Sun, 06 Mar 2011 21:03:32 +0000
Commit GSI patches from Wesley Terpstra
- Add JTAG capture pin
==> allows removing sensitivity to reg_update which caused clocking problems making JTAG unstable
- Use register file backed by RAM blocks
==> saves quite some area and speed on altera
... be sure to enable it using `define CFG_EBR_POSEDGE_REGISTER_FILE
- Fix a minor problem where compilation fails when interrupts are not supported
- Add support to flush icache and dcache per JTAG
- Fix wrong width assignments for PC
Multiplier patch has been left out for now; don't the design synthesizers (Quartus / Xst) split the multiply automatically?
Original-Author: Wesley Terpstra <w.terpsta gsi.de>
Original-Source: Milkymist mailing list postings, 2011-02-28 (11:19 and 13:32) and 2011-03-01
Original-Message-Ids: <4D6B84B5.9040604@gsi.de> <4D6BA3E4.3020609@gsi.de> <4D6CFFF2.6030703@gsi.de>
philpem@0 | 1 | // ============================================================================= |
philpem@0 | 2 | // COPYRIGHT NOTICE |
philpem@0 | 3 | // Copyright 2006 (c) Lattice Semiconductor Corporation |
philpem@0 | 4 | // ALL RIGHTS RESERVED |
philpem@0 | 5 | // This confidential and proprietary software may be used only as authorised by |
philpem@0 | 6 | // a licensing agreement from Lattice Semiconductor Corporation. |
philpem@0 | 7 | // The entire notice above must be reproduced on all authorized copies and |
philpem@0 | 8 | // copies may only be made to the extent permitted by a licensing agreement from |
philpem@0 | 9 | // Lattice Semiconductor Corporation. |
philpem@0 | 10 | // |
philpem@0 | 11 | // Lattice Semiconductor Corporation TEL : 1-800-Lattice (USA and Canada) |
philpem@0 | 12 | // 5555 NE Moore Court 408-826-6000 (other locations) |
philpem@0 | 13 | // Hillsboro, OR 97124 web : http://www.latticesemi.com/ |
philpem@0 | 14 | // U.S.A email: techsupport@latticesemi.com |
philpem@0 | 15 | // =============================================================================/ |
philpem@0 | 16 | // FILE DETAILS |
philpem@0 | 17 | // Project : LatticeMico32 |
philpem@0 | 18 | // File : lm_mc_arithmetic.v |
philpem@0 | 19 | // Title : Multi-cycle arithmetic unit. |
philpem@0 | 20 | // Dependencies : lm32_include.v |
philpem@0 | 21 | // Version : 6.1.17 |
philpem@0 | 22 | // : Initial Release |
philpem@0 | 23 | // Version : 7.0SP2, 3.0 |
philpem@0 | 24 | // : No Change |
philpem@0 | 25 | // Version : 3.1 |
philpem@0 | 26 | // : No Change |
philpem@0 | 27 | // ============================================================================= |
philpem@0 | 28 | |
philpem@0 | 29 | `include "lm32_include.v" |
philpem@0 | 30 | |
philpem@0 | 31 | `define LM32_MC_STATE_RNG 2:0 |
philpem@0 | 32 | `define LM32_MC_STATE_IDLE 3'b000 |
philpem@0 | 33 | `define LM32_MC_STATE_MULTIPLY 3'b001 |
philpem@0 | 34 | `define LM32_MC_STATE_MODULUS 3'b010 |
philpem@0 | 35 | `define LM32_MC_STATE_DIVIDE 3'b011 |
philpem@0 | 36 | `define LM32_MC_STATE_SHIFT_LEFT 3'b100 |
philpem@0 | 37 | `define LM32_MC_STATE_SHIFT_RIGHT 3'b101 |
philpem@0 | 38 | |
philpem@0 | 39 | ///////////////////////////////////////////////////// |
philpem@0 | 40 | // Module interface |
philpem@0 | 41 | ///////////////////////////////////////////////////// |
philpem@0 | 42 | |
philpem@0 | 43 | module lm32_mc_arithmetic ( |
philpem@0 | 44 | // ----- Inputs ----- |
philpem@0 | 45 | clk_i, |
philpem@0 | 46 | rst_i, |
philpem@0 | 47 | stall_d, |
philpem@0 | 48 | kill_x, |
philpem@0 | 49 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 50 | divide_d, |
philpem@0 | 51 | modulus_d, |
