[LTP] MD-RAID: Use seq_putc() in three status functions?

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>> Calling the function "seq_putc" will be more efficient than "seq_printf"
>> in this case because of the following reasons.
>>
>> 1. How does the distribution look like for supported processor architectures
>>    where the data transfer for bytes (as a function call parameter)
>>    is faster than for (string) pointers?
>>
> How would I know?

How many processor architecture characteristics do you know already?

* Is a string pointer often longer than a byte?

* I imagine that it can become also interesting to check byte level data access
  under constraints of machine word sizes and alignment.


> I would assume that _you_ did some measurements here;

How much would you trust in any concrete numbers I could present
for this use case?

Do you give more trust to a reference testing platform?


> I could easily claim that seq_printf() is more efficient than
> seq_putc(), and won't apply your patch.

This is also possible in principle.


> So _you_ have to prove that your patch is more efficient.

How many results would we like to clarify from various hardware
and software combinations?


>> 2. Did anybody measure already how many the execution times can vary
>>    for these functions?
>>
> Probably not.

Thanks for this information.

How important are the mentioned functions for you within the Linux
programming interface so far?


> Unless _you_ prove that _your_ patch is more efficient it won't get applied.

Which data would you accept as a "prove" in this case?


>>    Where do you get doubts about its efficiency for the data processing
>>    of a single character?
>>
> Because it's being called at the end of a function calling seq_printf() already.

Interesting view …


> So exchanging a single call is probably not helping anything,
> as the compiler will optimize it anyway.

How common is the discussed software transformation between implementations
for optimising compilers?


> Case in point: with your patch the x86_64 compiler generates nearly
> identical code for driver/md/raid1.c, but with one instruction _more_
> after your patch has been applied.

Which software versions and command parameters did you try out
for this information (from an unspecified run time environment)?


> So it's not immediately obvious that your patch is an improvement.

I agree that there are system properties and constraints which can be
considered further.

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Hannes Reinecke]

On 10/17/2016 11:00 AM, SF Markus Elfring wrote:
>>> Calling the function "seq_putc" will be more efficient than "seq_printf"
>>> in this case because of the following reasons.
>>>
>>> 1. How does the distribution look like for supported processor architectures
>>>    where the data transfer for bytes (as a function call parameter)
>>>    is faster than for (string) pointers?
>>>
>> How would I know?
> 
> How many processor architecture characteristics do you know already?
> 
x86, s390x, ppc/ppc64.

> * Is a string pointer often longer than a byte?
> 
Always.

(Which up to now I thought was basic programming knowledge...)

> * I imagine that it can become also interesting to check byte level data access
>   under constraints of machine word sizes and alignment.
> 
So, another test for you to do.

> 
>> I would assume that _you_ did some measurements here;
> 
> How much would you trust in any concrete numbers I could present
> for this use case?
> 
> Do you give more trust to a reference testing platform?
> 
At the moment _any_ test would do.
With every response from your side you just keep on asking further
questions. But so far you haven't delivered any answers nor measurements.

> 
>> I could easily claim that seq_printf() is more efficient than
>> seq_putc(), and won't apply your patch.
> 
> This is also possible in principle.
> 
No, this is what's going to happen if you don't show any measurements.

> 
>> So _you_ have to prove that your patch is more efficient.
> 
> How many results would we like to clarify from various hardware
> and software combinations?
> 
See above. At the moment _any_ test result from your side would do.

> 
>>> 2. Did anybody measure already how many the execution times can vary
>>>    for these functions?
>>>
>> Probably not.
> 
> Thanks for this information.
> 
> How important are the mentioned functions for you within the Linux
> programming interface so far?
> 
Not very. The interface is only used in a slow path, and the execution
time doesn't affect I/O performance in any way.

> 
>> Unless _you_ prove that _your_ patch is more efficient it won't get applied.
> 
> Which data would you accept as a "prove" in this case?
> 
Again: You want something from us. We don't have to prove anything, you
need to convince us. And it is really hard to convince anyone by asking
questions.

