Jacob Lifshay [Mon, 11 Sep 2023 21:32:10 +0000 (14:32 -0700)]
skip madd* tests since they're not implemented
Jacob Lifshay [Mon, 11 Sep 2023 20:30:26 +0000 (13:30 -0700)]
MASK was moved into ISACallerHelper class
Jacob Lifshay [Mon, 11 Sep 2023 20:29:51 +0000 (13:29 -0700)]
set parent pspec to class with XLEN = 64
Jacob Lifshay [Mon, 11 Sep 2023 20:28:46 +0000 (13:28 -0700)]
set 'soc' filter to filter out v3.1 insns
Jacob Lifshay [Mon, 11 Sep 2023 20:27:43 +0000 (13:27 -0700)]
fix ALU test by setting 'soc' filter to filter out addex
Jacob Lifshay [Mon, 11 Sep 2023 20:26:32 +0000 (13:26 -0700)]
mark src/soc/experiment/test/test_compldst_multi.py as broken
Jacob Lifshay [Mon, 11 Sep 2023 20:24:55 +0000 (13:24 -0700)]
mark test_pi2ls.py as broken
I tried to fix it, but a whole bunch needs changing so I gave up for now
Jacob Lifshay [Mon, 11 Sep 2023 20:23:56 +0000 (13:23 -0700)]
add cocotb to gitlab-ci
Jacob Lifshay [Mon, 11 Sep 2023 16:54:57 +0000 (09:54 -0700)]
add pytest-output-to-files's output dir to gitignore
Jacob Lifshay [Wed, 16 Aug 2023 02:10:31 +0000 (19:10 -0700)]
make CI work
Andrey Miroshnikov [Thu, 27 Jul 2023 11:35:00 +0000 (11:35 +0000)]
Makefile: Added rule for generating mw-compatible core with SVP64
issue_verilog.py: Added argument for mw+svp64.
It's intended to be an unintrusive change for testing. No default behaviour
has been changed.
Cesar Strauss [Sun, 1 Jan 2023 12:30:41 +0000 (09:30 -0300)]
Handle newer nMigen adding a "bench" hierarchy root in VCD files
Cesar Strauss [Tue, 15 Nov 2022 11:52:58 +0000 (08:52 -0300)]
Keep the valid signal from the formal engine ALU stable, until read
Cesar Strauss [Fri, 28 Oct 2022 13:15:34 +0000 (10:15 -0300)]
Check that exactly one ALU write is made, per instruction
Cesar Strauss [Fri, 28 Oct 2022 12:56:41 +0000 (09:56 -0300)]
Check cover and bmc in separate sub-tests
This allows generating a bmc trace even if cover failed.
Cesar Strauss [Wed, 26 Oct 2022 22:02:21 +0000 (19:02 -0300)]
Reset req_l latch on system reset
Even if the holding register of SRLatch is really reset-less, it is still
reset at startup, because the latch reset port has reset=1. But, this only
works if the latch reset port is driven by the sync domain. In case of
req_l, it's driven by the comb domain, so the latch is not reset on system
reset.
Generate a pulse on system reset, and combine it with the latch reset.
Cesar Strauss [Sun, 16 Oct 2022 13:01:37 +0000 (10:01 -0300)]
Reset src_l latch on issue_i
When the input operand is masked, there will be no go_i pulse to
reset the src_l latch, so we force it to reset on issue_i, as well.
Luke Kenneth Casson Leighton [Mon, 24 Oct 2022 21:05:54 +0000 (22:05 +0100)]
only NLnet sponsor
Cesar Strauss [Sun, 16 Oct 2022 11:52:41 +0000 (08:52 -0300)]
Move test to expose bug in MultiCompUnit
A bug was discovered in MultiCompUnit, which occurs only when an
instruction with zero_a or imm_ok is immediately followed by one with
a masked input port. If we move one test out of the way, we get exactly
that.
