956 |
|
|
// stalling between the two branches.
|
| 957 |
|
|
// BC X, BRA Y is still clear, since BRA Y
|
| 958 |
|
|
// is an early branch instruction.
|
| 959 |
|
|
// (This exception is commented out in
|
| 960 |
|
|
// order to help keep our logic simple, and
|
| 961 |
|
|
// because multiple conditional branches
|
| 962 |
|
|
// following each other constitutes a
|
| 963 |
|
|
// fairly unusualy code structure.)
|
| 964 |
|
|
//
|
| 965 |
|
|
((~dcd_zI)&&(opR == dcdB)&&(opR_wr))
|
| 966 |
|
|
// &&(opR != { op_gie, `CPU_PC_REG } )
|
| 967 |
|
|
// Stall following any instruction that will
|
| 968 |
|
|
// set the flags, if we're going to need the
|
| 969 |
|
|
// flags (CC) register for opB.
|
| 970 |
|
|
||((opF_wr)&&(dcdB_cc))
|
| 971 |
|
|
// Stall on any ongoing memory operation that
|
| 972 |
|
|
// will write to opB -- captured above
|
| 973 |
|
|
// ||((mem_busy)&&(~mem_we)&&(mem_last_reg==dcdB)&&(~dcd_zI))
|
| 974 |
|
|
);
|
| 975 |
|
|
assign dcdF_stall = ((~dcdF[3])
|
| 976 |
|
|
||((dcdA_rd)&&(dcdA_cc))
|
| 977 |
|
|
||((dcdB_rd)&&(dcdB_cc)))
|
| 978 |
|
|
&&(opvalid)&&(opR_cc);
|
| 979 |
|
|
// &&(dcdvalid) is checked for elsewhere
|
| 980 |
|
|
`else
|
| 981 |
|
|
// No stalls without pipelining, 'cause how can you have a pipeline
|
| 982 |
|
|
// hazard without the pipeline?
|
| 983 |
|
|
assign dcdB_stall = 1'b0;
|
| 984 |
|
|
assign dcdF_stall = 1'b0;
|
| 985 |
|
|
`endif
|
| 986 |
|
|
//
|
| 987 |
|
|
//
|
| 988 |
|
|
// PIPELINE STAGE #4 :: Apply Instruction
|
| 989 |
|
|
//
|
| 990 |
|
|
//
|
| 991 |
|
|
`ifdef OPT_NEW_INSTRUCTION_SET
|
| 992 |
|
|
cpuops #(IMPLEMENT_MPY) doalu(i_clk, i_rst, alu_ce,
|
| 993 |
|
|
(opvalid_alu), opn, opA, opB,
|
| 994 |
|
|
alu_result, alu_flags, alu_valid, alu_illegal_op,
|
| 995 |
|
|
alu_busy);
|
| 996 |
|
|
`else
|
| 997 |
|
|
cpuops_deprecated #(IMPLEMENT_MPY) doalu(i_clk, i_rst, alu_ce,
|
| 998 |
|
|
(opvalid_alu), opn, opA, opB,
|
| 999 |
|
|
alu_result, alu_flags, alu_valid, alu_illegal_op);
|
| 1000 |
|
|
assign alu_busy = 1'b0;
|
| 1001 |
|
|
`endif
|
| 1002 |
|
|
|
| 1003 |
|
|
generate
|
| 1004 |
|
|
if (IMPLEMENT_DIVIDE != 0)
|
| 1005 |
|
|
begin
|
| 1006 |
|
|
div thedivide(i_clk, (i_rst)||(clear_pipeline), div_ce, opn[0],
|
| 1007 |
|
|
opA, opB, div_busy, div_valid, div_error, div_result,
|
| 1008 |
|
|
div_flags);
|
| 1009 |
|
|
end else begin
|
| 1010 |
|
|
assign div_error = 1'b1;
|
| 1011 |
|
|
assign div_busy = 1'b0;
|
| 1012 |
|
|
assign div_valid = 1'b0;
|
| 1013 |
|
|
assign div_result= 32'h00;
|
| 1014 |
|
|
assign div_flags = 4'h0;
|
| 1015 |
|
|
end endgenerate
|
| 1016 |
|
|
|
| 1017 |
|
|
generate
|
| 1018 |
|
|
if (IMPLEMENT_FPU != 0)
|
| 1019 |
|
|
begin
|
| 1020 |
|
|
//
|
| 1021 |
|
|
// sfpu thefpu(i_clk, i_rst, fpu_ce,
|
| 1022 |
|
|
// opA, opB, fpu_busy, fpu_valid, fpu_err, fpu_result,
|
| 1023 |
|
|
// fpu_flags);
|
| 1024 |
|
|
//
|
| 1025 |
|
|
assign fpu_error = 1'b1;
|
| 1026 |
|
|
assign fpu_busy = 1'b0;
|
| 1027 |
|
|
assign fpu_valid = 1'b0;
|
| 1028 |
|
|
assign fpu_result= 32'h00;
|
| 1029 |
|
|
assign fpu_flags = 4'h0;
|
| 1030 |
|
|
end else begin
|
| 1031 |
|
|
assign fpu_error = 1'b1;
|
| 1032 |
|
|
assign fpu_busy = 1'b0;
|
| 1033 |
|
|
assign fpu_valid = 1'b0;
|
| 1034 |
|
|
assign fpu_result= 32'h00;
|
| 1035 |
|
|
assign fpu_flags = 4'h0;
|
| 1036 |
|
|
end endgenerate
|
| 1037 |
|
|
|
| 1038 |
|
|
|
| 1039 |
|
|
assign set_cond = ((opF[7:4]&opFl[3:0])==opF[3:0]);
|
| 1040 |
|
|
initial alF_wr = 1'b0;
|
| 1041 |
|
|
initial alu_wr = 1'b0;
|
| 1042 |
|
|
always @(posedge i_clk)
|
| 1043 |
|
|
if (i_rst)
|
| 1044 |
|
|
begin
|
| 1045 |
|
|
alu_wr <= 1'b0;
|
| 1046 |
|
|
alF_wr <= 1'b0;
|
| 1047 |
|
|
end else if (alu_ce)
|
| 1048 |
|
|
begin
|
| 1049 |
|
|
// alu_reg <= opR;
|
| 1050 |
|
|
alu_wr <= (opR_wr)&&(set_cond);
|
| 1051 |
|
|
alF_wr <= (opF_wr)&&(set_cond);
|
| 1052 |
|
|
end else if (~alu_busy) begin
|
| 1053 |
|
|
// These are strobe signals, so clear them if not
|
| 1054 |
|
|
// set for any particular clock
|
| 1055 |
|
|
alu_wr <= (i_halt)&&(i_dbg_we);
|
| 1056 |
|
|
alF_wr <= 1'b0;
|
| 1057 |
|
|
end
|
| 1058 |
