- clean up a few warnings
- use an ExprLib tool rather than redoing it ourselves in ParseTools.cc
19 lines of code changed in 4 files:
Changed how multi envrionment model is parsed and registered
54 lines of code changed in 4 files:
Reverting earlier change on diffusion coefficient. Turns out that
density was being applied upstream in mu_t.
However, keeping the factor of "2" in Dirichlet boundary treatment which
seems to be mathematically correct.
Answers will still change but the acid_base problem shouldn't diverge.
21 lines of code changed in 2 files:
remove restart on a couple of local tests.
2 lines of code changed in 1 file:
Fix the ghost node declaration for gmass in updateCohesiveZones.
2 lines of code changed in 1 file:
Improved error checking for command line arguments.
58 lines of code changed in 1 file:
rip out some dead code.
28 lines of code changed in 2 files:
Fix invalid device ordinal error.
1 lines of code changed in 1 file:
Fix PoissonGPU1 Example to work with non-UVA environments (compute capability <= 1.2).
We now support multiple Nvidia GPU compute capabilities (1.2, 1.3, 2.0, 2.1, 3.0), although 1.2 and 1.3 will be slower.
18 lines of code changed in 1 file:
Fix AC_HELP_STRING for --enable-gencode
Note: Configuring with CUDA is still only:
--with-cuda=<DIR with CUDA installation>
But but a specific architecture to generate code for can be specified with, e,g:
--with-cuda=/usr/local/cuda
--enable-gencode=20
4 lines of code changed in 3 files:
Modify CUDA support in build system to allow GPU code (PTX or CUBIN) generation for specific compute capabilities: 1.2, 1.3, 2.0, 2.1, 3.1
* We need to be able to explicitly generate code for Fermi and Kepler, and also for earlier compute capabilities, even non-UVA environments.
Generated new configure script.
33 lines of code changed in 3 files:
For now, don't use cudaMemcpyDefault when doing CUDA memcopies.
Explicitly use cudaMemcpyHostToDevice and cudaMemcpyDeviceToHost for non-UVA environments.
3 lines of code changed in 1 file:
Rearrangement of diffusion term to be more consistent. Also changing
the way diffusion is handled at Dirichlet boundaries to be more
consistent with the the bc. Also added a few comments.
This will change stuff:
MPMARCHES-opt tests
:heptane_pipe:
:coal_table_pipe:
:intrusion_test:
74 lines of code changed in 3 files:
Added if statement in multi environment to avoid NaNs
6 lines of code changed in 1 file:
Added aggregation kernel for the PBE in Wasatch
641 lines of code changed in 5 files:
Comment out some unused code, and replace a hardcoded value for a reference
pressure with a input parameter.
6 lines of code changed in 2 files:
Use subcycling of F all the time. This may change the answers in the RT.
27 lines of code changed in 1 file:
Making the previous changes related to the SUB_CYCLE_F option more efficient and
robust.
15 lines of code changed in 1 file:
Bug fix for QMOM
7 lines of code changed in 1 file:
Can't use a comma in a msg string.
2 lines of code changed in 2 files:
M configure
M configure.ac
Updated the PETSc check... specifically, the previous code was using a
for loop that actually had empty parameters (but ran once with a blank
'' input). Anyway, the logic and code was confusing, so hopefully it
is cleaner now. The main reason for this update is now configure will
check if you are using PETSc v3.x and if so, if you specify
PETSC_ARCH, it will error out and tell you not to do this.
Previously, it would just fail without a 'good' reason.
214 lines of code changed in 2 files:
add a regression test showing how one can use a symmetry boundary condition in Wasatch.
2 lines of code changed in 1 file:
Final touches on the post-table lookup, inert stream mixing. Users can
now specify N independent inert streams to the property calculation
using <post_mix> in the specific table lookup section of the input file.
174 lines of code changed in 4 files:
Problem setup for post-combustion, inert stream mixing. It doesn't do
anything at this point.
49 lines of code changed in 2 files:
Bug found in the temperature calculation for two-stream mixing.
This will change answers.
13 lines of code changed in 1 file:
Back out in CUDA include: device_launch_parameters.h
No guard on the header with CUDA distribution.
0 lines of code changed in 1 file:
add regression test for Smagorinsky model.
2 lines of code changed in 1 file:
Add another CUDA include: device_launch_parameters.h
1 lines of code changed in 1 file:
Added an effective viscosity expression for particle flow
372 lines of code changed in 3 files:
cleaned up some sloppy coding.
Fixed compiler warnings.
Added virtual destructor of base class
Now deleting base class
7 lines of code changed in 1 file:
initialize a temporary field in the shear stress calculation that was causing some instability in the LES model for certain problems. Alex, this should fix the issues you were having when running LES.
4 lines of code changed in 2 files:
use attach_dependency to add the turbulent viscosity to the molecular viscosity instead of pasing it to the stress tensor expression.
19 lines of code changed in 4 files:
Fix seg fault in optimized build of GPU Poisson1 example.
This is the last of the failing GPU regression tests since the switch to CUDA driver API. GPU tests should all pass now.
0 lines of code changed in 1 file:
* Finish converting from CUDA runtime to driver API.
* Build system now creates a "ptx" directory (at same level as opt or dbg) for Nvidia assembler/bytecode. PTX/CUBIN can be generically referenced and loaded from component code.
* All GPU-enabled component code (except ICE) has been converted to use this new method.
* Updated configure script.
28 lines of code changed in 8 files:
Fixing the density guess for intrusions that act as inlets.
87 lines of code changed in 5 files:
fixed a bug that vertically offset the image by 1 pixel
9 lines of code changed in 1 file:
The user can now specify any number of materials or 'a/all' for all materials
in the uda.
Now scaling data:
8bit: 0->255
16bit: 0->65025
53 lines of code changed in 1 file:
The #define SCATTER was left uncommented in the
last commit. It doesn't change the answers in
the RT either way, since the default is a scattering
coefficient of 0 (no scattering), but it is safer
and faster to have it undefined unless we're
actually running a case with scattering.
1 lines of code changed in 1 file:
Allow for scattering in all parts of Ray.cc (virtual radiometer,
divQ solver, flux solver) by passing the pointer to the MTwister object to
updateSumI.
17 lines of code changed in 2 files:
add function that calculates an approximate walldistance field. this uses the poisson equation framework recently committed. I still need to do proper normalization of the walldistance calculation.
187 lines of code changed in 3 files:
Added capability for benchmark 4 which is an
isotropic scattering case described by Siegel.
Benchmark 4 can be run with the input file below,
and by uncommenting the compiler directive to
define SCATTER in Ray.cc. I'll run one more test,
and this input file will be ready for the RT.
Also, sometime between March 24,2012 and yesterday,
the temperature for benchmark 3 was changed from
1000 * abskg[c]
to
1000 * abskg[c] * 1000 * abskg[c].
I changed it back in this commit.
18 lines of code changed in 2 files:
Second try at a new API for ExprLib FieldManager
402 lines of code changed in 34 files:
Updating the soot model to be consistent with the published version.
66 lines of code changed in 3 files:
Revert "Updates due to recent API change in ExprLib"
This reverts commit eeefe0834374a9fd83fb5ba17a179aeba22a8a1c.
385 lines of code changed in 33 files:
Initialize incident and net values for the
map outside the cell iterator for speed.
5 lines of code changed in 1 file:
Updates due to recent API change in ExprLib
356 lines of code changed in 33 files:
Minor inconsequential changes
1 lines of code changed in 2 files:
Update documentation for the Ellipse geometry.
9 lines of code changed in 1 file:
Further additions/optimizations related to CUDA Driver API approach:
* Added NVCC_MODULE_FLAGS for automatic PTX module optimization.
* Added #pragma unroll in kernels
* Updated configure script
This all works well on Keenleland IDS. Still need to test on TitanDev (allocation is spent)
34 lines of code changed in 6 files:
Add documentation in configVars.mk.in about multiple CUDA compiler (NVCC) phases, attempting to explain multiple compile lines and how the generated device-only PTX (Nvidia assembler) code is used in explicitly registering PTX modules via the CUDA Driver API for kernel (grid) launches. Also fixed some compiler warnings about unused variables.
16 lines of code changed in 2 files:
Whoops. Accidently deleted a bunch of code in that last commit.
132 lines of code changed in 1 file:
Bug fix for soot model with more than one material.
