[Gmp-commit] /home/hgfiles/gmp: Clean out some outdated comments.

[mercurial at gmplib.org]

details:   /home/hgfiles/gmp/rev/c833ceac4f33
changeset: 13405:c833ceac4f33
user:      Torbjorn Granlund <tege at gmplib.org>
date:      Thu Jan 28 08:41:23 2010 +0100
description:
Clean out some outdated comments.

diffstat:

 mpn/generic/mu_div_qr.c    |  14 --------------
 mpn/generic/mu_divappr_q.c |  18 ------------------
 2 files changed, 0 insertions(+), 32 deletions(-)

diffs (52 lines):

diff -r e1f73c9da681 -r c833ceac4f33 mpn/generic/mu_div_qr.c
--- a/mpn/generic/mu_div_qr.c	Thu Jan 28 08:09:31 2010 +0100
+++ b/mpn/generic/mu_div_qr.c	Thu Jan 28 08:41:23 2010 +0100
@@ -40,20 +40,6 @@
 
  Things to work on:
 
-  * Passing k isn't a great interface.  Either 'in' should be passed, or
-    determined by the code.
-
-  * The current mpn_mu_div_qr_itch isn't exactly scientifically written.
-    Scratch space buffer overruns are not unlikely before some analysis is
-    applied.  Since scratch requirements are expected to change, such an
-    analysis will have to wait til things settle.
-
-  * This isn't optimal when the remainder isn't needed, since the final
-    multiplication could be made special and take O(1) time on average, in that
-    case.  This is particularly bad when qn << dn.  At some level, code as in
-    GMP 4 mpn_tdiv_qr should be used, effectively dividing the leading 2qn
-    dividend limbs by the qn divisor limbs.
-
   * This isn't optimal when the quotient isn't needed, as it might take a lot
     of space.  The computation is always needed, though, so there is no time to
     save with special code.
diff -r e1f73c9da681 -r c833ceac4f33 mpn/generic/mu_divappr_q.c
--- a/mpn/generic/mu_divappr_q.c	Thu Jan 28 08:09:31 2010 +0100
+++ b/mpn/generic/mu_divappr_q.c	Thu Jan 28 08:41:23 2010 +0100
@@ -40,24 +40,6 @@
 
  Things to work on:
 
-  * Passing k isn't a great interface.  Either 'in' should be passed, or
-    determined by the code.
-
-  * The current mpn_mu_div_qr_itch isn't exactly scientifically written.
-    Scratch space buffer overruns are not unlikely before some analysis is
-    applied.  Since scratch requirements are expected to change, such an
-    analysis will have to wait til things settle.
-
-  * This isn't optimal when the remainder isn't needed, since the final
-    multiplication could be made special and take O(1) time on average, in that
-    case.  This is particularly bad when qn << dn.  At some level, code as in
-    GMP 4 mpn_tdiv_qr should be used, effectively dividing the leading 2qn
-    dividend limbs by the qn divisor limbs.
-
-  * This isn't optimal when the quotient isn't needed, as it might take a lot
-    of space.  The computation is always needed, though, so there is no time to
-    save with special code.
-
   * The itch/scratch scheme isn't perhaps such a good idea as it once seemed,
     demonstrated by the fact that the mpn_invertappr function's scratch needs
     mean that we need to keep a large allocation long after it is needed.