GeantV Geometry: SIMD abstraction and interfacing with CUDA · C-like abstract kernels Backend...

PH - SFT

GeantV Geometry: SIMD abstraction and interfacing with CUDA

Johannes de Fine Licht ([email protected]) Sandro Wenzel ([email protected]) !Thanks to Michal Husejko and Romain Wartel at TechLab for providing hardware.

mailto:[email protected]


Johannes de Fine Licht ([email protected])

PH - SFT

Abstracted algorithms

Write common code for multiple types of SIMD “backends” (Vc/Cilk/CUDA).

Eases code maintenance; requires only one kernel to be written for a given functionality

Must retain performance of a low level implementation

��2PH - SFT



PH - SFT ��3

Illustrating scalar/SIMD abstraction and kernelsSingle particle

interface

C-like abstract kernels

Vector interface

External CUDA kernels



PH - SFT ��3


interface


Vector interface


Vc Looper Thread access

Scalar Looper Cilk Plus Looper



PH - SFT ��3


interface


Backend instantiation

Cilk Plus Backend

Vector interface




Scalar Backend Vc Backend CUDA Backend



PH - SFT ��3


interface


Backend instantiation

Cilk Plus Backend

Vector interface




C-like specialized kernels

Scalar Backend Vc Backend CUDA Backend

Kernel instantiation



PH - SFT

Generic kernels on abstract types

��4

Generic programming; write algorithms in a generic way on abstract types

!Implementation depends on the backend !Backend determines the types on which the high level operations are performed



PH - SFT

Implementations wrapped in types

Operations are abstracted

Backends implement all operations

Vc already supports arithmetics

CUDA uses primitive types

Wrapper structs are implemented if needed

��5



PH - SFT

Interfacing with CUDA

��6

Involving the GPU should be easy !Principle of common code !Issue: GPU code needs nvcc, but we want a different compiler for CPU

vs.GNU/Clang:

C++11 Vc

Cilk

nvcc: CUDA



PH - SFT

Host memory !vecgeom::CudaManager

vecgeom::UnplacedBox vecgeom::LogicalVolume …

Dual namespaces

��7

GPU memory !vecgeom_cuda::UnplacedBox vecgeom_cuda::LogicalVolume …

CUDA kernel

namespace vecgeom

namespace vecgeom_cuda

namespace vecgeom

Utilize distinct namespaces but same code for CPU and GPU !Classes live in one or both environments !Provide abstraction to interface between environments



PH - SFT



Dual namespaces

��7


CUDA kernel

namespace vecgeom


Allocation

namespace vecgeom




PH - SFT



Dual namespaces

��7


Launch kernel CUDA kernel

namespace vecgeom


Instantiate

Allocation

Synchronize content

namespace vecgeom




PH - SFT

Compilation scheme

��8

namespace vecgeom .o

Compile for C++11 with vector backend

Source files .cu

Compile with NVCC with CUDA backend


.o

Optional modules (ROOT, benchmarking…)

Source files .cpp

+ CUDA interface .cu

+

Executable

Linking

Interface header



PH - SFT

Status

��9

Nu m b e r of p a rt ic le s loca te d

0 .0

0 .5

1 .0

1 .5

2 .0

2 .5

3 .0

3 .5

4 .0

4 .5

Sp

ee

du

p

Pa rt ic le loca t ion in b ox g e om e t ry

GPU

GPU with ove rh e a d

CPU

Preliminary benchmarks run in hybrid CPU/GPU environment for four-level box geometry.

Location of particles is a tree algorithm, so speedups are only seen at very high particle multiplicity.

Common code for box methods runs for scalar, Vc, Cilk and CUDA backends !Implemented GPU memory synchronization !Dual namespaces allows dispatching work to CPU or GPU !Location can run in either environment !Results are in agreement

!



PH - SFT

Appendix: CPU/GPU same interface

��10

Main executable compiled with C++ compiler

CUDA kernel compiled with NVCC

Same code used



PH - SFT

Appendix: Common code

��11



PH - SFT

Appendix: Template hierarchy

��12



PH - SFT ��13



Vector interface



PH - SFT ��13



Scalar Backend

Vc Backend

Cilk Plus Backend

CUDA Backend

Vector interface


GeantV Geometry: SIMD abstraction and interfacing with CUDA · C-like abstract kernels Backend...

Documents

Transcript of GeantV Geometry: SIMD abstraction and interfacing with CUDA · C-like abstract kernels Backend...