Use at-inline instead of _inline_meta.

Author: maleadt

Manifest.toml

@@ -77,17 +77,15 @@ version = "0.1.6"
 
 [[GPUArrays]]
 deps = ["Adapt", "LinearAlgebra", "Printf", "Random", "Serialization", "Statistics"]
-git-tree-sha1 = "59aa54826b2667e2a9161f6dbd9e37255fdb541b"
-repo-rev = "e1a4b3d"
-repo-url = "https://github.com/JuliaGPU/GPUArrays.jl.git"
+git-tree-sha1 = "7c39d767a9c55fafd01f7bc8b3fd0adf175fbc97"
 uuid = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
-version = "8.0.2"
+version = "8.1.0"
 
 [[GPUCompiler]]
 deps = ["ExprTools", "InteractiveUtils", "LLVM", "Libdl", "Logging", "TimerOutputs", "UUIDs"]
-git-tree-sha1 = "4ed2616d5e656c8716736b64da86755467f26cf5"
+git-tree-sha1 = "55ea723d032654a52671923fdce9d785e02ed577"
 uuid = "61eb1bfa-7361-4325-ad38-22787b887f55"
-version = "0.12.9"
+version = "0.13.0"
 
 [[InteractiveUtils]]
 deps = ["Markdown"]
@@ -106,9 +104,9 @@ version = "1.3.0"
 
 [[LLVM]]
 deps = ["CEnum", "LLVMExtra_jll", "Libdl", "Printf", "Unicode"]
-git-tree-sha1 = "23a47d417a3cd9c2e73c854bac7dd4731c105ef7"
+git-tree-sha1 = "36d95ecdfbc3240d728f68d73064d5b097fbf2ef"
 uuid = "929cbde3-209d-540e-8aea-75f648917ca0"
-version = "4.4.0"
+version = "4.5.2"
 
 [[LLVMExtra_jll]]
 deps = ["Artifacts", "JLLWrappers", "Libdl", "Pkg"]

Project.toml

@@ -33,7 +33,7 @@ BFloat16s = "0.1"
 CEnum = "0.2, 0.3, 0.4"
 ExprTools = "0.1"
 GPUArrays = "8"
-GPUCompiler = "0.12.6"
+GPUCompiler = "0.13.0"
 LLVM = "4.1.1"
 Random123 = "1.2"
 RandomNumbers = "1.5.3"

lib/cudadrv/execution.jl

@@ -6,14 +6,12 @@ export cudacall
 ## device
 
 # pack arguments in a buffer that CUDA expects
-@generated function pack_arguments(f::Function, args...)
+@inline @generated function pack_arguments(f::Function, args...)
     for arg in args
         isbitstype(arg) || throw(ArgumentError("Arguments to kernel should be bitstype."))
     end
 
-    ex = quote
-        Base.@_inline_meta
-    end
+    ex = quote end
 
     # If f has N parameters, then kernelParams needs to be an array of N pointers.
     # Each of kernelParams[0] through kernelParams[N-1] must point to a region of memory
@@ -78,12 +76,10 @@ end
 
 # convert the argument values to match the kernel's signature (specified by the user)
 # (this mimics `lower-ccall` in julia-syntax.scm)
-@generated function convert_arguments(f::Function, ::Type{tt}, args...) where {tt}
+@inline @generated function convert_arguments(f::Function, ::Type{tt}, args...) where {tt}
     types = tt.parameters
 
-    ex = quote
-        Base.@_inline_meta
-    end
+    ex = quote end
 
     converted_args = Vector{Symbol}(undef, length(args))
     arg_ptrs = Vector{Symbol}(undef, length(args))

lib/cudnn/util.jl

@@ -24,8 +24,7 @@ juliaDataType(a)=(a==CUDNN_DATA_HALF ? Float16 :
 tuple_strides(A::Tuple) = _strides((1,), A)
 _strides(out::Tuple{Int}, A::Tuple{}) = ()
 _strides(out::NTuple{N,Int}, A::NTuple{N}) where {N} = out
-function _strides(out::NTuple{M,Int}, A::Tuple) where M
-    Base.@_inline_meta
+@inline function _strides(out::NTuple{M,Int}, A::Tuple) where M
     _strides((out..., out[M]*A[M]), A)
 end
 

perf/volumerhs.jl

@@ -27,8 +27,7 @@ for (jlf, f) in zip((:+, :*, :-), (:add, :mul, :sub))
         """
         @eval begin
             # the @pure is necessary so that we can constant propagate.
-            Base.@pure function $jlf(a::$T, b::$T)
-                Base.@_inline_meta
+            @inline Base.@pure function $jlf(a::$T, b::$T)
                 Base.llvmcall($ir, $T, Tuple{$T, $T}, a, b)
             end
         end
@@ -46,10 +45,8 @@ let (jlf, f) = (:div_arcp, :div)
         """
         @eval begin
             # the @pure is necessary so that we can constant propagate.
-            Base.@pure function $jlf(a::$T, b::$T)
-                @Base._inline_meta
+            @inline Base.@pure function $jlf(a::$T, b::$T) =
                 Base.llvmcall($ir, $T, Tuple{$T, $T}, a, b)
-            end
         end
     end
     @eval function $jlf(args...)

src/compiler/execution.jl

@@ -174,7 +174,7 @@ The following keyword arguments are supported:
 """
 AbstractKernel
 
-@generated function call(kernel::AbstractKernel{F,TT}, args...; call_kwargs...) where {F,TT}
+@inline @generated function call(kernel::AbstractKernel{F,TT}, args...; call_kwargs...) where {F,TT}
     sig = Tuple{F, TT.parameters...}    # Base.signature_type with a function type
     args = (:(kernel.f), (:( args[$i] ) for i in 1:length(args))...)
 
@@ -197,8 +197,6 @@ AbstractKernel
     call_tt = Base.to_tuple_type(call_t)
 
     quote
-        Base.@_inline_meta
-
         cudacall(kernel.fun, $call_tt, $(call_args...); call_kwargs...)
     end
 end

src/device/intrinsics/dynamic_parallelism.jl

@@ -73,10 +73,9 @@ function launch(f::CuDeviceFunction, args::Vararg{Any,N}; blocks::CuDim=1, threa
     return
 end
 
-@generated function parameter_buffer(f::CuDeviceFunction, blocks, threads, shmem, args...)
+@inline @generated function parameter_buffer(f::CuDeviceFunction, blocks, threads, shmem, args...)
     # allocate a buffer
     ex = quote
-        Base.@_inline_meta
         buf = cudaGetParameterBufferV2(f, blocks, threads, shmem)
         ptr = Base.unsafe_convert(Ptr{UInt32}, buf)
     end

src/device/intrinsics/output.jl

@@ -117,7 +117,7 @@ const cuprint_specifiers = Dict(
     Cstring     => "%s",
 )
 
-@generated function _cuprint(parts...)
+@inline @generated function _cuprint(parts...)
     fmt = ""
     args = Expr[]
 
@@ -170,7 +170,6 @@ const cuprint_specifiers = Dict(
     end
 
     quote
-        Base.@_inline_meta
         @cuprintf($fmt, $(args...))
     end
 end