;;; reliable?
(defpackage #:sb-sprof
- (:use #:cl #:sb-ext #:sb-unix #:sb-alien #:sb-sys)
+ (:use #:cl #:sb-ext #:sb-unix #:sb-alien #:sb-sys :sb-int)
(:export #:*sample-interval* #:*max-samples* #:*alloc-interval*
#:*report-sort-by* #:*report-sort-order*
#:start-sampling #:stop-sampling #:with-sampling
(max-samples (sb-int:missing-arg) :type sb-int:index)
(sampled-threads nil :type list))
+(defmethod print-object ((samples samples) stream)
+ (print-unreadable-object (samples stream :type t :identity t)
+ (let ((*print-array* nil))
+ (call-next-method))))
+
(defmethod print-object ((call-graph call-graph) stream)
(print-unreadable-object (call-graph stream :type t :identity t)
(format stream "~d samples" (call-graph-nsamples call-graph))))
(defvar *sampling-mode* :cpu
"Default sampling mode. :CPU for cpu profiling, :ALLOC for allocation
-profiling")
+profiling, and :TIME for wallclock profiling.")
(declaim (type (member :cpu :alloc :time) *sampling-mode*))
(defvar *alloc-region-size*
;; Ensure that only one thread at a time will be doing profiling stuff.
(defvar *profiler-lock* (sb-thread:make-mutex :name "Statistical Profiler"))
(defvar *distribution-lock* (sb-thread:make-mutex :name "Wallclock profiling lock"))
-
+
#+sb-thread
(declaim (inline pthread-kill))
#+sb-thread
;;; interested using SIGPROF.
(defun thread-distribution-handler ()
(declare (optimize speed (space 0)))
- (when *sampling*
- #+sb-thread
- (let ((lock *distribution-lock*))
- ;; Don't flood the system with more interrupts if the last
- ;; set is still being delivered.
- (unless (sb-thread:mutex-value lock)
- (sb-thread::with-system-mutex (lock)
- (dolist (thread (profiled-threads))
- ;; This may occasionally fail to deliver the signal, but that
- ;; seems better then using kill_thread_safely with it's 1
- ;; second backoff.
- (let ((os-thread (sb-thread::thread-os-thread thread)))
- (when os-thread
- (pthread-kill os-thread sb-unix:sigprof)))))))
- #-sb-thread
- (unix-kill 0 sb-unix:sigprof)))
+ #+sb-thread
+ (let ((lock *distribution-lock*))
+ ;; Don't flood the system with more interrupts if the last
+ ;; set is still being delivered.
+ (unless (sb-thread:mutex-value lock)
+ (sb-thread::with-system-mutex (lock)
+ (dolist (thread (profiled-threads))
+ ;; This may occasionally fail to deliver the signal, but that
+ ;; seems better then using kill_thread_safely with it's 1
+ ;; second backoff.
+ (let ((os-thread (sb-thread::thread-os-thread thread)))
+ (when os-thread
+ (pthread-kill os-thread sb-unix:sigprof)))))))
+ #-sb-thread
+ (unix-kill 0 sb-unix:sigprof))
(defun sigprof-handler (signal code scp)
(declare (ignore signal code) (optimize speed (space 0))
(max-samples '*max-samples*)
(reset nil)
(mode '*sampling-mode*)
- (loop t)
+ (loop nil)
(max-depth most-positive-fixnum)
show-progress
(threads '(list sb-thread:*current-thread*))
(report nil report-p))
&body body)
- "Repeatedly evaluate BODY with statistical profiling turned on.
- In multi-threaded operation, only the thread in which WITH-PROFILING
- was evaluated will be profiled by default. If you want to profile
- multiple threads, invoke the profiler with START-PROFILING.
+ "Evaluate BODY with statistical profiling turned on. If LOOP is true,
+loop around the BODY until a sufficient number of samples has been collected.
+Returns the values from the last evaluation of BODY.
- The following keyword args are recognized:
+In multithreaded operation, only the thread in which WITH-PROFILING was
+evaluated will be profiled by default. If you want to profile multiple
+threads, invoke the profiler with START-PROFILING.
- :SAMPLE-INTERVAL <n>
- Take a sample every <n> seconds. Default is *SAMPLE-INTERVAL*.
+The following keyword args are recognized:
- :ALLOC-INTERVAL <n>
- Take a sample every time <n> allocation regions (approximately
- 8kB) have been allocated since the last sample. Default is
- *ALLOC-INTERVAL*.
+ :SAMPLE-INTERVAL <n>
+ Take a sample every <n> seconds. Default is *SAMPLE-INTERVAL*.
- :MODE <mode>
- If :CPU, run the profiler in CPU profiling mode. If :ALLOC, run the
- profiler in allocation profiling mode. If :TIME, run the profiler
- in wallclock profiling mode.
+ :ALLOC-INTERVAL <n>
+ Take a sample every time <n> allocation regions (approximately
+ 8kB) have been allocated since the last sample. Default is
+ *ALLOC-INTERVAL*.
- :MAX-SAMPLES <max>
- Repeat evaluating body until <max> samples are taken.
- Default is *MAX-SAMPLES*.
+ :MODE <mode>
+ If :CPU, run the profiler in CPU profiling mode. If :ALLOC, run the
+ profiler in allocation profiling mode. If :TIME, run the profiler
+ in wallclock profiling mode.
- :MAX-DEPTH <max>
- Maximum call stack depth that the profiler should consider. Only
- has an effect on x86 and x86-64.
+ :MAX-SAMPLES <max>
+ Repeat evaluating body until <max> samples are taken.
