Luke Kenneth Casson Leighton [Sat, 16 Jan 2021 19:02:25 +0000 (19:02 +0000)]
add PartitionedXOR to PartitionedSignal (xor operator)
Cesar Strauss [Sat, 16 Jan 2021 19:00:28 +0000 (16:00 -0300)]
Check PartitionedSignal.any().
Luke Kenneth Casson Leighton [Sat, 16 Jan 2021 18:51:34 +0000 (18:51 +0000)]
add first cut (untested) of PartitionedXOR
Luke Kenneth Casson Leighton [Sat, 16 Jan 2021 18:45:10 +0000 (18:45 +0000)]
convert EQCombiner to general-purpose, create XORCombiner
Luke Kenneth Casson Leighton [Sat, 16 Jan 2021 18:40:50 +0000 (18:40 +0000)]
add partitioned signal any() operator
Cesar Strauss [Sat, 16 Jan 2021 17:06:41 +0000 (14:06 -0300)]
Allow a variable number of operands in the proof driver
It can now check unary operations, as well as binary.
Check the "all" method by defining an operation that applies it to
its argument.
Cesar Strauss [Mon, 11 Jan 2021 19:28:38 +0000 (16:28 -0300)]
Apparently PartitionedEqGtGe is already conformant
Cesar Strauss [Sun, 10 Jan 2021 22:02:01 +0000 (19:02 -0300)]
Implement checks for all the rest of the comparison operarions
Cesar Strauss [Sun, 10 Jan 2021 21:42:58 +0000 (18:42 -0300)]
Generate and check expected values for all possible partition sizes
Cesar Strauss [Sun, 10 Jan 2021 21:33:10 +0000 (18:33 -0300)]
Generate shifted down input and outputs
Cesar Strauss [Sun, 10 Jan 2021 21:18:17 +0000 (18:18 -0300)]
Use styles in write_gtkw to simplify the trace description
Cesar Strauss [Sun, 10 Jan 2021 20:56:29 +0000 (17:56 -0300)]
Add the Gate Generator to the ComparisonOpDriver
Cesar Strauss [Sun, 10 Jan 2021 20:28:23 +0000 (17:28 -0300)]
Factor-out the code to make equally spaced partition points
Cesar Strauss [Sun, 10 Jan 2021 19:43:57 +0000 (16:43 -0300)]
Start proof for PartitionedSignal equals operator
The idea is to make use of operator.* to abstract away the operation.
Start with binary comparison operators.
Cesar Strauss [Sun, 10 Jan 2021 18:32:48 +0000 (15:32 -0300)]
Update Simulator interface to current nMigen
Cesar Strauss [Sun, 10 Jan 2021 17:47:26 +0000 (14:47 -0300)]
Use len(sig) instead of sig.shape()[0]
In latest nMigen, Shape is no longer a tuple. On the other hand,
len(sig) worked since the last stable nMigen release, so it should not
break for anyone.
Cesar Strauss [Sat, 9 Jan 2021 17:29:53 +0000 (14:29 -0300)]
Use the correct class name
Cesar Strauss [Sat, 9 Jan 2021 17:23:03 +0000 (14:23 -0300)]
Check all possible opcodes for PartitionedEqGtGe
Cesar Strauss [Sat, 9 Jan 2021 16:55:25 +0000 (13:55 -0300)]
Formal verification of PartitionedEqGtGe (EQ only)
Fix operation at EQ to begin with.
Code is mostly identical to the PartitionedEq proof.
Cesar Strauss [Sat, 9 Jan 2021 14:41:18 +0000 (11:41 -0300)]
Post-process PartitionedEq to ripple the LSB
This needs to be removed once PartitionedEq is compliant.
Cesar Strauss [Sat, 9 Jan 2021 13:56:28 +0000 (10:56 -0300)]
Implement formal verification of PartitionedEq based on GateGenerator
Cesar Strauss [Fri, 8 Jan 2021 21:35:10 +0000 (18:35 -0300)]
Calculate the expected value, test and assert
Cesar Strauss [Fri, 8 Jan 2021 21:17:02 +0000 (18:17 -0300)]
Split-out the gate generator from the proof
This reduces repetition among partition proofs.
Cesar Strauss [Thu, 7 Jan 2021 22:34:20 +0000 (19:34 -0300)]
Compare the expected output for all partition sizes
Optimize for the case where the partition points are equally spaced,
so the possible partition sizes are just multiple of the smaller one.
