From: Kyrill Tkachov <kyrylo.tkachov@foss.arm.com>
To: Wilco Dijkstra <Wilco.Dijkstra@arm.com>,
GCC Patches <gcc-patches@gcc.gnu.org>,
Richard Earnshaw <Richard.Earnshaw@arm.com>
Cc: nd <nd@arm.com>
Subject: Re: [PATCH][ARM] Cleanup 64-bit multiplies
Date: Wed, 18 Sep 2019 16:30:00 -0000 [thread overview]
Message-ID: <64f47d61-a838-fae9-246d-1f22e80e8dd1@foss.arm.com> (raw)
In-Reply-To: <VI1PR0801MB21277C99DAB935AD89EACA4A83B70@VI1PR0801MB2127.eurprd08.prod.outlook.com>
Hi Wilco,
On 9/9/19 6:08 PM, Wilco Dijkstra wrote:
> ping
>
>
> Cleanup 64-bit multiplies. Combine the expanders using iterators.
> Â Merge the signed/unsigned multiplies as well as the pre-Armv6 and Armv6
>  variants. Split DImode operands early into parallel sets inside the
> Â MULL/MLAL instructions - this improves register allocation and avoids
> Â subreg issues due to other DImode operations splitting early.
>
Hmm... quite a lot going on this patch. Perhaps breaking it into a
series would have been easier.
But I think I untangled it all and it looks like a good improvement.
Ok.
Thanks,
Kyrill
> Â Bootstrap OK on armhf, regress passes.
>
> Â ChangeLog:
>  2019-09-03 Wilco Dijkstra <wdijkstr@arm.com>
>
> Â Â Â Â Â Â Â Â * config/arm/arm.md (maddsidi4): Remove expander.
> Â Â Â Â Â Â Â Â (mulsidi3adddi): Remove pattern.
> Â Â Â Â Â Â Â Â (mulsidi3adddi_v6): Likewise.
> Â Â Â Â Â Â Â Â (mulsidi3_nov6): Likewise.
> Â Â Â Â Â Â Â Â (mulsidi3_v6): Likewise.
> Â Â Â Â Â Â Â Â (umulsidi3): Remove expander.
> Â Â Â Â Â Â Â Â (umulsidi3_nov6): Remove pattern.
> Â Â Â Â Â Â Â Â (umulsidi3_v6): Likewise.
> Â Â Â Â Â Â Â Â (umulsidi3adddi): Likewise.
> Â Â Â Â Â Â Â Â (umulsidi3adddi_v6): Likewise.
> Â Â Â Â Â Â Â Â (<Us>mulsidi3): Add combined expander.
> Â Â Â Â Â Â Â Â (<Us>maddsidi4): Likewise.
> Â Â Â Â Â Â Â Â (<US>mull): Add combined umull and smull pattern.
> Â Â Â Â Â Â Â Â (<US>mlal): Likewise.
> Â Â Â Â Â Â Â Â * config/arm/iterators.md (Us): Add new iterator.
> Â --
> Â diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
> Â index
> 1ab203810bf143927a8afa0d00d82537cd7c75ed..c1fea4abdbccedbbbed9a25cab133de5cacb1afb
> 100644
> Â --- a/gcc/config/arm/arm.md
> Â +++ b/gcc/config/arm/arm.md
> Â @@ -1636,144 +1636,80 @@ (define_insn "*mls"
> Â Â Â Â (set_attr "predicable" "yes")]
> Â )
>
> Â -(define_expand "maddsidi4"
> Â -Â [(set (match_operand:DI 0 "s_register_operand")
> Â -Â Â Â Â Â Â (plus:DI
> Â -Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 1 "s_register_operand"))
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand")))
> Â -Â Â Â Â Â Â Â (match_operand:DI 3 "s_register_operand")))]
> Â -Â "TARGET_32BIT"
> Â -Â "")
> Â -
> Â -(define_insn "*mulsidi3adddi"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=&r")
> Â -Â Â Â Â Â Â (plus:DI
> Â -Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand" "%r"))
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 3 "s_register_operand" "r")))
> Â -Â Â Â Â Â Â Â (match_operand:DI 1 "s_register_operand" "0")))]
> Â -Â "TARGET_32BIT && !arm_arch6"
> Â -Â "smlal%?\\t%Q0, %R0, %3, %2"
> Â -Â [(set_attr "type" "smlal")
> Â -Â Â (set_attr "predicable" "yes")]
> Â -)
> Â -
> Â -(define_insn "*mulsidi3adddi_v6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=r")
> Â -Â Â Â Â Â Â (plus:DI
> Â -Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand" "r"))
> Â -Â Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 3 "s_register_operand" "r")))
> Â -Â Â Â Â Â Â Â (match_operand:DI 1 "s_register_operand" "0")))]
> Â -Â "TARGET_32BIT && arm_arch6"
> Â -Â "smlal%?\\t%Q0, %R0, %3, %2"
> Â -Â [(set_attr "type" "smlal")
> Â -Â Â (set_attr "predicable" "yes")]
> Â -)
> Â -
> Â ;; 32x32->64 widening multiply.
