diff mbox

[Atom] Improve AGU stalls avoidance optimization

Message ID CAFULd4ZXsexGvEuMpWmHCGxSk3g4_5SGZRmYnNJ_pbHspOiFbA@mail.gmail.com
State New
Headers show

Commit Message

Uros Bizjak Sept. 3, 2011, 2:24 p.m. UTC 
... Sent again, with correct Cc and subject line ...

Hello!

> Here is a patch which adds few more splits for AGU stalls avoidance on
> Atom. It also fixes cost model and detects AGU stalls more
> efficiently.
>
> Bootstrapped and checked on x86_64-linux.
>
> 2011-09-02  Enkovich Ilya  <ilya.enkovich@intel.com>
>
>       * config/i386/i386-protos.h (ix86_lea_outperforms): New.
>       (ix86_avoid_lea_for_add): Likewise.
>       (ix86_avoid_lea_for_addr): Likewise.
>       (ix86_split_lea_for_addr): Likewise.
>
>       * config/i386/i386.c (LEA_MAX_STALL): New.
>       (increase_distance): Likewise.
>       (insn_defines_reg): Likewise.
>       (insn_uses_reg_mem): Likewise.
>       (distance_non_agu_define_in_bb): Likewise.
>       (distance_agu_use_in_bb): Likewise.
>       (ix86_lea_outperforms): Likewise.
>       (ix86_ok_to_clobber_flags): Likewise.
>       (ix86_avoid_lea_for_add): Likewise.
>       (ix86_avoid_lea_for_addr): Likewise.
>       (ix86_split_lea_for_addr): Likewise.
>       (distance_non_agu_define): Search in pred BBs added.
>       (distance_agu_use): Search in succ BBs added.
>       (IX86_LEA_PRIORITY): Value changed from 2 to 0.
>       (LEA_SEARCH_THRESHOLD): Now depends on LEA_MAX_STALL.
>       (ix86_lea_for_add_ok): Use ix86_lea_outperforms to make decision.
>
>       * config/i386/i386.md: Splits added to transform lea into a
>       sequence of instructions.

Did you also test on x32 ? H.J.'s x32 page [1] currently says that
Atom LEA optimization is disabled on x32 for some reason.

The patch looks OK to me, with a few nits below.

[1] https://sites.google.com/site/x32abi/

--- a/gcc/config/i386/i386-protos.h
+++ b/gcc/config/i386/i386-protos.h

+bool
+ix86_avoid_lea_for_addr (rtx insn, rtx operands[])
+{
+  unsigned int regno0 = true_regnum (operands[0]) ;
+  unsigned int regno1 = -1;
+  unsigned int regno2 = -1;

Use INVALID_REGNUM here.

+extern void
+ix86_split_lea_for_addr (rtx operands[], enum machine_mode mode)
+{

Missing comment.

Comments

Ilya Enkovich Sept. 6, 2011, 12:26 p.m. UTC | #1 
Hello,

Thanks for review!

2011/9/3 Uros Bizjak <ubizjak@gmail.com>:
> Did you also test on x32 ? H.J.'s x32 page [1] currently says that
> Atom LEA optimization is disabled on x32 for some reason.
No. I did not try to cover x32. It will be a separate work.

> +bool
> +ix86_avoid_lea_for_addr (rtx insn, rtx operands[])
> +{
> +  unsigned int regno0 = true_regnum (operands[0]) ;
> +  unsigned int regno1 = -1;
> +  unsigned int regno2 = -1;
>
> Use INVALID_REGNUM here.
Fixed. Also used  INVALID_REGNUM in other places where -1 was used as
invalid register number.

>
> +extern void
> +ix86_split_lea_for_addr (rtx operands[], enum machine_mode mode)
> +{
>
> Missing comment.
Fixed

> +;; Split non destructive adds if we cannot use lea.
> +(define_split
> +  [(set (match_operand:SWI48 0 "register_operand" "")
> +       (plus:SWI48 (match_operand:SWI48 1 "register_operand" "")
> +              (match_operand:SWI48 2 "nonmemory_operand" "")))
> +   (clobber (reg:CC FLAGS_REG))]
> +  "reload_completed && ix86_avoid_lea_for_add (insn, operands)"
> +  [(set (match_dup 0) (match_dup 1))
> +   (parallel [(set (match_dup 0) (plus:<MODE> (match_dup 0) (match_dup 2)))
> +             (clobber (reg:CC FLAGS_REG))])
> +  ]
> +)
>
> Put all closing braces on one line:
Fixed.

> +;; Split lea into one or more ALU instructions if profitable.
> +(define_split
> +  [(set (match_operand:SI 0 "register_operand" "")
> +       (subreg:SI (match_operand:DI 1 "lea_address_operand" "") 0))]
> +  "reload_completed && ix86_avoid_lea_for_addr (insn, operands)"
> +  [(const_int 0)]
> +{
> +  ix86_split_lea_for_addr (operands, SImode);
> +  DONE;
> +})
>
> This is valid only for TARGET_64BIT.
Fixed.