Otherwise, all combinations of start and end points would have to be
considered.
Cesar Strauss [Wed, 6 Jan 2021 09:41:24 +0000 (06:41 -0300)]
Generate shifted down outputs
Align the start of any selected partition to zero, so it can easily be
checked with the expected value.
For partitioned operations with inputs, it is also necessary to generate
shifted down versions of each input.
Cesar Strauss [Tue, 5 Jan 2021 20:32:40 +0000 (17:32 -0300)]
Generate the bit pattern of gates corresponding to a partition
Use Assume to constraint the pattern to conform to the given offset
and width. For each bit it is:
1) one if on the partition boundary
2) zero if it's inside the partition
3) don't care, otherwise
Cesar Strauss [Mon, 4 Jan 2021 21:38:51 +0000 (18:38 -0300)]
Work only with nibbles on the ascending cascade
Cesar Strauss [Mon, 4 Jan 2021 21:08:37 +0000 (18:08 -0300)]
Fill the second nibble of the pattern
It should encode the length of the partition.
The first nibble encodes the position within the partition.
Cesar Strauss [Mon, 4 Jan 2021 20:26:35 +0000 (17:26 -0300)]
Restart the pattern at partition boundaries
Cesar Strauss [Mon, 4 Jan 2021 19:56:25 +0000 (16:56 -0300)]
Start work on improving formal verification of PartitionedSignal
The goal is to have a generic approach, decoupled from the actual
implementation of each partitioned operation.
Luke Kenneth Casson Leighton [Mon, 28 Dec 2020 13:34:49 +0000 (13:34 +0000)]
partsig: redirect bool to any for now, and use a == Const(-1) for any
Luke Kenneth Casson Leighton [Sun, 27 Dec 2020 00:09:57 +0000 (00:09 +0000)]
comment PartitionedSignal lt/le
Jacob Lifshay [Mon, 20 Jul 2020 03:19:22 +0000 (20:19 -0700)]
disable faulty bit_width reduction logic in DivPipeCore
Jacob Lifshay [Mon, 20 Jul 2020 03:15:25 +0000 (20:15 -0700)]
fix mismatched comb process delays
Jacob Lifshay [Mon, 20 Jul 2020 03:15:01 +0000 (20:15 -0700)]
add process tracing
Luke Kenneth Casson Leighton [Wed, 15 Jul 2020 13:42:39 +0000 (14:42 +0100)]
remove FPPipeContext (moved to nmutil), name-substitute on import
see https://bugs.libre-soc.org/show_bug.cgi?id=431
Jacob Lifshay [Mon, 13 Jul 2020 03:49:59 +0000 (20:49 -0700)]
working on fixing DivPipeCore's test cases
this is as part of fixing the problems caused by Luke reducing the bit-width used without a corresponding change in the fract-width when DivPipeCore only supports UDivMod
Jacob Lifshay [Mon, 13 Jul 2020 02:45:11 +0000 (19:45 -0700)]
clean up DivPipeCoreConfig API
Luke Kenneth Casson Leighton [Fri, 10 Jul 2020 12:02:22 +0000 (13:02 +0100)]
whoops missed set up of temp variable bw
Luke Kenneth Casson Leighton [Fri, 10 Jul 2020 10:34:19 +0000 (11:34 +0100)]
only pass in lhs bit_width * 2 for UDivRem
Luke Kenneth Casson Leighton [Mon, 6 Jul 2020 14:07:36 +0000 (15:07 +0100)]