philpem@0 | 52 | `endif |
philpem@0 | 53 | `ifdef CFG_MC_MULTIPLY_ENABLED |
philpem@0 | 54 | multiply_d, |
philpem@0 | 55 | `endif |
philpem@0 | 56 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 57 | shift_left_d, |
philpem@0 | 58 | shift_right_d, |
philpem@0 | 59 | sign_extend_d, |
philpem@0 | 60 | `endif |
philpem@0 | 61 | operand_0_d, |
philpem@0 | 62 | operand_1_d, |
philpem@0 | 63 | // ----- Ouputs ----- |
philpem@0 | 64 | result_x, |
philpem@0 | 65 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 66 | divide_by_zero_x, |
philpem@0 | 67 | `endif |
philpem@0 | 68 | stall_request_x |
philpem@0 | 69 | ); |
philpem@0 | 70 | |
philpem@0 | 71 | ///////////////////////////////////////////////////// |
philpem@0 | 72 | // Inputs |
philpem@0 | 73 | ///////////////////////////////////////////////////// |
philpem@0 | 74 | |
philpem@0 | 75 | input clk_i; // Clock |
philpem@0 | 76 | input rst_i; // Reset |
philpem@0 | 77 | input stall_d; // Stall instruction in D stage |
philpem@0 | 78 | input kill_x; // Kill instruction in X stage |
philpem@0 | 79 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 80 | input divide_d; // Perform divide |
philpem@0 | 81 | input modulus_d; // Perform modulus |
philpem@0 | 82 | `endif |
philpem@0 | 83 | `ifdef CFG_MC_MULTIPLY_ENABLED |
philpem@0 | 84 | input multiply_d; // Perform multiply |
philpem@0 | 85 | `endif |
philpem@0 | 86 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 87 | input shift_left_d; // Perform left shift |
philpem@0 | 88 | input shift_right_d; // Perform right shift |
philpem@0 | 89 | input sign_extend_d; // Whether to sign-extend (arithmetic) or zero-extend (logical) |
philpem@0 | 90 | `endif |
philpem@0 | 91 | input [`LM32_WORD_RNG] operand_0_d; |
philpem@0 | 92 | input [`LM32_WORD_RNG] operand_1_d; |
philpem@0 | 93 | |
philpem@0 | 94 | ///////////////////////////////////////////////////// |
philpem@0 | 95 | // Outputs |
philpem@0 | 96 | ///////////////////////////////////////////////////// |
philpem@0 | 97 | |
philpem@0 | 98 | output [`LM32_WORD_RNG] result_x; // Result of operation |
philpem@0 | 99 | reg [`LM32_WORD_RNG] result_x; |
philpem@0 | 100 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 101 | output divide_by_zero_x; // A divide by zero was attempted |
philpem@0 | 102 | reg divide_by_zero_x; |
philpem@0 | 103 | `endif |
philpem@0 | 104 | output stall_request_x; // Request to stall pipeline from X stage back |
philpem@0 | 105 | wire stall_request_x; |
philpem@0 | 106 | |
philpem@0 | 107 | ///////////////////////////////////////////////////// |
philpem@0 | 108 | // Internal nets and registers |
philpem@0 | 109 | ///////////////////////////////////////////////////// |
philpem@0 | 110 | |
philpem@0 | 111 | reg [`LM32_WORD_RNG] p; // Temporary registers |
philpem@0 | 112 | reg [`LM32_WORD_RNG] a; |
philpem@0 | 113 | reg [`LM32_WORD_RNG] b; |
philpem@0 | 114 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 115 | wire [32:0] t; |
philpem@0 | 116 | `endif |
philpem@0 | 117 | |
philpem@0 | 118 | reg [`LM32_MC_STATE_RNG] state; // Current state of FSM |
philpem@0 | 119 | reg [5:0] cycles; // Number of cycles remaining in the operation |
philpem@0 | 120 | |
philpem@0 | 121 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 122 | reg sign_extend_x; // Whether to sign extend of zero extend right shifts |
philpem@0 | 123 | wire fill_value; // Value to fill with for right barrel-shifts |
philpem@0 | 124 | `endif |
philpem@0 | 125 | |
philpem@0 | 126 | ///////////////////////////////////////////////////// |
