调用 C 和 Fortran 代码

在数值计算领域,尽管有很多用 C 语言或 Fortran 写的高质量且成熟的库都可以用 Julia 重写,但为了便捷利用现有的 C 或 Fortran 代码,Julia 提供简洁且高效的调用方式。Julia 的哲学是 no boilerplate: Julia 可以直接调用 C/Fortran 的函数,不需要任何”胶水”代码,代码生成或其它编译过程 – 即使在交互式会话 (REPL/Jupyter notebook) 中使用也一样. 在 Julia 中,上述特性可以仅仅通过调用 ccall 实现,它的语法看起来就像是普通的函数调用。

The code to be called must be available as a shared library. Most C and Fortran libraries ship compiled as shared libraries already, but if you are compiling the code yourself using GCC (or Clang), you will need to use the -shared and -fPIC options. The machine instructions generated by Julia’s JIT are the same as a native C call would be, so the resulting overhead is the same as calling a library function from C code. [1]

Shared libraries and functions are referenced by a tuple of the form (:function, "library") or ("function", "library") where function is the C-exported function name, and library refers to the shared library name. Shared libraries available in the (platform-specific) load path will be resolved by name. The full path to the library may also be specified.

可以单独使用函数名来代替元组(只用 :function"function")。在这种情况下,函数名在当前进程中进行解析。这一调用形式可用于调用 C 库函数、Julia 运行时中的函数或链接到 Julia 的应用程序中的函数。

默认情况下,Fortran 编译器会进行名称修饰(例如,将函数名转换为小写或大写,通常会添加下划线),要通过 ccall 调用 Fortran 函数,传递的标识符必须与 Fortran 编译器名称修饰之后的一致。此外,在调用 Fortran 函数时,所有输入必须以指针形式传递,并已在堆或栈上分配内存。这不仅适用于通常是堆分配的数组及可变对象,而且适用于整数和浮点数等标量值,尽管这些值通常是栈分配的,且在使用 C 或 Julia 调用约定时通常是通过寄存器传递的。

Finally, you can use ccall to actually generate a call to the library function. The arguments to ccall are:

  1. A (:function, "library") pair (most common),

    a :function name symbol or "function" name string (for symbols in the current process or libc),

    一个函数指针(例如,从 dlsym 获得的指针)。

  2. The function’s return type

  3. A tuple of input types, corresponding to the function signature

  4. The actual argument values to be passed to the function, if any; each is a separate parameter.

Note

The (:function, "library") pair, return type, and input types must be literal constants (i.e., they can’t be variables, but see Non-constant Function Specifications below).

The remaining parameters are evaluated at compile time, when the containing method is defined.

Note

See below for how to map C types to Julia types.

As a complete but simple example, the following calls the clock function from the standard C library on most Unix-derived systems:

  1. julia> t = ccall(:clock, Int32, ())
  2. 2292761
  3. julia> t
  4. 2292761
  5. julia> typeof(ans)
  6. Int32

clock takes no arguments and returns an Int32. One common mistake is forgetting that a 1-tuple of argument types must be written with a trailing comma. For example, to call the getenv function to get a pointer to the value of an environment variable, one makes a call like this:

  1. julia> path = ccall(:getenv, Cstring, (Cstring,), "SHELL")
  2. Cstring(@0x00007fff5fbffc45)
  3. julia> unsafe_string(path)
  4. "/bin/bash"

Note that the argument type tuple must be written as (Cstring,), not (Cstring). This is because (Cstring) is just the expression Cstring surrounded by parentheses, rather than a 1-tuple containing Cstring:

  1. julia> (Cstring)
  2. Cstring
  3. julia> (Cstring,)
  4. (Cstring,)

In practice, especially when providing reusable functionality, one generally wraps ccall uses in Julia functions that set up arguments and then check for errors in whatever manner the C or Fortran function specifies. And if an error occurs it is thrown as a normal Julia exception. This is especially important since C and Fortran APIs are notoriously inconsistent about how they indicate error conditions. For example, the getenv C library function is wrapped in the following Julia function, which is a simplified version of the actual definition from env.jl:

  1. function getenv(var::AbstractString)
  2. val = ccall(:getenv, Cstring, (Cstring,), var)
  3. if val == C_NULL
  4. error("getenv: undefined variable: ", var)
  5. end
  6. return unsafe_string(val)
  7. end

C 函数 getenv 通过返回 NULL 的方式进行报错,但是其他 C 标准库函数也会通过多种不同的方式来报错,这包括返回 -1,0,1 以及其它特殊值。此封装能够明确地抛出异常信息,即是否调用者在尝试获取一个不存在的环境变量:

  1. julia> getenv("SHELL")
  2. "/bin/bash"
  3. julia> getenv("FOOBAR")
  4. getenv: undefined variable: FOOBAR

Here is a slightly more complex example that discovers the local machine’s hostname. In this example, the networking library code is assumed to be in a shared library named “libc”. In practice, this function is usually part of the C standard library, and so the “libc” portion should be omitted, but we wish to show here the usage of this syntax.

