From 0d629549f22595189fae7249a0e203d24e983031 Mon Sep 17 00:00:00 2001
From: Chris Elrod <elrodc@gmail.com>
Date: Wed, 25 May 2022 11:58:10 -0400
Subject: [PATCH] fix integer outer reduct initialization on
 simd_integer_register_size < register_size arches

---
 Project.toml            |  2 +-
 src/codegen/lowering.jl | 14 +++++++++++---
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/Project.toml b/Project.toml
index 367700aa9..82202603f 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <elrodc@gmail.com>"]
-version = "0.12.111"
+version = "0.12.112"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"
diff --git a/src/codegen/lowering.jl b/src/codegen/lowering.jl
index 9d011783e..df26ef26d 100644
--- a/src/codegen/lowering.jl
+++ b/src/codegen/lowering.jl
@@ -515,20 +515,28 @@ function pointerremcomparison(
   end
 end
 
-@generated function of_same_size(::Type{T}, ::Type{S}) where {T,S}
+@generated function of_same_size(::Type{T}, ::Type{S}, ::StaticInt{R}) where {T,S,R}
   sizeof_S = sizeof(S)
+  if T <: Integer && sizeof(T) == 8
+    # sizeof(T) == 8 && max(..., 4) to maybe demote Int64 -> Int32
+    # but otherwise, we're giving up too much with the demotion.
+    sizeof_S *= max(8 ÷ R, 4)
+  end
   sizeof(T) == sizeof_S && return T
   # Tfloat = T <: Union{Float32,Float64}
   if T <: Union{Float32,Float64}
     sizeof_S ≥ 8 ? Float64 : Float32
   elseif T <: Signed
-    Symbol(:Int, 8sizeof_S)
+    Symbol(:Int, sizeof_S)
   elseif (T <: Unsigned) | (T === Bool)
-    Symbol(:UInt, 8sizeof_S)
+    Symbol(:UInt, sizeof_S)
   else
     S
   end
 end
+@inline function of_same_size(::Type{T}, ::Type{S}) where {T,S}
+  of_same_size(T, S, VectorizationBase.register_size() ÷ VectorizationBase.simd_integer_register_size())
+end
 function outer_reduction_zero(
   op::Operation,
   u₁u::Bool,