Change handling of reductions to better accomdate their use within the loop body, and don't unroll inner reductions where the unrolled loop is the reduction loop if that reduction is already tiled.

Author: chriselrod

Manifest.toml

@@ -42,9 +42,9 @@ uuid = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 
 [[SIMDPirates]]
 deps = ["VectorizationBase"]
-git-tree-sha1 = "1a9cbe1be1f5d43ac49eeb38ca64dd78e50a0cc6"
+git-tree-sha1 = "8f89aa38f5e4e89f2a474ffdc850fc21d6ab9ed4"
 uuid = "21efa798-c60a-11e8-04d3-e1a92915a26a"
-version = "0.7.3"
+version = "0.7.4"
 
 [[SLEEFPirates]]
 deps = ["Libdl", "SIMDPirates", "VectorizationBase"]
@@ -69,6 +69,6 @@ version = "0.1.0"
 
 [[VectorizationBase]]
 deps = ["CpuId", "LinearAlgebra"]
-git-tree-sha1 = "83e32d4835fc4f4ecfd43eb59fa7fc00854b3d41"
+git-tree-sha1 = "2a83ab02d3fb6ad5de0c6d04103b0ca403d9a7d8"
 uuid = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
-version = "0.9.4"
+version = "0.9.5"

Project.toml

@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <[email protected]>"]
-version = "0.6.23"
+version = "0.6.24"
 
 [deps]
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
@@ -13,10 +13,10 @@ VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
 
 [compat]
 OffsetArrays = "1"
-SIMDPirates = "0.7.3"
+SIMDPirates = "0.7.4"
 SLEEFPirates = "0.4"
 UnPack = "0"
-VectorizationBase = "0.9.3"
+VectorizationBase = "0.9.5"
 julia = "1.1"
 
 [extras]