philpem@0 | 127 | // Combinational logic |
philpem@0 | 128 | ///////////////////////////////////////////////////// |
philpem@0 | 129 | |
philpem@0 | 130 | // Stall pipeline while any operation is being performed |
philpem@0 | 131 | assign stall_request_x = state != `LM32_MC_STATE_IDLE; |
philpem@0 | 132 | |
philpem@0 | 133 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 134 | // Subtraction |
philpem@0 | 135 | assign t = {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]} - b; |
philpem@0 | 136 | `endif |
philpem@0 | 137 | |
philpem@0 | 138 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 139 | // Determine fill value for right shift - Sign bit for arithmetic shift, or zero for logical shift |
philpem@0 | 140 | assign fill_value = (sign_extend_x == `TRUE) & b[`LM32_WORD_WIDTH-1]; |
philpem@0 | 141 | `endif |
philpem@0 | 142 | |
philpem@0 | 143 | ///////////////////////////////////////////////////// |
philpem@0 | 144 | // Sequential logic |
philpem@0 | 145 | ///////////////////////////////////////////////////// |
philpem@0 | 146 | |
philpem@0 | 147 | // Perform right shift |
philpem@0 | 148 | always @(posedge clk_i `CFG_RESET_SENSITIVITY) |
philpem@0 | 149 | begin |
philpem@0 | 150 | if (rst_i == `TRUE) |
philpem@0 | 151 | begin |
philpem@0 | 152 | cycles <= {6{1'b0}}; |
philpem@0 | 153 | p <= {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 154 | a <= {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 155 | b <= {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 156 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 157 | sign_extend_x <= 1'b0; |
philpem@0 | 158 | `endif |
philpem@0 | 159 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 160 | divide_by_zero_x <= `FALSE; |
philpem@0 | 161 | `endif |
philpem@0 | 162 | result_x <= {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 163 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 164 | end |
philpem@0 | 165 | else |
philpem@0 | 166 | begin |
philpem@0 | 167 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 168 | divide_by_zero_x <= `FALSE; |
philpem@0 | 169 | `endif |
philpem@0 | 170 | case (state) |
philpem@0 | 171 | `LM32_MC_STATE_IDLE: |
philpem@0 | 172 | begin |
philpem@0 | 173 | if (stall_d == `FALSE) |
philpem@0 | 174 | begin |
philpem@0 | 175 | cycles <= `LM32_WORD_WIDTH; |
philpem@0 | 176 | p <= 32'b0; |
philpem@0 | 177 | a <= operand_0_d; |
philpem@0 | 178 | b <= operand_1_d; |
philpem@0 | 179 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 180 | if (divide_d == `TRUE) |
philpem@0 | 181 | state <= `LM32_MC_STATE_DIVIDE; |
philpem@0 | 182 | if (modulus_d == `TRUE) |
philpem@0 | 183 | state <= `LM32_MC_STATE_MODULUS; |
philpem@0 | 184 | `endif |
philpem@0 | 185 | `ifdef CFG_MC_MULTIPLY_ENABLED |
philpem@0 | 186 | if (multiply_d == `TRUE) |
philpem@0 | 187 | state <= `LM32_MC_STATE_MULTIPLY; |
philpem@0 | 188 | `endif |
philpem@0 | 189 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 190 | if (shift_left_d == `TRUE) |
philpem@0 | 191 | begin |
philpem@0 | 192 | state <= `LM32_MC_STATE_SHIFT_LEFT; |
philpem@0 | 193 | sign_extend_x <= sign_extend_d; |
philpem@0 | 194 | cycles <= operand_1_d[4:0]; |
philpem@0 | 195 | a <= operand_0_d; |
philpem@0 | 196 | b <= operand_0_d; |
philpem@0 | 197 | end |
philpem@0 | 198 | if (shift_right_d == `TRUE) |
philpem@0 | 199 | begin |
philpem@0 | 200 | state <= `LM32_MC_STATE_SHIFT_RIGHT; |
philpem@0 | 201 | sign_extend_x <= sign_extend_d; |
philpem@0 | 202 | cycles <= operand_1_d[4:0]; |
philpem@0 | 203 | a <= operand_0_d; |
philpem@0 | 204 | b <= operand_0_d; |