|
|
|
| 1059 |
|
|
`ifdef OPT_VLIW
|
| 1060 |
|
|
reg r_alu_phase;
|
| 1061 |
|
|
initial r_alu_phase = 1'b0;
|
| 1062 |
|
|
always @(posedge i_clk)
|
| 1063 |
|
|
if (i_rst)
|
| 1064 |
|
|
r_alu_phase <= 1'b0;
|
| 1065 |
|
|
else if ((alu_ce)||(mem_ce)||(div_ce)||(fpu_ce))
|
| 1066 |
|
|
r_alu_phase <= op_phase;
|
| 1067 |
|
|
assign alu_phase = r_alu_phase;
|
| 1068 |
|
|
`else
|
| 1069 |
|
|
assign alu_phase = 1'b0;
|
| 1070 |
|
|
`endif
|
| 1071 |
|
|
|
| 1072 |
|
|
always @(posedge i_clk)
|
| 1073 |
|
|
if ((alu_ce)||(div_ce)||(fpu_ce))
|
| 1074 |
|
|
alu_reg <= opR;
|
| 1075 |
|
|
else if ((i_halt)&&(i_dbg_we))
|
| 1076 |
|
|
alu_reg <= i_dbg_reg;
|
| 1077 |
|
|
|
| 1078 |
|
|
reg [31:0] dbg_val;
|
| 1079 |
|
|
reg dbgv;
|
| 1080 |
|
|
always @(posedge i_clk)
|
| 1081 |
|
|
dbg_val <= i_dbg_data;
|
| 1082 |
|
|
initial dbgv = 1'b0;
|
| 1083 |
|
|
always @(posedge i_clk)
|
| 1084 |
|
|
dbgv <= (~i_rst)&&(~alu_ce)&&((i_halt)&&(i_dbg_we));
|
| 1085 |
|
|
always @(posedge i_clk)
|
| 1086 |
|
|
if ((alu_ce)||(mem_ce))
|
| 1087 |
|
|
alu_gie <= op_gie;
|
| 1088 |
|
|
always @(posedge i_clk)
|
| 1089 |
|
|
if ((alu_ce)||((master_ce)&&(opvalid_mem)&&(~clear_pipeline)
|
| 1090 |
|
|
&&(~mem_stalled)))
|
| 1091 |
|
|
alu_pc <= op_pc;
|
| 1092 |
|
|
|
| 1093 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 1094 |
|
|
reg r_alu_illegal;
|
| 1095 |
|
|
initial r_alu_illegal = 0;
|
| 1096 |
|
|
always @(posedge i_clk)
|
| 1097 |
|
|
if (clear_pipeline)
|
| 1098 |
|
|
r_alu_illegal <= 1'b0;
|
| 1099 |
|
|
else if ((alu_ce)||(mem_ce))
|
| 1100 |
|
|
r_alu_illegal <= op_illegal;
|
| 1101 |
|
|
assign alu_illegal = (alu_illegal_op)||(r_alu_illegal);
|
| 1102 |
|
|
`endif
|
| 1103 |
|
|
|
| 1104 |
|
|
// This _almost_ is equal to (alu_ce)||(mem_ce). The only
|
| 1105 |
|
|
// problem is that mem_ce is gated by the set_cond, and
|
| 1106 |
|
|
// the PC will be valid independent of the set condition. Hence, this
|
| 1107 |
|
|
// equals (alu_ce)||(everything in mem_ce but the set condition)
|
| 1108 |
|
|
initial alu_pc_valid = 1'b0;
|
| 1109 |
|
|
always @(posedge i_clk)
|
| 1110 |
|
|
alu_pc_valid <= ((alu_ce)
|
| 1111 |
|
|
||((master_ce)&&(opvalid_mem)&&(~clear_pipeline)&&(~mem_stalled)));
|
| 1112 |
|
|
|
| 1113 |
|
|
wire bus_lock;
|
| 1114 |
|
|
`ifdef OPT_PIPELINED
|
| 1115 |
|
|
generate
|
| 1116 |
|
|
if (IMPLEMENT_LOCK != 0)
|
| 1117 |
|
|
begin
|
| 1118 |
|
|
reg r_bus_lock;
|
| 1119 |
|
|
initial r_bus_lock = 1'b0;
|
| 1120 |
|
|
always @(posedge i_clk)
|
| 1121 |
|
|
if (i_rst)
|
| 1122 |
|
|
r_bus_lock <= 1'b0;
|
| 1123 |
|
|
else if ((op_ce)&&(op_lock))
|
| 1124 |
|
|
r_bus_lock <= 1'b1;
|
| 1125 |
|
|
else if (~opvalid_mem)
|
| 1126 |
|
|
r_bus_lock <= 1'b0;
|
| 1127 |
|
|
assign bus_lock = r_bus_lock;
|
| 1128 |
|
|
end else begin
|
| 1129 |
|
|
assign bus_lock = 1'b0;
|
| 1130 |
|
|
end endgenerate
|
| 1131 |
|
|
`else
|
| 1132 |
|
|
assign bus_lock = 1'b0;
|
| 1133 |
|
|
`endif
|
| 1134 |
|
|
|
| 1135 |
|
|
`ifdef OPT_PIPELINED_BUS_ACCESS
|
| 1136 |
|
|
pipemem #(AW,IMPLEMENT_LOCK) domem(i_clk, i_rst,(mem_ce)&&(set_cond), bus_lock,
|
| 1137 |
|
|
(opn[0]), opB, opA, opR,
|
| 1138 |
|
|
mem_busy, mem_pipe_stalled,
|
| 1139 |
|
|
mem_valid, bus_err, mem_wreg, mem_result,
|
| 1140 |
|
|
mem_cyc_gbl, mem_cyc_lcl,
|
| 1141 |
|
|
mem_stb_gbl, mem_stb_lcl,
|
| 1142 |
|
|
mem_we, mem_addr, mem_data,
|
| 1143 |
|
|
mem_ack, mem_stall, mem_err, i_wb_data);
|
| 1144 |
|
|
|
| 1145 |
|
|
`else // PIPELINED_BUS_ACCESS
|
| 1146 |
|
|
memops #(AW,IMPLEMENT_LOCK) domem(i_clk, i_rst,(mem_ce)&&(set_cond), bus_lock,
|
| 1147 |
|
|
(opn[0]), opB, opA, opR,
|
| 1148 |
|
|
mem_busy,
|
| 1149 |
|
|
mem_valid, bus_err, mem_wreg, mem_result,
|
| 1150 |
|
|
mem_cyc_gbl, mem_cyc_lcl,
|
| 1151 |
|
|
mem_stb_gbl, mem_stb_lcl,
|
| 1152 |
|
|
mem_we, mem_addr, mem_data,
|
| 1153 |
|
|
mem_ack, mem_stall, mem_err, i_wb_data);
|
| 1154 |
|
|
`endif // PIPELINED_BUS_ACCESS
|
| 1155 |
|
|
assign mem_rdbusy = ((mem_busy)&&(~mem_we));
|
| 1156 |
|
|
|
| 1157 |
|
|
// Either the prefetch or the instruction gets the memory bus, but
|
| 1158 |
|
|
// never both.
