[libre-riscv-dev] Fwd: [OpenPOWER-HDL-Cores] POWER9 ISA opcode usage statistics

[libre-riscv-dev] Fwd: [OpenPOWER-HDL-Cores] POWER9 ISA opcode usage statistics

[Luke Kenneth Casson Leighton]

can someone please capture this in a bugreport?

---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68


---------- Forwarded message ---------
From: Forest Crossman <cyrozap@gmail.com>
Date: Sun, May 24, 2020 at 12:31 AM
Subject: Re: [OpenPOWER-HDL-Cores] POWER9 ISA opcode usage statistics
To: Luke Kenneth Casson Leighton <lkcl@lkcl.net>
Cc: <openpower-hdl-cores@mailinglist.openpowerfoundation.org>


> whilst i appreciate that these statistics will be use-case-specific
> for certain workloads, does anyone know of any studies that have
> already been done, or of any libre-licensed tools that help with the
> same?

I realize it's been over three weeks since you posted this question,
so maybe you've found an answer already, but I had an idea on how to
do it so I decided to try it. I ended up with the following bash
script: https://paste.debian.net/hidden/10201ca1

I ran that script on my Talos II running Debian 8 and got the
following results: https://paste.debian.net/hidden/13696b59

Note that Debian for ppc64el is built without `-mcpu=power9`, so these
statistics won't include any POWER9 instructions. If you want to get
the stats for that, you'll want to build Debian (or another popular
distro) with the `-mcpu=power9` flag to make sure the POWER9
instructions and optimizations are enabled. It also might be a good
idea to gather instruction statistics for the kernel as well, since
kernel code might have a different distribution than the userspace
code I sampled.

_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev