Add support for Type arguments and fix static parameter codegen

ext/CUDAExt.jl

@@ -257,6 +257,7 @@ Other values pass through unchanged.
 """
 to_tile_arg(x) = x
 to_tile_arg(arr::AbstractArray) = TileArray(arr)
+to_tile_arg(t::Type) = Constant(t)
 
 # Tiled Broadcast — TiledStyle wins over CuArrayStyle
 BroadcastStyle(::cuTile.TiledStyle{N}, ::CuArrayStyle{M}) where {N,M} = cuTile.TiledStyle{max(N,M)}()

src/compiler/codegen/expressions.jl

@@ -20,6 +20,14 @@ function emit_expr!(ctx::CGCtx, expr::Expr, @nospecialize(result_type))
         # Bounds checking is always disabled in Tile IR kernels.
         # Emit false so IfOps referencing this SSA can resolve the condition.
         return emit_constant!(ctx, false, Bool)
+    elseif expr.head === :static_parameter
+        # Static type parameter reference (e.g., V in `f(::T{V}) where {V}`).
+        # Look up the concrete value from the method's sptypes.
+        idx = expr.args[1]::Int
+        sp = ctx.sci.sptypes[idx]
+        sptyp = sp isa CC.VarState ? sp.typ : sp
+        val = sptyp isa CC.Const ? sptyp.val : CC.widenconst(sptyp)
+        return emit_value!(ctx, val)
     elseif expr.head === :code_coverage_effect
         return nothing
     else

src/compiler/passes/canonicalize.jl

@@ -126,13 +126,15 @@ function scalar_elim_block!(block::Block)
 
         current_type = value_type(inst)
         current_type === nothing && continue
+        is_token_type(current_type) && continue
         T = CC.widenconst(current_type)
         T <: Tile && continue     # already tile-typed
         T <: Number || continue   # only promote scalar number types
 
         for op in ops
             op_type = value_type(block, op)
             op_type === nothing && continue
+            is_token_type(op_type) && continue
             OT = CC.widenconst(op_type)
             OT <: Tile || continue
             S = OT.parameters[2]
@@ -162,6 +164,7 @@ function scalar_elim_block!(block::Block)
     for inst in instructions(block)
         current_type = value_type(inst)
         current_type === nothing && continue
+        is_token_type(current_type) && continue
         new_type = promote_scalar_type(CC.widenconst(current_type))
         new_type === nothing && continue
         update_type!(block, inst, new_type)
@@ -170,6 +173,7 @@ function scalar_elim_block!(block::Block)
     # Phase 5: Promote block argument types (loop IVs, carries).
     # BlockArgument is immutable, so we create a new one and replace all uses.
     for (i, arg) in enumerate(block.args)
+        is_token_type(arg.type) && continue
         T = CC.widenconst(arg.type)
         T <: Tile && continue
         T <: Number || continue

src/compiler/reflection.jl

@@ -82,15 +82,21 @@ Returns `(stripped, nothing)` when no Constant types are present.
 function process_const_argtypes(@nospecialize(f), @nospecialize(argtypes))
     params = argtypes isa DataType ? argtypes.parameters :
              argtypes isa Tuple ? argtypes : fieldtypes(argtypes)
-    has_consts = any(T -> T <: Constant, params)
+    has_consts = any(T -> T <: Constant || CC.isconstType(T), params)
     stripped_params = map(params) do T
         T <: Constant ? constant_eltype(T) : T
     end
     stripped = Tuple{stripped_params...}
     const_argtypes = if has_consts
         cats = Any[CC.Const(f)]
         for T in params
-            push!(cats, T <: Constant ? CC.Const(constant_value(T)) : T)
+            if T <: Constant
+                push!(cats, CC.Const(constant_value(T)))
+            elseif CC.isconstType(T)
+                push!(cats, CC.Const(T.parameters[1]))
+            else
+                push!(cats, T)
+            end
         end
         cats
     else

src/compiler/utils.jl

@@ -77,7 +77,7 @@ end
 # walk_uses! extensions so that IRStructurizer's uses()/replace_uses! see
 # operands inside cuTile-specific IR nodes.
 IRStructurizer.walk_uses!(f, node::JoinTokensNode) =
-    for i in 1:length(node.tokens); f(IndexedUseRef(node.tokens, i)); end
+    for i in 1:length(node.tokens); f(IRStructurizer.IndexedUseRef(node.tokens, i)); end
 IRStructurizer.walk_uses!(f, ::TokenResultNode) = nothing
 IRStructurizer.walk_uses!(f, ::MakeTokenNode) = nothing
 

src/language/types.jl

@@ -316,8 +316,9 @@ argtypes = Tuple{Ptr{Float32}, Constant{Int, 16}}
 """
 struct Constant{T, V} end
 
-# Convenience constructor that infers type from value
+# Convenience constructors that infer type from value
 Constant(val::T) where {T} = Constant{T, val}()
+Constant(val::Type{T}) where {T} = Constant{Type{T}, T}()
 
 # Extract constant value - @inline ensures this folds to a constant in IR
 @inline Base.getindex(::Constant{T, V}) where {T, V} = V

test/codegen/integration.jl

@@ -1262,6 +1262,33 @@ end
     end
 end
 
+#=============================================================================
+ Constant Type Arguments
+=============================================================================#
+
+@testset "Constant Type Arguments" begin
+    spec = ct.ArraySpec{1}(16, true)
+
+    function _type_param_kernel(a, b, tile_size::Int, ::Type{T}) where T
+        pid = ct.bid(1)
+        tile = ct.load(a, pid, (tile_size,)) + zeros(T, (tile_size,))
+        ct.store(b, pid, tile)
+        return
+    end
+
+    @testset "Type parameter used in kernel body" begin
+        @test @filecheck begin
+            @check_label "entry"
+            @check "load_view_tko"
+            @check "addf"
+            @check "store_view_tko"
+            code_tiled(_type_param_kernel,
+                       Tuple{ct.TileArray{Float32,1,spec}, ct.TileArray{Float32,1,spec},
+                             ct.Constant{Int,16}, Type{Float32}})
+        end
+    end
+end
+
 #=============================================================================
  For Loops
 =============================================================================#

test/codegen/reflection.jl

@@ -104,3 +104,30 @@ end
         end
     end
 end
+
+@testset "Type args" begin
+    const_spec = ct.ArraySpec{1}(128, true, (0,), (32,))
+
+    @test ct.Constant(Int) isa ct.Constant{Type{Int}, Int}
+
+    @testset "code_tiled with Type parameter" begin
+        function reflect_type_param(a, b, c, tile_size::Int, ::Type{T}) where T
+            pid = ct.bid(1)
+            tile_a = ct.load(a; index=pid, shape=(tile_size,))
+            tile_b = ct.load(b; index=pid, shape=(tile_size,))
+            ct.store(c; index=pid, tile=tile_a + tile_b + zeros(T, (tile_size,)))
+            return
+        end
+
+        ConstTypeTT = Tuple{ct.TileArray{Float32,1,const_spec}, ct.TileArray{Float32,1,const_spec},
+                            ct.TileArray{Float32,1,const_spec}, ct.Constant{Int64, 16},
+                            Type{Float32}}
+
+        @test @filecheck begin
+            @check "load_view_tko"
+            @check "addf"
+            @check "store_view_tko"
+            ct.code_tiled(reflect_type_param, ConstTypeTT)
+        end
+    end
+end

test/device/core.jl

@@ -387,6 +387,26 @@ end
     @test Array(b) ≈ Array(a)
 end
 
+@testset "Type parameter (auto-wrapped)" begin
+    function vadd_type_param(a, b, c, tile_size::Int, ::Type{T}) where T
+        pid = ct.bid(1)
+        tile_a = ct.load(a; index=pid, shape=(tile_size,))
+        tile_b = ct.load(b; index=pid, shape=(tile_size,))
+        ct.store(c; index=pid, tile=T.(tile_a) + T.(tile_b))
+        return
+    end
+
+    n = 1024
+    tile_size = 32
+    a = CUDA.rand(Float16, n)
+    b = CUDA.rand(Float16, n)
+    c = CUDA.zeros(Float32, n)
+
+    ct.launch(vadd_type_param, cld(n, tile_size), a, b, c, ct.Constant(tile_size), Float32)
+
+    @test Array(c) ≈ Float32.(Array(a)) + Float32.(Array(b))
+end
+
 end
 
 @testset "TileArray auto-conversion" begin