|
| 1159 |
|
|
wbdblpriarb #(32,AW) pformem(i_clk, i_rst,
|
| 1160 |
|
|
// Memory access to the arbiter, priority position
|
| 1161 |
|
|
mem_cyc_gbl, mem_cyc_lcl, mem_stb_gbl, mem_stb_lcl,
|
| 1162 |
|
|
mem_we, mem_addr, mem_data, mem_ack, mem_stall, mem_err,
|
| 1163 |
|
|
// Prefetch access to the arbiter
|
| 1164 |
|
|
pf_cyc, 1'b0, pf_stb, 1'b0, pf_we, pf_addr, pf_data,
|
| 1165 |
|
|
pf_ack, pf_stall, pf_err,
|
| 1166 |
|
|
// Common wires, in and out, of the arbiter
|
| 1167 |
|
|
o_wb_gbl_cyc, o_wb_lcl_cyc, o_wb_gbl_stb, o_wb_lcl_stb,
|
| 1168 |
|
|
o_wb_we, o_wb_addr, o_wb_data,
|
| 1169 |
|
|
i_wb_ack, i_wb_stall, i_wb_err);
|
| 1170 |
|
|
|
| 1171 |
|
|
//
|
| 1172 |
|
|
//
|
| 1173 |
|
|
// PIPELINE STAGE #5 :: Write-back results
|
| 1174 |
|
|
//
|
| 1175 |
|
|
//
|
| 1176 |
|
|
// This stage is not allowed to stall. If results are ready to be
|
| 1177 |
|
|
// written back, they are written back at all cost. Sleepy CPU's
|
| 1178 |
|
|
// won't prevent write back, nor debug modes, halting the CPU, nor
|
| 1179 |
|
|
// anything else. Indeed, the (master_ce) bit is only as relevant
|
| 1180 |
|
|
// as knowinig something is available for writeback.
|
| 1181 |
|
|
|
| 1182 |
|
|
//
|
| 1183 |
|
|
// Write back to our generic register set ...
|
| 1184 |
|
|
// When shall we write back? On one of two conditions
|
| 1185 |
|
|
// Note that the flags needed to be checked before issuing the
|
| 1186 |
|
|
// bus instruction, so they don't need to be checked here.
|
| 1187 |
|
|
// Further, alu_wr includes (set_cond), so we don't need to
|
| 1188 |
|
|
// check for that here either.
|
| 1189 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 1190 |
|
|
assign wr_reg_ce = (~alu_illegal)&&
|
| 1191 |
|
|
(((alu_wr)&&(~clear_pipeline)
|
| 1192 |
|
|
&&((alu_valid)||(div_valid)||(fpu_valid)))
|
| 1193 |
|
|
||(mem_valid));
|
| 1194 |
|
|
`else
|
| 1195 |
|
|
assign wr_reg_ce = ((alu_wr)&&(~clear_pipeline))||(mem_valid)||(div_valid)||(fpu_valid);
|
| 1196 |
|
|
`endif
|
| 1197 |
|
|
// Which register shall be written?
|
| 1198 |
|
|
// COULD SIMPLIFY THIS: by adding three bits to these registers,
|
| 1199 |
|
|
// One or PC, one for CC, and one for GIE match
|
| 1200 |
|
|
// Note that the alu_reg is the register to write on a divide or
|
| 1201 |
|
|
// FPU operation.
|
| 1202 |
|
|
assign wr_reg_id = (alu_wr)?alu_reg:mem_wreg;
|
| 1203 |
|
|
// Are we writing to the CC register?
|
| 1204 |
|
|
assign wr_write_cc = (wr_reg_id[3:0] == `CPU_CC_REG);
|
| 1205 |
|
|
// Are we writing to the PC?
|
| 1206 |
|
|
assign wr_write_pc = (wr_reg_id[3:0] == `CPU_PC_REG);
|
| 1207 |
|
|
// What value to write?
|
| 1208 |
|
|
assign wr_reg_vl = ((mem_valid) ? mem_result
|
| 1209 |
|
|
:((div_valid|fpu_valid))
|
| 1210 |
|
|
? ((div_valid) ? div_result:fpu_result)
|
| 1211 |
|
|
:((dbgv) ? dbg_val : alu_result));
|
| 1212 |
|
|
always @(posedge i_clk)
|
| 1213 |
|
|
if (wr_reg_ce)
|
| 1214 |
|
|
regset[wr_reg_id] <= wr_reg_vl;
|
| 1215 |
|
|
|
| 1216 |
|
|
//
|
| 1217 |
|
|
// Write back to the condition codes/flags register ...
|
| 1218 |
|
|
// When shall we write to our flags register? alF_wr already
|
| 1219 |
|
|
// includes the set condition ...