> Â -;; As with mulsi3, the only difference between the v3-5 and v6+
> Â -;; versions of these patterns is the requirement that the output not
> Â -;; overlap the inputs, but that still means we have to have a named
> Â -;; expander and two different starred insns.
> Â +;; The only difference between the v3-5 and v6+ versions is the
> requirement
> Â +;; that the output does not overlap with either input.
>
> Â -(define_expand "mulsidi3"
> Â +(define_expand "<Us>mulsidi3"
> Â Â Â [(set (match_operand:DI 0 "s_register_operand")
> Â Â Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 1 "s_register_operand"))
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand"))))]
> Â +Â Â Â Â Â Â Â (SE:DI (match_operand:SI 1 "s_register_operand"))
> Â +Â Â Â Â Â Â Â (SE:DI (match_operand:SI 2 "s_register_operand"))))]
> Â Â Â "TARGET_32BIT"
> Â -Â ""
> Â -)
> Â -
> Â -(define_insn "*mulsidi3_nov6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=&r")
> Â -Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 1 "s_register_operand" "%r"))
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand"
> "r"))))]
> Â -Â "TARGET_32BIT && !arm_arch6"
> Â -Â "smull%?\\t%Q0, %R0, %1, %2"
> Â -Â [(set_attr "type" "smull")
> Â -Â Â (set_attr "predicable" "yes")]
> Â -)
> Â -
> Â -(define_insn "*mulsidi3_v6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=r")
> Â -Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 1 "s_register_operand" "r"))
> Â -Â Â Â Â Â Â Â (sign_extend:DI (match_operand:SI 2 "s_register_operand"
> "r"))))]
> Â -Â "TARGET_32BIT && arm_arch6"
> Â -Â "smull%?\\t%Q0, %R0, %1, %2"
> Â -Â [(set_attr "type" "smull")
> Â -Â Â (set_attr "predicable" "yes")]
> Â +Â {
> Â +Â Â Â Â Â emit_insn (gen_<US>mull (gen_lowpart (SImode, operands[0]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â gen_highpart (SImode, operands[0]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â operands[1], operands[2]));
> Â +Â Â Â Â Â DONE;
> Â +Â }
> Â )
>
> Â -(define_expand "umulsidi3"
> Â -Â [(set (match_operand:DI 0 "s_register_operand")
> Â -Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 1 "s_register_operand"))
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand"))))]
> Â +(define_insn "<US>mull"
> Â +Â [(set (match_operand:SI 0 "s_register_operand" "=r,&r")
> Â +Â Â Â Â Â Â (mult:SI
> Â +Â Â Â Â Â Â Â (match_operand:SI 2 "s_register_operand" "%r,r")
> Â +Â Â Â Â Â Â Â (match_operand:SI 3 "s_register_operand" "r,r")))
> Â +Â Â (set (match_operand:SI 1 "s_register_operand" "=r,&r")
> Â +Â Â Â Â Â Â (truncate:SI
> Â +Â Â Â Â Â Â Â (lshiftrt:DI
> Â +Â Â Â Â Â Â Â Â (mult:DI (SE:DI (match_dup 2)) (SE:DI (match_dup 3)))
> Â +Â Â Â Â Â Â Â Â (const_int 32))))]
> Â Â Â "TARGET_32BIT"
> Â -Â ""
> Â -)
> Â -
> Â -(define_insn "*umulsidi3_nov6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=&r")
> Â -Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 1 "s_register_operand" "%r"))
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand"
> "r"))))]
> Â -Â "TARGET_32BIT && !arm_arch6"
> Â -Â "umull%?\\t%Q0, %R0, %1, %2"
> Â -Â [(set_attr "type" "umull")
> Â -Â Â (set_attr "predicable" "yes")]
> Â -)
> Â -
> Â -(define_insn "*umulsidi3_v6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=r")
> Â -Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 1 "s_register_operand" "r"))
> Â -Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand"
> "r"))))]
> Â -Â "TARGET_32BIT && arm_arch6"
> Â -Â "umull%?\\t%Q0, %R0, %1, %2"
> Â +Â "<US>mull%?\\t%0, %1, %2, %3"
> Â Â Â [(set_attr "type" "umull")
> Â -Â Â (set_attr "predicable" "yes")]
> Â +Â Â (set_attr "predicable" "yes")
> Â +Â Â (set_attr "arch" "v6,nov6")]
> Â )
>
> Â -(define_expand "umaddsidi4"
> Â +(define_expand "<Us>maddsidi4"
> Â Â Â [(set (match_operand:DI 0 "s_register_operand")
> Â Â Â Â Â Â Â Â Â (plus:DI
> Â Â Â Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 1 "s_register_operand"))