  1. function gethostname()
  2. hostname = Vector{UInt8}(undef, 256) # MAXHOSTNAMELEN
  3. err = ccall((:gethostname, "libc"), Int32,
  4. (Ptr{UInt8}, Csize_t),
  5. hostname, sizeof(hostname))
  6. Base.systemerror("gethostname", err != 0)
  7. hostname[end] = 0 # ensure null-termination
  8. return unsafe_string(pointer(hostname))
  9. end

This example first allocates an array of bytes. It then calls the C library function gethostname to populate the array with the hostname. Finally, it takes a pointer to the hostname buffer, and converts the pointer to a Julia string, assuming that it is a NUL-terminated C string. It is common for C libraries to use this pattern of requiring the caller to allocate memory to be passed to the callee and populated. Allocation of memory from Julia like this is generally accomplished by creating an uninitialized array and passing a pointer to its data to the C function. This is why we don’t use the Cstring type here: as the array is uninitialized, it could contain NUL bytes. Converting to a Cstring as part of the ccall checks for contained NUL bytes and could therefore throw a conversion error.

创建和C兼容的Julia函数指针

可以将Julia函数传递给接受函数指针参数的原生C函数。例如,要匹配满足下面的C原型:

  1. typedef returntype (*functiontype)(argumenttype, ...)

The macro @cfunction generates the C-compatible function pointer for a call to a Julia function. The arguments to @cfunction are:

  1. A Julia function
  2. The function’s return type
  3. A tuple of input types, corresponding to the function signature

Note

As with ccall, the return type and tuple of input types must be literal constants.

Note

Currently, only the platform-default C calling convention is supported. This means that @cfunction-generated pointers cannot be used in calls where WINAPI expects a stdcall function on 32-bit Windows, but can be used on WIN64 (where stdcall is unified with the C calling convention).

一个典型的例子就是标准C库函数qsort,定义为:

  1. void qsort(void *base, size_t nmemb, size_t size,
  2. int (*compare)(const void*, const void*));

The base argument is a pointer to an array of length nmemb, with elements of size bytes each. compare is a callback function which takes pointers to two elements a and b and returns an integer less/greater than zero if a should appear before/after b (or zero if any order is permitted).

Now, suppose that we have a 1-d array A of values in Julia that we want to sort using the qsort function (rather than Julia’s built-in sort function). Before we consider calling qsort and passing arguments, we need to write a comparison function:

  1. julia> function mycompare(a, b)::Cint
  2. return (a < b) ? -1 : ((a > b) ? +1 : 0)
  3. end
  4. mycompare (generic function with 1 method)

qsort expects a comparison function that return a C int, so we annotate the return type to be Cint.

In order to pass this function to C, we obtain its address using the macro @cfunction:

  1. julia> mycompare_c = @cfunction(mycompare, Cint, (Ref{Cdouble}, Ref{Cdouble}));

@cfunction 需要三个参数: Julia函数 (mycompare), 返回值类型(Cint), 和一个输入参数类型的值元组, 此处是要排序的Cdouble(Float64) 元素的数组.

qsort的最终调用看起来是这样的:

  1. julia> A = [1.3, -2.7, 4.4, 3.1]
  2. 4-element Array{Float64,1}:
  3. 1.3
  4. -2.7
  5. 4.4
  6. 3.1
  7. julia> ccall(:qsort, Cvoid, (Ptr{Cdouble}, Csize_t, Csize_t, Ptr{Cvoid}),
  8. A, length(A), sizeof(eltype(A)), mycompare_c)
  9. julia> A
  10. 4-element Array{Float64,1}:
  11. -2.7
  12. 1.3
  13. 3.1
  14. 4.4

As the example shows, the original Julia array A has now been sorted: [-2.7, 1.3, 3.1, 4.4]. Note that Julia takes care of converting the array to a Ptr{Cdouble}), computing the size of the element type in bytes, and so on.

For fun, try inserting a println("mycompare($a, $b)") line into mycompare, which will allow you to see the comparisons that qsort is performing (and to verify that it is really calling the Julia function that you passed to it).

Mapping C Types to Julia

It is critical to exactly match the declared C type with its declaration in Julia. Inconsistencies can cause code that works correctly on one system to fail or produce indeterminate results on a different system.

Note that no C header files are used anywhere in the process of calling C functions: you are responsible for making sure that your Julia types and call signatures accurately reflect those in the C header file.[2]

Automatic Type Conversion

Julia automatically inserts calls to the Base.cconvert function to convert each argument to the specified type. For example, the following call:

  1. ccall((:foo, "libfoo"), Cvoid, (Int32, Float64), x, y)

will behave as if it were written like this:

  1. ccall((:foo, "libfoo"), Cvoid, (Int32, Float64),
  2. Base.unsafe_convert(Int32, Base.cconvert(Int32, x)),
  3. Base.unsafe_convert(Float64, Base.cconvert(Float64, y)))

Base.cconvert normally just calls convert, but can be defined to return an arbitrary new object more appropriate for passing to C. This should be used to perform all allocations of memory that will be accessed by the C code. For example, this is used to convert an Array of objects (e.g. strings) to an array of pointers.

Base.unsafe_convert handles conversion to Ptr types. It is considered unsafe because converting an object to a native pointer can hide the object from the garbage collector, causing it to be freed prematurely.

