lkml.org 
[lkml]   [2020]   [Sep]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 5/5] perf: arm_spe: Decode SVE events
On Tue, Sep 29, 2020 at 10:19:02AM +0800, Leo Yan wrote:
> On Mon, Sep 28, 2020 at 03:47:56PM +0100, Dave Martin wrote:
> > On Mon, Sep 28, 2020 at 02:59:34PM +0100, André Przywara wrote:
> > > On 28/09/2020 14:21, Dave Martin wrote:
> > >
> > > Hi Dave,
> > >
> > > > On Tue, Sep 22, 2020 at 11:12:25AM +0100, Andre Przywara wrote:
> > > >> The Scalable Vector Extension (SVE) is an ARMv8 architecture extension
> > > >> that introduces very long vector operations (up to 2048 bits).
> > > >
> > > > (8192, in fact, though don't expect to see that on real hardware any
> > > > time soon... qemu and the Arm fast model can do it, though.)

[...]

> > Mostly I'm curious because the encoding doesn't match the SVE
> > architecture: SVE requires 4 bits to specify the vector length, not 3.
> > This might have been a deliberate limitation in the SPE spec., but it
> > raises questions about what should happen when 3 bits is not enough.
> >
> > For SVE, valid vector lengths are 16 bytes * n
> > or equivalently 128 bits * n), where 1 <= n <= 16.
> >
> > The code here though cannot print EVLEN16 or EVLEN48 etc. This might
> > not be a bug, but I'd like to understand where it comes from...
>
> In the SPE's spec, the defined values for EVL are:
>
> 0b'000 -> EVLEN: 32 bits.
> 0b'001 -> EVLEN: 64 bits.
> 0b'010 -> EVLEN: 128 bits.
> 0b'011 -> EVLEN: 256 bits.
> 0b'100 -> EVLEN: 512 bits.
> 0b'101 -> EVLEN: 1024 bits.
> 0b'110 -> EVLEN: 2048 bits.
>
> Note that 0b'111 is reserved. In theory, I think SPE Operation packet
> can support up to 4196 bits (32 << 7) when the EVL field is 0b'111; but

OK, having looked at the spec I can now confirm that this look correct.
I was expecting a more direct correspondence between the SVE ISA and
these events, but it looks like SPE may report on a finer granularity
than whole instructions, hence showing effective vector lengths smaller
than 32; also SPE rounds the reported effective vector length up to a
power of two, which allows the full range of lengths to be reported via
the 3-bit EVL field.

> it's impossible to express vector length for 8192 bits as you mentioned.

Yes, ignore my comment about 8192-bit vectors: I was confusing myself
(the Linux API extensions support up to 8192 _bytes_ per vector in order
to have some expansion room just in case; however the SVE architecture
limits vectors to at most 2048 bits).

So I don't see any obvious issues.

It might be a good idea to explicitly reject the encoding 0b111, since
we can't be certain what it is going to mean -- however, I don't have a
strong opinion on this.

Cheers
---Dave

\
 
 \ /
  Last update: 2020-09-29 16:04    [W:0.140 / U:1.156 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site