+ Default is *MAX-SAMPLES*.
- :REPORT <type>
- If specified, call REPORT with :TYPE <type> at the end.
+ :MAX-DEPTH <max>
+ Maximum call stack depth that the profiler should consider. Only
+ has an effect on x86 and x86-64.
- :RESET <bool>
- It true, call RESET at the beginning.
+ :REPORT <type>
+ If specified, call REPORT with :TYPE <type> at the end.
- :THREADS <list-form>
- Form that evaluates to the list threads to profile, or :ALL to indicate
- that all threads should be profiled. Defaults to the current
- thread. (Note: START-PROFILING defaults to all threads.)
+ :RESET <bool>
+ It true, call RESET at the beginning.
- :THREADS has no effect on call-counting at the moment.
+ :THREADS <list-form>
+ Form that evaluates to the list threads to profile, or :ALL to indicate
+ that all threads should be profiled. Defaults to the current
+ thread. (Note: START-PROFILING defaults to all threads.)
- On some platforms (eg. Darwin) the signals used by the profiler are
- not properly delivered to threads in proportion to their CPU usage
- when doing :CPU profiling. If you see empty call graphs, or are obviously
- missing several samples from certain threads, you may be falling afoul
- of this.
+ :THREADS has no effect on call-counting at the moment.
+
+ On some platforms (eg. Darwin) the signals used by the profiler are
+ not properly delivered to threads in proportion to their CPU usage
+ when doing :CPU profiling. If you see empty call graphs, or are obviously
+ missing several samples from certain threads, you may be falling afoul
+ of this. In this case using :MODE :TIME is likely to work better.
- :LOOP <bool>
- If true (the default) repeatedly evaluate BODY. If false, evaluate
- if only once."
+ :LOOP <bool>
+ If false (the default), evaluate BODY only once. If true repeatedly
+ evaluate BODY."
(declare (type report-type report))
- `(let* ((*sample-interval* ,sample-interval)
- (*alloc-interval* ,alloc-interval)
- (*sampling* nil)
- (*sampling-mode* ,mode)
- (*max-samples* ,max-samples))
- ,@(when reset '((reset)))
- (unwind-protect
- (progn
- (start-profiling :max-depth ,max-depth :threads ,threads)
- (loop
- (when (>= (samples-trace-count *samples*)
- (samples-max-samples *samples*))
- (return))
- ,@(when show-progress
- `((format t "~&===> ~d of ~d samples taken.~%"
- (samples-trace-count *samples*)
- (samples-max-samples *samples*))))
- (let ((.last-index. (samples-index *samples*)))
- ,@body
- (when (= .last-index. (samples-index *samples*))
- (warn "No sampling progress; possibly a profiler bug.")
- (return)))
- (unless ,loop
- (return))))
- (stop-profiling))
- ,@(when report-p `((report :type ,report)))))
+ (check-type loop boolean)
+ (with-unique-names (values last-index oops)
+ `(let* ((*sample-interval* ,sample-interval)
+ (*alloc-interval* ,alloc-interval)
+ (*sampling* nil)
+ (*sampling-mode* ,mode)
+ (*max-samples* ,max-samples))
+ ,@(when reset '((reset)))
+ (flet ((,oops ()
+ (warn "~@<No sampling progress; run too short, sampling interval ~
+ too long, inappropriate set of sampled thread, or possibly ~
+ a profiler bug.~:@>")))
+ (unwind-protect
+ (progn
+ (start-profiling :max-depth ,max-depth :threads ,threads)
+ ,(if loop
+ `(let (,values)
+ (loop
+ (when (>= (samples-trace-count *samples*)
+ (samples-max-samples *samples*))
+ (return))
+ ,@(when show-progress
+ `((format t "~&===> ~d of ~d samples taken.~%"
+ (samples-trace-count *samples*)
+ (samples-max-samples *samples*))))
+ (let ((,last-index, (samples-index *samples*)))
+ (setf ,values (multiple-value-list (progn ,@body)))
+ (when (= ,last-index (samples-index *samples*))
+ (,oops)
+ (return))))
+ (values-list ,values))
+ `(let ((,last-index (samples-index *samples*)))
+ (multiple-value-prog1 (progn ,@body)
+ (when (= ,last-index (samples-index *samples*))
+ (,oops))))))
+ (stop-profiling)))
+ ,@(when report-p `((report :type ,report))))))
(defvar *timer* nil)
(sb-c:%more-arg-values more-context
0
more-count)))))))))
-
\f
-;;; silly examples
-
-(defun test-0 (n &optional (depth 0))
- (declare (optimize (debug 3)))
- (when (< depth n)
- (dotimes (i n)
- (test-0 n (1+ depth))
- (test-0 n (1+ depth)))))
-
-(defun test ()
- (with-profiling (:reset t :max-samples 1000 :report :graph)
- (test-0 7)))
-
-(defun consalot ()
- (let ((junk '()))
- (loop repeat 10000 do
- (push (make-array 10) junk))
- junk))
-
-(defun consing-test ()
- ;; 0.0001 chosen so that it breaks rather reliably when sprof does not
- ;; respect pseudo atomic.
- (with-profiling (:reset t :sample-interval 0.0001 :report :graph :loop nil)
- (let ((target (+ (get-universal-time) 15)))
- (princ #\.)
- (force-output)
- (loop
- while (< (get-universal-time) target)
- do (consalot)))))
-
-
-;;; provision
(provide 'sb-sprof)
-
-;;; end of file
projects / sbcl.git / blobdiff