@@ -70,7 +70,7 @@ function lower_block(
70
70
lower! (blockq, opsv1, ls, unrollsyms, u₁, u₂, t, mask & ! (dontmaskfirsttiles & (t < u₂ - 1 )), lowernonstore, lowerstore)
71
71
if iszero (t) && ! store # u₁ && !u₂
72
72
# for u ∈ 0:u₁-1
73
- lower! (blockq, ops[2 ,1 ,prepost,n], ls, unrollsyms, u₁, u₂, - 1 , mask, lowernonstore, lowerstore )
73
+ lower! (blockq, ops[2 ,1 ,prepost,n], ls, unrollsyms, u₁, u₂, - 1 , mask, true , true )
74
74
# end
75
75
end
76
76
# u₁ && u₂
@@ -504,7 +504,6 @@ function initialize_outer_reductions!(
504
504
u₁u, u₂u = isunrolled_sym (op, getloop (ls, us. u₁loopnum). itersymbol, getloop (ls, us. u₂loopnum). itersymbol, getloop (ls, us. vloopnum). itersymbol, ls)# , u₂)
505
505
z = outer_reduction_zero (op, u₁u, Umax, reduction_instruction_class (instruction (op)), rs)
506
506
mvar = variable_name (op, - 1 )
507
- # @show u₁, u₂, u₁u, _Umax
508
507
if (u₂ == - 1 )
509
508
push! (q. args, Expr (:(= ), Symbol (mvar, ' _' , _Umax), z))
510
509
elseif u₁u
@@ -549,7 +548,6 @@ function add_upper_outer_reductions(ls::LoopSet, loopq::Expr, Ulow::Int, Uhigh::
549
548
ifq = Expr (:block )
550
549
ifqlet = Expr (:block )
551
550
initialize_outer_reductions! (ifqlet, ls, Uhigh)
552
- # @show loopq
553
551
push! (ifq. args, loopq)
554
552
t = Expr (:tuple )
555
553
mvartu = Expr (:tuple )
738
736
maskexpr (looplimit) = Expr (:(= ), MASKSYMBOL, Expr (:call , lv (:mask ), VECTORWIDTHSYMBOL, looplimit))
739
737
@inline idiv_fast (a:: I , b:: I ) where {I <: Base.BitInteger } = Base. udiv_int (a, b)
740
738
@inline idiv_fast (a, b) = idiv_fast (Int (a), Int (b))
741
- # @inline idiv_fast(a, b) = idiv_fast(@show(Int(a)), @show(Int(b)))
742
739
function definemask (loop:: Loop )
743
740
isstaticloop (loop) && return maskexpr (length (loop))
744
741
# W = 4
@@ -863,7 +860,6 @@ ureduct(ls::LoopSet) = ls.ureduct
863
860
function lower_unrollspec (ls:: LoopSet )
864
861
us = ls. unrollspecification
865
862
@unpack vloopnum, u₁, u₂ = us
866
- # @show u₁, u₂
867
863
order = names (ls)
868
864
init_loop_map! (ls)
869
865
vectorized = order[vloopnum]
@@ -955,7 +951,6 @@ function isunrolled_sym(
955
951
u₁reduced = u₁loop ∈ reductops
956
952
u₂reduced = u₂loop ∈ reductops
957
953
# If they're being reduced, we want to only unroll the reduced variable along one of the two loops.
958
- # @show u₁reduced, u₂reduced
959
954
if u₂reduced
960
955
if u₁reduced# if both are reduced, we unroll u₁
961
956
if vloop === u₁loop
0 commit comments