The Bühlmann factors were cached in a thread-safe hashmap. It seemed
somewhat dubious that entering a critical section and doing a hash-lookup
would be significantly faster than a simple exp() call.
Indeed, in a very cache friendly test (16 entries, tight loop) calling the
factor() function 32 000 000 times from a different translation units we get:
- with cache: 604 ms
- without cache: 266 ms
Therefore, remove the cache. Given that 32 000 000 calls take only 266 ms,
it appears not sensible to try to optimize this function anyway.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
The parser had global state in the form of a linear regression
and the "plot_depth" variable. Collect that in the deco_state struct and
pass it down the call-chain. Move out the code to update the
regression data to not bother other callers of tissue_tolerance_calc().
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Signed-off-by: Robert C. Helling <helling@atdotde.de>
A number of architecture-dependent functions were declared in
dive.h. Move them to file.h so that not all file-manipulating
translation units have to include dive.h. This is a small step
in avoiding mass-recompilation on every change to dive.h
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Move the declarations of the "report_error()" and "set_error_cb()"
functions and the "verbose" variable to errorhelper.h.
Thus, error-reporting translation units don't have to import the
big dive.h header file.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
Remove a few cases of
void fun() {
...
}
While touching these functions, fix a few other whitespace
coding style violations.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
I incorrectly thought that 'ci_pointing_to_guiding_tissue' was the only
missing initialization, because that is the only one valgrind pointed at.
... that is, until I started looking at a few more dives, which showed
that there were other parts tht weren't initialized either, like
double tolerated_by_tissue[16];
double tissue_inertgas_saturation[16];
double crushing_onset_tension[16]; // total inert gas tension in the t* moment
so just make sure to clear the whole data structure, to avoid any random
behavior due to uninitialized deco state.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is another entry in the series to make more things
"const-clean" with the ultimate goal of merge_dive() take
const pointers.
This concerns functions taking pointers to events and
the fallout from making these const.
The somewhat debatable part of this commit might be
that get_next_event() is split in a two distinct
(const and non-const) versions with different names,
since C doesn't allow overloading. The linker should
recognize that these functions are identical and remove
one of them.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
In a previous commit, the get_gasmix_* functions were changed to
return by value. For consistency, also pass gasmix by value.
Note that on common 64-bit platforms struct gasmix is the size
of a pointer [2 * 32 bit vs. 64 bit] and therefore uses the
same space on the stack. On 32-bit platforms, the stack use
is probably doubled, but in return a dereference is avoided.
Supporting arbitrary gas-mixes (H2, Ar, ...) will be such an
invasive change that going back to pointers is probably the
least of our worries.
This commit is a step in const-ifying input parameters (passing
by value is the ultimate way of signaling that the input parameter
will not be changed [unless there are references to said parameter]).
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
...as the usuage is not anymore about a computer but
a momentary dive mode. Rename the end indicator as well.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
This is the second step for implementing bailout. The indirect
calls to fill_pressures through add_segment() (in deco.c) are
addressed. Bailout is now fully implemented in the dive log but
not in the dive planner.
1) The parameters to add_segment() are changed to take a
divemode as the second last parameter, and not a *dive.
2) Call to add_segment() in profile.c and in divelist.c are
adapted. In divelist.c some calls to add_segment were left
using dc-> divemode instead of possible bailout. This appears
tp be the most appropriate route.
3) The functions get_divemode_from_time() and get_next_divemodechange()
in dive.c have had some small changes.
4) The calls to get_segment(0 in planner.c were changed to reflect
the new parameter list, but not updated to reflect bailout. This
is the next step.
Signed-off-by: Willem Ferguson <willemferguson@zoology.up.ac.za>
Identify segements that fullfill the folllowing criteria for
the leading compartment:
He is off-gasing while N2 is on-gasing
Overall there is on-gasing
Add a line to the info box for those segments
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Since all qt-helpers are defined in qthelper.cpp, there seems to be
no reason to have two include files. By unifying the two files,
duplication and inconsistencies are removed. The C++-only part is
simply compiled away with #ifdefs.
Signed-off-by: Berthold Stoeger <bstoeger@mail.tuwien.ac.at>
For deep dives with long deco, the sum of deco stops could
overflow. This is prevent by turning it into long.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
For UI responsiveness, we need to be able to run the planner in the background. This needs the
planner state to be localized (and we need to pass a pointer around).
