Re: [PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
On Mon, Oct 19, 2020 at 10:54:40AM +0200, Christophe Leroy wrote: > Le 19/10/2020 à 10:32, Segher Boessenkool a écrit : > >The kernel should just use __always_inline if that is what it *wants*; > >that is true here most likely. GCC could perhaps improve its heuristics > >so that it no longer thinks these functions are often too big for > >inlining (they *are* pretty big, but not after basic optimisations with > >constant integer arguments). > > Yes I guess __always_inline is to be added on functions like this defined > in headers for exactly that, and that's the purpose of this patch. > > However I find it odd that get_order() is outlined by GCC even in some > object files that don't use it at all, for instance in fs/pipe.o It is (arguably) too big too always inline if you do not consider that __builtin_constant_p will remove half of the function one way or another. Not sure if that is what happens here, but now we have a PR (thanks!) and we will find out. Segher
Re: [PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
Le 19/10/2020 à 10:32, Segher Boessenkool a écrit :
On Mon, Oct 19, 2020 at 07:50:41AM +0200, Christophe Leroy wrote:
Le 19/10/2020 à 06:55, Joel Stanley a écrit :
In the old days, marking a function 'static inline' was forcing
GCC to inline, but since commit ac7c3e4ff401 ("compiler: enable
CONFIG_OPTIMIZE_INLINING forcibly") GCC may decide to not inline
a function.
It looks like GCC 10 is taking poor decisions on this.
1952 bytes smaller with your patch applied. Did you raise this with
anyone from GCC?
Yes I did, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97445
For the time being, it's at a standstill.
The kernel should just use __always_inline if that is what it *wants*;
that is true here most likely. GCC could perhaps improve its heuristics
so that it no longer thinks these functions are often too big for
inlining (they *are* pretty big, but not after basic optimisations with
constant integer arguments).
Yes I guess __always_inline is to be added on functions like this defined in headers for exactly
that, and that's the purpose of this patch.
However I find it odd that get_order() is outlined by GCC even in some object files that don't use
it at all, for instance in fs/pipe.o
Christophe
Re: [PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
On Mon, Oct 19, 2020 at 07:50:41AM +0200, Christophe Leroy wrote:
> Le 19/10/2020 à 06:55, Joel Stanley a écrit :
> >>In the old days, marking a function 'static inline' was forcing
> >>GCC to inline, but since commit ac7c3e4ff401 ("compiler: enable
> >>CONFIG_OPTIMIZE_INLINING forcibly") GCC may decide to not inline
> >>a function.
> >>
> >>It looks like GCC 10 is taking poor decisions on this.
> >1952 bytes smaller with your patch applied. Did you raise this with
> >anyone from GCC?
>
> Yes I did, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97445
>
> For the time being, it's at a standstill.
The kernel should just use __always_inline if that is what it *wants*;
that is true here most likely. GCC could perhaps improve its heuristics
so that it no longer thinks these functions are often too big for
inlining (they *are* pretty big, but not after basic optimisations with
constant integer arguments).
Segher
Re: [PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
Le 19/10/2020 à 06:55, Joel Stanley a écrit :
On Sat, 17 Oct 2020 at 15:55, Christophe Leroy
wrote:
When building mpc885_ads_defconfig with gcc 10.1,
the function get_order() appears 50 times in vmlinux:
[linux]# ppc-linux-objdump -x vmlinux | grep get_order | wc -l
50
[linux]# size vmlinux
textdata bss dec hex filename
3842620 675624 135160 4653404 47015c vmlinux
In the old days, marking a function 'static inline' was forcing
GCC to inline, but since commit ac7c3e4ff401 ("compiler: enable
CONFIG_OPTIMIZE_INLINING forcibly") GCC may decide to not inline
a function.
It looks like GCC 10 is taking poor decisions on this.
get_order() compiles into the following tiny function,
occupying 20 bytes of text.
007c :
7c: 38 63 ff ff addir3,r3,-1
80: 54 63 a3 3e rlwinm r3,r3,20,12,31
84: 7c 63 00 34 cntlzw r3,r3
88: 20 63 00 20 subfic r3,r3,32
8c: 4e 80 00 20 blr
By forcing get_order() to be __always_inline, the size of text is
reduced by 1940 bytes, that is almost twice the space occupied by
50 times get_order()
[linux-powerpc]# size vmlinux
textdata bss dec hex filename
3840680 675588 135176 4651444 46f9b4 vmlinux
I see similar results with GCC 10.2 building for arm32. There are 143
instances of get_order with aspeed_g5_defconfig.
Before:
9071838 2630138 186468 11888444 b5673c vmlinux
After:
9069886 2630126 186468 11886480 b55f90 vmlinux
1952 bytes smaller with your patch applied. Did you raise this with
anyone from GCC?
Yes I did, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97445
For the time being, it's at a standstill.
