mirror of
git://projects.qi-hardware.com/nn-usb-fpga.git
synced 2024-12-13 23:02:49 +02:00
324 lines
14 KiB
VHDL
324 lines
14 KiB
VHDL
---------------------------------------------------------------------
|
|
-- TITLE: Register Bank
|
|
-- AUTHOR: Steve Rhoads (rhoadss@yahoo.com)
|
|
-- DATE CREATED: 2/2/01
|
|
-- FILENAME: reg_bank.vhd
|
|
-- PROJECT: Plasma CPU core
|
|
-- COPYRIGHT: Software placed into the public domain by the author.
|
|
-- Software 'as is' without warranty. Author liable for nothing.
|
|
-- DESCRIPTION:
|
|
-- Implements a register bank with 32 registers that are 32-bits wide.
|
|
-- There are two read-ports and one write port.
|
|
---------------------------------------------------------------------
|
|
library ieee;
|
|
use ieee.std_logic_1164.all;
|
|
use ieee.std_logic_unsigned.all;
|
|
use work.mlite_pack.all;
|
|
--library UNISIM; --May need to uncomment for ModelSim
|
|
--use UNISIM.vcomponents.all; --May need to uncomment for ModelSim
|
|
|
|
entity reg_bank is
|
|
generic(memory_type : string := "XILINX_16X");
|
|
port(clk : in std_logic;
|
|
reset_in : in std_logic;
|
|
pause : in std_logic;
|
|
rs_index : in std_logic_vector(5 downto 0);
|
|
rt_index : in std_logic_vector(5 downto 0);
|
|
rd_index : in std_logic_vector(5 downto 0);
|
|
reg_source_out : out std_logic_vector(31 downto 0);
|
|
reg_target_out : out std_logic_vector(31 downto 0);
|
|
reg_dest_new : in std_logic_vector(31 downto 0);
|
|
intr_enable : out std_logic);
|
|
end; --entity reg_bank
|
|
|
|
|
|
--------------------------------------------------------------------
|
|
-- The ram_block architecture attempts to use TWO dual-port memories.
|
|
-- Different FPGAs and ASICs need different implementations.
|
|
-- Choose one of the RAM implementations below.
|
|
-- I need feedback on this section!
|
|
--------------------------------------------------------------------
|
|
architecture ram_block of reg_bank is
|
|
signal intr_enable_reg : std_logic;
|
|
type ram_type is array(31 downto 0) of std_logic_vector(31 downto 0);
|
|
|
|
--controls access to dual-port memories
|
|
signal addr_read1, addr_read2 : std_logic_vector(4 downto 0);
|
|
signal addr_write : std_logic_vector(4 downto 0);
|
|
signal data_out1, data_out2 : std_logic_vector(31 downto 0);
|
|
signal write_enable : std_logic;
|
|
|
|
begin
|
|
|
|
reg_proc: process(clk, rs_index, rt_index, rd_index, reg_dest_new,
|
|
intr_enable_reg, data_out1, data_out2, reset_in, pause)
|
|
begin
|
|
--setup for first dual-port memory
|
|
if rs_index = "101110" then --reg_epc CP0 14
|
|
addr_read1 <= "00000";
|
|
else
|
|
addr_read1 <= rs_index(4 downto 0);
|
|
end if;
|
|
case rs_index is
|
|
when "000000" => reg_source_out <= ZERO;
|
|
when "101100" => reg_source_out <= ZERO(31 downto 1) & intr_enable_reg;
|
|
--interrupt vector address = 0x3c
|
|
when "111111" => reg_source_out <= ZERO(31 downto 8) & "00111100";
|
|
when others => reg_source_out <= data_out1;
|
|
end case;
|
|
|
|
--setup for second dual-port memory
|
|
addr_read2 <= rt_index(4 downto 0);
|
|
case rt_index is
|
|
when "000000" => reg_target_out <= ZERO;
|
|
when others => reg_target_out <= data_out2;
|
|
end case;
|
|
|
|
--setup write port for both dual-port memories
|
|
if rd_index /= "000000" and rd_index /= "101100" and pause = '0' then
|
|
write_enable <= '1';
|
|
else
|
|
write_enable <= '0';
|
|
end if;
|
|
if rd_index = "101110" then --reg_epc CP0 14
|
|
addr_write <= "00000";
|
|
else
|
|
addr_write <= rd_index(4 downto 0);
|
|
end if;
|
|
|
|
if reset_in = '1' then
|
|
intr_enable_reg <= '0';
|
|
elsif rising_edge(clk) then
|
|
if rd_index = "101110" then --reg_epc CP0 14
|
|
intr_enable_reg <= '0'; --disable interrupts
|
|
elsif rd_index = "101100" then
|
|
intr_enable_reg <= reg_dest_new(0);
|
|
end if;
|
|
end if;
|
|
|
|
intr_enable <= intr_enable_reg;
|
|
end process;
|
|
|
|
|
|
--------------------------------------------------------------
|
|
---- Pick only ONE of the dual-port RAM implementations below!