> Please note that x32 adds quite some different LEA patterns (see
> i386.md, line 5466+). I suggest you merge your splitters with these
> define_insn patterns into define_insn_and_split, adding "&&
> reload_completed && ix86_avoid_lea_for_addr (insn, operands)" as a
> split condition.
Thanks for the note. I'll look at new patterns when we enable lea
optimization for x32.

>
> Uros.
>

Is fixed version OK?

Thanks,
Ilya
---
gcc/

2011-09-06  Enkovich Ilya  <ilya.enkovich@intel.com>

	* config/i386/i386-protos.h (ix86_lea_outperforms): New.
	(ix86_avoid_lea_for_add): Likewise.
	(ix86_avoid_lea_for_addr): Likewise.
	(ix86_split_lea_for_addr): Likewise.

	* config/i386/i386.c (LEA_MAX_STALL): New.
	(increase_distance): Likewise.
	(insn_defines_reg): Likewise.
	(insn_uses_reg_mem): Likewise.
	(distance_non_agu_define_in_bb): Likewise.
	(distance_agu_use_in_bb): Likewise.
	(ix86_lea_outperforms): Likewise.
	(ix86_ok_to_clobber_flags): Likewise.
	(ix86_avoid_lea_for_add): Likewise.
	(ix86_avoid_lea_for_addr): Likewise.
	(ix86_split_lea_for_addr): Likewise.
	(distance_non_agu_define): Search in pred BBs added.
	(distance_agu_use): Search in succ BBs added.
	(IX86_LEA_PRIORITY): Value changed from 2 to 0.
	(LEA_SEARCH_THRESHOLD): Now depends on LEA_MAX_STALL.
	(ix86_lea_for_add_ok): Use ix86_lea_outperforms to make decision.

	* config/i386/i386.md: Splits added to transform lea into a
	sequence of instructions.
Uros Bizjak Sept. 6, 2011, 3:40 p.m. UTC | #2 
On Tue, Sep 6, 2011 at 2:26 PM, Ilya Enkovich <enkovich.gnu@gmail.com> wrote:


> Is fixed version OK?
>
> Thanks,
> Ilya
> ---
> gcc/
>
> 2011-09-06  Enkovich Ilya  <ilya.enkovich@intel.com>
>
>        * config/i386/i386-protos.h (ix86_lea_outperforms): New.
>        (ix86_avoid_lea_for_add): Likewise.
>        (ix86_avoid_lea_for_addr): Likewise.
>        (ix86_split_lea_for_addr): Likewise.
>
>        * config/i386/i386.c (LEA_MAX_STALL): New.
>        (increase_distance): Likewise.
>        (insn_defines_reg): Likewise.
>        (insn_uses_reg_mem): Likewise.
>        (distance_non_agu_define_in_bb): Likewise.
>        (distance_agu_use_in_bb): Likewise.
>        (ix86_lea_outperforms): Likewise.
>        (ix86_ok_to_clobber_flags): Likewise.
>        (ix86_avoid_lea_for_add): Likewise.
>        (ix86_avoid_lea_for_addr): Likewise.
>        (ix86_split_lea_for_addr): Likewise.
>        (distance_non_agu_define): Search in pred BBs added.
>        (distance_agu_use): Search in succ BBs added.
>        (IX86_LEA_PRIORITY): Value changed from 2 to 0.
>        (LEA_SEARCH_THRESHOLD): Now depends on LEA_MAX_STALL.
>        (ix86_lea_for_add_ok): Use ix86_lea_outperforms to make decision.
>
>        * config/i386/i386.md: Splits added to transform lea into a
>        sequence of instructions.

Please merge your new splitters with corresponding LEA patterns.

OK with this change.

Thanks,
Uros.
Ilya Enkovich Sept. 6, 2011, 5:54 p.m. UTC | #3 
2011/9/6 Uros Bizjak <ubizjak@gmail.com>:
>
> Please merge your new splitters with corresponding LEA patterns.
>
> OK with this change.
>
> Thanks,
> Uros.
>

Fixed. Could please someone check it in if it's OK now?

Thanks,
Ilya
---
gcc/

2011-09-06  Enkovich Ilya  <ilya.enkovich@intel.com>

	* config/i386/i386-protos.h (ix86_lea_outperforms): New.
	(ix86_avoid_lea_for_add): Likewise.
	(ix86_avoid_lea_for_addr): Likewise.
	(ix86_split_lea_for_addr): Likewise.

