1
0
Commit Graph

3468 Commits

Author SHA1 Message Date
Andrew Waterman
5bc78aba99 Merge pull request #15 from terpstra/ahb
Ahb
2016-05-24 17:06:03 -07:00
Andrew Waterman
c49cb10c74 Merge pull request #42 from terpstra/ahb
Ahb
2016-05-24 17:02:15 -07:00
Andrew Waterman
4605b616c1 Fix bug in D$ AMO/storegen logic 2016-05-24 16:26:07 -07:00
Andrew Waterman
88cc91db75 Ignore way_en in MetadataArray for direct-mapped caches 2016-05-24 15:47:09 -07:00
Andrew Waterman
5dac7b818d Support set associativity in blocking D$ 2016-05-24 15:45:52 -07:00
Andrew Waterman
e0addb5723 Support uncached AMOs in blocking D$ 2016-05-24 15:45:35 -07:00
Andrew Waterman
f14d87e327 Support larger I$ sets when VM is disabled 2016-05-24 15:44:59 -07:00
Andrew Waterman
3b35c7470e Add uncached support to blocking D$ 2016-05-24 15:05:41 -07:00
Andrew Waterman
42f079ce57 JAL requires DW_XPR
This has been benign so far because of how the logic minimization worked.
2016-05-24 15:05:41 -07:00
Andrew Waterman
b92c73e361 Add LR/SC to blocking D$ 2016-05-24 15:05:41 -07:00
Andrew Waterman
0d93d1a1a0 Clean up pending store logic a bit 2016-05-24 15:05:41 -07:00
Andrew Waterman
0b8de578d4 Add additional D$ store buffering to prevent structural hazards 2016-05-24 15:05:41 -07:00
Andrew Waterman
354cb2d5ec Don't stall I$ response when resolving a branch misprediction
This avoids a fetch bubble.

Not clear if this is the best way to do it.  Perhaps this change should
instead be made to Frontend (i.e., ignore resp.ready when req.valid is
high), but that might exacerbate a critical path.
2016-05-24 15:05:41 -07:00
Andrew Waterman
d7790ac6a4 WIP on blocking D$ 2016-05-24 15:05:41 -07:00
Andrew Waterman
335e2c8a1e Support disabling atomics extension 2016-05-24 15:05:41 -07:00
Andrew Waterman
765b90f6a4 Stall on D$ lockups less conservatively 2016-05-24 15:05:41 -07:00
Andrew Waterman
a3061047e3 Instantiate blocking D$ when NMSHRS=0 2016-05-24 15:05:41 -07:00
Andrew Waterman
80482890fd Don't rely on tag value for nacks 2016-05-24 15:05:41 -07:00
Wesley W. Terpstra
e19c5e5d2c IOMSHR: support atomic operations 2016-05-24 15:00:50 -07:00
Wesley W. Terpstra
a012341d96 ahb: TileLink => AHB bridge, including atomics and bursts 2016-05-24 14:58:27 -07:00
Wesley W. Terpstra
ace9362d81 ahb: amoalu does not need so many parameters! (i want to reuse it) 2016-05-24 14:58:27 -07:00
Wesley W. Terpstra
b921bae107 ahb: eliminate trait abused for constants 2016-05-24 14:20:45 -07:00
Wesley W. Terpstra
200c69c106 ahb: support hmastlock acquistion of crossbar 2016-05-24 14:20:45 -07:00
Wesley W. Terpstra
e1e8eda419 ahb: add a test SRAM 2016-05-24 14:20:42 -07:00
Wesley W. Terpstra
1db40687c6 ahb: eliminate now-unnecesary non-standard hreadyin 2016-05-24 14:14:22 -07:00
Wesley W. Terpstra
15cad8414d ahb: put signals in the order they appear in signal traces in the spec 2016-05-24 14:14:22 -07:00
Wesley W. Terpstra
f30f8d9f79 ahb: reduce obsolete degenerate cases of a crossbar 2016-05-24 14:14:22 -07:00
Wesley W. Terpstra
0368b6db6b ahb: replace defective crossbar with a functional one
The previous crossbar had the following bugs:
  1. a bursting master could be preempted
     the AHB-lite spec requires a slave receive the entire burst

