Jacob Lifshay [Thu, 18 Mar 2021 02:45:39 +0000 (19:45 -0700)]
force opcode field to be always specified in binary for opint=True CSVs
Jacob Lifshay [Thu, 18 Mar 2021 02:44:58 +0000 (19:44 -0700)]
add simplev.py to .gitignore
Jacob Lifshay [Thu, 18 Mar 2021 02:32:07 +0000 (19:32 -0700)]
update submodule
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 22:33:15 +0000 (22:33 +0000)]
correct comments
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 22:29:07 +0000 (22:29 +0000)]
re-enable SVP64 ISACaller predicate tests
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 22:25:21 +0000 (22:25 +0000)]
add ascii graphic for extsw svp64 operation
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 22:21:12 +0000 (22:21 +0000)]
add more explanatory comments
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 22:15:58 +0000 (22:15 +0000)]
add twin-predicated extsw SVP64 ISACaller unit test
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 21:29:49 +0000 (21:29 +0000)]
add SVP64 dststep incrementing in PowerDecoder2, Testissuer and ISACaller
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 20:40:49 +0000 (20:40 +0000)]
add CR-based predication to ISACaller
Tobias Platen [Wed, 17 Mar 2021 19:13:40 +0000 (20:13 +0100)]
cleanup raduxmmu._walk_tree
Tobias Platen [Wed, 17 Mar 2021 18:59:12 +0000 (19:59 +0100)]
create iterative mmu lookup loop
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 15:20:02 +0000 (15:20 +0000)]
add SVP64 INT-style predication to ISACaller
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 14:13:02 +0000 (14:13 +0000)]
add predication SVP64 unit test
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 13:21:25 +0000 (13:21 +0000)]
add predication read ports (CR and INT)
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 13:16:24 +0000 (13:16 +0000)]
whoops shift has to be done at same bitwidth
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 13:14:07 +0000 (13:14 +0000)]
split out new_lookup function
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 13:00:28 +0000 (13:00 +0000)]
link up SVP64 RM Mode decoding into PowerDecoder2
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 12:46:54 +0000 (12:46 +0000)]
add priv and mode to RADIXMMU
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 12:34:04 +0000 (12:34 +0000)]
add instr_fetch mode to ISACaller Mem and RADIXMMU
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 12:22:44 +0000 (12:22 +0000)]
whitespace
Luke Kenneth Casson Leighton [Wed, 17 Mar 2021 12:19:06 +0000 (12:19 +0000)]
add in SVP64 RM Mode decoder
Tobias Platen [Tue, 16 Mar 2021 18:49:20 +0000 (19:49 +0100)]
radixmmu: detect badtree
Tobias Platen [Tue, 16 Mar 2021 17:35:35 +0000 (18:35 +0100)]
add valid, leaf to loop
Cesar Strauss [Tue, 16 Mar 2021 10:55:48 +0000 (07:55 -0300)]
Use symbolic values for subfields and bits
Cesar Strauss [Tue, 16 Mar 2021 10:48:05 +0000 (07:48 -0300)]
Add subfield and bit definitions for the SVP64 RM mode field
Cesar Strauss [Tue, 16 Mar 2021 00:12:11 +0000 (21:12 -0300)]
Define and initialise the mode variable, to be used later on
Cesar Strauss [Tue, 16 Mar 2021 00:02:28 +0000 (21:02 -0300)]
Rename class so it does not clash with the enum
Cesar Strauss [Mon, 15 Mar 2021 22:33:34 +0000 (19:33 -0300)]
Fix import
Tobias Platen [Mon, 15 Mar 2021 18:49:44 +0000 (19:49 +0100)]
add rpte bitfields valid and leaf
Luke Kenneth Casson Leighton [Sun, 14 Mar 2021 14:55:28 +0000 (14:55 +0000)]
remove "sv." and replace with "sv" in all SVP64Asm
Luke Kenneth Casson Leighton [Sun, 14 Mar 2021 14:54:45 +0000 (14:54 +0000)]
remove "sv." and replace with "sv" in all SVP64Asm
Cesar Strauss [Sun, 14 Mar 2021 13:46:08 +0000 (10:46 -0300)]
Activate the VL==0 loop with any SVP64 prefix whatsoever
Besides agreeing with documentation, this will ease doing the VL==0
loop entirely on the Fetch FSM, since no opcode decoding is needed.