> 
>>>    Where do you get doubts about its efficiency for the data processing
>>>    of a single character?
>>>
>> Because it's being called at the end of a function calling seq_printf() already.
> 
> Interesting view …
> 
> 
>> So exchanging a single call is probably not helping anything,
>> as the compiler will optimize it anyway.
> 
> How common is the discussed software transformation between implementations
> for optimising compilers?
> 
> 
>> Case in point: with your patch the x86_64 compiler generates nearly
>> identical code for driver/md/raid1.c, but with one instruction _more_
>> after your patch has been applied.
> 
> Which software versions and command parameters did you try out
> for this information (from an unspecified run time environment)?
> 
> 
# gcc --version
gcc (SUSE Linux) 4.8.5
Copyright (C) 2015 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is
NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR
PURPOSE.

git tree from git.kernel.org/mkp/u/4.10/scsi-queue

_I_ did some measurements.
I'm still waiting from results from your side.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>> * Is a string pointer often longer than a byte?
>>
> Always.

I have got doubts for this specific information.


> (Which up to now I thought was basic programming knowledge...)

By the way:
Run time environments still exist where the size of a pointer can be also
just one byte, don't they?


>> How many results would we like to clarify from various hardware
>> and software combinations?
>>
> See above. At the moment _any_ test result from your side would do.

I imagine that another single result might not be representative.
How many lessons from test statistics will usually be also relevant here?


>> How important are the mentioned functions for you within the Linux
>> programming interface so far?
>>
> Not very. The interface is only used in a slow path, and the execution
> time doesn't affect I/O performance in any way.

Thanks for another interesting information.


>>> Case in point: with your patch the x86_64 compiler generates nearly
>>> identical code for driver/md/raid1.c, but with one instruction _more_
>>> after your patch has been applied.
>>
>> Which software versions and command parameters did you try out
>> for this information (from an unspecified run time environment)?
>>
> # gcc --version
> gcc (SUSE Linux) 4.8.5

Thanks for this detail.

* Did you choose any special optimisation settings for your quick check?

* Will any compilation results matter if "optimisation" would be
  switched off there?


> I'm still waiting from results from your side.

Would any other software developers or testers dare to add related information?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Bernd Petrovitsch]

On Mon, 2016-10-17 at 13:10 +0200, SF Markus Elfring wrote:
[...]
> > (Which up to now I thought was basic programming knowledge...)
> 
> By the way:
> Run time environments still exist where the size of a pointer can
> be also just one byte, don't they?

In the context of the Linux kernel: No.

[ Side note: there might be some DSP out there with a running Linux
kernel which cannot really address a "byte" (meaning 8bits) but only in
register sized quantities (and also aligned for that). But no one cares
here really deeply as that is a so fundamental difference that the C-
compiler must cope with that anyways in the first place. ]

[...]
> > See above. At the moment _any_ test result from your side would do.
> 
> I imagine that another single result might not be representative.

Publish not only results but also everything (complete!) so that anyone
can *easily* follow it to check and reproduce the results - especially
if you want people with knowledge of other architectures to comment
(otherwise they probably won't bother).

Kind regards,
	Bernd
-- 
Bernd Petrovitsch                  Email : bernd@petrovitsch.priv.at
                     LUGA : http://www.luga.at

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>>> See above. At the moment _any_ test result from your side would do.
>>
>> I imagine that another single result might not be representative.
> 
> Publish not only results but also everything (complete!) so that anyone
> can *easily* follow it to check and reproduce the results - especially
> if you want people with knowledge of other architectures to comment
> (otherwise they probably won't bother).

Am I the only software developer so far who would dare to reconsider
implementation details from three status functions?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Hannes Reinecke]

On 10/17/2016 01:43 PM, SF Markus Elfring wrote:
>>>> See above. At the moment _any_ test result from your side would do.
>>>
>>> I imagine that another single result might not be representative.
>>
>> Publish not only results but also everything (complete!) so that anyone
>> can *easily* follow it to check and reproduce the results - especially
>> if you want people with knowledge of other architectures to comment
>> (otherwise they probably won't bother).
> 
> Am I the only software developer so far who would dare to reconsider
> implementation details from three status functions?
> 
No.
But we're waiting for you showing is that it is an improvement.
Which at the moment we don't see.
Hence we're waiting from a proof or validation from your side.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>> Am I the only software developer so far who would dare to reconsider
>> implementation details from three status functions?
>>
> No.

Thanks for this kind of promising feedback.


> But we're waiting for you showing is that it is an improvement.

Can this aspect also be clarified to some degree from a logical point of view?

* Would you really like to know under which circumstances data processing
  will be faster for a single character instead of using a string pointer
  and corresponding two characters?

* Do you care for a changed memory allocation characteristic?

