SROA: generalize unswitchtupleunion optimization

[aviatesk]

This commit improves SROA pass by extending the unswitchtupleunion optimization to handle the general parametric types, e.g.:

julia> struct A{T}
           x::T
       end;

julia> function foo(a1, a2, c)
           t = c ? A(a1) : A(a2)
           return getfield(t, :x)
       end;

julia> only(Base.code_ircode(foo, (Int,Float64,Bool); optimize_until="SROA"))

Before

2 1 ─      goto #3 if not _4                                          │
  2 ─ %2 = %new(A{Int64}, _2)::A{Int64}                               │╻ A
  └──      goto #4                                                    │
  3 ─ %4 = %new(A{Float64}, _3)::A{Float64}                           │╻ A
  4 ┄ %5 = φ (#2 => %2, #3 => %4)::Union{A{Float64}, A{Int64}}        │
3 │   %6 = Main.getfield(%5, :x)::Union{Float64, Int64}               │
  └──      return %6                                                  │
   => Union{Float64, Int64}

After

julia> only(Base.code_ircode(foo, (Int,Float64,Bool); optimize_until="SROA"))
2 1 ─      goto #3 if not _4                                           │
  2 ─      nothing::A{Int64}                                           │╻ A
  └──      goto #4                                                     │
  3 ─      nothing::A{Float64}                                         │╻ A
  4 ┄ %8 = φ (#2 => _2, #3 => _3)::Union{Float64, Int64}               │
  │        nothing::Union{A{Float64}, A{Int64}}
3 │   %6 = %8::Union{Float64, Int64}                                   │
  └──      return %6                                                   │
   => Union{Float64, Int64}

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[Keno]

Sure, why not

[maleadt]

This also caused many PkgEval regressions, so good that it will be reverted:

EDIT: this PR actually fixes them.

Internal error: encountered unexpected error in runtime:
BoundsError(a=svec(), i=1)
ijl_bounds_error_int at /cache/build/default-amdci4-3/julialang/julia-master/src/rtutils.c:187
jl_f__svec_ref at /cache/build/default-amdci4-3/julialang/julia-master/src/builtins.c:1668
getindex at ./essentials.jl:764 [inlined]
unswitchtypeunion at ./compiler/typeutils.jl:345
unswitchtupleunion at ./compiler/typeutils.jl:321 [inlined]
sroa_pass! at ./compiler/ssair/passes.jl:1111
run_passes at ./compiler/optimize.jl:486
run_passes at ./compiler/optimize.jl:501 [inlined]
optimize at ./compiler/optimize.jl:450
jfptr_optimize_37641.1 at /opt/julia/lib/julia/sys.so (unknown line)
_jl_invoke at /cache/build/default-amdci4-3/julialang/julia-master/src/gf.c:2890 [inlined]
ijl_apply_generic at /cache/build/default-amdci4-3/julialang/julia-master/src/gf.c:3072
_typeinf at ./compiler/typeinfer.jl:265
typeinf at ./compiler/typeinfer.jl:216

[KristofferC]

(it hasnt been reverted)

[aviatesk]

This also caused many PkgEval regressions, so good that it has been reverted:

@maleadt At which package does the regression happen?

[maleadt]

See the latest daily, https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_date/2023-07/11/report.html, most of the crashes point to this PR. For example, StaticArrayInterface, VectorizationBase, StrideArraysCore, ImageMorphology, ...

[aviatesk]

Thanks!

[aviatesk]

The regressions are introduced by #50483, but then fixed by #50502 (this PR). So they don't reproduce on the latest master.

[maleadt]

Ah, OK. I saw unswitchtupleunion and only thought of this PR.