Launch interface

Author: christiangnrd

src/KernelAbstractions.jl

@@ -244,10 +244,10 @@ Declare storage that is local to a workgroup.
 """
 macro localmem(T, dims)
     # Stay in sync with CUDAnative
-    id = gensym("static_shmem")
+    # id = gensym("static_shmem")
 
     return quote
-        $SharedMemory($(esc(T)), Val($(esc(dims))), Val($(QuoteNode(id))))
+        $SharedMemory($(esc(T)), Val($(esc(dims))))#, Val($(QuoteNode(id))))
     end
 end
 
@@ -814,7 +814,8 @@ include("macros.jl")
 ###
 
 function Scratchpad end
-SharedMemory(t::Type{T}, dims::Val{Dims}, id::Val{Id}) where {T, Dims, Id} = KernelIntrinsics.localmemory(t, dims, id)
+# SharedMemory(t::Type{T}, dims::Val{Dims}, id::Val{Id}) where {T, Dims, Id} = KernelIntrinsics.localmemory(t, dims, id)
+SharedMemory(t::Type{T}, dims::Val{Dims}) where {T, Dims} = KernelIntrinsics.localmemory(t, dims)
 
 __synchronize() = KernelIntrinsics.barrier()
 

src/intrinsics.jl

@@ -51,10 +51,47 @@ Returns the unique group ID.
 """
 function get_group_id end
 
-function localmemory end
+"""
+    localmemory(T, dims)
+
+Declare memory that is local to a workgroup.
+
+!!! note
+    Backend implementations **must** implement:
+    ```
+    localmemory(T::DataType, ::Val{Dims}) where {T, Dims}
+    ```
+    As well as the on-device functionality.
+"""
+localmemory(::Type{T}, dims) where T = localmemory(T, Val(dims))
+# @inline localmemory(::Type{T}, dims::Val{Dims}) where {T, Dims} = localmemory(T, dims, Val(gensym("static_shmem")))
+
 function barrier()
     error("Group barrier used outside kernel or not captured")
 end
 function print end
 
+
+"""
+    KIKernel{Backend, BKern}
+
+KIKernel closure struct that is used to represent the backend
+kernel on the host.
+
+!!! note
+    Backend implementations **must** implement:
+    ```
+    KI.KIKernel(::NewBackend, f, args...; kwargs...)
+    (kernel::KIKernel{<:NewBackend})(args...; numworkgroups=nothing, workgroupsize=nothing)
+    ```
+    As well as the on-device functionality.
+"""
+struct KIKernel{Backend, BKern}
+    backend::Backend
+    kern::BKern
+end
+
+function kernel_max_work_group_size end
+function max_work_group_size end
+function multiprocessor_count end
 end

src/pocl/backend.jl

@@ -178,7 +178,7 @@ end
 
 ## Shared and Scratch Memory
 
-@device_override @inline function KI.localmemory(::Type{T}, ::Val{Dims}, ::Val{Id}) where {T, Dims, Id}
+@device_override @inline function KI.localmemory(::Type{T}, ::Val{Dims}) where {T, Dims}
     ptr = POCL.emit_localmemory(T, Val(prod(Dims)))
     CLDeviceArray(Dims, ptr)
 end