1 ---------------------------------------------------------------------
\r
2 -- TITLE: Register Bank
\r
3 -- AUTHOR: Steve Rhoads (rhoadss@yahoo.com)
\r
4 -- DATE CREATED: 2/2/01
\r
5 -- FILENAME: reg_bank.vhd
\r
6 -- PROJECT: Plasma CPU core
\r
7 -- COPYRIGHT: Software placed into the public domain by the author.
\r
8 -- Software 'as is' without warranty. Author liable for nothing.
\r
10 -- Implements a register bank with 32 registers that are 32-bits wide.
\r
11 -- There are two read-ports and one write port.
\r
12 ---------------------------------------------------------------------
\r
14 use ieee.std_logic_1164.all;
\r
15 use ieee.std_logic_unsigned.all;
\r
16 use work.mlite_pack.all;
\r
17 --library UNISIM; --May need to uncomment for ModelSim
\r
18 --use UNISIM.vcomponents.all; --May need to uncomment for ModelSim
\r
21 generic(memory_type : string := "XILINX_16X");
\r
22 port(clk : in std_logic;
\r
23 reset_in : in std_logic;
\r
24 pause : in std_logic;
\r
25 rs_index : in std_logic_vector(5 downto 0);
\r
26 rt_index : in std_logic_vector(5 downto 0);
\r
27 rd_index : in std_logic_vector(5 downto 0);
\r
28 reg_source_out : out std_logic_vector(31 downto 0);
\r
29 reg_target_out : out std_logic_vector(31 downto 0);
\r
30 reg_dest_new : in std_logic_vector(31 downto 0);
\r
31 intr_enable : out std_logic);
\r
32 end; --entity reg_bank
\r
35 --------------------------------------------------------------------
\r
36 -- The ram_block architecture attempts to use TWO dual-port memories.
\r
37 -- Different FPGAs and ASICs need different implementations.
\r
38 -- Choose one of the RAM implementations below.
\r
39 -- I need feedback on this section!
\r
40 --------------------------------------------------------------------
\r
41 architecture ram_block of reg_bank is
\r
42 signal intr_enable_reg : std_logic;
\r
43 type ram_type is array(31 downto 0) of std_logic_vector(31 downto 0);
\r
45 --controls access to dual-port memories
\r
46 signal addr_read1, addr_read2 : std_logic_vector(4 downto 0);
\r
47 signal addr_write : std_logic_vector(4 downto 0);
\r
48 signal data_out1, data_out2 : std_logic_vector(31 downto 0);
\r
49 signal write_enable : std_logic;
\r
53 reg_proc: process(clk, rs_index, rt_index, rd_index, reg_dest_new,
\r
54 intr_enable_reg, data_out1, data_out2, reset_in, pause)
\r
56 --setup for first dual-port memory
\r
57 if rs_index = "101110" then --reg_epc CP0 14
\r
58 addr_read1 <= "00000";
\r
60 addr_read1 <= rs_index(4 downto 0);
\r
63 when "000000" => reg_source_out <= ZERO;
\r
64 when "101100" => reg_source_out <= ZERO(31 downto 1) & intr_enable_reg;
\r
65 --interrupt vector address = 0x3c
\r
66 when "111111" => reg_source_out <= ZERO(31 downto 8) & "00111100";
\r
67 when others => reg_source_out <= data_out1;
\r
70 --setup for second dual-port memory
\r
71 addr_read2 <= rt_index(4 downto 0);
\r
73 when "000000" => reg_target_out <= ZERO;
\r
74 when others => reg_target_out <= data_out2;
\r
77 --setup write port for both dual-port memories
\r
78 if rd_index /= "000000" and rd_index /= "101100" and pause = '0' then
\r
79 write_enable <= '1';
\r
81 write_enable <= '0';
\r
83 if rd_index = "101110" then --reg_epc CP0 14
\r
84 addr_write <= "00000";
\r
86 addr_write <= rd_index(4 downto 0);
\r
89 if reset_in = '1' then
\r
90 intr_enable_reg <= '0';
\r
91 elsif rising_edge(clk) then
\r
92 if rd_index = "101110" then --reg_epc CP0 14
\r
93 intr_enable_reg <= '0'; --disable interrupts
\r
94 elsif rd_index = "101100" then
\r
95 intr_enable_reg <= reg_dest_new(0);
\r
99 intr_enable <= intr_enable_reg;
\r
103 --------------------------------------------------------------
\r
104 ---- Pick only ONE of the dual-port RAM implementations below!
