From: Kito Cheng <kito.cheng@gmail.com>
To: juzhe.zhong@rivai.ai
Cc: gcc-patches@gcc.gnu.org, palmer@dabbelt.com
Subject: Re: [PATCH] RISC-V: Fix backward_propagate_worthwhile_p
Date: Fri, 27 Jan 2023 03:13:22 +0800 [thread overview]
Message-ID: <CA+yXCZB-qRC7B-OCJyqFUhY-CHo=uPeDyAFmvDj3SCLF63acOQ@mail.gmail.com> (raw)
In-Reply-To: <20230103071641.149958-1-juzhe.zhong@rivai.ai>
[-- Attachment #1: Type: text/plain, Size: 4803 bytes --]
committed, thanks.
On Tue, Jan 3, 2023 at 3:17 PM <juzhe.zhong@rivai.ai> wrote:
> From: Ju-Zhe Zhong <juzhe.zhong@rivai.ai>
>
> gcc/ChangeLog:
>
> * config/riscv/riscv-vsetvl.cc (loop_basic_block_p): Adjust
> function.
> (backward_propagate_worthwhile_p): Fix non-worthwhile.
>
> ---
> gcc/config/riscv/riscv-vsetvl.cc | 91 +++++++++++++++++++++++++-------
> 1 file changed, 71 insertions(+), 20 deletions(-)
>
> diff --git a/gcc/config/riscv/riscv-vsetvl.cc
> b/gcc/config/riscv/riscv-vsetvl.cc
> index ad0457ed89d..fe76bea297e 100644
> --- a/gcc/config/riscv/riscv-vsetvl.cc
> +++ b/gcc/config/riscv/riscv-vsetvl.cc
> @@ -116,10 +116,27 @@ vlmax_avl_insn_p (rtx_insn *rinsn)
> || INSN_CODE (rinsn) == CODE_FOR_vlmax_avldi);
> }
>
> +/* Return true if the block is a loop itself:
> + local_dem
> + __________
> + ____|____ |
> + | | |
> + |________| |
> + |_________|
> + reaching_out
> +*/
> static bool
> loop_basic_block_p (const basic_block cfg_bb)
> {
> - return JUMP_P (BB_END (cfg_bb)) && any_condjump_p (BB_END (cfg_bb));
> + if (JUMP_P (BB_END (cfg_bb)) && any_condjump_p (BB_END (cfg_bb)))
> + {
> + edge e;
> + edge_iterator ei;
> + FOR_EACH_EDGE (e, ei, cfg_bb->succs)
> + if (e->dest->index == cfg_bb->index)
> + return true;
> + }
> + return false;
> }
>
> /* Return true if it is an RVV instruction depends on VTYPE global
> @@ -271,26 +288,60 @@ backward_propagate_worthwhile_p (const basic_block
> cfg_bb,
> {
> if (loop_basic_block_p (cfg_bb))
> {
> - if (block_info.local_dem.compatible_p (block_info.reaching_out))
> - return true;
> -
> - /* There is a obvious case that is not worthwhile and meaningless
> - to propagate the demand information:
> - local_dem
> - __________
> - ____|____ |
> - | | |
> - |________| |
> - |_________|
> - reaching_out
> - Header is incompatible with reaching_out and the block is loop
> itself,
> - we don't backward propagate the local_dem since we can't avoid
> emit
> - vsetvl for the local_dem. */
> - edge e;
> - edge_iterator ei;
> - FOR_EACH_EDGE (e, ei, cfg_bb->succs)
> - if (e->dest->index == cfg_bb->index)
> + if (block_info.reaching_out.valid_or_dirty_p ())
> + {
> + if (block_info.local_dem.compatible_p (block_info.reaching_out))
> + {
> + /* Case 1 (Can backward propagate):
> + ....
> + bb0:
> + ...
> + for (int i = 0; i < n; i++)
> + {
> + vint16mf4_t v = __riscv_vle16_v_i16mf4 (in + i + 5,
> 7);
> + __riscv_vse16_v_i16mf4 (out + i + 5, v, 7);
> + }
> + The local_dem is compatible with reaching_out. Such case
> is
> + worthwhile backward propagation. */
> + return true;
> + }
> + else
> + {
> + /* Case 2 (Don't backward propagate):
> + ....
> + bb0:
> + ...
> + for (int i = 0; i < n; i++)
> + {
> + vint16mf4_t v = __riscv_vle16_v_i16mf4 (in + i +
> 5, 7);
> + __riscv_vse16_v_i16mf4 (out + i + 5, v, 7);
> + vint16mf2_t v2 = __riscv_vle16_v_i16mf2 (in + i +
> 6, 8);
> + __riscv_vse16_v_i16mf2 (out + i + 6, v, 8);
> + }
> + The local_dem is incompatible with reaching_out.
> + It makes no sense to backward propagate the local_dem
> since we
> + can't avoid VSETVL inside the loop. */
> + return false;
> + }
> + }
> + else
> + {
> + gcc_assert (block_info.reaching_out.unknown_p ());
> + /* Case 3 (Don't backward propagate):
> + ....
> + bb0:
> + ...
> + for (int i = 0; i < n; i++)
> + {
> + vint16mf4_t v = __riscv_vle16_v_i16mf4 (in + i + 5, 7);
> + __riscv_vse16_v_i16mf4 (out + i + 5, v, 7);
> + fn3 ();
> + }
> + The local_dem is VALID, but the reaching_out is UNKNOWN.
> + It makes no sense to backward propagate the local_dem since we
> + can't avoid VSETVL inside the loop. */
> return false;
> + }
> }
>
> return true;
> --
> 2.36.3
>
>
prev parent reply other threads:[~2023-01-26 19:13 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-03 7:16 juzhe.zhong
2023-01-26 19:13 ` Kito Cheng [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='CA+yXCZB-qRC7B-OCJyqFUhY-CHo=uPeDyAFmvDj3SCLF63acOQ@mail.gmail.com' \
--to=kito.cheng@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=juzhe.zhong@rivai.ai \
--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).