Merge pull request #183 from omlins/data

Fix Data module type replacement in interactive session

Author: omlins

.github/workflows/ci.yml

@@ -27,7 +27,7 @@ jobs:
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
-      - uses: actions/cache@v1
+      - uses: actions/cache@v4
         env:
           cache-name: cache-artifacts
         with:

Project.toml

@@ -25,7 +25,7 @@ ParallelStencil_MetalExt = "Metal"
 [compat]
 AMDGPU = "0.6, 0.7, 0.8, 0.9, 1"
 CUDA = "3.12, 4, 5"
-CellArrays = "0.3"
+CellArrays = "0.3.2"
 Enzyme = "0.12, 0.13"
 MacroTools = "0.5"
 Metal = "1.2"

src/ParallelKernel/CUDAExt/allocators.jl

@@ -9,7 +9,7 @@ ParallelStencil.ParallelKernel.fill_cuda(::Type{T}, blocklength, args...) where
 
 ParallelStencil.ParallelKernel.zeros_cuda(::Type{T}, blocklength, args...) where {T<:Union{SArray,FieldArray}}    = (check_datatype_cuda(T); fill_cuda(T, blocklength, 0, args...))
 ParallelStencil.ParallelKernel.ones_cuda(::Type{T}, blocklength, args...) where {T<:Union{SArray,FieldArray}}    = (check_datatype_cuda(T); fill_cuda(T, blocklength, 1, args...))
-ParallelStencil.ParallelKernel.rand_cuda(::Type{T}, ::Val{B},    dims)    where {T<:Union{SArray,FieldArray}, B} = (check_datatype_cuda(T, Bool, Enum); blocklen = (B == 0) ? prod(dims) : B; CellArray{T,length(dims),B, CUDA.CuArray{eltype(T),3}}(CUDA.rand(eltype(T), blocklen, prod(size(T)), ceil(Int,prod(dims)/(blocklen))), dims))
+ParallelStencil.ParallelKernel.rand_cuda(::Type{T}, ::Val{B},    dims)    where {T<:Union{SArray,FieldArray}, B} = (check_datatype_cuda(T, Bool, Enum); blocklen = (B == 0) ? prod(dims) : B; CuCellArray{T,length(dims),B, eltype(T)}(CUDA.rand(eltype(T), blocklen, prod(size(T)), ceil(Int,prod(dims)/(blocklen))), dims))
 ParallelStencil.ParallelKernel.rand_cuda(::Type{T}, blocklength, dims...) where {T<:Union{SArray,FieldArray}}    = rand_cuda(T, blocklength, dims)
 ParallelStencil.ParallelKernel.falses_cuda(::Type{T}, blocklength, args...) where {T<:Union{SArray,FieldArray}}    = fill_cuda(T, blocklength, false, args...)
 ParallelStencil.ParallelKernel.trues_cuda(::Type{T}, blocklength, args...) where {T<:Union{SArray,FieldArray}}    = fill_cuda(T, blocklength, true, args...)

src/ParallelKernel/parallel.jl

@@ -388,8 +388,8 @@ function simplify_conditions(caller::Module, expr::Expr)
                 elseif (@capture(cond, a_ <= ixyz_ - c_ <= b_) && ixyz in INDICES) cond = :($a + $c <= $ixyz <= $b + $c)
                 end
                 if @capture(cond, a_ < x_ < b_) || @capture(cond, a_ < x_ <= b_) || @capture(cond, a_ <= x_ < b_) || @capture(cond, a_ <= x_ <= b_)
-                    a_val = eval_try(caller, a)
-                    b_val = eval_try(caller, b)
+                    a_val = eval_try(caller, a; when_interactive=false)
+                    b_val = eval_try(caller, b; when_interactive=false)
                     if !isnothing(a_val) cond = substitute(cond, a, :($a_val), inQuoteNode=true) end
                     if !isnothing(b_val) cond = substitute(cond, b, :($b_val), inQuoteNode=true) end
                 end

