By Koichi Hirai, Fujitsu Post-K use Arm based super computer. But there are not too many Arm based servers for HPC. Therefore we think to need to build Arm HPC Ecosystem until Post-K release. In this presentation, we describe our collaboration efforts to build the Arm HPC Ecosystem. For more info on The Linaro High Performance Computing (HPC) visit https://www.linaro.org/sig/hpc/

1. Kouichi Hirai
FUJITSU LIMITED
Dec 12th, 2017
Post-K:
Building the Arm HPC Ecosystem
0 Copyright 2017 FUJITSU LIMITEDLinaro Work Shop, Dec. 12, 2017

2. Post-K: Building up Arm HPC Ecosystem
 Fujitsu’s approach for HPC
 For making the Post-K a resounding success
 The high performance compiler increases software portability
 Summary
Copyright 2017 FUJITSU LIMITEDLinaro Work Shop, Dec. 12, 2017 1

3. Fujitsu HPC Solutions to Meet Customer Demands
 Supercomputers, both Fujitsu-developed CPUs and x86
 Single system image operation w/ Fujitsu system software
 High performance, high availability, and high reliability
Copyright 2017 FUJITSU LIMITED
x86 Cluster
RX2530/RX2540 CX600CX400
High scalability
with Fujitsu-
developed CPU
and interconnect
PRIMERGY x86
cluster systems
support the
latest CPUs and
accelerators
Under Development
w/ RIKEN
High
-end
Divisional
Departmental
Workgroup
PRIMEHPC FX10 PRIMEHPC FX100 Post-KK computer
Co-developed with RIKEN
© RIKEN
Large-Scale
SMP System
RX900
Linaro Work Shop, Dec. 12, 2017 2

4. Fujitsu High-end Supercomputers Development
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
PRIMEHPC FX10
 1.8x CPU perf. of K
 Easier installation
 4x(DP) / 8x(SP) CPU per. of K, Tofu2
 High-density pkg & lower energy
App.
review
FS
projects
HPCI strategic apps program
Operation of K computerDevelopment
Japan’s National Projects
FUJITSU
Post-K computer development
PRIMEHPC FX100
K computer and PRIMEHPC
FX10/FX100 in operation
The CPU and interconnect of
FX10/FX100 inherit the K computer
architectural concept, featuring state-of-
the-art technologies
System software “TCS” supports Fujitsu
supercomputer with originally introduced
technologies
Many applications are currently running
and being developed for science and
various industries
RIKEN and Fujitsu are working together
to provide a successor to K computer
with application R&D teams using co-
design approach
Technical Computing Suite (TCS)
Handles millions of parallel jobs
FEFS: super scalable file system
MPI: Ultra scalable collective
communication libraries
 OS: Lower OS jitter w/
assistant core
Copyright 2017 FUJITSU LIMITED
Post-K supercomputer
Post-K
Linaro Work Shop, Dec. 12, 2017 3

5. Post-K Features and Status
 Fujitsu CPU core (w/ Arm SVE) and Tofu maintain the programming models
and provide high application performance
 RIKEN & Fujitsu system software enable high performance and low power
consumption with flexible operations
 Apps from 9 “priority issues” & many “exploratory challenges” are being
optimized for the Post-K
Functions & architecture
Post-K FX100 FX10 K
CPU Core
Instruction set architecture Armv8-A SPARC V9
SIMD width 512bit 256bit 128bit 128bit
Double precision (64bit) ✔ ✔ ✔ ✔
Single precision (32bit) ✔ ✔ ✔ ✔
Half precision (16bit) ✔ - - -
Interconnect Tofu interconnect Enhanced Tofu2 Tofu Tofu
Copyright 2017 FUJITSU LIMITED
Post-K
Linaro Work Shop, Dec. 12, 2017 4

6. Post-K Software Stack
 Valuable feedbacks through “co-design” from application R&D teams
Post-K System Hardware
FUJITSU Technical Computing Suite / RIKEN Advanced System Software
Linux OS / McKernel (Lightweight Kernel)
Post-K Applications
System management
for highly available & power
saving operation
Job management for higher
system utilization & power
efficiency
Lustre-based
distributed file system
FEFS
OpenMP, COARRAY, Math Libs
Compilers (C, C++, Fortran)
Debugging and tuning tools
Management Software Programming EnvironmentHierarchical File I/O Software
MPI (Open MPI, MPICH)
XcalableMP
Application-oriented
file I/O middleware
Post-K
Under Development
w/ RIKEN
Copyright 2017 FUJITSU LIMITEDLinaro Work Shop, Dec. 12, 2017 5

7. Post-K to be More Useful?
 More apps from OSS & ISVs
High performance on “real” applications
Lower TCO
• Low power consumption
• Water cooling
De-facto standards
• Lowering barriers in developing and porting
Ecosystem
• More Arm platforms
• More partners
• More knowledge/experience inside/outside of communities
Copyright 2017 FUJITSU LIMITEDLinaro Work Shop, Dec. 12, 2017 6

