diff options
author | Kevin Ryde <user42@zip.com.au> | 2001-06-12 03:37:57 +0200 |
---|---|---|
committer | Kevin Ryde <user42@zip.com.au> | 2001-06-12 03:37:57 +0200 |
commit | 883723ae30a880f6075da3f868d40753c30af809 (patch) | |
tree | 6d7756094fc37cf150f04e969f5df3a0db320adc /mpn | |
parent | 4fd49ab7b8e4d442d4feb79eb9c043c5de26eee9 (diff) | |
download | gmp-883723ae30a880f6075da3f868d40753c30af809.tar.gz |
* mpn/x86/divrem_1.asm: Amend some comments about P5 speed.
Diffstat (limited to 'mpn')
-rw-r--r-- | mpn/x86/divrem_1.asm | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/mpn/x86/divrem_1.asm b/mpn/x86/divrem_1.asm index 9bf7dfb70..ee62cbe6a 100644 --- a/mpn/x86/divrem_1.asm +++ b/mpn/x86/divrem_1.asm @@ -82,13 +82,13 @@ C but that algorithm has been found to suffer from the releatively poor C carry handling on K6 and too many auxiliary instructions. The C fractional part however could be done at about 13 c/l. C -C P5: Moving the load down to pair with the store might save 1 cycle, but -C that doesn't seem worth bothering with, since it'd be only a 2.2% -C saving. -C -C Again here the auxiliary instructions hinder a multiply-by-inverse, +C P5: Again here the auxiliary instructions hinder a multiply-by-inverse, C though there might be a 10-15% speedup available - +C +C It might be thought that moving the load down to pair with the store +C would save 1 cycle, but that doesn't seem to happen in practice, and +C in any case would be a mere 2.2% saving, so it hardly worth bothering +C about. defframe(PARAM_CARRY, 24) defframe(PARAM_DIVISOR,20) |