Type Correspondences

First, let’s review some relevant Julia type terminology:

语法 / 关键字例子描述
mutable structBitSetLeaf Type:包含 type-tag 的一组相关数据,由 Julia GC 管理,通过 object-identity 来定义。为了保证实例可以被构造,Leaf Type 必须是完整定义的,即不允许使用 TypeVars
abstract typeAny, AbstractArray{T, N}, Complex{T}Super Type:用于描述一组类型,它不是 Leaf-Type,也无法被实例化。
T{A}Vector{Int}Type Parameter:某种类型的一种具体化,通常用于分派或存储优化。
TypeVarType parameter 声明中的 T 是一个 TypeVar,它是类型变量的简称。
primitive typeInt, Float64Primitive Type:一种没有成员变量的类型,但是它有大小。It is stored and defined by-value.
structPair{Int, Int}“Struct” :: A type with all fields defined to be constant. It is defined by-value, and may be stored with a type-tag.
ComplexF64 (isbits)“Is-Bits” :: A primitive type, or a struct type where all fields are other isbits types. It is defined by-value, and is stored without a type-tag.
struct …; endnothingSingleton:没有成员变量的 Leaf TypeStruct
(…) or tuple(…)(1, 2, 3)“Tuple” :: an immutable data-structure similar to an anonymous struct type, or a constant array. Represented as either an array or a struct.

Bits Types

There are several special types to be aware of, as no other type can be defined to behave the same:

  • Float32

    和C语言中的 float 类型完全对应(以及Fortran中的 REAL*4

  • Float64

    和C语言中的 double 类型完全对应(以及Fortran中的 REAL*8

  • ComplexF32

    和C语言中的 complex float 类型完全对应(以及Fortran中的 COMPLEX*8

  • ComplexF64

    和C语言中的 complex double 类型完全对应(以及Fortran中的 COMPLEX*16

  • Signed

    和C语言中的 signed 类型标识完全对应(以及Fortran中的任意 INTEGER 类型) Julia中任何不是Signed 的子类型的类型,都会被认为是unsigned类型。

  • Ref{T}

    Ptr{T} 行为相同,能通过Julia的GC管理其内存。

  • Array{T,N}

    When an array is passed to C as a Ptr{T} argument, it is not reinterpret-cast: Julia requires that the element type of the array matches T, and the address of the first element is passed.

    因此,如果一个 Array 中的数据格式不正确,它必须被显式地转换 ,通过类似 trunc(Int32, a) 的函数。

    若要将一个数组 A 以不同类型的指针传递,而不提前转换数据, (比如,将一个 Float64 数组传给一个处理原生字节的函数时),你 可以将这一参数声明为 Ptr{Cvoid}

    如果一个元素类型为 Ptr{T} 的数组作为 Ptr{Ptr{T}} 类型的参数传递, Base.cconvert 将会首先尝试进行 null-terminated copy(即直到下一个元素为null才停止复制),并将每一个元素使用其通过 Base.cconvert 转换后的版本替换。 这允许,比如,将一个 argv 的指针数组,其类型为 Vector{String} ,传递给一个类型为 Ptr{Ptr{Cchar}} 的参数。

On all systems we currently support, basic C/C++ value types may be translated to Julia types as follows. Every C type also has a corresponding Julia type with the same name, prefixed by C. This can help when writing portable code (and remembering that an int in C is not the same as an Int in Julia).

System Independent Types

C 类型Fortran 类型标准 Julia 别名Julia 基本类型
unsigned charCHARACTERCucharUInt8
bool (_Bool in C99+)CucharUInt8
shortINTEGER2, LOGICAL2CshortInt16
unsigned shortCushortUInt16
int, BOOL (C, typical)INTEGER4, LOGICAL4CintInt32
unsigned intCuintUInt32
long longINTEGER8, LOGICAL8ClonglongInt64
unsigned long longCulonglongUInt64
intmax_tCintmax_tInt64
uintmax_tCuintmax_tUInt64
floatREAL4iCfloatFloat32
doubleREAL8CdoubleFloat64
complex floatCOMPLEX8ComplexF32Complex{Float32}
complex doubleCOMPLEX16ComplexF64Complex{Float64}
ptrdiff_tCptrdiff_tInt
ssize_tCssize_tInt
size_tCsize_tUInt
voidCvoid
void and [[noreturn]] or _NoreturnUnion{}
voidPtr{Cvoid}
T (where T represents an appropriately defined type)Ref{T}
char (or char[], e.g. a string)CHARACTERNCstring if NUL-terminated, or Ptr{UInt8} if not
char (or char[])Ptr{Ptr{UInt8}}
jl_value_t (any Julia Type)Any
jl_value_t (a reference to a Julia Type)Ref{Any}
va_argNot supported
(variadic function specification)T… (where T is one of the above types, variadic functions of different argument types are not supported)

The Cstring type is essentially a synonym for Ptr{UInt8}, except the conversion to Cstring throws an error if the Julia string contains any embedded NUL characters (which would cause the string to be silently truncated if the C routine treats NUL as the terminator). If you are passing a char* to a C routine that does not assume NUL termination (e.g. because you pass an explicit string length), or if you know for certain that your Julia string does not contain NUL and want to skip the check, you can use Ptr{UInt8} as the argument type. Cstring can also be used as the ccall return type, but in that case it obviously does not introduce any extra checks and is only meant to improve readability of the call.

System Dependent Types

C 类型标准 Julia 别名Julia 基本类型
charCcharInt8 (x86, x86_64), UInt8 (powerpc, arm)
longClongInt (UNIX), Int32 (Windows)
unsigned longCulongUInt (UNIX), UInt32 (Windows)
wchar_tCwchar_tInt32 (UNIX), UInt16 (Windows)

Note

When calling Fortran, all inputs must be passed by pointers to heap- or stack-allocated values, so all type correspondences above should contain an additional Ptr{..} or Ref{..} wrapper around their type specification.

