Original Redmine Comment
Author Name: Todd Strader (@toddstrader)
Original Date: 2015-09-30T00:30:40Z

I tried the simple thing:

$ git diff src/Verilator.cpp
diff --git a/src/Verilator.cpp b/src/Verilator.cpp
index 6a820c2..dcd2b94 100644
--- a/src/Verilator.cpp
+++ b/src/Verilator.cpp
@@ -170,6 +170,13 @@ void process () {
 
      if (v3Global.opt.stats()) V3Stats::statsStageAll(v3Global.rootp(), "Link");
 
+// TAS -- moved this ---\
+    // Calculate and check widths, edit tree to TRUNC/EXTRACT any width mismatches
+    V3Width::width(v3Global.rootp());
+
+    V3Error::abortIfErrors();
+// TAS -- moved this ---/
+
      // Remove parameters by cloning modules to de-parameterized versions
      //   This requires some width calculations and constant propagation
      V3Param::param(v3Global.rootp());
@@ -180,10 +187,10 @@ void process () {
      V3Dead::deadifyModules(v3Global.rootp());
      v3Global.checkTree();
 
-    // Calculate and check widths, edit tree to TRUNC/EXTRACT any width mismatches
-    V3Width::width(v3Global.rootp());
-
-    V3Error::abortIfErrors();
+//    // Calculate and check widths, edit tree to TRUNC/EXTRACT any width mismatches
+//    V3Width::width(v3Global.rootp());
+//
+//    V3Error::abortIfErrors();
 
      // Commit to the widths we've chosen; Make widthMin==width
      V3Width::widthCommit(v3Global.rootp());

And while I got enum item values to work with in the paramlink step, I also got:

TESTS Passed 587  Unsup 18  Skipped 24  Failed 82  Time 3:56

Perhaps I can extract the enum value assignment from the width step and run that before paramlink? That's what I'll try next.

Original Redmine Comment
Author Name: Wilson Snyder (@wsnyder)
Original Date: 2015-09-30T00:46:22Z

Constification causes calls to width as each individual item requires it. I suspect V3Const's AstEnumItemRef call is missing something, probably the enum item hasn't been made into a constant. Probably code line V3Const's AstFuncRef "if (m_param)" is needed in AstEnumItemRef. That is, when we see an EnumItemRef and there's no value yet, and we're parameter constifying, we need to iterate on the EnumItem to constify it.

Original Redmine Comment
Author Name: Todd Strader (@toddstrader)
Original Date: 2015-09-30T22:16:19Z

Thanks for the pointers. It seems that once again, things are simpler than I assumed.

However, I am a bit confused by what I'm currently seeing. I added an EnumItemRef visitor to V3Simulate with some debug just to see what's going on (pushed to GitHub). When I run the test while debugging V3Simulate, I get:

- V3Simulate.h:349:      simulate enumitemref nodep ENUMITEMREF 0x10b7740 <e156#> {e16} @dt=0x10bf7f0@(sw32)  LOW_FIVE -> ENUMITEM 0x10bf730 <e142#> {e10} @dt=0x10bf7f0@(sw32)  LOW_FIVE
- V3Simulate.h:351:      simulate enumitemref itemp ENUMITEM 0x10bf730 <e142#> {e10} @dt=0x10bf7f0@(sw32)  LOW_FIVE
- V3Simulate.h:353:      simulate enumitemref valuep CONST 0x10bbd00 <e144#> {e10} @dt=0x10bf530@(G/w32)  32'hdeadbef0

This seems strange to me because LOW_FIVE already seems to have the right enum item value (0xdeadbeef + 1), but I can't find this node in any of the tree dumps. Vt_static_elab_enum_007_link.tree shows :

     1:2:1:2: ENUMITEM 0x10bf610 <e10> {e9} @dt=0@  HIGH_FIVE
     1:2:1:2:2: CONST 0x10bf3f0 <e9> {e9} @dt=0x10bf530@(G/w32)  32'hdeadbeef
     1:2:1:2: ENUMITEM 0x10bf730 <e12> {e10} @dt=0@  LOW_FIVE
     1:2: FUNC 0x10bfc50 <e34> {e13} @dt=0@  gimme_five

In fact, no one has heard of this guy:

tstrader@sark:~/git/verilator_static_elab_2/test_regress$ grep 0x10bbd00 obj_dir/t_static_elab_enum/*tree
tstrader@sark:~/git/verilator_static_elab_2/test_regress$ grep deadbef0 obj_dir/t_static_elab_enum/*tree
tstrader@sark:~/git/verilator_static_elab_2/test_regress$ 

Where did LOW_FIVE's CONST node come from? And is it a problem that it doesn't show up in any of the tree dumps? Does it get created during paramlink? If so, why does paramlink say that there are no changes? Is this because the build fails?

Sorry for all the questions, just trying to get my head around what is happening here.

Original Redmine Comment
Author Name: Wilson Snyder (@wsnyder)
Original Date: 2015-09-30T23:57:17Z

Since the constant isn't part of the tree it was probably created by V3Simulate. It passes constants across itself to represent variable values related to the nodes it's simulating.

Original Redmine Comment
Author Name: Todd Strader (@toddstrader)
Original Date: 2015-10-01T22:17:10Z

GitHub repo should be ready to go now. I combined the struct and enum parameter tests since I've seen a a few tests that roll up a few things like that. Please let me know if this isn't desirable and I'll break them apart.

Original Redmine Comment
Author Name: Wilson Snyder (@wsnyder)
Original Date: 2015-10-02T01:16:01Z

Oh, that makes sense, simpler than I thought too.

Fixed in git towards 3.877.

Original Redmine Comment
Author Name: Wilson Snyder (@wsnyder)
Original Date: 2015-11-01T13:23:03Z

In 3.878.