Can 750 user-mode binaries run on a 603e core?

Can 750 user-mode binaries run on a 603e core?

[Patrick J. Kelsey]

SGksDQogDQpJZiBoYXZlIHVzZXItbW9kZSBiaW5hcmllcyBidWlsdCBmb3IgYSA3NTAgY29yZSwg
c2F5IHdpdGggZ2NjIC1tY3B1PTc1MCwgc2hvdWxkIEkgZXhwZWN0IHRoZW0gdG8gcnVuIG9uIGEg
NjAzZSBjb3JlPyAgV291bGQgSSBoYXZlIHRvIHdvcnJ5IGFib3V0IHVuc3VwcG9ydGVkIGluc3Ry
dWN0aW9ucyBvciBpbnN0cnVjdGlvbiBzY2hlZHVsaW5nIGlzc3Vlcz8NCiANCklmIHRoZXJlJ3Mg
YSBiZXR0ZXIgbGlzdCBmb3IgdGhpcyBxdWVzdGlvbiwgcGxlYXNlIGFkdmlzZS4NCiANClRoYW5r
cywNClBhdA0K

Re: Can 750 user-mode binaries run on a 603e core?

[Kumar Gala]

On Jul 26, 2006, at 5:27 PM, Patrick J. Kelsey wrote:

> Hi,
>
> If have user-mode binaries built for a 750 core, say with gcc - 
> mcpu=750, should I expect them to run on a 603e core?  Would I have  
> to worry about unsupported instructions or instruction scheduling  
> issues?
>
> If there's a better list for this question, please advise.

Sure, there isn't different in the user space instruction sets  
between the 750 and 603e core.  The -mcpu is only tweaking  
instruction scheduling and making it specific for the 750.  The  
instruction scheduling differences aren't that huge if memory serves  
me correctly.  The 750 probably takes fewer cycles on multiple and  
divides.  I forget if the 750 has any additional execution units that  
the 603e didn't have (additional integer units, etc.).

- kumar

RE: Can 750 user-mode binaries run on a 603e core?

[Patrick J. Kelsey]
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RE: Can 750 user-mode binaries run on a 603e core?

[Patrick J. Kelsey]

