From: David Edelsohn <dje.gcc@gmail.com>
To: Bill Schmidt <wschmidt@linux.ibm.com>,
Michael Meissner <meissner@linux.ibm.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>,
Segher Boessenkool <segher@kernel.crashing.org>,
Peter Bergner <bergner@linux.ibm.com>,
Will Schmidt <will_schmidt@vnet.ibm.com>
Subject: Re: [PATCH] Fix float128-call.c test for power8 IEEE 128 and power10.
Date: Fri, 27 Aug 2021 12:47:35 -0400 [thread overview]
Message-ID: <CAGWvnymUQVkbykBC3JYdcHkwhD3CJnbAKiwBbWS0UF018fVhZw@mail.gmail.com> (raw)
In-Reply-To: <b60faed2-d550-b7fd-8a10-d2fd9ab06ad1@linux.ibm.com>
This patch is okay.
Thanks, David
On Fri, Aug 27, 2021 at 12:41 PM Bill Schmidt <wschmidt@linux.ibm.com> wrote:
>
> Hi Mike,
>
> Thanks for this clean-up!
>
> On 8/25/21 5:09 PM, Michael Meissner wrote:
> > From 327273dfeec5c000f3c33ca7b88ee0097fd33586 Mon Sep 17 00:00:00 2001
> > From: Michael Meissner <meissner@linux.ibm.com>
> > Date: Wed, 25 Aug 2021 00:31:35 -0400
> > Subject: [PATCH] Fix float128-call.c test for power8 IEEE 128 and power10.
> >
> > I built a compiler on a little endian power8 system where the default long
> > double was IEEE 128-bit instead of IBM 128-bit. I discovered that on
> > power8, we would generate a lxvd2x and xxpermdi to deal with the endianess
> > instead of the Altivec lxv.
> >
> > In addition, I noticed the constant that was being loaded (1.0q) could be
> > loaded by the lxvkq instruction.
> >
> > I rewrote the test to handle all forms of vector load and store that can
> > be generated. And I changed the constant to be one that lxvkq does not
> > support.
> >
> > I did bootstrap tests on the following systems, and the the test ran in all
> > environments (each of the systems were configured for the cpu mentioned):
> >
> > 1) Little endian power9 with IBM 128-bit long double
> > 2) Little endian power9 with IEEE 128-bit long double
> > 3) Little endian power8 with IBM 128-bit long double
> > 4) Little endian power8 with IEEE 128-bit long double
> > 5) Little endian power10 with IBM 128-bit long double
> > 6) Little endian power10 with IEEE 128-bit long double
> > 7) Big endian power8 with IBM 128-bit long double
> >
> > Can I check this patch into the master branch and later backport it to GCC-11?
> >
> > 2021-08-25 Michael Meissner <meissner@linux.ibm.com>
> >
> > gcc/testsuite/
> > * gcc.target/powerpc/float128-call.c: Fix test for IEEE 128-bit
> > long double and power10.
> > ---
> > .../gcc.target/powerpc/float128-call.c | 27 +++++++++++++------
> > 1 file changed, 19 insertions(+), 8 deletions(-)
> >
> > diff --git a/gcc/testsuite/gcc.target/powerpc/float128-call.c b/gcc/testsuite/gcc.target/powerpc/float128-call.c
> > index b64ffc68bfa..d1cf47e4298 100644
> > --- a/gcc/testsuite/gcc.target/powerpc/float128-call.c
> > +++ b/gcc/testsuite/gcc.target/powerpc/float128-call.c
> > @@ -6,22 +6,33 @@
> > #error "-mfloat128 is not supported."
> > #endif
> >
> > +/* Pick a constant to load that cannot be generated by the power10 lxvkq
> > + instruction. */
> > #ifdef __LONG_DOUBLE_IEEE128__
> > #define TYPE long double
> > -#define ONE 1.0L
> > +#define TEN 10.0L
> >
> > #else
> > #define TYPE __float128
> > -#define ONE 1.0Q
> > +#define TEN 10.0Q
> > #endif
> >
> > /* Test to make sure vector registers are used for passing IEEE 128-bit
> > floating point values and returning them. Also make sure the 'q' suffix is
> > - handled. */
> > -TYPE one (void) { return ONE; }
> > + handled for __float128. */
> > +TYPE one (void) { return TEN; }
>
> This amuses me, and I want to keep it this way. :-)
> > void store (TYPE a, TYPE *p) { *p = a; }
> >
> > -/* { dg-final { scan-assembler {\mlxvd2x 34\M} {target be} } } */
> > -/* { dg-final { scan-assembler {\mstxvd2x 34\M} {target be} } } */
> > -/* { dg-final { scan-assembler {\mlvx 2\M} {target le} } } */
> > -/* { dg-final { scan-assembler {\mstvx 2\M} {target le} } } */
> > +/* This regexp captures the different vector load/stores that can be generated:
> > +
> > + lxvd2x -- big endian power7/power8, little endian power8
> > + lvx -- Altivec
> > + lxv -- power9
> > + plxv -- power10
> > + lxvx -- X-form variant.
> > + stxvd2x -- big endian power7/power8, little endian power8
> > + stvx -- Altivec
>
> For symmetry, also mention stxvx as an X-form variant?
>
> Looks fine to me, recommend approval.
>
> Thanks,
> Bill
>
> > + lxvx -- power9/power10. */
> > +
> > +/* { dg-final { scan-assembler {\mlxvd2x 34\M|\mlvx 2\M|\mp?lxvx? 34\M} } } */
> > +/* { dg-final { scan-assembler {\mstxvd2x 34\M|\mstvx 2\M|\mstxvx 34\M} } } */
next prev parent reply other threads:[~2021-08-27 16:47 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-25 22:09 Michael Meissner
2021-08-27 16:41 ` Bill Schmidt
2021-08-27 16:47 ` David Edelsohn [this message]
2021-08-27 17:11 ` Michael Meissner
2021-08-27 17:29 ` Segher Boessenkool
2021-08-27 20:11 ` Michael Meissner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAGWvnymUQVkbykBC3JYdcHkwhD3CJnbAKiwBbWS0UF018fVhZw@mail.gmail.com \
--to=dje.gcc@gmail.com \
--cc=bergner@linux.ibm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=meissner@linux.ibm.com \
--cc=segher@kernel.crashing.org \
--cc=will_schmidt@vnet.ibm.com \
--cc=wschmidt@linux.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).