Re: [PATCH][ARM] Remove support for MULS - Richard Earnshaw (lists)

public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed

From: "Richard Earnshaw (lists)" <Richard.Earnshaw@arm.com>
To: Kyrill Tkachov <kyrylo.tkachov@foss.arm.com>,
	Wilco Dijkstra <Wilco.Dijkstra@arm.com>,
	GCC Patches <gcc-patches@gcc.gnu.org>
Cc: nd <nd@arm.com>
Subject: Re: [PATCH][ARM] Remove support for MULS
Date: Thu, 19 Sep 2019 09:52:00 -0000	[thread overview]
Message-ID: <519898ff-e3b5-79b0-c5ab-d2c844ee024c@arm.com> (raw)
In-Reply-To: <4061d3a8-202c-eb7b-02f4-c88d0da7e2c0@foss.arm.com>

On 18/09/2019 17:31, Kyrill Tkachov wrote:
> Hi Wilco,
> 
> On 9/9/19 6:07 PM, Wilco Dijkstra wrote:
>> ping
>>
>>
>> Remove various MULS/MLAS patterns which are enabled when optimizing for
>> Â size.Â  However the codesize gain from these patterns is so minimal that
>> Â there is no point in keeping them.
>>
> I disagree. If they still trigger and generate better code than without 
> we should keep them.
> 
> What kind of code is *common* varies greatly from user to user.

Also, the main reason for restricting their use was that in the 'olden 
days', when we had multi-cycle implementations of the multiply 
instructions with short-circuit fast termination when the result was 
completed, the flag setting variants would never short-circuit.

These days we have fixed cycle counts for multiply instructions, so this 
is no-longer a penalty.  In the thumb2 case in particular we can often 
reduce mul-cmp (6 bytes) to muls (2 bytes), that's a 66% saving on this 
sequence and definitely worth exploiting when we can, even if it's not 
all that common.

In fact, give the latest architectural implementations, I think we 
should look again at re-enabling these when doing normal optimization as 
well.

R.

