Remote Mini-Workshop Series - Revision history

98.228.48.56: update static_tov runnability statement

2017-01-28T21:49:08Z

update static_tov runnability statement

198.254.245.169: remove claim that static TOV works

2016-12-29T14:16:52Z

remove claim that static TOV works

198.254.245.169: add snapshot functionality statement

2016-12-28T16:47:06Z

add snapshot functionality statement

Rhaas: add branch information about Zelmani branch

2016-12-27T15:12:29Z

add branch information about Zelmani branch

70.29.74.45: /* Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST */

2016-12-24T10:36:34Z

‎Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST

Eschnett: /* To Do */

2016-12-14T12:23:16Z

‎To Do

Eschnett: /* Source code */

2016-12-14T11:46:02Z

‎Source code

Eschnett: /* To-Do */

2016-12-14T09:30:24Z

‎To-Do

Eschnett: /* Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST */

2016-12-14T09:19:05Z

‎Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST

38.104.158.162: /* Mini-Workshop #1: Wed, Dec 7, 2016, 9:00 EST */

2016-12-14T09:16:57Z

‎Mini-Workshop #1: Wed, Dec 7, 2016, 9:00 EST

←Older revision		Revision as of 21:49, 28 January 2017
Line 283:		Line 283:
	A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).		A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).

−	~~<strike>~~The snapshot is able to evolve static_tov without crashing.~~</strike>~~	+	The snapshot in 138ef6bb of Zelmani is able to evolve static_tov without crashing.

←Older revision		Revision as of 14:16, 29 December 2016
Line 283:		Line 283:
	A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).		A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).

−	The snapshot is able to evolve static_tov without crashing.	+	<strike>The snapshot is able to evolve static_tov without crashing.</strike>

←Older revision		Revision as of 16:47, 28 December 2016
Line 282:		Line 282:

	A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).		A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).
		+
		+	The snapshot is able to evolve static_tov without crashing.

←Older revision		Revision as of 15:12, 27 December 2016
Line 280:		Line 280:

	As describe above, this would not yet work since the RHS requires con2prim to have run, and con2prim runs pointwise, so it needs to run after the state vector boundary conditions have been applied. Our current rough idea is to run con2prim only in the interior (and thus before the state vector boundary conditions), and synchronize the primitive variables together with the hydro state vector.		As describe above, this would not yet work since the RHS requires con2prim to have run, and con2prim runs pointwise, so it needs to run after the state vector boundary conditions have been applied. Our current rough idea is to run con2prim only in the interior (and thus before the state vector boundary conditions), and synchronize the primitive variables together with the hydro state vector.
		+
		+	A snapshot of the hydro code is in the rhaas/hydrofuntoy branch of Zelmani (git clone -b rhaas/hydrofuntoy git@bitbucket.org:zelmani/zelmani.git).

←Older revision		Revision as of 10:36, 24 December 2016
Line 257:		Line 257:
	* Update benchmark results		* Update benchmark results
	* Next meeting: Next Wed, 12:00 EST (tentatively); continue with FunHPC in Cactus		* Next meeting: Next Wed, 12:00 EST (tentatively); continue with FunHPC in Cactus
		+
		+	==Mini-Workshop #3 at Perimeter: Dec 21-22, 2016==
		+
		+	Roland Haas and Erik Schnetter met at Perimeter to design a parallelization scheme for relativistic hydro code with FunHPC. Here is a rough description of the outcome of our discussions.
		+
		+	In principle, this follows the scheme outlined above:
		+
		+	- the code is written in a mostly point-wise manner
		+	- we determined that the code accesses about 200 Bytes of memory per grid point, and thus a per-thread block size of 8x4x4 should work well
		+	- each such block is scheduled as a separate FunHPC thread
		+	- each scheduled routine then loops over the blocks and starts the threads
		+
		+	To overlap communication and computation, one needs to ensure that communication and computation occur "at the same time", i.e. are scheduled next to each other. This means that the hydro RHS needs to be scheduled in MoL_PostStep where the hydro state vector is synchronized. The details how this can happen are slightly involved since this involves also McLachlan setting the ADMBase variable, setting Tmunu, calling con2prim, etc., and this part hasn't been implemented yet.
		+
		+	The general design is:
		+
		+	1. Schedule the RHS calculations for each block. The blocks that need boundary data are scheduled, but wait for a trigger. Interior blocks can start right away.
		+	2. Schedule state vector synchronization.
		+	3. Toggle the trigger that allows the boundary block threads to start.
		+	4. Possibly schedule some other, independent routines.
		+	5. Wait for all RHS threads to finish.
		+
		+	As describe above, this would not yet work since the RHS requires con2prim to have run, and con2prim runs pointwise, so it needs to run after the state vector boundary conditions have been applied. Our current rough idea is to run con2prim only in the interior (and thus before the state vector boundary conditions), and synchronize the primitive variables together with the hydro state vector.

