From: "Kewen.Lin" <linkw@linux.ibm.com>
To: HAO CHEN GUI <guihaoc@linux.ibm.com>
Cc: Segher Boessenkool <segher@kernel.crashing.org>,
David <dje.gcc@gmail.com>, Peter Bergner <bergner@linux.ibm.com>,
gcc-patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH, rs6000] xfail float128 comparison test case that fails on powerpc64 [PR108728]
Date: Fri, 14 Apr 2023 13:11:38 +0800 [thread overview]
Message-ID: <b338d5e1-79a7-4a14-5846-9406b9d72d22@linux.ibm.com> (raw)
In-Reply-To: <84c7ea10-5f15-03b8-3354-487665e7e409@linux.ibm.com>
Hi Haochen,
on 2023/4/14 09:34, HAO CHEN GUI wrote:
> Hi Kewen,
>
> 在 2023/4/13 16:32, Kewen.Lin 写道:
>> xfail all powerpc*-*-* can have some XPASSes on those ENVs with
>> software emulation. Since the related hw insn xscmpuqp is guarded
>> with TARGET_FLOAT128_HW, could we use the effective target
>> ppc_float128_hw instead?
>
> Thanks for your review comments. It's tricky. It invokes "__lekf2"
> with "-mno-float128_hw". But it doesn't always pass the check.
> With math library on P8, it can. With the library on P9, it fails.
Math library doesn't provide it, __lekf2 is from libgcc (GCC itself).
The reason why the __lekf2 behaves differently on P8 and P9 is that
we have SW and HW versions for __lekf2, when the underlying "CPU
"supports 128-bit IEEE binary floating point instructions.", it will
use __lekf2_hw instead of __lekf2_sw, the former still adopts
insn xscmpuqp, then it fails.
> So it's totally depended on the version of library which is not
> controlled by GCC. What's your opinion?
So
1) for ppc_float128_hw, it generates xscmpuqp then fails.
2) for !ppc_float128_hw, it uses __lekf2 but the underlying ENV
supports __builtin_cpu_supports ("ieee128"), it exploits
xscmpuqp, then fails.
Ideally we should use one effective target like ppc_ieee128_hw to
indicate the underlying ENV supports __builtin_cpu_supports ("ieee128")
but I think it may not be worth to adding that at this stage, so
I'd suggest xfail-ing it for "ppc_float128_hw || (ppc_cpu_supports_hw && p9vector_hw)".
BR,
Kewen
>
> Test result on P9
> make check-gcc-c RUNTESTFLAGS="--target_board=unix'{-mno-float128-hardware}' dg-torture.exp=float128-cmp-invalid.c"
>
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O0 execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O1 execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O2 execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O3 -g execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -Os execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O2 -flto -fno-use-linker-plugin -flto-partition=none execution test
> FAIL: gcc.dg/torture/float128-cmp-invalid.c -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects execution test
>
> === gcc Summary ===
>
> # of expected passes 7
> # of unexpected failures 7
>
> Gui Haochen
> Thanks
prev parent reply other threads:[~2023-04-14 5:11 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-04-12 2:27 HAO CHEN GUI
2023-04-13 8:32 ` Kewen.Lin
2023-04-14 1:34 ` HAO CHEN GUI
2023-04-14 5:11 ` Kewen.Lin [this message]
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=b338d5e1-79a7-4a14-5846-9406b9d72d22@linux.ibm.com \
--to=linkw@linux.ibm.com \
--cc=bergner@linux.ibm.com \
--cc=dje.gcc@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=guihaoc@linux.ibm.com \
--cc=segher@kernel.crashing.org \
/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).