Cesar Strauss [Wed, 12 Oct 2022 13:32:49 +0000 (10:32 -0300)]
Check invariant for instruction operands
For every instruction issued, at any point in time, each operand
was either:
1) Already read
2) Not read yet, but the read is pending (rel_o high)
3) Masked
Cesar Strauss [Wed, 12 Oct 2022 12:41:42 +0000 (09:41 -0300)]
If the ALU is idle, do not assert valid
Cesar Strauss [Sun, 9 Oct 2022 21:01:36 +0000 (18:01 -0300)]
Count zero_a and imm_data.ok as masked read transactions
Cesar Strauss [Sun, 9 Oct 2022 10:08:13 +0000 (07:08 -0300)]
Don't issue while busy
Otherwise, the formal engine, in some cases, will happily assert issue_i
while busy_o is high. MultiCompUnit can't handle that.
Cesar Strauss [Sat, 8 Oct 2022 21:57:57 +0000 (18:57 -0300)]
Add count of masked reads
Cesar Strauss [Sat, 8 Oct 2022 21:08:48 +0000 (18:08 -0300)]
Add ALU read transaction counter
Cesar Strauss [Sat, 8 Oct 2022 21:05:32 +0000 (18:05 -0300)]
Add ALU write transaction counter
Cesar Strauss [Sat, 8 Oct 2022 20:36:39 +0000 (17:36 -0300)]
Add write transaction counter
Cesar Strauss [Wed, 5 Oct 2022 22:11:09 +0000 (19:11 -0300)]
Fix duplicate line (copy & paste error)
Cesar Strauss [Sat, 1 Oct 2022 22:08:00 +0000 (19:08 -0300)]
Add counter for operand reads
Cesar Strauss [Sat, 1 Oct 2022 22:06:23 +0000 (19:06 -0300)]
Avoid toggling go_i when rel_o is low
This must be enforced on the scoreboard, in its proof.
Cesar Strauss [Sat, 1 Oct 2022 22:00:55 +0000 (19:00 -0300)]
Leave shadow / die proof for last
Cesar Strauss [Sat, 1 Oct 2022 14:48:19 +0000 (11:48 -0300)]
Start of formal proof of MultiCompUnit
Create a "random" ALU, controlled by the formal engine.
Add an "issue" transaction counter and ask the fomal engine to
give an example.
Jacob Lifshay [Tue, 16 Aug 2022 06:43:13 +0000 (23:43 -0700)]
change goldschmidt_div_sqrt to use nmutil.plain_data rather than dataclasses
Luke Kenneth Casson Leighton [Sun, 14 Aug 2022 21:09:13 +0000 (22:09 +0100)]
grr not a yield function
Luke Kenneth Casson Leighton [Sun, 14 Aug 2022 18:50:46 +0000 (19:50 +0100)]
add get_fpregs stub function to HDLstate
Luke Kenneth Casson Leighton [Wed, 6 Jul 2022 13:22:54 +0000 (14:22 +0100)]
update pinmux submodule, rename to "fabric"
Luke Kenneth Casson Leighton [Wed, 6 Jul 2022 08:11:47 +0000 (09:11 +0100)]
add fabric compatibility mode
Luke Kenneth Casson Leighton [Tue, 5 Jul 2022 13:20:11 +0000 (14:20 +0100)]
MulOutputData was only 64-bit output not 128-bit
meaning that mulhd is truncated (always zero)
https://bugs.libre-soc.org/show_bug.cgi?id=855
Luke Kenneth Casson Leighton [Mon, 4 Jul 2022 15:12:46 +0000 (16:12 +0100)]
add signal for resetting trap internal state (kaivb cache)
Luke Kenneth Casson Leighton [Mon, 4 Jul 2022 15:07:29 +0000 (16:07 +0100)]
set msr_o.data not msr_o Record in trap main_stage.py
Luke Kenneth Casson Leighton [Sun, 26 Jun 2022 11:57:33 +0000 (12:57 +0100)]
adapt TRAP function in main state pipeline to put KAIVB
into top bits
Luke Kenneth Casson Leighton [Sun, 26 Jun 2022 11:38:52 +0000 (12:38 +0100)]
store KAIVB SPR 850 in TRAP Pipeline
by redirecting OP_MTSPR and OP_MFSPR to it
https://bugs.libre-soc.org/show_bug.cgi?id=859
