lp:adol-c

Branch merges

This import branch has no branches proposed for merge into it.

No branches dependent on this one.

Related source package recipes

No recipes using this branch. (?)

Create packaging recipe

Related snap packages

Related bugs

Related blueprints

221. By awalther on 2012-08-28

set derivative values array pointer to NULL in MittelmannDistCntrlNeumA_sparse

220. By awalther on 2012-08-28

set derivative values array pointer to NULL in LuksanVlek1_sparse

219. By utke on 2012-08-16

int and size_t are incompatible on some platforms - so be consistent and
list it in the manual as declared in the code with size_t

218. By utke on 2012-08-16

initialize for sanity so the 0 checks in forward/reverse make sense

217. By utke on 2012-08-16

put comments (doxygen style) to explain the use of the pointers

216. By utke on 2012-08-16

apparently is defined also for FOV

215. By utke on 2012-08-16

various typos (advance -> advanced) , and rephrased the section on differetiating external functions trying to clarify things

214. By kulshres on 2012-08-06

add missing files

Signed-off-by: Kshitij Kulshreshtha <email address hidden>

213. By kulshres on 2012-08-06

Merge branch 'master' of 'gitclone' into svn

It may require some work for hessian patterns and advector to work
properly, since they were changed simultaneously by Andrea and I

Signed-off-by: Kshitij Kulshreshtha <email address hidden>

The following commits were merged:

commit aeb288e939c7657aeaf75a8cbb833f9096ee2ce0
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Aug 6 16:46:20 2012 +0200

    correct compilation error in message printing

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 0d6e681b19d116bf0aca1d918fb567aed49e1766
Author: Benjamin Letschert <email address hidden>
Date: Tue May 29 15:53:45 2012 +0200

    for floor_op and ceil_op in INDO+TIGHT mode ind_dom should be set zero

commit 35689c69c49e02fb1b211e4f873851ce734c839e
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri May 4 16:34:49 2012 +0200

    add a comment for clarification

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 710b93301d98610239f17dc40ba8cba6651f2ea5
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri May 4 14:21:49 2012 +0200

    add copyright notices

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 830f2cc75ba823c2343f4a524eb3e90ab02bc278
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu May 3 13:59:32 2012 +0200

    Use the block allocation and don't write all locations to tape

    Now that we ensure block structure while allocating an advector
    we don't need to write all locations to tape in order to do
    subscripting operations. We can use arithmetic.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 8599c889e5cd684a7066600835df5f0c033eb339
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu May 3 13:57:39 2012 +0200

    Allocate advector ensuring contiguous locations

    use a dummy blocker to call the function before calling the vector
    constructor. Also use the fill constructor since the initialise
    constructor is experimental (for some reason) and a fill constructor
    is called internally by gcc anyway. This way we won't have to ensure a
    n+1 block and get holes.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit d62777e6070cf8b80b7b32616a874186e6463919
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu May 3 13:50:48 2012 +0200

    Always put the last freed block to the front

    This will make it efficienter if a block is deallocated right
    one after the other. There will only be a search the first time
    and all others will just work with the front of the list.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 33ae71a35fc04da062c4ba85659cbaf474caf1b8
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri Apr 27 17:36:41 2012 +0200

    changes in StoreManagemerLocintBlock to make it work right

    add StoreManagerLocintBlock to be the default StoreManager in
    tape_handling.cpp

    All existing ADOL-C programs need to be tested again.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit af30b922ee0a267555fdcdfe66d0823aa98688be
Author: Benjamin Letschert <email address hidden>
Date: Fri Apr 27 13:09:56 2012 +0200

    Adding new storem anagement StoreManagerLocintBlock to storemanager.h
    and to tape_handling.cpp

commit 7875317a03b08a2729879edfaec7ae5cd1363ef2
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Feb 29 08:49:25 2012 +0100

    convert tape stats to size_t from uint

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 11e535ae503f5961d0c62336d6942c96e4b7c3ba
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Jan 12 13:22:27 2012 +0100

    Revert "add missing arguments to ext_fct calls"

    This reverts commit 00622f92b5444e81f0c3760eeee0e14d4e8e4787.