|
| 1220 |
|
|
assign wr_flags_ce = ((alF_wr)||(div_valid)||(fpu_valid))&&(~clear_pipeline)&&(~alu_illegal);
|
| 1221 |
|
|
assign w_uflags = { uhalt_phase, ufpu_err_flag,
|
| 1222 |
|
|
udiv_err_flag, ubus_err_flag, trap, ill_err_u,
|
| 1223 |
|
|
1'b0, step, 1'b1, sleep,
|
| 1224 |
|
|
((wr_flags_ce)&&(alu_gie))?alu_flags:flags };
|
| 1225 |
|
|
assign w_iflags = { ihalt_phase, ifpu_err_flag,
|
| 1226 |
|
|
idiv_err_flag, ibus_err_flag, trap, ill_err_i,
|
| 1227 |
|
|
break_en, 1'b0, 1'b0, sleep,
|
| 1228 |
|
|
((wr_flags_ce)&&(~alu_gie))?alu_flags:iflags };
|
| 1229 |
|
|
|
| 1230 |
|
|
|
| 1231 |
|
|
// What value to write?
|
| 1232 |
|
|
always @(posedge i_clk)
|
| 1233 |
|
|
// If explicitly writing the register itself
|
| 1234 |
|
|
if ((wr_reg_ce)&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1235 |
|
|
flags <= wr_reg_vl[3:0];
|
| 1236 |
|
|
// Otherwise if we're setting the flags from an ALU operation
|
| 1237 |
|
|
else if ((wr_flags_ce)&&(alu_gie))
|
| 1238 |
|
|
flags <= (div_valid)?div_flags:((fpu_valid)?fpu_flags
|
| 1239 |
|
|
: alu_flags);
|
| 1240 |
|
|
|
| 1241 |
|
|
always @(posedge i_clk)
|
| 1242 |
|
|
if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_cc))
|
| 1243 |
|
|
iflags <= wr_reg_vl[3:0];
|
| 1244 |
|
|
else if ((wr_flags_ce)&&(~alu_gie))
|
| 1245 |
|
|
iflags <= (div_valid)?div_flags:((fpu_valid)?fpu_flags
|
| 1246 |
|
|
: alu_flags);
|
| 1247 |
|
|
|
| 1248 |
|
|
// The 'break' enable bit. This bit can only be set from supervisor
|
| 1249 |
|
|
// mode. It control what the CPU does upon encountering a break
|
| 1250 |
|
|
// instruction.
|
| 1251 |
|
|
//
|
| 1252 |
|
|
// The goal, upon encountering a break is that the CPU should stop and
|
| 1253 |
|
|
// not execute the break instruction, choosing instead to enter into
|
| 1254 |
|
|
// either interrupt mode or halt first.
|
| 1255 |
|
|
// if ((break_en) AND (break_instruction)) // user mode or not
|
| 1256 |
|
|
// HALT CPU
|
| 1257 |
|
|
// else if (break_instruction) // only in user mode
|
| 1258 |
|
|
// set an interrupt flag, go to supervisor mode
|
| 1259 |
|
|
// allow supervisor to step the CPU.
|
| 1260 |
|
|
// Upon a CPU halt, any break condition will be reset. The
|
| 1261 |
|
|
// external debugger will then need to deal with whatever
|
| 1262 |
|
|
// condition has taken place.
|
| 1263 |
|
|
initial break_en = 1'b0;
|
| 1264 |
|
|
always @(posedge i_clk)
|
| 1265 |
|
|
if ((i_rst)||(i_halt))
|
| 1266 |
|
|
break_en <= 1'b0;
|
| 1267 |
|
|
else if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_cc))
|
| 1268 |
|
|
break_en <= wr_reg_vl[`CPU_BREAK_BIT];
|
| 1269 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 1270 |
|
|
assign o_break = ((break_en)||(~op_gie))&&(op_break)
|
| 1271 |
|
|
&&(~alu_valid)&&(~mem_valid)&&(~mem_busy)
|
| 1272 |
|
|
&&(~div_busy)&&(~fpu_busy)
|
| 1273 |
|
|
&&(~clear_pipeline)
|
| 1274 |
|
|
||((~alu_gie)&&(bus_err))
|
| 1275 |
|
|
||((~alu_gie)&&(div_valid)&&(div_error))
|
| 1276 |
|
|
||((~alu_gie)&&(fpu_valid)&&(fpu_error))
|
| 1277 |
|
|
||((~alu_gie)&&(alu_pc_valid)&&(alu_illegal));
|
| 1278 |
|
|
`else
|
| 1279 |
|
|
assign o_break = (((break_en)||(~op_gie))&&(op_break)
|
| 1280 |
|
|
&&(~alu_valid)&&(~mem_valid)&&(~mem_busy)
|
| 1281 |
|
|
&&(~clear_pipeline))
|
| 1282 |
|
|
||((~alu_gie)&&(bus_err));
|
| 1283 |
|
|
`endif
|
| 1284 |
|
|
|
| 1285 |
|
|
|
| 1286 |
|
|
// The sleep register. Setting the sleep register causes the CPU to
|
| 1287 |
|
|
// sleep until the next interrupt. Setting the sleep register within
|
| 1288 |
|
|
// interrupt mode causes the processor to halt until a reset. This is
|
| 1289 |
|
|
// a panic/fault halt. The trick is that you cannot be allowed to
|
| 1290 |
|
|
// set the sleep bit and switch to supervisor mode in the same
|
| 1291 |
|
|
// instruction: users are not allowed to halt the CPU.
|
| 1292 |
|
|
always @(posedge i_clk)
|
| 1293 |
|
|
if ((i_rst)||(w_switch_to_interrupt))
|
| 1294 |
|
|
sleep <= 1'b0;
|
| 1295 |
|
|
else if ((wr_reg_ce)&&(wr_write_cc)&&(~alu_gie))
|
| 1296 |
|
|
// In supervisor mode, we have no protections. The
|
| 1297 |
|
|
// supervisor can set the sleep bit however he wants.
|
| 1298 |
|
|
// Well ... not quite. Switching to user mode and
|
| 1299 |
|
|
// sleep mode shouold only be possible if the interrupt
|
| 1300 |
|
|
// flag isn't set.
|
| 1301 |
|
|
// Thus: if (i_interrupt)&&(wr_reg_vl[GIE])
|
| 1302 |
|
|
// don't set the sleep bit
|
| 1303 |
|
|
// otherwise however it would o.w. be set
|
| 1304 |
|
|
sleep <= (wr_reg_vl[`CPU_SLEEP_BIT])
|
| 1305 |
|
|
&&((~i_interrupt)||(~wr_reg_vl[`CPU_GIE_BIT]));
|
| 1306 |
|
|
else if ((wr_reg_ce)&&(wr_write_cc)&&(wr_reg_vl[`CPU_GIE_BIT]))
|
| 1307 |
|
|
// In user mode, however, you can only set the sleep
|
| 1308 |
|
|
// mode while remaining in user mode. You can't switch
|
| 1309 |
|
|
// to sleep mode *and* supervisor mode at the same
|
| 1310 |
|
|
// time, lest you halt the CPU.
