Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1637966 > unrolled thread

Re: [PATCH v6 2/7] perf/x86/intel: Record branch type

Started byJiri Olsa <jolsa@redhat.com>
First post2017-05-09 10:30 +0200
Last post2017-05-10 02:20 +0200
Articles 4 — 2 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v6 2/7] perf/x86/intel: Record branch type Jiri Olsa <jolsa@redhat.com> - 2017-05-09 10:30 +0200
    Re: [PATCH v6 2/7] perf/x86/intel: Record branch type "Jin, Yao" <yao.jin@linux.intel.com> - 2017-05-09 14:00 +0200
      Re: [PATCH v6 2/7] perf/x86/intel: Record branch type Jiri Olsa <jolsa@redhat.com> - 2017-05-09 14:40 +0200
        Re: [PATCH v6 2/7] perf/x86/intel: Record branch type "Jin, Yao" <yao.jin@linux.intel.com> - 2017-05-10 02:20 +0200

#1637966 — Re: [PATCH v6 2/7] perf/x86/intel: Record branch type

FromJiri Olsa <jolsa@redhat.com>
Date2017-05-09 10:30 +0200
SubjectRe: [PATCH v6 2/7] perf/x86/intel: Record branch type
Message-ID<tF8GR-7BJ-1@gated-at.bofh.it>
On Mon, Apr 24, 2017 at 08:47:14AM +0800, Jin, Yao wrote:
> 
> 
> On 4/23/2017 9:55 PM, Jiri Olsa wrote:
> > On Thu, Apr 20, 2017 at 08:07:50PM +0800, Jin Yao wrote:
> > 
> > SNIP
> > 
> > > +#define X86_BR_TYPE_MAP_MAX	16
> > > +
> > > +static int
> > > +common_branch_type(int type)
> > > +{
> > > +	int i, mask;
> > > +	const int branch_map[X86_BR_TYPE_MAP_MAX] = {
> > > +		PERF_BR_CALL,		/* X86_BR_CALL */
> > > +		PERF_BR_RET,		/* X86_BR_RET */
> > > +		PERF_BR_SYSCALL,	/* X86_BR_SYSCALL */
> > > +		PERF_BR_SYSRET,		/* X86_BR_SYSRET */
> > > +		PERF_BR_INT,		/* X86_BR_INT */
> > > +		PERF_BR_IRET,		/* X86_BR_IRET */
> > > +		PERF_BR_JCC,		/* X86_BR_JCC */
> > > +		PERF_BR_JMP,		/* X86_BR_JMP */
> > > +		PERF_BR_IRQ,		/* X86_BR_IRQ */
> > > +		PERF_BR_IND_CALL,	/* X86_BR_IND_CALL */
> > > +		PERF_BR_NONE,		/* X86_BR_ABORT */
> > > +		PERF_BR_NONE,		/* X86_BR_IN_TX */
> > > +		PERF_BR_NONE,		/* X86_BR_NO_TX */
> > > +		PERF_BR_CALL,		/* X86_BR_ZERO_CALL */
> > > +		PERF_BR_NONE,		/* X86_BR_CALL_STACK */
> > > +		PERF_BR_IND_JMP,	/* X86_BR_IND_JMP */
> > > +	};
> > > +
> > > +	type >>= 2; /* skip X86_BR_USER and X86_BR_KERNEL */
> > > +	mask = ~(~0 << 1);
> > is that a fancy way to get 1 into the mask? what do I miss?

you did not comment on this one

> > 
> > > +
> > > +	for (i = 0; i < X86_BR_TYPE_MAP_MAX; i++) {
> > > +		if (type & mask)
> > > +			return branch_map[i];
> > I wonder some bit search would be faster in here, but maybe not big deal
> > 
> > jirka
> 
> I just think the branch_map[] doesn't contain many entries (16 entries
> here), so maybe checking 1 bit one time should be acceptable. I just want to
> keep the code simple.
> 
> But if the number of entries is more (e.g. 64), maybe it'd better check 2 or
> 4 bits one time.

ook

jirka

[toc] | [next] | [standalone]

