Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 23:57:33 +0000 (23:57 +0000)]
another major bug, CacheTagArray valid was only 1 bit not NUM_WAYS
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 23:40:40 +0000 (23:40 +0000)]
tidyup: move hit_set to DCachePendingHit in dcache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 23:35:44 +0000 (23:35 +0000)]
dcache.py tidyup
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 23:17:35 +0000 (23:17 +0000)]
rename dtlb to dtlb_valid and tidyup
remove dtlb argument (not needed) because dtlb_valid is now localised
to DTLBUpdate Module
also put in some code-comments
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 22:43:29 +0000 (22:43 +0000)]
convert TLBArray to TLBValidArray
(because PTE and TAG are now each in a Memory)
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 22:41:58 +0000 (22:41 +0000)]
convert DTLBUpdate to use a pair of Memorys
one for PTEs and one for TAGs
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 20:37:37 +0000 (20:37 +0000)]
more signals local to DTLBUpdate
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 20:30:47 +0000 (20:30 +0000)]
more signals local to DTLBUpdate
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 20:25:58 +0000 (20:25 +0000)]
update DTLBUpdate to reflect internal API now
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 17:40:12 +0000 (17:40 +0000)]
ooo nasty bug. used tlb_hit.way instead of tlb_hit.valid
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 17:33:31 +0000 (17:33 +0000)]
move DTLB Tags/Valids/PTEs into DTLBUpdate module
drastically simplifies DCache graphviz and also makes it clear that
TLBs have to be a Memory SRAM
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 17:14:45 +0000 (17:14 +0000)]
start moving TLBArray into DTLBUpdate
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 17:00:29 +0000 (17:00 +0000)]
PLRUs were selecting an output index, only one selected
therefore move the selection from the PLRUs into the module.
simplified
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 16:01:13 +0000 (16:01 +0000)]
repeated copies of read/write addr/sel to Cache SRAMs
moved rd/wr addr/sel outside of loops, only creates one MUX set now
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 15:45:40 +0000 (15:45 +0000)]
move bank of PLRUs to their own submodule in both dcache.py and icache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 14:49:20 +0000 (14:49 +0000)]
code-comments
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 14:41:32 +0000 (14:41 +0000)]
use binary-to-unary encoders in dcache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 14:11:45 +0000 (14:11 +0000)]
global (one) do_read signal in cache_rams dcache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 14:01:56 +0000 (14:01 +0000)]
use one-hot binary-to-unary in dcache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 13:50:48 +0000 (13:50 +0000)]
use i_in.req to gate hit_way via Decoder in icache.py
Luke Kenneth Casson Leighton [Mon, 6 Dec 2021 13:36:04 +0000 (13:36 +0000)]
use Decoder (binary-to-unary) in icache.py to deal with CAM creation
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 23:16:28 +0000 (23:16 +0000)]
use unary encoding (one-hot) for replace_way hit_way etc.
otherwise it produces binary CAM-compares
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 22:02:49 +0000 (22:02 +0000)]
code-comments
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 21:52:25 +0000 (21:52 +0000)]
whitespace and minor cleanup of D-Cache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 21:41:41 +0000 (21:41 +0000)]
more use of TLBHit Record in D-Cache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 21:33:20 +0000 (21:33 +0000)]
correct tlb_hit_way and index sizes, use TLBHit Record in D-Cache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 21:11:27 +0000 (21:11 +0000)]
use TLBRecord in D-Cache for which TLB is selected
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:48:35 +0000 (20:48 +0000)]
split out TLBRecord, correct number of valid bits
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:39:15 +0000 (20:39 +0000)]
use Record in DCache for TLB
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:31:26 +0000 (20:31 +0000)]
use Record in D-Cache Cache Tags
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:20:27 +0000 (20:20 +0000)]
whitespace
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:18:46 +0000 (20:18 +0000)]
use Record for I-Cache Cache Tag/Valid
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:05:47 +0000 (20:05 +0000)]
whitespace
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 20:02:18 +0000 (20:02 +0000)]
use Record for ICache TLB
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 14:24:48 +0000 (14:24 +0000)]
sorting out test_mmu_dcache.py to use wb_get
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 14:19:27 +0000 (14:19 +0000)]
convert icache.py to standard wishbone Interface
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 13:38:37 +0000 (13:38 +0000)]
fake up wishbone stall signal in icache.