Luke Kenneth Casson Leighton [Sun, 26 Jun 2022 11:11:42 +0000 (12:11 +0100)]
reduce icache/dcache TLB sizes
Luke Kenneth Casson Leighton [Sun, 26 Jun 2022 11:11:14 +0000 (12:11 +0100)]
update trap test_pipe_caller.py to use up-to-date test method
Luke Kenneth Casson Leighton [Sun, 26 Jun 2022 11:10:35 +0000 (12:10 +0100)]
missing module argument to TestRunner execute
Luke Kenneth Casson Leighton [Fri, 17 Jun 2022 14:13:11 +0000 (15:13 +0100)]
convert trap test_pipe_caller.py to consistent format
(like alu test_pipe_caller.py)
Andrey Miroshnikov [Mon, 23 May 2022 22:17:07 +0000 (22:17 +0000)]
Change usage of WB sel for individual control
Luke Kenneth Casson Leighton [Sun, 1 May 2022 17:56:27 +0000 (18:56 +0100)]
split out front of div into separate stage, still too long combinatorial
blocks
Luke Kenneth Casson Leighton [Sat, 30 Apr 2022 21:20:02 +0000 (22:20 +0100)]
add missing module
Luke Kenneth Casson Leighton [Sat, 30 Apr 2022 21:17:58 +0000 (22:17 +0100)]
split off CR0/XER production in DIV Function Unit into separate stage
due to massive combinatorial chains
Luke Kenneth Casson Leighton [Sat, 30 Apr 2022 12:52:00 +0000 (13:52 +0100)]
clear out DEC in core.cur_state.dec due to spurious interrupt.
this is slightly complicated. the STATE regfile contains pc, msr, svstate,
dec, and tb, being a reflection of CoreState. reading from STATE regfile
is on a one-clock delay. the DEC/TB FSM needs to decrement DEC and increment
TB, by reading from the STATE regfile and then writing a new value.
of course, the SPR pipeline has to get a word in edgeways as well.
but...
the complication comes in that it is the PowerDecoder2 which receives
a *cached* copy of DEC, and this cached copy is what has (up until now)
been out-of-date with what is in the STATE regfile.
the hack-job-solution is to zero-out the cached copy when the SPR pipeline
writes a new value to DEC. the DEC/TB FSM will then rewrite a correct
value into it.
given that PowerDecoder2 only uses the MSB of DEC (and the EE bit of MSR)
to determine whether to fire an interrupt, this should be perfectly fine.
Cesar Strauss [Sat, 30 Apr 2022 11:08:27 +0000 (08:08 -0300)]
Implement transparent read port option on the XOR wrapper SRAM
Add a bypass path when simultaneously reading and writing to the same
address.
There needs to be an independent Mux for each write lane, since we may not be
writing on all lanes, necessarily.
Jacob Lifshay [Fri, 29 Apr 2022 18:56:49 +0000 (11:56 -0700)]
fix waay-too-precise error requirements
reduces 128x64->64-bit divider to 220k cells rather than 700k cells
Luke Kenneth Casson Leighton [Fri, 29 Apr 2022 09:32:24 +0000 (10:32 +0100)]
add option to set small cache sizes in
issuer_verilog.py
Jacob Lifshay [Fri, 29 Apr 2022 06:05:08 +0000 (23:05 -0700)]
add comment
Jacob Lifshay [Fri, 29 Apr 2022 06:04:05 +0000 (23:04 -0700)]
fix so HDL works for 5, 8, 16, 32, and 64-bits.
Jacob Lifshay [Fri, 29 Apr 2022 05:40:32 +0000 (22:40 -0700)]
HDL works for io_width=5
Cesar Strauss [Thu, 28 Apr 2022 22:37:20 +0000 (19:37 -0300)]
Test simultaneous transparent reads and partial writes
Should catch some corner cases.