  2. a waited master could be replaced
     the AHB-lite spec requires haddr/etc to remain unchanged

  3. hmastlock did no ensure exclusive access
     atomic operations could be pre-empted
2016-05-24 14:14:22 -07:00
Wesley W. Terpstra
2b37f37335 ahb: helper methods 2016-05-24 14:14:21 -07:00
Wesley W. Terpstra
ef2aae26a8 ahb: rename hreadyout to standard hready, mark hreadyin for death 2016-05-24 14:14:21 -07:00
Wesley W. Terpstra
2f8a77f27a ahb: include all AHB-lite constants 2016-05-24 14:14:21 -07:00
Wesley W. Terpstra
7896c4157e ahb: parameterize poci 2016-05-24 14:14:21 -07:00
Wesley W. Terpstra
93447eb274 ahb: make hasti parameters location sensitive 2016-05-24 14:14:17 -07:00
Wesley W. Terpstra
00d31dc5c5 bram: use new hasti definitions 2016-05-24 13:35:16 -07:00
Albert Ou
ee0acc1d07 Fix BRAM assertion condition 2016-05-23 13:19:53 -07:00
Matthew Naylor
05c0808ff2 Update trace generation and checking scripts
Pass the elf file (that specifies the tohost and fromhost addresses)
to the emulator in the trace generator & checker scripts.
2016-05-23 17:02:15 +01:00
Andrew Waterman
7bc38383de add (non-working) blocking data cache 2016-05-20 18:59:05 -07:00
Colin Schmidt
3e0b5d6fd9 Ensure that a TSHR doesn't see a valid Acquire if that is blocked by a Release,
but would otherwise be allocated.

Closes #45
2016-05-20 16:35:30 -07:00
Ken McMillan
fd83d20857 Use a def instead of a lazy val in ManagerCoherenceAgent.
Prevents C++ emulator from randomizing inputs in unit testing.

Closes #44
2016-05-20 16:31:12 -07:00
Howard Mao
f228309bd1 add assertion to make sure SimpleHellaCacheIF doesn't get exception 2016-05-20 16:30:27 -07:00
Ken McMillan
d69446e177 Add config classes to drive unit testing of L2 TileLink agents.
Closes #43
2016-05-20 16:15:43 -07:00
Howard Mao
87be2bcd60 make sure TraceGen addresses are correct 2016-05-20 16:12:11 -07:00
Howard Mao
f52fc655a5 remove zscale 2016-05-19 09:43:15 -07:00
Howard Mao
4f84d8f757 make sure to hook up finish in ClientTileLinkEnqueuer 2016-05-18 13:13:34 -07:00
Colin Schmidt
abb0e2921b return non-zero exit codes when an assertion fires
This ensures that assertion failures, which currently print a message to
the console but return a successful exit code, now will cause non-zero
exit code. This is meant to help automated tools like travis and
buildbot do a better job at catching assertions.
This impacts the various run-* targets in the simulation
directories.
2016-05-18 12:57:58 -07:00
Colin Schmidt
b396c68577 bump torture for priv-1.9 test-env 2016-05-17 22:02:38 -07:00
Colin Schmidt
b0ae003981 add firrtl to regression makefile
this makes it easier to test chisel3 builds
e.g. the local buildbot can now test a bunch of configs
2016-05-17 17:34:20 -07:00
Andrew Waterman
4aef567a80 Fix MMIO bug: replay_next wasn't set 2016-05-13 17:59:53 -07:00
Andrew Waterman
742c05d6a7 Pipeline D$->I$ control paths
These stretch the miss latency by a cycle in exchange for slack.
The current implementation also adds a cycle to mul/div latency,
which can be worked around for more hardware (possibly gated by
the FastMulDiv option).
2016-05-13 17:07:28 -07:00
Andrew Waterman
684d902059 Fix PLIC instantiation when S-mode is disabled 2016-05-13 11:22:46 -07:00