From: "Maciej W. Rozycki" <macro@embecosm.com>
To: Kito Cheng <kito.cheng@sifive.com>
Cc: gcc-patches@gcc.gnu.org, kito.cheng@gmail.com,
Jim Wilson <jim.wilson.gcc@gmail.com>,
Palmer Dabbelt <palmer@dabbelt.com>,
andrew@sifive.com, juzhe.zhong@rivai.ai,
joseph@codesourcery.com
Subject: Re: [PATCH v2 1/2] RISC-V: Support _Float16 type.
Date: Mon, 5 Dec 2022 13:05:22 +0000 (GMT) [thread overview]
Message-ID: <alpine.DEB.2.20.2212051305120.7841@tpp.orcam.me.uk> (raw)
In-Reply-To: <20220810154456.101086-1-kito.cheng@sifive.com>
Hi Kito,
I came across this issue while inspecting code and I have been wondering
what the reason was to downgrade current FMV.X.W and FMW.W.X instructions
to their older FMV.S.W and FMV.W.S variants here:
On Wed, 10 Aug 2022, Kito Cheng wrote:
> diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
> index 5a0adffb5ce..47e6110767c 100644
> --- a/gcc/config/riscv/riscv.cc
> +++ b/gcc/config/riscv/riscv.cc
> @@ -2308,10 +2310,19 @@ riscv_output_move (rtx dest, rtx src)
> if (dest_code == REG && GP_REG_P (REGNO (dest)))
> {
> if (src_code == REG && FP_REG_P (REGNO (src)))
> - return dbl_p ? "fmv.x.d\t%0,%1" : "fmv.x.w\t%0,%1";
> + switch (width)
> + {
> + case 2:
> + /* Using fmv.x.s + sign-extend to emulate fmv.x.h. */
> + return "fmv.x.s\t%0,%1;slli\t%0,%0,16;srai\t%0,%0,16";
> + case 4:
> + return "fmv.x.s\t%0,%1";
> + case 8:
> + return "fmv.x.d\t%0,%1";
> + }
and here:
> @@ -2353,18 +2364,24 @@ riscv_output_move (rtx dest, rtx src)
> return "mv\t%0,%z1";
>
> if (FP_REG_P (REGNO (dest)))
> - {
> - if (!dbl_p)
> - return "fmv.w.x\t%0,%z1";
> - if (TARGET_64BIT)
> - return "fmv.d.x\t%0,%z1";
> - /* in RV32, we can emulate fmv.d.x %0, x0 using fcvt.d.w */
> - gcc_assert (src == CONST0_RTX (mode));
> - return "fcvt.d.w\t%0,x0";
> - }
> + switch (width)
> + {
> + case 2:
> + /* High 16 bits should be all-1, otherwise HW will treated
> + as a n-bit canonical NaN, but isn't matter for softfloat. */
> + return "fmv.s.x\t%0,%1";
> + case 4:
> + return "fmv.s.x\t%0,%z1";
> + case 8:
> + if (TARGET_64BIT)
> + return "fmv.d.x\t%0,%z1";
> + /* in RV32, we can emulate fmv.d.x %0, x0 using fcvt.d.w */
(Incorrect comment formatting here as well.)
> + gcc_assert (src == CONST0_RTX (mode));
> + return "fcvt.d.w\t%0,x0";
> + }
Was it intentional or just an oversight in review? If intentional, I'd
expect such a change to happen on its own rather than sneaked in with a
large functional update.
Maciej
next prev parent reply other threads:[~2022-12-05 13:05 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-10 15:44 Kito Cheng
2022-08-10 15:44 ` [PATCH v2 2/2] RISC-V: Support zfh and zfhmin extension Kito Cheng
2022-08-10 23:19 ` 钟居哲
2022-12-05 13:05 ` Maciej W. Rozycki [this message]
2022-12-07 14:15 ` [PATCH v2 1/2] RISC-V: Support _Float16 type Kito Cheng
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=alpine.DEB.2.20.2212051305120.7841@tpp.orcam.me.uk \
--to=macro@embecosm.com \
--cc=andrew@sifive.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jim.wilson.gcc@gmail.com \
--cc=joseph@codesourcery.com \
--cc=juzhe.zhong@rivai.ai \
--cc=kito.cheng@gmail.com \
--cc=kito.cheng@sifive.com \
--cc=palmer@dabbelt.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).