\r
105 --------------------------------------------------------------
\r
108 -- One tri-port RAM, two read-ports, one write-port
\r
109 -- 32 registers 32-bits wide
\r
111 if memory_type = "TRI_PORT_X" generate
\r
112 ram_proc: process(clk, addr_read1, addr_read2,
\r
113 addr_write, reg_dest_new, write_enable)
\r
114 variable tri_port_ram : ram_type := (others => ZERO);
\r
116 data_out1 <= tri_port_ram(conv_integer(addr_read1));
\r
117 data_out2 <= tri_port_ram(conv_integer(addr_read2));
\r
118 if rising_edge(clk) then
\r
119 if write_enable = '1' then
\r
120 tri_port_ram(conv_integer(addr_write)) := reg_dest_new;
\r
124 end generate; --tri_port_mem
\r
128 -- Two dual-port RAMs, each with one read-port and one write-port
\r
130 if memory_type = "DUAL_PORT_" generate
\r
131 ram_proc2: process(clk, addr_read1, addr_read2,
\r
132 addr_write, reg_dest_new, write_enable)
\r
133 variable dual_port_ram1 : ram_type := (others => ZERO);
\r
134 variable dual_port_ram2 : ram_type := (others => ZERO);
\r
136 data_out1 <= dual_port_ram1(conv_integer(addr_read1));
\r
137 data_out2 <= dual_port_ram2(conv_integer(addr_read2));
\r
138 if rising_edge(clk) then
\r
139 if write_enable = '1' then
\r
140 dual_port_ram1(conv_integer(addr_write)) := reg_dest_new;
\r
141 dual_port_ram2(conv_integer(addr_write)) := reg_dest_new;
\r
145 end generate; --dual_port_mem
\r
149 -- RAM16X1D: 16 x 1 positive edge write, asynchronous read dual-port
\r
150 -- distributed RAM for all Xilinx FPGAs
\r
151 -- From library UNISIM; use UNISIM.vcomponents.all;
\r
153 if memory_type = "XILINX_16X" generate
\r
154 signal data_out1A, data_out1B : std_logic_vector(31 downto 0);
\r
155 signal data_out2A, data_out2B : std_logic_vector(31 downto 0);
\r
156 signal weA, weB : std_logic;
\r
157 signal no_connect : std_logic_vector(127 downto 0);
\r
159 weA <= write_enable and not addr_write(4); --lower 16 registers
\r
160 weB <= write_enable and addr_write(4); --upper 16 registers
\r
162 reg_loop: for i in 0 to 31 generate
\r
164 --Read port 1 lower 16 registers
\r
165 reg_bit1a : RAM16X1D
\r
167 WCLK => clk, -- Port A write clock input
\r
168 WE => weA, -- Port A write enable input
\r
169 A0 => addr_write(0), -- Port A address[0] input bit
\r
170 A1 => addr_write(1), -- Port A address[1] input bit
\r
171 A2 => addr_write(2), -- Port A address[2] input bit
\r
172 A3 => addr_write(3), -- Port A address[3] input bit
\r
173 D => reg_dest_new(i), -- Port A 1-bit data input
\r
174 DPRA0 => addr_read1(0), -- Port B address[0] input bit
\r
175 DPRA1 => addr_read1(1), -- Port B address[1] input bit
\r
176 DPRA2 => addr_read1(2), -- Port B address[2] input bit
\r
177 DPRA3 => addr_read1(3), -- Port B address[3] input bit
\r
178 DPO => data_out1A(i), -- Port B 1-bit data output
\r
179 SPO => no_connect(i) -- Port A 1-bit data output
\r
181 --Read port 1 upper 16 registers
\r
182 reg_bit1b : RAM16X1D
\r
184 WCLK => clk, -- Port A write clock input
\r
185 WE => weB, -- Port A write enable input
\r
186 A0 => addr_write(0), -- Port A address[0] input bit
\r
187 A1 => addr_write(1), -- Port A address[1] input bit
\r
188 A2 => addr_write(2), -- Port A address[2] input bit
\r
189 A3 => addr_write(3), -- Port A address[3] input bit
\r
190 D => reg_dest_new(i), -- Port A 1-bit data input
\r
191 DPRA0 => addr_read1(0), -- Port B address[0] input bit
\r
192 DPRA1 => addr_read1(1), -- Port B address[1] input bit