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand")))
> Â +Â Â Â Â Â Â Â Â (SE:DI (match_operand:SI 1 "s_register_operand"))
> Â +Â Â Â Â Â Â Â Â (SE:DI (match_operand:SI 2 "s_register_operand")))
> Â Â Â Â Â Â Â Â Â Â (match_operand:DI 3 "s_register_operand")))]
> Â Â Â "TARGET_32BIT"
> Â -Â "")
> Â -
> Â -(define_insn "*umulsidi3adddi"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=&r")
> Â -Â Â Â Â Â Â (plus:DI
> Â -Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand" "%r"))
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 3 "s_register_operand" "r")))
> Â -Â Â Â Â Â Â Â (match_operand:DI 1 "s_register_operand" "0")))]
> Â -Â "TARGET_32BIT && !arm_arch6"
> Â -Â "umlal%?\\t%Q0, %R0, %3, %2"
> Â -Â [(set_attr "type" "umlal")
> Â -Â Â (set_attr "predicable" "yes")]
> Â +Â {
> Â +Â Â Â Â Â emit_insn (gen_<US>mlal (gen_lowpart (SImode, operands[0]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â gen_lowpart (SImode, operands[3]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â gen_highpart (SImode, operands[0]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â gen_highpart (SImode, operands[3]),
> Â +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â operands[1], operands[2]));
> Â +Â Â Â Â Â DONE;
> Â +Â }
> Â )
>
> Â -(define_insn "*umulsidi3adddi_v6"
> Â -Â [(set (match_operand:DI 0 "s_register_operand" "=r")
> Â -Â Â Â Â Â Â (plus:DI
> Â -Â Â Â Â Â Â Â (mult:DI
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 2 "s_register_operand" "r"))
> Â -Â Â Â Â Â Â Â Â (zero_extend:DI (match_operand:SI 3 "s_register_operand" "r")))
> Â -Â Â Â Â Â Â Â (match_operand:DI 1 "s_register_operand" "0")))]
> Â -Â "TARGET_32BIT && arm_arch6"
> Â -Â "umlal%?\\t%Q0, %R0, %3, %2"
> Â +(define_insn "<US>mlal"
> Â +Â [(set (match_operand:SI 0 "s_register_operand" "=r,&r")
> Â +Â Â Â Â Â Â (plus:SI
> Â +Â Â Â Â Â Â Â (mult:SI
> Â +Â Â Â Â Â Â Â Â (SE:DI (match_operand:SI 4 "s_register_operand" "%r,r"))
> Â +Â Â Â Â Â Â Â Â (SE:DI (match_operand:SI 5 "s_register_operand" "r,r")))
> Â +Â Â Â Â Â Â Â (match_operand:SI 1 "s_register_operand" "0,0")))
> Â +Â Â (set (match_operand:SI 2 "s_register_operand" "=r,&r")
> Â +Â Â Â Â Â Â (plus:SI
> Â +Â Â Â Â Â Â Â (truncate:SI
> Â +Â Â Â Â Â Â Â Â (lshiftrt:DI
> Â +Â Â Â Â Â Â Â Â Â (plus:DI
> Â +Â Â Â Â Â Â Â Â Â Â (mult:DI (SE:DI (match_dup 4)) (SE:DI (match_dup 5)))
> Â +Â Â Â Â Â Â Â Â Â Â (zero_extend:DI (match_dup 1)))
> Â +Â Â Â Â Â Â Â Â Â (const_int 32)))
> Â +Â Â Â Â Â Â Â (match_operand:SI 3 "s_register_operand" "2,2")))]
> Â +Â "TARGET_32BIT"
> Â +Â "<US>mlal%?\\t%0, %2, %4, %5"
> Â Â Â [(set_attr "type" "umlal")
> Â -Â Â (set_attr "predicable" "yes")]
> Â +Â Â (set_attr "predicable" "yes")
> Â +Â Â (set_attr "arch" "v6,nov6")]
> Â )
>
> Â (define_expand "<US>mulsi3_highpart"
> Â diff --git a/gcc/config/arm/iterators.md b/gcc/config/arm/iterators.md
> Â index
> fa6f0c0529d5364b1e1df705cb1029868578e38c..c29897a3b70d342b025c72b8c032bb3bb831040f
> 100644
> Â --- a/gcc/config/arm/iterators.md
> Â +++ b/gcc/config/arm/iterators.md
> Â @@ -796,6 +796,7 @@ (define_code_attr optab [(ltu "ltu") (geu "geu")])
>
> Â ;; Assembler mnemonics for signedness of widening operations.
> Â (define_code_attr US [(sign_extend "s") (zero_extend "u")])
> Â +(define_code_attr Us [(sign_extend "") (zero_extend "u")])
>
> Â ;; Signedness suffix for float->fixed conversions. Empty for signed
> Â ;; conversion.
prev parent reply other threads:[~2019-09-18 16:30 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-03 15:42 Wilco Dijkstra
2019-09-09 17:08 ` Wilco Dijkstra
2019-09-18 16:30 ` Kyrill Tkachov [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=64f47d61-a838-fae9-246d-1f22e80e8dd1@foss.arm.com \
--to=kyrylo.tkachov@foss.arm.com \
--cc=Richard.Earnshaw@arm.com \
--cc=Wilco.Dijkstra@arm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=nd@arm.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).