←Older revision		Revision as of 12:23, 14 December 2016
Line 256:		Line 256:
	* Check whether a vanilla checkout builds		* Check whether a vanilla checkout builds
	* Update benchmark results		* Update benchmark results
		+	* Next meeting: Next Wed, 12:00 EST (tentatively); continue with FunHPC in Cactus

@@ Line 244: / Line 244: @@
 ===Source code===
 * FunHPC and its dependencies, as described last week (see above)
-* FunHPC arrangements, basic thorns and examples (get it at https://bitbucket.org/eschnett/funhpc.cactus.git)
+* FunHPC arrangement, basic thorns and examples (get it at https://bitbucket.org/eschnett/funhpc.cactus.git)
 * Cactus flesh, branch eschnett/funhpc (mostly for startup)
 * Carpet, branch eschnett/funhpc (mostly to disable OpenMP)

@@ Line 213: / Line 213: @@
 * Maybe: Make FunHPC compile with Clang on Darwin
 * Maybe: Set up FunHPC on Bethe or Fermi (if Frank can't get access to Wheeler)
 * Add pointers to http://cppreference.com to wiki (for async, future)
@@ Line 226: / Line 225: @@
 * Add pointers to package web sites to build instructions
 * Put loop parallelization example onto wiki (and make it compile)
 ==Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST==

@@ Line 229: / Line 229: @@
 ==Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST==
-(Note the later time.)
+(By popular request, later in the day.)
 Topic: FunHPC in Cactus
@@ Line 238: / Line 238: @@
 * FunHPC vs. OpenMP (recap from last week)
 * Overview of current proof-of-concept code
 * How to download, build, run, benchmark
 * Next steps
@@ Line 248: / Line 249: @@
 * Kranc, branch eschnett/funhpc (to generate FunHPC-parallelized code)
 * McLachlan, branch eschnett/funhpc (contains new FunHPC-parallelized thorns with "_FH" suffix)

←Older revision		Revision as of 09:16, 14 December 2016
Line 226:		Line 226:
	* Add pointers to package web sites to build instructions		* Add pointers to package web sites to build instructions
	* Put loop parallelization example onto wiki (and make it compile)		* Put loop parallelization example onto wiki (and make it compile)
		+
		+	==Mini-Workshop #2: Wed, Dec 14, 2016, 12:00 EST==
		+
		+	(Note the later time.)
		+
		+	Topic: FunHPC in Cactus
		+
		+	Venue: Google Hangouts https://hangouts.google.com/call/jjkffrrvmnbhrooiyjxhfeb2ume
		+
		+	Agenda:
		+	* FunHPC vs. OpenMP (recap from last week)
		+	* Overview of current proof-of-concept code
		+	* How to download, build, run, benchmark
		+	* Next steps
		+
		+	===Source code===
		+	* FunHPC and its dependencies, as described last week (see above)
		+	* FunHPC arrangements, basic thorns and examples (get it at https://bitbucket.org/eschnett/funhpc.cactus.git)
		+	* Cactus flesh, branch eschnett/funhpc (mostly for startup)
		+	* Carpet, branch eschnett/funhpc (mostly to disable OpenMP)
		+	* Kranc, branch eschnett/funhpc (to generate FunHPC-parallelized code)
		+	* McLachlan, branch eschnett/funhpc (contains new FunHPC-parallelized thorns with "_FH" suffix)