* Will it occasionally be useful to avoid the storage for another string literal?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Hannes Reinecke]

On 10/17/2016 04:30 PM, SF Markus Elfring wrote:
>>> Am I the only software developer so far who would dare to reconsider
>>> implementation details from three status functions?
>>>
>> No.
>
> Thanks for this kind of promising feedback.
>
>
>> But we're waiting for you showing is that it is an improvement.
>
> Can this aspect also be clarified to some degree from a logical point of view?
>
I sincerely doubt that.
We've discussed the logical implications already, and failed to come to 
a consensus. So we need some proof (as in: on this architecture I'm 
seeing this and that performance improvements).
Which you have to deliver.

> * Would you really like to know under which circumstances data processing
>   will be faster for a single character instead of using a string pointer
>   and corresponding two characters?
>
It's not a problem of the interface, it's a problem of the resulting 
code (ie assembler output). We can discuss all we like, if the compiler 
decides to throw in an optimisation none of the arguments even apply.

> * Do you care for a changed memory allocation characteristic?
>
> * Will it occasionally be useful to avoid the storage for another string literal?
>
Occasionally: yes.
In this particular case: hardly.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>> * Would you really like to know under which circumstances data processing
>>   will be faster for a single character instead of using a string pointer
>>   and corresponding two characters?
>>
> It's not a problem of the interface, it's a problem of the resulting code
> (ie assembler output).

How do you think about to discuss concrete generated code any further?


> We can discuss all we like, if the compiler decides to throw in
> an optimisation none of the arguments even apply.

Would it make sense to clarify assembler output with optimisation switched off?

Do you eventually care for code from non-optimising compilers?


>> * Will it occasionally be useful to avoid the storage for another string literal?
>>
> Occasionally: yes.
> In this particular case: hardly.

I am curious when such a software design aspect can become more relevant.
Would it be nice to get rid of three questionable string terminators (null bytes)
for example?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Hannes Reinecke]

On 10/17/2016 06:08 PM, SF Markus Elfring wrote:
>>> * Would you really like to know under which circumstances data processing
>>>   will be faster for a single character instead of using a string pointer
>>>   and corresponding two characters?
>>>
>> It's not a problem of the interface, it's a problem of the resulting code
>> (ie assembler output).
>
> How do you think about to discuss concrete generated code any further?
>
Sure. Show me the generated code and point out where the benefits are.

>> We can discuss all we like, if the compiler decides to throw in
>> an optimisation none of the arguments even apply.
>
> Would it make sense to clarify assembler output with optimisation switched off?
>
> Do you eventually care for code from non-optimising compilers?
>
No. This is the linux kernel. There is a very, _very_ limited benefit of 
trying to use a non-standard compiler.

>
>>> * Will it occasionally be useful to avoid the storage for another string literal?
>>>
>> Occasionally: yes.
>> In this particular case: hardly.
>
> I am curious when such a software design aspect can become more relevant.
> Would it be nice to get rid of three questionable string terminators (null bytes)
> for example?
>
Again, all this does it trying to out-guess what the compiler might be 
doing during compilation. For which the easiest method is checking.
So back to the original task for you: Show me in the generated output 
where the benefits are.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

> So back to the original task for you: Show me in the generated output where the benefits are.

I can offer a bit more information for this software development discussion.


The afffected source files can be compiled for the processor architecture "x86_64"
by a tool like "GCC 6.2.1+r239849-1.3" from the software distribution
"openSUSE Tumbleweed" with the following command examples.

git checkout next-20161014 && my_cc=/usr/bin/gcc-6 && make HOSTCC="${my_cc}" allmodconfig && make -j6 HOSTCC="${my_cc}" drivers/md/
git checkout next_usage_of_seq_putc_in_md_raid_1 && my_cc=/usr/bin/gcc-6 && make HOSTCC="${my_cc}" allmodconfig && make -j6 HOSTCC="${my_cc}" drivers/md/


The tool "objdump" from the software package "binutils 2.27-1.3" can be used
to get corresponding disassemblies for a file like "drivers/md/raid1.obj"
which can then be compared as follows.