src/ParallelKernel/shared.jl

@@ -377,8 +377,8 @@ function eval_arg(caller::Module, arg)
     end
 end
 
-function eval_try(caller::Module, expr)
-    if isinteractive() # NOTE: this is required to avoid that this function returns non-constant values in interactive sessions.
+function eval_try(caller::Module, expr; when_interactive::Bool=true)
+    if !when_interactive && isinteractive() # NOTE: this is required to avoid that this function returns non-constant values in interactive sessions, when not appropriate (e.g. in for optimization)
         return nothing
     else
         try
@@ -562,9 +562,9 @@ end
 
 
 function interpolate(sym::Symbol, vals::NTuple, block::Expr)
-    return quote
+    return flatten(unblock(quote
         $((substitute(block, sym, val; inQuoteNode=true, inString=true) for val in vals)...)
-    end
+    end))
 end
 
 interpolate(sym::Symbol, vals_expr::Expr, block::Expr) = interpolate(sym, (extract_tuple(vals_expr)...,), block)

test/ParallelKernel/test_parallel.jl

@@ -138,7 +138,7 @@ eval(:(
                     @test Array(Ā) ≈ Ā_ref
                     @test Array(B̄) ≈ B̄_ref
                 end
-            end
+            end;
             @testset "@parallel_indices" begin
                 @testset "inbounds" begin
                     expansion = @prettystring(1, @parallel_indices (ix) inbounds=true f(A) = (2*A; return))
@@ -147,27 +147,27 @@ eval(:(
                     @test !occursin("Base.@inbounds begin", expansion)
                     expansion = @prettystring(1, @parallel_indices (ix) f(A) = (2*A; return))
                     @test !occursin("Base.@inbounds begin", expansion)
-                end
+                end;
                 @testset "addition of range arguments" begin
                     expansion = @gorgeousstring(1, @parallel_indices (ix,iy) f(a::T, b::T) where T <: Union{Array{Float32}, Array{Float64}} = (println("a=$a, b=$b)"); return))
                     @test occursin("f(a::T, b::T, ranges::Tuple{UnitRange, UnitRange, UnitRange}, rangelength_x::Int64, rangelength_y::Int64, rangelength_z::Int64", expansion)
-                end
-                $(interpolate(:__T__, ARRAYTYPES, :(
+                end;
+                @testset "Data.T to Data.Device.T" $(interpolate(:__T__, ARRAYTYPES, :(
                     @testset "Data.__T__ to Data.Device.__T__" begin
                         @static if @isgpu($package)
                             expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::Data.__T__, B::Data.__T__, c::T) where T <: Integer = (A[ix,iy] = B[ix,iy]^c; return))
                             @test occursin("f(A::Data.Device.__T__, B::Data.Device.__T__,", expansion)
                         end
-                    end
-                )))
-                $(interpolate(:__T__, FIELDTYPES, :(
+                    end;
+                )));
+                @testset "Data.Fields.T to Data.Fields.Device.T" $(interpolate(:__T__, FIELDTYPES, :(
                     @testset "Data.Fields.__T__ to Data.Fields.Device.__T__" begin
                         @static if @isgpu($package)
                             expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::Data.Fields.__T__, B::Data.Fields.__T__, c::T) where T <: Integer = (A[ix,iy] = B[ix,iy]^c; return))
                             @test occursin("f(A::Data.Fields.Device.__T__, B::Data.Fields.Device.__T__,", expansion)
                         end
-                    end
-                )))
+                    end;
+                )));
                 # NOTE: the following GPU tests fail, because the Fields module cannot be imported.
                 # @testset "Fields.Field to Data.Fields.Device.Field" begin
                 #     @static if @isgpu($package)
@@ -183,22 +183,22 @@ eval(:(
                 #             @test occursin("f(A::Data.Fields.Device.Field, B::Data.Fields.Device.Field,", expansion)
                 #     end
                 # end
-                $(interpolate(:__T__, ARRAYTYPES, :(
+                @testset "TData.T to TData.Device.T" $(interpolate(:__T__, ARRAYTYPES, :(
                     @testset "TData.__T__ to TData.Device.__T__" begin
