blob: 280c845bacfef988cd880a8ce24ef07f05cf00c8 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
|
Dynamic VLIW
Allows a dynamic number of instructions to be queued for simultaneous execution
All inputs within the same queue will be calculated before outputs
`mov R0, R1` and `mov R1, R0` put into the same queue will result in swapping R0 and R1
If any exception occurs during the execution of the queued instructions, all output will be discarded
If multiple exceptions would have theoretically occurred, which one is triggered is undefined
If outputs from multiple instructions overlap:
Flag saving will always take priority
Conditional instructions will be preferred above all other instructions
Indirect writes, such as `jmp` and `push`, will be preferred over direct writes
If a conflict still remains, which value will be written to the overlapping area is undefined
If an exception would be generated by attempting to write one value there but not the other, the exception might not occur if the other value is what is written
Format:
size_of_instructions // size so it won't have to parse each to find where the immediate references start
for each:
opcode
num_instructions
for each * params per instruction:
value
size_of_immediate_references
for each:
value
num_flag_save_definitions
for each:
instruction index in queue
which flags to save (1 bit per flag type available)
where to save them
bitwise offset in the destination (log2(bits_for_current_mode) bits)
A minimum of <n> instructions per queue must be accepted by all CPUs
TODO: define n
Specific implementations may accept more, up to the maximum specifiable by the format
|
