Fix use counts for mutable struct SROA

PR #28478 moved the computation of the use counts before the finish call.
to fix #28444. However, the early parts of the finish call fixes up phi
node arguments, which fail to get counted if we look at use counts before
that fixup is performed. This causes #30594 where the only non-trivial use
is on the backedge of the phi and would thus incorrectly fail to get accounted
for. Fix that by taking the use count after phi fixup but before dce.

(cherry picked from commit f8f2045)

Author: Keno

base/compiler/ssair/passes.jl

@@ -694,10 +694,14 @@ function getfield_elim_pass!(ir::IRCode, domtree::DomTree)
         compact[idx] = val === nothing ? nothing : val.x
     end
 
-    # Copy the use count, `finish` may modify it and for our predicate
-    # below we need it consistent with the state of the IR here.
+
+    non_dce_finish!(compact)
+    # Copy the use count, `simple_dce!` may modify it and for our predicate
+    # below we need it consistent with the state of the IR here (after tracking
+    # phi node arguments, but before dce).
     used_ssas = copy(compact.used_ssas)
-    ir = finish(compact)
+    simple_dce!(compact)
+    ir = complete(compact)
     # Now go through any mutable structs and see which ones we can eliminate
     for (idx, (intermediaries, defuse)) in defuses
         intermediaries = collect(intermediaries)

test/compiler/irpasses.jl

@@ -2,6 +2,32 @@
 
 using Test
 
+# Tests for SROA
+
+mutable struct Foo30594; x::Float64; end
+Base.copy(x::Foo30594) = Foo30594(x.x)
+function add!(p::Foo30594, off::Foo30594)
+    p.x += off.x
+    return p
+end
+Base.:(+)(a::Foo30594, b::Foo30594) = add!(copy(a), b)
+
+let results = Float64[]
+    @noinline use30594(x) = push!(results, x.x); nothing
+    function foo30594(cnt::Int, dx::Int)
+        step = Foo30594(dx)
+        curr = step + Foo30594(1)
+        for i in 1:cnt
+            use30594(curr)
+            curr = curr + step
+        end
+        nothing
+    end
+
+    foo30594(4, -1)
+    @test results == [0.0, -1.0, -2.0, -3.0]
+end
+
 # Issue #29983
 # This one is a bit hard to trigger, but the key is to create a case
 # where SROA needs to introduce an intermediate type-unstable phi node