Luke Kenneth Casson Leighton [Sat, 8 Oct 2022 09:29:38 +0000 (10:29 +0100)]
add addex to csv and sv_analysis db. also needs CryIn.OV enum
added quick test_pysvp64dis.py test too
Luke Kenneth Casson Leighton [Sat, 8 Oct 2022 09:10:15 +0000 (10:10 +0100)]
misnamed instruction, lfiwzx
Luke Kenneth Casson Leighton [Fri, 7 Oct 2022 12:47:03 +0000 (13:47 +0100)]
more work on inssort. add useful reg-dump in ISACaller
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 21:41:04 +0000 (22:41 +0100)]
nope. failfirst needs to always save the result, but truncate VL *after*.
https://bugs.libre-soc.org/show_bug.cgi?id=936
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 19:19:42 +0000 (20:19 +0100)]
fix fail-first to exclude failed element in VLi=0 mode
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 17:34:38 +0000 (18:34 +0100)]
sort out CROPs fail-first in ISACaller. needed to take a copy of CR
for when sv.cmp (and other pseudocode) *overwrites* CR and it needs
restoring (when VLI=0). also needed to identify 3-bit and 5-bit
ffirst mode, and extract bottom 2 bits of BF
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 16:26:24 +0000 (17:26 +0100)]
make fail-first cope with sv.cmp which uses CR[BF]
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 14:59:54 +0000 (15:59 +0100)]
add insert sort svp64 test
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 14:58:11 +0000 (15:58 +0100)]
search for BF in registers to over-ride Vector lookup into CR Register
CR[32+BF...] is used, so it is more complex
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 13:36:23 +0000 (14:36 +0100)]
starting to add sv.cmp support and failfirst, had to add
SVMode to SVP64RMModeDecode to identify the different RM modes first
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 12:16:28 +0000 (13:16 +0100)]
add PredicateBaseRM decode to CR Ops Simple mode as well as ff=3-bit
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 11:57:57 +0000 (12:57 +0100)]
add vli mode to ff=5 CR ops
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 11:51:42 +0000 (12:51 +0100)]
whoops must only be PredicateBaseRM in CROpFF5RM
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 11:43:00 +0000 (12:43 +0100)]
add sv.cmp (ffirst-5) decode/encode asm support
* sv/trans/svp64.py needed a totally different ffirst handling
* CROpFF5RM needs to derive from FFPRRc0BaseRM and PredicateWidthBaseRM
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 11:18:30 +0000 (12:18 +0100)]
slightly different crops failfirst mode bits
Luke Kenneth Casson Leighton [Thu, 6 Oct 2022 10:53:59 +0000 (11:53 +0100)]
add sv.cmp and try fail-first test_pysvp64dist.py
Luke Kenneth Casson Leighton [Sun, 2 Oct 2022 11:38:32 +0000 (12:38 +0100)]
remove complaints about standard Cray-style Vectors for the past 40 years
being able to have VL set to zero dynamically at runtime.
it is not appropriate to have complaints about 40-year-old standard
canonical behaviour of Cray Vectors in source code comments
VL=0 being set dynamically at runtime based on an algorithm input
sets the operations to nop because that is expected behaviour.
to not have VL=0 would be catastrophically inconvenient: it would
require either Illegal-Instruction traps to be raised, or Condition
Codes to be set and followed up with instructions to test and
branch for the dynamic condition when RA was zero, or to pre-test
for RA or CTR pre-being-zero prior to entry into a loop.
Data-Dependent Fail-First would become irrevocably damaged as well
as it is possible for VL to be set to zero at that time (first
test fails)
sv.bc would also be irrevocably damaged as it would no longer
be possible in VLSET mode to have VL become truncated to zero
dynamically based on a Condition Code.
in all it is pretty catastrophic to the entire Cray-Vector paradigm
and would severely damage SimpleV to disallow VL=0 purely for
arbitrary "convenience"
Luke Kenneth Casson Leighton [Sun, 2 Oct 2022 02:29:39 +0000 (03:29 +0100)]
comments for why preinc is called for svstep
Luke Kenneth Casson Leighton [Sun, 2 Oct 2022 02:21:08 +0000 (03:21 +0100)]
comment out selectableint getitem logs
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 23:35:08 +0000 (00:35 +0100)]
skip svstate_pre_inc on svremap
should not be needed
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 23:26:23 +0000 (00:26 +0100)]
no svstate instruction
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 23:24:25 +0000 (00:24 +0100)]
svstep calls SVSTATE_NEXT so needs svstate_pre_inc
setvl no longer calls SVSTATE_NEXT so does not
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 23:10:15 +0000 (00:10 +0100)]
remove special case from setvl calling SVSTATE_NEXT,
only accessible through svstep now
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 21:40:21 +0000 (22:40 +0100)]
replacing setvl-svstep with just svstep
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 21:36:04 +0000 (22:36 +0100)]
replacing setvl-svstep with just svstep
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 21:27:25 +0000 (22:27 +0100)]
replacing setvl-svstep with just svstep
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 16:00:06 +0000 (17:00 +0100)]
comments
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 15:55:13 +0000 (16:55 +0100)]
comments
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 15:44:56 +0000 (16:44 +0100)]
minor cleanup in ISACaller on result handling
create a dictionary matched with output reg names, adapt handle_carry
handle_overflow etc. to use it. mostly
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 13:43:24 +0000 (14:43 +0100)]
simplify ISACaller execute_one
Luke Kenneth Casson Leighton [Sat, 1 Oct 2022 13:21:44 +0000 (14:21 +0100)]
simplify setting default SVSHAPE SPRs to zero
Jacob Lifshay [Sat, 1 Oct 2022 00:10:44 +0000 (17:10 -0700)]
increase pcdec. output compression by skipping impossible codes
Jacob Lifshay [Fri, 30 Sep 2022 23:08:53 +0000 (16:08 -0700)]
prefix codes tests pass
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 17:39:45 +0000 (18:39 +0100)]
no need for ctr mode on sv.bc
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 17:37:35 +0000 (18:37 +0100)]
ctr mode not needed, just use unconditional CTR dec
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 13:39:27 +0000 (14:39 +0100)]
set srcstep/dststep to zero in StepLoop (ISACaller) at loopend
see if a different style of looping can be used
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 13:03:20 +0000 (14:03 +0100)]
add sv.bc vlset-inverted test
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 12:59:05 +0000 (13:59 +0100)]
comments/variables-cleanup
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 11:28:19 +0000 (12:28 +0100)]
add sv.bc vlset-inverted test
this one inverts all the logic (some instead of all, LE instead of GT,
VL-truncate if FAIL)
and thus can swap the success-fail branch point. avoids one branch in loops
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 10:56:55 +0000 (11:56 +0100)]
add sv.bc/vs - VLset - test. truncates VL at the vector-condition-fail point
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 09:47:41 +0000 (10:47 +0100)]
add code-comments in variance_svp64_real.s on how to use sv.bc/ctr/all
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 09:37:57 +0000 (10:37 +0100)]
add new sv.bc CTR-loop test, subtracts VL from CTR
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 08:14:26 +0000 (09:14 +0100)]
whitespace
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 08:12:41 +0000 (09:12 +0100)]
use regs variables in get_predint
Luke Kenneth Casson Leighton [Fri, 30 Sep 2022 08:06:58 +0000 (09:06 +0100)]
comments
Jacob Lifshay [Fri, 30 Sep 2022 03:49:16 +0000 (20:49 -0700)]
fix pcdec. assembly -- merge into va_form() since it's no longer VA2-form
Jacob Lifshay [Fri, 30 Sep 2022 03:46:38 +0000 (20:46 -0700)]
fix pcdec.'s form
Jacob Lifshay [Fri, 30 Sep 2022 03:17:50 +0000 (20:17 -0700)]
rewrite pcdec. pseudocode to work better for JPEG
the pcdec. unittests aren't updated yet
Jacob Lifshay [Fri, 30 Sep 2022 01:44:47 +0000 (18:44 -0700)]
add lookup table generation for JPEG decode
Jacob Lifshay [Fri, 30 Sep 2022 01:44:23 +0000 (18:44 -0700)]
allow logging function to be overridden for Mem.log_fancy
Jacob Lifshay [Thu, 29 Sep 2022 23:21:22 +0000 (16:21 -0700)]
convert svp64 bigint unittests to use TestAccumulatorBase
Jacob Lifshay [Thu, 29 Sep 2022 22:46:48 +0000 (15:46 -0700)]
finish changing to use adde, not addeo for bigint add
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 18:00:30 +0000 (19:00 +0100)]
sv.adde not sv.addeo
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 14:06:52 +0000 (15:06 +0100)]
destination for maddedu and divmod2du for RS defaults to RC for scalar
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 13:59:45 +0000 (14:59 +0100)]
wowser, complex. implementing maddedu implicit RC/RS rules.