In order to not let too many lines overrun (and to save typing in the future)
I have renamed instances of struct deco_state to ds. Yes this should have gone
to a separate commit but I accidentally commit --amend'ed it.
Computing of planner variations is temporarily disabled.
Unlock the planner when returning early
So we don't deadlock in add dive and recreational mode (which
use the planner without actually planning).
Signed-off-by: Robert C. Helling <helling@atdotde.de>
... and reset deco information in profile ceiling computation.
The planner test then needs to know about the struct holding the deco
state.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
This option should have never been there. This is not how
gradient factors are supposed to work. It would only trick
users to use the wrong value..
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Not using lrint(f) when converting double/float to int
creates rounding errors.
This error was detected by TestParse::testParseDM4 failure
on Windows. It was creating rounding inconsistencies
on Linux too, see change in TestDiveDM4.xml.
Enable -Wfloat-conversion for gcc version greater than 4.9.0
Signed-off-by: Jeremie Guichard <djebrest@gmail.com>
This patch eliminates the difference between the saturation and
desaturation rates. This was probably once meant as a conservative
measure but the desaturation rate was increased rather than the
saturation rate (which is probably a typo, as reported by Stefan).
Since there is no good basis for this anyway, this patch sets
both factors to 1.0 (and if accepted the whole factor business
should be removed).
This makes our deco times slightly longer. But in the past,
we had introduced a 1.2% fudge factor in the critical radius
calculation to add conservatism and match the benchmark better.
Removing this fudge factor brings us close to the benchmarks.
Expected test values updated.
Reported-by: Stefan <sjti@gmx.net>
Signed-off-by: Robert C. Helling <helling@atdotde.de>
This resets the maximum crushing pressures and the maximal
ambient pressure between repetitive dives to prevent anomalies
that a dive produces a shorter deco when following another one
than without.
Reported-by: sfuchs@gmx.de
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
We have two prefernces determining the deco_mode (BUEHLMANN vs VPMB
vs RECREATIONAL): One for the planner (deco_mode) and one for
displaying dives (display_deco_mode). The former is set in the planner
settings while the latter is set in the preferences.
This patch clears up a confusion which of the two to use by introducing
a helper function that selects the correct variable.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
This patch fixes two bugs:
1) It first computes the effective gradient factors and then
composes the notes with the diveplan rather than the other way
around.
2) It does not try to fit a line through a single point.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
For each stop, this computes an effective gradient factor
that gives the same ceiling. Then, it does linear regression
to find values for GFlow and GFhigh that give a similar deco
profile.
Note that this optimises the average gradient factor. The
runtime however depends strongly at the gradient factor at
the last depth. So we don't necessarily to get the runtime
right.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Calculate gfline using the gradient factor that is set by the planner
preferences when in the planner, and by the general prefs when not in the
planner. This is achieved by doing the gradient factor calculation in dive.c,
where buehlmann_config is defined.
Previously, the gfline was calculated using the general preferences gfhigh and
gflow, even when in the planner.
Signed-off-by: Rick Walsh <rickmwalsh@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
... as those are needed for the heatmap and not only
for Buehlmann ceilings.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Separate the VPM-B conservatism preference into diveplan.vpmb_conservatism for
planning dives and prefs.vpmb_conservatism for profile ceiling display of
saved dives.
Signed-off-by: Rick Walsh <rickmwalsh@gmail.com>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
... and not just for Buehlmann. This makes the saturation
graphs meaningful for VPM-B.
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Change runtime table string from ZHL-16B to ZHL-16C to reflect he fact
that we use 5min as half-time for the fastest compartment rather than
4min.
Further more trade pow(2.0, ...) for exp().
Signed-off-by: Robert C. Helling <helling@atdotde.de>
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Having subsurface-core as a directory name really messes with
autocomplete and is obviously redundant. Simmilarly, qt-mobile caused an
autocomplete conflict and also was inconsistent with the desktop-widget
name for the directory containing the "other" UI.
And while cleaning up the resulting change in the path name for include
files, I decided to clean up those even more to make them consistent
overall.
This could have been handled in more commits, but since this requires a
make clean before the build, it seemed more sensible to do it all in one.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