--- ../disassembly-md-raid1-next-20161014-1.txt	2016-10-18 18:00:12.341222741 +0200
+++ ../disassembly-md-raid1-seq_putc-1.txt	2016-10-18 18:03:54.135887333 +0200
@@ -3349,7 +3349,7 @@
     37ad:	85 c0                	test   %eax,%eax
     37af:	74 0d                	je     37be <raid1_status+0x9e>
     37b1:	80 3d 00 00 00 00 00 	cmpb   $0x0,0x0(%rip)        # 37b8 <raid1_status+0x98>
-    37b8:	0f 84 1d 01 00 00    	je     38db <raid1_status+0x1bb>
+    37b8:	0f 84 1b 01 00 00    	je     38d9 <raid1_status+0x1b9>
     37be:	4c 89 ff             	mov    %r15,%rdi
     37c1:	31 db                	xor    %ebx,%ebx
     37c3:	e8 00 00 00 00       	callq  37c8 <raid1_status+0xa8>
@@ -3404,42 +3404,43 @@
     3891:	85 c0                	test   %eax,%eax
     3893:	74 09                	je     389e <raid1_status+0x17e>
     3895:	80 3d 00 00 00 00 00 	cmpb   $0x0,0x0(%rip)        # 389c <raid1_status+0x17c>
-    389c:	74 6e                	je     390c <raid1_status+0x1ec>
+    389c:	74 6c                	je     390a <raid1_status+0x1ea>
     389e:	48 c7 c2 00 00 00 00 	mov    $0x0,%rdx
     38a5:	be 01 00 00 00       	mov    $0x1,%esi
     38aa:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
     38b1:	65 ff 0d 00 00 00 00 	decl   %gs:0x0(%rip)        # 38b8 <raid1_status+0x198>
     38b8:	e8 00 00 00 00       	callq  38bd <raid1_status+0x19d>
     38bd:	4c 89 f7             	mov    %r14,%rdi
-    38c0:	48 c7 c6 00 00 00 00 	mov    $0x0,%rsi
-    38c7:	e8 00 00 00 00       	callq  38cc <raid1_status+0x1ac>
-    38cc:	48 83 c4 10          	add    $0x10,%rsp
-    38d0:	5b                   	pop    %rbx
-    38d1:	41 5c                	pop    %r12
-    38d3:	41 5d                	pop    %r13
-    38d5:	41 5e                	pop    %r14
-    38d7:	41 5f                	pop    %r15
-    38d9:	5d                   	pop    %rbp
-    38da:	c3                   	retq   
-    38db:	e8 00 00 00 00       	callq  38e0 <raid1_status+0x1c0>
-    38e0:	84 c0                	test   %al,%al
-    38e2:	0f 85 d6 fe ff ff    	jne    37be <raid1_status+0x9e>
-    38e8:	48 c7 c2 00 00 00 00 	mov    $0x0,%rdx
-    38ef:	be 69 03 00 00       	mov    $0x369,%esi
-    38f4:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
-    38fb:	c6 05 00 00 00 00 01 	movb   $0x1,0x0(%rip)        # 3902 <raid1_status+0x1e2>
-    3902:	e8 00 00 00 00       	callq  3907 <raid1_status+0x1e7>
-    3907:	e9 b2 fe ff ff       	jmpq   37be <raid1_status+0x9e>
-    390c:	e8 00 00 00 00       	callq  3911 <raid1_status+0x1f1>
-    3911:	84 c0                	test   %al,%al
-    3913:	75 89                	jne    389e <raid1_status+0x17e>
-    3915:	48 c7 c2 00 00 00 00 	mov    $0x0,%rdx
-    391c:	be 9c 03 00 00       	mov    $0x39c,%esi
-    3921:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
-    3928:	c6 05 00 00 00 00 01 	movb   $0x1,0x0(%rip)        # 392f <raid1_status+0x20f>
-    392f:	e8 00 00 00 00       	callq  3934 <raid1_status+0x214>
-    3934:	e9 65 ff ff ff       	jmpq   389e <raid1_status+0x17e>
-    3939:	0f 1f 80 00 00 00 00 	nopl   0x0(%rax)
+    38c0:	be 5d 00 00 00       	mov    $0x5d,%esi
+    38c5:	e8 00 00 00 00       	callq  38ca <raid1_status+0x1aa>
+    38ca:	48 83 c4 10          	add    $0x10,%rsp
+    38ce:	5b                   	pop    %rbx
+    38cf:	41 5c                	pop    %r12
+    38d1:	41 5d                	pop    %r13
+    38d3:	41 5e                	pop    %r14
+    38d5:	41 5f                	pop    %r15
+    38d7:	5d                   	pop    %rbp
+    38d8:	c3                   	retq   
+    38d9:	e8 00 00 00 00       	callq  38de <raid1_status+0x1be>
+    38de:	84 c0                	test   %al,%al
+    38e0:	0f 85 d8 fe ff ff    	jne    37be <raid1_status+0x9e>
+    38e6:	48 c7 c2 00 00 00 00 	mov    $0x0,%rdx
+    38ed:	be 69 03 00 00       	mov    $0x369,%esi
+    38f2:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
+    38f9:	c6 05 00 00 00 00 01 	movb   $0x1,0x0(%rip)        # 3900 <raid1_status+0x1e0>
+    3900:	e8 00 00 00 00       	callq  3905 <raid1_status+0x1e5>
+    3905:	e9 b4 fe ff ff       	jmpq   37be <raid1_status+0x9e>
+    390a:	e8 00 00 00 00       	callq  390f <raid1_status+0x1ef>
+    390f:	84 c0                	test   %al,%al
+    3911:	75 8b                	jne    389e <raid1_status+0x17e>
+    3913:	48 c7 c2 00 00 00 00 	mov    $0x0,%rdx
+    391a:	be 9c 03 00 00       	mov    $0x39c,%esi
+    391f:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
+    3926:	c6 05 00 00 00 00 01 	movb   $0x1,0x0(%rip)        # 392d <raid1_status+0x20d>
+    392d:	e8 00 00 00 00       	callq  3932 <raid1_status+0x212>
+    3932:	e9 67 ff ff ff       	jmpq   389e <raid1_status+0x17e>
+    3937:	66 0f 1f 84 00 00 00 	nopw   0x0(%rax,%rax,1)
+    393e:	00 00 
 
