Fix PR67639

Fix PR67639

[Bernd Schmidt]

[-- Attachment #1: Type: text/plain, Size: 847 bytes --]

In an earlier fix, the following change was made in varasm.c for invalid 
register variables:

--- trunk/gcc/varasm.c	2014/08/26 14:59:59	214525
+++ trunk/gcc/varasm.c	2014/08/26 17:06:31	214526
@@ -1371,6 +1371,11 @@ make_decl_rtl (tree decl)
  	  /* As a register variable, it has no section.  */
  	  return;
  	}
+      /* Avoid internal errors from invalid register
+	 specifications.  */
+      SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
+      DECL_HARD_REGISTER (decl) = 0;
+      return;
      }

As seen in PR67639, this makes the IL inconsistent and triggers another 
internal error where we expect to see an SSA_NAME instead of a VAR_DECL.

The following patch extends the above slightly, by also setting 
DECL_EXTERNAL to pretend that the erroneous variable is actually a global.

Bootstrapped and tested on x86_64-linux, ok?


Bernd

[-- Attachment #2: 67639.diff --]
[-- Type: text/x-patch, Size: 1138 bytes --]

	PR middle-end/67639
	* varasm.c (make_decl_rtl): Mark invalid register vars as
	DECL_EXTERNAL.

testsuite/
	PR middle-end/67639
	* c-c++-common/pr67639.c: New test.

Index: gcc/testsuite/c-c++-common/pr67639.c
===================================================================
--- gcc/testsuite/c-c++-common/pr67639.c	(revision 0)
+++ gcc/testsuite/c-c++-common/pr67639.c	(working copy)
@@ -0,0 +1,8 @@
+/* { dg-do compile } */
+void
+foo (int p)
+{
+  int t; 
+  register long x asm ("rhubarb") = p; /* { dg-error "register name" } */
+  __asm ("" : "=r" (t), "=r" (t), "=r" (t), "=r" (x) : "0" (x));
+}
Index: gcc/varasm.c
===================================================================
--- gcc/varasm.c	(revision 231653)
+++ gcc/varasm.c	(working copy)
@@ -1420,6 +1420,9 @@ make_decl_rtl (tree decl)
 	 specifications.  */
       SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
       DECL_HARD_REGISTER (decl) = 0;
+      /* Also avoid SSA inconsistencies by pretending this is an external
+	 decl now.  */
+      DECL_EXTERNAL (decl) = 1;
       return;
     }
   /* Now handle ordinary static variables and functions (in memory).

Re: Fix PR67639

[Jeff Law]

On 12/18/2015 11:38 AM, Bernd Schmidt wrote:
> In an earlier fix, the following change was made in varasm.c for invalid
> register variables:
>
> --- trunk/gcc/varasm.c    2014/08/26 14:59:59    214525
> +++ trunk/gcc/varasm.c    2014/08/26 17:06:31    214526
> @@ -1371,6 +1371,11 @@ make_decl_rtl (tree decl)
>         /* As a register variable, it has no section.  */
>         return;
>       }
> +      /* Avoid internal errors from invalid register
> +     specifications.  */
> +      SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
> +      DECL_HARD_REGISTER (decl) = 0;
> +      return;
>       }
>
> As seen in PR67639, this makes the IL inconsistent and triggers another
> internal error where we expect to see an SSA_NAME instead of a VAR_DECL.
>
> The following patch extends the above slightly, by also setting
> DECL_EXTERNAL to pretend that the erroneous variable is actually a global.
>
> Bootstrapped and tested on x86_64-linux, ok?
OK.
jeff

Re: Fix PR67639

[Bernd Schmidt]

