1 |
18 |
Agner |
//////////////////////////////////////////////////////////////////////////////////
|
2 |
|
|
// Engineer: Agner Fog
|
3 |
|
|
//
|
4 |
|
|
// Create Date: 2020-06-03
|
5 |
|
|
// Last modified: 2021-08-02
|
6 |
|
|
// Module Name: data_cache
|
7 |
|
|
// Project Name: ForwardCom soft core
|
8 |
|
|
// Target Devices: Artix 7
|
9 |
|
|
// Tool Versions: Vivado v. 2020.1
|
10 |
|
|
// License: CERN-OHL-W v. 2 or later
|
11 |
|
|
// Description: data memory or data cache for read/write data
|
12 |
|
|
//
|
13 |
|
|
//////////////////////////////////////////////////////////////////////////////////
|
14 |
|
|
`include "defines.vh"
|
15 |
|
|
|
16 |
|
|
|
17 |
|
|
// read/write data memory or cache, (2**`DATA_ADDR_WIDTH) bytes = 2**16 = 64kB
|
18 |
|
|
module data_memory (
|
19 |
|
|
input clock, // clock
|
20 |
|
|
input clock_enable, // clock enable. Used when single-stepping
|
21 |
|
|
input [`COMMON_ADDR_WIDTH-1:0] read_write_addr, // Address for reading and writing from/to ram
|
22 |
|
|
// The lower 3 bits of read_write_addr indicate a byte within an 8 bytes line
|
23 |
|
|
input read_enable, // read enable
|
24 |
|
|
input [1:0] read_data_size, // 8, 16, 32, or 64 bits read
|
25 |
|
|
input [7:0] write_enable, // write enable for each byte separately
|
26 |
|
|
input [63:0] write_data_in, // Data in. Always 64 bits. Any part of the write bus can be used when the data size is less than 64 bits
|
27 |
|
|
`ifdef DISTRIBUTED_RAM // Distributed RAM takes a lot of FPGA resources
|
28 |
|
|
output reg [`RB1:0] read_data_out // Data out
|
29 |
|
|
`else // Block RAM
|
30 |
|
|
output logic [`RB1:0] read_data_out // Data out
|
31 |
|
|
`endif
|
32 |
|
|
);
|
33 |
|
|
|
34 |
|
|
// read/write data ram
|
35 |
|
|
reg [63:0] dataram [0:(2**(`DATA_ADDR_WIDTH-3))-1]; // 64kB RAM
|
36 |
|
|
|
37 |
|
|
// split read/write address into double-word index, and byte index
|
38 |
|
|
logic [`DATA_ADDR_WIDTH-4:0] address_hi;
|
39 |
|
|
logic [2:0] address_lo;
|
40 |
|
|
logic address_valid;
|
41 |
|
|
|
42 |
|
|
always_comb begin
|
43 |
|
|
address_hi = read_write_addr[`DATA_ADDR_WIDTH-1:3]; // index to 64-bit lines
|
44 |
|
|
address_lo = read_write_addr[2:0]; // index to byte within line
|
45 |
|
|
address_valid = read_write_addr[`COMMON_ADDR_WIDTH-1:`DATA_ADDR_WIDTH] == 0; // exclude code addresses
|
46 |
|
|
end
|
47 |
|
|
|
48 |
|
|
|
49 |
|
|
// Data write:
|
50 |
|
|
always_ff @(posedge clock) if (clock_enable & address_valid) begin
|
51 |
|
|
// write data to RAM. Each byte enabled separately
|
52 |
|
|
if (write_enable[0]) dataram[address_hi][ 7: 0] <= write_data_in[ 7: 0];
|
53 |
|
|
if (write_enable[1]) dataram[address_hi][15: 8] <= write_data_in[15: 8];
|
54 |
|
|
if (write_enable[2]) dataram[address_hi][23:16] <= write_data_in[23:16];
|
55 |
|
|
if (write_enable[3]) dataram[address_hi][31:24] <= write_data_in[31:24];
|
56 |
|
|
if (write_enable[4]) dataram[address_hi][39:32] <= write_data_in[39:32];
|
57 |
|
|
if (write_enable[5]) dataram[address_hi][47:40] <= write_data_in[47:40];
|
58 |
|
|
if (write_enable[6]) dataram[address_hi][55:48] <= write_data_in[55:48];
|
59 |
|
|
if (write_enable[7]) dataram[address_hi][63:56] <= write_data_in[63:56];
|
60 |
|
|
end
|
61 |
|
|
|
62 |
|
|
|
63 |
|
|
// data read. Must have natural alignment
|
64 |
|
|
|
65 |
|
|
`ifdef DISTRIBUTED_RAM
|
66 |
|
|
// The multiplexer comes before the register. This is only possible with distributed RAM.
|
67 |
|
|
// Distributed RAM takes a lot of FPGA resources but may allow a slightly higher clock frequency.
|
68 |
|
|
|
69 |
|
|
always_ff @(posedge clock) if (clock_enable & address_valid & read_enable) begin
|
70 |
|
|
|
71 |
|
|
// Each 64-bit RAM line may be divided into
|
72 |
|
|
// eight bytes, four 16-bit halfwords, two 32-bit words, or one 64-bit double word:
|
73 |
|
|
case (address_lo)
|
74 |
|
|
0: read_data_out[7:0] <= dataram[address_hi][ 7: 0];
|
75 |
|
|
1: read_data_out[7:0] <= dataram[address_hi][15: 8];
|
76 |
|
|
2: read_data_out[7:0] <= dataram[address_hi][23:16];
|
77 |
|
|
3: read_data_out[7:0] <= dataram[address_hi][31:24];
|
78 |
|
|
4: read_data_out[7:0] <= dataram[address_hi][39:32];
|
79 |
|
|
5: read_data_out[7:0] <= dataram[address_hi][47:40];
|
80 |
|
|
6: read_data_out[7:0] <= dataram[address_hi][55:48];
|
81 |
|
|
7: read_data_out[7:0] <= dataram[address_hi][63:56];
|
82 |
|
|
endcase
|
83 |
|
|
|
84 |
|
|
case (address_lo[2:1])
|
85 |
|
|
0: read_data_out[15:8] <= dataram[address_hi][15: 8];
|
86 |
|
|
1: read_data_out[15:8] <= dataram[address_hi][31:24];
|
87 |
|
|
2: read_data_out[15:8] <= dataram[address_hi][47:40];
|
88 |
|
|
3: read_data_out[15:8] <= dataram[address_hi][63:56];
|
89 |
|
|
endcase
|
90 |
|
|
|
91 |
|
|
case (address_lo[2])
|
92 |
|
|
0: read_data_out[31:16] <= dataram[address_hi][31:16];
|
93 |
|
|
1: read_data_out[31:16] <= dataram[address_hi][63:48];
|
94 |
|
|
endcase
|
95 |
|
|
|
96 |
|
|
`ifdef SUPPORT_64BIT
|
97 |
|
|
read_data_out[63:32] <= dataram[address_hi][63:32];
|
98 |
|
|
`endif
|
99 |
|
|
end
|
100 |
|
|
|
101 |
|
|
`else
|
102 |
|
|
// block RAM. The multiplexer must come after the register
|
103 |
|
|
|
104 |
|
|
reg [63:0] read_data; // a whole line read from the RAM
|
105 |
|
|
reg [2:0] address_lo2; // address_lo saved
|
106 |
|
|
|
107 |
|
|
|
108 |
|
|
always_ff @(posedge clock) if (clock_enable & address_valid & read_enable) begin
|
109 |
|
|
read_data <= dataram[address_hi]; // read a 64 bits line from ram
|
110 |
|
|
address_lo2 <= address_lo; // save low part of address
|
111 |
|
|
end
|
112 |
|
|
|
113 |
|
|
always_comb begin
|
114 |
|
|
// Each 64-bit RAM line may be divided into eight bytes, four 16-bit halfwords,
|
115 |
|
|
// two 32-bit words, or one 64-bit double word.
|
116 |
|
|
// The speed of this multiplexer is very critical because it adds to the delay
|
117 |
|
|
// in the execution unit. We are saving time by not setting unused parts of
|
118 |
|
|
// read_data_out to zero.
|
119 |
|
|
case (address_lo2)
|
120 |
|
|
0: read_data_out[7:0] = read_data[ 7: 0];
|
121 |
|
|
1: read_data_out[7:0] = read_data[15: 8];
|
122 |
|
|
2: read_data_out[7:0] = read_data[23:16];
|
123 |
|
|
3: read_data_out[7:0] = read_data[31:24];
|
124 |
|
|
4: read_data_out[7:0] = read_data[39:32];
|
125 |
|
|
5: read_data_out[7:0] = read_data[47:40];
|
126 |
|
|
6: read_data_out[7:0] = read_data[55:48];
|
127 |
|
|
7: read_data_out[7:0] = read_data[63:56];
|
128 |
|
|
endcase
|
129 |
|
|
|
130 |
|
|
case (address_lo2[2:1])
|
131 |
|
|
0: read_data_out[15:8] = read_data[15: 8];
|
132 |
|
|
1: read_data_out[15:8] = read_data[31:24];
|
133 |
|
|
2: read_data_out[15:8] = read_data[47:40];
|
134 |
|
|
3: read_data_out[15:8] = read_data[63:56];
|
135 |
|
|
endcase
|
136 |
|
|
|
137 |
|
|
case (address_lo2[2])
|
138 |
|
|
0: read_data_out[31:16] = read_data[31:16];
|
139 |
|
|
1: read_data_out[31:16] = read_data[63:48];
|
140 |
|
|
endcase
|
141 |
|
|
|
142 |
|
|
`ifdef SUPPORT_64BIT
|
143 |
|
|
read_data_out[63:32] = read_data[63:32];
|
144 |
|
|
`endif
|
145 |
|
|
end
|
146 |
|
|
|
147 |
|
|
`endif
|
148 |
|
|
|
149 |
|
|
endmodule
|