| Commit message (Collapse) | Author | Age |
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
| |
breakpoint for first cycle after resume.
Instruction for stage is updated when given stage is flushed as well.
But other signals are left intact, it is duty of memory stage
to discard effect of instruction causing interrupt.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
| |
It works like real inserted breakpoint on hardware.
Breakpoint has to be removed to allow code continue
because else instruction is refetch and breakpoint
triggers again. The single step function should
resolve temporal masking of the breakpoint.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
| |
stages.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
| |
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
| |
opcode.
Because individual cache maintenance operations are not decoded,
be on safe side and flush all caches when CACHE operation
is processed,
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
| |
BGEZALL.
GCC generates these opcodes for default compilation mode.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
|
| |
delay slot.
When exception occurs at instruction in delay slot, the address
of branch/delay instruction is stored to EPC instead of address
of instruction causing the exception. So that address has
to be delivered to exception handling object.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
| |
The new Qt5 syntax is used to create connections because
old syntax does not work with multiple arguments for some
unresolved reason.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Memory stage is chosen to be exception commit stage.
Instructions flow postponed and stages holding following
instructions are cleaned. Processing of syscall at decode
stage as jump to the handler would be better solution
in real hardware but for future emulated syscalls
it is better to reach consistent state of registers.
Memory access caused exceptions would require cleanup
even in real hardware.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
| |
independent.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
| |
Remaining are MOVZ and MOVN in the execution phase
and all branch and jump operations in handle_pc().
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
| |
instructions.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
| |
This allows to specify requirement for RS and RD on instruction
basis even for T_R / ALU instructions.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
| |
When any variant of cache instruction is detected
flush and invalidate whole data cache.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
|
|
|
| |
instructions.
The previous code worked by chance only because decode has been fully
processed including forwarding from M and W before PC processing
started. But in real hardware the PC processing runs in parallel
with ALU and cannot read its results in the same cycle.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
| |
The JR, BEQ, BNE are most probably incorrect still.
There is missing forwarding for pipelined execution.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
| |
zero-extended immediate.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
| |
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
|
|
|
|
|
|
|
| |
The shorter loop has priority. This is achieved by later
processing when it replaces possible result from longer
loop over W stage.
Signed-off-by: Pavel Pisa <pisa@cmp.felk.cvut.cz>
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
THere are exceptions when we care about forwarding and when we don't.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
In reality this internally allows us to see stages even it we are not
using pipelining but that is hidden from outside simply to not confuse
user.
|
| |
|
| |
|