#1638066

From"Jin, Yao" <yao.jin@linux.intel.com>
Date2017-05-09 14:00 +0200
Message-ID<tFbY5-1dU-17@gated-at.bofh.it>
In reply to#1637966

On 5/9/2017 4:26 PM, Jiri Olsa wrote:
> On Mon, Apr 24, 2017 at 08:47:14AM +0800, Jin, Yao wrote:
>>
>> On 4/23/2017 9:55 PM, Jiri Olsa wrote:
>>> On Thu, Apr 20, 2017 at 08:07:50PM +0800, Jin Yao wrote:
>>>
>>> SNIP
>>>
>>>> +#define X86_BR_TYPE_MAP_MAX	16
>>>> +
>>>> +static int
>>>> +common_branch_type(int type)
>>>> +{
>>>> +	int i, mask;
>>>> +	const int branch_map[X86_BR_TYPE_MAP_MAX] = {
>>>> +		PERF_BR_CALL,		/* X86_BR_CALL */
>>>> +		PERF_BR_RET,		/* X86_BR_RET */
>>>> +		PERF_BR_SYSCALL,	/* X86_BR_SYSCALL */
>>>> +		PERF_BR_SYSRET,		/* X86_BR_SYSRET */
>>>> +		PERF_BR_INT,		/* X86_BR_INT */
>>>> +		PERF_BR_IRET,		/* X86_BR_IRET */
>>>> +		PERF_BR_JCC,		/* X86_BR_JCC */
>>>> +		PERF_BR_JMP,		/* X86_BR_JMP */
>>>> +		PERF_BR_IRQ,		/* X86_BR_IRQ */
>>>> +		PERF_BR_IND_CALL,	/* X86_BR_IND_CALL */
>>>> +		PERF_BR_NONE,		/* X86_BR_ABORT */
>>>> +		PERF_BR_NONE,		/* X86_BR_IN_TX */
>>>> +		PERF_BR_NONE,		/* X86_BR_NO_TX */
>>>> +		PERF_BR_CALL,		/* X86_BR_ZERO_CALL */
>>>> +		PERF_BR_NONE,		/* X86_BR_CALL_STACK */
>>>> +		PERF_BR_IND_JMP,	/* X86_BR_IND_JMP */
>>>> +	};
>>>> +
>>>> +	type >>= 2; /* skip X86_BR_USER and X86_BR_KERNEL */
>>>> +	mask = ~(~0 << 1);
>>> is that a fancy way to get 1 into the mask? what do I miss?
> you did not comment on this one

Sorry, I misunderstood that this comment and the next comment had the 
same meaning.

In the previous version, I used the switch/case to convert from X86_BR 
to PERF_BR. I got a comment from community that it'd better use a lookup 
table for conversion.

Since each bit in type represents a X86_BR type so I use a mask (0x1) to 
filter the bit. Yes, it looks I can also directly set 0x1 to mask.

I write the code "mask = ~(~0 << 1)" according to my coding habits. If 
you think I should change the code to "mask = 0x1", that's OK  :)

>>>> +
>>>> +	for (i = 0; i < X86_BR_TYPE_MAP_MAX; i++) {
>>>> +		if (type & mask)
>>>> +			return branch_map[i];
>>> I wonder some bit search would be faster in here, but maybe not big deal
>>>
>>> jirka
>> I just think the branch_map[] doesn't contain many entries (16 entries
>> here), so maybe checking 1 bit one time should be acceptable. I just want to
>> keep the code simple.
>>
>> But if the number of entries is more (e.g. 64), maybe it'd better check 2 or
>> 4 bits one time.
> ook
>
> jirka
Sorry, what's the meaning of ook? Does it mean "OK"?