 0000000000003940 <print_conf>:
     3940:	e8 00 00 00 00       	callq  3945 <print_conf+0x5>
@@ -11134,7 +11135,7 @@
 
 0000000000000000 <_GLOBAL__sub_D_65535_0_raid1.c>:
    0:	55                   	push   %rbp
-   1:	be 35 00 00 00       	mov    $0x35,%esi
+   1:	be 34 00 00 00       	mov    $0x34,%esi
    6:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
    d:	48 89 e5             	mov    %rsp,%rbp
   10:	e8 00 00 00 00       	callq  15 <_GLOBAL__sub_D_65535_0_raid1.c+0x15>
@@ -11145,7 +11146,7 @@
 
 0000000000000000 <_GLOBAL__sub_I_65535_1_raid1.c>:
    0:	55                   	push   %rbp
-   1:	be 35 00 00 00       	mov    $0x35,%esi
+   1:	be 34 00 00 00       	mov    $0x34,%esi
    6:	48 c7 c7 00 00 00 00 	mov    $0x0,%rdi
    d:	48 89 e5             	mov    %rsp,%rbp
   10:	e8 00 00 00 00       	callq  15 <_GLOBAL__sub_I_65535_1_raid1.c+0x15>


Does this kind of data display contain differences which are worth for further considerations?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>> So back to the original task for you: Show me in the generated output where the benefits are.

I can offer another bit of information for this software development discussion.

The following build settings were active in my "Makefile" for this Linux test case.

…
HOSTCFLAGS   = -Wall -Wmissing-prototypes -Wstrict-prototypes -O0 -fomit-frame-pointer -std=gnu89
…


The afffected source files can be compiled for the processor architecture "x86_64"
by a tool like "GCC 6.2.1+r239849-1.4" from the software distribution
"openSUSE Tumbleweed" with the following command example.

my_original=${my_build_dir}unchanged/test/ \
&& my_fixing=${my_build_dir}patched/test/ \
&& mkdir -p ${my_original} ${my_fixing} \
&& my_cc=/usr/bin/gcc-6 \
&& my_module=drivers/md/raid1.s \
&& git checkout next-20161014 \
&& make -j6 O="${my_original}" HOSTCC="${my_cc}" allmodconfig ${my_module} \
&& git checkout next_usage_of_seq_putc_in_md_raid_1 \
&& make -j6 O="${my_fixing}" HOSTCC="${my_cc}" allmodconfig ${my_module} \
&& diff -u "${my_original}${my_module}" "${my_fixing}${my_module}" > "${my_build_dir}assembler_code_comparison_$(date -I)_1.diff"


Unfortunately, the generated file got the size "311 KiB". I guess that
this is too big to send such a file around on the Linux mailing list.

