| Table of Contents | ||
|---|---|---|
|
Attendance
Committee Members
| Name | Present |
|---|---|
| Kanta Vekaria (OCTO, Linaro) |  |
| Martin Stadtler (Director of LEG, Linaro) | |
Darren Cepulis (ARM) |
|
| Eric Van Hensbergen (ARM) |  |
Tianhong Ding (HiSilicon) |
|
Larry Wikelius (Cavium) |
|
Gary Yurcak (Qualcomm) |
|
Scott Hara (Qualcomm) |
|
Hongbo (HXT) |
|
Jon Masters (RedHat) |
|
Takeharu Kato (Fujitsu) |
|
| Koichi Hirai (Fujitsu) | |
Guests
| Name | Present |
|---|---|
David Rusling (CTO, Linaro) |
|
| Andrea Gallo (VP of Segment Groups, Linaro) |
|
| Anoop Saxena (Project Manager, Segments) |
|
Renato Golin (Linaro) |
|
| Matt Spencer (ARM) Elsie Wahlig (Qualcomm) Grant Likely (ARM) |
|
| |
| |
Agenda
- Round the table SWOT discussion
Minutes
Kanta : Kickoff with introduction
- Scott Hara
- Need a class leading server available for development
- Need to prepare open src libraries. To tune libraries and components
- Note the headcount in Linaro is low
- Perhaps look into HPC Toolkit?
Larry
Issue is to have OpenHPC equivalent to x86. And challenge is micro optimization
Want to see open src GCC + LLVM toolchains. I'm confused with the ARM model and when their LLVM changes will go upstream. We would like to see LLVM caught up with GCC
Inifinband - Don't think this is a good investment for Linaro. It's niche expertise.expertise – but still required for the market
No GPU compute/offload solution is supported on ARM today – major gap
Jon
Concern customer adapting HPC solution. Inifiniband solutions no one is really looking into it
LInaro doing good job by highlighting a problem
Martin: We are 4# engineers working on HPC
Jon: Go for core implementation and should not go and investing much on fixing bug
Tianhong
- Tianhong: port x86 libraries to ARM. Action to send a list
- Infiniband is important. And companies like Huawei can support to linaro on this
Kato San
5 things are needed
optimising Compiler for SVE
HW implementation not optimized for HPC
Need to prepare applications
We need gcc and LLVM compatibility needs to be included such as intrinsics and encourage application porting
See good openHPC ecosystem
Kanta
Porting of libraries
Larry
gcc 7, most of the things are breaking. And good if someone is looking into it. note that this is for all architectures, not just ARM – but we are seeing it first as we are pushing customers to move more quickly to GCC7
Kantalist of Status of libraries and
- RenatoARM is hard on upstreaming the SVE implementation. May miss even in gcc 67
- Compiler work: All agree that that is a priority Action: Linaro to investigate and scope work in this area. Possible Lead project which many can collaborate around