                         @static if @isgpu($package)
                             expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::TData.__T__, B::TData.__T__, c::T) where T <: Integer = (A[ix,iy] = B[ix,iy]^c; return))
                             @test occursin("f(A::TData.Device.__T__, B::TData.Device.__T__,", expansion)
                         end
-                    end
-                )))
-                $(interpolate(:__T__, FIELDTYPES, :(
+                    end;
+                )));
+                @testset "TData.Fields.T to TData.Fields.Device.T" $(interpolate(:__T__, FIELDTYPES, :(
                     @testset "TData.Fields.__T__ to TData.Fields.Device.__T__" begin
                         @static if @isgpu($package)
                             expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::TData.Fields.__T__, B::TData.Fields.__T__, c::T) where T <: Integer = (A[ix,iy] = B[ix,iy]^c; return))
                             @test occursin("f(A::TData.Fields.Device.__T__, B::TData.Fields.Device.__T__,", expansion)
                         end
-                    end
-                )))
+                    end;
+                )));
                 # NOTE: the following GPU tests fail, because the Fields module cannot be imported.
                 # @testset "Fields.Field to TData.Fields.Device.Field" begin
                 #     @static if @isgpu($package)
@@ -214,14 +214,14 @@ eval(:(
                 #             @test occursin("f(A::TData.Fields.Device.Field, B::TData.Fields.Device.Field,", expansion)
                 #     end
                 # end
-                $(interpolate(:__T__, ARRAYTYPES, :(
+                @testset "Nested Data.T to Data.Device.T" $(interpolate(:__T__, ARRAYTYPES, :(
                     @testset "Nested Data.__T__ to Data.Device.__T__" begin
                         @static if @isgpu($package)
                             expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::NamedTuple{T1, NTuple{T2,T3}} where {T1,T2} where T3 <: Data.__T__, c::T) where T <: Integer = (A[ix,iy] = B[ix,iy]^c; return))
                             @test occursin("f(A::((NamedTuple{T1, NTuple{T2, T3}} where {T1, T2}) where T3 <: Data.Device.__T__),", expansion)
                         end
-                    end
-                )))
+                    end;
+                )));
                 @testset "@parallel_indices (1D)" begin
                     A  = @zeros(4)
                     @parallel_indices (ix) function write_indices!(A)
@@ -422,22 +422,22 @@ eval(:(
             @require !@is_initialized()
             @init_parallel_kernel(package = $package)
             @require @is_initialized
-            $(interpolate(:__T__, ARRAYTYPES, :(
+            @testset "Data.T{T2} to Data.Device.T{T2}" $(interpolate(:__T__, ARRAYTYPES, :(
                 @testset "Data.__T__{T2} to Data.Device.__T__{T2}" begin
                     @static if @isgpu($package)
                         expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::Data.__T__{T2}, B::Data.__T__{T2}, c<:Integer) where T2 <: Union{Float32, Float64}  = (A[ix,iy] = B[ix,iy]^c; return))
                         @test occursin("f(A::Data.Device.__T__{T2}, B::Data.Device.__T__{T2},", expansion)
                     end
                 end;
-            )))
-            $(interpolate(:__T__, FIELDTYPES, :(
+            )));
+            @testset "Data.Fields.T{T2} to Data.Fields.Device.T{T2}" $(interpolate(:__T__, FIELDTYPES, :(
                 @testset "Data.Fields.__T__{T2} to Data.Fields.Device.__T__{T2}" begin
                     @static if @isgpu($package)
                         expansion = @prettystring(1, @parallel_indices (ix,iy) f(A::Data.Fields.__T__{T2}, B::Data.Fields.__T__{T2}, c<:Integer) where T2 <: Union{Float32, Float64}  = (A[ix,iy] = B[ix,iy]^c; return))
                         @test occursin("f(A::Data.Fields.Device.__T__{T2}, B::Data.Fields.Device.__T__{T2},", expansion)
                     end
                 end;
-            )))
+            )));
             @reset_parallel_kernel()
         end;
         @testset "5. Exceptions" begin