add arguments to MulPipe_8_16_32_64
Luke Kenneth Casson Leighton [Mon, 6 Jul 2020 14:07:13 +0000 (15:07 +0100)]
fix test_mul_pipe.py unit test
Luke Kenneth Casson Leighton [Sat, 4 Jul 2020 10:05:48 +0000 (11:05 +0100)]
continue reducing length of signals in div core
Luke Kenneth Casson Leighton [Sat, 4 Jul 2020 10:00:30 +0000 (11:00 +0100)]
attempt to get simulation for div test_core.py running
Luke Kenneth Casson Leighton [Sat, 4 Jul 2020 10:00:03 +0000 (11:00 +0100)]
reduce compare lengths to *2 rather than *3
Luke Kenneth Casson Leighton [Sat, 4 Jul 2020 09:43:52 +0000 (10:43 +0100)]
whoops set pass_flag[0] always true
rather than pass_flag[-1] always false
Luke Kenneth Casson Leighton [Sat, 4 Jul 2020 09:25:39 +0000 (10:25 +0100)]
add feedback_width argument to runfp for testing
Luke Kenneth Casson Leighton [Fri, 3 Jul 2020 14:42:51 +0000 (15:42 +0100)]
allow arrangement for feedback loops to be possible on ReservationStations in unit test
Luke Kenneth Casson Leighton [Fri, 3 Jul 2020 04:26:17 +0000 (05:26 +0100)]
cut top trial comparison
Luke Kenneth Casson Leighton [Fri, 3 Jul 2020 03:45:04 +0000 (04:45 +0100)]
remove use of Array, replace with treereduce
Luke Kenneth Casson Leighton [Fri, 3 Jul 2020 03:32:30 +0000 (04:32 +0100)]
cut root_times_radicand if not doing Sqrt
Luke Kenneth Casson Leighton [Fri, 3 Jul 2020 03:09:36 +0000 (04:09 +0100)]
add "supported" option to div core
Luke Kenneth Casson Leighton [Tue, 9 Jun 2020 17:44:13 +0000 (18:44 +0100)]
fix imports (allows test command to be run from non-top-level directory)
Jacob Lifshay [Thu, 4 Jun 2020 00:26:04 +0000 (17:26 -0700)]
move mulAddRecFN.py and nmigen_div_experiment.py to unused dir
Jacob Lifshay [Fri, 22 May 2020 22:32:06 +0000 (15:32 -0700)]
add DivPipeOp in fpdiv/op.py
Michael Nolan [Sat, 16 May 2020 15:16:29 +0000 (11:16 -0400)]
Fix handling of FPPipeContext.ports()
Michael Nolan [Fri, 8 May 2020 17:01:38 +0000 (13:01 -0400)]
Allow partsig to take in a PartitionPoints directly
Michael Nolan [Tue, 5 May 2020 18:08:58 +0000 (14:08 -0400)]
Add in FPPipeContext to the cordic pipeline
Michael Nolan [Tue, 5 May 2020 15:16:21 +0000 (11:16 -0400)]
Use clz.py from nmutil
Luke Kenneth Casson Leighton [Tue, 5 May 2020 14:59:23 +0000 (15:59 +0100)]
track down error in CORDIC pipe_data, "yield from" used instead of just "yield"
Luke Kenneth Casson Leighton [Tue, 5 May 2020 14:52:35 +0000 (15:52 +0100)]
whoops cant output .il at the moment
Luke Kenneth Casson Leighton [Tue, 5 May 2020 14:51:52 +0000 (15:51 +0100)]
add progress counter
Luke Kenneth Casson Leighton [Tue, 5 May 2020 14:48:14 +0000 (15:48 +0100)]
allow rtlil file to be created in cordic
Michael Nolan [Tue, 5 May 2020 14:21:02 +0000 (10:21 -0400)]
Add renormalize.py (oops!)
Luke Kenneth Casson Leighton [Tue, 5 May 2020 14:01:30 +0000 (15:01 +0100)]
remove extraneous whitespace
Michael Nolan [Mon, 4 May 2020 22:00:09 +0000 (18:00 -0400)]
Remove print statements from clz.py
Michael Nolan [Mon, 4 May 2020 21:59:55 +0000 (17:59 -0400)]
Add asserts to fp pipe test
Michael Nolan [Mon, 4 May 2020 20:52:41 +0000 (16:52 -0400)]
Sorta working FP renormalization in cordic
Michael Nolan [Mon, 4 May 2020 20:01:20 +0000 (16:01 -0400)]
Extend clz to work with odd widths
Michael Nolan [Mon, 4 May 2020 19:56:24 +0000 (15:56 -0400)]
Extend CLZ to work over even, non powers of 2
Michael Nolan [Mon, 4 May 2020 19:34:03 +0000 (15:34 -0400)]
Add proof for clz.py
Michael Nolan [Mon, 4 May 2020 18:54:26 +0000 (14:54 -0400)]
Add count leading zeros module (should probably go somewhere else)
Michael Nolan [Mon, 4 May 2020 17:40:10 +0000 (13:40 -0400)]
Have sin_cos pipeline use bigfloat calculated atan table
Michael Nolan [Tue, 28 Apr 2020 17:55:53 +0000 (13:55 -0400)]
Add cordic stages to fp cordic pipeline
Michael Nolan [Tue, 28 Apr 2020 17:44:53 +0000 (13:44 -0400)]
Add stage to convert input float to fixed point number
Michael Nolan [Tue, 28 Apr 2020 17:21:20 +0000 (13:21 -0400)]
Begin adding floating point cordic pipeline
Michael Nolan [Fri, 17 Apr 2020 15:21:57 +0000 (11:21 -0400)]
Allow cordic to work with 64 bit floats
Luke Kenneth Casson Leighton [Fri, 1 May 2020 13:29:20 +0000 (14:29 +0100)]
move ripple.py to nmutil
Jacob Lifshay [Fri, 17 Apr 2020 21:32:50 +0000 (14:32 -0700)]
add libmpfr-dev to .gitlab-ci.yml
Michael Nolan [Fri, 17 Apr 2020 20:29:09 +0000 (16:29 -0400)]
Change tabs to spaces in .gitlab-ci.yml
Michael Nolan [Fri, 17 Apr 2020 18:40:10 +0000 (14:40 -0400)]
Add libgmp-dev to .gitlab-ci.yml
Michael Nolan [Fri, 17 Apr 2020 17:46:05 +0000 (13:46 -0400)]
Use python bigfloat to calculate atan tables
Michael Nolan [Fri, 17 Apr 2020 15:22:24 +0000 (11:22 -0400)]
Revert "Use higher precision arithmetic when calculating the atan2 table"
This reverts commit
7aea44d471226db693ee99fd272504248d57375b.
Michael Nolan [Fri, 17 Apr 2020 14:44:34 +0000 (10:44 -0400)]
Use higher precision arithmetic when calculating the atan2 table
Michael Nolan [Thu, 16 Apr 2020 21:02:09 +0000 (17:02 -0400)]
change test_fpsin to use 32 bit floats
Michael Nolan [Thu, 16 Apr 2020 19:18:07 +0000 (15:18 -0400)]
Run the cordic for the proper number of iterations
Michael Nolan [Thu, 16 Apr 2020 18:57:56 +0000 (14:57 -0400)]
Working (ish) fpsin iterative cordic
Michael Nolan [Thu, 16 Apr 2020 17:14:43 +0000 (13:14 -0400)]
Assert that fpsin_cos converts floats to fixed correctly
Michael Nolan [Thu, 16 Apr 2020 15:17:36 +0000 (11:17 -0400)]
Begin working on multi cycle float cordic
Michael Nolan [Tue, 14 Apr 2020 13:46:23 +0000 (09:46 -0400)]
Add experiment testing the effects of adding extra bits to cordic
Luke Kenneth Casson Leighton [Mon, 13 Apr 2020 17:22:56 +0000 (18:22 +0100)]
add rtlil conversion and fix yield from in Cordic Data
Jacob Lifshay [Fri, 3 Apr 2020 07:07:25 +0000 (00:07 -0700)]
run tests in parallel
Jacob Lifshay [Fri, 3 Apr 2020 05:52:16 +0000 (22:52 -0700)]
building sfpy works
Jacob Lifshay [Fri, 3 Apr 2020 05:27:49 +0000 (22:27 -0700)]
add not yet working .gitlab-ci.yml
Michael Nolan [Thu, 2 Apr 2020 17:01:46 +0000 (13:01 -0400)]
Combine a selectable number of rounds into one pipeline stage
Michael Nolan [Wed, 1 Apr 2020 17:29:01 +0000 (13:29 -0400)]
Cleanup
Michael Nolan [Wed, 1 Apr 2020 17:26:09 +0000 (13:26 -0400)]
Combine initial stage with first cordic stage
Michael Nolan [Wed, 1 Apr 2020 17:17:40 +0000 (13:17 -0400)]
Add actual tests to test_pipeline.py
Michael Nolan [Wed, 1 Apr 2020 17:17:27 +0000 (13:17 -0400)]
Register each pipeline stage
Michael Nolan [Wed, 1 Apr 2020 16:08:45 +0000 (12:08 -0400)]
Add test for sin_cos_pipe (kinda working)