Christophe
Reviewed-by: Joel Stanley
Signed-off-by: Christophe Leroy
---
include/asm-generic/getorder.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/asm-generic/getorder.h b/include/asm-generic/getorder.h
index e9f20b813a69..f2979e3a96b6 100644
--- a/include/asm-generic/getorder.h
+++ b/include/asm-generic/getorder.h
@@ -26,7 +26,7 @@
*
* The result is undefined if the size is 0.
*/
-static inline __attribute_const__ int get_order(unsigned long size)
+static __always_inline __attribute_const__ int get_order(unsigned long size)
{
if (__builtin_constant_p(size)) {
if (!size)
--
2.25.0
Re: [PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
On Sat, 17 Oct 2020 at 15:55, Christophe Leroy
wrote:
>
> When building mpc885_ads_defconfig with gcc 10.1,
> the function get_order() appears 50 times in vmlinux:
>
> [linux]# ppc-linux-objdump -x vmlinux | grep get_order | wc -l
> 50
>
> [linux]# size vmlinux
>textdata bss dec hex filename
> 3842620 675624 135160 4653404 47015c vmlinux
>
> In the old days, marking a function 'static inline' was forcing
> GCC to inline, but since commit ac7c3e4ff401 ("compiler: enable
> CONFIG_OPTIMIZE_INLINING forcibly") GCC may decide to not inline
> a function.
>
> It looks like GCC 10 is taking poor decisions on this.
>
> get_order() compiles into the following tiny function,
> occupying 20 bytes of text.
>
> 007c :
> 7c: 38 63 ff ff addir3,r3,-1
> 80: 54 63 a3 3e rlwinm r3,r3,20,12,31
> 84: 7c 63 00 34 cntlzw r3,r3
> 88: 20 63 00 20 subfic r3,r3,32
> 8c: 4e 80 00 20 blr
>
> By forcing get_order() to be __always_inline, the size of text is
> reduced by 1940 bytes, that is almost twice the space occupied by
> 50 times get_order()
>
> [linux-powerpc]# size vmlinux
>textdata bss dec hex filename
> 3840680 675588 135176 4651444 46f9b4 vmlinux
I see similar results with GCC 10.2 building for arm32. There are 143
instances of get_order with aspeed_g5_defconfig.
Before:
9071838 2630138 186468 11888444 b5673c vmlinux
After:
9069886 2630126 186468 11886480 b55f90 vmlinux
1952 bytes smaller with your patch applied. Did you raise this with
anyone from GCC?
Reviewed-by: Joel Stanley
> Signed-off-by: Christophe Leroy
> ---
> include/asm-generic/getorder.h | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/include/asm-generic/getorder.h b/include/asm-generic/getorder.h
> index e9f20b813a69..f2979e3a96b6 100644
> --- a/include/asm-generic/getorder.h
> +++ b/include/asm-generic/getorder.h
> @@ -26,7 +26,7 @@
> *
> * The result is undefined if the size is 0.
> */
> -static inline __attribute_const__ int get_order(unsigned long size)
> +static __always_inline __attribute_const__ int get_order(unsigned long size)
> {
> if (__builtin_constant_p(size)) {
> if (!size)
> --
> 2.25.0
>
[PATCH] asm-generic: Force inlining of get_order() to work around gcc10 poor decision
When building mpc885_ads_defconfig with gcc 10.1,
the function get_order() appears 50 times in vmlinux:
[linux]# ppc-linux-objdump -x vmlinux | grep get_order | wc -l
50
[linux]# size vmlinux
textdata bss dec hex filename
3842620 675624 135160 4653404 47015c vmlinux
In the old days, marking a function 'static inline' was forcing
GCC to inline, but since commit ac7c3e4ff401 ("compiler: enable
CONFIG_OPTIMIZE_INLINING forcibly") GCC may decide to not inline
a function.
It looks like GCC 10 is taking poor decisions on this.
get_order() compiles into the following tiny function,
occupying 20 bytes of text.
007c :
7c: 38 63 ff ff addir3,r3,-1
80: 54 63 a3 3e rlwinm r3,r3,20,12,31
84: 7c 63 00 34 cntlzw r3,r3
88: 20 63 00 20 subfic r3,r3,32
8c: 4e 80 00 20 blr
By forcing get_order() to be __always_inline, the size of text is
reduced by 1940 bytes, that is almost twice the space occupied by
50 times get_order()
[linux-powerpc]# size vmlinux
textdata bss dec hex filename
3840680 675588 135176 4651444 46f9b4 vmlinux
Signed-off-by: Christophe Leroy
---
include/asm-generic/getorder.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/asm-generic/getorder.h b/include/asm-generic/getorder.h
index e9f20b813a69..f2979e3a96b6 100644
--- a/include/asm-generic/getorder.h
+++ b/include/asm-generic/getorder.h
@@ -26,7 +26,7 @@
*
* The result is undefined if the size is 0.
*/
-static inline __attribute_const__ int get_order(unsigned long size)
+static __always_inline __attribute_const__ int get_order(unsigned long size)
{
if (__builtin_constant_p(size)) {
if (!size)
--
2.25.0