KHNvcnJ5IGFib3V0IHRoYXQgbGFzdCBvbmUuIGhhZCBhIGJpdCBvZiB0cm91YmxlIHdpdGggYSBj
ZXJ0YWluIHdlYiBiYXNlZCBtYWlsIGNsaWVudC4uLikgDQogDQpUaGFua3MgZm9yIHRoZSByZXBs
eSwgS3VtYXIuDQogDQpUaGF0IHNvdW5kcyBlbmNvdXJhZ2luZy4gIE9uZSBvZiB0aGUgdGhpbmdz
IEkgd2FzIHdvcnJpZWQgYWJvdXQgd2l0aCBzY2hlZHVsaW5nIGRpZmZlcmVuY2VzIHdvdWxkIGJl
IGEgZGlmZmVyaW5nIG51bWJlciBvZiBicmFuY2ggZGVsYXkgc2xvdHMgYmV0d2VlbiB0aGUgdHdv
IGNvcmUgdmVyc2lvbnMuICBJJ20gc3RpbGwgYSBiaXQgbmV3IHRvIHRoZSBkZXRhaWxzIG9mIHRo
ZSBQb3dlclBDIGFyY2hpdGVjdHVyZSwgYW5kIGF0IHRoaXMgcG9pbnQgSSdtIG5vdCBldmVuIHN1
cmUgaWYgdGhlcmUgYXJlIGJyYW5jaCBkZWxheSBzbG90cywgYWx0aG91Z2ggaXQgZG9lcyBzZWVt
IGZyb20gbXkgcmVhZGluZyB0aGF0IHRoZSA2MDNlIGFuZCA3NTAgcGlwZWxpbmVzIGFyZSB0aGUg
c2FtZSwgaW4gd2hpY2ggY2FzZSB0aGVyZSB3b3VsZCBjZXJhdGlubHkgYmUgbm8gd29ycmllcyBo
ZXJlLg0KIA0KQXQgdGhpcyBwb2ludCwgSSdtIG5vdCBjb25jZXJuZWQgc28gbXVjaCBhYm91dCBh
biBpbmVmZmljaWVudCBzY2hlZHVsZSByZXN1bHRpbmcgZnJvbSBydW5uaW5nIC1tY3B1PTc1MCBj
b2RlIG9uIGEgNjAzZSBhcyBsb25nIGFzIHRoZSBleGVjdXRpb24gaXMgY29ycmVjdC4NCiANClBh
dA0KIA0KDQoJLS0tLS1PcmlnaW5hbCBNZXNzYWdlLS0tLS0gDQoJRnJvbTogbGludXhwcGMtZGV2
LWJvdW5jZXMra2Vsc2V5cGo9ZmFpcm1vdW50YXV0b21hdGlvbi5jb21Ab3psYWJzLm9yZyBvbiBi
ZWhhbGYgb2YgS3VtYXIgR2FsYSANCglTZW50OiBXZWQgNy8yNi8yMDA2IDg6MDQgUE0gDQoJVG86
IFBhdHJpY2sgSi4gS2Vsc2V5IA0KCUNjOiBsaW51eHBwYy1kZXZAb3psYWJzLm9yZyANCglTdWJq
ZWN0OiBSZTogQ2FuIDc1MCB1c2VyLW1vZGUgYmluYXJpZXMgcnVuIG9uIGEgNjAzZSBjb3JlPw0K
CQ0KCQ0KDQoNCglPbiBKdWwgMjYsIDIwMDYsIGF0IDU6MjcgUE0sIFBhdHJpY2sgSi4gS2Vsc2V5
IHdyb3RlOg0KCQ0KCT4gSGksDQoJPg0KCT4gSWYgaGF2ZSB1c2VyLW1vZGUgYmluYXJpZXMgYnVp
bHQgZm9yIGEgNzUwIGNvcmUsIHNheSB3aXRoIGdjYyAtDQoJPiBtY3B1PTc1MCwgc2hvdWxkIEkg
ZXhwZWN0IHRoZW0gdG8gcnVuIG9uIGEgNjAzZSBjb3JlPyAgV291bGQgSSBoYXZlIA0KCT4gdG8g
d29ycnkgYWJvdXQgdW5zdXBwb3J0ZWQgaW5zdHJ1Y3Rpb25zIG9yIGluc3RydWN0aW9uIHNjaGVk
dWxpbmcgDQoJPiBpc3N1ZXM/DQoJPg0KCT4gSWYgdGhlcmUncyBhIGJldHRlciBsaXN0IGZvciB0
aGlzIHF1ZXN0aW9uLCBwbGVhc2UgYWR2aXNlLg0KCQ0KCVN1cmUsIHRoZXJlIGlzbid0IGRpZmZl
cmVudCBpbiB0aGUgdXNlciBzcGFjZSBpbnN0cnVjdGlvbiBzZXRzIA0KCWJldHdlZW4gdGhlIDc1
MCBhbmQgNjAzZSBjb3JlLiAgVGhlIC1tY3B1IGlzIG9ubHkgdHdlYWtpbmcgDQoJaW5zdHJ1Y3Rp
b24gc2NoZWR1bGluZyBhbmQgbWFraW5nIGl0IHNwZWNpZmljIGZvciB0aGUgNzUwLiAgVGhlIA0K
CWluc3RydWN0aW9uIHNjaGVkdWxpbmcgZGlmZmVyZW5jZXMgYXJlbid0IHRoYXQgaHVnZSBpZiBt
ZW1vcnkgc2VydmVzIA0KCW1lIGNvcnJlY3RseS4gIFRoZSA3NTAgcHJvYmFibHkgdGFrZXMgZmV3
ZXIgY3ljbGVzIG9uIG11bHRpcGxlIGFuZCANCglkaXZpZGVzLiAgSSBmb3JnZXQgaWYgdGhlIDc1
MCBoYXMgYW55IGFkZGl0aW9uYWwgZXhlY3V0aW9uIHVuaXRzIHRoYXQgDQoJdGhlIDYwM2UgZGlk
bid0IGhhdmUgKGFkZGl0aW9uYWwgaW50ZWdlciB1bml0cywgZXRjLikuDQoJDQoJLSBrdW1hcg0K
CV9fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fDQoJTGludXhw
cGMtZGV2IG1haWxpbmcgbGlzdA0KCUxpbnV4cHBjLWRldkBvemxhYnMub3JnDQoJaHR0cHM6Ly9v
emxhYnMub3JnL21haWxtYW4vbGlzdGluZm8vbGludXhwcGMtZGV2DQoJDQoNCg==

RE: Can 750 user-mode binaries run on a 603e core?