	* config/i386/i386.c (LEA_MAX_STALL): New.
	(increase_distance): Likewise.
	(insn_defines_reg): Likewise.
	(insn_uses_reg_mem): Likewise.
	(distance_non_agu_define_in_bb): Likewise.
	(distance_agu_use_in_bb): Likewise.
	(ix86_lea_outperforms): Likewise.
	(ix86_ok_to_clobber_flags): Likewise.
	(ix86_avoid_lea_for_add): Likewise.
	(ix86_avoid_lea_for_addr): Likewise.
	(ix86_split_lea_for_addr): Likewise.
	(distance_non_agu_define): Search in pred BBs added.
	(distance_agu_use): Search in succ BBs added.
	(IX86_LEA_PRIORITY): Value changed from 2 to 0.
	(LEA_SEARCH_THRESHOLD): Now depends on LEA_MAX_STALL.
	(ix86_lea_for_add_ok): Use ix86_lea_outperforms to make decision.

	* config/i386/i386.md: Split added to transform non destructive
	add into move and add.
	(lea_1): transformed into insn_and_split to avoid AGU stalls.
	(lea<mode>_2): Likewise.
H.J. Lu Sept. 8, 2011, 1:42 p.m. UTC | #4 
On Tue, Sep 6, 2011 at 10:54 AM, Ilya Enkovich <enkovich.gnu@gmail.com> wrote:
> 2011/9/6 Uros Bizjak <ubizjak@gmail.com>:
>>
>> Please merge your new splitters with corresponding LEA patterns.
>>
>> OK with this change.
>>
>> Thanks,
>> Uros.
>>
>
> Fixed. Could please someone check it in if it's OK now?
>
> Thanks,
> Ilya
> ---
> gcc/
>
> 2011-09-06  Enkovich Ilya  <ilya.enkovich@intel.com>
>
>        * config/i386/i386-protos.h (ix86_lea_outperforms): New.
>        (ix86_avoid_lea_for_add): Likewise.
>        (ix86_avoid_lea_for_addr): Likewise.
>        (ix86_split_lea_for_addr): Likewise.
>
>        * config/i386/i386.c (LEA_MAX_STALL): New.
>        (increase_distance): Likewise.
>        (insn_defines_reg): Likewise.
>        (insn_uses_reg_mem): Likewise.
>        (distance_non_agu_define_in_bb): Likewise.
>        (distance_agu_use_in_bb): Likewise.
>        (ix86_lea_outperforms): Likewise.
>        (ix86_ok_to_clobber_flags): Likewise.
>        (ix86_avoid_lea_for_add): Likewise.
>        (ix86_avoid_lea_for_addr): Likewise.
>        (ix86_split_lea_for_addr): Likewise.
>        (distance_non_agu_define): Search in pred BBs added.
>        (distance_agu_use): Search in succ BBs added.
>        (IX86_LEA_PRIORITY): Value changed from 2 to 0.
>        (LEA_SEARCH_THRESHOLD): Now depends on LEA_MAX_STALL.
>        (ix86_lea_for_add_ok): Use ix86_lea_outperforms to make decision.
>
>        * config/i386/i386.md: Split added to transform non destructive
>        add into move and add.
>        (lea_1): transformed into insn_and_split to avoid AGU stalls.
>        (lea<mode>_2): Likewise.
>

I checked it into trunk for you.

Thanks.
diff mbox

Patch

--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -5777,6 +5777,41 @@ 
        (const_string "none")))
   (set_attr "mode" "QI")])

+;; Split non destructive adds if we cannot use lea.
+(define_split
+  [(set (match_operand:SWI48 0 "register_operand" "")
+       (plus:SWI48 (match_operand:SWI48 1 "register_operand" "")
+              (match_operand:SWI48 2 "nonmemory_operand" "")))
+   (clobber (reg:CC FLAGS_REG))]
+  "reload_completed && ix86_avoid_lea_for_add (insn, operands)"
+  [(set (match_dup 0) (match_dup 1))
+   (parallel [(set (match_dup 0) (plus:<MODE> (match_dup 0) (match_dup 2)))
+             (clobber (reg:CC FLAGS_REG))])
+  ]
+)

Put all closing braces on one line:

             (clobber (reg:CC FLAGS_REG))])])

+;; Split lea into one or more ALU instructions if profitable.
+(define_split
+  [(set (match_operand:SI 0 "register_operand" "")
+       (subreg:SI (match_operand:DI 1 "lea_address_operand" "") 0))]
+  "reload_completed && ix86_avoid_lea_for_addr (insn, operands)"
+  [(const_int 0)]
+{
+  ix86_split_lea_for_addr (operands, SImode);
+  DONE;
+})

This is valid only for TARGET_64BIT.

Please note that x32 adds quite some different LEA patterns (see
i386.md, line 5466+). I suggest you merge your splitters with these
define_insn patterns into define_insn_and_split, adding "&&
reload_completed && ix86_avoid_lea_for_addr (insn, operands)" as a
split condition.

Uros.