Repair setup4 hier case 1 buffer #6048

andyfox-rushc · 2024-10-29T13:29:03Z

Changes to resizer to support hierarchy for rebuffer and split load operations only

New test cases: split_load_hier.tcl (shows split load across hierarchy with factoring of nets to minimize punch through), resize1_hier.tcl (shows rebuffering and resizing with hierarchical design).

Key changes:

Journalling (dbJournal.cpp) to support hierarchical net connection/disconnection.
Propagation of modnets in rebuffering. (eg rebuffer.cc, RepairSetup.cc)
Wherever parasitics updated make sure a modnet is not used.

This is a substantial change so marked as draft until Matt and I have gone through changes.

Signed-off-by: andyfox-rushc <[email protected]>

…ck for max fanout limit to avoid split_load Signed-off-by: andyfox-rushc <[email protected]>

Signed-off-by: andyfox-rushc <[email protected]>

github-actions

clang-tidy made some suggestions

src/dbSta/src/dbNetwork.cc

src/rsz/src/Rebuffer.cc

Signed-off-by: andyfox-rushc <[email protected]>

github-actions

clang-tidy made some suggestions

src/rsz/src/Rebuffer.cc

Signed-off-by: andyfox-rushc <[email protected]>

github-actions

clang-tidy made some suggestions

src/dbSta/src/dbNetwork.cc

Signed-off-by: andyfox-rushc <[email protected]>

github-actions · 2024-10-29T20:04:55Z

clang-tidy review says "All clean, LGTM! 👍"

Signed-off-by: andyfox-rushc <[email protected]>

github-actions · 2024-10-29T21:42:16Z

clang-tidy review says "All clean, LGTM! 👍"

andyfox-rushc · 2024-10-29T22:10:53Z

Matt, ready for review. This one just has split_loads and rebuffer support for hierarchy.

Signed-off-by: andyfox-rushc <[email protected]>

github-actions · 2024-10-29T22:32:22Z

clang-tidy review says "All clean, LGTM! 👍"

maliberty · 2024-10-31T18:09:01Z

secure CI testing started

maliberty

partial review

maliberty · 2024-10-31T18:12:57Z

src/odb/src/db/dbModNet.cpp

+  delete (obj->_name);
+  obj->_name = strdup(new_name);


free is the opposite of strdup not delete.

I also don't see a free call in dbModNet::destroy

done.
Note the tbldestroy goes off and calls the destructor ~_dbModNet which inturn calls free on the name .
block->_modnet_tbl->destroy(_modnet);

maliberty · 2024-10-31T18:13:33Z

src/odb/src/db/dbModNet.cpp

+//
+// Support for renaming hierarchical nets
+//
+void dbModNet::reName(const char* new_name)


rename is one word so no need a capital in reName.

agreed, done

maliberty · 2024-10-31T18:14:00Z

src/odb/src/db/dbModNet.cpp

+  obj->_name = strdup(new_name);
+  _dbBlock* block = (_dbBlock*) obj->getOwner();
+  _dbModule* parent = block->_module_tbl->getPtr(obj->_parent);
+  parent->_modnet_hash[new_name] = obj->getOID();


You haven't removed the old name from the hash.

maliberty · 2024-10-31T18:21:32Z

src/odb/src/db/dbJournal.cpp

@@ -1618,6 +1634,7 @@ void dbJournal::undo_connectObject()
      dbITerm* iterm = dbITerm::getITerm(_block, iterm_id);
      uint net_id;
      _log.pop(net_id);
+      // disconnects everything modnet and bnet)


Poor comment, cleaned up.

maliberty · 2024-10-31T18:26:09Z

src/dbSta/src/dbNetwork.cc

+    std::string port_name_str = moditerm->getName();
+    size_t last_idx = port_name_str.find_last_of('/');
+    if (last_idx != string::npos) {
+      port_name_str = port_name_str.substr(last_idx + 1);
+    }


Why is this done by string manipulation? Is this different from getChildModBTerm()->getName()? Why does moditerm even need to store its name?