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 17:07:09 +0000 (17:07 +0000)]
add setvl unit test assertions, add 2nd test
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 16:10:02 +0000 (16:10 +0000)]
get first revision setvl operational in ISACaller
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 13:35:30 +0000 (13:35 +0000)]
add setvl-to-long converter in SVP64Asm (sigh)
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 12:22:16 +0000 (12:22 +0000)]
add setvl unit test
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 12:20:02 +0000 (12:20 +0000)]
update submodule to include simplev setvl
Luke Kenneth Casson Leighton [Sat, 13 Mar 2021 11:33:53 +0000 (11:33 +0000)]
include SVSTATE in namespace, passing to ISACaller
Jacob Lifshay [Fri, 12 Mar 2021 22:55:24 +0000 (14:55 -0800)]
update submodule
Jacob Lifshay [Fri, 12 Mar 2021 22:52:16 +0000 (14:52 -0800)]
add setvl to decoder
Jacob Lifshay [Fri, 12 Mar 2021 22:50:01 +0000 (14:50 -0800)]
autoformat code
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 21:50:44 +0000 (21:50 +0000)]
add OP_SETVL to MicrOp in power_enums.py
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 21:41:57 +0000 (21:41 +0000)]
add ability to set and distinguish RT=0 (RT_OR_ZERO) to OutSel enum
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 15:02:06 +0000 (15:02 +0000)]
use PowerDecoder2.loop_continue instead of no_out_vec
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 14:17:17 +0000 (14:17 +0000)]
remove old code
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 14:16:10 +0000 (14:16 +0000)]
remove old code
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 14:12:41 +0000 (14:12 +0000)]
remove old code
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 12:13:58 +0000 (12:13 +0000)]
add more sophisticated checking of whether SVP64 loop should continue
PowerDecoder2
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 12:03:39 +0000 (12:03 +0000)]
**FOR NOW** LD/ST relies on detection of twin-predication to determine
if it should continue looping.
this needs double-checking
Luke Kenneth Casson Leighton [Fri, 12 Mar 2021 12:00:58 +0000 (12:00 +0000)]
decoding of svp64 reg by name has to occur after immediate is extracted
otherwise tries to identify D(RA) as a GPR which of course fails
Jacob Lifshay [Fri, 12 Mar 2021 05:50:42 +0000 (21:50 -0800)]
add forgotten PO (primary opcode) field to DecodeFields
Cesar Strauss [Thu, 11 Mar 2021 22:52:52 +0000 (19:52 -0300)]
Bring a few test cases from test_caller_64.py
1) Test early out when destination is not a vector
2) Do not increment source register number for scalar operand
Cesar Strauss [Thu, 11 Mar 2021 22:40:47 +0000 (19:40 -0300)]
Test case for two successive SV instructions
This checks that SRCSTEP is reset properly between instructions.
