Merge pull request #181 from DilumAluthge/dpa/avx-coverage

Include the original `LineNumberNode`s from the original expression in the final expression

Author: chriselrod

Project.toml

@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <[email protected]>"]
-version = "0.9.12"
+version = "0.9.13"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"

src/LoopVectorization.jl

@@ -17,8 +17,8 @@ using VectorizationBase: REGISTER_SIZE, REGISTER_COUNT, data,
 
 using IfElse: ifelse
 
-# missing: stridedpointer_for_broadcast, noalias!, gepbyte, 
-# using SIMDPirates: VECTOR_SYMBOLS, evadd, evsub, evmul, evfdiv, vrange, 
+# missing: stridedpointer_for_broadcast, noalias!, gepbyte,
+# using SIMDPirates: VECTOR_SYMBOLS, evadd, evsub, evmul, evfdiv, vrange,
 #     reduced_add, reduced_prod, reduce_to_add, reduced_max, reduced_min, vsum, vprod, vmaximum, vminimum,
 #     sizeequivalentfloat, sizeequivalentint, vadd!, vsub!, vmul!, vfdiv!, vfmadd!, vfnmadd!, vfmsub!, vfnmsub!,
 #     vfmadd231, vfmsub231, vfnmadd231, vfnmsub231, sizeequivalentfloat, sizeequivalentint, #prefetch,
@@ -67,6 +67,7 @@ include("add_compute.jl")
 include("add_constants.jl")
 include("add_ifelse.jl")
 include("determinestrategy.jl")
+include("line_number_nodes.jl")
 include("loopstartstopmanager.jl")
 include("lower_compute.jl")
 include("lower_constant.jl")

src/condense_loopset.jl

@@ -210,7 +210,7 @@ end
 # Try to condense in type stable manner
 function generate_call(ls::LoopSet, inline_unroll::NTuple{3,Int8}, debug::Bool = false)
     operation_descriptions = Expr(:curly, :Tuple)
-    varnames = Symbol[]; ids = Vector{Int}(undef, length(operations(ls))) 
+    varnames = Symbol[]; ids = Vector{Int}(undef, length(operations(ls)))
     for op ∈ operations(ls)
         instr = instruction(op)
         push!(operation_descriptions.args, QuoteNode(instr.mod))
@@ -315,7 +315,7 @@ function setup_call_inline(ls::LoopSet, inline::Int8 = zero(Int8), U::Int8 = zer
 end
 function setup_call_debug(ls::LoopSet)
     # avx_loopset(instr, ops, arf, AM, LB, vargs)
-    
+
     pushpreamble!(ls, generate_call(ls, (zero(Int8),zero(Int8),zero(Int8)), true))
     Expr(:block, ls.prepreamble, ls.preamble)
 end
@@ -325,8 +325,11 @@ function setup_call(ls::LoopSet, q = nothing, inline::Int8 = zero(Int8), check_e
     # the generated function must be inlined into the initial loop preamble for performance reasons.
     # Creating an anonymous function and calling it also achieves the outlining, while still
     # inlining the generated function into the loop preamble.
+    lnns = extract_all_lnns(q)
     call = setup_call_inline(ls, inline, u₁, u₂)
     call = check_empty ? check_if_empty(ls, call) : call
     isnothing(q) && return Expr(:block, ls.prepreamble, call)
-    Expr(:block, ls.prepreamble, Expr(:if, check_args_call(ls), call, make_crashy(make_fast(q))))
+    result = Expr(:block, ls.prepreamble, Expr(:if, check_args_call(ls), call, make_crashy(make_fast(q))))
+    prepend_lnns!(result, lnns)
+    return result
 end

src/line_number_nodes.jl

@@ -0,0 +1,29 @@
+function extract_all_lnns(x)
+    lnns = Vector{LineNumberNode}(undef, 0)
+    extract_all_lnns!(lnns, x)
+    return lnns
+end
+
+function extract_all_lnns!(lnns::AbstractVector{<:LineNumberNode}, lnn::LineNumberNode)
+    push!(lnns, lnn)
+    return lnns
+end
+function extract_all_lnns!(lnns::AbstractVector{<:LineNumberNode}, ex::Expr)
+    for arg in ex.args
+        extract_all_lnns!(lnns::AbstractVector{<:LineNumberNode}, arg)
+    end
+    return lnns
+end
+function extract_all_lnns!(lnns::AbstractVector{<:LineNumberNode}, ::Any)
+    return lnns
+end
+
+function prepend_lnns!(ex::Expr, lnns::AbstractVector{<:LineNumberNode})
+    return prepend_lnns!(ex, lnns, Val(ex.head))
+end
+function prepend_lnns!(ex::Expr, lnns::AbstractVector{<:LineNumberNode}, ::Val{:block})
+    for lnn in lnns
+        pushfirst!(ex.args, Expr(:block, lnn, :(nothing)))
+    end
+    return ex
+end