Agreed, I have put in the getChildModBTerm and remove the string manipulation.
I am sure not sure why but in the opensta both pins and port names are stored. So we do indeed store names in the modIterm and modBterm. I have updated the code and put in a comment about this in the dbSta/dbReadVerilog.cc. We can always get the ModBTerm for a modIterm using the getChildModBTerm api.

maliberty · 2024-10-31T18:28:33Z

src/rsz/src/EstimateWireParasitics.cc

+  odb::dbNet* db_net = nullptr;
+  odb::dbModNet* db_modnet = nullptr;
+  db_network_->staToDb(net, db_net, db_modnet);


What is the purpose? Nothing seems to use the results.

redundant code, removed.

maliberty · 2024-10-31T18:29:39Z

src/rsz/src/OdbCallBack.cc

@@ -150,7 +150,11 @@ void OdbCallBack::inDbInstSwapMasterAfter(dbInst* inst)
  while (pin_iter->hasNext()) {
    Pin* pin = pin_iter->next();
    Net* net = network_->net(pin);
-    resizer_->parasiticsInvalid(net);
+    // we can only update parasitics for low level net


I don't follow the comment. There is only one dbNet attached to the pin.

Another poor comment I am afraid. Cleaned up to call the new flatNet api, so it is clear that the parasiticsInvalid is applied to the dbNet only

maliberty · 2024-10-31T22:21:20Z

src/rsz/src/SteinerTree.cc

+  odb::dbNet* db_net;
+  odb::dbModNet* db_mod_net;
+  db_network_->net(drvr_pin, db_net, db_mod_net);


This pattern repeats a lot. It seems better to just make an api like odb::dbNet* db_net = db_network_->flat_net(drvr_pin);

Good idea. I called the api flatNet. Two versions: one for net and one for pin.

maliberty · 2024-10-31T22:22:06Z

src/rsz/src/RepairSetup.cc

+/*void
+writeVerilog(const char *filename,
+             bool sort,
+             bool include_pwr_gnd,
+             sta::CellSeq *remove_cells,
+             sta::Network *network);
+*/


yep, sorry some old debug code (was looking at verilog at various points in flow).

Signed-off-by: andyfox-rushc <[email protected]>

github-actions

clang-tidy made some suggestions

github-actions · 2024-10-31T23:50:07Z

src/dbSta/include/db_sta/dbNetwork.hh

@@ -132,6 +132,7 @@ class dbNetwork : public ConcreteNetwork

  dbNet* staToDb(const Net* net) const;
  void staToDb(const Net* net, dbNet*& dnet, dbModNet*& modnet) const;
+  dbNet* flatNet(const Net* pin) const;


warning: function 'sta::dbNetwork::flatNet' has a definition with different parameter names [readability-inconsistent-declaration-parameter-name]

dbNet* flatNet(const Net* pin) const; ^

Additional context

src/dbSta/src/dbNetwork.cc:2198: the definition seen here

dbNet* dbNetwork::flatNet(const Net* net) const ^

src/dbSta/include/db_sta/dbNetwork.hh:134: differing parameters are named here: ('pin'), in definition: ('net')

dbNet* flatNet(const Net* pin) const; ^

Signed-off-by: andyfox-rushc <[email protected]>

github-actions · 2024-11-01T18:41:19Z

clang-tidy review says "All clean, LGTM! 👍"

Signed-off-by: andyfox-rushc <[email protected]>

github-actions · 2024-11-01T19:19:13Z

clang-tidy review says "All clean, LGTM! 👍"

andyfox-rushc added 11 commits October 26, 2024 11:08

Work in progress: hierarchical support in rebuffering

f5adaf0

Signed-off-by: andyfox-rushc <[email protected]>

wip on repair setup4 with hierarchy

8fd68e2

Signed-off-by: andyfox-rushc <[email protected]>

Reformatted, removed hack for return on RepairSetup.cc but left in ha…

f412b55

…ck for max fanout limit to avoid split_load Signed-off-by: andyfox-rushc <[email protected]>

