26.6 Monitoring of Performance

26.6.1 SPY (Unix only)

The Unix profil system call provides a facility to get information about the cpu consumption in a LISP program. A simple interface was build to be able to turn profiling on and off and to map the results on to the LISP address space. For details, please refer to the source code in $pxu/spy.sl.

(spywhole N:Integer): void expr

• Turns on profiling for the whole bps space. N must be a power of 2, defining the grain for the profil call. Typical values are 4 and 8. Returns nil.

(spyprint): void expr

• Turns off the profiler. Prints profiler information after mapping it to the LISP address space, i.e the names of LISP functions are printed together with their relative cpu consumption. The variable &spyminimum& defines the cutoff for spyprint, i.e. functions with less than &spyminimum& ticks will not be printed. Returns nil.

(spyon fwa lwa tslice bucketsize power): void expr

• This is a complete interface for the profil call. Allocates Warray space for the Unix profiler and turns on profiling. fwa and lwa are the addresses of the first and last word of code memory which should be inspected, tslice is the timeslice and not suported on most Unix systems. It is a parameter for compatibility reasons. Bucketsize must be a power of 2 and power is the base 2 logarithm of bucketsize. Typical values are 4 or 8 for bucketsize and 2 or 3 for power resp. Returns nil.

(spyoff): void expr

• Turns off profiling using the Unix profil system call. Returns nil.

Example:

It is easy to see that the command (null (setq aa(expt 3000 10000))) leads to a lot of bignum computations. The spyprint output proves that. Please note that INITCODE sums up all the functions that belong to PSL’s C runtime support. In this case the high percentage is due to the fact that the (old) SPARC implementation performs integer multiply and divide in software i.e. in non LISP code.

26.6.2 Qualified timing

Qualified counting is used to measure the cpu time spend between call and return from a function, not necessarily spend within the body of the function. It gives an overview about the costs that a call of particular function causes. The sum shown as result may be bigger that 100 percent, if a function and its callee are both measured or if gc time is accumulated.

(qualtime S:list): void macro

• Starts qualified timing for the list of functions contained in list. Returns nil.

(print!-qualtime): void expr

• Prints the accumulated timings and number of calls for all functions. Resets counters. Returns nil.

Example:

26.6.3 Qualified counting

For our own applications it had turned out to be useful to know the callers of a function especially when the function is called very, very often, e.g. basic routines like generic arithmetic. To provide this a module qualified counting was generated using a similar syntax like qualified timing.

(qualcount S:list): void macro

• Starts qualified counting for the list of functions contained in list. Returns nil.

(print!-qualcount): void expr

• Prints the accumulated call counting and the callers for all functions. Resets counters. Printing will be suppressed if the number of calls is less than bordervalue (default = 20) Returns nil.

(reset!-qualcount): void expr

• Resets the accumulated call counting and the callers for all functions. Returns nil.

Example: