PoC.cache.par2
Cache with parallel tag-unit and data memory. For the data memory, PoC.mem.ocram.sp is used.
Configuration
Parameter |
Description |
---|---|
REPLACEMENT_POLICY |
Replacement policy. For supported policies see PoC.cache_replacement_policy. |
CACHE_LINES |
Number of cache lines. |
ASSOCIATIVITY |
Associativity of the cache. |
ADDR_BITS |
Number of address bits. Each address identifies exactly one cache line in memory. |
DATA_BITS |
Size of a cache line in bits. DATA_BITS must be divisible by 8. |
Command truth table
Request |
ReadWrite |
Invalidate |
Replace |
Command |
---|---|---|---|---|
0 |
0 |
0 |
0 |
None |
1 |
0 |
0 |
0 |
Read cache line |
1 |
1 |
0 |
0 |
Update cache line |
1 |
0 |
1 |
0 |
Read cache line and discard it |
1 |
1 |
1 |
0 |
Write cache line and discard it |
0 |
0 |
0 |
1 |
Read cache line before replace. |
0 |
1 |
0 |
1 |
Replace cache line. |
Operation
All inputs are synchronous to the rising-edge of the clock clock.
All commands use Address
to lookup (request) or replace a cache line.
Address
and OldAddress
do not include the word/byte select part.
Each command is completed within one clock cycle, but outputs are delayed as
described below.
Upon requests, the outputs CacheMiss
and CacheHit
indicate (high-active)
whether the Address
is stored within the cache, or not. Both outputs have a
latency of one clock cycle (pipelined) if HIT_MISS_REG
is true, otherwise the
result is outputted immediately (combinational).
Upon writing a cache line, the new content is given by CacheLineIn
.
Only the bytes which are not masked, i.e. the corresponding bit in WriteMask
is ‘0’, are actually written.
Upon reading a cache line, the current content is outputed on CacheLineOut
with a latency of one clock cycle.
Replacing a cache line requires two steps, both with Replace = '1'
:
Read old contents of cache line by setting
ReadWrite
to ‘0’. The old content is outputed onCacheLineOut
and the old tag onOldAddress
, both with a latency of one clock cycle.Write new cache line by setting
ReadWrite
to ‘1’. The new content is given byCacheLineIn
. All bytes shall be written, i.e.WriteMask = 0
. The new cache line content will be outputed again onCacheLineOut
in the next clock cycle (latency = 1).
Warning
If the design is synthesized with Xilinx ISE / XST, then the synthesis option “Keep Hierarchy” must be set to SOFT or TRUE.
Entity Declaration:
1 generic (
2 REPLACEMENT_POLICY : string := "LRU";
3 CACHE_LINES : positive := 32;
4 ASSOCIATIVITY : positive := 32;
5 ADDR_BITS : positive := 8;
6 DATA_BITS : positive := 8;
7 HIT_MISS_REG : boolean := true -- must be true for Cocotb.
8 );
9 port (
10 Clock : in std_logic;
11 Reset : in std_logic;
12
13 Request : in std_logic;
14 ReadWrite : in std_logic;
15 WriteMask : in std_logic_vector(DATA_BITS/8 - 1 downto 0) := (others => '0');
16 Invalidate : in std_logic;
17 Replace : in std_logic;
18 Address : in std_logic_vector(ADDR_BITS-1 downto 0);
19
20 CacheLineIn : in std_logic_vector(DATA_BITS - 1 downto 0);
21 CacheLineOut : out std_logic_vector(DATA_BITS - 1 downto 0);
22 CacheHit : out std_logic := '0';
23 CacheMiss : out std_logic := '0';
24 OldAddress : out std_logic_vector(ADDR_BITS-1 downto 0)
25 );
26end entity;