Ensure refs alias stores on new lines

Author: chriselrod

Project.toml

@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <[email protected]>"]
-version = "0.12.58"
+version = "0.12.59"
 
 [deps]
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"

src/modeling/graphs.jl

@@ -1128,16 +1128,18 @@ function add_operation!(
     end
 end
 
-function prepare_rhs_for_storage!(ls::LoopSet, RHS::Union{Symbol,Expr}, array, rawindices, elementbytes::Int, position::Int)
-    RHS isa Symbol && return add_store!(ls, RHS, array, rawindices, elementbytes)
-    mpref = array_reference_meta!(ls, array, rawindices, elementbytes)
-    cachedparents = copy(mpref.parents)
-    ref = mpref.mref.ref
-    lrhs = gensym!(ls, "RHS")
-    mpref.varname = lrhs
-    add_operation!(ls, lrhs, RHS, mpref, elementbytes, position)
-    mpref.parents = cachedparents
-    add_store!(ls, mpref, elementbytes)
+function prepare_rhs_for_storage!(ls::LoopSet, RHS::Union{Symbol,Expr}, array, rawindices, elementbytes::Int, position::Int)::Operation
+  RHS isa Symbol && return add_store!(ls, RHS, array, rawindices, elementbytes)
+  mpref = array_reference_meta!(ls, array, rawindices, elementbytes)
+  cachedparents = copy(mpref.parents)
+  ref = mpref.mref.ref
+  lrhs = gensym!(ls, "RHS")
+  mpref.varname = lrhs
+  add_operation!(ls, lrhs, RHS, mpref, elementbytes, position)
+  mpref.parents = cachedparents
+  op = add_store!(ls, mpref, elementbytes)
+  ls.syms_aliasing_refs[findfirst(==(mpref.mref), ls.refs_aliasing_syms)] = lrhs
+  return op
 end
 
 function add_assignment!(ls::LoopSet, LHS, RHS, elementbytes::Int, position::Int)

test/convolutions.jl

@@ -0,0 +1,38 @@
+
+function compute_update0!(H2, dHdtau, H, min_dxy2, _dx, _dy, dx, dy)
+  @inbounds @fastmath for iy=1:size(dHdtau,2)
+    for ix=1:size(dHdtau,1)
+      dHdtau[ix, iy] = (-((-((H[(ix + 1) + 1, iy + 1] - H[ix + 1, iy + 1])) * _dx - -((H[ix + 1, iy + 1] - H[ix, iy + 1])) * _dx)) * _dx - (-((H[ix + 1, (iy + 1) + 1] - H[ix + 1, iy + 1])) * _dy - -((H[ix + 1, iy + 1] - H[ix + 1, iy])) * _dy) * _dy) + 0.6 * dHdtau[ix, iy]
+      H2[ix+1,iy+1] = H[ix+1,iy+1] + min_dxy2/4.1*dHdtau[ix,iy]  # sets the BC as H[1]=H[end]=0
+    end
+  end
+  return
+end
+function compute_update1!(H2, dHdtau, H, min_dxy2, _dx, _dy, dx, dy)
+  @turbo for iy=1:size(dHdtau,2)
+    for ix=1:size(dHdtau,1)
+      dHdtau[ix, iy] = (-((-((H[(ix + 1) + 1, iy + 1] - H[ix + 1, iy + 1])) * _dx - -((H[ix + 1, iy + 1] - H[ix, iy + 1])) * _dx)) * _dx - (-((H[ix + 1, (iy + 1) + 1] - H[ix + 1, iy + 1])) * _dy - -((H[ix + 1, iy + 1] - H[ix + 1, iy])) * _dy) * _dy) + 0.6 * dHdtau[ix, iy]
+      H2[ix+1,iy+1] = H[ix+1,iy+1] + min_dxy2/4.1*dHdtau[ix,iy]  # sets the BC as H[1]=H[end]=0
+    end
+  end
+  return
+end
+
+@testset "conv" begin
+  lx, ly = 10.0, 10.0   # domain size
+  nx, ny = 12, 12     # number of grid points
+  nt     = 1e3          # max number of iterations
+  dx, dy = lx/nx, ly/ny
+  xc, yc = LinRange(dx/2, lx-dx/2, nx), LinRange(dy/2, ly-dy/2, ny)
+  dHdtau = zeros(nx-2, ny-2)
+  dHdtau2 = zeros(nx-2, ny-2)
+  H      = exp.(.-(xc.-lx/2).^2 .-(yc.-ly/2)'.^2)
+  H1     = copy(H)
+  H2     = copy(H)
+  min_dxy2 = min(dx,dy)^2
+  _dx, _dy = 1.0/dx, 1.0/dy
+  compute_update0!(H1, dHdtau, H, min_dxy2, _dx, _dy, dx, dy)
+  compute_update1!(H2, dHdtau2, H, min_dxy2, _dx, _dy, dx, dy)
+  @test H1 ≈ H2
+  @test dHdtau ≈ dHdtau2
+end

test/grouptests.jl

@@ -47,6 +47,8 @@ const START_TIME = time()
     @time include("manyloopreductions.jl")
 
     @time include("simplemisc.jl")
+
+    @time include("convolutions.jl")
   end
 
   @time if LOOPVECTORIZATION_TEST == "all" || LOOPVECTORIZATION_TEST == "part3"