8. Making the Post-K a Resounding Success
 Recapping the goal & requirements
 High performance HW and SW complying open standards
 Apps in quality & variety
 Environments – rich, modern, and comprehensive
 Our approach
 Arm architecture (w/ Fujitsu’s proven microarchitecture)
• SBSA: Server Base System Architecture
• SBBR: Server Base Boot Requirements
• VLA: Vector-Length Agnostic
 Fujitsu enhanced/maintained system software
• Based on Linux & OSSs
• Single source for x86 & Arm
• Open MPI, OpenMP, Libraries,
• Performance analyzer, Debugger
 Powerful but original compilers --- will be aligned to be useful & popular
Copyright 2017 FUJITSU LIMITED
Assure binary compatibility
Lowering barriers for single
source development
Linaro Work Shop, Dec. 12, 2017 7

9.  Transform our original & powerful compilers to be all-around
 Working and contributing for the Clang project to satisfy both high
performance and portability
 Fujitsu’s back-end advantage
 Auto-parallelization for many-core architecture
 Auto-vectorization for Scalable Vector Extension
 Strong software pipelining with loop fission
Compilers to Increase Software Portability
Copyright 2017 FUJITSU LIMITED
Utilize Post-K μArch:
• Rich & wide SIMD
• Sector cache…
Software:
Apps, Middleware,
and Basics (written
in variety of styles)
Portable
binariesFujitsu original
front-end
Fujitsu original
back-end from
knowledge of
CPU
development
Clang front-end Clang back-end
Linaro Work Shop, Dec. 12, 2017 8

10. Auto-vectorization for Arm SVE
 4 Byte x 16 SIMD List Memory Access by utilizing 512bit Register
 Various Types of SIMD Optimization by Utilizing Predicate Registers
Copyright 2017 FUJITSU LIMITED
for (int i=0; i<n; ++i) {
if (mask[i] !=0) { a[i] = b[i]; }
}
for (int i=0; i<VL/2; ++i) {
a[i] = b[i] * c[i];
}
do {
b[i] = a[i];
} while(a[i++] != 0);
Loop including IF clause
Small Loop less
than SIMD length
While Loop with
Data Dependency
SVE
Reg. dest.
Reg. index
int index[n]
float P[n], Q[n];
for (i=0; i<n; ++i) {
P[i] = Q[index[i]];
}
Q[14] Q[1] ・ Q[13] ・ Q[0] Q[3] Q[15] Q[2]
14 1 ・ 13 ・ 0 3 15 2
Memory Q [15] [14] [13] ・ ・ [3] [2] [1] [0]1
2
3
4
5
6
7
1
2
3
1
2
3
1
2
3
Linaro Work Shop, Dec. 12, 2017 9

11. Fujitsu Compiler Back-end Optimization Flow
 Loop Fission reduces required resources, such as registers
 Software Pipelining and Register Allocation
 Best utilization of hardware functions and resources
Copyright 2017 FUJITSU LIMITED
Back-end optimization pipeline
Portable
Arm
binaries
SIMDize
Loop
Fission
Software
Pipelining
Register
Allocation
Instruction
Scheduling
for (...) {
}
// Reduced # of Regs.
for (...) {
}
// Reduced # of Regs.
for (...) {
}
// Higher ILP
for (...) {
}
// Higher ILP
for (...) {
}
Software pipelined #1
Software pipelined #2
Divided # 1
Divided# 2
Original
1
2
3
4
5
6
7
8
9
1
2
3
4
5
6
7
8
9
1
2
3
4
5
6
7
8
9
Linaro Work Shop, Dec. 12, 2017 10

12. Copyright 2017 FUJITSU LIMITED
Effectiveness of SWP w/ Loop Fission and SoA
 Runs on FX100 w/ 32 registers
 72% speed-up per core is observed
 >2x speed-up compared w/ K computer
 Software Pipelining w/ Loop Fission
utilizes CPU resources
 SoA-style layout extracts more
NICAM* single core performance on FX100 w/ 32 regs
(Source: http://www.riken.jp/pr/topics/2013/20130920_1/)
CPUclocksnormalized
byKcomputer
*NICAM-DC-MINI: Climate simulations with fine mesh, https://github.com/fiber-miniapp/nicam-dc-mini
SWP w/
Loop
fission
+ SoA
style
72% speedup w/ loop fission + SoA
Without
Loop
fission
Linaro Work Shop, Dec. 12, 2017 11

13. Summary
 Fujitsu’s Approach to HPC
 Supporting high-end supercomputers with original CPU & x86 clusters
 Developing the Post-K for app performance and low power consumption
 Expecting more apps from OSS & ISVs through growing ecosystem
 Keys for Post-K Success
 High performance standard-compliant HW and SW
 All-around high performance compiler with binary compatibility
 Many and varied high quality apps with x86 software compatibility
 Open & Highly Optimized Compilers
 Clang + Fujitsu technologies
 Tentative evaluation results are encouraging
Copyright 2017 FUJITSU LIMITEDLinaro Work Shop, Dec. 12, 2017 12

14. Copyright 2017 FUJITSU LIMITED