\r
193 DPRA2 => addr_read1(2), -- Port B address[2] input bit
\r
194 DPRA3 => addr_read1(3), -- Port B address[3] input bit
\r
195 DPO => data_out1B(i), -- Port B 1-bit data output
\r
196 SPO => no_connect(32+i) -- Port A 1-bit data output
\r
198 --Read port 2 lower 16 registers
\r
199 reg_bit2a : RAM16X1D
\r
201 WCLK => clk, -- Port A write clock input
\r
202 WE => weA, -- Port A write enable input
\r
203 A0 => addr_write(0), -- Port A address[0] input bit
\r
204 A1 => addr_write(1), -- Port A address[1] input bit
\r
205 A2 => addr_write(2), -- Port A address[2] input bit
\r
206 A3 => addr_write(3), -- Port A address[3] input bit
\r
207 D => reg_dest_new(i), -- Port A 1-bit data input
\r
208 DPRA0 => addr_read2(0), -- Port B address[0] input bit
\r
209 DPRA1 => addr_read2(1), -- Port B address[1] input bit
\r
210 DPRA2 => addr_read2(2), -- Port B address[2] input bit
\r
211 DPRA3 => addr_read2(3), -- Port B address[3] input bit
\r
212 DPO => data_out2A(i), -- Port B 1-bit data output
\r
213 SPO => no_connect(64+i) -- Port A 1-bit data output
\r
215 --Read port 2 upper 16 registers
\r
216 reg_bit2b : RAM16X1D
\r
218 WCLK => clk, -- Port A write clock input
\r
219 WE => weB, -- Port A write enable input
\r
220 A0 => addr_write(0), -- Port A address[0] input bit
\r
221 A1 => addr_write(1), -- Port A address[1] input bit
\r
222 A2 => addr_write(2), -- Port A address[2] input bit
\r
223 A3 => addr_write(3), -- Port A address[3] input bit
\r
224 D => reg_dest_new(i), -- Port A 1-bit data input
\r
225 DPRA0 => addr_read2(0), -- Port B address[0] input bit
\r
226 DPRA1 => addr_read2(1), -- Port B address[1] input bit
\r
227 DPRA2 => addr_read2(2), -- Port B address[2] input bit
\r
228 DPRA3 => addr_read2(3), -- Port B address[3] input bit
\r
229 DPO => data_out2B(i), -- Port B 1-bit data output
\r
230 SPO => no_connect(96+i) -- Port A 1-bit data output
\r
232 end generate; --reg_loop
\r
234 data_out1 <= data_out1A when addr_read1(4)='0' else data_out1B;
\r
235 data_out2 <= data_out2A when addr_read2(4)='0' else data_out2B;
\r
236 end generate; --xilinx_16x1d
\r
240 -- RAM32X1D: 32 x 1 positive edge write, asynchronous read dual-port
\r
241 -- distributed RAM for 5-LUT Xilinx FPGAs such as Virtex-5
\r
242 -- From library UNISIM; use UNISIM.vcomponents.all;
\r
244 if memory_type = "XILINX_32X" generate
\r
245 signal no_connect : std_logic_vector(63 downto 0);
\r
247 reg_loop: for i in 0 to 31 generate
\r
250 reg_bit1 : RAM32X1D
\r
252 WCLK => clk, -- Port A write clock input
\r
253 WE => write_enable, -- Port A write enable input
\r
254 A0 => addr_write(0), -- Port A address[0] input bit
\r
255 A1 => addr_write(1), -- Port A address[1] input bit
\r
256 A2 => addr_write(2), -- Port A address[2] input bit
\r
257 A3 => addr_write(3), -- Port A address[3] input bit
\r
258 A4 => addr_write(4), -- Port A address[4] input bit
\r
259 D => reg_dest_new(i), -- Port A 1-bit data input
\r
260 DPRA0 => addr_read1(0), -- Port B address[0] input bit
\r
261 DPRA1 => addr_read1(1), -- Port B address[1] input bit
\r
262 DPRA2 => addr_read1(2), -- Port B address[2] input bit
\r
263 DPRA3 => addr_read1(3), -- Port B address[3] input bit
\r
264 DPRA4 => addr_read1(4), -- Port B address[4] input bit
\r
265 DPO => data_out1(i), -- Port B 1-bit data output
\r
266 SPO => no_connect(i) -- Port A 1-bit data output
\r
269 reg_bit2 : RAM32X1D
\r