philpem@0 | 205 | end |
philpem@0 | 206 | `endif |
philpem@0 | 207 | end |
philpem@0 | 208 | end |
philpem@0 | 209 | `ifdef CFG_MC_DIVIDE_ENABLED |
philpem@0 | 210 | `LM32_MC_STATE_DIVIDE: |
philpem@0 | 211 | begin |
philpem@0 | 212 | if (t[32] == 1'b0) |
philpem@0 | 213 | begin |
philpem@0 | 214 | p <= t[31:0]; |
philpem@0 | 215 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b1}; |
philpem@0 | 216 | end |
philpem@0 | 217 | else |
philpem@0 | 218 | begin |
philpem@0 | 219 | p <= {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]}; |
philpem@0 | 220 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0}; |
philpem@0 | 221 | end |
philpem@0 | 222 | result_x <= a; |
philpem@0 | 223 | if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE)) |
philpem@0 | 224 | begin |
philpem@0 | 225 | // Check for divide by zero |
philpem@0 | 226 | divide_by_zero_x <= b == {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 227 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 228 | end |
philpem@0 | 229 | cycles <= cycles - 1'b1; |
philpem@0 | 230 | end |
philpem@0 | 231 | `LM32_MC_STATE_MODULUS: |
philpem@0 | 232 | begin |
philpem@0 | 233 | if (t[32] == 1'b0) |
philpem@0 | 234 | begin |
philpem@0 | 235 | p <= t[31:0]; |
philpem@0 | 236 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b1}; |
philpem@0 | 237 | end |
philpem@0 | 238 | else |
philpem@0 | 239 | begin |
philpem@0 | 240 | p <= {p[`LM32_WORD_WIDTH-2:0], a[`LM32_WORD_WIDTH-1]}; |
philpem@0 | 241 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0}; |
philpem@0 | 242 | end |
philpem@0 | 243 | result_x <= p; |
philpem@0 | 244 | if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE)) |
philpem@0 | 245 | begin |
philpem@0 | 246 | // Check for divide by zero |
philpem@0 | 247 | divide_by_zero_x <= b == {`LM32_WORD_WIDTH{1'b0}}; |
philpem@0 | 248 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 249 | end |
philpem@0 | 250 | cycles <= cycles - 1'b1; |
philpem@0 | 251 | end |
philpem@0 | 252 | `endif |
philpem@0 | 253 | `ifdef CFG_MC_MULTIPLY_ENABLED |
philpem@0 | 254 | `LM32_MC_STATE_MULTIPLY: |
philpem@0 | 255 | begin |
philpem@0 | 256 | if (b[0] == 1'b1) |
philpem@0 | 257 | p <= p + a; |
philpem@0 | 258 | b <= {1'b0, b[`LM32_WORD_WIDTH-1:1]}; |
philpem@0 | 259 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0}; |
philpem@0 | 260 | result_x <= p; |
philpem@0 | 261 | if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE)) |
philpem@0 | 262 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 263 | cycles <= cycles - 1'b1; |
philpem@0 | 264 | end |
philpem@0 | 265 | `endif |
philpem@0 | 266 | `ifdef CFG_MC_BARREL_SHIFT_ENABLED |
philpem@0 | 267 | `LM32_MC_STATE_SHIFT_LEFT: |
philpem@0 | 268 | begin |
philpem@0 | 269 | a <= {a[`LM32_WORD_WIDTH-2:0], 1'b0}; |
philpem@0 | 270 | result_x <= a; |
philpem@0 | 271 | if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE)) |
philpem@0 | 272 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 273 | cycles <= cycles - 1'b1; |
philpem@0 | 274 | end |
philpem@0 | 275 | `LM32_MC_STATE_SHIFT_RIGHT: |
philpem@0 | 276 | begin |
philpem@0 | 277 | b <= {fill_value, b[`LM32_WORD_WIDTH-1:1]}; |
philpem@0 | 278 | result_x <= b; |
philpem@0 | 279 | if ((cycles == `LM32_WORD_WIDTH'd0) || (kill_x == `TRUE)) |
philpem@0 | 280 | state <= `LM32_MC_STATE_IDLE; |
philpem@0 | 281 | cycles <= cycles - 1'b1; |
philpem@0 | 282 | end |
philpem@0 | 283 | `endif |
philpem@0 | 284 | endcase |
philpem@0 | 285 | end |
philpem@0 | 286 | end |
philpem@0 | 287 | |
philpem@0 | 288 | endmodule |