diff options
author | vlefevre <vlefevre@280ebfd0-de03-0410-8827-d642c229c3f4> | 2016-03-08 15:57:43 +0000 |
---|---|---|
committer | vlefevre <vlefevre@280ebfd0-de03-0410-8827-d642c229c3f4> | 2016-03-08 15:57:43 +0000 |
commit | ac298c75b3b4f67ca66b0e933f7a5ef839522322 (patch) | |
tree | f1a5703c01f7c48e2828b05e753f8e2d405b6907 /doc/sum.txt | |
parent | 3838940d79f6c27ae76e196234d07c41e2adfee2 (diff) | |
download | mpfr-ac298c75b3b4f67ca66b0e933f7a5ef839522322.tar.gz |
[doc/sum.txt] Corrected 2 typos found by PZ.
git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@10210 280ebfd0-de03-0410-8827-d642c229c3f4
Diffstat (limited to 'doc/sum.txt')
-rw-r--r-- | doc/sum.txt | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/doc/sum.txt b/doc/sum.txt index bcb5bf2b3..5f33a15f6 100644 --- a/doc/sum.txt +++ b/doc/sum.txt @@ -265,7 +265,7 @@ memory area. Two's complement is preferred to the sign + magnitude representation because the signs of the temporary (and final) results are not known -in advanced, and the computations (additions and subtractions) in +in advance, and the computations (additions and subtractions) in two's complement are more natural here; the usual sign + magnitude representation for the accumulator would have been possible with about the same number of elementary operations (even in case of cancellation), @@ -328,7 +328,7 @@ In summary: The sum_raw function will work in a fixed-point accumulator, having a fixed precision (= size in limbs * GMP_NUMB_BITS). It will accumulate the bits of the inputs whose exponents are in ⟦minexp,maxexp⟦, where -maxexp − minexp is less that the precision of the accumulator: as said +maxexp − minexp is less than the precision of the accumulator: as said above, we need some additional bits in order to avoid overflows during the accumulation. On the entry, the accumulator may already contain a value from previous computations (it is the caller that clears it if |