same thing is done in dcache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 13:38:19 +0000 (13:38 +0000)]
fix icache row store issue
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 12:41:56 +0000 (12:41 +0000)]
using same tag/row functions as in dcache.py
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 12:35:20 +0000 (12:35 +0000)]
more signal sizes in icache.py
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 12:26:43 +0000 (12:26 +0000)]
incorrect Signal sizes in icache.py,
e.g. using NUM_WAYS instead of WAY_BITS
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 12:11:39 +0000 (12:11 +0000)]
sorting out icache.py, used to work
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 11:47:52 +0000 (11:47 +0000)]
remove redundant code
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 11:46:53 +0000 (11:46 +0000)]
add I-Cache standard bus (not used yet)
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 11:25:10 +0000 (11:25 +0000)]
remove yet another duplicate copy of wb_get, possible (again) due to
adding wishbone standard interface to D-Cache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 11:23:07 +0000 (11:23 +0000)]
replace yet another duplicate copy of wb_get, possible after renaming
the D-Cache wishbone bus
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 01:29:58 +0000 (01:29 +0000)]
wishbone bus convert on dcache
Luke Kenneth Casson Leighton [Sun, 5 Dec 2021 00:26:50 +0000 (00:26 +0000)]
correct import of wg_get function
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:37:58 +0000 (18:37 +0000)]
remove yet another duplicated copy of wb_get and add some better
testing in misaligned mmu test
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:27:46 +0000 (18:27 +0000)]
rename function which needs replacing
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:22:06 +0000 (18:22 +0000)]
should have been using common version of wb_get, not 8 duplicates
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:17:13 +0000 (18:17 +0000)]
should not have been duplicating wb_get function in 5 places
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:15:11 +0000 (18:15 +0000)]
get test_mmu_dcache.py working again
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:02:02 +0000 (18:02 +0000)]
remove wb_get, should not have been duplicated
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 18:00:01 +0000 (18:00 +0000)]
remove wb_get, should not have been massively duplicated. moved to openpower-isa
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 17:55:15 +0000 (17:55 +0000)]
fix return results from pi_ld
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 15:27:42 +0000 (15:27 +0000)]
wark-wark, broke mmu with removing rin. reverted
Tobias Platen [Sat, 4 Dec 2021 15:19:52 +0000 (16:19 +0100)]
fixed wait_addr to exit immediately on exception
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 15:11:07 +0000 (15:11 +0000)]
tidyup, comments
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 15:05:18 +0000 (15:05 +0000)]
tidyup mmu
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 14:59:52 +0000 (14:59 +0000)]
sigh in MMU FSM use direct access to ldst.dar/dsisr for OP_MFSPR
and likewise to mmu.
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 13:09:17 +0000 (13:09 +0000)]
remove DAR from PortInterface (where is the data going? there is no place
to put DAR if transmitted over PortInterface? what receives it? nothing
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 13:05:35 +0000 (13:05 +0000)]
stop using dar_o from PortInterface, get DAR directly from LoadStore1
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 13:05:05 +0000 (13:05 +0000)]
put DSISR and DAR publicly accessible in LoadStore1
these should ONLY be READ, NOT written to
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 13:01:26 +0000 (13:01 +0000)]
whoops fix up exception happened if alignment triggers from LoadStore1
set_wr_addr or set_rd_addr
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 12:54:22 +0000 (12:54 +0000)]
fix pi_st which should not be trying to wait for the address
when an exception occurs.
TODO: fix wait_addr so that it exits if an exception occurs
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 12:03:07 +0000 (12:03 +0000)]
fixing DAR updating from exceptions
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 11:46:57 +0000 (11:46 +0000)]
whoops
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 11:46:34 +0000 (11:46 +0000)]
MMU lookup DSISR load bit inverted in LoadStore1
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 11:45:59 +0000 (11:45 +0000)]
store DAR in LoadStore1
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 04:06:53 +0000 (04:06 +0000)]
not busy if excrption occurs on MMU_LOOKUP in loadstore.py
Luke Kenneth Casson Leighton [Sat, 4 Dec 2021 01:25:05 +0000 (01:25 +0000)]
add means to update dsisr from MMU FSM. TODO: add a back-communication
to allow LoadStore1 to update a copy of dsisr in the MMU FSM
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 19:35:41 +0000 (19:35 +0000)]
priv_mode/virt_mode are set in the request, which is passed through
to the MMU, via the PortInterface "pr" parameter.