|
| 1311 |
|
|
sleep <= wr_reg_vl[`CPU_SLEEP_BIT];
|
| 1312 |
|
|
|
| 1313 |
|
|
always @(posedge i_clk)
|
| 1314 |
|
|
if ((i_rst)||(w_switch_to_interrupt))
|
| 1315 |
|
|
step <= 1'b0;
|
| 1316 |
|
|
else if ((wr_reg_ce)&&(~alu_gie)&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1317 |
|
|
step <= wr_reg_vl[`CPU_STEP_BIT];
|
| 1318 |
|
|
else if ((alu_pc_valid)&&(step)&&(gie))
|
| 1319 |
|
|
step <= 1'b0;
|
| 1320 |
|
|
|
| 1321 |
|
|
// The GIE register. Only interrupts can disable the interrupt register
|
| 1322 |
|
|
assign w_switch_to_interrupt = (gie)&&(
|
| 1323 |
|
|
// On interrupt (obviously)
|
| 1324 |
|
|
((i_interrupt)&&(~alu_phase)&&(~bus_lock))
|
| 1325 |
|
|
// If we are stepping the CPU
|
| 1326 |
|
|
||((alu_pc_valid)&&(step)&&(~alu_phase)&&(~bus_lock))
|
| 1327 |
|
|
// If we encounter a break instruction, if the break
|
| 1328 |
|
|
// enable isn't set.
|
| 1329 |
|
|
||((master_ce)&&(~mem_rdbusy)&&(~div_busy)&&(~fpu_busy)
|
| 1330 |
|
|
&&(op_break)&&(~break_en))
|
| 1331 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 1332 |
|
|
// On an illegal instruction
|
| 1333 |
|
|
||((alu_pc_valid)&&(alu_illegal))
|
| 1334 |
|
|
`endif
|
| 1335 |
|
|
// On division by zero. If the divide isn't
|
| 1336 |
|
|
// implemented, div_valid and div_error will be short
|
| 1337 |
|
|
// circuited and that logic will be bypassed
|
| 1338 |
|
|
||((div_valid)&&(div_error))
|
| 1339 |
|
|
// Same thing on a floating point error.
|
| 1340 |
|
|
||((fpu_valid)&&(fpu_error))
|
| 1341 |
|
|
//
|
| 1342 |
|
|
||(bus_err)
|
| 1343 |
|
|
// If we write to the CC register
|
| 1344 |
|
|
||((wr_reg_ce)&&(~wr_reg_vl[`CPU_GIE_BIT])
|
| 1345 |
|
|
&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1346 |
|
|
);
|
| 1347 |
|
|
assign w_release_from_interrupt = (~gie)&&(~i_interrupt)
|
| 1348 |
|
|
// Then if we write the CC register
|
| 1349 |
|
|
&&(((wr_reg_ce)&&(wr_reg_vl[`CPU_GIE_BIT])
|
| 1350 |
|
|
&&(~wr_reg_id[4])&&(wr_write_cc))
|
| 1351 |
|
|
);
|
| 1352 |
|
|
always @(posedge i_clk)
|
| 1353 |
|
|
if (i_rst)
|
| 1354 |
|
|
gie <= 1'b0;
|
| 1355 |
|
|
else if (w_switch_to_interrupt)
|
| 1356 |
|
|
gie <= 1'b0;
|
| 1357 |
|
|
else if (w_release_from_interrupt)
|
| 1358 |
|
|
gie <= 1'b1;
|
| 1359 |
|
|
|
| 1360 |
|
|
initial trap = 1'b0;
|
| 1361 |
|
|
always @(posedge i_clk)
|
| 1362 |
|
|
if (i_rst)
|
| 1363 |
|
|
trap <= 1'b0;
|
| 1364 |
|
|
else if ((alu_gie)&&(wr_reg_ce)&&(~wr_reg_vl[`CPU_GIE_BIT])
|
| 1365 |
|
|
&&(wr_write_cc)) // &&(wr_reg_id[4]) implied
|
| 1366 |
|
|
trap <= 1'b1;
|
| 1367 |
|
|
else if (w_release_from_interrupt)
|
| 1368 |
|
|
trap <= 1'b0;
|
| 1369 |
|
|
|
| 1370 |
|
|
`ifdef OPT_ILLEGAL_INSTRUCTION
|
| 1371 |
|
|
initial ill_err_i = 1'b0;
|
| 1372 |
|
|
always @(posedge i_clk)
|
| 1373 |
|
|
if (i_rst)
|
| 1374 |
|
|
ill_err_i <= 1'b0;
|
| 1375 |
|
|
// The debug interface can clear this bit
|
| 1376 |
|
|
else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
|
| 1377 |
|
|
&&(~wr_reg_vl[`CPU_ILL_BIT]))
|
| 1378 |
|
|
ill_err_i <= 1'b0;
|
| 1379 |
|
|
else if ((alu_pc_valid)&&(alu_illegal)&&(~alu_gie))
|
| 1380 |
|
|
ill_err_i <= 1'b1;
|
| 1381 |
|
|
initial ill_err_u = 1'b0;
|
| 1382 |
|
|
always @(posedge i_clk)
|
| 1383 |
|
|
if (i_rst)
|
| 1384 |
|
|
ill_err_u <= 1'b0;
|
| 1385 |
|
|
// The bit is automatically cleared on release from interrupt
|
| 1386 |
|
|
else if (w_release_from_interrupt)
|
| 1387 |
|
|
ill_err_u <= 1'b0;
|
| 1388 |
|
|
// If the supervisor writes to this register, clearing the
|
| 1389 |
|
|
// bit, then clear it
|
| 1390 |
|
|
else if (((~alu_gie)||(dbgv))
|
| 1391 |
|
|
&&(wr_reg_ce)&&(~wr_reg_vl[`CPU_ILL_BIT])
|
| 1392 |
|
|
&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1393 |
|
|
ill_err_u <= 1'b0;
|
| 1394 |
|
|
else if ((alu_pc_valid)&&(alu_illegal)&&(alu_gie))
|
| 1395 |
|
|
ill_err_u <= 1'b1;
|
| 1396 |
|
|
`else
|
| 1397 |
|
|
assign ill_err_u = 1'b0;
|
| 1398 |
|
|
assign ill_err_i = 1'b0;
|
| 1399 |
|
|
`endif
|
| 1400 |
|
|
// Supervisor/interrupt bus error flag -- this will crash the CPU if
|
| 1401 |
|
|
// ever set.