Reformatted, support for modnets in dbJournal

8ee3bbf

Signed-off-by: andyfox-rushc <[email protected]>

Added repair_setup4_hier case

2bc3085

Signed-off-by: andyfox-rushc <[email protected]>

Folded in latest master

bfdf1a0

Signed-off-by: andyfox-rushc <[email protected]>

Factor hierarchical connections to mimimize punch through

ee1f044

Signed-off-by: andyfox-rushc <[email protected]>

Removed debug code

20ffe7a

Signed-off-by: andyfox-rushc <[email protected]>

Split Load Hierarchy Unit test

efd06d1

Signed-off-by: andyfox-rushc <[email protected]>

Cleaned up duplicate names in test case

57400a3

Signed-off-by: andyfox-rushc <[email protected]>

Folded in latest master

a6bfdef

Signed-off-by: andyfox-rushc <[email protected]>

andyfox-rushc marked this pull request as draft October 29, 2024 13:29

github-actions bot reviewed Oct 29, 2024

View reviewed changes

src/dbSta/src/dbNetwork.cc Outdated Show resolved Hide resolved

src/rsz/src/Rebuffer.cc Outdated Show resolved Hide resolved

src/rsz/src/Rebuffer.cc Outdated Show resolved Hide resolved

src/rsz/src/Rebuffer.cc Outdated Show resolved Hide resolved

andyfox-rushc added 2 commits October 29, 2024 06:58

Removed unused vars

8d0f598

Signed-off-by: andyfox-rushc <[email protected]>

ctidy clean ups

bc34e51

Signed-off-by: andyfox-rushc <[email protected]>

github-actions bot reviewed Oct 29, 2024

View reviewed changes

src/rsz/src/Rebuffer.cc Show resolved Hide resolved

ctidy fixes

6c775ae

Signed-off-by: andyfox-rushc <[email protected]>

github-actions bot reviewed Oct 29, 2024

View reviewed changes

src/dbSta/src/dbNetwork.cc Outdated Show resolved Hide resolved

Ctidy suggestion, remove virtual for override

6a2bfbe

Signed-off-by: andyfox-rushc <[email protected]>

Regolden split_load_hier to remove extraneous print

e1f0444

Signed-off-by: andyfox-rushc <[email protected]>

andyfox-rushc marked this pull request as ready for review October 29, 2024 22:10

Removed debug code checking for bad update parasitic on modnet

1b27603

Signed-off-by: andyfox-rushc <[email protected]>

maliberty requested changes Oct 31, 2024

View reviewed changes

First round of review comments

9c89a1a

Signed-off-by: andyfox-rushc <[email protected]>

github-actions bot reviewed Oct 31, 2024

View reviewed changes

andyfox-rushc added 2 commits November 1, 2024 10:56

setChildModBTerm for moditerm when creating hierachy

b31759e

Signed-off-by: andyfox-rushc <[email protected]>

ctidy flatNet

842e502

Signed-off-by: andyfox-rushc <[email protected]>

Regolden readdb_hier

80284f5

Signed-off-by: andyfox-rushc <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repair setup4 hier case 1 buffer #6048

Repair setup4 hier case 1 buffer #6048

andyfox-rushc commented Oct 29, 2024

github-actions bot left a comment

github-actions bot left a comment

github-actions bot left a comment

github-actions bot commented Oct 29, 2024

github-actions bot commented Oct 29, 2024

andyfox-rushc commented Oct 29, 2024

github-actions bot commented Oct 29, 2024

maliberty commented Oct 31, 2024

maliberty left a comment

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

maliberty Oct 31, 2024

andyfox-rushc Oct 31, 2024

github-actions bot left a comment

github-actions bot Oct 31, 2024

github-actions bot commented Nov 1, 2024

github-actions bot commented Nov 1, 2024

Repair setup4 hier case 1 buffer #6048

Are you sure you want to change the base?

Repair setup4 hier case 1 buffer #6048

Conversation

andyfox-rushc commented Oct 29, 2024

github-actions bot left a comment

Choose a reason for hiding this comment

github-actions bot left a comment

Choose a reason for hiding this comment

github-actions bot left a comment

Choose a reason for hiding this comment

github-actions bot commented Oct 29, 2024

github-actions bot commented Oct 29, 2024

andyfox-rushc commented Oct 29, 2024

github-actions bot commented Oct 29, 2024

maliberty commented Oct 31, 2024

maliberty left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

github-actions bot left a comment

Choose a reason for hiding this comment

github-actions bot Oct 31, 2024

Choose a reason for hiding this comment

github-actions bot commented Nov 1, 2024

github-actions bot commented Nov 1, 2024