Is this kind of assembler code comparison still useful to clarify relevant
differences further?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[SF Markus Elfring]

>>> So back to the original task for you: Show me in the generated output where the benefits are.

I can offer another bit of information for this software development discussion.

The following build settings were active in my "Makefile" for this Linux test case.

…
HOSTCFLAGS   = -Wall -Wmissing-prototypes -Wstrict-prototypes -O0 -fomit-frame-pointer -std=gnu89
…


The afffected source files can be compiled for the processor architecture "x86_64"
by a tool like "GCC 6.2.1+r239849-1.5" from the software distribution
"openSUSE Tumbleweed" with the following command example.

my_original=${my_build_dir}unchanged/test/ \
&& my_fixing=${my_build_dir}patched/test/ \
&& mkdir -p ${my_original} ${my_fixing} \
&& my_cc=/usr/bin/gcc-6 \
&& my_module=drivers/md/raid1.s \
&& git checkout next-20161014 \
&& make -j6 O="${my_original}" CC="${my_cc}" HOSTCC="${my_cc}" allmodconfig "${my_module}" \
&& git checkout next_usage_of_seq_putc_in_md_raid_1 \
&& make -j6 O="${my_fixing}" CC="${my_cc}" HOSTCC="${my_cc}" allmodconfig "${my_module}" \
&& diff -u "${my_original}${my_module}" "${my_fixing}${my_module}" > "${my_build_dir}assembler_code_comparison_$(date -I)_3.diff"


The generated file got the size "25.4 KiB" this time. I guess that only
the following two diff hunks are interesting then to show desired effects
for the suggested software refactoring around data output of a single character
(instead of a similar string).


…
@@ -4402,10 +4402,6 @@
 .LC19:
 	.string	"%s"
 	.zero	61
-	.align 32
-.LC20:
-	.string	"]"
-	.zero	62
 	.text
 	.p2align 4,,15
 	.type	raid1_status, @function
@@ -4564,8 +4560,8 @@
 	movq	$rcu_lock_map, %rdi	#,
 	call	lock_release	#
 	movq	%r14, %rdi	# seq,
-	movq	$.LC20, %rsi	#,
-	call	seq_printf	#
+	movl	$93, %esi	#,
+	call	seq_putc	#
 	addq	$16, %rsp	#,
 	popq	%rbx	#
 	popq	%r12	#
…


* Is this kind of assembler code comparison useful to clarify relevant
  differences further?

* Are any software development concerns left over for such a transformation?

Regards,
Markus

[LTP] MD-RAID: Use seq_putc() in three status functions?

[Hannes Reinecke]

On 10/17/2016 01:10 PM, SF Markus Elfring wrote:
>>> * Is a string pointer often longer than a byte?
>>>
>> Always.
> 
> I have got doubts for this specific information.
> 
> 
>> (Which up to now I thought was basic programming knowledge...)
> 
> By the way:
> Run time environments still exist where the size of a pointer can be also
> just one byte, don't they?
> 
Really? Name one.
You can only fit a point in one byte if you are on an 8-bit system.
Which I don't think linux is running on.

>>> How many results would we like to clarify from various hardware
>>> and software combinations?
>>>
>> See above. At the moment _any_ test result from your side would do.
> 
> I imagine that another single result might not be representative.
> How many lessons from test statistics will usually be also relevant here?
> 
> 
As said above, _any_ statistic will do at this point.

>>> How important are the mentioned functions for you within the Linux
>>> programming interface so far?
>>>
>> Not very. The interface is only used in a slow path, and the execution
>> time doesn't affect I/O performance in any way.
> 
> Thanks for another interesting information.
> 
> 
>>>> Case in point: with your patch the x86_64 compiler generates nearly
>>>> identical code for driver/md/raid1.c, but with one instruction _more_
>>>> after your patch has been applied.
>>>
>>> Which software versions and command parameters did you try out
>>> for this information (from an unspecified run time environment)?
>>>
>> # gcc --version
>> gcc (SUSE Linux) 4.8.5
> 
> Thanks for this detail.
> 
> * Did you choose any special optimisation settings for your quick check?
> 
> * Will any compilation results matter if "optimisation" would be
>   switched off there?
> 
> 
These were the results when calling 'make' in the kernel source tree.
With all applicable options.

>> I'm still waiting from results from your side.
> 
> Would any other software developers or testers dare to add related information?
> 
No. It's your patch, _you_ have to do it.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		   Teamlead Storage & Networking
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)