still TODO
<!-- bit 8 of EXTRA is clear: RS.[s|v]=RT.[s|v]+MAXVL
<!-- bit 8 of EXTRA is set : RS.[s|v]=RC.[s|v]
actually it is currently "if RC is scalar then RS=RC" which is more
sensible
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 13:57:26 +0000 (14:57 +0100)]
add carry-roll-over-vector-mul-with-add (!) unit test
test_caller_svp64_bigint.py
https://bugs.libre-soc.org/show_bug.cgi?id=937
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 11:09:43 +0000 (12:09 +0100)]
comments
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 10:56:20 +0000 (11:56 +0100)]
add shift-left and shift-right scalar-to-vector tests
Luke Kenneth Casson Leighton [Thu, 29 Sep 2022 01:39:55 +0000 (02:39 +0100)]
update iterators in ISACaller, not used yet
Jacob Lifshay [Thu, 29 Sep 2022 03:10:05 +0000 (20:10 -0700)]
rename madded->maddedu for consistency with PowerISA maddhdu instruction
Jacob Lifshay [Thu, 29 Sep 2022 03:05:02 +0000 (20:05 -0700)]
rename divrem2du->divmod2du for consistency with PowerISA mod* instructions
Jacob Lifshay [Thu, 29 Sep 2022 02:46:54 +0000 (19:46 -0700)]
add bigint tests and fix madded pseudocode
Jacob Lifshay [Thu, 29 Sep 2022 02:46:15 +0000 (19:46 -0700)]
add bigint ops
Jacob Lifshay [Thu, 29 Sep 2022 02:45:00 +0000 (19:45 -0700)]
fill out dsld/dsrd pseudocode
Jacob Lifshay [Thu, 29 Sep 2022 02:43:16 +0000 (19:43 -0700)]
add missing DRAFT comment
Jacob Lifshay [Thu, 29 Sep 2022 02:39:27 +0000 (19:39 -0700)]
fix test_minor_30
Jacob Lifshay [Thu, 29 Sep 2022 02:39:07 +0000 (19:39 -0700)]
format code
Jacob Lifshay [Thu, 29 Sep 2022 02:33:00 +0000 (19:33 -0700)]
clean up bigint instruction naming
Jacob Lifshay [Thu, 29 Sep 2022 02:29:20 +0000 (19:29 -0700)]
remove unnecesary commented code
Jacob Lifshay [Thu, 29 Sep 2022 02:08:07 +0000 (19:08 -0700)]
add unofficial and comment2 fields to minor_31.csv
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 23:49:46 +0000 (00:49 +0100)]
srcstep
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 21:14:17 +0000 (22:14 +0100)]
rename iterators init function
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 21:04:16 +0000 (22:04 +0100)]
redundant comment
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 21:00:23 +0000 (22:00 +0100)]
split out svstate update in ISACaller
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 20:44:54 +0000 (21:44 +0100)]
move failfirst check to separate function in ISACaller
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 20:00:10 +0000 (21:00 +0100)]
new revision of dsld
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 18:33:48 +0000 (19:33 +0100)]
add double-sld pseudocode, first draft
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 18:33:00 +0000 (19:33 +0100)]
add limit argument to MASK() helper
