Mercurial > hg > xemacs-beta
view lib-src/qsort.c @ 5882:bbe4146603db
Reduce regexp usage, now CL-oriented non-regexp code available, core Lisp
lisp/ChangeLog addition:
2015-04-01 Aidan Kehoe <kehoea@parhasard.net>
When calling #'string-match with a REGEXP without regular
expression special characters, call #'search, #'mismatch, #'find,
etc. instead, making our code less likely to side-effect other
functions' match data and a little faster.
* apropos.el (apropos-command):
* apropos.el (apropos):
Call (position ?\n ...) rather than (string-match "\n" ...) here.
* buff-menu.el:
* buff-menu.el (buffers-menu-omit-invisible-buffers):
Don't fire up the regexp engine just to check if a string starts
with a space.
* buff-menu.el (select-buffers-tab-buffers-by-mode):
Don't fire up the regexp engine just to compare mode basenames.
* buff-menu.el (format-buffers-tab-line):
* buff-menu.el (build-buffers-tab-internal): Moved to being a
label within the following.
* buff-menu.el (buffers-tab-items): Use the label.
* bytecomp.el (byte-compile-log-1):
Don't fire up the regexp engine just to look for a newline.
* cus-edit.el (get):
Ditto.
* cus-edit.el (custom-variable-value-create):
Ditto, but for a colon.
* descr-text.el (describe-text-sexp):
Ditto.
* descr-text.el (describe-char-unicode-data):
Use #'split-string-by-char given that we're just looking for a
semicolon.
* descr-text.el (describe-char):
Don't fire up the regexp engine just to look for a newline.
* disass.el (disassemble-internal):
Ditto.
* files.el (file-name-sans-extension):
Implement this using #'position.
* files.el (file-name-extension):
Correct this function's docstring, implement it in terms of
#'position.
* files.el (insert-directory):
Don't fire up the regexp engine to split a string by space; don't
reverse the list of switches, this is actually a longstand bug as
far as I can see.
* gnuserv.el (gnuserv-process-filter):
Use #'position here, instead of consing inside #'split-string
needlessly.
* gtk-file-dialog.el (gtk-file-dialog-update-dropdown):
Use #'split-string-by-char here, don't fire up #'split-string for
directory-sep-char.
* gtk-font-menu.el (hack-font-truename):
Implement this more cheaply in terms of #'find,
#'split-string-by-char, #'equal, rather than #'string-match,
#'split-string, #'string-equal.
* hyper-apropos.el (hyper-apropos-grok-functions):
* hyper-apropos.el (hyper-apropos-grok-variables):
Look for a newline using #'position rather than #'string-match in
these functions.
* info.el (Info-insert-dir):
* info.el (Info-insert-file-contents):
* info.el (Info-follow-reference):
* info.el (Info-extract-menu-node-name):
* info.el (Info-menu):
Look for fixed strings using #'position or #'search as appropriate
in this file.
* ldap.el (ldap-decode-string):
* ldap.el (ldap-encode-string):
#'encode-coding-string, #'decode-coding-string are always
available, don't check if they're fboundp.
* ldap.el (ldap-decode-address):
* ldap.el (ldap-encode-address):
Use #'split-string-by-char in these functions.
* lisp-mnt.el (lm-creation-date):
* lisp-mnt.el (lm-last-modified-date):
Don't fire up the regexp engine just to look for spaces in this file.
* menubar-items.el (default-menubar):
Use (not (mismatch ...)) rather than #'string-match here, for
simple regexp.
Use (search "beta" ...) rather than (string-match "beta" ...)
* menubar-items.el (sort-buffers-menu-alphabetically):
* menubar-items.el (sort-buffers-menu-by-mode-then-alphabetically):
* menubar-items.el (group-buffers-menu-by-mode-then-alphabetically):
Don't fire up the regexp engine to check if a string starts with
a space or an asterisk.
Use the more fine-grained results of #'compare-strings; compare
case-insensitively for the buffer menu.
* menubar-items.el (list-all-buffers):
* menubar-items.el (tutorials-menu-filter):
Use #'equal rather than #'string-equal, which, in this context,
has the drawback of not having a bytecode, and no redeeming
features.
* minibuf.el:
* minibuf.el (un-substitute-in-file-name):
Use #'count, rather than counting the occurences of $ using the
regexp engine.
* minibuf.el (read-file-name-internal-1):
Don't fire up the regexp engine to search for ?=.
* mouse.el (mouse-eval-sexp):
Check for newline with #'find.
* msw-font-menu.el (mswindows-reset-device-font-menus):
Split a string by newline with #'split-string-by-char.
* mule/japanese.el:
* mule/japanese.el ("Japanese"):
Use #'search rather than #'string-match; canoncase before
comparing; fix a bug I had introduced where I had been making case
insensitive comparisons where the case mattered.
* mule/korea-util.el (default-korean-keyboard):
Look for ?3 using #'find, not #'string-march.
* mule/korea-util.el (quail-hangul-switch-hanja):
Search for a fixed string using #'search.
* mule/mule-cmds.el (set-locale-for-language-environment):
#'position, #'substitute rather than #'string-match,
#'replace-in-string.
* newcomment.el (comment-make-extra-lines):
Use #'search rather than #'string-match for a simple string.
* package-get.el (package-get-remote-filename):
Use #'position when looking for ?@
* process.el (setenv):
* process.el (read-envvar-name):
Use #'position when looking for ?=.
* replace.el (map-query-replace-regexp):
Use #'split-string-by-char instead of using an inline
implementation of it.
* select.el (select-convert-from-cf-text):
* select.el (select-convert-from-cf-unicodetext):
Use #'position rather than #'string-match in these functions.
* setup-paths.el (paths-emacs-data-root-p):
Use #'search when looking for simple string.
* sound.el (load-sound-file):
Use #'split-string-by-char rather than an inline reimplementation
of same.
* startup.el (splash-screen-window-body):
* startup.el (splash-screen-tty-body):
Search for simple strings using #'search.
* version.el (emacs-version):
Ditto.
* x-font-menu.el (hack-font-truename):
Implement this more cheaply in terms of #'find,
#'split-string-by-char, #'equal, rather than #'string-match,
#'split-string, #'string-equal.
* x-font-menu.el (x-reset-device-font-menus-core):
Use #'split-string-by-char here.
* x-init.el (x-initialize-keyboard):
Search for a simple string using #'search.
author | Aidan Kehoe <kehoea@parhasard.net> |
---|---|
date | Wed, 01 Apr 2015 14:28:20 +0100 |
parents | 061f4f90f874 |
children |
line wrap: on
line source
/* Plug-compatible replacement for UNIX qsort. Copyright (C) 1989 Free Software Foundation, Inc. Written by Douglas C. Schmidt (schmidt@ics.uci.edu) This file is part of GNU CC. GNU QSORT is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. GNU QSORT is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with GNU QSORT. If not, see <http://www.gnu.org/licenses/>. */ /* Synched up with: FSF 19.28. */ #ifdef sparc #include <alloca.h> #endif /* Invoke the comparison function, returns either 0, < 0, or > 0. */ #define CMP(A,B) ((*cmp)((A),(B))) /* Byte-wise swap two items of size SIZE. */ #define SWAP(A,B,SIZE) do {int sz = (SIZE); char *a = (A); char *b = (B); \ do { char _temp = *a;*a++ = *b;*b++ = _temp;} while (--sz);} while (0) /* Copy SIZE bytes from item B to item A. */ #define COPY(A,B,SIZE) {int sz = (SIZE); do { *(A)++ = *(B)++; } while (--sz); } /* This should be replaced by a standard ANSI macro. */ #define BYTES_PER_WORD 8 /* The next 4 #defines implement a very fast in-line stack abstraction. */ #define STACK_SIZE (BYTES_PER_WORD * sizeof (long)) #define PUSH(LOW,HIGH) do {top->lo = LOW;top++->hi = HIGH;} while (0) #define POP(LOW,HIGH) do {LOW = (--top)->lo;HIGH = top->hi;} while (0) #define STACK_NOT_EMPTY (stack < top) /* Discontinue quicksort algorithm when partition gets below this size. This particular magic number was chosen to work best on a Sun 4/260. */ #define MAX_THRESH 4 /* Stack node declarations used to store unfulfilled partition obligations. */ typedef struct { char *lo; char *hi; } stack_node; /* Order size using quicksort. This implementation incorporates four optimizations discussed in Sedgewick: 1. Non-recursive, using an explicit stack of pointer that store the next array partition to sort. To save time, this maximum amount of space required to store an array of MAX_INT is allocated on the stack. Assuming a 32-bit integer, this needs only 32 * sizeof (stack_node) == 136 bits. Pretty cheap, actually. 2. Choose the pivot element using a median-of-three decision tree. This reduces the probability of selecting a bad pivot value and eliminates certain extraneous comparisons. 3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving insertion sort to order the MAX_THRESH items within each partition. This is a big win, since insertion sort is faster for small, mostly sorted array segments. 4. The larger of the two sub-partitions is always pushed onto the stack first, with the algorithm then concentrating on the smaller partition. This *guarantees* no more than log (n) stack size is needed (actually O(1) in this case)! */ int qsort (base_ptr, total_elems, size, cmp) char *base_ptr; int total_elems; int size; int (*cmp)(); { /* Allocating SIZE bytes for a pivot buffer facilitates a better algorithm below since we can do comparisons directly on the pivot. */ char *pivot_buffer = (char *) alloca (size); int max_thresh = MAX_THRESH * size; if (total_elems > MAX_THRESH) { char *lo = base_ptr; char *hi = lo + size * (total_elems - 1); stack_node stack[STACK_SIZE]; /* Largest size needed for 32-bit int!!! */ stack_node *top = stack + 1; while (STACK_NOT_EMPTY) { char *left_ptr; char *right_ptr; { char *pivot = pivot_buffer; { /* Select median value from among LO, MID, and HI. Rearrange LO and HI so the three values are sorted. This lowers the probability of picking a pathological pivot value and skips a comparison for both the LEFT_PTR and RIGHT_PTR. */ char *mid = lo + size * ((hi - lo) / size >> 1); if (CMP (mid, lo) < 0) SWAP (mid, lo, size); if (CMP (hi, mid) < 0) SWAP (mid, hi, size); else goto jump_over; if (CMP (mid, lo) < 0) SWAP (mid, lo, size); jump_over: COPY (pivot, mid, size); pivot = pivot_buffer; } left_ptr = lo + size; right_ptr = hi - size; /* Here's the famous ``collapse the walls'' section of quicksort. Gotta like those tight inner loops! They are the main reason that this algorithm runs much faster than others. */ do { while (CMP (left_ptr, pivot) < 0) left_ptr += size; while (CMP (pivot, right_ptr) < 0) right_ptr -= size; if (left_ptr < right_ptr) { SWAP (left_ptr, right_ptr, size); left_ptr += size; right_ptr -= size; } else if (left_ptr == right_ptr) { left_ptr += size; right_ptr -= size; break; } } while (left_ptr <= right_ptr); } /* Set up pointers for next iteration. First determine whether left and right partitions are below the threshold size. If so, ignore one or both. Otherwise, push the larger partition's bounds on the stack and continue sorting the smaller one. */ if ((right_ptr - lo) <= max_thresh) { if ((hi - left_ptr) <= max_thresh) /* Ignore both small partitions. */ POP (lo, hi); else /* Ignore small left partition. */ lo = left_ptr; } else if ((hi - left_ptr) <= max_thresh) /* Ignore small right partition. */ hi = right_ptr; else if ((right_ptr - lo) > (hi - left_ptr)) /* Push larger left partition indices. */ { PUSH (lo, right_ptr); lo = left_ptr; } else /* Push larger right partition indices. */ { PUSH (left_ptr, hi); hi = right_ptr; } } } /* Once the BASE_PTR array is partially sorted by quicksort the rest is completely sorted using insertion sort, since this is efficient for partitions below MAX_THRESH size. BASE_PTR points to the beginning of the array to sort, and END_PTR points at the very last element in the array (*not* one beyond it!). */ #define MIN(X,Y) ((X) < (Y) ? (X) : (Y)) { char *end_ptr = base_ptr + size * (total_elems - 1); char *run_ptr; char *tmp_ptr = base_ptr; char *thresh = MIN (end_ptr, base_ptr + max_thresh); /* Find smallest element in first threshold and place it at the array's beginning. This is the smallest array element, and the operation speeds up insertion sort's inner loop. */ for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size) if (CMP (run_ptr, tmp_ptr) < 0) tmp_ptr = run_ptr; if (tmp_ptr != base_ptr) SWAP (tmp_ptr, base_ptr, size); /* Insertion sort, running from left-hand-side up to `right-hand-side.' Pretty much straight out of the original GNU qsort routine. */ for (run_ptr = base_ptr + size; (tmp_ptr = run_ptr += size) <= end_ptr; ) { while (CMP (run_ptr, tmp_ptr -= size) < 0) ; if ((tmp_ptr += size) != run_ptr) { char *trav; for (trav = run_ptr + size; --trav >= run_ptr;) { char c = *trav; char *hi, *lo; for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo) *hi = *lo; *hi = c; } } } } return 1; }