    This was temporary and wasn't supposed to be uploaded, some parts will
    go into trunk.

commit 37ca7b37f3538153054dee40d3a75d560b3aa81f
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Jan 12 13:21:13 2012 +0100

    Revert "add safeguards and indo mode to ext_diff"

    This reverts commit 507df83d988e3d64848eb0350369094b0b2ce742.

    This was temporary for debugging and wasn't supposed to be uploaded

commit 507df83d988e3d64848eb0350369094b0b2ce742
Author: Kshitij Kulshreshtha <email address hidden>
Date: Sat Dec 31 23:46:23 2011 +0100

    add safeguards and indo mode to ext_diff

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 00622f92b5444e81f0c3760eeee0e14d4e8e4787
Author: Kshitij Kulshreshtha <email address hidden>
Date: Sat Dec 31 00:47:17 2011 +0100

    add missing arguments to ext_fct calls

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 1c1c7347867e583509716f7e2a8395875d0e3d53
Author: Benjamin Letschert <email address hidden>
Date: Fri Dec 16 14:18:24 2011 +0100

    Bugfix in copy_index_domain

commit 69df20b8f0240e80515c7f0df456f124cef70cf8
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Dec 5 11:32:58 2011 +0100

    shift around some code and add another constructor

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit b2c13089d4f5b45e7626e04c17b2f7bee08adc6a
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Nov 29 17:29:47 2011 +0100

    const no more needed after last bugfix

    add check for gradient w.r.t. rhs vector.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 78ea1e83eb17163fa0703e7b2b289659c4c6d42b
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Nov 28 17:13:23 2011 +0100

    bugfix in reverse mode during ref_copyout

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 3a7c70061118c89ad66afeec45d957dbdaf002b3
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Nov 28 17:12:39 2011 +0100

    Add one more assignment operator

    this one is used when an adubref is copied into another adubref
    as in from one advector to another advector, as this needs taping
    however we don't need to override copy-constructor, because
    construction can only be done when returning adubref.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 9c150d9aa825cf1e5964cd15c7d38b9a34984fce
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri Nov 25 12:49:14 2011 +0100

    changed aTmp*TRES to (aTmp==0)?0:(aTmp*TRES) and similar occurences to avoid nans by 0*inf

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 0efabbcb41d8b17f0fe55a35ca63ba93913bccbe
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri Nov 25 12:32:36 2011 +0100

    compile with -std=gnu99 to get both _GNU_SOURCE and _ISOC99_SOURCE

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 45fa79bd4d8ae0f4ef1e39b32f96271134e886f5
Author: Kshitij Kulshreshtha <email address hidden>
Date: Fri Nov 25 12:20:02 2011 +0100

    cast result of trunc to avoid warnings

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit b97ae47a80f7c0034b8f29d14656f75abcd51b4d
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Nov 24 17:53:11 2011 +0100

    counting down with unsigned variables is dangerous

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 5dd6a95ce0403f9ac4b262a9546f86e95a5489ac
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Nov 24 17:52:40 2011 +0100

    reverse the order of indep vars and add reverse call

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 2d794980215e1ae41af0e3cc6ed8a5e97507a2f7
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Nov 24 17:49:25 2011 +0100

    Add functions to print L and R and compute scalar product

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 889d35f671f2d2eea4dda182b0d903746994fbcd
Author: Kshitij Kulshreshtha <email address hidden>
Date: Thu Nov 24 17:47:31 2011 +0100

    bugfixes in luexam

    use less adoubles, loop variables for constant sized loops should
    remain inactive size_t. counting down with size_t is bad.

    const-ness of advectors matters during active subscripting,
    espacially when the lhs is a subscripted advector too, unless
    there is a expression involving badouble on the rhs.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 0dadd11bf743e7283a26613c1503778c6997654a
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Nov 23 17:18:05 2011 +0100

    tweak the example a bit

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 898afff0d43d86941125dd2e71857c619e3f28a4
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Nov 23 17:17:43 2011 +0100

    If location buffer must contain n+2 free spaces

    in case location buffer goes onto disk then check and write that
    block before doing a subscript operation
    if location buffer is smaller than n+2 then we cannot continue.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 1c2e4771972f0ce4db38bd959c23d18c061be568
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Nov 23 15:38:50 2011 +0100

    one more cast

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 71b506ae55f8d6754f5c2a1f800220cb5a5bb974
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Nov 23 15:27:53 2011 +0100

    Compilation fixes for sparse configuration

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 73334f20d51e707ee709a3602060aa3351dc6c29
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Nov 22 18:23:37 2011 +0100

    add an example that uses advector with permutations

    Lu factorization with pivoting is implemented here. The same traced
    procedure can be used for different matrices that require different
    pivots.