On 12/21/2015 08:39 PM, Jeff Law wrote:
> On 12/18/2015 11:38 AM, Bernd Schmidt wrote:
>> In an earlier fix, the following change was made in varasm.c for invalid
>> register variables:
>>
>> --- trunk/gcc/varasm.c    2014/08/26 14:59:59    214525
>> +++ trunk/gcc/varasm.c    2014/08/26 17:06:31    214526
>> @@ -1371,6 +1371,11 @@ make_decl_rtl (tree decl)
>>         /* As a register variable, it has no section.  */
>>         return;
>>       }
>> +      /* Avoid internal errors from invalid register
>> +     specifications.  */
>> +      SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
>> +      DECL_HARD_REGISTER (decl) = 0;
>> +      return;
>>       }
>>
>> As seen in PR67639, this makes the IL inconsistent and triggers another
>> internal error where we expect to see an SSA_NAME instead of a VAR_DECL.
>>
>> The following patch extends the above slightly, by also setting
>> DECL_EXTERNAL to pretend that the erroneous variable is actually a
>> global.
>>
>> Bootstrapped and tested on x86_64-linux, ok?
> OK.

Turns out 65702 is a dup and this should go into gcc-5 as well. Ok to 
backport?


Bernd

Re: Fix PR67639

[Jeff Law]

On 02/08/2016 07:26 AM, Bernd Schmidt wrote:
> On 12/21/2015 08:39 PM, Jeff Law wrote:
>> On 12/18/2015 11:38 AM, Bernd Schmidt wrote:
>>> In an earlier fix, the following change was made in varasm.c for invalid
>>> register variables:
>>>
>>> --- trunk/gcc/varasm.c    2014/08/26 14:59:59    214525
>>> +++ trunk/gcc/varasm.c    2014/08/26 17:06:31    214526
>>> @@ -1371,6 +1371,11 @@ make_decl_rtl (tree decl)
>>>         /* As a register variable, it has no section.  */
>>>         return;
>>>       }
>>> +      /* Avoid internal errors from invalid register
>>> +     specifications.  */
>>> +      SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
>>> +      DECL_HARD_REGISTER (decl) = 0;
>>> +      return;
>>>       }
>>>
>>> As seen in PR67639, this makes the IL inconsistent and triggers another
>>> internal error where we expect to see an SSA_NAME instead of a VAR_DECL.
>>>
>>> The following patch extends the above slightly, by also setting
>>> DECL_EXTERNAL to pretend that the erroneous variable is actually a
>>> global.
>>>
>>> Bootstrapped and tested on x86_64-linux, ok?
>> OK.
>
> Turns out 65702 is a dup and this should go into gcc-5 as well. Ok to
> backport?
Yes.

Thanks,
jeff

Re: Fix PR67639

[Matthias Klose]

On 08.02.2016 15:26, Bernd Schmidt wrote:
> On 12/21/2015 08:39 PM, Jeff Law wrote:
>> On 12/18/2015 11:38 AM, Bernd Schmidt wrote:
>>> In an earlier fix, the following change was made in varasm.c for invalid
>>> register variables:
>>>
>>> --- trunk/gcc/varasm.c    2014/08/26 14:59:59    214525
>>> +++ trunk/gcc/varasm.c    2014/08/26 17:06:31    214526
>>> @@ -1371,6 +1371,11 @@ make_decl_rtl (tree decl)
>>>         /* As a register variable, it has no section.  */
>>>         return;
>>>       }
>>> +      /* Avoid internal errors from invalid register
>>> +     specifications.  */
>>> +      SET_DECL_ASSEMBLER_NAME (decl, NULL_TREE);
>>> +      DECL_HARD_REGISTER (decl) = 0;
>>> +      return;
>>>       }
>>>
>>> As seen in PR67639, this makes the IL inconsistent and triggers another
>>> internal error where we expect to see an SSA_NAME instead of a VAR_DECL.
>>>
>>> The following patch extends the above slightly, by also setting
>>> DECL_EXTERNAL to pretend that the erroneous variable is actually a
>>> global.
>>>
>>> Bootstrapped and tested on x86_64-linux, ok?
>> OK.
>
> Turns out 65702 is a dup and this should go into gcc-5 as well. Ok to backport?

ChangeLog entry is not backported.