Open komatits opened 9 years ago
Hi Dimitri and Etienne,
I've added a new directory, src/postprocess
, and a new routine to go in it, sum_kernels_ascii.f90
. In so far as similarities between packages allow, the routine is modeled after sum_kernels.f90
from SPECFEM3D. Also, it has the nice property that it is independent of any Par_file parameters (aside from SAVE_ASCII_KERNELS).
To write 2D versions of the other utilities, we would probably need to change the kernel binary file format. With this change, we would lose of a certain nice symmetry between ascii and binary formats, but without this change, we would not be able to use any existing command line interfaces, and would have to resort to a not very clean and not very flexible approach.
What do you think, is it still alright on your end if I were to make this change to the kernel binary formats?
I had wanted to make this change a while before but got sidetracked--sorry for the delay.
Ryan
Hi Ryan,
Yes, very good idea to make that change, it will be very useful.
Thanks, Dimitri.
On 03/06/2015 03:03 AM, rmodrak wrote:
Hi Dimitri and Etienne,
I've added a new directory, |src/postprocess|, and a new routine to go in it, |sum_kernels_ascii.f90|. In so far as similarities between packages allow, the routine is modeled after |sum_kernels.f90| from SPECFEM3D. Also, it has the nice property that is independent of any Par_file parameters (aside from SAVE_ASCII_KERNELS).
To write 2D versions of the other utilities, we would probably need to change the kernel binary file format. With this change, we would lose of a certain nice symmetry between ascii and binary formats, but without this change, we would not be able to use any existing command line interfaces, and would have to resort to a not very clean and not very flexible approach.
What do you think, is it still alright on your end if I were to make this change to the kernel binary formats?
I had wanted to make this change a while before but got sidetracked--sorry for the delay.
Ryan
— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem2d/issues/223#issuecomment-77493407.
Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr
Hi Dimitri and Etienne,
Thanks, I've started making these changes.
To minimize disruption, I've added a new parameter to constants.h, NEW_BINARY_FORMAT, currently turned off by default, which will allow for the old format to be phased out and the new format to be phased in.
Ryan
Hi Ryan,
Excellent, thanks a lot. Very useful.
Dimitri.
On 03/08/2015 09:04 PM, rmodrak wrote:
Hi Dimitri and Etienne,
Thanks, I've started making these changes.
To minimize disruption, I've added a new parameter to constants.h, NEW_BINARY_FORMAT, currently turned off by default, which will allow for the old format to be phased out and the new format to be phased in.
Ryan
— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem2d/issues/223#issuecomment-77771001.
Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr
Etienne is currently doing it I think (or else using Ryan's tools).
Vadim @vmont is currently doing it.
Let us first wait for the new 3D tomography tools to be ready and released.
Will be done by Etienne @EtienneBachmann when he arrives at Princeton, based on the nice merged tool we have with Vadim @vmont on our private SVN repository (soon to be merged into the official Git repository of SPECFEM).
Done by Vadim @vmont under the name "SOUNDVIEW", see https://doi.org/10.1088/1361-6560/aa7e5a
When we switch to Vadim's unified inversion framework in the future (see e.g. our "SOUNDVIEW" inversion framework at https://doi.org/10.1088/1361-6560/aa7e5a) this will be done / fixed automatically.
There is currently no 2D version of these tools in Git, although there was probably one at some point (Tromp Tape Liu 2005 was in 2D); but it seems these tools have been lost (or never existed, since Tromp et al (2005) showed kernels but (if I remember correctly) did not use them to iterate i.e. did not perform inversion).
Ryan @rmodrak talked about writing one based on the 3D tools, once these 3D tools are cleaned and unified (there are currently two different versions of these 3D tools: one in 3D_Cartesian and one in 3D_GLOBE). Etienne Bachmann @EtienneBachmann is also interested and will participate.