|
|
--------------------------------------------------------------
|
|
|
|
-- Option #1
|
|
-- One tri-port RAM, two read-ports, one write-port
|
|
-- 32 registers 32-bits wide
|
|
tri_port_mem:
|
|
if memory_type = "TRI_PORT_X" generate
|
|
ram_proc: process(clk, addr_read1, addr_read2,
|
|
addr_write, reg_dest_new, write_enable)
|
|
variable tri_port_ram : ram_type := (others => ZERO);
|
|
begin
|
|
data_out1 <= tri_port_ram(conv_integer(addr_read1));
|
|
data_out2 <= tri_port_ram(conv_integer(addr_read2));
|
|
if rising_edge(clk) then
|
|
if write_enable = '1' then
|
|
tri_port_ram(conv_integer(addr_write)) := reg_dest_new;
|
|
end if;
|
|
end if;
|
|
end process;
|
|
end generate; --tri_port_mem
|
|
|
|
|
|
-- Option #2
|
|
-- Two dual-port RAMs, each with one read-port and one write-port
|
|
dual_port_mem:
|
|
if memory_type = "DUAL_PORT_" generate
|
|
ram_proc2: process(clk, addr_read1, addr_read2,
|
|
addr_write, reg_dest_new, write_enable)
|
|
variable dual_port_ram1 : ram_type := (others => ZERO);
|
|
variable dual_port_ram2 : ram_type := (others => ZERO);
|
|
begin
|
|
data_out1 <= dual_port_ram1(conv_integer(addr_read1));
|
|
data_out2 <= dual_port_ram2(conv_integer(addr_read2));
|
|
if rising_edge(clk) then
|
|
if write_enable = '1' then
|
|
dual_port_ram1(conv_integer(addr_write)) := reg_dest_new;
|
|
dual_port_ram2(conv_integer(addr_write)) := reg_dest_new;
|
|
end if;
|
|
end if;
|
|
end process;
|
|
end generate; --dual_port_mem
|
|
|
|
|
|
-- Option #3
|
|
-- RAM16X1D: 16 x 1 positive edge write, asynchronous read dual-port
|
|
-- distributed RAM for all Xilinx FPGAs
|
|
-- From library UNISIM; use UNISIM.vcomponents.all;
|
|
xilinx_16x1d:
|
|
if memory_type = "XILINX_16X" generate
|
|
signal data_out1A, data_out1B : std_logic_vector(31 downto 0);
|
|
signal data_out2A, data_out2B : std_logic_vector(31 downto 0);
|
|
signal weA, weB : std_logic;
|
|
signal no_connect : std_logic_vector(127 downto 0);
|
|
begin
|
|
weA <= write_enable and not addr_write(4); --lower 16 registers
|
|
weB <= write_enable and addr_write(4); --upper 16 registers
|
|
|
|
reg_loop: for i in 0 to 31 generate
|
|
begin
|
|
--Read port 1 lower 16 registers
|
|
reg_bit1a : RAM16X1D
|
|
port map (
|
|
WCLK => clk, -- Port A write clock input
|
|
WE => weA, -- Port A write enable input
|
|
A0 => addr_write(0), -- Port A address[0] input bit
|
|
A1 => addr_write(1), -- Port A address[1] input bit
|
|
A2 => addr_write(2), -- Port A address[2] input bit
|
|
A3 => addr_write(3), -- Port A address[3] input bit
|
|
D => reg_dest_new(i), -- Port A 1-bit data input
|
|
DPRA0 => addr_read1(0), -- Port B address[0] input bit
|
|
DPRA1 => addr_read1(1), -- Port B address[1] input bit
|
|
DPRA2 => addr_read1(2), -- Port B address[2] input bit
|
|
DPRA3 => addr_read1(3), -- Port B address[3] input bit
|
|
DPO => data_out1A(i), -- Port B 1-bit data output
|
|
SPO => no_connect(i) -- Port A 1-bit data output
|
|
);
|
|
--Read port 1 upper 16 registers
|
|
reg_bit1b : RAM16X1D
|
|