[Matt Sealey]

The 603/603e has 1 integer unit. The 604/604e and 7xx lines had
2.

Basically if you optimize for the 750 it may queue up integer
ops assuming there is room to execute them simultaneously, when
in fact there isn't.

But as kumar said, userspace is absolutely positively identical,
the chips from 603e to the latest G4 are entirely binary
compatible for integer and FPU code. Most people just optimize
for the 603e and let the CPU handle the rest.

-- 
Matt Sealey <matt@genesi-usa.com>
Manager, Genesi, Developer Relations
 

> -----Original Message-----
> From: linuxppc-dev-bounces+matt=genesi-usa.com@ozlabs.org 
> [mailto:linuxppc-dev-bounces+matt=genesi-usa.com@ozlabs.org] 
> On Behalf Of Patrick J. Kelsey
> Sent: Thursday, July 27, 2006 5:54 AM
> To: linuxppc-dev@ozlabs.org
> Subject: RE: Can 750 user-mode binaries run on a 603e core?
> 
> (sorry about that last one. had a bit of trouble with a 
> certain web based mail client...) 
>  
> Thanks for the reply, Kumar.
>  
> That sounds encouraging.  One of the things I was worried 
> about with scheduling differences would be a differing number 
> of branch delay slots between the two core versions.  I'm 
> still a bit new to the details of the PowerPC architecture, 
> and at this point I'm not even sure if there are branch delay 
> slots, although it does seem from my reading that the 603e 
> and 750 pipelines are the same, in which case there would 
> ceratinly be no worries here.
>  
> At this point, I'm not concerned so much about an inefficient 
> schedule resulting from running -mcpu=750 code on a 603e as 
> long as the execution is correct.
>  
> Pat
>  
> 
> 	-----Original Message----- 
> 	From: 
> linuxppc-dev-bounces+kelseypj=fairmountautomation.com@ozlabs.o
> rg on behalf of Kumar Gala 
> 	Sent: Wed 7/26/2006 8:04 PM 
> 	To: Patrick J. Kelsey 
> 	Cc: linuxppc-dev@ozlabs.org 
> 	Subject: Re: Can 750 user-mode binaries run on a 603e core?
> 	
> 	
> 
> 
> 	On Jul 26, 2006, at 5:27 PM, Patrick J. Kelsey wrote:
> 	
> 	> Hi,
> 	>
> 	> If have user-mode binaries built for a 750 core, say 
> with gcc -
> 	> mcpu=750, should I expect them to run on a 603e core? 
>  Would I have 
> 	> to worry about unsupported instructions or 
> instruction scheduling 
> 	> issues?
> 	>
> 	> If there's a better list for this question, please advise.
> 	
> 	Sure, there isn't different in the user space instruction sets 
> 	between the 750 and 603e core.  The -mcpu is only tweaking 
> 	instruction scheduling and making it specific for the 750.  The 
> 	instruction scheduling differences aren't that huge if 
> memory serves 
> 	me correctly.  The 750 probably takes fewer cycles on 
> multiple and 
> 	divides.  I forget if the 750 has any additional 
> execution units that 
> 	the 603e didn't have (additional integer units, etc.).
> 	
> 	- kumar
> 	_______________________________________________
> 	Linuxppc-dev mailing list
> 	Linuxppc-dev@ozlabs.org
> 	https://ozlabs.org/mailman/listinfo/linuxppc-dev
> 	
> 
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev
> 

RE: Can 750 user-mode binaries run on a 603e core?

[Patrick J. Kelsey]
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Re: Can 750 user-mode binaries run on a 603e core?

[Jerry Van Baren]

Patrick J. Kelsey wrote:
> (sorry about that last one. had a bit of trouble with a certain web based mail client...) 
>  
> Thanks for the reply, Kumar.
>  
> That sounds encouraging.  One of the things I was worried about with
> scheduling differences would be a differing number of branch delay
> slots between the two core versions.  I'm still a bit new to the
> details of the PowerPC architecture, and at this point I'm not even
> sure if there are branch delay slots, although it does seem from my
> reading that the 603e and 750 pipelines are the same, in which case
> there would ceratinly be no worries here.

