Patchwork [4.7,google] Support for getting CPU type and feature information at run-time. (issue4893046)

login
register
mail settings
Submitter Sriraman Tallam
Date Aug. 26, 2011, 12:37 a.m.
Message ID <CAAs8Hmz1PKZ42jHq7rw_EnCn+kwKtsnBmYHoNWoeJAwzjTkYpg@mail.gmail.com>
Download mbox | patch
Permalink /patch/111693/
State New
Headers show

Comments

Sriraman Tallam - Aug. 26, 2011, 12:37 a.m.
Hi,

  Thanks for all the comments. I am attaching a new patch
incorporating all of the changes mentioned, mainly :

1) Make __cpu_indicator_init a constructor in libgcc and guard to call
it only once.
2) Add symbol versions.
3) Move all builtins to the i386 port.
4) Add check for atom processor.
5) No separate passes to fold the builtins.

Please let me know what you think.
Thanks,
-Sri.

	* config/i386/i386.c (build_struct_with_one_bit_fields): New function.
	(make_var_decl): New function.
	(get_field_from_struct): New function.
	(fold_builtin_target): New function.
	(ix86_fold_builtin): New function.
	(ix86_expand_builtin): Expand new builtins by folding them.
	(TARGET_FOLD_BUILTIN): New macro.
	(IX86_BUILTIN_CPU_SUPPORTS_CMOV): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_MMX): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSE): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSE2): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSE3): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSSE3): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSE4_1): New enum value.
	(IX86_BUILTIN_CPU_SUPPORTS_SSE4_2): New enum value.
	(IX86_BUILTIN_CPU_IS_AMD): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL_ATOM): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL_CORE2): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE): New enum value.
	(IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE): New enum value.
	(IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA): New enum value.
	(IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI): New enum value.
	(IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL): New enum value.
	* config/i386/libgcc-glibc.ver (__cpu_indicator_init): Export symbol.
	(__cpu_model): Export symbol.
	(__cpu_features): Export symbol.
	* config/i386/i386-builtin-types.def: New function type.

	* config/i386/i386-cpuinfo.c: New file.
	* config/i386/t-cpuinfo: New file.
	* config.host: Add t-cpuinfo to link i386-cpuinfo.o with libgcc

	* gcc.dg/builtin_target.c: New test.




On Tue, Aug 23, 2011 at 4:35 AM, Michael Matz <matz@suse.de> wrote:
> Hi,
>
> On Mon, 22 Aug 2011, H.J. Lu wrote:
>
>> > void __attribute__((constructor)) bla(void)
>> > {
>> >  __cpu_indicator_init ();
>> > }
>> >
>> > I don't see any complication.?
>> >
>>
>> Order of constructors.  A constructor may call functions
>> which use __cpu_indicator.
>
> That's why I wrote also:
>
>> The initializer function has to be callable from pre-.init contexts, e.g.
>> ifunc dispatchers.
>
> It obviously has to be guarded against multiple calls.  The ctor in libgcc
> would be mere convenience because then non-ctor code can rely on the data
> being initialized, and only (potential) ctor code has to check and call
> the init function on demand.
>
>
> Ciao,
> Michael.
H.J. Lu - Aug. 26, 2011, 1:02 a.m.
On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
> Hi,
>
>  Thanks for all the comments. I am attaching a new patch
> incorporating all of the changes mentioned, mainly :
>
> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
> it only once.

This is unreliable and you don't need 3 symbols from libgcc. You can use

static struct cpu_indicator
{
  feature
  model
  status
} cpu_indicator;

struct cpu_indicator *
__get_cpu_indicator ()
{
   if cpu_indicator is uninitialized; then
      initialize cpu_indicator;
  return &cpu_indicator;
}

You can simply call __get_cpu_indicator to
get a pointer to cpu_indicator;
Sriraman Tallam - Aug. 26, 2011, 5:06 p.m.
On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>> Hi,
>>
>>  Thanks for all the comments. I am attaching a new patch
>> incorporating all of the changes mentioned, mainly :
>>
>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>> it only once.
>
> This is unreliable and you don't need 3 symbols from libgcc. You can use

Do you mean it is unreliable because of the constructor ordering problem?

>
> static struct cpu_indicator
> {
>  feature
>  model
>  status
> } cpu_indicator;
>
> struct cpu_indicator *
> __get_cpu_indicator ()
> {
>   if cpu_indicator is uninitialized; then
>      initialize cpu_indicator;
>  return &cpu_indicator;
> }
>
> You can simply call __get_cpu_indicator to
> get a pointer to cpu_indicator;
>
> --
> H.J.
>
H.J. Lu - Aug. 26, 2011, 5:10 p.m.
On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>> Hi,
>>>
>>>  Thanks for all the comments. I am attaching a new patch
>>> incorporating all of the changes mentioned, mainly :
>>>
>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>> it only once.
>>
>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>
> Do you mean it is unreliable because of the constructor ordering problem?
>

You do not have total control when __cpu_indicator_init is called.

Also you shouldn't use bitfield in

struct __processor_model
+{
+  unsigned int __cpu_is_amd : 1;
+  unsigned int __cpu_is_intel : 1;
+  unsigned int __cpu_is_intel_atom : 1;
+  unsigned int __cpu_is_intel_core2 : 1;
+  unsigned int __cpu_is_intel_corei7_nehalem : 1;
+  unsigned int __cpu_is_intel_corei7_westmere : 1;
+  unsigned int __cpu_is_intel_corei7_sandybridge : 1;
+  unsigned int __cpu_is_amdfam10_barcelona : 1;
+  unsigned int __cpu_is_amdfam10_shanghai : 1;
+  unsigned int __cpu_is_amdfam10_istanbul : 1;
+} __cpu_model = {0};
+

A processor can't be both Atom and Core 2.
Sriraman Tallam - Aug. 26, 2011, 5:17 p.m.
On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>> Hi,
>>>>
>>>>  Thanks for all the comments. I am attaching a new patch
>>>> incorporating all of the changes mentioned, mainly :
>>>>
>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>> it only once.
>>>
>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>
>> Do you mean it is unreliable because of the constructor ordering problem?
>>
>
> You do not have total control when __cpu_indicator_init is called.

Like  discussed before, for non-ctor functions, which in my opinion is
the common use case, it works out great because __cpu_indicator_init
is guaranteed to be called and I save doing an extra check. It is only
for other ctors where this is a problem. So other ctors call this
explicitly.  What did I miss?

