Luke Kenneth Casson Leighton [Mon, 26 Sep 2022 18:55:38 +0000 (19:55 +0100)]
get pack/unpack tests to use sv.ori to copy sequence
01234567
https://bugs.libre-soc.org/show_bug.cgi?id=871
Luke Kenneth Casson Leighton [Mon, 26 Sep 2022 19:19:49 +0000 (20:19 +0100)]
finally got pack/unpack working
https://bugs.libre-soc.org/show_bug.cgi?id=871
Luke Kenneth Casson Leighton [Mon, 26 Sep 2022 17:45:52 +0000 (18:45 +0100)]
code-morph on loop-end detection in ISACaller
there is a bit of a problem in Pack/Unpack in that the end-of-loop
detection is overrunning.
Luke Kenneth Casson Leighton [Mon, 26 Sep 2022 16:14:26 +0000 (17:14 +0100)]
explicit test of src/dststep end-condition in ISACaller iterators
Luke Kenneth Casson Leighton [Mon, 26 Sep 2022 11:03:07 +0000 (12:03 +0100)]
swap RA/RB so that RA|0 is used not RB|0
RB|0 would need a new flag to be passed down to ALUs in HDL
Konstantinos Margaritis [Sun, 25 Sep 2022 17:11:58 +0000 (17:11 +0000)]
fix variables in memory copy
Konstantinos Margaritis [Sun, 25 Sep 2022 16:56:31 +0000 (16:56 +0000)]
comment out debug dumps
Konstantinos Margaritis [Sun, 25 Sep 2022 16:56:08 +0000 (16:56 +0000)]
Fixed SVP64 implentation
Konstantinos Margaritis [Sun, 25 Sep 2022 16:55:43 +0000 (16:55 +0000)]
remove functions as not relevant for this test
Konstantinos Margaritis [Sun, 25 Sep 2022 16:55:15 +0000 (16:55 +0000)]
clean up, convert from uint64 for python due to rounding in Python, fix copying functions
Konstantinos Margaritis [Sun, 25 Sep 2022 16:54:22 +0000 (16:54 +0000)]
add prototypes
Konstantinos Margaritis [Sun, 25 Sep 2022 16:54:04 +0000 (16:54 +0000)]
fix finalize function, clean ups
Konstantinos Margaritis [Sun, 25 Sep 2022 16:53:22 +0000 (16:53 +0000)]
remove unimplemented tests, lower iterations
Konstantinos Margaritis [Sun, 25 Sep 2022 16:52:50 +0000 (16:52 +0000)]
use sv.maddled/mr, cleanup
Konstantinos Margaritis [Sat, 24 Sep 2022 19:56:09 +0000 (19:56 +0000)]
add header
Dmitry Selyutin [Sun, 25 Sep 2022 16:03:30 +0000 (19:03 +0300)]
test_pysvp64dis: sort ld/st idx stride specs
Dmitry Selyutin [Sun, 25 Sep 2022 16:02:58 +0000 (19:02 +0300)]
power_insn: always provide els for ld/st idx stride
Dmitry Selyutin [Sun, 25 Sep 2022 16:02:00 +0000 (19:02 +0300)]
pysvp64asm: fix VLi attribute access
Dmitry Selyutin [Sun, 25 Sep 2022 11:05:14 +0000 (14:05 +0300)]
power_insn: fix and unify /vli specifier
Luke Kenneth Casson Leighton [Sun, 25 Sep 2022 12:50:35 +0000 (13:50 +0100)]
have to sanity-check dz/zz after full qualifier-processing in branch-mode
Luke Kenneth Casson Leighton [Sun, 25 Sep 2022 12:48:23 +0000 (13:48 +0100)]
add dz/sz assertion in is_bc mode
Luke Kenneth Casson Leighton [Sun, 25 Sep 2022 12:46:22 +0000 (13:46 +0100)]
whitespace
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:24:35 +0000 (17:24 +0100)]
move sea check to after all qualifiers are checked
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:21:41 +0000 (17:21 +0100)]
check variable rather than explicit == LDST_IDX
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:16:44 +0000 (17:16 +0100)]
add elstrided/sea on ldst_idx mode
Dmitry Selyutin [Sat, 24 Sep 2022 15:17:59 +0000 (18:17 +0300)]
test_pysvp64dis: test ld/st idx SEA (simple)
Dmitry Selyutin [Sat, 24 Sep 2022 14:51:55 +0000 (17:51 +0300)]
power_insn: support SEA specifier
