[PATCH 1/2] sched/deadline: Add cpudl_maximum_dl()

[PATCH 1/2] sched/deadline: Add cpudl_maximum_dl()

[Byungchul Park]

Current code uses cpudl_maximum() to get the root node's cpu, while it
directly accesses the root node using 'cp->elements[0].dl' to get the
root node's dl. It would be better and more readible if a function to
get the dl is added. So add it.

Signed-off-by: Byungchul Park <byungchul.park@lge.com>
---
 kernel/sched/cpudeadline.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index fba235c..d4a6963 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -108,11 +108,16 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
 		cpudl_heapify_down(cp, idx);
 }
 
-static inline int cpudl_maximum(struct cpudl *cp)
+static inline int cpudl_maximum_cpu(struct cpudl *cp)
 {
 	return cp->elements[0].cpu;
 }
 
+static inline u64 cpudl_maximum_dl(struct cpudl *cp)
+{
+	return cp->elements[0].dl;
+}
+
 /*
  * cpudl_find - find the best (later-dl) CPU in the system
  * @cp: the cpudl max-heap context
@@ -131,9 +136,9 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
 	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
 		best_cpu = cpumask_any(later_mask);
 		goto out;
-	} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
-			dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
-		best_cpu = cpudl_maximum(cp);
+	} else if (cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
+			dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
+		best_cpu = cpudl_maximum_cpu(cp);
 		if (later_mask)
 			cpumask_set_cpu(best_cpu, later_mask);
 	}
-- 
1.9.1

[PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

When the heap tree is empty, cp->elements[0].cpu has meaningless value.
We need to consider the case.

Signed-off-by: Byungchul Park <byungchul.park@lge.com>
---
 kernel/sched/cpudeadline.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index d4a6963..9b314a9 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -110,7 +110,8 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
 
 static inline int cpudl_maximum_cpu(struct cpudl *cp)
 {
-	return cp->elements[0].cpu;
+	int cpu = cp->elements[0].cpu;
+	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
 }
 
 static inline u64 cpudl_maximum_dl(struct cpudl *cp)
-- 
1.9.1

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Juri Lelli]

Hi,

On 02/06/17 16:31, Byungchul Park wrote:
> When the heap tree is empty, cp->elements[0].cpu has meaningless value.
> We need to consider the case.
> 
> Signed-off-by: Byungchul Park <byungchul.park@lge.com>
> ---
>  kernel/sched/cpudeadline.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index d4a6963..9b314a9 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -110,7 +110,8 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
>  
>  static inline int cpudl_maximum_cpu(struct cpudl *cp)
>  {
> -	return cp->elements[0].cpu;
> +	int cpu = cp->elements[0].cpu;
> +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;

Mmm, don't we get a WARN from cpumask_check() if we return -1 here?

Thanks,

- Juri

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> Hi,
> 
> On 02/06/17 16:31, Byungchul Park wrote:
> > When the heap tree is empty, cp->elements[0].cpu has meaningless value.

Hi,

The meaningless value is 0.

> > We need to consider the case.
> > 
> > Signed-off-by: Byungchul Park <byungchul.park@lge.com>
> > ---
> >  kernel/sched/cpudeadline.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> > index d4a6963..9b314a9 100644
> > --- a/kernel/sched/cpudeadline.c
> > +++ b/kernel/sched/cpudeadline.c
> > @@ -110,7 +110,8 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
> >  
> >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> >  {
> > -	return cp->elements[0].cpu;
> > +	int cpu = cp->elements[0].cpu;
> > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> 
> Mmm, don't we get a WARN from cpumask_check() if we return -1 here?

The function does not return -1 without my patch.

Right?

> 
> Thanks,
> 
> - Juri

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

On Wed, Jun 07, 2017 at 08:42:24AM +0900, Byungchul Park wrote:
> On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> > Hi,
> > 
> > On 02/06/17 16:31, Byungchul Park wrote:
> > > When the heap tree is empty, cp->elements[0].cpu has meaningless value.
> 
> Hi,
> 
> The meaningless value is 0.
> 
> > > We need to consider the case.
> > > 
> > > Signed-off-by: Byungchul Park <byungchul.park@lge.com>
> > > ---
> > >  kernel/sched/cpudeadline.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> > > index d4a6963..9b314a9 100644
> > > --- a/kernel/sched/cpudeadline.c
> > > +++ b/kernel/sched/cpudeadline.c
> > > @@ -110,7 +110,8 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
> > >  
> > >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> > >  {
> > > -	return cp->elements[0].cpu;
> > > +	int cpu = cp->elements[0].cpu;
> > > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> > 
> > Mmm, don't we get a WARN from cpumask_check() if we return -1 here?
> 
> The function does not return -1 without my patch.
> 
> Right?

