Cheri-research/Toooba
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity
763 Commits 7 Branches 0 Tags
bbdcf6dde9eb8a4970684a0c78983db7bac79418
Commit Graph

521 Commits

Author SHA1 Message Date
Jonathan Woodruff
bbdcf6dde9 A more sophisticated RAS protection delay. 
Saturate down, and also predict with old push value if there is more
than 6 cycles of delay.
 2022-03-08 15:17:54 +00:00
Jonathan Woodruff
064cd22114 Inteligently delay Decode (for up to 8 cycles) if we encounter a pop 
while waiting for a push.
 2022-03-07 17:52:11 +00:00
Jonathan Woodruff
c17874bdba Add options to disable each of 5 aspects of enforcment of the Branching 
Speculation Contract.
 2022-02-22 16:21:53 +00:00
Jonathan Woodruff
e889bb95fa Add a "delay" for fixups after a push. 
A push updates the head pointer from Execute, but the instructions
between Decode and execute will have an out-of-date head pointer.
If a branch in that window mispredicts, it will get the head pointer
out of sync by resetting it to ignore the push.
By delaying by one branch, we seem to eliminate over 60% of the misses,
and reduce the cycle overhead from 0.5% to 0.2% in CoreMark, and still
well ahead of the baseline with no fixup.
 2022-02-18 18:00:02 +00:00
Jonathan Woodruff
75d1afe8b4 Only push to RAS when the linking instruction is non-speculative. 
That is, don't push in Decode, push from the training structure coming
from Execute, which is delayed until the instruction that issued it is
non-speculative.
 2022-02-18 13:22:02 +00:00
Jonathan Woodruff
2b9ebb5b40 Ugly hack to avoid redirection epoch from being killed by the killAll. 
Commit stage redirections are post-prediction and should need to somehow
be treated specially in the SpecFifo for epochs in the fetch stage.
They are post-commit, so should be unkillable somehow.  I guess
SpecFifos are not meant to hold post-commit state currently.
 2022-02-16 12:23:05 +00:00
Jonathan Woodruff
4961d321b9 Merge branch 'CHERI' into jdw57_prediction 2022-02-16 10:10:00 +00:00
Jonathan Woodruff
05df76073b Revert some unintended changes from the last commit. 2022-02-01 11:08:37 +00:00
Jonathan Woodruff
01132d165a Enable tracing of physical addresses for RVFI (if PADDR_RVFI is 
defined).
This is done by adding a vector of padder lookup interfaces to the load
store queue.
This should help with targeted memory translation debugging, and also
with compatability with the current Sail RVFI implemenation which is
probably incorrect with respect to the RVFI spec which calls for virtual
addresses in the maddr field.
 2022-02-01 09:10:48 +00:00
Jonathan Woodruff
ad5d5aca0d Also pause branch execution when there is a pending wrong speculation. 2022-01-31 12:36:17 +00:00
Jonathan Woodruff
a56deab46c Simpelest strategy for making wrongSpec not conflict with loads of 
rules.
This triggers some assertions in the ROB what should be dealt with
properly if this is going to work.
 2022-01-31 12:36:17 +00:00
Jonathan Woodruff
4fbb7b5648 A version that seems to actually run. 
The buffer in GlobalSpecUpdate has to be a SpecFifo.
 2022-01-31 12:09:38 +00:00
Jonathan Woodruff
c1c4a46b1e Simpelest strategy for making wrongSpec not conflict with loads of 
rules.
This triggers some assertions in the ROB what should be dealt with
properly if this is going to work.
 2022-01-31 12:09:38 +00:00
gameboo
3543056f82 Added a CORE_MINI configuration (with @jdw57) 2022-01-28 12:14:34 +00:00
Jonathan Woodruff
1ee961bb83 Simplify timing optimisation by including in the root of the original 
loop.
 2022-01-27 17:13:11 +00:00
Jonathan Woodruff
4cfd652b0d Use mkRegOR in register file. 2022-01-25 20:42:16 +00:00
Jonathan Woodruff
6ea5d34d06 Bump Bluestuff. 2022-01-25 17:05:34 +00:00
Jonathan Woodruff
678c5bcf49 Fix RVFI_DII which had inadvertently been broken. 2022-01-20 17:36:53 +00:00
Jonathan Woodruff
ef6945cc07 More readable version of previous optimisation, with input from 
Alexandre.
 2022-01-19 15:28:45 +00:00
Jonathan Woodruff
e497091e18 A timing optimisation for the direction predictor. 
Add a "nextPc" interface so that lookup can begin in the previous cycle.
Remove the "pc" operands in the vector of lookup interfaces,
but rather derive the PC from lookup from the first PC being looked up
in that cycle. That is, by adding 4*(interface number) to the beginning
lookup PC.  As this might not actually be the PC of the instruction,
pass the index you used with the training info so that you are certain
to train with the same index you looked up.

This has less than 1% overhead in CoreMark.