Warning

For string arguments (char*) the Julia type should be Cstring (if NUL- terminated data is expected), or either Ptr{Cchar} or Ptr{UInt8} otherwise (these two pointer types have the same effect), as described above, not String. Similarly, for array arguments (T[] or T*), the Julia type should again be Ptr{T}, not Vector{T}.

Warning

Julia’s Char type is 32 bits, which is not the same as the wide character type (wchar_t or wint_t) on all platforms.

Warning

A return type of Union{} means the function will not return, i.e., C++11 [[noreturn]] or C11 _Noreturn (e.g. jl_throw or longjmp). Do not use this for functions that return no value (void) but do return, use Cvoid instead.

Note

For wchar_t* arguments, the Julia type should be Cwstring (if the C routine expects a NUL-terminated string), or Ptr{Cwchar_t} otherwise. Note also that UTF-8 string data in Julia is internally NUL-terminated, so it can be passed to C functions expecting NUL-terminated data without making a copy (but using the Cwstring type will cause an error to be thrown if the string itself contains NUL characters).

Note

C functions that take an argument of type char** can be called by using a Ptr{Ptr{UInt8}} type within Julia. For example, C functions of the form:

  1. int main(int argc, char **argv);

can be called via the following Julia code:

  1. argv = [ "a.out", "arg1", "arg2" ]
  2. ccall(:main, Int32, (Int32, Ptr{Ptr{UInt8}}), length(argv), argv)

Note

For Fortran functions taking variable length strings of type character(len=*) the string lengths are provided as hidden arguments. Type and position of these arguments in the list are compiler specific, where compiler vendors usually default to using Csize_t as type and append the hidden arguments at the end of the argument list. While this behaviour is fixed for some compilers (GNU), others optionally permit placing hidden arguments directly after the character argument (Intel, PGI). For example, Fortran subroutines of the form

  1. subroutine test(str1, str2)
  2. character(len=*) :: str1,str2

can be called via the following Julia code, where the lengths are appended

  1. str1 = "foo"
  2. str2 = "bar"
  3. ccall(:test, Cvoid, (Ptr{UInt8}, Ptr{UInt8}, Csize_t, Csize_t),
  4. str1, str2, sizeof(str1), sizeof(str2))

Warning

Fortran compilers may also add other hidden arguments for pointers, assumed-shape (:) and assumed-size (*) arrays. Such behaviour can be avoided by using ISO_C_BINDING and including bind(c) in the definition of the subroutine, which is strongly recommended for interoperable code. In this case there will be no hidden arguments, at the cost of some language features (e.g. only character(len=1) will be permitted to pass strings).

Note

A C function declared to return Cvoid will return the value nothing in Julia.

Struct Type Correspondences

Composite types such as struct in C or TYPE in Fortran90 (or STRUCTURE / RECORD in some variants of F77), can be mirrored in Julia by creating a struct definition with the same field layout.

When used recursively, isbits types are stored inline. All other types are stored as a pointer to the data. When mirroring a struct used by-value inside another struct in C, it is imperative that you do not attempt to manually copy the fields over, as this will not preserve the correct field alignment. Instead, declare an isbits struct type and use that instead. Unnamed structs are not possible in the translation to Julia.

Packed structs and union declarations are not supported by Julia.

You can get an approximation of a union if you know, a priori, the field that will have the greatest size (potentially including padding). When translating your fields to Julia, declare the Julia field to be only of that type.

Arrays of parameters can be expressed with NTuple. For example, the struct in C notation written as

  1. struct B {
  2. int A[3];
  3. };
  4. b_a_2 = B.A[2];

can be written in Julia as

  1. struct B
  2. A::NTuple{3, Cint}
  3. end
  4. b_a_2 = B.A[3] # note the difference in indexing (1-based in Julia, 0-based in C)

Arrays of unknown size (C99-compliant variable length structs specified by [] or [0]) are not directly supported. Often the best way to deal with these is to deal with the byte offsets directly. For example, if a C library declared a proper string type and returned a pointer to it:

  1. struct String {
  2. int strlen;
  3. char data[];
  4. };

In Julia, we can access the parts independently to make a copy of that string:

  1. str = from_c::Ptr{Cvoid}
  2. len = unsafe_load(Ptr{Cint}(str))
  3. unsafe_string(str + Core.sizeof(Cint), len)

Type Parameters

The type arguments to ccall and @cfunction are evaluated statically, when the method containing the usage is defined. They therefore must take the form of a literal tuple, not a variable, and cannot reference local variables.

This may sound like a strange restriction, but remember that since C is not a dynamic language like Julia, its functions can only accept argument types with a statically-known, fixed signature.

However, while the type layout must be known statically to compute the intended C ABI, the static parameters of the function are considered to be part of this static environment. The static parameters of the function may be used as type parameters in the call signature, as long as they don’t affect the layout of the type. For example, f(x::T) where {T} = ccall(:valid, Ptr{T}, (Ptr{T},), x) is valid, since Ptr is always a word-size primitive type. But, g(x::T) where {T} = ccall(:notvalid, T, (T,), x) is not valid, since the type layout of T is not known statically.

SIMD 值

Note: This feature is currently implemented on 64-bit x86 and AArch64 platforms only.