Or the following patch would be needed, instead.

----->8-----

>From cada1345bf0ff8e6b5743999509d2abcacd79a9e Mon Sep 17 00:00:00 2001
From: Byungchul Park <byungchul.park@lge.com>
Date: Wed, 7 Jun 2017 09:05:34 +0900
Subject: [PATCH 2/2] sched/deadline: Initialize cp->elements[].cpu to an
 invalid value

Currently, when the heap tree is empty, cpudl_maximum_cpu() returns 0,
which causes unnecessary migration. It has to return an invalid value
e.g. -1 to prevent that.

Signed-off-by: Byungchul Park <byungchul.park@lge.com>
---
 kernel/sched/cpudeadline.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index d4a6963..0f67cea 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -266,8 +266,10 @@ int cpudl_init(struct cpudl *cp)
 		return -ENOMEM;
 	}
 
-	for_each_possible_cpu(i)
+	for_each_possible_cpu(i) {
+		cp->elements[i].cpu = -1;
 		cp->elements[i].idx = IDX_INVALID;
+	}
 
 	return 0;
 }
-- 
1.9.1

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Juri Lelli]

On 07/06/17 09:14, Byungchul Park wrote:
> On Wed, Jun 07, 2017 at 08:42:24AM +0900, Byungchul Park wrote:
> > On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> > > Hi,
> > > 
> > > On 02/06/17 16:31, Byungchul Park wrote:

[...]

> > > >  
> > > >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> > > >  {
> > > > -	return cp->elements[0].cpu;
> > > > +	int cpu = cp->elements[0].cpu;
> > > > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> > > 
> > > Mmm, don't we get a WARN from cpumask_check() if we return -1 here?
> > 
> > The function does not return -1 without my patch.
> > 
> > Right?
> 

That's actually my point: with the change you are proposing we will
start returning -1 and it looks to me that the WARN will start to fire.

What about the below instead (properly splitted in 2 patches I guess,
and I'm not sure at all the macro thing is pretty at all) ?

--->8---
 kernel/sched/cpudeadline.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index fba235c7d026..32e3dcef2b81 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -108,11 +108,17 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
 		cpudl_heapify_down(cp, idx);
 }
 
-static inline int cpudl_maximum(struct cpudl *cp)
-{
-	return cp->elements[0].cpu;
+#define cpudl_maximum(field)				\
+static inline int cpudl_maximum_##field			\
+(struct cpudl *cp)					\
+{							\
+	return cp->elements[0].field;			\
 }
 
+cpudl_maximum(cpu);
+cpudl_maximum(dl);
+cpudl_maximum(idx);
+
 /*
  * cpudl_find - find the best (later-dl) CPU in the system
  * @cp: the cpudl max-heap context
@@ -131,9 +137,10 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
 	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
 		best_cpu = cpumask_any(later_mask);
 		goto out;
-	} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
-			dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
-		best_cpu = cpudl_maximum(cp);
+	} else if (cpudl_maximum_idx(cp) != IDX_INVALID &&
+		   cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
+		   dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
+		best_cpu = cpudl_maximum_cpu(cp);
 		if (later_mask)
 			cpumask_set_cpu(best_cpu, later_mask);
 	}

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

On Thu, Jun 08, 2017 at 03:02:43PM +0100, Juri Lelli wrote:
> On 07/06/17 09:14, Byungchul Park wrote:
> > On Wed, Jun 07, 2017 at 08:42:24AM +0900, Byungchul Park wrote:
> > > On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> > > > Hi,
> > > > 
> > > > On 02/06/17 16:31, Byungchul Park wrote:
> 
> [...]
> 
> > > > >  
> > > > >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> > > > >  {
> > > > > -	return cp->elements[0].cpu;
> > > > > +	int cpu = cp->elements[0].cpu;
> > > > > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> > > > 
> > > > Mmm, don't we get a WARN from cpumask_check() if we return -1 here?
> > > 
> > > The function does not return -1 without my patch.
> > > 
> > > Right?
> > 
> 
> That's actually my point: with the change you are proposing we will
> start returning -1 and it looks to me that the WARN will start to fire.

