diff options
author | Kevin Ryde <user42@zip.com.au> | 2001-07-10 01:50:37 +0200 |
---|---|---|
committer | Kevin Ryde <user42@zip.com.au> | 2001-07-10 01:50:37 +0200 |
commit | 4eeb1fd014f5033e96edf9ffefe3d1ab1fb60fd9 (patch) | |
tree | 62e78e30443b3784bd75eaebca30ab0a8ba05862 /gmp-impl.h | |
parent | bfc1db42872f0afc735ca3f6c49fbc62b57abdd0 (diff) | |
download | gmp-4eeb1fd014f5033e96edf9ffefe3d1ab1fb60fd9.tar.gz |
More of:
* gmp-impl.h (MPN_NORMALIZE): Add notes on x86 repe/scasl slow.
Diffstat (limited to 'gmp-impl.h')
-rw-r--r-- | gmp-impl.h | 16 |
1 files changed, 9 insertions, 7 deletions
diff --git a/gmp-impl.h b/gmp-impl.h index e065cee1d..eaf928e9d 100644 --- a/gmp-impl.h +++ b/gmp-impl.h @@ -644,13 +644,15 @@ _MPN_COPY (d, s, n) mp_ptr d; mp_srcptr s; mp_size_t n; /* On the x86s repe/scasl doesn't seem useful, since it takes many cycles to start up and would need to strip a lot of zeros before it'd be faster - than a simple cmpl loop. std/repe/scasl/cld stripping no zeros runs as - follows, - cycles - P5 18 - P6 46 - K6 36 - K7 21 + than a simple cmpl loop. Here are some times in cycles for + std/repe/scasl/cld and cld/repe/scasl (the latter would be for stripping + low zeros). + + std cld + P5 18 16 + P6 46 38 + K6 36 13 + K7 21 20 */ #ifndef MPN_NORMALIZE #define MPN_NORMALIZE(DST, NLIMBS) \ |