From: "Kewen.Lin" <linkw@linux.ibm.com>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Cc: Segher Boessenkool <segher@kernel.crashing.org>,
David Edelsohn <dje.gcc@gmail.com>,
Peter Bergner <bergner@linux.ibm.com>
Subject: PING^1 [PATCH] rs6000: Fix vector_set_var_p9 by considering BE [PR108807]
Date: Mon, 20 Mar 2023 14:35:43 +0800 [thread overview]
Message-ID: <52394650-aa6b-c7a1-8a7f-691870309829@linux.ibm.com> (raw)
In-Reply-To: <737a5392-29f8-763c-8dc7-b48c36edb1a7@linux.ibm.com>
Hi,
I'd like to gentle ping this:
https://gcc.gnu.org/pipermail/gcc-patches/2023-February/612213.html
It's to fix one regression, I think it's stage 4 content.
BR,
Kewen
on 2023/2/17 17:55, Kewen.Lin via Gcc-patches wrote:
> Hi,
>
> As PR108807 exposes, the current handling in function
> rs6000_expand_vector_set_var_p9 doesn't take care of big
> endianness. Currently the function is to rotate the
> target vector by moving element to-be-set to element 0,
> set element 0 with the given val, then rotate back. To
> get the permutation control vector for the rotation, it
> makes use of lvsr and lvsl, but the element ordering is
> different for BE and LE (like element 0 is the most
> significant one on BE while the least significant one on
> LE), this patch is to add consideration for BE and make
> sure permutation control vectors for rotations are expected.
>
> As tested, it helped to fix the below failures:
>
> FAIL: gcc.target/powerpc/pr79251-run.p9.c execution test
> FAIL: gcc.target/powerpc/pr89765-mc.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-10d.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-11d.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-14d.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-16d.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-18d.c execution test
> FAIL: gcc.target/powerpc/vsx-builtin-9d.c execution test
>
> Bootstrapped and regtested on powerpc64-linux-gnu P{8,9}
> and powerpc64le-linux-gnu P10.
>
> Is it ok for trunk?
>
> BR,
> Kewen
> -----
> PR target/108807
>
> gcc/ChangeLog:
>
> * config/rs6000/rs6000.cc (rs6000_expand_vector_set_var_p9): Fix gen
> function for permutation control vector by considering big endianness.
> ---
> gcc/config/rs6000/rs6000.cc | 48 +++++++++++++++++++++----------------
> 1 file changed, 28 insertions(+), 20 deletions(-)
>
> diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
> index 16ca3a31757..774eb2963d9 100644
> --- a/gcc/config/rs6000/rs6000.cc
> +++ b/gcc/config/rs6000/rs6000.cc
> @@ -7235,22 +7235,26 @@ rs6000_expand_vector_set_var_p9 (rtx target, rtx val, rtx idx)
>
> machine_mode shift_mode;
> rtx (*gen_ashl)(rtx, rtx, rtx);
> - rtx (*gen_lvsl)(rtx, rtx);
> - rtx (*gen_lvsr)(rtx, rtx);
> + rtx (*gen_pcvr1)(rtx, rtx);
> + rtx (*gen_pcvr2)(rtx, rtx);
>
> if (TARGET_POWERPC64)
> {
> shift_mode = DImode;
> gen_ashl = gen_ashldi3;
> - gen_lvsl = gen_altivec_lvsl_reg_di;
> - gen_lvsr = gen_altivec_lvsr_reg_di;
> + gen_pcvr1 = BYTES_BIG_ENDIAN ? gen_altivec_lvsl_reg_di
> + : gen_altivec_lvsr_reg_di;
> + gen_pcvr2 = BYTES_BIG_ENDIAN ? gen_altivec_lvsr_reg_di
> + : gen_altivec_lvsl_reg_di;
> }
> else
> {
> shift_mode = SImode;
> gen_ashl = gen_ashlsi3;
> - gen_lvsl = gen_altivec_lvsl_reg_si;
> - gen_lvsr = gen_altivec_lvsr_reg_si;
> + gen_pcvr1 = BYTES_BIG_ENDIAN ? gen_altivec_lvsl_reg_si
> + : gen_altivec_lvsr_reg_si;
> + gen_pcvr2 = BYTES_BIG_ENDIAN ? gen_altivec_lvsr_reg_si
> + : gen_altivec_lvsl_reg_si;
> }
> /* Generate the IDX for permute shift, width is the vector element size.
> idx = idx * width. */
> @@ -7259,25 +7263,29 @@ rs6000_expand_vector_set_var_p9 (rtx target, rtx val, rtx idx)
>
> emit_insn (gen_ashl (tmp, idx, GEN_INT (shift)));
>
> - /* lvsr v1,0,idx. */
> - rtx pcvr = gen_reg_rtx (V16QImode);
> - emit_insn (gen_lvsr (pcvr, tmp));
> -
> - /* lvsl v2,0,idx. */
> - rtx pcvl = gen_reg_rtx (V16QImode);
> - emit_insn (gen_lvsl (pcvl, tmp));
> + /* Generate one permutation control vector used for rotating the element
> + at to-insert position to element zero in target vector. lvsl is
> + used for big endianness while lvsr is used for little endianness:
> + lvs[lr] v1,0,idx. */
> + rtx pcvr1 = gen_reg_rtx (V16QImode);
> + emit_insn (gen_pcvr1 (pcvr1, tmp));
>
> rtx sub_target = simplify_gen_subreg (V16QImode, target, mode, 0);
> + rtx perm1 = gen_altivec_vperm_v8hiv16qi (sub_target, sub_target, sub_target,
> + pcvr1);
> + emit_insn (perm1);
>
> - rtx permr
> - = gen_altivec_vperm_v8hiv16qi (sub_target, sub_target, sub_target, pcvr);
> - emit_insn (permr);
> -
> + /* Insert val into element 0 of target vector. */
> rs6000_expand_vector_set (target, val, const0_rtx);
>
> - rtx perml
> - = gen_altivec_vperm_v8hiv16qi (sub_target, sub_target, sub_target, pcvl);
> - emit_insn (perml);
> + /* Rotate back with a reversed permutation control vector generated from:
> + lvs[rl] v2,0,idx. */
> + rtx pcvr2 = gen_reg_rtx (V16QImode);
> + emit_insn (gen_pcvr2 (pcvr2, tmp));
> +
> + rtx perm2 = gen_altivec_vperm_v8hiv16qi (sub_target, sub_target, sub_target,
> + pcvr2);
> + emit_insn (perm2);
> }
>
> /* Insert VAL into IDX of TARGET, VAL size is same of the vector element, IDX
> --
> 2.39.1
next prev parent reply other threads:[~2023-03-20 6:35 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-02-17 9:55 Kewen.Lin
2023-03-20 6:35 ` Kewen.Lin [this message]
2023-04-03 11:44 ` Segher Boessenkool
2023-04-04 5:19 ` Kewen.Lin
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=52394650-aa6b-c7a1-8a7f-691870309829@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=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).