FWIIW, the PowerPC architecture has hardware instruction interlocking 
and scheduling and doesn't require the compiler to implement the branch 
delay slots like, for instance, the MIPS architecture.  This is much 
more compiler and portability friendly, but at the expense of more logic 
in the processor (the PowerPC is not nearly as minimalistic as the MIPS).

> At this point, I'm not concerned so much about an inefficient
> schedule resulting from running -mcpu=750 code on a 603e as long as
> the execution is correct.
>  
> Pat

gvb

Re: Can 750 user-mode binaries run on a 603e core?

[Andy Fleming]

Branch delay slots are evil, and PowerPC did not use them.

'nuff said.

Pretty much any powerpc binary should work on any PowerPC system.   
Actually, we use a 603-compiled root file system for our e500  
processors (in our lab), and don't run into any problems.  This is

On Jul 27, 2006, at 07:36, Jerry Van Baren wrote:

> Patrick J. Kelsey wrote:
>> (sorry about that last one. had a bit of trouble with a certain  
>> web based mail client...)
>>
>> Thanks for the reply, Kumar.
>>
>> That sounds encouraging.  One of the things I was worried about with
>> scheduling differences would be a differing number of branch delay
>> slots between the two core versions.  I'm still a bit new to the
>> details of the PowerPC architecture, and at this point I'm not even
>> sure if there are branch delay slots, although it does seem from my
>> reading that the 603e and 750 pipelines are the same, in which case
>> there would ceratinly be no worries here.
>
> FWIIW, the PowerPC architecture has hardware instruction interlocking
> and scheduling and doesn't require the compiler to implement the  
> branch
> delay slots like, for instance, the MIPS architecture.  This is much
> more compiler and portability friendly, but at the expense of more  
> logic
> in the processor (the PowerPC is not nearly as minimalistic as the  
> MIPS).
>
>> At this point, I'm not concerned so much about an inefficient
>> schedule resulting from running -mcpu=750 code on a 603e as long as
>> the execution is correct.
>>
>> Pat
>
> gvb
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev

Re: Can 750 user-mode binaries run on a 603e core?

[Kumar Gala]

On Jul 27, 2006, at 10:30 AM, Andy Fleming wrote:

> Branch delay slots are evil, and PowerPC did not use them.
>
> 'nuff said.
>
> Pretty much any powerpc binary should work on any PowerPC system.
> Actually, we use a 603-compiled root file system for our e500
> processors (in our lab), and don't run into any problems.  This is

Let's be clear, that is done by emulating the floating point  
instructions in the kernel.  (And there are a few user space integer/ 
control instructions that e500 doesn't implement, but your rarely run  
into them).

- k

> On Jul 27, 2006, at 07:36, Jerry Van Baren wrote:
>
>> Patrick J. Kelsey wrote:
>>> (sorry about that last one. had a bit of trouble with a certain
>>> web based mail client...)
>>>
>>> Thanks for the reply, Kumar.
>>>
>>> That sounds encouraging.  One of the things I was worried about with
>>> scheduling differences would be a differing number of branch delay
>>> slots between the two core versions.  I'm still a bit new to the
>>> details of the PowerPC architecture, and at this point I'm not even
>>> sure if there are branch delay slots, although it does seem from my
>>> reading that the 603e and 750 pipelines are the same, in which case
>>> there would ceratinly be no worries here.
>>
>> FWIIW, the PowerPC architecture has hardware instruction interlocking
>> and scheduling and doesn't require the compiler to implement the
>> branch
>> delay slots like, for instance, the MIPS architecture.  This is much
>> more compiler and portability friendly, but at the expense of more
>> logic
>> in the processor (the PowerPC is not nearly as minimalistic as the
>> MIPS).
>>
>>> At this point, I'm not concerned so much about an inefficient
>>> schedule resulting from running -mcpu=750 code on a 603e as long as
>>> the execution is correct.
>>>
>>> Pat
>>
>> gvb
>> _______________________________________________
>> Linuxppc-dev mailing list
>> Linuxppc-dev@ozlabs.org
>> https://ozlabs.org/mailman/listinfo/linuxppc-dev
>
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev