This is hard to do well. In git, a branch is just an auto-advancing pointer to a commit, and a commit can have any number of branch names on it. Consider this case:
your master: y1---y2---y3
/
master: a---b---c---d---e
\
feature: f1---f2---f3---f4
You checked out branch "master" at c, and committed y1, y2, and y3. Your history thus looks like a b c y1 y2 y3. Meanwhile master has advanced to d and e, but someone created a feature branch and committed f1 through f4 based on c. Git has no way to determine that your branch came from master instead of feature, so at best you will have a choice of branches to merge.
Were you to do this automatically you would have to apply a heuristic about picking the shortest branch, or the longest branch, or the one with the most/least commits, or something else like that. Naturally, since there are so many options, it's not really a good choice for a git built-in function. However, using git's "plumbing" functions, you can write your own:
#!/bin/bash
# Tries to determine a good merge base from among all local branches.
# Here used, a "good" merge base is the one sharing the most recent commit
# on this branch. This function will exit 1 if no branch is found,
# or exit 2 in a tie.
#
# Untested - use at your own risk.
MAX_SEARCH=20 # only search the last 20 commits on this branch
FOUND=0
LAST_BRANCH=
# iterate through the commits, most recent first
for COMMIT in $(git rev-list --max-count=$MAX_SEARCH HEAD); do
# check every local branch
for BRANCH in $(git for-each-ref --format="%(refname)" refs/heads); do
# look for the commit in that branch's history
if (git rev-list $BRANCH | fgrep -q COMMIT); then
echo $BRANCH
FOUND=$((FOUND + 1))
LAST_BRANCH="$BRANCH"
fi
done
if [ $FOUND -gt 1 ]; then
# more than one choice; exit
exit 2
elif [ $FOUND -eq 1 ]; then
git merge $LAST_BRANCH
exit 0
fi
done
exit 1 # could not find a parent
