From: Takayuki 'January June' Suwa <jjsuwa_sys3175@yahoo.co.jp>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: [PATCH 4/5] xtensa: Eliminate unwanted reg-reg moves during DFmode input reloads
Date: Tue, 14 Jun 2022 12:39:49 +0900 [thread overview]
Message-ID: <0ebde82d-4449-c731-1656-0e1b29d5ed9c@yahoo.co.jp> (raw)
When spilled DFmode registers are reloaded in, once loaded into a pair of
SImode regs and then copied from that regs. Such unwanted reg-reg moves
seems not to be eliminated at the "cprop_hardreg" stage, despite no problem
in output reloads.
Luckily it is easy to resolve such inefficiencies, with the use of peephole2
pattern.
gcc/ChangeLog:
* config/xtensa/predicates.md (reload_operand):
New predicate.
* config/xtensa/xtensa.md: New peephole2 pattern.
---
gcc/config/xtensa/predicates.md | 13 +++++++++++++
gcc/config/xtensa/xtensa.md | 31 +++++++++++++++++++++++++++++++
2 files changed, 44 insertions(+)
diff --git a/gcc/config/xtensa/predicates.md b/gcc/config/xtensa/predicates.md
index d63a6cf034c..edd13ae41b9 100644
--- a/gcc/config/xtensa/predicates.md
+++ b/gcc/config/xtensa/predicates.md
@@ -165,6 +165,19 @@
(and (match_code "const_int")
(match_test "xtensa_mem_offset (INTVAL (op), SFmode)")))
+(define_predicate "reload_operand"
+ (match_code "mem")
+{
+ const_rtx addr = XEXP (op, 0);
+ if (REG_P (addr))
+ return REGNO (addr) == A1_REG;
+ if (GET_CODE (addr) == PLUS)
+ return REG_P (XEXP (addr, 0))
+ && REGNO (XEXP (addr, 0)) == A1_REG
+ && CONST_INT_P (XEXP (addr, 1));
+ return false;
+})
+
(define_predicate "branch_operator"
(match_code "eq,ne,lt,ge"))
diff --git a/gcc/config/xtensa/xtensa.md b/gcc/config/xtensa/xtensa.md
index 9588a829136..e6f5594762f 100644
--- a/gcc/config/xtensa/xtensa.md
+++ b/gcc/config/xtensa/xtensa.md
@@ -2724,3 +2724,34 @@
(if_then_else (match_test "TARGET_DENSITY")
(const_int 5)
(const_int 6)))])
+
+(define_peephole2
+ [(set (match_operand:SI 0 "register_operand")
+ (match_operand:SI 6 "reload_operand"))
+ (set (match_operand:SI 1 "register_operand")
+ (match_operand:SI 7 "reload_operand"))
+ (set (match_operand:SF 2 "register_operand")
+ (match_operand:SF 4 "register_operand"))
+ (set (match_operand:SF 3 "register_operand")
+ (match_operand:SF 5 "register_operand"))]
+ "REGNO (operands[0]) == REGNO (operands[4])
+ && REGNO (operands[1]) == REGNO (operands[5])
+ && peep2_reg_dead_p (4, operands[0])
+ && peep2_reg_dead_p (4, operands[1])"
+ [(set (match_dup 2)
+ (match_dup 6))
+ (set (match_dup 3)
+ (match_dup 7))]
+{
+ uint32_t check = 0;
+ int i;
+ for (i = 0; i <= 3; ++i)
+ {
+ uint32_t mask = (uint32_t)1 << REGNO (operands[i]);
+ if (check & mask)
+ FAIL;
+ check |= mask;
+ }
+ operands[6] = gen_rtx_MEM (SFmode, XEXP (operands[6], 0));
+ operands[7] = gen_rtx_MEM (SFmode, XEXP (operands[7], 0));
+})
--
2.20.1
next reply other threads:[~2022-06-14 3:54 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-06-14 3:39 Takayuki 'January June' Suwa [this message]
2022-06-16 0:02 ` Max Filippov
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=0ebde82d-4449-c731-1656-0e1b29d5ed9c@yahoo.co.jp \
--to=jjsuwa_sys3175@yahoo.co.jp \
--cc=gcc-patches@gcc.gnu.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).