Fixes for nightly

lib/cudadrv/module/linker.jl

@@ -66,29 +66,30 @@ end
 Add PTX code to a pending link operation.
 """
 function add_data!(link::CuLink, name::String, code::String)
-    data = unsafe_wrap(Vector{UInt8}, code)
-
-    # there shouldn't be any embedded NULLs
-    checked_data = Base.unsafe_convert(Cstring, data)
-
-    res = unsafe_cuLinkAddData_v2(link, JIT_INPUT_PTX, pointer(checked_data), length(data),
-                                  name, 0, C_NULL, C_NULL)
-    if res == ERROR_NO_BINARY_FOR_GPU ||
-       res == ERROR_INVALID_IMAGE ||
-       res == ERROR_INVALID_PTX
-        throw(CuError(res, unsafe_string(pointer(link.options[JIT_ERROR_LOG_BUFFER]))))
-    elseif res != SUCCESS
-        throw_api_error(res)
+    GC.@preserve code begin
+        # cuLinkAddData takes a Ptr{Cvoid} instead of a Cstring, because it accepts both
+        # source and binary, so do the conversion (ensuring no embedded NULLs) ourselves
+        data = Base.unsafe_convert(Cstring, code)
+
+        res = unsafe_cuLinkAddData_v2(link, JIT_INPUT_PTX, pointer(data), length(code),
+                                      name, 0, C_NULL, C_NULL)
+        if res == ERROR_NO_BINARY_FOR_GPU ||
+        res == ERROR_INVALID_IMAGE ||
+        res == ERROR_INVALID_PTX
+            throw(CuError(res, unsafe_string(pointer(link.options[JIT_ERROR_LOG_BUFFER]))))
+        elseif res != SUCCESS
+            throw_api_error(res)
+        end
     end
 end
 
 """
-    add_data!(link::CuLink, name::String, data::Vector{UInt8}, type::CUjitInputType)
+    add_data!(link::CuLink, name::String, data::Vector{UInt8})
 
 Add object code to a pending link operation.
 """
 function add_data!(link::CuLink, name::String, data::Vector{UInt8})
-    res = unsafe_cuLinkAddData_v2(link, JIT_INPUT_OBJECT, pointer(data), length(data),
+    res = unsafe_cuLinkAddData_v2(link, JIT_INPUT_OBJECT, data, length(data),
                                   name, 0, C_NULL, C_NULL)
     if res == ERROR_NO_BINARY_FOR_GPU ||
        res == ERROR_INVALID_IMAGE ||
@@ -97,8 +98,6 @@ function add_data!(link::CuLink, name::String, data::Vector{UInt8})
     elseif res != SUCCESS
         throw_api_error(res)
     end
-
-    return
 end
 
 """

src/array.jl

@@ -512,7 +512,14 @@ end
 
 ## memory copying
 
+if VERSION >= v"1.11.0-DEV.753"
+function typetagdata(a::Array, i=1)
+  ptr_or_offset = Int(a.ref.ptr_or_offset)
+  @ccall(jl_genericmemory_typetagdata(a.ref.mem::Any)::Ptr{UInt8}) + ptr_or_offset + i - 1
+end
+else
 typetagdata(a::Array, i=1) = ccall(:jl_array_typetagdata, Ptr{UInt8}, (Any,), a) + i - 1
+end
 function typetagdata(a::CuArray, i=1; type=Mem.Device)
   PT = if type == Mem.Device
     CuPtr{UInt8}

src/compiler/execution.jl

@@ -419,7 +419,8 @@ No keyword arguments are supported.
     DeviceKernel{F,tt}(f, fun, kernel_state())
 end
 
-(kernel::DeviceKernel)(args...; kwargs...) = call(kernel, args...; kwargs...)
+@inline (kernel::DeviceKernel)(args::Vararg{Any,N}; kwargs...) where {N} =
+    call(kernel, args...; kwargs...)
 
 # re-use the parent kernel's seed to avoid need for the RNG
 make_seed(::DeviceKernel) = kernel_state().random_seed

src/device/intrinsics/output.jl

@@ -168,7 +168,7 @@ const cuprint_specifiers = Dict(
             fmt *= string(T.parameters[1])
         else
             @warn("@cuprint does not support values of type $T")
-            fmt *= "$(string(T.parameters[1]))(...)"
+            fmt *= "$(T)(...)"
         end
     end
 

src/indexing.jl

@@ -12,10 +12,13 @@ using Base.Cartesian
 # TODO: it should still be possible to use the same technique;
 #       Base.LogicalIndex basically contains the same as our `findall` here does.
 Base.to_index(::CuArray, I::AbstractArray{Bool}) = findall(I)
-## same for the trailing Array{Bool} optimization (see `_maybe_linear_logical_index` in Base)
-Base.to_indices(A::CuArray, inds,
-                I::Tuple{Union{Array{Bool,N}, BitArray{N}}}) where {N} =
-    (Base.to_index(A, I[1]),)
+if VERSION >= v"1.11.0-DEV.1157"
+    Base.to_indices(A::CuArray, I::Tuple{AbstractArray{Bool}}) = (Base.to_index(A, I[1]),)
+else
+    Base.to_indices(A::CuArray, inds,
+                    I::Tuple{Union{Array{Bool,N}, BitArray{N}}}) where {N} =
+        (Base.to_index(A, I[1]),)
+end
 
 
 ## find*