lkml.org 
[lkml]   [2020]   [Oct]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 14/14] perf arm-spe: Add support for ARMv8.3-SPE
On Tue, Oct 20, 2020 at 10:54:44PM +0100, André Przywara wrote:
> On 29/09/2020 14:39, Leo Yan wrote:
>
> Hi,
>
> > From: Wei Li <liwei391@huawei.com>
> >
> > This patch is to support Armv8.3 extension for SPE, it adds alignment
> > field in the Events packet and it supports the Scalable Vector Extension
> > (SVE) for Operation packet and Events packet with two additions:
> >
> > - The vector length for SVE operations in the Operation Type packet;
> > - The incomplete predicate and empty predicate fields in the Events
> > packet.
> >
> > Signed-off-by: Wei Li <liwei391@huawei.com>
> > Signed-off-by: Leo Yan <leo.yan@linaro.org>
> > ---
> > .../arm-spe-decoder/arm-spe-pkt-decoder.c | 84 ++++++++++++++++++-
> > .../arm-spe-decoder/arm-spe-pkt-decoder.h | 6 ++
> > 2 files changed, 87 insertions(+), 3 deletions(-)
> >
> > diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c b/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
> > index 05a4c74399d7..3ec381fddfcb 100644
> > --- a/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
> > +++ b/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
> > @@ -342,14 +342,73 @@ int arm_spe_pkt_desc(const struct arm_spe_pkt *packet, char *buf,
> > return ret;
> > }
> > }
> > + if (idx > 2) {
>
> As I mentioned in the other patch, I doubt this extra comparison is
> useful. Does that protect us from anything?

It's the same reason with Event packet which have explained for replying
patch 10, the condition is to respect the SPE specifiction:

E[11], byte 1, bit [11], when SZ == 0b10 , or SZ == 0b11
Alignment.
...
Otherwise this bit reads-as-zero.

So we gives higher priority for checking payload size than the Event
bit setting; if you have other thinking for this, please let me know.

> > + if (payload & SPE_EVT_PKT_ALIGNMENT) {
>
> Mmh, but this is bit 11, right?

Yes.

> So would need to go into the (idx > 1)
> section (covering bits 8-15)? Another reason to ditch this comparison above.

As has explained in patch 10, idx is not the same thing with "sz"
field; "idx" stands for payload length in bytes, so:

idx = 1 << sz

The spec defines the sz is 2 or 3, thus idx is 4 or 8; so this is why
here use the condition "(idx > 2)".

I think here need to refine code for more explict expression so can
avoid confusion. So I think it's better to condition such like:

if (payload_len >= 4) {

...

}

> > + ret = snprintf(buf, buf_len, " ALIGNMENT");
> > + if (ret < 0)
> > + return ret;
> > + buf += ret;
> > + blen -= ret;
>
> Shouldn't we use the new arm_spe_pkt_snprintf() function here as well?
> Or is there a reason that this doesn't work?

Goot point. Will change to use arm_spe_pkt_snprintf().

> > + }
> > + if (payload & SPE_EVT_PKT_SVE_PARTIAL_PREDICATE) {
> > + ret = snprintf(buf, buf_len, " SVE-PARTIAL-PRED");
> > + if (ret < 0)
> > + return ret;
> > + buf += ret;
> > + blen -= ret;
> > + }
> > + if (payload & SPE_EVT_PKT_SVE_EMPTY_PREDICATE) {
> > + ret = snprintf(buf, buf_len, " SVE-EMPTY-PRED");
> > + if (ret < 0)
> > + return ret;
> > + buf += ret;
> > + blen -= ret;
> > + }
> > + }
> > +
> > return buf_len - blen;
> >
> > case ARM_SPE_OP_TYPE:
> > switch (idx) {
> > case SPE_OP_PKT_HDR_CLASS_OTHER:
> > - return arm_spe_pkt_snprintf(&buf, &blen,
> > - payload & SPE_OP_PKT_OTHER_SUBCLASS_COND ?
> > - "COND-SELECT" : "INSN-OTHER");
> > + if ((payload & SPE_OP_PKT_OTHER_SVE_SUBCLASS_MASK) ==
> > + SPE_OP_PKT_OTHER_SUBCLASS_SVG_OP) {
> > +
> > + ret = arm_spe_pkt_snprintf(&buf, &blen, "SVE-OTHER");
> > + if (ret < 0)
> > + return ret;
> > +
> > + /* Effective vector length: step is 32 bits */
> > + ret = arm_spe_pkt_snprintf(&buf, &blen, " EVLEN %d",
> > + 32 << ((payload & SPE_OP_PKT_SVE_EVL_MASK) >>
> > + SPE_OP_PKT_SVE_EVL_SHIFT));
>
> Can you move this into a macro, and add a comment about how this works?
> People might get confused over the "32 << something".

Yeah, will refine for it.

Thanks,
Leo

\
 
 \ /
  Last update: 2020-10-21 07:11    [W:0.129 / U:2.812 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site