Thanks,
-Sri.

>
> Also you shouldn't use bitfield in
>
> struct __processor_model
> +{
> +  unsigned int __cpu_is_amd : 1;
> +  unsigned int __cpu_is_intel : 1;
> +  unsigned int __cpu_is_intel_atom : 1;
> +  unsigned int __cpu_is_intel_core2 : 1;
> +  unsigned int __cpu_is_intel_corei7_nehalem : 1;
> +  unsigned int __cpu_is_intel_corei7_westmere : 1;
> +  unsigned int __cpu_is_intel_corei7_sandybridge : 1;
> +  unsigned int __cpu_is_amdfam10_barcelona : 1;
> +  unsigned int __cpu_is_amdfam10_shanghai : 1;
> +  unsigned int __cpu_is_amdfam10_istanbul : 1;
> +} __cpu_model = {0};
> +
>
> A processor can't be both Atom and Core 2.
>
> --
> H.J.
>
H.J. Lu - Aug. 26, 2011, 5:24 p.m.
On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>> Hi,
>>>>>
>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>> incorporating all of the changes mentioned, mainly :
>>>>>
>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>> it only once.
>>>>
>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>
>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>
>>
>> You do not have total control when __cpu_indicator_init is called.
>
> Like  discussed before, for non-ctor functions, which in my opinion is
> the common use case, it works out great because __cpu_indicator_init
> is guaranteed to be called and I save doing an extra check. It is only
>> for other ctors where this is a problem. So other ctors call this
> explicitly.  What did I miss?
>

I have

static void foo ( void ) __attribute__((constructor));

static void foo ( void )
{
   ...
   call bar ();
   ...
}

in my application. bar () uses those cpu specific functions.
foo () is called before __cpu_indicator_init.  Since IFUNC
returns the cpu specific function address only for the
first call, the proper cpu specific functions will never be used.
Sriraman Tallam - Aug. 26, 2011, 5:37 p.m.
On Fri, Aug 26, 2011 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>> Hi,
>>>>>>
>>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>>> incorporating all of the changes mentioned, mainly :
>>>>>>
>>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>>> it only once.
>>>>>
>>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>>
>>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>>
>>>
>>> You do not have total control when __cpu_indicator_init is called.
>>
>> Like  discussed before, for non-ctor functions, which in my opinion is
>> the common use case, it works out great because __cpu_indicator_init
>> is guaranteed to be called and I save doing an extra check. It is only
>>> for other ctors where this is a problem. So other ctors call this
>> explicitly.  What did I miss?
>>
>
> I have
>
> static void foo ( void ) __attribute__((constructor));
>
> static void foo ( void )
> {
>   ...
>   call bar ();
>   ...
> }
>
> in my application. bar () uses those cpu specific functions.
> foo () is called before __cpu_indicator_init.  Since IFUNC
> returns the cpu specific function address only for the
> first call, the proper cpu specific functions will never be used.

Please correct me if I am wrong since I did not follow the IFUNC part
you mentioned.  However, it looks like this could be solved with
adding an explicit call to __cpu_indicator_init from within the ctor
foo. To me, it seems like the pain of adding this call explicitly in
other ctors is worth it because it works cleanly for non-ctors.

static void foo ( void ) __attribute__((constructor));

static void foo ( void )
{
  ...
  __cpu_indicator_init ();
  call bar ();
  ...
}

Will this work?

Thanks,
-Sri.

>
>
> --
> H.J.
>
H.J. Lu - Aug. 26, 2011, 5:45 p.m.
On Fri, Aug 26, 2011 at 10:37 AM, Sriraman Tallam <tmsriram@google.com> wrote:
> On Fri, Aug 26, 2011 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>>>> incorporating all of the changes mentioned, mainly :
>>>>>>>
>>>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>>>> it only once.
>>>>>>
>>>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>>>
>>>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>>>
>>>>
>>>> You do not have total control when __cpu_indicator_init is called.
>>>
>>> Like  discussed before, for non-ctor functions, which in my opinion is
>>> the common use case, it works out great because __cpu_indicator_init
>>> is guaranteed to be called and I save doing an extra check. It is only
>>>> for other ctors where this is a problem. So other ctors call this
>>> explicitly.  What did I miss?
>>>
>>
>> I have
>>
>> static void foo ( void ) __attribute__((constructor));
>>
>> static void foo ( void )
>> {
>>   ...
>>   call bar ();
>>   ...
>> }
>>
>> in my application. bar () uses those cpu specific functions.
>> foo () is called before __cpu_indicator_init.  Since IFUNC
>> returns the cpu specific function address only for the
>> first call, the proper cpu specific functions will never be used.
>
> Please correct me if I am wrong since I did not follow the IFUNC part
> you mentioned.  However, it looks like this could be solved with
> adding an explicit call to __cpu_indicator_init from within the ctor
> foo. To me, it seems like the pain of adding this call explicitly in
> other ctors is worth it because it works cleanly for non-ctors.
>
> static void foo ( void ) __attribute__((constructor));
>
> static void foo ( void )
> {
>  ...
>  __cpu_indicator_init ();
>  call bar ();
>  ...
> }
>
> Will this work?
>
>

Do I have to do that in every constructor, including
C++ global constructors?  It is ridiculous.
Xinliang David Li - Aug. 26, 2011, 5:58 p.m.
Is there a standard way to force this init function to be called
before all ctors?  Adding a ctor in one crtx.o ?

David

On Fri, Aug 26, 2011 at 10:45 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Aug 26, 2011 at 10:37 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>> On Fri, Aug 26, 2011 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>>>>> incorporating all of the changes mentioned, mainly :
>>>>>>>>
>>>>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>>>>> it only once.
>>>>>>>
>>>>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>>>>
>>>>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>>>>
>>>>>
>>>>> You do not have total control when __cpu_indicator_init is called.
>>>>
>>>> Like  discussed before, for non-ctor functions, which in my opinion is
>>>> the common use case, it works out great because __cpu_indicator_init
>>>> is guaranteed to be called and I save doing an extra check. It is only
>>>>> for other ctors where this is a problem. So other ctors call this
>>>> explicitly.  What did I miss?
>>>>
>>>
>>> I have
>>>
>>> static void foo ( void ) __attribute__((constructor));
>>>
>>> static void foo ( void )
>>> {
>>>   ...
>>>   call bar ();
>>>   ...
>>> }
>>>
>>> in my application. bar () uses those cpu specific functions.
>>> foo () is called before __cpu_indicator_init.  Since IFUNC
>>> returns the cpu specific function address only for the
>>> first call, the proper cpu specific functions will never be used.
>>
>> Please correct me if I am wrong since I did not follow the IFUNC part
>> you mentioned.  However, it looks like this could be solved with
>> adding an explicit call to __cpu_indicator_init from within the ctor
>> foo. To me, it seems like the pain of adding this call explicitly in
>> other ctors is worth it because it works cleanly for non-ctors.
>>
>> static void foo ( void ) __attribute__((constructor));
>>
>> static void foo ( void )
>> {
>>  ...
>>  __cpu_indicator_init ();
>>  call bar ();
>>  ...
>> }
>>
>> Will this work?
>>
>>
>
> Do I have to do that in every constructor, including
> C++ global constructors?  It is ridiculous.
>
> --
> H.J.
>
Sriraman Tallam - Aug. 26, 2011, 6:20 p.m.
On Fri, Aug 26, 2011 at 10:45 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Aug 26, 2011 at 10:37 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>> On Fri, Aug 26, 2011 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>> On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>>>>> incorporating all of the changes mentioned, mainly :
>>>>>>>>
>>>>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>>>>> it only once.
>>>>>>>
>>>>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>>>>
>>>>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>>>>
>>>>>
>>>>> You do not have total control when __cpu_indicator_init is called.
>>>>
>>>> Like  discussed before, for non-ctor functions, which in my opinion is
>>>> the common use case, it works out great because __cpu_indicator_init
>>>> is guaranteed to be called and I save doing an extra check. It is only
>>>>> for other ctors where this is a problem. So other ctors call this
>>>> explicitly.  What did I miss?
>>>>
>>>
>>> I have
>>>
>>> static void foo ( void ) __attribute__((constructor));
>>>
>>> static void foo ( void )
>>> {
>>>   ...
>>>   call bar ();
>>>   ...
>>> }
>>>
>>> in my application. bar () uses those cpu specific functions.
>>> foo () is called before __cpu_indicator_init.  Since IFUNC
>>> returns the cpu specific function address only for the
>>> first call, the proper cpu specific functions will never be used.
>>
>> Please correct me if I am wrong since I did not follow the IFUNC part
>> you mentioned.  However, it looks like this could be solved with
>> adding an explicit call to __cpu_indicator_init from within the ctor
>> foo. To me, it seems like the pain of adding this call explicitly in
>> other ctors is worth it because it works cleanly for non-ctors.
>>
>> static void foo ( void ) __attribute__((constructor));
>>
>> static void foo ( void )
>> {
>>  ...
>>  __cpu_indicator_init ();
>>  call bar ();
>>  ...
>> }
>>
>> Will this work?
>>
>>
>
> Do I have to do that in every constructor, including
> C++ global constructors?  It is ridiculous.

It seems like libgcc is on the link line after user code in the
command-line and so __cpu_indicator_init should fire first, both when
statically and dynamically linked.
Example:

foo.cc:
int  __attribute__ ((constructor))
foo ()
{
  return 0;
}


However, with something like this :

g++ -Wl,--u,__cpu_indicator_init  -lgcc foo.cc

foo gets called ahead of __cpu_indicator_init. For these abnormal link
usages, call it explicitly. So, can you please give me a common use
case where __cpu_inidicator_init will get called after a constructor.

Thanks,
-Sri.

>
> --
> H.J.
>
Xinliang David Li - Aug. 26, 2011, 8:44 p.m.
IFUNC selector will need to call get_cpu_indicator (as proposed by HJ
or something similar), while in other contexts, the implementation
should find a way to make sure the indicator is already initialized
such that the builtins accessing the features can be directly used
(See also Michael and Richard's previous comments).  The runtime
penalty is much smaller.

david

On Fri, Aug 26, 2011 at 10:37 AM, Sriraman Tallam <tmsriram@google.com> wrote:
> On Fri, Aug 26, 2011 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Fri, Aug 26, 2011 at 10:17 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>> On Fri, Aug 26, 2011 at 10:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>> On Fri, Aug 26, 2011 at 10:06 AM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>> On Thu, Aug 25, 2011 at 6:02 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>>>>> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>>  Thanks for all the comments. I am attaching a new patch
>>>>>>> incorporating all of the changes mentioned, mainly :
>>>>>>>
>>>>>>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>>>>>>> it only once.
>>>>>>
>>>>>> This is unreliable and you don't need 3 symbols from libgcc. You can use
>>>>>
>>>>> Do you mean it is unreliable because of the constructor ordering problem?
>>>>>
>>>>
>>>> You do not have total control when __cpu_indicator_init is called.
>>>
>>> Like  discussed before, for non-ctor functions, which in my opinion is
>>> the common use case, it works out great because __cpu_indicator_init
>>> is guaranteed to be called and I save doing an extra check. It is only
>>>> for other ctors where this is a problem. So other ctors call this
>>> explicitly.  What did I miss?
>>>
>>
>> I have
>>
>> static void foo ( void ) __attribute__((constructor));
>>
>> static void foo ( void )
>> {
>>   ...
>>   call bar ();
>>   ...
>> }
>>
>> in my application. bar () uses those cpu specific functions.
>> foo () is called before __cpu_indicator_init.  Since IFUNC
>> returns the cpu specific function address only for the
>> first call, the proper cpu specific functions will never be used.
>
> Please correct me if I am wrong since I did not follow the IFUNC part
> you mentioned.  However, it looks like this could be solved with
> adding an explicit call to __cpu_indicator_init from within the ctor
> foo. To me, it seems like the pain of adding this call explicitly in
> other ctors is worth it because it works cleanly for non-ctors.
>
> static void foo ( void ) __attribute__((constructor));
>
> static void foo ( void )
> {
>  ...
>  __cpu_indicator_init ();
>  call bar ();
>  ...
> }
>
> Will this work?
>
> Thanks,
> -Sri.
>
>>
>>
>> --
>> H.J.
>>
>
Mike Stump - Aug. 28, 2011, 5:41 p.m.
On Aug 25, 2011, at 6:02 PM, H.J. Lu wrote:
> On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
>> 
>> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
>> it only once.
> 
> This is unreliable and you don't need 3 symbols from libgcc.

I'll add that once you start adding ctors, it is hard to impossible to ever get rid of them, and some people actually care about start up time and have static warnings for any code that has _any_ global constructors.  By having it hidden behind an api, at least only that that want it, pay the price for it.
Mike Stump - Aug. 28, 2011, 5:43 p.m.
On Aug 26, 2011, at 10:58 AM, Xinliang David Li wrote:
> Is there a standard way to force this init function to be called
> before all ctors?  Adding a ctor in one crtx.o ?

Including the ctors that want to run before all other ctors?  Think about that carefully.  :-)
Xinliang David Li - Aug. 29, 2011, 12:31 a.m.
Sri, please add a new api to do cpu_indicator initialization on demand
to be used in IFUNC context. Perhaps also add some debug check to make
sure no conflicting cpu model is set.

Ok for google branches for now while the discussion continues.

thanks,

David

On Thu, Aug 25, 2011 at 5:37 PM, Sriraman Tallam <tmsriram@google.com> wrote:
> Hi,
>
>  Thanks for all the comments. I am attaching a new patch
> incorporating all of the changes mentioned, mainly :
>
> 1) Make __cpu_indicator_init a constructor in libgcc and guard to call
> it only once.
> 2) Add symbol versions.
> 3) Move all builtins to the i386 port.
> 4) Add check for atom processor.
> 5) No separate passes to fold the builtins.
>
> Please let me know what you think.
> Thanks,
> -Sri.
>
>        * config/i386/i386.c (build_struct_with_one_bit_fields): New function.
>        (make_var_decl): New function.
>        (get_field_from_struct): New function.
>        (fold_builtin_target): New function.
>        (ix86_fold_builtin): New function.
>        (ix86_expand_builtin): Expand new builtins by folding them.
>        (TARGET_FOLD_BUILTIN): New macro.
>        (IX86_BUILTIN_CPU_SUPPORTS_CMOV): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_MMX): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSE): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSE2): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSE3): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSSE3): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSE4_1): New enum value.
>        (IX86_BUILTIN_CPU_SUPPORTS_SSE4_2): New enum value.
>        (IX86_BUILTIN_CPU_IS_AMD): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL_ATOM): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL_CORE2): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE): New enum value.
>        (IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE): New enum value.
>        (IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA): New enum value.
>        (IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI): New enum value.
>        (IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL): New enum value.
>        * config/i386/libgcc-glibc.ver (__cpu_indicator_init): Export symbol.
>        (__cpu_model): Export symbol.
>        (__cpu_features): Export symbol.
>        * config/i386/i386-builtin-types.def: New function type.
>
>        * config/i386/i386-cpuinfo.c: New file.
>        * config/i386/t-cpuinfo: New file.
>        * config.host: Add t-cpuinfo to link i386-cpuinfo.o with libgcc
>
>        * gcc.dg/builtin_target.c: New test.
>
>
>
>
> On Tue, Aug 23, 2011 at 4:35 AM, Michael Matz <matz@suse.de> wrote:
>> Hi,
>>
>> On Mon, 22 Aug 2011, H.J. Lu wrote:
>>
>>> > void __attribute__((constructor)) bla(void)
>>> > {
>>> >  __cpu_indicator_init ();
>>> > }
>>> >
>>> > I don't see any complication.?
>>> >
>>>
>>> Order of constructors.  A constructor may call functions
>>> which use __cpu_indicator.
>>
>> That's why I wrote also:
>>
>>> The initializer function has to be callable from pre-.init contexts, e.g.
>>> ifunc dispatchers.
>>
>> It obviously has to be guarded against multiple calls.  The ctor in libgcc
>> would be mere convenience because then non-ctor code can rely on the data
>> being initialized, and only (potential) ctor code has to check and call
>> the init function on demand.
>>
>>
>> Ciao,
>> Michael.
>

Patch

Index: libgcc/config.host
===================================================================
--- libgcc/config.host	(revision 177767)
+++ libgcc/config.host	(working copy)
@@ -609,7 +609,7 @@  case ${host} in
 i[34567]86-*-linux* | x86_64-*-linux* | \
   i[34567]86-*-kfreebsd*-gnu | i[34567]86-*-knetbsd*-gnu | \
   i[34567]86-*-gnu*)
-	tmake_file="${tmake_file} t-tls"
+	tmake_file="${tmake_file} t-tls i386/t-cpuinfo"
 	if test "$libgcc_cv_cfi" = "yes"; then
 		tmake_file="${tmake_file} t-stack i386/t-stack-i386"
 	fi
Index: libgcc/config/i386/t-cpuinfo
===================================================================
--- libgcc/config/i386/t-cpuinfo	(revision 0)
+++ libgcc/config/i386/t-cpuinfo	(revision 0)
@@ -0,0 +1 @@ 
+LIB2ADD += $(srcdir)/config/i386/i386-cpuinfo.c
Index: libgcc/config/i386/i386-cpuinfo.c
===================================================================
--- libgcc/config/i386/i386-cpuinfo.c	(revision 0)
+++ libgcc/config/i386/i386-cpuinfo.c	(revision 0)
@@ -0,0 +1,245 @@ 
+/* Get CPU type and Features for x86 processors.
+   Copyright (C) 2011 Free Software Foundation, Inc.
+   Contributed by Sriraman Tallam (tmsriram@google.com)
+
+This file is part of GCC.
+
+GCC is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 3, or (at your option) any later
+version.
+
+GCC is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with GCC; see the file COPYING3.  If not see
+<http://www.gnu.org/licenses/>. */
+
+#include "cpuid.h"
+
+void __cpu_indicator_init (void) __attribute__ ((constructor));
+
+enum vendor_signatures
+{
+  SIG_INTEL =	0x756e6547 /* Genu */,
+  SIG_AMD =	0x68747541 /* Auth */
+};
+
+
+/* Features supported. */
+
+struct __processor_features
+{
+  unsigned int __cpu_cmov : 1;
+  unsigned int __cpu_mmx : 1;
+  unsigned int __cpu_popcnt : 1;
+  unsigned int __cpu_sse : 1;
+  unsigned int __cpu_sse2 : 1;
+  unsigned int __cpu_sse3 : 1;
+  unsigned int __cpu_ssse3 : 1;
+  unsigned int __cpu_sse4_1 : 1;
+  unsigned int __cpu_sse4_2 : 1;
+} __cpu_features = {0};
+
+/* Processor Model. */
+
+struct __processor_model
+{
+  unsigned int __cpu_is_amd : 1;
+  unsigned int __cpu_is_intel : 1;
+  unsigned int __cpu_is_intel_atom : 1;
+  unsigned int __cpu_is_intel_core2 : 1;
+  unsigned int __cpu_is_intel_corei7_nehalem : 1;
+  unsigned int __cpu_is_intel_corei7_westmere : 1;
+  unsigned int __cpu_is_intel_corei7_sandybridge : 1;
+  unsigned int __cpu_is_amdfam10_barcelona : 1;
+  unsigned int __cpu_is_amdfam10_shanghai : 1;
+  unsigned int __cpu_is_amdfam10_istanbul : 1;
+} __cpu_model = {0};
+
+/* Get the specific type of AMD CPU.  */
+
+static void
+get_amd_cpu (unsigned int family, unsigned int model)
+{
+  switch (family)
+    {
+    case 0x10:
+      switch (model)
+	{
+	case 0x2:
+	  __cpu_model.__cpu_is_amdfam10_barcelona = 1;
+	  break;
+	case 0x4:
+	  __cpu_model.__cpu_is_amdfam10_shanghai = 1;
+	  break;
+	case 0x8:
+	  __cpu_model.__cpu_is_amdfam10_istanbul = 1;
+	  break;
+	default:
+	  break;
+	}
+      break;
+    default:
+      break;
+    }
+}
+
+/* Get the specific type of Intel CPU.  */
+
+static void
+get_intel_cpu (unsigned int family, unsigned int model, unsigned int brand_id)
+{
+  /* Parse family and model only if brand ID is 0. */
+  if (brand_id == 0)
+    {
+      switch (family)
+	{
+	case 0x5:
+	  /* Pentium.  */
+	  break;
+	case 0x6:
+	  switch (model)
+	    {
+	    case 0x1c:
+	    case 0x26:
+	      /* Atom.  */
+	      __cpu_model.__cpu_is_intel_atom = 1;
+	      break;
+	    case 0x1a:
+	    case 0x1e:
+	    case 0x1f:
+	    case 0x2e:
+	      /* Nehalem.  */
+	      __cpu_model.__cpu_is_intel_corei7_nehalem = 1;
+	      break;
+	    case 0x25:
+	    case 0x2c:
+	    case 0x2f:
+	      /* Westmere.  */
+	      __cpu_model.__cpu_is_intel_corei7_westmere = 1;
+	      break;
+	    case 0x2a:
+	      /* Sandy Bridge.  */
+	      __cpu_model.__cpu_is_intel_corei7_sandybridge = 1;
+	      break;
+	    case 0x17:
+	    case 0x1d:
+	      /* Penryn.  */
+	    case 0x0f:
+	      /* Merom.  */
+	      __cpu_model.__cpu_is_intel_core2 = 1;
+	      break;
+	    default:
+	      break;
+	    }
+	  break;
+	default:
+	  /* We have no idea.  */
+	  break;
+	}
+    }
+}	             	
+
+static void
+get_available_features (unsigned int ecx, unsigned int edx)
+{
+  __cpu_features.__cpu_cmov = (edx & bit_CMOV) ? 1 : 0;
+  __cpu_features.__cpu_mmx = (edx & bit_MMX) ? 1 : 0;
+  __cpu_features.__cpu_sse = (edx & bit_SSE) ? 1 : 0;
+  __cpu_features.__cpu_sse2 = (edx & bit_SSE2) ? 1 : 0;
+  __cpu_features.__cpu_popcnt = (ecx & bit_POPCNT) ? 1 : 0;
+  __cpu_features.__cpu_sse3 = (ecx & bit_SSE3) ? 1 : 0;
+  __cpu_features.__cpu_ssse3 = (ecx & bit_SSSE3) ? 1 : 0;
+  __cpu_features.__cpu_sse4_1 = (ecx & bit_SSE4_1) ? 1 : 0;
+  __cpu_features.__cpu_sse4_2 = (ecx & bit_SSE4_2) ? 1 : 0;
+}
+
+/* A noinline function calling __get_cpuid. Having many calls to
+   cpuid in one function in 32-bit mode causes GCC to complain:
+   "can’t find a register in class ‘CLOBBERED_REGS’".  This is
+   related to PR rtl-optimization 44174. */
+
+static int __attribute__ ((noinline))
+__get_cpuid_output (unsigned int __level,
+		    unsigned int *__eax, unsigned int *__ebx,
+		    unsigned int *__ecx, unsigned int *__edx)
+{
+  return __get_cpuid (__level, __eax, __ebx, __ecx, __edx);
+}
+
+
+/* A constructor function that is sets __cpu_model and __cpu_features with
+   the right values.  This needs to run only once.
+   If another constructor needs to use these values, explicitly call this
+   function from the other constructor.  Otherwise, the ordering of
+   constructors could make this constructor run later.  */
+
+void __attribute__ ((constructor))
+__cpu_indicator_init (void)
+{
+  unsigned int eax, ebx, ecx, edx;
+
+  int max_level = 5;
+  unsigned int vendor;
+  unsigned int model, family, brand_id;
+  static int called = 0;
+
+  /* This function needs to run just once.  */
+  if (called)
+    return;
+  else
+    called = 1;
+
+  /* Assume cpuid insn present. Run in level 0 to get vendor id. */
+  if (!__get_cpuid_output (0, &eax, &ebx, &ecx, &edx))
+    return;
+
+  vendor = ebx;
+  max_level = eax;
+
+  if (max_level < 1)
+    return;
+
+  if (!__get_cpuid_output (1, &eax, &ebx, &ecx, &edx))
+    return;
+
+  model = (eax >> 4) & 0x0f;
+  family = (eax >> 8) & 0x0f;
+  brand_id = ebx & 0xff;
+
+  /* Adjust model and family for Intel CPUS. */
+  if (vendor == SIG_INTEL)
+    {
+      unsigned int extended_model, extended_family;
+
+      extended_model = (eax >> 12) & 0xf0;
+      extended_family = (eax >> 20) & 0xff;
+      if (family == 0x0f)
+	{
+	  family += extended_family;
+	  model += extended_model;
+	}
+      else if (family == 0x06)
+	model += extended_model;
+    }
+
+  /* Find CPU model. */
+
+  if (vendor == SIG_AMD)
+    {
+      __cpu_model.__cpu_is_amd = 1;
+      get_amd_cpu (family, model);
+    }
+  else if (vendor == SIG_INTEL)
+    {
+      __cpu_model.__cpu_is_intel = 1;
+      get_intel_cpu (family, model, brand_id);
+    }
+
+  /* Find available features. */
+  get_available_features (ecx, edx);
+}
Index: gcc/testsuite/gcc.dg/builtin_target.c
===================================================================
--- gcc/testsuite/gcc.dg/builtin_target.c	(revision 0)
+++ gcc/testsuite/gcc.dg/builtin_target.c	(revision 0)
@@ -0,0 +1,53 @@ 
+/* This test checks if the __builtin_cpu_* calls are recognized. */
+
+/* { dg-do run } */
+
+int
+fn1 ()
+{
+  if (__builtin_cpu_supports_cmov () < 0)
+    return -1;
+  if (__builtin_cpu_supports_mmx () < 0)
+    return -1;
+  if (__builtin_cpu_supports_popcount () < 0)
+    return -1;
+  if (__builtin_cpu_supports_sse () < 0)
+    return -1;
+  if (__builtin_cpu_supports_sse2 () < 0)
+    return -1;
+  if (__builtin_cpu_supports_sse3 () < 0)
+    return -1;
+  if (__builtin_cpu_supports_ssse3 () < 0)
+    return -1;
+  if (__builtin_cpu_supports_sse4_1 () < 0)
+    return -1;
+  if (__builtin_cpu_supports_sse4_2 () < 0)
+    return -1;
+  if (__builtin_cpu_is_amd () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel_atom () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel_core2 () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel_corei7_nehalem () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel_corei7_westmere () < 0)
+    return -1;
+  if (__builtin_cpu_is_intel_corei7_sandybridge () < 0)
+    return -1;
+  if (__builtin_cpu_is_amdfam10_barcelona () < 0)
+    return -1;
+  if (__builtin_cpu_is_amdfam10_shanghai () < 0)
+    return -1;
+  if (__builtin_cpu_is_amdfam10_istanbul () < 0)
+    return -1;
+
+  return 0;
+}
+
+int main ()
+{
+  return fn1 ();
+}
Index: gcc/config/i386/i386-builtin-types.def
===================================================================
--- gcc/config/i386/i386-builtin-types.def	(revision 177767)
+++ gcc/config/i386/i386-builtin-types.def	(working copy)
@@ -131,6 +131,7 @@  DEF_FUNCTION_TYPE (UINT64)
 DEF_FUNCTION_TYPE (UNSIGNED)
 DEF_FUNCTION_TYPE (VOID)
 DEF_FUNCTION_TYPE (PVOID)
+DEF_FUNCTION_TYPE (INT)
 
 DEF_FUNCTION_TYPE (FLOAT, FLOAT)
 DEF_FUNCTION_TYPE (FLOAT128, FLOAT128)
Index: gcc/config/i386/libgcc-glibc.ver
===================================================================
--- gcc/config/i386/libgcc-glibc.ver	(revision 177767)
+++ gcc/config/i386/libgcc-glibc.ver	(working copy)
@@ -147,6 +147,12 @@  GCC_4.3.0 {
   __trunctfxf2
   __unordtf2
 }
+
+GCC_4.6.0 {
+  __cpu_indicator_init
+  __cpu_model
+  __cpu_features
+}
 %else
 GCC_4.4.0 {
   __addtf3
@@ -183,4 +189,10 @@  GCC_4.4.0 {
 GCC_4.5.0 {
   __extendxftf2
 }
+
+GCC_4.6.0 {
+  __cpu_indicator_init
+  __cpu_model
+  __cpu_features
+}
 %endif
Index: gcc/config/i386/i386.c
===================================================================
--- gcc/config/i386/i386.c	(revision 177767)
+++ gcc/config/i386/i386.c	(working copy)
@@ -58,6 +58,8 @@  along with GCC; see the file COPYING3.  If not see
 #include "sched-int.h"
 #include "sbitmap.h"
 #include "fibheap.h"
+#include "tree-flow.h"
+#include "tree-pass.h"
 
 enum upper_128bits_state
 {
@@ -24443,6 +24445,28 @@  enum ix86_builtins
   /* CFString built-in for darwin */
   IX86_BUILTIN_CFSTRING,
 
+  /* Builtins to get CPU features. */
+  IX86_BUILTIN_CPU_SUPPORTS_CMOV,
+  IX86_BUILTIN_CPU_SUPPORTS_MMX,
+  IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT,
+  IX86_BUILTIN_CPU_SUPPORTS_SSE,
+  IX86_BUILTIN_CPU_SUPPORTS_SSE2,
+  IX86_BUILTIN_CPU_SUPPORTS_SSE3,
+  IX86_BUILTIN_CPU_SUPPORTS_SSSE3,
+  IX86_BUILTIN_CPU_SUPPORTS_SSE4_1,
+  IX86_BUILTIN_CPU_SUPPORTS_SSE4_2,
+  /* Builtins to get CPU type. */
+  IX86_BUILTIN_CPU_IS_AMD,
+  IX86_BUILTIN_CPU_IS_INTEL,
+  IX86_BUILTIN_CPU_IS_INTEL_ATOM,
+  IX86_BUILTIN_CPU_IS_INTEL_CORE2,
+  IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM,
+  IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE,
+  IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE,
+  IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA,
+  IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI,
+  IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL,
+
   IX86_BUILTIN_MAX
 };
 
@@ -25809,6 +25833,316 @@  ix86_init_mmx_sse_builtins (void)
     }
 }
 
+/* Returns a struct type with name NAME and number of fields equal to
+   NUM_FIELDS.  Each field is a unsigned int bit field of length 1 bit. */
+
+static tree
+build_struct_with_one_bit_fields (int num_fields, const char *name)
+{
+  int i;
+  char field_name [10];
+  tree field = NULL_TREE, field_chain = NULL_TREE;
+  tree type = make_node (RECORD_TYPE);
+
+  strcpy (field_name, "k_field");
+
+  for (i = 0; i < num_fields; i++)
+    {
+      /* Name the fields, 0_field, 1_field, ... */
+      field_name [0] = '0' + i;
+      field = build_decl (UNKNOWN_LOCATION, FIELD_DECL,
+			  get_identifier (field_name), unsigned_type_node);
+      DECL_BIT_FIELD (field) = 1;
+      DECL_SIZE (field) = bitsize_one_node;
+      if (field_chain != NULL_TREE)
+	DECL_CHAIN (field) = field_chain;
+      field_chain = field;
+    }
+  finish_builtin_struct (type, name, field_chain, NULL_TREE);
+  return type;
+}
+
+/* Returns a extern, comdat VAR_DECL of type TYPE and name NAME. */
+
+static tree
+make_var_decl (tree type, const char *name)
+{
+  tree new_decl;
+  struct varpool_node *vnode;
+
+  new_decl = build_decl (UNKNOWN_LOCATION,
+	                 VAR_DECL,
+	  	         get_identifier(name),
+		         type);
+
+  DECL_EXTERNAL (new_decl) = 1;
+  TREE_STATIC (new_decl) = 1;
+  TREE_PUBLIC (new_decl) = 1;
+  DECL_INITIAL (new_decl) = 0;
+  DECL_ARTIFICIAL (new_decl) = 0;
+  DECL_PRESERVE_P (new_decl) = 1;
+
+  make_decl_one_only (new_decl, DECL_ASSEMBLER_NAME (new_decl));
+  assemble_variable (new_decl, 0, 0, 0);
+
+  vnode = varpool_node (new_decl);
+  gcc_assert (vnode != NULL);
+  /* Set finalized to 1, otherwise it asserts in function "write_symbol" in
+     lto-streamer-out.c. */
+  vnode->finalized = 1;
+
+  return new_decl;
+}
+
+/* Traverses the chain of fields in STRUCT_TYPE and returns the FIELD_NUM
+   numbered field. */
+
+static tree
+get_field_from_struct (tree struct_type, int field_num)
+{
+  int i;
+  tree field = TYPE_FIELDS (struct_type);
+
+  for (i = 0; i < field_num; i++, field = DECL_CHAIN(field))
+    {
+      gcc_assert (field != NULL_TREE);
+    }
+
+  return field;
+}
+
+/* FNDECL is a __builtin_cpu_* call that is folded into an integer defined
+   in libgcc/config/i386/i386-cpuinfo.c */
+
+static tree 
+fold_builtin_cpu (enum ix86_builtins fn_code)
+{
+  /* This is the order of bit-fields in __processor_features in
+     i386-cpuinfo.c */
+  enum processor_features
+  {
+    F_CMOV = 0,
+    F_MMX,
+    F_POPCNT,
+    F_SSE,
+    F_SSE2,
+    F_SSE3,
+    F_SSSE3,
+    F_SSE4_1,
+    F_SSE4_2,
+    F_MAX
+  };
+
+  /* This is the order of bit-fields in __processor_model in
+     i386-cpuinfo.c */
+  enum processor_model
+  {
+    M_AMD = 0,
+    M_INTEL,
+    M_INTEL_ATOM,
+    M_INTEL_CORE2,
+    M_INTEL_COREI7_NEHALEM,
+    M_INTEL_COREI7_WESTMERE,
+    M_INTEL_COREI7_SANDYBRIDGE,
+    M_AMDFAM10_BARCELONA,
+    M_AMDFAM10_SHANGHAI,
+    M_AMDFAM10_ISTANBUL,
+    M_MAX
+  };
+
+  static tree __processor_features_type = NULL_TREE;
+  static tree __cpu_features_var = NULL_TREE;
+  static tree __processor_model_type = NULL_TREE;
+  static tree __cpu_model_var = NULL_TREE;
+  static tree field;
+  static tree which_struct;
+
+  if (__processor_features_type == NULL_TREE)
+    __processor_features_type = build_struct_with_one_bit_fields (F_MAX,
+ 			          "__processor_features");
+
+  if (__processor_model_type == NULL_TREE)
+    __processor_model_type = build_struct_with_one_bit_fields (M_MAX,
+ 			          "__processor_model");
+
+  if (__cpu_features_var == NULL_TREE)
+    __cpu_features_var = make_var_decl (__processor_features_type,
+					"__cpu_features");
+
+  if (__cpu_model_var == NULL_TREE)
+    __cpu_model_var = make_var_decl (__processor_model_type,
+				     "__cpu_model");
+
+  /* Look at the code to identify the field requested. */ 
+  switch (fn_code)
+    {
+    case IX86_BUILTIN_CPU_SUPPORTS_CMOV:
+      field = get_field_from_struct (__processor_features_type, F_CMOV);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_MMX:
+      field = get_field_from_struct (__processor_features_type, F_MMX);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT:
+      field = get_field_from_struct (__processor_features_type, F_POPCNT);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE:
+      field = get_field_from_struct (__processor_features_type, F_SSE);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE2:
+      field = get_field_from_struct (__processor_features_type, F_SSE2);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE3:
+      field = get_field_from_struct (__processor_features_type, F_SSE3);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSSE3:
+      field = get_field_from_struct (__processor_features_type, F_SSE3);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE4_1:
+      field = get_field_from_struct (__processor_features_type, F_SSE4_1);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE4_2:
+      field = get_field_from_struct (__processor_features_type, F_SSE4_2);
+      which_struct = __cpu_features_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_AMD:
+      field = get_field_from_struct (__processor_model_type, M_AMD);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL:
+      field = get_field_from_struct (__processor_model_type, M_INTEL);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL_ATOM:
+      field = get_field_from_struct (__processor_model_type, M_INTEL_ATOM);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL_CORE2:
+      field = get_field_from_struct (__processor_model_type, M_INTEL_CORE2);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM:
+      field = get_field_from_struct (__processor_model_type,
+				     M_INTEL_COREI7_NEHALEM);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE:
+      field = get_field_from_struct (__processor_model_type,
+				     M_INTEL_COREI7_WESTMERE);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE:
+      field = get_field_from_struct (__processor_model_type,
+				     M_INTEL_COREI7_SANDYBRIDGE);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA:
+      field = get_field_from_struct (__processor_model_type,
+				     M_AMDFAM10_BARCELONA);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI:
+      field = get_field_from_struct (__processor_model_type,
+				     M_AMDFAM10_SHANGHAI);
+      which_struct = __cpu_model_var;
+      break;
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL:
+      field = get_field_from_struct (__processor_model_type,
+				     M_AMDFAM10_ISTANBUL);
+      which_struct = __cpu_model_var;
+      break;
+    default:
+      return NULL_TREE;
+    }
+
+  return build3 (COMPONENT_REF, TREE_TYPE (field), which_struct, field, NULL_TREE);
+}
+
+static tree
+ix86_fold_builtin (tree fndecl, int n_args ATTRIBUTE_UNUSED,
+		   tree *args ATTRIBUTE_UNUSED, bool ignore ATTRIBUTE_UNUSED)
+{
+  const char* decl_name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
+  if (DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_MD
+      && strstr(decl_name, "__builtin_cpu") != NULL)
+    {
+      enum ix86_builtins code = (enum ix86_builtins)
+				DECL_FUNCTION_CODE (fndecl);
+      return fold_builtin_cpu (code);
+    }
+  return NULL_TREE;
+}
+
+/* A builtin to return the cpu type or feature.  Returns an integer
+   and is a const. */
+
+static void
+make_platform_builtin (const char* name, int code)
+{
+  tree decl;
+  tree type;
+
+  type = ix86_get_builtin_func_type (INT_FTYPE_VOID);
+  decl = add_builtin_function (name, type, code, BUILT_IN_MD,
+			       NULL, NULL_TREE);
+  gcc_assert (decl != NULL_TREE);
+  ix86_builtins[(int) code] = decl;
+  /* Mark this as a const function. */
+  TREE_READONLY (decl) = 1;
+} 
+
+/* Builtins to get CPU type and features supported. */
+
+static void
+ix86_init_platform_type_builtins (void)
+{
+  make_platform_builtin ("__builtin_cpu_supports_cmov",
+			    IX86_BUILTIN_CPU_SUPPORTS_CMOV);
+  make_platform_builtin ("__builtin_cpu_supports_mmx",
+			    IX86_BUILTIN_CPU_SUPPORTS_MMX);
+  make_platform_builtin ("__builtin_cpu_supports_popcount",
+			    IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT);
+  make_platform_builtin ("__builtin_cpu_supports_sse",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSE);
+  make_platform_builtin ("__builtin_cpu_supports_sse2",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSE2);
+  make_platform_builtin ("__builtin_cpu_supports_sse3",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSE3);
+  make_platform_builtin ("__builtin_cpu_supports_ssse3",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSSE3);
+  make_platform_builtin ("__builtin_cpu_supports_sse4_1",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSE4_1);
+  make_platform_builtin ("__builtin_cpu_supports_sse4_2",
+			    IX86_BUILTIN_CPU_SUPPORTS_SSE4_2);
+  make_platform_builtin ("__builtin_cpu_is_amd",
+			    IX86_BUILTIN_CPU_IS_AMD);
+  make_platform_builtin ("__builtin_cpu_is_intel_atom",
+			    IX86_BUILTIN_CPU_IS_INTEL_ATOM);
+  make_platform_builtin ("__builtin_cpu_is_intel_core2",
+			    IX86_BUILTIN_CPU_IS_INTEL_CORE2);
+  make_platform_builtin ("__builtin_cpu_is_intel",
+			    IX86_BUILTIN_CPU_IS_INTEL);
+  make_platform_builtin ("__builtin_cpu_is_intel_corei7_nehalem",
+			    IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM);
+  make_platform_builtin ("__builtin_cpu_is_intel_corei7_westmere",
+			    IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE);
+  make_platform_builtin ("__builtin_cpu_is_intel_corei7_sandybridge",
+			    IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE);
+  make_platform_builtin ("__builtin_cpu_is_amdfam10_barcelona",
+			    IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA);
+  make_platform_builtin ("__builtin_cpu_is_amdfam10_shanghai",
+			    IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI);
+  make_platform_builtin ("__builtin_cpu_is_amdfam10_istanbul",
+			    IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL);
+}
+
 /* Internal method for ix86_init_builtins.  */
 
 static void
@@ -25892,6 +26226,9 @@  ix86_init_builtins (void)
 
   ix86_init_builtin_types ();
 
+  /* Builtins to get CPU type and features. */
+  ix86_init_platform_type_builtins ();
+
   /* TFmode support builtins.  */
   def_builtin_const (0, "__builtin_infq",
 		     FLOAT128_FTYPE_VOID, IX86_BUILTIN_INFQ);
@@ -27351,6 +27688,35 @@  ix86_expand_builtin (tree exp, rtx target, rtx sub
   enum machine_mode mode0, mode1, mode2;
   unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
 
+  /* For CPU builtins that can be folded, fold first and expand the fold.  */
+  switch (fcode)
+    {
+    case IX86_BUILTIN_CPU_SUPPORTS_CMOV:
+    case IX86_BUILTIN_CPU_SUPPORTS_MMX:
+    case IX86_BUILTIN_CPU_SUPPORTS_POPCOUNT:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE2:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE3:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSSE3:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE4_1:
+    case IX86_BUILTIN_CPU_SUPPORTS_SSE4_2:
+    case IX86_BUILTIN_CPU_IS_AMD:
+    case IX86_BUILTIN_CPU_IS_INTEL:
+    case IX86_BUILTIN_CPU_IS_INTEL_ATOM:
+    case IX86_BUILTIN_CPU_IS_INTEL_CORE2:
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_NEHALEM:
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_WESTMERE:
+    case IX86_BUILTIN_CPU_IS_INTEL_COREI7_SANDYBRIDGE:
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_BARCELONA:
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_SHANGHAI:
+    case IX86_BUILTIN_CPU_IS_AMDFAM10_ISTANBUL:
+      {
+        tree fold_expr = fold_builtin_cpu ((enum ix86_builtins) fcode);
+	gcc_assert (fold_expr != NULL_TREE);
+        return expand_expr (fold_expr, target, mode, EXPAND_NORMAL);
+      }
+    }
+
   /* Determine whether the builtin function is available under the current ISA.
      Originally the builtin was not created if it wasn't applicable to the
      current ISA based on the command line switches.  With function specific
@@ -35097,6 +35463,9 @@  ix86_autovectorize_vector_sizes (void)
 #undef TARGET_BUILD_BUILTIN_VA_LIST
 #define TARGET_BUILD_BUILTIN_VA_LIST ix86_build_builtin_va_list
 
+#undef TARGET_FOLD_BUILTIN
+#define TARGET_FOLD_BUILTIN ix86_fold_builtin
+
 #undef TARGET_ENUM_VA_LIST_P
 #define TARGET_ENUM_VA_LIST_P ix86_enum_va_list