Remove unnessary branch for static length loops.

Author: chriselrod

Project.toml

@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <[email protected]>"]
-version = "0.8.16"
+version = "0.8.17"
 
 [deps]
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"

src/lowering.jl

@@ -340,19 +340,19 @@ function lower_unrolled_dynamic(ls::LoopSet, us::UnrollSpecification, n::Int, in
     usorig = ls.unrollspecification[]
     # tc = (usorig.u₁ == us.u₁) && (usorig.u₂ == us.u₂) && !loopisstatic && !inclmask && !ls.loadelimination[] ? expect(tc) : tc
     body = lower_block(ls, us, n, inclmask, UF)
-    q = if loopisstatic
+    if loopisstatic
         iters = length(loop) ÷ UFW
         if iters*UF ≤ 16 # Let's set a limit on total unrolling
             q = Expr(:block)
             foreach(_ -> push!(q.args, body), 1:iters)
-            q
         else
-            Expr(:while, tc, body)
+            q = Expr(:while, tc, body)
         end
+        remblock = Expr(:block)
     else
-        Expr(:while, tc, body)
+        remblock = init_remblock(loop, ls.lssm[], n)#loopsym)
+        q = Expr(:while, tc, body)
     end
-    remblock = init_remblock(loop, ls.lssm[], n)#loopsym)
     q = if unsigned(Ureduct) < unsigned(UF) # unsigned(-1) == typemax(UInt); is logic relying on twos-complement bad?
         UF_cleanup = UF - Ureduct
         us_cleanup = nisunrolled ? UnrollSpecification(us, UF_cleanup, u₂) : UnrollSpecification(us, u₁, UF_cleanup)

test/runtests.jl

@@ -5,7 +5,7 @@ using LinearAlgebra
 # const START_TIME = time()
 # exceeds_time_limit() = (time() - START_TIME) > 35 * 60
 
-function clenshaw(x,coeff)
+function clenshaw(x, coeff)
     len_c = length(coeff)
     tmp = zero(x)
     ret = zero(x)