caleb-brown/redict
1
0
Fork 0
mirror of https://codeberg.org/redict/redict.git synced 2025-01-23 00:28:26 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

ARM: Avoid fast path for BITOP.

Browse Source 
GCC will produce certain unaligned multi load-store instructions
that will be trapped by the Linux kernel since ARM v6 cannot
handle them with unaligned addresses. Better to use the slower
but safer implementation instead of generating the exception which
should be anyway very slow.
This commit is contained in:
Salvatore Sanfilippo 2017-02-19 15:07:08 +00:00
parent 4e9cf4cc7e
commit 7329cc3981
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/bitops.c
View File

@ -654,8 +654,11 @@ void bitopCommand(client *c) {
/* Fast path: as far as we have data for all the input bitmaps we /* Fast path: as far as we have data for all the input bitmaps we
* can take a fast path that performs much better than the * can take a fast path that performs much better than the
* vanilla algorithm. */ * vanilla algorithm. On ARM we skip the fast path since it will
* result in GCC compiling the code using multiple-words load/store
* operations that are not supported even in ARM >= v6. */
j = 0; j = 0;
#ifndef __arm__
if (minlen >= sizeof(unsigned long)*4 && numkeys <= 16) { if (minlen >= sizeof(unsigned long)*4 && numkeys <= 16) {
unsigned long *lp[16]; unsigned long *lp[16];
unsigned long *lres = (unsigned long*) res; unsigned long *lres = (unsigned long*) res;
@ -716,6 +719,7 @@ void bitopCommand(client *c) {
} }
} }
} }
#endif
/* j is set to the next byte to process by the previous loop. */ /* j is set to the next byte to process by the previous loop. */
for (; j < maxlen; j++) { for (; j < maxlen; j++) {