Progress Toward Accelerating CAM-SE

Progress Toward Accelerating CAM-SE.Jeff Larkin <larkin@cray.com>Along with:Rick Archibald, Ilene Carpenter , Kate Evans, Paulius Micikevicius , Jim Rosinski, Jim Schwarzmeier, Mark Taylor

BackgroundIn 2009 ORNL asked many of their top users: What sort of science would you do on a 20 Petaflops machine in 2012?Answer to come on next slideCenter for Accelerated Application Research (CAAR) established to determine:Can a set of codes from various disciplines be made to effectively use GPU accelerators with the combined efforts of domain scientists and vendorsEach team has a science lead, code lead, members from ORNL, Cray, Nvidia, and elsewhere

CAM-SE Target Problem1/8 degree CAM, using CAM-SE dynamical core and Mozart tropospheric chemistry. Why is acceleration needed to “do” the problem?When including all the tracers associated with Mozart atmospheric chemistry, the simulation is too expensive to run at high resolution on today’s systems. What unrealized parallelism needs to be exposed?In many parts of the dynamics, parallelism needs to include levels (k) and chemical constituents (q).

Profile of Runtime% of Runtime

Next StepsOnce the dominant routines were identified, standalone kernels were created for each.Early efforts tested PGI & HMPP directive, plus CUDA C, CUDA Fortran, and OpenCLDirectives-based compiler were too immature at the timePoor support for Fortran modules and derived typesDid not allow implementation at a high enough levelCUDA Fortran provided good performance while allowing us to remain in Fortran

Identifying ParallelismHOMME parallelizes both MPI and OpenMP over elementsMost of the tracer advection can also parallelize over tracers (q) and levels (k)Vertical remap is the exception, due to vertical dependence in levels.Parallelizing over tracers and sometimes levels while threading over quadrature points (nv) provides ample parallelism within each element to utilize GPU effectively.

StatusEuler_step & laplace_sphere_wk were straightforward to rewrite in CUDA FortranVertical Remap was rewritten to be more amenable to GPU (made it vectorize)Resulting code is 2X faster on CPU than original code and has been given back to the communityEdge Packing/Unpacking for boundary exchange needs to be rewritten (Ilene talked about this already)Designed for 1 element per MPI rank, but we plan to run with moreOnce this is node-aware, it can also be device-aware and greatly reduce PCIe transfersSomeone said yesterday: “As with many kernels, the ratio of FLOPS per by transfer determines successful acceleration.”

Status (cont.)Kernels were put back into HOMME and validation tests were run and passedThis version did nothing to reduce data movement, only tested kernel accuracyIn process of porting forward to current trunk and do more intelligent data movementCurrently reevaluating directives now that compilers have maturedDirectives-based vertical remap now slightly outperforms hand-tuned CUDAStill working around derived_type issues

ChallengesData Structures (Object-Oriented Fortran)Every node has an array of element derived types, which contains more arraysWe only care about some of these arrays, so data movement isn’t very naturalWe must essentially change many non-contiguous CPU arrays into a contiguous GPU arrayParallelism occurs at various levels of the calltree, not just leaf routines, so compiler must be able to inline leaves in order to use directivesCray compiler handles this via whole program analysis, PGI compiler may support this via inline library

Challenges (cont.)CUDA Fortran requires everything live in the same moduleMust duplicate some routines and data structures from several module in our “cuda_mod”Insert ifdefs that hijack CPU routine calls and forward the request to matching cuda_mod routinesSimple for user, but developer must maintain duplicate routinesHey Dave, when will this get changed? ;)

Until the Boundary Exchange is rewritten, euler_step performance is hampered by data movement. Streaming over elements helps, but may not be realistic for the full code.

With data transfer, laplace_sphere_wk is a wash, but since all necessary data is already resident from euler_step, kernel only time is realistic.

Vertical remap rewrite is 2X faster on the CPU and still faster on GPU. All data already resident on device from euler_step, so kernel-only time is realistic.

Future WorkUse CUDA 4.0 dynamic pinning of memory to allow overlapping & better PCIe performanceMove forward to CAM5/CESM1No chance of our work being used otherwiseSome additional, small kernels are needed to allow data to remain residentCheaper to run these on the GPU than to copy the dataReprofile with accelerated application to identify next most important routinesChemisty implicit solver is expected to be nextPhysics is expected to require mature, directives-based compilerRinse, repeat

ConclusionsMuch has been done, much remainsFor a fairly new, cleanly written code, CUDA Fortran was tractable.HOMME has very similar loop nests throughout, that was key to making this possibleStill results in multiple code paths to maintain, so we’d prefer to move to directives for the long-runWe believe GPU accelerators will be beneficial for the selected problemWe hope that it will also benefit a wider audience (CAM5 should help this)

Progress Toward Accelerating CAM-SE

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Progress Toward Accelerating CAM-SE (20)

More from Jeff Larkin (12)

Recently uploaded (20)

Progress Toward Accelerating CAM-SE

Editor's Notes