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 20:01:47 +0000 (20:01 +0000)]
add link of RA_OR_ZERO SVP64 detection
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 19:50:19 +0000 (19:50 +0000)]
add detection of whether *full* 7-bit of RA is zero/non-zero
this because RA_OR_ZERO in PowerDecoder2 needs to test if the full
SVP64-extended register is zero
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 18:32:51 +0000 (18:32 +0000)]
add in SVP64 LD/ST basic test for ISACaller
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 18:29:26 +0000 (18:29 +0000)]
whoops sort out when svstate not active in ISACaller
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 18:19:43 +0000 (18:19 +0000)]
whoops PIDR is defined as 32-bits in SPRs.csv (and spec)
Luke Kenneth Casson Leighton [Thu, 11 Mar 2021 18:15:06 +0000 (18:15 +0000)]
add understanding of LDST immediates to SVP64ASM
Tobias Platen [Thu, 11 Mar 2021 17:56:48 +0000 (18:56 +0100)]
fix runtime error
Tobias Platen [Wed, 10 Mar 2021 18:41:11 +0000 (19:41 +0100)]
radix: reading first page table entry
Luke Kenneth Casson Leighton [Wed, 10 Mar 2021 16:32:28 +0000 (16:32 +0000)]
add walk_tree arguments it needs
see https://bugs.libre-soc.org/show_bug.cgi?id=604#c13
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 19:40:19 +0000 (19:40 +0000)]
fix address must convert to SelectableInt
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 19:38:24 +0000 (19:38 +0000)]
call decode_ptre on address to obtain shift, mbits, and pgbase
Tobias Platen [Tue, 9 Mar 2021 19:03:10 +0000 (20:03 +0100)]
whitespace
Tobias Platen [Tue, 9 Mar 2021 18:59:51 +0000 (19:59 +0100)]
RADIX: call self._walk_tree in ld and st
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 18:09:53 +0000 (18:09 +0000)]
debug radix mmu ISACaller
Tobias Platen [Tue, 9 Mar 2021 17:06:59 +0000 (18:06 +0100)]
comment out broken spr code
Tobias Platen [Tue, 9 Mar 2021 16:34:17 +0000 (17:34 +0100)]
_walk_tree: access sprs
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 13:09:35 +0000 (13:09 +0000)]
create first check_perms RADIX ISACaller function
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 12:56:49 +0000 (12:56 +0000)]
move Mem class out of ISACaller
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 12:41:58 +0000 (12:41 +0000)]
cleanup imports
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 12:38:08 +0000 (12:38 +0000)]
move ISACaller RADIX MMU class to separate module
Luke Kenneth Casson Leighton [Tue, 9 Mar 2021 12:30:02 +0000 (12:30 +0000)]
add pgtable and pte calculation to RADIX ISACaller
Cesar Strauss [Tue, 9 Mar 2021 11:00:04 +0000 (08:00 -0300)]
Enable VL==0 vector instruction skip test case
Cesar Strauss [Tue, 9 Mar 2021 10:57:41 +0000 (07:57 -0300)]
Add some extra debug traces to the GTKWave document
Cesar Strauss [Tue, 9 Mar 2021 10:49:03 +0000 (07:49 -0300)]
Create a new signal for the Simulator to wait on
We wait on "core busy" before simulating an instruction. Trouble is, on a
VL==0 loop, there is no issue, so busy is never toggled. As a solution,
export a new insn_done signal with is pulsed either at end of Execute, or
when going back to Fetch due to skipping a vector instruction.
Luke Kenneth Casson Leighton [Mon, 8 Mar 2021 22:43:18 +0000 (22:43 +0000)]
start adding _get_prtable_addr
Luke Kenneth Casson Leighton [Mon, 8 Mar 2021 17:21:34 +0000 (17:21 +0000)]
actually make it possible to disable svp64 on commandline of test_issuer.py
Luke Kenneth Casson Leighton [Mon, 8 Mar 2021 17:00:36 +0000 (17:00 +0000)]
add option in TestRunner to disable svp64 via commandline test_runner.py nosvp64
currently does nothing
Luke Kenneth Casson Leighton [Mon, 8 Mar 2021 12:10:00 +0000 (12:10 +0000)]
add option to cut out SVP64 from PowerDecoder2
Luke Kenneth Casson Leighton [Mon, 8 Mar 2021 12:07:50 +0000 (12:07 +0000)]
correct comments in sv.add rc=1
Cesar Strauss [Mon, 8 Mar 2021 10:22:57 +0000 (07:22 -0300)]
Remove the unused internal insn_done signal
This was used previously to enable writing to the PC register, but it's
done now within a state transition.
Cesar Strauss [Sun, 7 Mar 2021 22:32:45 +0000 (19:32 -0300)]
Fix argument order to match function declaration
No harm was done, since the second inversion undid the first.
Just the VCD traces were switched.
Cesar Strauss [Sun, 7 Mar 2021 20:55:39 +0000 (17:55 -0300)]
Fix missing NIA update on ISACaller
The effect of this bug was mostly hidden because NIA is later updated at
the end of the SV Loop, in call(). However, in a VL==0 loop, the effect
is apparent, as PC is incremented by 4 instead of 8.