Thanks
Jin Yao

[toc] | [prev] | [next] | [standalone]

#1638093

FromJiri Olsa <jolsa@redhat.com>
Date2017-05-09 14:40 +0200
Message-ID<tFcAO-1J6-17@gated-at.bofh.it>
In reply to#1638066
On Tue, May 09, 2017 at 07:57:11PM +0800, Jin, Yao wrote:

SNIP

> > > > > +
> > > > > +	type >>= 2; /* skip X86_BR_USER and X86_BR_KERNEL */
> > > > > +	mask = ~(~0 << 1);
> > > > is that a fancy way to get 1 into the mask? what do I miss?
> > you did not comment on this one
> 
> Sorry, I misunderstood that this comment and the next comment had the same
> meaning.
> 
> In the previous version, I used the switch/case to convert from X86_BR to
> PERF_BR. I got a comment from community that it'd better use a lookup table
> for conversion.
> 
> Since each bit in type represents a X86_BR type so I use a mask (0x1) to
> filter the bit. Yes, it looks I can also directly set 0x1 to mask.
> 
> I write the code "mask = ~(~0 << 1)" according to my coding habits. If you
> think I should change the code to "mask = 0x1", that's OK  :)

im ok with that.. was just wondering for the reason
I guess compiler will make it single constant assignment anyway

> 
> > > > > +
> > > > > +	for (i = 0; i < X86_BR_TYPE_MAP_MAX; i++) {
> > > > > +		if (type & mask)
> > > > > +			return branch_map[i];
> > > > I wonder some bit search would be faster in here, but maybe not big deal
> > > > 
> > > > jirka
> > > I just think the branch_map[] doesn't contain many entries (16 entries
> > > here), so maybe checking 1 bit one time should be acceptable. I just want to
> > > keep the code simple.
> > > 
> > > But if the number of entries is more (e.g. 64), maybe it'd better check 2 or
> > > 4 bits one time.
> > ook
> > 
> > jirka
> Sorry, what's the meaning of ook? Does it mean "OK"?

just means ok ;-)

thanks,
jirka

[toc] | [prev] | [next] | [standalone]

#1638463

From"Jin, Yao" <yao.jin@linux.intel.com>
Date2017-05-10 02:20 +0200
Message-ID<tFnwd-Jl-5@gated-at.bofh.it>
In reply to#1638093

On 5/9/2017 8:39 PM, Jiri Olsa wrote:
> On Tue, May 09, 2017 at 07:57:11PM +0800, Jin, Yao wrote:
>
> SNIP
>
>>>>>> +
>>>>>> +	type >>= 2; /* skip X86_BR_USER and X86_BR_KERNEL */
>>>>>> +	mask = ~(~0 << 1);
>>>>> is that a fancy way to get 1 into the mask? what do I miss?
>>> you did not comment on this one
>> Sorry, I misunderstood that this comment and the next comment had the same
>> meaning.
>>
>> In the previous version, I used the switch/case to convert from X86_BR to
>> PERF_BR. I got a comment from community that it'd better use a lookup table
>> for conversion.
>>
>> Since each bit in type represents a X86_BR type so I use a mask (0x1) to
>> filter the bit. Yes, it looks I can also directly set 0x1 to mask.
>>
>> I write the code "mask = ~(~0 << 1)" according to my coding habits. If you
>> think I should change the code to "mask = 0x1", that's OK  :)
> im ok with that.. was just wondering for the reason
> I guess compiler will make it single constant assignment anyway

I think so.  The compiler should be clever enough for this optimization.
>>>>>> +
>>>>>> +	for (i = 0; i < X86_BR_TYPE_MAP_MAX; i++) {
>>>>>> +		if (type & mask)
>>>>>> +			return branch_map[i];
>>>>> I wonder some bit search would be faster in here, but maybe not big deal
>>>>>
>>>>> jirka
>>>> I just think the branch_map[] doesn't contain many entries (16 entries
>>>> here), so maybe checking 1 bit one time should be acceptable. I just want to
>>>> keep the code simple.
>>>>
>>>> But if the number of entries is more (e.g. 64), maybe it'd better check 2 or
>>>> 4 bits one time.
>>> ook
>>>
>>> jirka
>> Sorry, what's the meaning of ook? Does it mean "OK"?
> just means ok ;-)
>
> thanks,
> jirka

Thanks so much!

Jin Yao

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web