Dmitry Selyutin [Sat, 24 Sep 2022 14:40:03 +0000 (17:40 +0300)]
pysvp64asm: support /sea specifier
Dmitry Selyutin [Sat, 24 Sep 2022 14:39:24 +0000 (17:39 +0300)]
consts: introduce SEA field
Dmitry Selyutin [Sat, 24 Sep 2022 14:06:44 +0000 (17:06 +0300)]
pysvp64asm: fix comment layout
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:08:59 +0000 (17:08 +0100)]
set sv_mode to 0b01 in element-strided
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:03:58 +0000 (17:03 +0100)]
frickin frick
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:00:28 +0000 (17:00 +0100)]
add assert to stop failfirst+sea
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:31:00 +0000 (17:31 +0100)]
add extra RC1 test, without VLI.
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 16:17:55 +0000 (17:17 +0100)]
add RC1 support to ISACaller.
this involves:
* reading Rc=0 and substituting RC1 in its place OR
* for non-Rc instructions just putting RC1 in place of Rc
* reading VLi flag and adding it to srcstep to put into VL on ffirst hit
* setting the cr-bit to test to EQ in RC1 mode
Dmitry Selyutin [Sat, 24 Sep 2022 13:22:14 +0000 (16:22 +0300)]
power_insn: slightly change table checking style
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 13:25:09 +0000 (14:25 +0100)]
add extra test_pysvp64dis.py test for ff=~RC1/vli mode
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 13:22:09 +0000 (14:22 +0100)]
whoops got mask/match test wrong in power_insn.py
should be value & mask == search & mask
Luke Kenneth Casson Leighton [Sat, 24 Sep 2022 10:25:04 +0000 (11:25 +0100)]
comment inv,CRbit swap in decode_bo
Dmitry Selyutin [Sat, 24 Sep 2022 09:09:32 +0000 (12:09 +0300)]
sv_binutils: support RS opindex
Dmitry Selyutin [Fri, 23 Sep 2022 07:37:47 +0000 (10:37 +0300)]
power_insn: reorder mode tables to match the spec
Dmitry Selyutin [Fri, 23 Sep 2022 07:36:37 +0000 (10:36 +0300)]
power_insn: rename smr to mr
Dmitry Selyutin [Thu, 22 Sep 2022 21:49:46 +0000 (00:49 +0300)]
sv_binutils: provide Boolean class and Rc field
Dmitry Selyutin [Thu, 22 Sep 2022 21:49:29 +0000 (00:49 +0300)]
power_insn: provide Record.Rc field
Dmitry Selyutin [Thu, 22 Sep 2022 21:30:59 +0000 (00:30 +0300)]
power_insn: simplify rsvd naming; drop unused rsvd
Dmitry Selyutin [Thu, 22 Sep 2022 21:30:32 +0000 (00:30 +0300)]
power_insn: replace Record.function with Record.mode
Dmitry Selyutin [Wed, 21 Sep 2022 19:35:08 +0000 (22:35 +0300)]
pysvp64asm: expand vector register macros
Dmitry Selyutin [Wed, 21 Sep 2022 15:07:01 +0000 (18:07 +0300)]
sv_binutils: support opcodes offset representation
Dmitry Selyutin [Wed, 21 Sep 2022 15:06:32 +0000 (18:06 +0300)]
sv_binutils: fix fields traversal
Dmitry Selyutin [Wed, 21 Sep 2022 15:04:53 +0000 (18:04 +0300)]
power_insn: sort database finally
Dmitry Selyutin [Wed, 21 Sep 2022 10:21:46 +0000 (13:21 +0300)]
power_insn: provide missing cr_in2 properties
Dmitry Selyutin [Wed, 21 Sep 2022 08:39:00 +0000 (11:39 +0300)]
sv_binutils: generate svp64_cr_in2 opindices
Dmitry Selyutin [Wed, 21 Sep 2022 08:37:46 +0000 (11:37 +0300)]
sv_binutils: generate BA opindex
Dmitry Selyutin [Tue, 20 Sep 2022 20:50:43 +0000 (23:50 +0300)]
power_fields: restore class-oriented traversal
Jacob Lifshay [Sat, 24 Sep 2022 00:12:47 +0000 (17:12 -0700)]
pcdec. works!
