From: Carl Love <cel@us.ibm.com>
To: Pedro Alves <pedro@palves.net>,
"gdb-patches@sourceware.org" <gdb-patches@sourceware.org>,
Will Schmidt <will_schmidt@vnet.ibm.com>,
Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
Subject: RE: [PATCH] Add IEEE FLOAT128 support to test gdb.base/whatis-ptype-typedefs.exp
Date: Mon, 18 Jul 2022 08:40:25 -0700 [thread overview]
Message-ID: <35d6420eb96ff01c3624a3de05659dece2b6a60b.camel@us.ibm.com> (raw)
In-Reply-To: <8ae117dd-17c2-382d-637a-64a3be7b185b@palves.net>
On Mon, 2022-07-18 at 16:22 +0100, Pedro Alves wrote:
> On 2022-07-13 8:38 p.m., Carl Love via Gdb-patches wrote:
>
> > GCC enabled IEEE FLOAT 128-bit support starting with GCC 12 by
> > default.
> > Previously long double was the default for 128-bit floating point
> > support.
> > This patch updates the expected result for the "long doube" tests
> > to
> > _Float128 if GCC 12 is used. The previous "long double" result is
> > expected for GCC 11 and older versions.
>
> This whole paragraph reads as if all archs were changed, but it was
> just Power, right?
Yes, it should say the default on PowerPC changed to IEEE.
Per comments from Ulrich, I need to investigate further to figure out
where the _Float128 is coming from as the string doesn't occur in the
source code. From my initial look, I don't see the string in the gdb
code or the test case source code, the source DWARF or objdump output.
In function check_typedef (gdb/gdbtpyes.c) there is a while loop that
iterates thru the type removing the typedefs. Initially the type->name
= long_double_typedef but then changes to _Float128. I don't know
where the typedef all gets setup but I would think it comes from
reading the source code DWARF. Anyway, I am working on figuring out
where and how the _Float128 gets setup.
>
> And isn't it the case that after the change, for GCC, "long double"
> _IS_ IEEE FLOAT 128-bit?
>
> Doesn't that mean that GCC and GDB now have a mismatch of what they
> think "long double" is?
> Like for example, if the user copies an expression from their program
> and evaluates in gdb,
> and that expression uses "long double", then GDB will evaluate it
> differently from how GCC would?
>
> Shouldn't GDB instead adjust its "long double" type depending on the
> ABI, mapping "long double"
> to the right format? And then, the testcase (probably) wouldn't
> change?
>
> BTW, I was expecting to see GCC's default change mentioned at:
>
>
> https://gcc.gnu.org/gcc-12/changes.html
>
>
> but I can't seem to find it there, other than a Fortran reference
> which seems related. I did find:
>
>
> https://gcc.gnu.org/wiki/Ieee128PowerPC#Transition
>
>
> > The patch has been run on Power 7 (gcc version 4.8.5), Power 10
> > (gcc
> > version 12.1) and Intel x86_64 (gcc version 11.2.0) to verify the
> > patch
> > fixes the 74 test failures on Power 10 with GCC 12 and does not
> > introduce
> > any regression failures on older versions of GCC.
>
> Seems like x86 with gcc 12 would better be tested as well.
>
> > ---
> > .../gdb.base/whatis-ptype-typedefs.exp | 26 ++++++++++++++-
> > ----
> > 1 file changed, 20 insertions(+), 6 deletions(-)
> >
> > diff --git a/gdb/testsuite/gdb.base/whatis-ptype-typedefs.exp
> > b/gdb/testsuite/gdb.base/whatis-ptype-typedefs.exp
> > index be76183ca79..a17084a19ec 100644
> > --- a/gdb/testsuite/gdb.base/whatis-ptype-typedefs.exp
> > +++ b/gdb/testsuite/gdb.base/whatis-ptype-typedefs.exp
> > @@ -77,6 +77,12 @@ proc prepare {lang} {
> > #
> > # This can be "c" or "c++".
> > #
> > +
> > +# GCC 12 uses IEEE 128-bit floating point as the default starting
> > with GCC 12.
>
> On all archs?
>
> > +# The table below consists of the compiler independent tests. The
> > GCC version
> > +# specific tests are appended to the end of the table based on the
> > compiler
> > +# version.
> > +
> > # Columns in the table represent:
> > # EXP # whatis # ptype #
> > language
> > set table {
> > @@ -97,12 +103,6 @@ set table {
> > {"double_typedef2" "double_typedef" "double"}
> > {"v_double_typedef" "double_typedef" "double"}
> > {"v_double_typedef2" "double_typedef2" "double"}
> > -
> > - {"long_double_typedef" "long double" "long
> > double"}
> > - {"long_double_typedef2" "long_double_typedef" "long
> > double"}
> > - {"v_long_double_typedef" "long_double_typedef" "long
> > double"}
> > - {"v_long_double_typedef2" "long_double_typedef2" "long
> > double"}
> > -
> > {"colors_typedef" "(enum )?colors" "enum colors( :
> > unsigned int)? {red, green, blue}"}
> > {"colors_typedef2" "colors_typedef" "enum colors( :
> > unsigned int)? {red, green, blue}"}
> > {"v_colors_typedef" "colors_typedef" "enum colors( :
> > unsigned int)? {red, green, blue}"}
> > @@ -151,6 +151,20 @@ set table {
> > "c++"}
> > }
> >
> > +# Add the long double tests on the version of GCC
> > +if { [test_compiler_info gcc-*] && [gcc_major_version] >= 12 } {
>
> What about x86 + gcc 12 ?
>
> > + lappend table {"long_double_typedef" "long
> > double" "_Float128"}
> > + lappend table
> > {"long_double_typedef2" "long_double_typedef" "_Float128"}
> > + lappend table
> > {"v_long_double_typedef" "long_double_typedef" "_Float128"}
> > + lappend table {"v_long_double_typedef2"
> > "long_double_typedef2" "_Float128"}
> > +
> > +} else {
> > + lappend table {"long_double_typedef" "long
> > double" "long double"}
> > + lappend table
> > {"long_double_typedef2" "long_double_typedef" "long double"}
> > + lappend table
> > {"v_long_double_typedef" "long_double_typedef" "long double"}
> > + lappend table {"v_long_double_typedef2"
> > "long_double_typedef2" "long double"}
> > +}
>
>
next prev parent reply other threads:[~2022-07-18 15:40 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-13 19:38 Carl Love
2022-07-15 11:42 ` Ulrich Weigand
2022-07-18 15:22 ` Pedro Alves
2022-07-18 15:40 ` Carl Love [this message]
2022-07-18 15:55 ` Carl Love
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=35d6420eb96ff01c3624a3de05659dece2b6a60b.camel@us.ibm.com \
--to=cel@us.ibm.com \
--cc=Ulrich.Weigand@de.ibm.com \
--cc=gdb-patches@sourceware.org \
--cc=pedro@palves.net \
--cc=will_schmidt@vnet.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).