If a C/C++ routine has an argument or return value that is a native SIMD type, the corresponding Julia type is a homogeneous tuple of VecElement that naturally maps to the SIMD type. Specifically:

  • The tuple must be the same size as the SIMD type. For example, a tuple representing an __m128 on x86 must have a size of 16 bytes.
  • The element type of the tuple must be an instance of VecElement{T} where T is a primitive type that is 1, 2, 4 or 8 bytes.

For instance, consider this C routine that uses AVX intrinsics:

  1. #include <immintrin.h>
  2. __m256 dist( __m256 a, __m256 b ) {
  3. return _mm256_sqrt_ps(_mm256_add_ps(_mm256_mul_ps(a, a),
  4. _mm256_mul_ps(b, b)));
  5. }

The following Julia code calls dist using ccall:

  1. const m256 = NTuple{8, VecElement{Float32}}
  2. a = m256(ntuple(i -> VecElement(sin(Float32(i))), 8))
  3. b = m256(ntuple(i -> VecElement(cos(Float32(i))), 8))
  4. function call_dist(a::m256, b::m256)
  5. ccall((:dist, "libdist"), m256, (m256, m256), a, b)
  6. end
  7. println(call_dist(a,b))

The host machine must have the requisite SIMD registers. For example, the code above will not work on hosts without AVX support.

内存所有权

malloc/free

Memory allocation and deallocation of such objects must be handled by calls to the appropriate cleanup routines in the libraries being used, just like in any C program. Do not try to free an object received from a C library with Libc.free in Julia, as this may result in the free function being called via the wrong library and cause the process to abort. The reverse (passing an object allocated in Julia to be freed by an external library) is equally invalid.

何时使用 T、Ptr{T} 以及 Ref{T}

In Julia code wrapping calls to external C routines, ordinary (non-pointer) data should be declared to be of type T inside the ccall, as they are passed by value. For C code accepting pointers, Ref{T} should generally be used for the types of input arguments, allowing the use of pointers to memory managed by either Julia or C through the implicit call to Base.cconvert. In contrast, pointers returned by the C function called should be declared to be of output type Ptr{T}, reflecting that the memory pointed to is managed by C only. Pointers contained in C structs should be represented as fields of type Ptr{T} within the corresponding Julia struct types designed to mimic the internal structure of corresponding C structs.

In Julia code wrapping calls to external Fortran routines, all input arguments should be declared as of type Ref{T}, as Fortran passes all variables by pointers to memory locations. The return type should either be Cvoid for Fortran subroutines, or a T for Fortran functions returning the type T.

Mapping C Functions to Julia

ccall / @cfunction argument translation guide

For translating a C argument list to Julia:

  • T, where T is one of the primitive types: char, int, long, short, float, double, complex, enum or any of their typedef equivalents

    • T, where T is an equivalent Julia Bits Type (per the table above)
    • if T is an enum, the argument type should be equivalent to Cint or Cuint
    • argument value will be copied (passed by value)
  • struct T (including typedef to a struct)

    • T, where T is a Julia leaf type
    • argument value will be copied (passed by value)
  • void*

    • depends on how this parameter is used, first translate this to the intended pointer type, then determine the Julia equivalent using the remaining rules in this list
    • this argument may be declared as Ptr{Cvoid}, if it really is just an unknown pointer
  • jl_value_t*

    • Any
    • argument value must be a valid Julia object
  • jl_value_t**

    • Ref{Any}
    • argument value must be a valid Julia object (or C_NULL)
  • T*

    • Ref{T}, where T is the Julia type corresponding to T
    • argument value will be copied if it is an isbits type otherwise, the value must be a valid Julia object
  • T (*)(...) (e.g. a pointer to a function)

    • Ptr{Cvoid} (you may need to use @cfunction explicitly to create this pointer)
  • ... (e.g. a vararg)

    • T..., where T is the Julia type
    • currently unsupported by @cfunction
  • va_arg

    • not supported by ccall or @cfunction

ccall / @cfunction return type translation guide

For translating a C return type to Julia:

  • void

    • Cvoid (this will return the singleton instance nothing::Cvoid)
  • T, where T is one of the primitive types: char, int, long, short, float, double, complex, enum or any of their typedef equivalents

    • T, where T is an equivalent Julia Bits Type (per the table above)
    • if T is an enum, the argument type should be equivalent to Cint or Cuint
    • argument value will be copied (returned by-value)
  • struct T (including typedef to a struct)

    • T, where T is a Julia Leaf Type
    • argument value will be copied (returned by-value)
  • void*

    • depends on how this parameter is used, first translate this to the intended pointer type, then determine the Julia equivalent using the remaining rules in this list
    • this argument may be declared as Ptr{Cvoid}, if it really is just an unknown pointer
  • jl_value_t*

    • Any
    • argument value must be a valid Julia object
  • jl_value_t**

    • Ptr{Any} (Ref{Any} is invalid as a return type)
    • argument value must be a valid Julia object (or C_NULL)
  • T*

    • If the memory is already owned by Julia, or is an isbits type, and is known to be non-null:

      • Ref{T}, where T is the Julia type corresponding to T
      • a return type of Ref{Any} is invalid, it should either be Any (corresponding to jl_value_t*) or Ptr{Any} (corresponding to jl_value_t**)
      • C MUST NOT modify the memory returned via Ref{T} if T is an isbits type
    • If the memory is owned by C:

