Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 21:04:45 +0000 (22:04 +0100)]
comments
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 21:03:44 +0000 (22:03 +0100)]
create cos table independent, outside of the inner loops
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 20:58:35 +0000 (21:58 +0100)]
cleanup
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 20:49:23 +0000 (21:49 +0100)]
add iterative list-reversing algorithm, replace recursive variant.
actually really simple (to implement in hardware)
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 16:04:22 +0000 (17:04 +0100)]
pre-reverse order of data indices in DCT so that *after* the inner
butterfly is done the data is in the correct order for the outer one.
this so that no data-swaps are needed
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 14:56:43 +0000 (15:56 +0100)]
temporary reordering after the DCT schedule is carried out, this removes
the need for *data* swaps.
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 13:08:06 +0000 (14:08 +0100)]
realised that SVSHAPE0-3 is not privileged
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 13:06:04 +0000 (14:06 +0100)]
add inner sub-loop testing from svstep Rc=1
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 09:01:08 +0000 (10:01 +0100)]
comments
Luke Kenneth Casson Leighton [Tue, 20 Jul 2021 08:58:09 +0000 (09:58 +0100)]
comments
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 19:45:25 +0000 (20:45 +0100)]
bit of a reorg, adding option to test end of inner loops of SVSTATE(s)
needed to pass the immediate to svstep as an option of which
SVSTATE0-3 to test
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 15:24:48 +0000 (16:24 +0100)]
do in-place swap
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:59:03 +0000 (15:59 +0100)]
annoying: missed out something in the unit test, not working yet
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:53:17 +0000 (15:53 +0100)]
simplify DCT code
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:50:23 +0000 (15:50 +0100)]
create coefficient table for DCT outside of loops
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:48:17 +0000 (15:48 +0100)]
update comments
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:13:25 +0000 (15:13 +0100)]
more comments
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 14:05:26 +0000 (15:05 +0100)]
update comments and license
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 13:47:59 +0000 (14:47 +0100)]
remove unneeded code
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 13:46:25 +0000 (14:46 +0100)]
no need for len(j) > 1 test, half of 1 is zero which stops swap anyway
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 13:44:20 +0000 (14:44 +0100)]
remove unneeded code
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 13:41:04 +0000 (14:41 +0100)]
swap the indices rather than the data in DCT top half: bizarrely this works!
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 12:25:51 +0000 (13:25 +0100)]
remove copy, use in-place with post-inner-loop swap
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 11:56:51 +0000 (12:56 +0100)]
add experimental order-reversing code (commented out) to DCT
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 11:08:38 +0000 (12:08 +0100)]
code comments
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 11:08:00 +0000 (12:08 +0100)]
whitespace cleanup
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 10:57:37 +0000 (11:57 +0100)]
move bit-reversing to before MULs in DCT
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 10:53:33 +0000 (11:53 +0100)]
reverse bit-order of in-place outer DCT butterfly
Luke Kenneth Casson Leighton [Mon, 19 Jul 2021 10:48:45 +0000 (11:48 +0100)]
finallygot the DCT outer butterfly correct
Luke Kenneth Casson Leighton [Sun, 18 Jul 2021 20:29:36 +0000 (21:29 +0100)]
got cos intermediate working on iterative dct
Luke Kenneth Casson Leighton [Sun, 18 Jul 2021 16:24:28 +0000 (17:24 +0100)]
experimenting to create iterative version of dct
Luke Kenneth Casson Leighton [Sun, 18 Jul 2021 10:09:29 +0000 (11:09 +0100)]
use lists rather than list incomprehension
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 17:03:30 +0000 (18:03 +0100)]
print out some debug statements in fastdctlee
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 16:07:29 +0000 (17:07 +0100)]
whitespace
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 16:06:38 +0000 (17:06 +0100)]
add naive dct, remove fft variant
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 15:29:11 +0000 (16:29 +0100)]
add nayuki dct
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 13:59:55 +0000 (14:59 +0100)]
working RADIX-2 FFT with bit-reversed LD/ST
Luke Kenneth Casson Leighton [Sat, 17 Jul 2021 12:07:40 +0000 (13:07 +0100)]