This is a general improvement for reasonable timing which should be upstreamed.
 2022-01-18 11:05:09 +00:00
Jonathan Woodruff
dac5c99100 Perform a number of decode and prediction operations unconditionally on 
the vector of instructions outside of the decode loop to improve timing.
Specifically, the branch direction prediction is lifted out and depends
only on whether there is agreement between direction prediction and a
branch being predicted by fetch, not on the actual predicted next PC.
This appears to keep the global history clean enough to
perform on-par with the old design in CoreMark.
This is a general timing improvement and should be upstreamed.
 2022-01-14 11:53:40 +00:00
Jonathan Woodruff
1f8181d595 Move to (recently merged) non-pipelined Divide. 2022-01-13 11:14:20 +00:00
Jonathan Woodruff
37f6dbde93 Timing optimisation for Tournament Predictor. 
Start global-history-indexed lookups in the previous cycle.
This is a general timing optimisation that should be pushed upstream at
some point.
 2022-01-12 16:08:33 +00:00
Jonathan Woodruff
1d6633f423 Buffer redirection from Commit to reduce conflicts in the pipeline. 
This is a general improvement, and should be upstreamed to master if
there are no issues discovered.
 2022-01-10 17:18:01 +00:00
Jonathan Woodruff
e600fd7d38 Delay BTB update by a cycle for timing. 
This should not be a primary degredation of performance since the
redirection gets on seperately from this update which will only affect
future predictions.
 2022-01-08 10:16:50 +00:00
Jonathan Woodruff
b84562a027 Improve scheduling for LatencyTimer to improve timing path discovered on 
DE10.
 2022-01-07 09:52:43 +00:00
Jonathan Woodruff
f619f4b0a1 Make the associative BTB compressed. 
That is, only store the bottom 16 bits of the target if the upper bits
of pc and nextPc match.
Have a single "way" dedicated to full targets.
 2021-12-20 17:24:55 +00:00
Jonathan Woodruff
54d1a204f0 Don't decode redirected streams until they are non-speculative. 2021-12-17 16:29:14 +00:00
Jonathan Woodruff
4cc86b1e56 Revert more changes to buffer speculative training, as previous 
buffering of training from execute should be sufficient for the BTB as
well.
 2021-12-16 15:01:46 +00:00
Jonathan Woodruff
3322d1d3d7 Revert "An attempt at wiring up the specUpdate interface for the Btb." 
This reverts commit 022df36199.
 2021-12-16 12:10:50 +00:00
Jonathan Woodruff
022df36199 An attempt at wiring up the specUpdate interface for the Btb. 2021-12-15 18:40:58 +00:00
Jonathan Woodruff
696f546ecb Version using searchableSpecFifo in BTB with actual searching, but 
without killing speculative elements yet. This one is full performance
in CoreMark.
 2021-12-15 18:27:45 +00:00
Jonathan Woodruff
657453a0a0 Two modules (that build) that should be helpful in delaying the commit 
of training data.
 2021-12-15 17:06:46 +00:00
Jonathan Woodruff
af69f2e5db Merge branch 'CHERI' into jdw57_prediction 2021-12-15 16:47:42 +00:00
Jonathan Woodruff
34c6ee1894 Specifically give priority to wrong_spec when they conflict. 
This appears to yield 10% performance improvement as our default setup
appears to choose a wrong priority.
 2021-12-03 12:36:40 +00:00
Jonathan Woodruff
e761f18c26 Code to trace RAS predictions for measuring accuracy in simulation. 2021-11-29 12:19:00 +00:00
Franz Fuchs
37074525e1 removed unnecessary execute permissions for Decode.bsv, Exec.bsv, ProcTypes.bsv, ReorderBuffer.bsv, and AluExePipeline.bsv 2021-11-29 10:38:23 +00:00
Jonathan Woodruff
59836cc9e5 Hopefully optimise this function to have just one call to modifyOffset. 2021-11-26 11:07:12 +00:00
Jonathan Woodruff
ad149db5c0 Hopefully optimise this function to have just one call to modifyOffset. 2021-11-25 17:39:46 +00:00
Jonathan Woodruff
4e0da006e8 Make non-speculative training optional. 
Removing non-speculative training actually improves performance in
CoreMark.  While it's "safer", I guess this isn't required for BSC.
 2021-11-24 17:11:36 +00:00
Jonathan Woodruff
cc0114f645 Swap write priority, as suggested by Franz. 2021-11-16 15:40:01 +00:00
Jonathan Woodruff
7b472d4d3c Reduce area overhead by using an unguarded specfifo that does not have 
to be worst-case-sized, but can drop elements if it is full.
 2021-11-16 13:53:34 +00:00
Jonathan Woodruff
618dea1225 Feed head of RAS through to execute with training info to reset the head 
on a misprediction.
 2021-11-16 11:00:04 +00:00
Jonathan Woodruff
d56045effe Make the associative BTB compressed. 
That is, only store the bottom 16 bits of the target if the upper bits
of pc and nextPc match.
Have a single "way" dedicated to full targets.
 2021-11-15 11:50:57 +00:00
Jonathan Woodruff
54c0ccadf0 Only train from branches that have all previous branches resolved. 2021-11-12 17:55:04 +00:00
Marno
0f86f09b6a Updated copyright statement 2021-11-11 12:15:45 +00:00
Marno
6ae92f786b Checking tag and seal on rVal2 and checking type on rVal1 in a way that does not use undefined behavior. 2021-11-11 11:36:45 +00:00
Marno
4d1dc29f79 Added ccopytype_bypass 2021-11-10 11:53:26 +00:00
Jonathan Woodruff
04a5d82ff0 Go back to standard tournament predictor (which performs the best on 
CoreMark), though cleaning up slightly to use standard functions.
 2021-11-10 09:54:34 +00:00
Jonathan Woodruff
8379d77cc5 Don't use a synthesis barrier for Bht so that we don't have to add it to 
XML.  The direction predictor already has one anyway.
 2021-11-08 13:51:52 +00:00