      • Ptr{T}, where T is the Julia type corresponding to T
  • T (*)(...) (e.g. a pointer to a function)

    • Ptr{Cvoid} (you may need to use @cfunction explicitly to create this pointer)

Passing Pointers for Modifying Inputs

Because C doesn’t support multiple return values, often C functions will take pointers to data that the function will modify. To accomplish this within a ccall, you need to first encapsulate the value inside a Ref{T} of the appropriate type. When you pass this Ref object as an argument, Julia will automatically pass a C pointer to the encapsulated data:

  1. width = Ref{Cint}(0)
  2. range = Ref{Cfloat}(0)
  3. ccall(:foo, Cvoid, (Ref{Cint}, Ref{Cfloat}), width, range)

Upon return, the contents of width and range can be retrieved (if they were changed by foo) by width[] and range[]; that is, they act like zero-dimensional arrays.

C Wrapper Examples

Let’s start with a simple example of a C wrapper that returns a Ptr type:

  1. mutable struct gsl_permutation
  2. end
  3. # The corresponding C signature is
  4. # gsl_permutation * gsl_permutation_alloc (size_t n);
  5. function permutation_alloc(n::Integer)
  6. output_ptr = ccall(
  7. (:gsl_permutation_alloc, :libgsl), # name of C function and library
  8. Ptr{gsl_permutation}, # output type
  9. (Csize_t,), # tuple of input types
  10. n # name of Julia variable to pass in
  11. )
  12. if output_ptr == C_NULL # Could not allocate memory
  13. throw(OutOfMemoryError())
  14. end
  15. return output_ptr
  16. end

The GNU Scientific Library (here assumed to be accessible through :libgsl) defines an opaque pointer, gsl_permutation *, as the return type of the C function gsl_permutation_alloc. As user code never has to look inside the gsl_permutation struct, the corresponding Julia wrapper simply needs a new type declaration, gsl_permutation, that has no internal fields and whose sole purpose is to be placed in the type parameter of a Ptr type. The return type of the ccall is declared as Ptr{gsl_permutation}, since the memory allocated and pointed to by output_ptr is controlled by C.

The input n is passed by value, and so the function’s input signature is simply declared as (Csize_t,) without any Ref or Ptr necessary. (If the wrapper was calling a Fortran function instead, the corresponding function input signature would instead be (Ref{Csize_t},), since Fortran variables are passed by pointers.) Furthermore, n can be any type that is convertible to a Csize_t integer; the ccall implicitly calls Base.cconvert(Csize_t, n).

Here is a second example wrapping the corresponding destructor:

  1. # The corresponding C signature is
  2. # void gsl_permutation_free (gsl_permutation * p);
  3. function permutation_free(p::Ref{gsl_permutation})
  4. ccall(
  5. (:gsl_permutation_free, :libgsl), # name of C function and library
  6. Cvoid, # output type
  7. (Ref{gsl_permutation},), # tuple of input types
  8. p # name of Julia variable to pass in
  9. )
  10. end

Here, the input p is declared to be of type Ref{gsl_permutation}, meaning that the memory that p points to may be managed by Julia or by C. A pointer to memory allocated by C should be of type Ptr{gsl_permutation}, but it is convertible using Base.cconvert and therefore

Now if you look closely enough at this example, you may notice that it is incorrect, given our explanation above of preferred declaration types. Do you see it? The function we are calling is going to free the memory. This type of operation cannot be given a Julia object (it will crash or cause memory corruption). Therefore, it may be preferable to declare the p type as Ptr{gsl_permutation }, to make it harder for the user to mistakenly pass another sort of object there than one obtained via gsl_permutation_alloc.

If the C wrapper never expects the user to pass pointers to memory managed by Julia, then using p::Ptr{gsl_permutation} for the method signature of the wrapper and similarly in the ccall is also acceptable.

Here is a third example passing Julia arrays:

  1. # The corresponding C signature is
  2. # int gsl_sf_bessel_Jn_array (int nmin, int nmax, double x,
  3. # double result_array[])
  4. function sf_bessel_Jn_array(nmin::Integer, nmax::Integer, x::Real)
  5. if nmax < nmin
  6. throw(DomainError())
  7. end
  8. result_array = Vector{Cdouble}(undef, nmax - nmin + 1)
  9. errorcode = ccall(
  10. (:gsl_sf_bessel_Jn_array, :libgsl), # name of C function and library
  11. Cint, # output type
  12. (Cint, Cint, Cdouble, Ref{Cdouble}),# tuple of input types
  13. nmin, nmax, x, result_array # names of Julia variables to pass in
  14. )
  15. if errorcode != 0
  16. error("GSL error code $errorcode")
  17. end
  18. return result_array
  19. end

The C function wrapped returns an integer error code; the results of the actual evaluation of the Bessel J function populate the Julia array result_array. This variable is declared as a Ref{Cdouble}, since its memory is allocated and managed by Julia. The implicit call to Base.cconvert(Ref{Cdouble}, result_array) unpacks the Julia pointer to a Julia array data structure into a form understandable by C.