|
| 1402 |
|
|
initial ibus_err_flag = 1'b0;
|
| 1403 |
|
|
always @(posedge i_clk)
|
| 1404 |
|
|
if (i_rst)
|
| 1405 |
|
|
ibus_err_flag <= 1'b0;
|
| 1406 |
|
|
else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
|
| 1407 |
|
|
&&(~wr_reg_vl[`CPU_BUSERR_BIT]))
|
| 1408 |
|
|
ibus_err_flag <= 1'b0;
|
| 1409 |
|
|
else if ((bus_err)&&(~alu_gie))
|
| 1410 |
|
|
ibus_err_flag <= 1'b1;
|
| 1411 |
|
|
// User bus error flag -- if ever set, it will cause an interrupt to
|
| 1412 |
|
|
// supervisor mode.
|
| 1413 |
|
|
initial ubus_err_flag = 1'b0;
|
| 1414 |
|
|
always @(posedge i_clk)
|
| 1415 |
|
|
if (i_rst)
|
| 1416 |
|
|
ubus_err_flag <= 1'b0;
|
| 1417 |
|
|
else if (w_release_from_interrupt)
|
| 1418 |
|
|
ubus_err_flag <= 1'b0;
|
| 1419 |
|
|
// else if ((i_halt)&&(i_dbg_we)&&(~i_dbg_reg[4])
|
| 1420 |
|
|
// &&(i_dbg_reg == {1'b1, `CPU_CC_REG})
|
| 1421 |
|
|
// &&(~i_dbg_data[`CPU_BUSERR_BIT]))
|
| 1422 |
|
|
// ubus_err_flag <= 1'b0;
|
| 1423 |
|
|
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
|
| 1424 |
|
|
&&(~wr_reg_vl[`CPU_BUSERR_BIT])
|
| 1425 |
|
|
&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1426 |
|
|
ubus_err_flag <= 1'b0;
|
| 1427 |
|
|
else if ((bus_err)&&(alu_gie))
|
| 1428 |
|
|
ubus_err_flag <= 1'b1;
|
| 1429 |
|
|
|
| 1430 |
|
|
generate
|
| 1431 |
|
|
if (IMPLEMENT_DIVIDE != 0)
|
| 1432 |
|
|
begin
|
| 1433 |
|
|
reg r_idiv_err_flag, r_udiv_err_flag;
|
| 1434 |
|
|
|
| 1435 |
|
|
// Supervisor/interrupt divide (by zero) error flag -- this will
|
| 1436 |
|
|
// crash the CPU if ever set. This bit is thus available for us
|
| 1437 |
|
|
// to be able to tell if/why the CPU crashed.
|
| 1438 |
|
|
initial r_idiv_err_flag = 1'b0;
|
| 1439 |
|
|
always @(posedge i_clk)
|
| 1440 |
|
|
if (i_rst)
|
| 1441 |
|
|
r_idiv_err_flag <= 1'b0;
|
| 1442 |
|
|
else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
|
| 1443 |
|
|
&&(~wr_reg_vl[`CPU_DIVERR_BIT]))
|
| 1444 |
|
|
r_idiv_err_flag <= 1'b0;
|
| 1445 |
|
|
else if ((div_error)&&(div_valid)&&(~alu_gie))
|
| 1446 |
|
|
r_idiv_err_flag <= 1'b1;
|
| 1447 |
|
|
// User divide (by zero) error flag -- if ever set, it will
|
| 1448 |
|
|
// cause a sudden switch interrupt to supervisor mode.
|
| 1449 |
|
|
initial r_udiv_err_flag = 1'b0;
|
| 1450 |
|
|
always @(posedge i_clk)
|
| 1451 |
|
|
if (i_rst)
|
| 1452 |
|
|
r_udiv_err_flag <= 1'b0;
|
| 1453 |
|
|
else if (w_release_from_interrupt)
|
| 1454 |
|
|
r_udiv_err_flag <= 1'b0;
|
| 1455 |
|
|
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
|
| 1456 |
|
|
&&(~wr_reg_vl[`CPU_DIVERR_BIT])
|
| 1457 |
|
|
&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1458 |
|
|
r_udiv_err_flag <= 1'b0;
|
| 1459 |
|
|
else if ((div_error)&&(alu_gie)&&(div_valid))
|
| 1460 |
|
|
r_udiv_err_flag <= 1'b1;
|
| 1461 |
|
|
|
| 1462 |
|
|
assign idiv_err_flag = r_idiv_err_flag;
|
| 1463 |
|
|
assign udiv_err_flag = r_udiv_err_flag;
|
| 1464 |
|
|
end else begin
|
| 1465 |
|
|
assign idiv_err_flag = 1'b0;
|
| 1466 |
|
|
assign udiv_err_flag = 1'b0;
|
| 1467 |
|
|
end endgenerate
|
| 1468 |
|
|
|
| 1469 |
|
|
generate
|
| 1470 |
|
|
if (IMPLEMENT_FPU !=0)
|
| 1471 |
|
|
begin
|
| 1472 |
|
|
// Supervisor/interrupt floating point error flag -- this will
|
| 1473 |
|
|
// crash the CPU if ever set.
|
| 1474 |
|
|
reg r_ifpu_err_flag, r_ufpu_err_flag;
|
| 1475 |
|
|
initial r_ifpu_err_flag = 1'b0;
|
| 1476 |
|
|
always @(posedge i_clk)
|
| 1477 |
|
|
if (i_rst)
|
| 1478 |
|
|
r_ifpu_err_flag <= 1'b0;
|
| 1479 |
|
|
else if ((dbgv)&&(wr_reg_id == {1'b0, `CPU_CC_REG})
|
| 1480 |
|
|
&&(~wr_reg_vl[`CPU_FPUERR_BIT]))
|
| 1481 |
|
|
r_ifpu_err_flag <= 1'b0;
|
| 1482 |
|
|
else if ((fpu_error)&&(fpu_valid)&&(~alu_gie))
|
| 1483 |
|
|
r_ifpu_err_flag <= 1'b1;
|
| 1484 |
|
|
// User floating point error flag -- if ever set, it will cause
|
| 1485 |
|
|
// a sudden switch interrupt to supervisor mode.
