[FIRRTL][Inliner] Support per-field syms, update local users.

[dtzSiFive]

No description provided.

[dtzSiFive]

cc #5598 -- TODO: link issue from commit/PR, include test. ✔️

[prithayan]

Shouldn't the following @nla2 annotation become a local annotation, instead of being dropped,

firrtl.circuit "CollidingSymbolsFields" {
  hw.hierpath private @nla1 [@CollidingSymbolsFields::@foo, @Foo::@bar, @Bar]
  hw.hierpath private @nla2 [@CollidingSymbolsFields::@foo, @Foo::@b_0]
  firrtl.module @Bar() attributes {annotations = [{circt.nonlocal = @nla1, class = "nla1"}]} {}
  firrtl.module @Foo(in %x : !firrtl.bundle<a: uint<1>, b: uint<1>> sym [<@b_0,1,public>,<@foo,2,public>]) attributes {annotations = [{class = "firrtl.passes.InlineAnnotation"}, {circt.nonlocal = @nla2, class = "nla2"}]} {
    %b = firrtl.wire sym [<@b,1,public>,<@bar_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
    firrtl.instance bar sym @bar @Bar()
    %1 = firrtl.ref.rwprobe <@Foo::@b_0> : !firrtl.uint<1>
    %2 = firrtl.ref.rwprobe <@Foo::@bar_0> : !firrtl.uint<1>
    %3 = firrtl.ref.rwprobe <@Foo::@foo> : !firrtl.uint<1>
  }
  firrtl.module @CollidingSymbolsFields(in %x : !firrtl.bundle<a: uint<1>> sym [<@b_0,1,public>]) {
    firrtl.instance foo sym @foo @Foo(in x : !firrtl.bundle<a: uint<1>, b: uint<1>>)
    %collision_b = firrtl.wire sym @b : !firrtl.uint<1>
    %collision_bar = firrtl.wire sym [<@bar,1,public>,<@bar_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
  }
}

output with this PR:

firrtl.circuit "CollidingSymbolsFields" {
    hw.hierpath private @nla1 [@CollidingSymbolsFields::@bar_1, @Bar]
    firrtl.module @Bar() attributes {annotations = [{circt.nonlocal = @nla1, class = "nla1"}]} {
    }
    firrtl.module @Foo(in %x: !firrtl.bundle<a: uint<1>, b: uint<1>> sym [<@b_0,1,public>, <@foo,2,public>]) attributes {annotations = [{circt.nonlocal = @nla2, class = "nla2"}]} {
      %b = firrtl.wire sym [<@b,1,public>, <@bar_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
      firrtl.instance bar sym @bar @Bar()
      %0 = firrtl.ref.rwprobe <@Foo::@b_0> : !firrtl.uint<1>
      %1 = firrtl.ref.rwprobe <@Foo::@bar_0> : !firrtl.uint<1>
      %2 = firrtl.ref.rwprobe <@Foo::@foo> : !firrtl.uint<1>
    }
    firrtl.module @CollidingSymbolsFields(in %x: !firrtl.bundle<a: uint<1>> sym [<@b_0,1,public>]) {
      %foo_x = firrtl.wire sym [<@b_0_0,1,public>, <@foo_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
      %foo_b = firrtl.wire sym [<@b_1,1,public>, <@bar_0_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
      firrtl.instance foo_bar sym @bar_1 @Bar()
      %0 = firrtl.ref.rwprobe <@CollidingSymbolsFields::@b_0_0> : !firrtl.uint<1>
      %1 = firrtl.ref.rwprobe <@CollidingSymbolsFields::@bar_0_0> : !firrtl.uint<1>
      %2 = firrtl.ref.rwprobe <@CollidingSymbolsFields::@foo_0> : !firrtl.uint<1>
      %collision_b = firrtl.wire sym @b : !firrtl.uint<1>
      %collision_bar = firrtl.wire sym [<@bar,1,public>, <@bar_0,2,public>] : !firrtl.bundle<a: uint<1>, b: uint<1>>
    }
  }

[dtzSiFive]

Shouldn't the following @nla2 annotation become a local annotation, instead of being dropped,

Thank you for pushing for breakages, this is a tricky thing to handle! Really appreciate it!

Regarding the example-- ah, yes. Okay so I somewhat intentionally didn't bother trying to make this work (note it presently crashes on your input) re:updating old-style NLA's (leaf reference isn't just the module, but points to the symbol in particular) since that looked super hard and I /think/ we don't need that anymore?

I'm not sure, though, will investigate a bit more. Please LMK if that sounds wrong! If not, I actually meant to push on (and perhaps bug you about) whether no longer needing to support leaf references (old-style NLA's) could significantly simplify the pass? The only symbols in NLA's we should need to update these days, I think/hope, are the ones on instances (which can't have per-field symbols anyway). WDYT?

[prithayan]

I think I agree, we need not complicate the implementation further to support old-style NLAs.
LGTM.

[dtzSiFive]

Thanks!