Luke Kenneth Casson Leighton [Sun, 7 Mar 2021 21:35:46 +0000 (21:35 +0000)]
whoops should be "make gitupdate"
Tobias Platen [Sun, 7 Mar 2021 18:22:57 +0000 (19:22 +0100)]
RADIX: read SPRs
Tobias Platen [Sun, 7 Mar 2021 16:39:23 +0000 (17:39 +0100)]
RADIX: implement memassign and call
Luke Kenneth Casson Leighton [Sun, 7 Mar 2021 15:05:24 +0000 (15:05 +0000)]
add SVSTATE read to DMI interface
Cesar Strauss [Sun, 7 Mar 2021 11:49:55 +0000 (08:49 -0300)]
Merge WAIT_RESET into INSN_FETCH on the Issue FSM
In a VL==0 loop, while we are skipping vector instructions, there needs to
be a way to stop the core. Unfortunately, this means duplicating the
corresponding code at instruction end, since there is no state in common
on either loop (the VL==0 instruction skip loop and the VL>1 vector loop).
This does makes it a little non-deterministic.
Normally, we would stop the core at instruction end, but could instead end
up stopping at instruction start. For this to happen, you need to stop the
core at the right moment, just after the instruction ended and before
the next instruction begins.
A way to avoid this, if necessary, would be to create a duplicate of the
INSN_FETCH state, that doesn't wait on "core stop" release.
Since we are now waiting on "core stop" release at instruction start
anyway, there is no need for the special WAIT_RESET state anymore.
Luke Kenneth Casson Leighton [Sun, 7 Mar 2021 11:34:32 +0000 (11:34 +0000)]
move DMI stuff to separate function in issuer.py
Luke Kenneth Casson Leighton [Sun, 7 Mar 2021 11:30:34 +0000 (11:30 +0000)]
update comments in issuer.py
Luke Kenneth Casson Leighton [Sun, 7 Mar 2021 11:12:36 +0000 (11:12 +0000)]
add Rc=1 SVP64 unit test to svp64_cases.py
Cesar Strauss [Sun, 7 Mar 2021 09:41:47 +0000 (06:41 -0300)]
Implement the VL==0 loop
Just after decode, decide whether we proceed to Execute, or shortcut it
directly to the next Fetch.
Cesar Strauss [Sat, 6 Mar 2021 22:38:00 +0000 (19:38 -0300)]
Allow updating the PC and SVSTATE registers while stopped
While the fetch address was overridden by a PC reset, the PC register
itself was updated (with NIA) only after the first instruction ended. Use
the time while the core is stopped to recognise and update the PC and
SVSTATE registers, before the first instruction starts.
Cesar Strauss [Sat, 6 Mar 2021 19:39:14 +0000 (16:39 -0300)]
Enable the Simple-V loop test case
Cesar Strauss [Sat, 6 Mar 2021 19:29:34 +0000 (16:29 -0300)]
Begin to implement the Simple-V loop
After returning from executing an instruction, decide whether to return
to Fetch, or go repeat Execute again.
1) If PC or SVSTATE were updated, go directly to Fetch, without updating
either
2) If there is no vector output, or it's the last VL loop iteration, go
back to Fetch as well, but update the PC. In the latter case, also
reset SRCSTEP
3) Otherwise, we are still in the loop, so increment SVSTEP, and go back
to Execute. But, first, pass through a new state, DECODE_SV, so the new
register numbers can be decoded.
Cesar Strauss [Sat, 6 Mar 2021 17:12:08 +0000 (14:12 -0300)]
Do not reset pc_changed and sv_changed at instruction end
We need these outputs to hold stable, so the Issue FSM can know whether
it can return to the Simple-V loop, or must return to Fetch. A good place
to reset these is at the start, before any instruction is executed.
Cesar Strauss [Sat, 6 Mar 2021 16:46:50 +0000 (13:46 -0300)]
Make the raw opcode input port of the decoder stay stable
During a Simple-V loop, the decoder will be reused repeatedly, so its
raw opcode input needs to hold stable. An alternate way would be to
pass the raw opcode and the SVP64 RM field to the issue FSM, so it could
supply these decoder inputs when needed.