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 18:04:06 +0000 (19:04 +0100)]
add Z23 shift-mode fields.txt
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 16:20:07 +0000 (17:20 +0100)]
bugfix reset remaps and get subvl early
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 13:34:21 +0000 (14:34 +0100)]
comments on horizontal-or
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 13:27:21 +0000 (14:27 +0100)]
make matrix horizontal-remap example more generic
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 13:18:42 +0000 (14:18 +0100)]
add horizontal-or-reduction example that thoroughly abuses the way
that Matrix REMAP works, ignoring the B Matrix entirely
Luke Kenneth Casson Leighton [Wed, 28 Sep 2022 13:11:55 +0000 (14:11 +0100)]
whoops VL incorrect in svshape markdown RTL for matrix REMAP
Jacob Lifshay [Wed, 28 Sep 2022 02:25:46 +0000 (19:25 -0700)]
extracting demo JPEG bitstream works
Luke Kenneth Casson Leighton [Tue, 27 Sep 2022 16:53:00 +0000 (17:53 +0100)]
add unpack predicated unit test
Luke Kenneth Casson Leighton [Tue, 27 Sep 2022 16:25:05 +0000 (17:25 +0100)]
hack to check skipping on predicate being all-zero.
HOWEVER... this will not work on sv.branches
Luke Kenneth Casson Leighton [Tue, 27 Sep 2022 15:40:32 +0000 (16:40 +0100)]
sort out predicate loop-skip on pack/unpack
Luke Kenneth Casson Leighton [Tue, 27 Sep 2022 14:39:18 +0000 (15:39 +0100)]
adapt loops to include predicate-mask skipping in ISACaller
currently not working, investigating (disabled for now)
Konstantinos Margaritis [Tue, 27 Sep 2022 10:23:13 +0000 (10:23 +0000)]
fix typo
Konstantinos Margaritis [Tue, 27 Sep 2022 10:08:09 +0000 (10:08 +0000)]
comment out more debug messages and reference C function
Konstantinos Margaritis [Tue, 27 Sep 2022 10:07:02 +0000 (10:07 +0000)]
comment out debug messages
Konstantinos Margaritis [Tue, 27 Sep 2022 10:04:49 +0000 (10:04 +0000)]
Working version of VP8 DCT4x4 in SVP64
Konstantinos Margaritis [Tue, 27 Sep 2022 10:03:12 +0000 (10:03 +0000)]
remove unused prototypes
Jacob Lifshay [Tue, 27 Sep 2022 04:05:38 +0000 (21:05 -0700)]
add WIP jpeg decoder demo
this includes a tiny test jpeg that's <2kB, so should be fine to be in git.
Jacob Lifshay [Mon, 26 Sep 2022 23:01:03 +0000 (16:01 -0700)]
add more tests and fix missing corner case
Jacob Lifshay [Mon, 26 Sep 2022 22:59:33 +0000 (15:59 -0700)]
pcdec.: change CR0.eq to be early-stop-needed to fit with data-dependent fail-first
Jacob Lifshay [Mon, 26 Sep 2022 22:20:13 +0000 (15:20 -0700)]
add checks for pcdec. once=1
Jacob Lifshay [Mon, 26 Sep 2022 21:50:34 +0000 (14:50 -0700)]
more cleanup after swapping RA/RB for pcdec.
Jacob Lifshay [Mon, 26 Sep 2022 21:48:55 +0000 (14:48 -0700)]
clean up after lkcl swapped RA/RB for pcdec.