MMU should not itself be attempting to set d_in.priv_mode/virt_mode
this fixes case_5_ldst_exception
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 19:19:36 +0000 (19:19 +0000)]
in loadstore.py set align_intr from request which comes from PortInterface
misalign, set it globally rather than from latched copy ldst_r
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 19:17:11 +0000 (19:17 +0000)]
driver conflict on priv_mode and virt_mode, do not understand right now,
commenting them out in loadstore.py
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 17:41:33 +0000 (17:41 +0000)]
fix up test_loadstore1.py
numerous things:
1) pi_ld and pi_st now return the exception information (full status)
because pi_ld and pi_st both reset the PortInterface: once that
happens the exception information is DESTROYED... therefore the
exception information MUST be obtained INSIDE pi_ld and pi_st
2) cleanup of whitespace
3) added some (probably unnecessary) yields in between LD/ST in the
exceptions test, to make the gtkwave output clearer
4) fixed the exceptions test and enabled it by default, now
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 17:38:26 +0000 (17:38 +0000)]
in loadstore.py, when an exception is done or if the FSM
is done, reset back to idle and indicate "not busy"
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 17:37:26 +0000 (17:37 +0000)]
fix PortInterfaceBase
* setting busy latch LOW on exception is not ok: this creates a
combinatorial loop
* setting st_done permanently was a bug, it left st_done permanently HI
* setting st_done when an exception is raised was also a bug
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 17:35:16 +0000 (17:35 +0000)]
fix up LDST test functions pi_ld and pi_st to respect timing
(combinatorial setting followed by test followed by immediate unsetting
is *not* ok. it has to be set WAIT (yield one clock) and *then* clear)
added a simulation function get_exception_info() which returns a
LDSTExceptionTuple containing the yield()ed exception information
reason: at the time that the exception was raised, THAT is when
the exception information must be obtained. by the time that
the LDST port is_ld_i/is_st_i is lowered, that is far too late
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 15:51:04 +0000 (15:51 +0000)]
whitespace
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 15:22:10 +0000 (15:22 +0000)]
add misaligned ld/st to trigger an exception
Luke Kenneth Casson Leighton [Fri, 3 Dec 2021 15:09:22 +0000 (15:09 +0000)]
comment out dsisr and dar in mmu FSM for now
Tobias Platen [Thu, 2 Dec 2021 19:01:56 +0000 (20:01 +0100)]
Merge branch 'master' of ssh://git.libre-riscv.org:922/soc
Tobias Platen [Thu, 2 Dec 2021 18:57:43 +0000 (19:57 +0100)]
cleanup test_compldst_multi_mmu.py
Luke Kenneth Casson Leighton [Thu, 2 Dec 2021 16:56:58 +0000 (16:56 +0000)]
add a bitvector remap function, the plan is to use it to reduce the size of
regfile bitvector hazards
Tobias Platen [Thu, 2 Dec 2021 16:25:12 +0000 (17:25 +0100)]
fix test_random in test_loadstore1
Luke Kenneth Casson Leighton [Thu, 2 Dec 2021 15:07:38 +0000 (15:07 +0000)]
use new namedtuple in core when calling regspec_decode()
Luke Kenneth Casson Leighton [Thu, 2 Dec 2021 14:56:50 +0000 (14:56 +0000)]
add module parameter to regspec_decode and therefore to get_byregfiles
as well
Jacob Lifshay [Thu, 2 Dec 2021 03:03:26 +0000 (19:03 -0800)]
remove bitmanip fu cuz ternlogi (the only instruction implemented so far) is in shiftrot instead
Jacob Lifshay [Thu, 2 Dec 2021 03:01:12 +0000 (19:01 -0800)]
add ternlogi to shiftrot
Jacob Lifshay [Thu, 2 Dec 2021 02:12:42 +0000 (18:12 -0800)]
format code
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 19:47:27 +0000 (19:47 +0000)]
stack of changes to MultiCompUnit to speed it up
a lot of comb settings were changed to sync as a hack-experiment
when TestIssuer (ls180) was being developed, to try to speed up ECP5
timing. those need reverting to reduce latency
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 16:25:10 +0000 (16:25 +0000)]
experimenting with option to shorten MultiCompUnit delays
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 16:24:50 +0000 (16:24 +0000)]
create single-stage ALU pipeline, shorten latency on in-order core
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 15:59:27 +0000 (15:59 +0000)]
allow MultiCompUnit to set read and write latches to run on sync or comb
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 13:41:21 +0000 (13:41 +0000)]
FunctionUnitBaseMulti which derives from ReservationStations2
*requires* the muxid to be set for each RS. these can be set
combinatorially no problem, to indicate to ReservationStations2 where
to route the data after it pops out the end of the "real" ALU
without these muxids being set everything is royally screwed
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 13:01:01 +0000 (13:01 +0000)]
better name for read latch in core.py
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 13:00:41 +0000 (13:00 +0000)]
use m.submodules[name] instead of getattr
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 12:20:23 +0000 (12:20 +0000)]
remove redundant / mis-named variable in core
Luke Kenneth Casson Leighton [Wed, 1 Dec 2021 12:19:33 +0000 (12:19 +0000)]
code-comments