|
| 1486 |
|
|
initial r_ufpu_err_flag = 1'b0;
|
| 1487 |
|
|
always @(posedge i_clk)
|
| 1488 |
|
|
if (i_rst)
|
| 1489 |
|
|
r_ufpu_err_flag <= 1'b0;
|
| 1490 |
|
|
else if (w_release_from_interrupt)
|
| 1491 |
|
|
r_ufpu_err_flag <= 1'b0;
|
| 1492 |
|
|
else if (((~alu_gie)||(dbgv))&&(wr_reg_ce)
|
| 1493 |
|
|
&&(~wr_reg_vl[`CPU_FPUERR_BIT])
|
| 1494 |
|
|
&&(wr_reg_id[4])&&(wr_write_cc))
|
| 1495 |
|
|
r_ufpu_err_flag <= 1'b0;
|
| 1496 |
|
|
else if ((fpu_error)&&(alu_gie)&&(fpu_valid))
|
| 1497 |
|
|
r_ufpu_err_flag <= 1'b1;
|
| 1498 |
|
|
|
| 1499 |
|
|
assign ifpu_err_flag = r_ifpu_err_flag;
|
| 1500 |
|
|
assign ufpu_err_flag = r_ufpu_err_flag;
|
| 1501 |
|
|
end else begin
|
| 1502 |
|
|
assign ifpu_err_flag = 1'b0;
|
| 1503 |
|
|
assign ufpu_err_flag = 1'b0;
|
| 1504 |
|
|
end endgenerate
|
| 1505 |
|
|
|
| 1506 |
|
|
`ifdef OPT_VLIW
|
| 1507 |
|
|
reg r_ihalt_phase, r_uhalt_phase;
|
| 1508 |
|
|
|
| 1509 |
|
|
initial r_ihalt_phase = 0;
|
| 1510 |
|
|
initial r_uhalt_phase = 0;
|
| 1511 |
|
|
always @(posedge i_clk)
|
| 1512 |
|
|
if (~alu_gie)
|
| 1513 |
|
|
r_ihalt_phase <= alu_phase;
|
| 1514 |
|
|
always @(posedge i_clk)
|
| 1515 |
|
|
if (alu_gie)
|
| 1516 |
|
|
r_uhalt_phase <= alu_phase;
|
| 1517 |
|
|
else if (w_release_from_interrupt)
|
| 1518 |
|
|
r_uhalt_phase <= 1'b0;
|
| 1519 |
|
|
|
| 1520 |
|
|
assign ihalt_phase = r_ihalt_phase;
|
| 1521 |
|
|
assign uhalt_phase = r_uhalt_phase;
|
| 1522 |
|
|
`else
|
| 1523 |
|
|
assign ihalt_phase = 1'b0;
|
| 1524 |
|
|
assign uhalt_phase = 1'b0;
|
| 1525 |
|
|
`endif
|
| 1526 |
|
|
|
| 1527 |
|
|
//
|
| 1528 |
|
|
// Write backs to the PC register, and general increments of it
|
| 1529 |
|
|
// We support two: upc and ipc. If the instruction is normal,
|
| 1530 |
|
|
// we increment upc, if interrupt level we increment ipc. If
|
| 1531 |
|
|
// the instruction writes the PC, we write whichever PC is appropriate.
|
| 1532 |
|
|
//
|
| 1533 |
|
|
// Do we need to all our partial results from the pipeline?
|
| 1534 |
|
|
// What happens when the pipeline has gie and ~gie instructions within
|
| 1535 |
|
|
// it? Do we clear both? What if a gie instruction tries to clear
|
| 1536 |
|
|
// a non-gie instruction?
|
| 1537 |
|
|
always @(posedge i_clk)
|
| 1538 |
|
|
if ((wr_reg_ce)&&(wr_reg_id[4])&&(wr_write_pc))
|
| 1539 |
|
|
upc <= wr_reg_vl[(AW-1):0];
|
| 1540 |
|
|
else if ((alu_gie)&&(alu_pc_valid)&&(~clear_pipeline))
|
| 1541 |
|
|
upc <= alu_pc;
|
| 1542 |
|
|
|
| 1543 |
|
|
always @(posedge i_clk)
|
| 1544 |
|
|
if (i_rst)
|
| 1545 |
|
|
ipc <= RESET_ADDRESS;
|
| 1546 |
|
|
else if ((wr_reg_ce)&&(~wr_reg_id[4])&&(wr_write_pc))
|
| 1547 |
|
|
ipc <= wr_reg_vl[(AW-1):0];
|
| 1548 |
|
|
else if ((~alu_gie)&&(alu_pc_valid)&&(~clear_pipeline))
|
| 1549 |
|
|
ipc <= alu_pc;
|
| 1550 |
|
|
|
| 1551 |
|
|
always @(posedge i_clk)
|
| 1552 |
|
|
if (i_rst)
|
| 1553 |
|
|
pf_pc <= RESET_ADDRESS;
|
| 1554 |
|
|
else if (w_switch_to_interrupt)
|
| 1555 |
|
|
pf_pc <= ipc;
|
| 1556 |
|
|
else if (w_release_from_interrupt)
|
| 1557 |
|
|
pf_pc <= upc;
|
| 1558 |
|
|
else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
|
| 1559 |
|
|
pf_pc <= wr_reg_vl[(AW-1):0];
|
| 1560 |
|
|
`ifdef OPT_PIPELINED
|
| 1561 |
|
|
else if ((dcd_early_branch)&&(~clear_pipeline))
|
| 1562 |
|
|
pf_pc <= dcd_branch_pc + 1;
|
| 1563 |
|
|
else if ((new_pc)||((~dcd_stalled)&&(pf_valid)))
|
| 1564 |
|
|
pf_pc <= pf_pc + {{(AW-1){1'b0}},1'b1};
|
| 1565 |
|
|
`else
|
| 1566 |
|
|
else if ((alu_pc_valid)&&(~clear_pipeline))
|
| 1567 |
|
|