271 WCLK => clk, -- Port A write clock input
\r
272 WE => write_enable, -- Port A write enable input
\r
273 A0 => addr_write(0), -- Port A address[0] input bit
\r
274 A1 => addr_write(1), -- Port A address[1] input bit
\r
275 A2 => addr_write(2), -- Port A address[2] input bit
\r
276 A3 => addr_write(3), -- Port A address[3] input bit
\r
277 A4 => addr_write(4), -- Port A address[4] input bit
\r
278 D => reg_dest_new(i), -- Port A 1-bit data input
\r
279 DPRA0 => addr_read2(0), -- Port B address[0] input bit
\r
280 DPRA1 => addr_read2(1), -- Port B address[1] input bit
\r
281 DPRA2 => addr_read2(2), -- Port B address[2] input bit
\r
282 DPRA3 => addr_read2(3), -- Port B address[3] input bit
\r
283 DPRA4 => addr_read2(4), -- Port B address[4] input bit
\r
284 DPO => data_out2(i), -- Port B 1-bit data output
\r
285 SPO => no_connect(32+i) -- Port A 1-bit data output
\r
287 end generate; --reg_loop
\r
288 end generate; --xilinx_32x1d
\r
292 -- Altera LPM_RAM_DP
\r
294 if memory_type = "ALTERA_LPM" generate
\r
295 signal clk_delayed : std_logic;
\r
296 signal addr_reg : std_logic_vector(4 downto 0);
\r
297 signal data_reg : std_logic_vector(31 downto 0);
\r
298 signal q1 : std_logic_vector(31 downto 0);
\r
299 signal q2 : std_logic_vector(31 downto 0);
\r
301 -- Altera dual port RAMs must have the addresses registered (sampled
\r
302 -- at the rising edge). This is very unfortunate.
\r
303 -- Therefore, the dual port RAM read clock must delayed so that
\r
304 -- the read address signal can be sent from the mem_ctrl block.
\r
305 -- This solution also delays the how fast the registers are read so the
\r
306 -- maximum clock speed is cut in half (12.5 MHz instead of 25 MHz).
\r
308 clk_delayed <= not clk; --Could be delayed by 1/4 clock cycle instead
\r
309 dpram_bypass: process(clk, addr_write, reg_dest_new, write_enable)
\r
311 if rising_edge(clk) and write_enable = '1' then
\r
312 addr_reg <= addr_write;
\r
313 data_reg <= reg_dest_new;
\r
315 end process; --dpram_bypass
\r
317 -- Bypass dpram if reading what was just written (Altera limitation)
\r
318 data_out1 <= q1 when addr_read1 /= addr_reg else data_reg;
\r
319 data_out2 <= q2 when addr_read2 /= addr_reg else data_reg;
\r
321 lpm_ram_dp_component1 : lpm_ram_dp
\r
325 --LPM_NUMWORDS => 0,
\r
326 LPM_INDATA => "REGISTERED",
\r
327 LPM_OUTDATA => "UNREGISTERED",
\r
328 LPM_RDADDRESS_CONTROL => "REGISTERED",
\r
329 LPM_WRADDRESS_CONTROL => "REGISTERED",
\r
330 LPM_FILE => "UNUSED",
\r
331 LPM_TYPE => "LPM_RAM_DP",
\r
333 INTENDED_DEVICE_FAMILY => "UNUSED",
\r
334 RDEN_USED => "FALSE",
\r
335 LPM_HINT => "UNUSED")
\r
337 RDCLOCK => clk_delayed,
\r
339 RDADDRESS => addr_read1,
\r
341 DATA => reg_dest_new,
\r
342 WRADDRESS => addr_write,
\r
343 WREN => write_enable,
\r
347 lpm_ram_dp_component2 : lpm_ram_dp
\r
351 --LPM_NUMWORDS => 0,
\r
352 LPM_INDATA => "REGISTERED",
\r
353 LPM_OUTDATA => "UNREGISTERED",
\r
354 LPM_RDADDRESS_CONTROL => "REGISTERED",
\r
355 LPM_WRADDRESS_CONTROL => "REGISTERED",
\r
356 LPM_FILE => "UNUSED",
\r
357 LPM_TYPE => "LPM_RAM_DP",
\r
359 INTENDED_DEVICE_FAMILY => "UNUSED",
\r
360 RDEN_USED => "FALSE",
\r
361 LPM_HINT => "UNUSED")
\r
363 RDCLOCK => clk_delayed,
\r
365 RDADDRESS => addr_read2,
\r
367 DATA => reg_dest_new,
\r
368 WRADDRESS => addr_write,
\r
369 WREN => write_enable,
\r
373 end generate; --altera_mem
\r
375 end; --architecture ram_block
\r