    Zos_forward has been verified, WIP to verify derivatives.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit c54da9e8cd03e99d9362cc1a7eb551101b5a4cb8
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Nov 22 18:20:35 2011 +0100

    some more inline helpers for advector

    the destructor need not contain anything.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit d403b2d2b7e591db7c353c8c3ce551e7d6174df0
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Nov 22 18:19:58 2011 +0100

    implement condassign for adubref

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 696e18b5dcf9534f7609243f085a508fba178b6c
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Nov 22 18:19:09 2011 +0100

    bugfix in -= operator for adubref

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit b7dd8285c6fe8262d07d0a5393c15bcca7eeae96
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Nov 21 13:47:48 2011 +0100

    implement reference operations in forward and reverse mode

    the referenced location is recorded in the taylor buffer of the
    reference variable, and read from the taylor buffer before any
    operation is performed. The operations are then performed directly
    on the referenced location.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit d546c517676a42065a98df782898f9e3472836b2
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Nov 21 13:45:49 2011 +0100

    add new operations for the tape to work with references

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 07508c193f53c6c5912485dcb517ca7c67b75234
Author: Kshitij Kulshreshtha <email address hidden>
Date: Mon Nov 21 13:45:17 2011 +0100

    Add a class to work as a reference to adouble and adub

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit e60d733c951a8a1d621a1401744fa03db78e9b92
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Mar 2 15:32:28 2011 +0100

    introduce active subscript operation in forward and reverse

    subscript operation as previously defined on advector was not quite
    correct in case of computed index that changed while being evaluated
    from tape. In this case we need the tape to record the subscript
    operation so that the correct taylor and adjoint values are propagated
    even if the index changed. This is especially true for index computation
    via the lookupindex() function, where the index depends on the arguments
    x and y which may change their values.

    This active subscript is a copy_value operation not a copy_ref one.
    So the result should not be used as an lvalue for any further operation.
    In case the result is used as an lvalue the original entry in the
    vector will not be altered, only the copied out version will be altered.

    Also the safe mode of integer propagation for sparsity pattern
    computation is not supported, only the tight mode is supported.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 5764beb91000483e02d4c6b52e67d177fff04a42
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Mar 2 13:59:12 2011 +0100

    zeroing out the store is not necessary at all

    and saves us some time by reducing memory access.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 29d3460e0c7a6d2839a25d8896c37fff98ad6a3c
Author: Kshitij Kulshreshtha <email address hidden>
Date: Wed Mar 2 13:57:33 2011 +0100

    oplate.h is not needed after installation

    so don't install it, and don't make it public api. it is totally
    internal to ADOL-C itself.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit 0e3ac08ba9c6ac0aa7d3820def9e62e079a83e44
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Mar 1 16:00:17 2011 +0100

    add missing include

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit b7364c7c2164981ce9e198b69f6aef8684388b43
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Mar 1 15:46:56 2011 +0100

    Add new files to buildsystem.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

commit aaefbdef3a6e00792ce0c57acff67e4f4ee3224b
Author: Kshitij Kulshreshtha <email address hidden>
Date: Tue Mar 1 15:38:59 2011 +0100

    New class advector for index lookup and subscripting

    This is a container using std::vector<adouble>. The subscripting
    operations return references to adoubles, nothing needs to be written
    onto the tape for this since the following operation using the adouble
    reference will do the right thing.
    The index lookup operation uses a sequence of n condassign_s ops which
    are taped, and returns the index which matches the requirement as another
    adouble.

    Signed-off-by: Kshitij Kulshreshtha <email address hidden>

212. By kulshres on 2012-08-06

remove DOS EOL characters

Signed-off-by: Kshitij Kulshreshtha <email address hidden>