General

Profile

Patrick Gartung

Issues

Projects

Activity

11/20/2020

08:30 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
The c7, e19 and e20 qualifiers are built with cxxflags += -std=cxx17 and cflags += -std=c11

11/18/2020

04:44 PM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Hi Andy,
There is now an official build of libtorch v1_6_0 installed on larsoft cvmfs. I tried the build on scisof...

11/16/2020

11:55 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Running on cluck, which has neither AVX nor AVX2 instructions I get the unsupported hardware warning.
NNPACK will ...

11/10/2020

03:16 PM LArSoft Support #24669: LibTorch v1.5.1 efficiency
The unsupported hardware message happens when running in a VirtualBox VM even if the host support avx and avx2.
Wh...
12:11 PM LArSoft Support #24669: LibTorch v1.5.1 efficiency
On my linux desktop with has a Haswell cpu I don't get the unsupported hardware error and the performance is shown ...
11:39 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Running in a VirtualBox VM on my laptop which has a Skylake cpu and supports AVX2, I get the unsupported hardware err...

11/09/2020

01:56 PM LArSoft Support #24669: LibTorch v1.5.1 efficiency
It appears that the NNPACK static library always compiles the same functions whether the host has AVX2 or not. Runnin...
09:57 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Running on a machine that supports AVX2 seems to be the key for performance. The only further optimization is the ad...

11/06/2020

09:32 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Andrew Chappell wrote:
> Hi Patrick,
>
> Here's what I see running against your libtorch 1.6 build:
>
> real 0...
09:05 AM LArSoft Support #24669: LibTorch v1.5.1 efficiency
Hi Andy,
Could you try again with...

Also available in: Atom