mirror of
git://git.code.sf.net/p/sbcl/sbcl
synced 2026-09-10 07:26:40 -04:00
68d68fe3cd
This got about 5% speedup in the FIND-SYMBOL benchmark for those CPUs, less than on x86-64, but I'll bet they run the string hashing step more slowly, so the remainder operation in %LOOKUP-SYMBOL is a smaller contributor to the total time. Amdahl's law pertains. Also implement fastrem-64 in case we need it.
61 lines
2.6 KiB
Common Lisp
61 lines
2.6 KiB
Common Lisp
(in-package sb-impl)
|
|
|
|
(defun find-sym-timing (niter &aux (res 0))
|
|
(let* ((pkg (find-package "KEYWORD"))
|
|
(externals (package-external-symbols pkg))
|
|
(vect (remove-if-not #'symbolp (symtbl-cells externals))))
|
|
(dotimes (i niter)
|
|
(dovector (sym vect)
|
|
(let ((str (symbol-name (truly-the symbol sym))))
|
|
(when (find-symbol str pkg)
|
|
(incf res))))))
|
|
res)
|
|
|
|
(find-sym-timing 5000)
|
|
|
|
#|
|
|
;; This benchmarks shows just how large the effect is of removing
|
|
;; one IDIV instruction from the hash -> bin calculation in SYMBOL-HASHSET.
|
|
;; As expected, we take a bunch more branches and run more instructions,
|
|
;; but the overall instructions/cycle is higher, and CPU time is lower.
|
|
;; These results are completely repeatable.
|
|
|
|
% perf stat path/to/sbcl --noinform --noprint --no-sysinit --no-userinit < benchmarks/find-symbol.lisp
|
|
|
|
Without fast remainder vop
|
|
==========================
|
|
1,963.00 msec task-clock:u # 0.999 CPUs utilized
|
|
0 context-switches:u # 0.000 /sec
|
|
0 cpu-migrations:u # 0.000 /sec
|
|
1,094 page-faults:u # 557.311 /sec
|
|
5,433,686,072 cycles:u # 2.768 GHz
|
|
2,067,083,081 stalled-cycles-frontend:u # 38.04% frontend cycles idle
|
|
8,180,495,043 instructions:u # 1.51 insn per cycle
|
|
# 0.25 stalled cycles per insn
|
|
1,648,928,710 branches:u # 840.006 M/sec
|
|
36,906,902 branch-misses:u # 2.24% of all branches
|
|
|
|
1.965377442 seconds time elapsed
|
|
|
|
1.956049000 seconds user
|
|
0.008000000 seconds sys
|
|
|
|
With fast remainder vop
|
|
=======================
|
|
1,689.64 msec task-clock:u # 0.999 CPUs utilized
|
|
0 context-switches:u # 0.000 /sec
|
|
0 cpu-migrations:u # 0.000 /sec
|
|
1,018 page-faults:u # 602.494 /sec
|
|
4,670,096,246 cycles:u # 2.764 GHz
|
|
1,540,757,518 stalled-cycles-frontend:u # 32.99% frontend cycles idle
|
|
8,636,783,323 instructions:u # 1.85 insn per cycle
|
|
# 0.18 stalled cycles per insn
|
|
1,707,772,211 branches:u # 1.011 G/sec
|
|
38,296,405 branch-misses:u # 2.24% of all branches
|
|
|
|
1.691727650 seconds time elapsed
|
|
|
|
1.678228000 seconds user
|
|
0.011958000 seconds sys
|
|
|#
|