Fortran Wrapper Example

The following example utilizes ccall to call a function in a common Fortran library (libBLAS) to computes a dot product. Notice that the argument mapping is a bit different here than above, as we need to map from Julia to Fortran. On every argument type, we specify Ref or Ptr. This mangling convention may be specific to your fortran compiler and operating system, and is likely undocumented. However, wrapping each in a Ref (or Ptr, where equivalent) is a frequent requirement of Fortran compiler implementations:

  1. function compute_dot(DX::Vector{Float64}, DY::Vector{Float64})
  2. @assert length(DX) == length(DY)
  3. n = length(DX)
  4. incx = incy = 1
  5. product = ccall((:ddot_, "libLAPACK"),
  6. Float64,
  7. (Ref{Int32}, Ptr{Float64}, Ref{Int32}, Ptr{Float64}, Ref{Int32}),
  8. n, DX, incx, DY, incy)
  9. return product
  10. end

垃圾回收安全

When passing data to a ccall, it is best to avoid using the pointer function. Instead define a convert method and pass the variables directly to the ccall. ccall automatically arranges that all of its arguments will be preserved from garbage collection until the call returns. If a C API will store a reference to memory allocated by Julia, after the ccall returns, you must ensure that the object remains visible to the garbage collector. The suggested way to do this is to make a global variable of type Array{Ref,1} to hold these values, until the C library notifies you that it is finished with them.

Whenever you have created a pointer to Julia data, you must ensure the original data exists until you have finished using the pointer. Many methods in Julia such as unsafe_load and String make copies of data instead of taking ownership of the buffer, so that it is safe to free (or alter) the original data without affecting Julia. A notable exception is unsafe_wrap which, for performance reasons, shares (or can be told to take ownership of) the underlying buffer.

The garbage collector does not guarantee any order of finalization. That is, if a contained a reference to b and both a and b are due for garbage collection, there is no guarantee that b would be finalized after a. If proper finalization of a depends on b being valid, it must be handled in other ways.

Non-constant Function Specifications

A (name, library) function specification must be a constant expression. However, it is possible to use computed values as function names by staging through eval as follows:

  1. @eval ccall(($(string("a", "b")), "lib"), ...

This expression constructs a name using string, then substitutes this name into a new ccall expression, which is then evaluated. Keep in mind that eval only operates at the top level, so within this expression local variables will not be available (unless their values are substituted with $). For this reason, eval is typically only used to form top-level definitions, for example when wrapping libraries that contain many similar functions. A similar example can be constructed for @cfunction.

However, doing this will also be very slow and leak memory, so you should usually avoid this and instead keep reading. The next section discusses how to use indirect calls to efficiently achieve a similar effect.

非直接调用

The first argument to ccall can also be an expression evaluated at run time. In this case, the expression must evaluate to a Ptr, which will be used as the address of the native function to call. This behavior occurs when the first ccall argument contains references to non-constants, such as local variables, function arguments, or non-constant globals.

For example, you might look up the function via dlsym, then cache it in a shared reference for that session. For example:

  1. macro dlsym(func, lib)
  2. z = Ref{Ptr{Cvoid}}(C_NULL)
  3. quote
  4. let zlocal = $z[]
  5. if zlocal == C_NULL
  6. zlocal = dlsym($(esc(lib))::Ptr{Cvoid}, $(esc(func)))::Ptr{Cvoid}
  7. $z[] = $zlocal
  8. end
  9. zlocal
  10. end
  11. end
  12. end
  13. mylibvar = Libdl.dlopen("mylib")
  14. ccall(@dlsym("myfunc", mylibvar), Cvoid, ())

Closure cfunctions

The first argument to @cfunction can be marked with a $, in which case the return value will instead be a struct CFunction which closes over the argument. You must ensure that this return object is kept alive until all uses of it are done. The contents and code at the cfunction pointer will be erased via a finalizer when this reference is dropped and atexit. This is not usually needed, since this functionality is not present in C, but can be useful for dealing with ill-designed APIs which don’t provide a separate closure environment parameter.

  1. function qsort(a::Vector{T}, cmp) where T
  2. isbits(T) || throw(ArgumentError("this method can only qsort isbits arrays"))
  3. callback = @cfunction $cmp Cint (Ref{T}, Ref{T})
  4. # Here, `callback` isa Base.CFunction, which will be converted to Ptr{Cvoid}
  5. # (and protected against finalization) by the ccall
  6. ccall(:qsort, Cvoid, (Ptr{T}, Csize_t, Csize_t, Ptr{Cvoid}),
  7. a, length(a), Base.elsize(a), callback)
  8. # We could instead use:
  9. # GC.@preserve callback begin
  10. # use(Base.unsafe_convert(Ptr{Cvoid}, callback))
  11. # end
  12. # if we needed to use it outside of a `ccall`
  13. return a
  14. end

Note

Closure @cfunction rely on LLVM trampolines, which are not available on all platforms (for example ARM and PowerPC).

关闭库

It is sometimes useful to close (unload) a library so that it can be reloaded. For instance, when developing C code for use with Julia, one may need to compile, call the C code from Julia, then close the library, make an edit, recompile, and load in the new changes. One can either restart Julia or use the Libdl functions to manage the library explicitly, such as:

  1. lib = Libdl.dlopen("./my_lib.so") # Open the library explicitly.
  2. sym = Libdl.dlsym(lib, :my_fcn) # Get a symbol for the function to call.
  3. ccall(sym, ...) # Use the pointer `sym` instead of the (symbol, library) tuple (remaining arguments are the
  4. same). Libdl.dlclose(lib) # Close the library explicitly.

Note that when using ccall with the tuple input (e.g., ccall((:my_fcn, "./my_lib.so"), ...)), the library is opened implicitly and it may not be explicitly closed.