Jacob Lifshay [Thu, 28 Apr 2022 09:26:09 +0000 (02:26 -0700)]
add docs for clz
Jacob Lifshay [Thu, 28 Apr 2022 09:19:01 +0000 (02:19 -0700)]
add WIP HDL version of goldschmidt division -- it's currently broken
Luke Kenneth Casson Leighton [Thu, 28 Apr 2022 08:21:42 +0000 (09:21 +0100)]
notes added to setup.py - absolute paranoia is needed on
what dependencies the HDL has. dropping USD 16 million on 7nm Mask Charges
you absolutely cannot have arbitrary software downloaded off the internet
from external sources.
Jacob Lifshay [Thu, 28 Apr 2022 05:52:41 +0000 (22:52 -0700)]
move GoldschmidtDivState
Jacob Lifshay [Thu, 28 Apr 2022 05:50:42 +0000 (22:50 -0700)]
add FIXME comments
Jacob Lifshay [Thu, 28 Apr 2022 05:39:08 +0000 (22:39 -0700)]
add the goldschmidt sqrt/rsqrt algorithm, still need code to calculate good parameters
Jacob Lifshay [Wed, 27 Apr 2022 23:21:46 +0000 (16:21 -0700)]
switch cached-property dependency to using libre-soc's git repo
Jacob Lifshay [Wed, 27 Apr 2022 06:50:37 +0000 (23:50 -0700)]
improved goldschmidt division algorithm parameter optimization algorithm
Jacob Lifshay [Wed, 27 Apr 2022 04:57:40 +0000 (21:57 -0700)]
split out non-derived params into separate class without all the accuracy checking stuff.
Jacob Lifshay [Wed, 27 Apr 2022 04:52:08 +0000 (21:52 -0700)]
split out n_hat as separate property
Jacob Lifshay [Wed, 27 Apr 2022 04:41:05 +0000 (21:41 -0700)]
add default_cost_fn
Jacob Lifshay [Wed, 27 Apr 2022 04:36:23 +0000 (21:36 -0700)]
move GoldschmidtDivParams.get to bottom of class
Jacob Lifshay [Wed, 27 Apr 2022 04:27:29 +0000 (21:27 -0700)]
rename _goldschmidt_div_ops to GoldschmidtDivState.__make_ops
Jacob Lifshay [Tue, 26 Apr 2022 03:56:30 +0000 (20:56 -0700)]
goldschmidt division works! still needs better parameter selection tho...
Jacob Lifshay [Tue, 26 Apr 2022 02:00:01 +0000 (19:00 -0700)]
fix goofed __init__.py file name
Jacob Lifshay [Mon, 25 Apr 2022 08:44:31 +0000 (01:44 -0700)]
working on goldschmidt_div_sqrt.py
Jacob Lifshay [Mon, 25 Apr 2022 08:44:13 +0000 (01:44 -0700)]
add cached_property dependency
Jacob Lifshay [Sat, 23 Apr 2022 02:28:37 +0000 (19:28 -0700)]
working on goldschmidt division algorithm
Luke Kenneth Casson Leighton [Fri, 22 Apr 2022 13:13:21 +0000 (14:13 +0100)]
whitespace
Jacob Lifshay [Fri, 22 Apr 2022 07:58:17 +0000 (00:58 -0700)]
add WIP goldschmidt division algorithm
Cesar Strauss [Sun, 17 Apr 2022 22:27:58 +0000 (19:27 -0300)]
Implement a 1W/1R register file, XOR style
Test case seems to be working.
Total 6x 1RW SRAMs were used, but no DFF RAM was needed.
Still needs a bypass for a transparent read option.
Cesar Strauss [Sun, 17 Apr 2022 20:19:32 +0000 (17:19 -0300)]
Formal proof of pseudo 1W/2R SRAM
Mostly taken from the proof of the 1W/1R SRAM. Sorry for all the
duplication, they are all variations of the same theme.
Cesar Strauss [Sun, 17 Apr 2022 18:44:23 +0000 (15:44 -0300)]
Add transparent option for the full read port
Cesar Strauss [Sun, 17 Apr 2022 18:13:52 +0000 (15:13 -0300)]
Implement a pseudo 1W/2R memory
This is intended for a XOR-based 6x1RW implementation of a 1R/1W
register file.
It costs 3 x 1RW memory blocks.