add FP LOAD bit-reversed operations to ISACaller simulator
Luke Kenneth Casson Leighton [Fri, 16 Jul 2021 19:03:17 +0000 (20:03 +0100)]
make a test for if size of FFT is less than 8, if so call a separate function
idea is to replace that function with assembler
Luke Kenneth Casson Leighton [Fri, 16 Jul 2021 17:15:31 +0000 (18:15 +0100)]
add fsins and fcoss to simulator
WARNING: THESE ARE **NOT** APPROVED BY OPF ISA WG, CONSIDER TO BE DRAFT
in particular, they have had to be added to minor opcode 59 (alongside
fcfids) because there is not enough space to add them to "sandbox" 22
Luke Kenneth Casson Leighton [Fri, 16 Jul 2021 16:23:26 +0000 (17:23 +0100)]
code clean-up, simplify, use float not double
Luke Kenneth Casson Leighton [Fri, 16 Jul 2021 16:14:06 +0000 (17:14 +0100)]
add fft makefile
Luke Kenneth Casson Leighton [Fri, 16 Jul 2021 15:59:41 +0000 (16:59 +0100)]
add nayuki project reference code
https://www.nayuki.io/page/free-small-fft-in-multiple-languages
Luke Kenneth Casson Leighton [Thu, 15 Jul 2021 18:07:38 +0000 (19:07 +0100)]
use coincidence of svremap "persistence" to remove one more instruction
from FFT example
Luke Kenneth Casson Leighton [Thu, 15 Jul 2021 17:58:34 +0000 (18:58 +0100)]
enable use of svremap "persist" mode, remove 4 instructions from FFT example
Luke Kenneth Casson Leighton [Thu, 15 Jul 2021 17:49:45 +0000 (18:49 +0100)]
stop using MSR vfirst bit, move to SVSTATE bit 63 instead
Luke Kenneth Casson Leighton [Thu, 15 Jul 2021 17:44:13 +0000 (18:44 +0100)]
add extra "persistence" bit to svremap instruction
Luke Kenneth Casson Leighton [Thu, 15 Jul 2021 12:51:08 +0000 (13:51 +0100)]
big intrusive update: merge SVREMAP with SVSTATE, remove SVREMAP
this also involved creating an SVP64State class similar to the REMAPP
class (now removed).
unit tests had to be altered to the new API
Luke Kenneth Casson Leighton [Wed, 14 Jul 2021 20:37:38 +0000 (21:37 +0100)]
use fmadds and fmsubs in complex fft example
reduces inner loop instruction count by 2
Luke Kenneth Casson Leighton [Wed, 14 Jul 2021 19:01:24 +0000 (20:01 +0100)]
update SVSTATE to 64 bit length
Luke Kenneth Casson Leighton [Wed, 14 Jul 2021 18:29:49 +0000 (19:29 +0100)]
subtract one from SVi field for setvl assembler
update all uses to match
Luke Kenneth Casson Leighton [Tue, 13 Jul 2021 16:24:30 +0000 (17:24 +0100)]
change order of log printout for "writing gpr NN"
Luke Kenneth Casson Leighton [Mon, 12 Jul 2021 20:02:55 +0000 (21:02 +0100)]
successful complex FFT butterfly, in-place, using Vertical-First SVP64
Luke Kenneth Casson Leighton [Mon, 12 Jul 2021 14:20:02 +0000 (15:20 +0100)]
add a Discrete FFT butterfly unit test as an intermediary (incremental)
step towards full complex butterfly.
this test uses fp24 as an intermediary for storing the result of the
multiply, source from butterfly-scheduled operands.
that scalar reg is then used as a source in a butterfly-scheduled
twin +/- in-place ADD-SUB
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 23:02:00 +0000 (00:02 +0100)]
minor reordering of setvl and svshape: svshape is now capable of setting VL
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 19:17:51 +0000 (20:17 +0100)]
add svremap instruction into ISACaller
alter FFT and Matrix-Multiply SVP64 tests to use new svremap
generic redirect of any register to a SVSHAPE(0-3) rather than hard-coded
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 17:49:42 +0000 (18:49 +0100)]
update svremap instruction to correctly store immediate args in SPR
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 16:11:27 +0000 (17:11 +0100)]
whoops
0b00002 is not binary
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 16:09:44 +0000 (17:09 +0100)]
add SVREMAP new Form / Fields and CSV entry
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 15:08:00 +0000 (16:08 +0100)]
add SVREMAP SPR to ISACaller and parser
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 15:02:15 +0000 (16:02 +0100)]
add SVREMAP SPR
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 11:35:08 +0000 (12:35 +0100)]
add SVP64REMAP Record
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 11:34:50 +0000 (12:34 +0100)]
rename SVP64REMAP to SVP64SHAPE
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 11:22:03 +0000 (12:22 +0100)]
rename svremap to svshape
Luke Kenneth Casson Leighton [Sun, 11 Jul 2021 11:17:50 +0000 (12:17 +0100)]
rename svremap to svshape
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 16:25:25 +0000 (17:25 +0100)]
add scalar ffadds unit test
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 16:24:10 +0000 (17:24 +0100)]
whoops forgot elif in SVP64Asm translation, detection of ffmadds overwritten
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 16:08:24 +0000 (17:08 +0100)]