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 23:00:20 +0000 (00:00 +0100)]
check svstate (vl) in failfirst test
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 22:15:23 +0000 (23:15 +0100)]
grr annoying recurrence of svshape bug, mscale starts with 6 bits
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 21:44:14 +0000 (22:44 +0100)]
add data-dependent fail-first mode, Rc=1 variant not RC1 yet
first unit test passes, not Vertical-First Mode
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 21:42:49 +0000 (22:42 +0100)]
whoops consistent inversion of inv,CRbit was CRbit,inv
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 20:17:52 +0000 (21:17 +0100)]
extra failfirst dis tests
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 19:20:39 +0000 (20:20 +0100)]
remove barse-ackwardsness, use SelectableInt() in decode_bo
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 19:12:05 +0000 (20:12 +0100)]
put back the barse-ackward decode_bo inversion of {inv||CR_bit}
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 17:40:33 +0000 (18:40 +0100)]
remove need for explicit-hack for "pcdec." - rc column in minor_4.csv file
can be set "rc=ONE" which tells ISACaller (and PowerDecoder2) to
*always* write to CR0
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 17:38:34 +0000 (18:38 +0100)]
lots of really bad hacks, here
https://bugs.libre-soc.org/show_bug.cgi?id=933
1) rename to "pcdec." because it always sets CR0. following the convention
set by "stbcx." etc.
2) hacked ISACaller into submission because this is the first instruction
supported with "." at the end which is not Rc=1
3) handle_comparison was bypassed when CR0 is detected as explicitly
an output: there is no point computing Rc=1 EQ/LT/GT/SO when CR0
is supplied by the pseudocode
4) the test case case_pcdec_simple() was not making explicit deepcopy()
of the registers, which causes problems
5) various places in actually getting the instruction from the insn
dictionary, have to special-case "pcdec."
6) sv/trans/svp64.py updated to name "pcdec."
.
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 17:05:48 +0000 (18:05 +0100)]
fix/hack some bugs in prefix_codes_cases
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:49:45 +0000 (17:49 +0100)]
add (sigh) to the hack-job get_pdecode_idx_out2() in ISACaller
really should be relying on PowerDecoder2 but hey
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:43:31 +0000 (17:43 +0100)]
change variablename dec2.use_svp64_fft to implicit_rs
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:43:13 +0000 (17:43 +0100)]
add match on implicit_rc for pcdec
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:35:16 +0000 (17:35 +0100)]
rename all "fft" variables in PowerDecoder2 because they are
to be used for pcdec as well.