Luke Kenneth Casson Leighton [Sat, 16 Apr 2022 22:03:27 +0000 (23:03 +0100)]
reduce dcache/icache number of ways, to fit into ECP5 45k resource
Tobias Platen [Sat, 16 Apr 2022 20:50:29 +0000 (22:50 +0200)]
part two of issuer_fix: read pspec.microwatt_old and pspec.microwatt_debug
Cesar Strauss [Sat, 16 Apr 2022 20:44:30 +0000 (17:44 -0300)]
Check non-transparent 1W/1R SRAM wrapper
Add special case for the non-transparent case, where a simultaneous write
returns the old value of the holding register.
This completes the proof for this 1W/1R SRAM wrapper
Cesar Strauss [Sat, 16 Apr 2022 20:36:04 +0000 (17:36 -0300)]
Enable read port for non-transparent memories
For some reason, the reset value (=1) of the port enable is not recognized
by the formal engine, even if the simulation is fine with it.
Tobias Platen [Sat, 16 Apr 2022 20:23:14 +0000 (22:23 +0200)]
Merge ssh://git.libre-riscv.org:922/soc
Tobias Platen [Sat, 16 Apr 2022 20:21:34 +0000 (22:21 +0200)]
part one of issuer_fix: add parameter to issuer_verilog.py
Cesar Strauss [Sat, 16 Apr 2022 17:33:04 +0000 (14:33 -0300)]
Add port declarations to the SRAM wrappers
Cesar Strauss [Sat, 16 Apr 2022 17:25:14 +0000 (14:25 -0300)]
Change write lane signal from one-hot to binary
It's simpler to write, and actually runs faster, somehow...
Luke Kenneth Casson Leighton [Sat, 16 Apr 2022 16:21:05 +0000 (17:21 +0100)]
whoops, WBASyncBridge ack signal not wired up!
Luke Kenneth Casson Leighton [Sat, 16 Apr 2022 15:26:33 +0000 (16:26 +0100)]
select width is data_width // data granularity.
(common mistake to make, confused the hell out of me too)
Cesar Strauss [Sat, 16 Apr 2022 14:33:08 +0000 (11:33 -0300)]
Synchronize LVT state, completing the induction proof
Cesar Strauss [Sat, 16 Apr 2022 14:29:05 +0000 (11:29 -0300)]
Sync proof state with downstream memories
Luke Kenneth Casson Leighton [Sat, 16 Apr 2022 13:51:47 +0000 (14:51 +0100)]
put the old microwatt compatibility back
PLEASE DO NOT CHANGE THIS WITHOUT CONSULTATION
Luke Kenneth Casson Leighton [Sat, 16 Apr 2022 12:33:33 +0000 (13:33 +0100)]
blegh.
* add in a new PRE_IDLE state into issue FSM (to allow pc_at_reset to
get a word in edgeways)
* set nia to pc_at_reset on core_rst (not 0x0)
* add in a terrible hack where the STATE regfile entry for PC is
shifted down by one.
absolutely no frickin idea why that is needed
Cesar Strauss [Fri, 15 Apr 2022 19:43:56 +0000 (16:43 -0300)]
Complete moving the induction support into the DUT
That way, we pass the state down, to be checked by single code at
the lowest level, instead of having to check all variations of the
lower level states, at the top level.
Had to increment the proof depth, for some reason.
Cesar Strauss [Fri, 15 Apr 2022 19:36:05 +0000 (16:36 -0300)]
Fix incorrect signal widths
dbg_data is the width of a write lane (granularity) and dbg_wrote is a
single bit.
Cesar Strauss [Fri, 15 Apr 2022 16:55:00 +0000 (13:55 -0300)]
Move part of formal proof to the implementation
For induction, some state has to be synchronized with the
device under test. Make the device itself responsible for checking
this, to avoid exposing its internal workings, and making the proof
more generic.
This is done for PhasedDualPortRegfile, the rest will follow.
Luke Kenneth Casson Leighton [Thu, 14 Apr 2022 13:13:23 +0000 (14:13 +0100)]
add option Spec to XICS ICP/ICS to be able to activate (in a horrible
hack way) make_wb_layout to add stall
this is done by spotting "spec.microwatt_compat" adds stall
blegh