调用规约

The second argument to ccall can optionally be a calling convention specifier (immediately preceding return type). Without any specifier, the platform-default C calling convention is used. Other supported conventions are: stdcall, cdecl, fastcall, and thiscall (no-op on 64-bit Windows). For example (from base/libc.jl) we see the same gethostnameccall as above, but with the correct signature for Windows:

  1. hn = Vector{UInt8}(undef, 256)
  2. err = ccall(:gethostname, stdcall, Int32, (Ptr{UInt8}, UInt32), hn, length(hn))

请参阅 LLVM Language Reference 来获得更多信息。

There is one additional special calling convention llvmcall, which allows inserting calls to LLVM intrinsics directly. This can be especially useful when targeting unusual platforms such as GPGPUs. For example, for CUDA, we need to be able to read the thread index:

  1. ccall("llvm.nvvm.read.ptx.sreg.tid.x", llvmcall, Int32, ())

As with any ccall, it is essential to get the argument signature exactly correct. Also, note that there is no compatibility layer that ensures the intrinsic makes sense and works on the current target, unlike the equivalent Julia functions exposed by Core.Intrinsics.

访问全局变量

Global variables exported by native libraries can be accessed by name using the cglobal function. The arguments to cglobal are a symbol specification identical to that used by ccall, and a type describing the value stored in the variable:

  1. julia> cglobal((:errno, :libc), Int32)
  2. Ptr{Int32} @0x00007f418d0816b8

The result is a pointer giving the address of the value. The value can be manipulated through this pointer using unsafe_load and unsafe_store!.

Note

This errno symbol may not be found in a library named “libc”, as this is an implementation detail of your system compiler. Typically standard library symbols should be accessed just by name, allowing the compiler to fill in the correct one. Also, however, the errno symbol shown in this example is special in most compilers, and so the value seen here is probably not what you expect or want. Compiling the equivalent code in C on any multi-threaded-capable system would typically actually call a different function (via macro preprocessor overloading), and may give a different result than the legacy value printed here.

Accessing Data through a Pointer

The following methods are described as “unsafe” because a bad pointer or type declaration can cause Julia to terminate abruptly.

Given a Ptr{T}, the contents of type T can generally be copied from the referenced memory into a Julia object using unsafe_load(ptr, [index]). The index argument is optional (default is 1), and follows the Julia-convention of 1-based indexing. This function is intentionally similar to the behavior of getindex and setindex! (e.g. [] access syntax).

The return value will be a new object initialized to contain a copy of the contents of the referenced memory. The referenced memory can safely be freed or released.

If T is Any, then the memory is assumed to contain a reference to a Julia object (a jl_value_t*), the result will be a reference to this object, and the object will not be copied. You must be careful in this case to ensure that the object was always visible to the garbage collector (pointers do not count, but the new reference does) to ensure the memory is not prematurely freed. Note that if the object was not originally allocated by Julia, the new object will never be finalized by Julia’s garbage collector. If the Ptr itself is actually a jl_value_t*, it can be converted back to a Julia object reference by unsafe_pointer_to_objref(ptr). (Julia values v can be converted to jl_value_t* pointers, as Ptr{Cvoid}, by calling pointer_from_objref(v).)

The reverse operation (writing data to a Ptr{T}), can be performed using unsafe_store!(ptr, value, [index]). Currently, this is only supported for primitive types or other pointer-free (isbits) immutable struct types.

Any operation that throws an error is probably currently unimplemented and should be posted as a bug so that it can be resolved.

If the pointer of interest is a plain-data array (primitive type or immutable struct), the function unsafe_wrap(Array, ptr,dims, own = false) may be more useful. The final parameter should be true if Julia should “take ownership” of the underlying buffer and call free(ptr) when the returned Array object is finalized. If the own parameter is omitted or false, the caller must ensure the buffer remains in existence until all access is complete.

Arithmetic on the Ptr type in Julia (e.g. using +) does not behave the same as C’s pointer arithmetic. Adding an integer to a Ptr in Julia always moves the pointer by some number of bytes, not elements. This way, the address values obtained from pointer arithmetic do not depend on the element types of pointers.

线程安全

Some C libraries execute their callbacks from a different thread, and since Julia isn’t thread-safe you’ll need to take some extra precautions. In particular, you’ll need to set up a two-layered system: the C callback should only schedule (via Julia’s event loop) the execution of your “real” callback. To do this, create an AsyncCondition object and wait on it:

  1. cond = Base.AsyncCondition()
  2. wait(cond)

传递给 C 的回调应该只通过 ccallcond.handle 作为参数传递给 :uv_async_send 并调用,注意避免任何内存分配操作或与 Julia 运行时的其他交互。

注意,事件可能会合并,因此对 uv_async_send 的多个调用可能会导致对该条件的单个唤醒通知。

关于 Callbacks 的更多内容

关于如何传递 callback 到 C 库的更多细节,请参考此博客

C++

如需要直接易用的C++接口,即直接用Julia写封装代码,请参考 Cxx。如需封装C++库的工具,即用C++写封装/胶水代码,请参考CxxWrap

  • 1Non-library function calls in both C and Julia can be inlined and thus may have even less overhead than calls to shared library functions. The point above is that the cost of actually doing foreign function call is about the same as doing a call in either native language.
  • 2The Clang package can be used to auto-generate Julia code from a C header file.