Hi,

I see what you talk about. You are talking about WARN in cpumask_check().
Sorry for missing your words.

> What about the below instead (properly splitted in 2 patches I guess,
> and I'm not sure at all the macro thing is pretty at all) ?
> 
> --->8---
>  kernel/sched/cpudeadline.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index fba235c7d026..32e3dcef2b81 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -108,11 +108,17 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
>  		cpudl_heapify_down(cp, idx);
>  }
>  
> -static inline int cpudl_maximum(struct cpudl *cp)
> -{
> -	return cp->elements[0].cpu;
> +#define cpudl_maximum(field)				\
> +static inline int cpudl_maximum_##field			\
> +(struct cpudl *cp)					\
> +{							\
> +	return cp->elements[0].field;			\
>  }
>  
> +cpudl_maximum(cpu);
> +cpudl_maximum(dl);
> +cpudl_maximum(idx);
> +
>  /*
>   * cpudl_find - find the best (later-dl) CPU in the system
>   * @cp: the cpudl max-heap context
> @@ -131,9 +137,10 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>  	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
>  		best_cpu = cpumask_any(later_mask);
>  		goto out;
> -	} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> -			dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
> -		best_cpu = cpudl_maximum(cp);
> +	} else if (cpudl_maximum_idx(cp) != IDX_INVALID &&
> +		   cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
> +		   dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
> +		best_cpu = cpudl_maximum_cpu(cp);

This would also work and avoid unnecessary warning. I missed the check
to avoid it. https://lkml.org/lkml/2017/3/23/175 was an original patch
doing it.

By the way, frankly speaking, I don't like accessing the cpudl instant
several times without protection. I rather prefer the following..

But whatever. I like both.

Thnaks,
Byungchul

----->8-----
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index 9b314a9..1d369cf 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -137,11 +137,17 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
 	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
 		best_cpu = cpumask_any(later_mask);
 		goto out;
-	} else if (cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
-			dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
-		best_cpu = cpudl_maximum_cpu(cp);
-		if (later_mask)
-			cpumask_set_cpu(best_cpu, later_mask);
+	} else {
+		int max_cpu = cpudl_maximum_cpu(cp);
+		u64 max_dl = cpudl_maximum_dl(cp);
+
+		if (max_cpu != -1 &&
+		    cpumask_test_cpu(max_cpu, &p->cpus_allowed) &&
+		    dl_time_before(dl_se->deadline, max_dl)) {
+			best_cpu = max_cpu;
+			if (later_mask)
+				cpumask_set_cpu(best_cpu, later_mask);
+		}
 	}
 
 out:

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Juri Lelli]

On 09/06/17 11:43, Byungchul Park wrote:
> On Thu, Jun 08, 2017 at 03:02:43PM +0100, Juri Lelli wrote:
> > On 07/06/17 09:14, Byungchul Park wrote:
> > > On Wed, Jun 07, 2017 at 08:42:24AM +0900, Byungchul Park wrote:
> > > > On Tue, Jun 06, 2017 at 04:12:25PM +0100, Juri Lelli wrote:
> > > > > Hi,
> > > > > 
> > > > > On 02/06/17 16:31, Byungchul Park wrote:
> > 
> > [...]
> > 
> > > > > >  
> > > > > >  static inline int cpudl_maximum_cpu(struct cpudl *cp)
> > > > > >  {
> > > > > > -	return cp->elements[0].cpu;
> > > > > > +	int cpu = cp->elements[0].cpu;
> > > > > > +	return cp->elements[cpu].idx == IDX_INVALID ? -1 : cpu;
> > > > > 
> > > > > Mmm, don't we get a WARN from cpumask_check() if we return -1 here?
> > > > 
> > > > The function does not return -1 without my patch.
> > > > 
> > > > Right?
> > > 
> > 
> > That's actually my point: with the change you are proposing we will
> > start returning -1 and it looks to me that the WARN will start to fire.
> 
> Hi,
> 
> I see what you talk about. You are talking about WARN in cpumask_check().
> Sorry for missing your words.
> 
> > What about the below instead (properly splitted in 2 patches I guess,
> > and I'm not sure at all the macro thing is pretty at all) ?
> > 
> > --->8---
> >  kernel/sched/cpudeadline.c | 19 +++++++++++++------
> >  1 file changed, 13 insertions(+), 6 deletions(-)
> > 
> > diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> > index fba235c7d026..32e3dcef2b81 100644
> > --- a/kernel/sched/cpudeadline.c
> > +++ b/kernel/sched/cpudeadline.c
> > @@ -108,11 +108,17 @@ static void cpudl_heapify(struct cpudl *cp, int idx)
> >  		cpudl_heapify_down(cp, idx);
> >  }
> >  
> > -static inline int cpudl_maximum(struct cpudl *cp)
> > -{
> > -	return cp->elements[0].cpu;
> > +#define cpudl_maximum(field)				\
> > +static inline int cpudl_maximum_##field			\
> > +(struct cpudl *cp)					\
> > +{							\
> > +	return cp->elements[0].field;			\
> >  }
> >  
> > +cpudl_maximum(cpu);
> > +cpudl_maximum(dl);
> > +cpudl_maximum(idx);
> > +
> >  /*
> >   * cpudl_find - find the best (later-dl) CPU in the system
> >   * @cp: the cpudl max-heap context
> > @@ -131,9 +137,10 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
> >  	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
> >  		best_cpu = cpumask_any(later_mask);
> >  		goto out;
> > -	} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> > -			dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
> > -		best_cpu = cpudl_maximum(cp);
> > +	} else if (cpudl_maximum_idx(cp) != IDX_INVALID &&
> > +		   cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
> > +		   dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
> > +		best_cpu = cpudl_maximum_cpu(cp);
> 
> This would also work and avoid unnecessary warning. I missed the check
> to avoid it. https://lkml.org/lkml/2017/3/23/175 was an original patch
> doing it.
> 
> By the way, frankly speaking, I don't like accessing the cpudl instant
> several times without protection. I rather prefer the following..
> 
> But whatever. I like both.
> 
> Thnaks,
> Byungchul
> 
> ----->8-----
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index 9b314a9..1d369cf 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -137,11 +137,17 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>  	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
>  		best_cpu = cpumask_any(later_mask);
>  		goto out;
> -	} else if (cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
> -			dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
> -		best_cpu = cpudl_maximum_cpu(cp);
> -		if (later_mask)
> -			cpumask_set_cpu(best_cpu, later_mask);
> +	} else {
> +		int max_cpu = cpudl_maximum_cpu(cp);
> +		u64 max_dl = cpudl_maximum_dl(cp);
> +
> +		if (max_cpu != -1 &&
> +		    cpumask_test_cpu(max_cpu, &p->cpus_allowed) &&
> +		    dl_time_before(dl_se->deadline, max_dl)) {

Don't we access cp 3 times both ways?

Re: [PATCH 2/2] sched/deadline: Don't return invalid cpu in cpudl_maximum_cpu()

[Byungchul Park]

On Fri, Jun 09, 2017 at 12:42:06PM +0100, Juri Lelli wrote:
> > 
> > This would also work and avoid unnecessary warning. I missed the check
> > to avoid it. https://lkml.org/lkml/2017/3/23/175 was an original patch
> > doing it.
> > 
> > By the way, frankly speaking, I don't like accessing the cpudl instant
> > several times without protection. I rather prefer the following..
> > 
> > But whatever. I like both.
> > 
> > Thnaks,
> > Byungchul
> > 
> > ----->8-----
> > diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> > index 9b314a9..1d369cf 100644
> > --- a/kernel/sched/cpudeadline.c
> > +++ b/kernel/sched/cpudeadline.c
> > @@ -137,11 +137,17 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
> >  	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
> >  		best_cpu = cpumask_any(later_mask);
> >  		goto out;
> > -	} else if (cpumask_test_cpu(cpudl_maximum_cpu(cp), &p->cpus_allowed) &&
> > -			dl_time_before(dl_se->deadline, cpudl_maximum_dl(cp))) {
> > -		best_cpu = cpudl_maximum_cpu(cp);
> > -		if (later_mask)
> > -			cpumask_set_cpu(best_cpu, later_mask);
> > +	} else {
> > +		int max_cpu = cpudl_maximum_cpu(cp);
> > +		u64 max_dl = cpudl_maximum_dl(cp);
> > +
> > +		if (max_cpu != -1 &&
> > +		    cpumask_test_cpu(max_cpu, &p->cpus_allowed) &&
> > +		    dl_time_before(dl_se->deadline, max_dl)) {
> 
> Don't we access cp 3 times both ways?

I wonder if I miss something.. As you said, in most cases it will, but
I am not sure if we can guarentee that, regardless of arches or
compile optimization level. Sorry if I am wrong.