> 
> Thanks,
> 
> Kyrill
> 
> 
>> Â Bootstrap OK on armhf, regress passes.
>>
>> Â ChangeLog:
>> Â 2019-09-03Â  Wilco DijkstraÂ  <wdijkstr@arm.com>
>>
>> Â Â Â Â Â Â Â Â  * config/arm/arm.md (mulsi3_compare0): Remove pattern.
>> Â Â Â Â Â Â Â Â  (mulsi3_compare0_v6): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi_compare0_scratch): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi_compare0_scratch_v6): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi3addsi_compare0): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi3addsi_compare0_v6): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi3addsi_compare0_scratch): Likewise.
>> Â Â Â Â Â Â Â Â  (mulsi3addsi_compare0_scratch_v6): Likewise.
>> Â Â Â Â Â Â Â Â  * config/arm/thumb2.md (thumb2_mulsi_short_compare0): Remove 
>> pattern.
>> Â Â Â Â Â Â Â Â  (thumb2_mulsi_short_compare0_scratch): Likewise.
>>
>> Â --
>> Â diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
>> Â index 
>> 738d42fd164f117f1dec1108a824d984ccd70d09..66dafdc47b7cfc37c131764e482d47bcaab90538 
>> 100644
>> Â --- a/gcc/config/arm/arm.md
>> Â +++ b/gcc/config/arm/arm.md
>> Â @@ -1618,60 +1618,6 @@ (define_insn "*arm_mulsi3_v6"
>> Â Â Â Â  (set_attr "predicable_short_it" "yes,yes,no")]
>> Â  )
>>
>> Â -(define_insn "*mulsi3_compare0"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV (mult:SI
>> 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  (match_operand:SI 2 "s_register_operand" 
>> "r,r")
>> 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  (match_operand:SI 1 "s_register_operand" 
>> "%0,r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (set (match_operand:SI 0 "s_register_operand" "=&r,&r")
>> Â -Â Â Â Â Â Â  (mult:SI (match_dup 2) (match_dup 1)))]
>> Â -Â  "TARGET_ARM && !arm_arch6"
>> Â -Â  "muls%?\\t%0, %2, %1"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "muls")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi3_compare0_v6"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV (mult:SI
>> 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  (match_operand:SI 2 "s_register_operand" "r")
>> 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  (match_operand:SI 1 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (set (match_operand:SI 0 "s_register_operand" "=r")
>> Â -Â Â Â Â Â Â  (mult:SI (match_dup 2) (match_dup 1)))]
>> Â -Â  "TARGET_ARM && arm_arch6 && optimize_size"
>> Â -Â  "muls%?\\t%0, %2, %1"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "muls")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi_compare0_scratch"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV (mult:SI
>> 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  (match_operand:SI 2 "s_register_operand" 
>> "r,r")
>> 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  (match_operand:SI 1 "s_register_operand" 
>> "%0,r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (clobber (match_scratch:SI 0 "=&r,&r"))]
>> Â -Â  "TARGET_ARM && !arm_arch6"
>> Â -Â  "muls%?\\t%0, %2, %1"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "muls")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi_compare0_scratch_v6"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV (mult:SI
>> 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  (match_operand:SI 2 "s_register_operand" "r")
>> 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  (match_operand:SI 1 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (clobber (match_scratch:SI 0 "=r"))]
>> Â -Â  "TARGET_ARM && arm_arch6 && optimize_size"
>> Â -Â  "muls%?\\t%0, %2, %1"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "muls")]
>> Â -)
>> Â -
>> Â  ;; Unnamed templates to match MLA instruction.
>>
>> Â  (define_insn "*mulsi3addsi"
>> Â @@ -1698,70 +1644,6 @@ (define_insn "*mulsi3addsi_v6"
>> Â Â Â Â  (set_attr "predicable" "yes")]
>> Â  )
>>
>> Â -(define_insn "*mulsi3addsi_compare0"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â  (plus:SI (mult:SI
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "s_register_operand" "r,r,r,r")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 1 "s_register_operand" "%0,r,0,r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 3 "s_register_operand" "r,r,0,0"))
>> Â -Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (set (match_operand:SI 0 "s_register_operand" "=&r,&r,&r,&r")
>> Â -Â Â Â Â Â Â  (plus:SI (mult:SI (match_dup 2) (match_dup 1))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_dup 3)))]
>> Â -Â  "TARGET_ARM && arm_arch6"
>> Â -Â  "mlas%?\\t%0, %2, %1, %3"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "mlas")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi3addsi_compare0_v6"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â  (plus:SI (mult:SI
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "s_register_operand" "r")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 1 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 3 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (set (match_operand:SI 0 "s_register_operand" "=r")
>> Â -Â Â Â Â Â Â  (plus:SI (mult:SI (match_dup 2) (match_dup 1))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_dup 3)))]
>> Â -Â  "TARGET_ARM && arm_arch6 && optimize_size"
>> Â -Â  "mlas%?\\t%0, %2, %1, %3"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "mlas")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi3addsi_compare0_scratch"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â  (plus:SI (mult:SI
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "s_register_operand" "r,r,r,r")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 1 "s_register_operand" "%0,r,0,r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 3 "s_register_operand" "?r,r,0,0"))
>> Â -Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (clobber (match_scratch:SI 0 "=&r,&r,&r,&r"))]
>> Â -Â  "TARGET_ARM && !arm_arch6"
>> Â -Â  "mlas%?\\t%0, %2, %1, %3"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "mlas")]
>> Â -)
>> Â -
>> Â -(define_insn "*mulsi3addsi_compare0_scratch_v6"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â  (plus:SI (mult:SI
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "s_register_operand" "r")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 1 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 3 "s_register_operand" "r"))
>> Â -Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (clobber (match_scratch:SI 0 "=r"))]
>> Â -Â  "TARGET_ARM && arm_arch6 && optimize_size"
>> Â -Â  "mlas%?\\t%0, %2, %1, %3"
>> Â -Â  [(set_attr "conds" "set")
>> Â -Â Â  (set_attr "type" "mlas")]
>> Â -)
>> Â -
>> Â  (define_insn "*mulsi3subsi"
>> Â Â Â  [(set (match_operand:SI 0 "s_register_operand" "=r")
>> Â Â Â Â Â Â Â Â Â  (minus:SI
>> Â diff --git a/gcc/config/arm/thumb2.md b/gcc/config/arm/thumb2.md
>> Â index 
>> 6ccc875e2b4e7b8ce256e52da966dfe220c6f5d6..8e26689b66263e7304a0da6163ceccfb4483d3e7 
>> 100644
>> Â --- a/gcc/config/arm/thumb2.md
>> Â +++ b/gcc/config/arm/thumb2.md
>> Â @@ -1381,31 +1381,6 @@ (define_insn "*thumb2_mulsi_short"
>> Â Â Â Â  (set_attr "length" "2")
>> Â Â Â Â  (set_attr "type" "muls")])
>>
>> Â -(define_insn "*thumb2_mulsi_short_compare0"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â Â  (mult:SI (match_operand:SI 1 "register_operand" "%0")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "register_operand" "l"))
>> Â -Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (set (match_operand:SI 0 "register_operand" "=l")
>> Â -Â Â Â Â Â Â  (mult:SI (match_dup 1) (match_dup 2)))]
>> Â -Â  "TARGET_THUMB2 && optimize_size"
>> Â -Â  "muls\\t%0, %2, %0"
>> Â -Â  [(set_attr "length" "2")
>> Â -Â Â  (set_attr "type" "muls")])
>> Â -
>> Â -(define_insn "*thumb2_mulsi_short_compare0_scratch"
>> Â -Â  [(set (reg:CC_NOOV CC_REGNUM)
>> Â -Â Â Â Â Â Â Â  (compare:CC_NOOV
>> Â -Â Â Â Â Â Â Â Â  (mult:SI (match_operand:SI 1 "register_operand" "%0")
>> Â -Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (match_operand:SI 2 "register_operand" "l"))
>> Â -Â Â Â Â Â Â Â Â  (const_int 0)))
>> Â -Â Â  (clobber (match_scratch:SI 0 "=l"))]
>> Â -Â  "TARGET_THUMB2 && optimize_size"
>> Â -Â  "muls\\t%0, %2, %0"
>> Â -Â  [(set_attr "length" "2")
>> Â -Â Â  (set_attr "type" "muls")])
>> Â -
>> Â  (define_insn "*thumb2_cbz"
>> Â Â Â  [(set (pc) (if_then_else
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  (eq (match_operand:SI 0 "s_register_operand" "l,?r")

next prev parent reply	other threads:[~2019-09-19  9:52 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-09-03 15:37 Wilco Dijkstra
2019-09-09 17:07 ` Wilco Dijkstra
2019-09-18 16:31   ` Kyrill Tkachov
2019-09-19  9:52     ` Richard Earnshaw (lists) [this message]
2019-09-19 16:47       ` Wilco Dijkstra
2019-10-10 17:20         ` Wilco Dijkstra
2020-02-04 10:58           ` Wilco Dijkstra

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=519898ff-e3b5-79b0-c5ab-d2c844ee024c@arm.com \
    --to=richard.earnshaw@arm.com \
    --cc=Wilco.Dijkstra@arm.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=kyrylo.tkachov@foss.arm.com \
    --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).