pf_pc <= alu_pc;
|
| 1568 |
|
|
`endif
|
| 1569 |
|
|
|
| 1570 |
|
|
initial new_pc = 1'b1;
|
| 1571 |
|
|
always @(posedge i_clk)
|
| 1572 |
|
|
if ((i_rst)||(i_clear_pf_cache))
|
| 1573 |
|
|
new_pc <= 1'b1;
|
| 1574 |
|
|
else if (w_switch_to_interrupt)
|
| 1575 |
|
|
new_pc <= 1'b1;
|
| 1576 |
|
|
else if (w_release_from_interrupt)
|
| 1577 |
|
|
new_pc <= 1'b1;
|
| 1578 |
|
|
else if ((wr_reg_ce)&&(wr_reg_id[4] == gie)&&(wr_write_pc))
|
| 1579 |
|
|
new_pc <= 1'b1;
|
| 1580 |
|
|
else
|
| 1581 |
|
|
new_pc <= 1'b0;
|
| 1582 |
|
|
|
| 1583 |
|
|
//
|
| 1584 |
|
|
// The debug interface
|
| 1585 |
|
|
generate
|
| 1586 |
|
|
if (AW<32)
|
| 1587 |
|
|
begin
|
| 1588 |
|
|
always @(posedge i_clk)
|
| 1589 |
|
|
begin
|
| 1590 |
|
|
o_dbg_reg <= regset[i_dbg_reg];
|
| 1591 |
|
|
if (i_dbg_reg[3:0] == `CPU_PC_REG)
|
| 1592 |
|
|
o_dbg_reg <= {{(32-AW){1'b0}},(i_dbg_reg[4])?upc:ipc};
|
| 1593 |
|
|
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
|
| 1594 |
|
|
begin
|
| 1595 |
|
|
o_dbg_reg[13:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
|
| 1596 |
|
|
o_dbg_reg[`CPU_GIE_BIT] <= gie;
|
| 1597 |
|
|
end
|
| 1598 |
|
|
end
|
| 1599 |
|
|
end else begin
|
| 1600 |
|
|
always @(posedge i_clk)
|
| 1601 |
|
|
begin
|
| 1602 |
|
|
o_dbg_reg <= regset[i_dbg_reg];
|
| 1603 |
|
|
if (i_dbg_reg[3:0] == `CPU_PC_REG)
|
| 1604 |
|
|
o_dbg_reg <= (i_dbg_reg[4])?upc:ipc;
|
| 1605 |
|
|
else if (i_dbg_reg[3:0] == `CPU_CC_REG)
|
| 1606 |
|
|
begin
|
| 1607 |
|
|
o_dbg_reg[13:0] <= (i_dbg_reg[4])?w_uflags:w_iflags;
|
| 1608 |
|
|
o_dbg_reg[`CPU_GIE_BIT] <= gie;
|
| 1609 |
|
|
end
|
| 1610 |
|
|
end
|
| 1611 |
|
|
end endgenerate
|
| 1612 |
|
|
|
| 1613 |
|
|
always @(posedge i_clk)
|
| 1614 |
|
|
o_dbg_cc <= { o_break, bus_err, gie, sleep };
|
| 1615 |
|
|
|
| 1616 |
|
|
always @(posedge i_clk)
|
| 1617 |
|
|
o_dbg_stall <= (i_halt)&&(
|
| 1618 |
|
|
(pf_cyc)||(mem_cyc_gbl)||(mem_cyc_lcl)||(mem_busy)
|
| 1619 |
|
|
||((~opvalid)&&(~i_rst))
|
| 1620 |
|
|
||((~dcdvalid)&&(~i_rst)));
|
| 1621 |
|
|
|
| 1622 |
|
|
//
|
| 1623 |
|
|
//
|
| 1624 |
|
|
// Produce accounting outputs: Account for any CPU stalls, so we can
|
| 1625 |
|
|
// later evaluate how well we are doing.
|
| 1626 |
|
|
//
|
| 1627 |
|
|
//
|
| 1628 |
|
|
assign o_op_stall = (master_ce)&&(op_stall);
|
| 1629 |
|
|
assign o_pf_stall = (master_ce)&&(~pf_valid);
|
| 1630 |
|
|
assign o_i_count = (alu_pc_valid)&&(~clear_pipeline);
|
| 1631 |
|
|
|
| 1632 |
|
|
`ifdef DEBUG_SCOPE
|
| 1633 |
|
|
always @(posedge i_clk)
|
| 1634 |
|
|
o_debug <= {
|
| 1635 |
|
|
pf_pc[3:0], flags,
|
| 1636 |
|
|
pf_valid, dcdvalid, opvalid, alu_valid, mem_valid,
|
| 1637 |
|
|
op_ce, alu_ce, mem_ce,
|
| 1638 |
|
|
//
|
| 1639 |
|
|
master_ce, opvalid_alu, opvalid_mem,
|
| 1640 |
|
|
//
|
| 1641 |
|
|
alu_stall, mem_busy, op_pipe, mem_pipe_stalled,
|
| 1642 |
|
|
mem_we,
|
| 1643 |
|
|
// ((opvalid_alu)&&(alu_stall))
|
| 1644 |
|
|
// ||((opvalid_mem)&&(~op_pipe)&&(mem_busy))
|
| 1645 |
|
|
// ||((opvalid_mem)&&( op_pipe)&&(mem_pipe_stalled)));
|
| 1646 |
|
|
// opA[23:20], opA[3:0],
|
| 1647 |
|
|
gie, sleep,
|
| 1648 |
|
|
wr_reg_ce, wr_reg_vl[4:0]
|
| 1649 |
|
|
/*
|
| 1650 |
|
|
i_rst, master_ce, (new_pc),
|
| 1651 |
|
|
((dcd_early_branch)&&(dcdvalid)),
|
| 1652 |
|
|
pf_valid, pf_illegal,
|
| 1653 |
|
|
op_ce, dcd_ce, dcdvalid, dcd_stalled,
|
| 1654 |
|
|
pf_cyc, pf_stb, pf_we, pf_ack, pf_stall, pf_err,
|
| 1655 |
|
|
pf_pc[7:0], pf_addr[7:0]
|
| 1656 |
|
|
*/
|
| 1657 |
|
|
};
|
| 1658 |
|
|
`endif
|
| 1659 |
|
|
|
| 1660 |
|
|
endmodule
|