in scalar case do not increment RB for FFT mode
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 11:37:20 +0000 (12:37 +0100)]
add sv.ffadds unit test, inversion of subtract needed in svfparith pseudocode
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 11:36:43 +0000 (12:36 +0100)]
more generic allow fft mode 2nd output detection. REALLY need a CSV Out2 column
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 11:11:19 +0000 (12:11 +0100)]
add ffadds decoding:
- SVP64 trans manual creation of opcode with XO=
0b01101
- add to power enums ISA list
- add to minor 59 and SVP64 CSV
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 11:10:08 +0000 (12:10 +0100)]
add more generic detection of FFT mode, really needs to be a new column in CSV file for Out2
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 11:09:10 +0000 (12:09 +0100)]
add (disabled) FFT complex unit test under development
Luke Kenneth Casson Leighton [Sat, 10 Jul 2021 10:46:49 +0000 (11:46 +0100)]
corrections to remaining fft madd/msub
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 18:42:16 +0000 (19:42 +0100)]
update comments
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 16:12:37 +0000 (17:12 +0100)]
for scalar destination or scalar source on ffmadd, only offset by one
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 16:03:31 +0000 (17:03 +0100)]
add svstep variant of fpmadds fft test
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 15:52:37 +0000 (16:52 +0100)]
add "odd" SVP64 unit tests which alter SVSTATE
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 11:48:57 +0000 (12:48 +0100)]
comments in unit test
Luke Kenneth Casson Leighton [Fri, 9 Jul 2021 11:44:22 +0000 (12:44 +0100)]
add Vertical-First explicit branch-loop using svstep, with an sv.add. works!
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 22:09:49 +0000 (23:09 +0100)]
end SVP64 "Vertical First" mode on rollover when end of svstep reached
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 22:04:51 +0000 (23:04 +0100)]
add CR0 setting and unit test on svstep
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 21:34:29 +0000 (22:34 +0100)]
whoops asmcode length (number of instructions) went over 256, caused
asmcode in simulator to "wrap" and get the wrong instruction name,
and then execute totally the wrong simulated instruction
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 21:26:30 +0000 (22:26 +0100)]
test MSR.SVF bit set after setvl Vertical-First mode set
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 21:10:03 +0000 (22:10 +0100)]
whoops must not reset last_op_svstate except when out of SVP64 mode
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 21:03:55 +0000 (22:03 +0100)]
testing new setvl "svstep" mode
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 20:46:56 +0000 (21:46 +0100)]
whoops sv.lfs registers must be even numbers (match SVP6 EXTRA2 encoding)
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 18:02:37 +0000 (19:02 +0100)]
add ability to explicitly increment SVSTATE srcstep/dststep
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 15:21:49 +0000 (16:21 +0100)]
add (new) extra argument to setvl (Vertical-First mode, set to zero here)
Luke Kenneth Casson Leighton [Thu, 8 Jul 2021 15:12:14 +0000 (16:12 +0100)]
add in extra "vertical" mode into SVP64 setvl
Luke Kenneth Casson Leighton [Wed, 7 Jul 2021 15:43:40 +0000 (16:43 +0100)]
clean up imports and unit test name
Luke Kenneth Casson Leighton [Wed, 7 Jul 2021 15:31:08 +0000 (16:31 +0100)]
ffmuls test, had to add to b not a in expected results
Luke Kenneth Casson Leighton [Wed, 7 Jul 2021 15:27:46 +0000 (16:27 +0100)]
get butterfly RADIX2 SVP64 example working, breaks the fpmadds one though
Luke Kenneth Casson Leighton [Wed, 7 Jul 2021 13:07:27 +0000 (14:07 +0100)]
add in DFT test from Project Nayuki to verify results of butterfly schedule
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 20:15:02 +0000 (21:15 +0100)]
change coefficients in FFT REMAP example so as not to get alternating zeros
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 20:14:33 +0000 (21:14 +0100)]
when REMAP shape is zero, skip it in ISACaller.
also use srcstep only for FFT butterfly mode (for now)
bit of a hack
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 20:13:50 +0000 (21:13 +0100)]
on FRS for FFT with REMAP use scheduled offset
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 19:34:40 +0000 (20:34 +0100)]
add FFT REMAP butterfly unit test
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 18:55:41 +0000 (19:55 +0100)]
add FFT SHAPE pseudocode in svremap, and a schedule in ISACaller
Luke Kenneth Casson Leighton [Tue, 6 Jul 2021 16:03:53 +0000 (17:03 +0100)]
add FFT butterfly iteration to SVSHAPE REMAP class for ISACaller