NAG Library Manual

A thread is a basic entity to which an operating system allocates CPU time. A thread has its own registers, stack and process resources. Threads provide a convenient way of allowing an application to maximise its usage of CPU resources in a system, especially in a multiple processor configuration. A routine is termed ‘thread safe’ if it can be called from two or more concurrently running threads without compromising results.
The remainder of this document describes thread safety within the context of the NAG Library and provides guidelines for calling Library routines from multithreaded applications.

## 2  Thread Safety and the NAG Library

It is essential that you refer to the Users' Note for details of whether the Library has been compiled in a manner that facilitates the use of multiple threads.

In a Fortran 77 context the constructs that prohibit thread safety are, potentially, DATA, SAVE, COMMON and EQUIVALENCE. This is because such constructs define data that may be shared by different threads, perhaps leading to unwanted interactions between them: for example, the possibility that one thread may be modifying the contents of a COMMON block at the same time as another thread is reading it. You are therefore advised to use such constructs with great care and to avoid their use wherever possible within multithreaded applications.
At Mark 20 of the NAG Library the thread safe provision was significantly enhanced by
 (a) eliminating unsafe constructs wherever possible to make the majority of routines safe for use in multithreaded applications; (b) providing equivalent thread safe routines with the same functionality where complete removal of unsafe constructs would affect interface design. Two approaches have been taken to provide thread safe equivalents; see Section 2.2 for further details.
At subsequent Marks of the NAG Library, the list of potentially thread unsafe routines will be reduced as these routines are replaced by thread safe equivalents. Within a small number of Marks there will be no potentially thread unsafe routines in the Library.
See Section 3.2 for a list of the remaining routines that are currently thread unsafe with no thread safe equivalent. It should be noted that it is always safe to call any NAG Library routine in one thread (only) of a multithreaded application.

### 2.2  Library Routines with Thread Safe Equivalents

At Marks 21 and 22 of the NAG Library two approaches have been taken to provide thread safe equivalents to routines containing unsafe constructs. In the first approach a close connection between the original routine and the thread safe equivalent can be maintained, allowing the two routines to appear as a pair and share the same root name. In the second approach more fundamental changes in interface design have been made such that the correspondence between a routine and its thread safe equivalent cannot be maintained through the root name.

#### 2.2.1  Routine and thread safe equivalent sharing the same root name

At Mark 22 of the NAG Library there are pairs of routines which share the same root name, for example, the routines E04UCF/E04UCA. Each routine in the pair has exactly the same functionality, except that one of them has additional parameters in order to make it safe for use in multithreaded applications. The routine that is safe for use in multithreaded applications has a different last character in the name in place of the usual character (typically ‘A’ instead of ‘F’). Such pairs are documented via one routine document. If the pair of routines contain a user-supplied routine argument in their interface then the ‘A’ routine will have array arguments that enable you to pass information to the user-supplied routine without the need for COMMON blocks. In some cases the ‘A’ routine may need to be initialized by a separate initialization routine; this requirement will be clearly documented.

### 2.3  Routines with Routine Arguments

Some Library routines require you to supply a routine and to pass the name of the routine as an actual argument in the call to the Library routine. For many of these Library routines, the supplied routine interface includes array parameters (called IUSER and RUSER) specifically for you to pass information to the supplied routine. However, there remain some Library routines for which you may need to supply your provided routine with more information than can be given via the interface argument list.
You are advised to check, in the relevant chapter introduction, whether the Library routines you intend to call have equivalent reverse communication interfaces. These have been designed specifically for problems where user-supplied routine interfaces are not flexible enough for a given problem, and their use should eliminate the need to provide data through COMMON blocks. Where reverse communication interfaces are not available, it is usual to define a COMMON block containing the required data in the supplied routine (and also in the calling program). It is threadsafe to do this only if no data referenced in the defined COMMON block is updated (thus avoiding the possibility of simultaneous modification by different threads). Where separate calls are made to a Library routine by different threads and these calls require different data sets to be passed through COMMON blocks to user-supplied routines, these routines and the COMMON blocks defined within them must have different names.
Thread safety of user-supplied routines is also an issue with a number of routines in the NAG Library for SMP and Multicore, which may internally parallelize around the calls to the user-supplied routines, affecting not just COMMON blocks but also how the IUSER and RUSER arrays may be used. In these cases, use of thread-unsafe features within the user-supplied routines should be avoided wherever possible, and care should be taken to ensure any usage of IUSER and RUSER is undertaken in a thread-safe manner. If they cannot be avoided entirely, you are advised to contact NAG for assistance on how to manage possible data dependencies within shared data structures.

### 2.4  Input/Output

The Library contains routines for setting the current error and advisory message unit numbers (X04AAF and X04ABF). These routines use the SAVE statement to retain the values of the current unit numbers between calls. It is therefore not advisable for different threads of a multithreaded program to set the message unit numbers to different values. A consequence of this is that error or advisory messages output simultaneously may become garbled, and in any event there is no indication of which thread produces which message. You are therefore advised always to select the ‘soft failure’ mechanism without any error message ($\mathrm{IFAIL}=+1$, see Section 3.3 in the Essential Introduction) on entry to each NAG Library routine called from a multithreaded application; it is then essential that the value of IFAIL be tested on return to the application.
A related problem is that of multiple threads writing to or reading from files from your multithreaded application. You are advised to make different threads use different unit numbers for opening files and to give these files different names (perhaps by appending an index number to the file basename). The only alternative to this is for you to protect each write to a file or unit number; for example, by putting each WRITE statement in a critical region.

### 2.5  Implementation Issues

In some implementations of the NAG Library calls are made to vendor BLAS and/or LAPACK Library routines. Although NAG perform tests to ensure that these calls are behaving correctly on multiple threads, NAG cannot guarantee the thread safety of the vendor BLAS and LAPACK routines. You are advised to refer to the Users' Note for details of whether the Library is to be linked with vendor BLAS and/or LAPACK Libraries.