port map (
|
|
WCLK => clk, -- Port A write clock input
|
|
WE => weB, -- Port A write enable input
|
|
A0 => addr_write(0), -- Port A address[0] input bit
|
|
A1 => addr_write(1), -- Port A address[1] input bit
|
|
A2 => addr_write(2), -- Port A address[2] input bit
|
|
A3 => addr_write(3), -- Port A address[3] input bit
|
|
D => reg_dest_new(i), -- Port A 1-bit data input
|
|
DPRA0 => addr_read1(0), -- Port B address[0] input bit
|
|
DPRA1 => addr_read1(1), -- Port B address[1] input bit
|
|
DPRA2 => addr_read1(2), -- Port B address[2] input bit
|
|
DPRA3 => addr_read1(3), -- Port B address[3] input bit
|
|
DPO => data_out1B(i), -- Port B 1-bit data output
|
|
SPO => no_connect(32+i) -- Port A 1-bit data output
|
|
);
|
|
--Read port 2 lower 16 registers
|
|
reg_bit2a : RAM16X1D
|
|
port map (
|
|
WCLK => clk, -- Port A write clock input
|
|
WE => weA, -- Port A write enable input
|
|
A0 => addr_write(0), -- Port A address[0] input bit
|
|
A1 => addr_write(1), -- Port A address[1] input bit
|
|
A2 => addr_write(2), -- Port A address[2] input bit
|
|
A3 => addr_write(3), -- Port A address[3] input bit
|
|
D => reg_dest_new(i), -- Port A 1-bit data input
|
|
DPRA0 => addr_read2(0), -- Port B address[0] input bit
|
|
DPRA1 => addr_read2(1), -- Port B address[1] input bit
|
|
DPRA2 => addr_read2(2), -- Port B address[2] input bit
|
|
DPRA3 => addr_read2(3), -- Port B address[3] input bit
|
|
DPO => data_out2A(i), -- Port B 1-bit data output
|
|
SPO => no_connect(64+i) -- Port A 1-bit data output
|
|
);
|
|
--Read port 2 upper 16 registers
|
|
reg_bit2b : RAM16X1D
|
|
port map (
|
|
WCLK => clk, -- Port A write clock input
|
|
WE => weB, -- Port A write enable input
|
|
A0 => addr_write(0), -- Port A address[0] input bit
|
|
A1 => addr_write(1), -- Port A address[1] input bit
|
|
A2 => addr_write(2), -- Port A address[2] input bit
|
|
A3 => addr_write(3), -- Port A address[3] input bit
|
|
D => reg_dest_new(i), -- Port A 1-bit data input
|
|
DPRA0 => addr_read2(0), -- Port B address[0] input bit
|
|
DPRA1 => addr_read2(1), -- Port B address[1] input bit
|
|
DPRA2 => addr_read2(2), -- Port B address[2] input bit
|
|
DPRA3 => addr_read2(3), -- Port B address[3] input bit
|
|
DPO => data_out2B(i), -- Port B 1-bit data output
|
|
SPO => no_connect(96+i) -- Port A 1-bit data output
|
|
);
|
|
end generate; --reg_loop
|
|
|
|
data_out1 <= data_out1A when addr_read1(4)='0' else data_out1B;
|
|
data_out2 <= data_out2A when addr_read2(4)='0' else data_out2B;
|
|
end generate; --xilinx_16x1d
|
|
|
|
|
|
-- Option #4
|
|
-- Altera LPM_RAM_DP
|
|
altera_mem:
|
|
if memory_type = "ALTERA_LPM" generate
|
|
signal clk_delayed : std_logic;
|
|
signal addr_reg : std_logic_vector(4 downto 0);
|
|
signal data_reg : std_logic_vector(31 downto 0);
|
|
signal q1 : std_logic_vector(31 downto 0);
|
|
signal q2 : std_logic_vector(31 downto 0);
|
|
begin
|
|
-- Altera dual port RAMs must have the addresses registered (sampled
|
|
-- at the rising edge). This is very unfortunate.