https://bugs.libre-soc.org/show_bug.cgi?id=933
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:21:00 +0000 (17:21 +0100)]
add sv.maddld/mr unit test example with expected results
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 16:12:56 +0000 (17:12 +0100)]
add expected results for sv.maddld in openpower/test/mul_cases.py
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 10:21:40 +0000 (11:21 +0100)]
reduce field name lengths (not in use)
Luke Kenneth Casson Leighton [Fri, 23 Sep 2022 09:13:39 +0000 (10:13 +0100)]
whoops offset-tracking on 3-in 2-out supposed to be by MAXVL not VL
Konstantinos Margaritis [Fri, 23 Sep 2022 07:49:56 +0000 (07:49 +0000)]
removed unneeded file
Konstantinos Margaritis [Fri, 23 Sep 2022 07:33:44 +0000 (07:33 +0000)]
update Makefile
Konstantinos Margaritis [Fri, 23 Sep 2022 07:33:37 +0000 (07:33 +0000)]
add new SVP64 function
Konstantinos Margaritis [Fri, 23 Sep 2022 07:33:01 +0000 (07:33 +0000)]
reduce iterations, taking too long in the simulator
Konstantinos Margaritis [Fri, 23 Sep 2022 07:32:42 +0000 (07:32 +0000)]
rename reference functions with _c suffix, add header
Konstantinos Margaritis [Fri, 23 Sep 2022 07:31:58 +0000 (07:31 +0000)]
renamed variance_svp64.c to variancefuncs_svp64.c
Jacob Lifshay [Fri, 23 Sep 2022 03:05:35 +0000 (20:05 -0700)]
add pcdec -- doesn't yet work due to broken ISACaller RT/RS output handling
Jacob Lifshay [Fri, 23 Sep 2022 03:22:45 +0000 (20:22 -0700)]
fix maddld pseudo-code
Jacob Lifshay [Fri, 23 Sep 2022 03:02:02 +0000 (20:02 -0700)]
add missing minor_4 decoder
Jacob Lifshay [Fri, 23 Sep 2022 03:00:52 +0000 (20:00 -0700)]
fix 'write reg ' log call
Jacob Lifshay [Fri, 23 Sep 2022 02:59:40 +0000 (19:59 -0700)]
add RC input to isa/caller.py
Jacob Lifshay [Fri, 23 Sep 2022 02:56:11 +0000 (19:56 -0700)]
format code
Jacob Lifshay [Fri, 23 Sep 2022 01:02:59 +0000 (18:02 -0700)]
maddhd[u]/maddld are official ops
Jacob Lifshay [Fri, 23 Sep 2022 00:37:33 +0000 (17:37 -0700)]
format code
Luke Kenneth Casson Leighton [Thu, 22 Sep 2022 23:47:26 +0000 (00:47 +0100)]
add first (correctly-working) ctr-mode sv.bc test
Luke Kenneth Casson Leighton [Thu, 22 Sep 2022 12:41:23 +0000 (13:41 +0100)]
comment need for waiting on binutils update
Konstantinos Margaritis [Thu, 22 Sep 2022 11:05:34 +0000 (11:05 +0000)]
fix no of iterations in comment, harmless but wrong
Konstantinos Margaritis [Thu, 22 Sep 2022 08:43:46 +0000 (08:43 +0000)]
dump memory
Konstantinos Margaritis [Thu, 22 Sep 2022 08:43:26 +0000 (08:43 +0000)]
better handling of memory copies, fix vpx_get4x4sse_cs_svp64
Konstantinos Margaritis [Thu, 22 Sep 2022 08:42:05 +0000 (08:42 +0000)]
remove extra setvl instruction
Luke Kenneth Casson Leighton [Wed, 21 Sep 2022 19:49:44 +0000 (20:49 +0100)]
add series of double-stride options to test_caller_svp64_dct.py
Luke Kenneth Casson Leighton [Wed, 21 Sep 2022 19:17:41 +0000 (20:17 +0100)]
do not set striding on costables, keep them contiguous.
not totally sure this is a good idea, but hey
Konstantinos Margaritis [Wed, 21 Sep 2022 18:33:24 +0000 (18:33 +0000)]
getting better, get rid of the ctr, group src/ref loads
Luke Kenneth Casson Leighton [Wed, 21 Sep 2022 17:18:14 +0000 (18:18 +0100)]
scale-up svshape pseudo-code for striding in DCT/FFT
Luke Kenneth Casson Leighton [Wed, 21 Sep 2022 16:56:57 +0000 (17:56 +0100)]
fix dct/fft test-functions with new "scaling" parameter
https://bugs.libre-soc.org/show_bug.cgi?id=930
Luke Kenneth Casson Leighton [Wed, 21 Sep 2022 14:53:53 +0000 (15:53 +0100)]
missed setting zdim in svshape on DCT modes
Konstantinos Margaritis [Wed, 21 Sep 2022 15:33:49 +0000 (15:33 +0000)]
use sv.subf