Need to allocate scalars in dummy solve before the variance calculation
(which needs the scalar).
Curse you dummySolve!
17 lines of code changed in 2 files:
cleanup some incorrect doxygen tags and merge some documentation.
24 lines of code changed in 11 files:
Allow users to solve poisson equations of their choice in Wasatch. Users can specify a RHS expression and feed it into the poisson system. They can specify boundary conditions for each poisson equation as well as different solver parameters.\n
757 lines of code changed in 11 files:
Simplify nightly vs local GPU RT.
1 lines of code changed in 1 file:
Fix non-CUDA configured compiler errors.
6 lines of code changed in 2 files:
* Convert from CUDA runtime API to driver API. Eliminates the issues with NVCC's inability to process GNU tr1/type_traits headers.
* This change allows using ANY host compiler, and also separates host from device code entirely. No more <<<arg1, arg2, agr3, arg4>>> to invoke kernels.
* .cu files solely contain kernels and device functions now.
* GPU-enabled ICE code is out of the mix temporarily. Will re-integrate once this is all working well.
* Build system's CUDA support has been altered considerably, with new configure script.
NOTE: this commit will break GPU RT tests until a generic/robust way to locate the generated PTX (Nvidia bytecode) can be finished.
1748 lines of code changed in 22 files:
Fix a warning in UCNH, reduce spew from the factories.
9 lines of code changed in 3 files:
Fix some computes/requires issues. Good to run your code from a debug build sometimes...
Also, suppress some of the spew from the EOS factory.
19 lines of code changed in 4 files:
Added a multi environment mixing model for the moment transport equations, along with a sample input file
625 lines of code changed in 8 files:
Committed a cleaned version of Arenisca with detailed comments
1067 lines of code changed in 1 file:
Add framework for LES eddy viscosity models. Implement constant Smagorinsky model and the Wall-Adapating Local Eddy-Viscosity model (WALE).
995 lines of code changed in 16 files:
AMRSimulationController.cc
computeStableTimestep()
- added comment
- added spaces
ImpMPM
computeStableTimestep()
- compute a delT on every level even if no tasks are run on that level by this component.
I think this is a bug at the infrastructure level.
- formatting
62 lines of code changed in 2 files:
Preliminary scattering capabilities for
isotropic scattering. This method assumes
homogeneous scattering coefficients. Alternative
methods exist that can handle non-homogeneous
scattering coefficients. Such methods require
a random number call
for every step of every ray, whereas the current
method requires only one random number for scattering
per ray. After further verification testing, I will
submit an input file for regression testing.
52 lines of code changed in 1 file:
Re-enable cudaHostUnregister() for host memory that was page-locked with cudaHostRegister().
Also add a little documentation.
11 lines of code changed in 2 files:
Back out changes to MPM PetscSolver for now.
Will fix this issue offline and re-commit.
39 lines of code changed in 1 file:
-added bulletproofing to inputs.
- Write out image so origin is in bottom left corner
- yanked 'orientation' option.
- fixed sample format for 8 & 16-bit images.
find_CC_ave()
- allocate for max_matl_index instead of matl.size().
45 lines of code changed in 1 file:
Added ability to write out 8, 16 or 32 bit tiff images.
Added ability to write out a collection of image slices as a volume or
as individual files.
The verification test is now a command line option.
473 lines of code changed in 1 file:
Added struct to allow for incident and net
flux computations.
16 lines of code changed in 2 files:
Cleaned up comments and indentation. Renamed a variable
for clarity in preparation to add net flux as well
as incident flux.
120 lines of code changed in 1 file:
Explicit stdio.h include for fprintf() in cuda_defs.h for CUDA error handling macros.
This is needed for gcc 4.6.x on TitanDev.
2 lines of code changed in 1 file:
Enable cudaHostRegister() code in GPU scheduler. This works well with CUDA 4.1. Should be a slight performance boost.
The issue with 4.0 was that the allocated host memory needed to be page aligned to work with cudaHostRegister()
Also eliminated explicit linking of libcuda until we're using driver API. Still need libcudart link.
- New configure script generated.
25 lines of code changed in 4 files:
Documentation for use of maps for storage of boundary fluxes.
0 lines of code changed in 4 files:
May 2012 »