|
|
-- Therefore, the dual port RAM read clock must delayed so that
|
|
-- the read address signal can be sent from the mem_ctrl block.
|
|
-- This solution also delays the how fast the registers are read so the
|
|
-- maximum clock speed is cut in half (12.5 MHz instead of 25 MHz).
|
|
|
|
clk_delayed <= not clk; --Could be delayed by 1/4 clock cycle instead
|
|
dpram_bypass: process(clk, addr_write, reg_dest_new)
|
|
begin
|
|
if rising_edge(clk) and write_enable = '1' then
|
|
addr_reg <= addr_write;
|
|
data_reg <= reg_dest_new;
|
|
end if;
|
|
end process; --dpram_bypass
|
|
|
|
-- Bypass dpram if reading what was just written (Altera limitation)
|
|
data_out1 <= q1 when addr_read1 /= addr_reg else data_reg;
|
|
data_out2 <= q2 when addr_read2 /= addr_reg else data_reg;
|
|
|
|
lpm_ram_dp_component1 : lpm_ram_dp
|
|
generic map (
|
|
LPM_WIDTH => 32,
|
|
LPM_WIDTHAD => 5,
|
|
--LPM_NUMWORDS => 0,
|
|
LPM_INDATA => "REGISTERED",
|
|
LPM_OUTDATA => "UNREGISTERED",
|
|
LPM_RDADDRESS_CONTROL => "REGISTERED",
|
|
LPM_WRADDRESS_CONTROL => "REGISTERED",
|
|
LPM_FILE => "UNUSED",
|
|
LPM_TYPE => "LPM_RAM_DP",
|
|
USE_EAB => "ON",
|
|
INTENDED_DEVICE_FAMILY => "UNUSED",
|
|
RDEN_USED => "FALSE",
|
|
LPM_HINT => "UNUSED")
|
|
port map (
|
|
RDCLOCK => clk_delayed,
|
|
RDCLKEN => '1',
|
|
RDADDRESS => addr_read1,
|
|
RDEN => '1',
|
|
DATA => reg_dest_new,
|
|
WRADDRESS => addr_write,
|
|
WREN => write_enable,
|
|
WRCLOCK => clk,
|
|
WRCLKEN => '1',
|
|
Q => q1);
|
|
lpm_ram_dp_component2 : lpm_ram_dp
|
|
generic map (
|
|
LPM_WIDTH => 32,
|
|
LPM_WIDTHAD => 5,
|
|
--LPM_NUMWORDS => 0,
|
|
LPM_INDATA => "REGISTERED",
|
|
LPM_OUTDATA => "UNREGISTERED",
|
|
LPM_RDADDRESS_CONTROL => "REGISTERED",
|
|
LPM_WRADDRESS_CONTROL => "REGISTERED",
|
|
LPM_FILE => "UNUSED",
|
|
LPM_TYPE => "LPM_RAM_DP",
|
|
USE_EAB => "ON",
|
|
INTENDED_DEVICE_FAMILY => "UNUSED",
|
|
RDEN_USED => "FALSE",
|
|
LPM_HINT => "UNUSED")
|
|
port map (
|
|
RDCLOCK => clk_delayed,
|
|
RDCLKEN => '1',
|
|
RDADDRESS => addr_read2,
|
|
RDEN => '1',
|
|
DATA => reg_dest_new,
|
|
WRADDRESS => addr_write,
|
|
WREN => write_enable,
|
|
WRCLOCK => clk,
|
|
WRCLKEN => '1',
|
|
Q => q2);
|
|
end generate; --altera_mem
|
|
|
|
end; --architecture ram_block
|