Re: Fix more of C/fortran canonical type issues

[Richard Biener <rguenther at suse dot de>]

On Mon, 8 Jun 2015, Jan Hubicka wrote:

> Hi,
> this is a variant of patch that globs also the rest of integer types.
> Note that we will still get false warnings out of lto-symtab when the
> values are not wrapped up in structures.  This is because lto-symtab
> uses types_compatible_p that in turn uses useless_type_conversion and
> that one needs to honor signedness.
> 
> I suppose we need a way to test representation compatibility and TBAA
> compatiblity. I will give it a more tought how to reorganize the code.
> Basically we need

representation compatibility is TYPE_CANONICAL equivalence, TBAA
compatibility is get_alias_set equivalence.

So you have to be careful when mangling TYPE_CANONICAL according to
get_alias_set and make sure to only apply this (signedness for example)
for aggregate type components.

Can you please split out the string-flag change?  It is approved.

I'm not sure the C standard mandates compatibility between

struct { int i; } and struct { unsigned i; } for purposes of TBAA.
Joseph?

Thanks,
Richard.

>   - way to decide if two types have compatible memory representation 
>     (to test in lto-symtab and for some cases in ipa-icf (contructors/pure
>      moves))
>     operands_equal_p/compare_constant/ipa-icf::sem_variable all implements
>     bit of this. copmare_constant seems to be most complete.
>   - way to decide if two types match by TBAA oracle
>     (to test in lto-symtab merging and for ipa-icf memory operations)
>   - way to decide if one type is semantically compatible to other
>     (useless_type_conversion_p)
>   - way to decide if two types are same for canonical type computation
>     (gimple_type_compatible_p).  This may be sensitive to the set of languages
>     we are merging and enable/disable various globbing as required.
> 
> It may make sense to refactor the type walkers and get this more organized.
> But before playing with this I think we want to get something conservatively
> correct according to language standards and get a reasonable body of testcases.
> 
> This is a variant of patch that removes TYPE_UNSIGNED testing completely.
> I am fine with both variants.
> 
> Bootstrapped/regtested ppc64le-linux.
> 
> Honza
> 
> 	* gimple-expr.c (useless_type_conversion_p): Move
> 	INTEGER_TYPE handling ahead.
> 	* tree.c (gimple_canonical_types_compatible_p): Do not compare
> 	TYPE_UNSIGNED for size_t and char compatible types; do not hash
> 	STRING_FLAG on integer types.
> 
> 	* lto.c (hash_canonical_type): Do not hash TYPE_UNSIGNED for size_t
> 	and char compatible types; do not hash STRING_FLAG on integer types.
> 
> 	* gfortran.dg/lto/bind_c-2_0.f90: New testcase.
> 	* gfortran.dg/lto/bind_c-2_1.c: New testcase.
> 	* gfortran.dg/lto/bind_c-3_0.f90: New testcase.
> 	* gfortran.dg/lto/bind_c-3_1.c: New testcase.
> 	* gfortran.dg/lto/bind_c-4_0.f90: New testcase.
> 	* gfortran.dg/lto/bind_c-4_1.c: New testcase.
> Index: gimple-expr.c
> ===================================================================
> --- gimple-expr.c	(revision 224201)
> +++ gimple-expr.c	(working copy)
> @@ -91,30 +91,14 @@
>  	       || TREE_CODE (TREE_TYPE (inner_type)) == METHOD_TYPE))
>  	return false;
>      }
> -
> -  /* From now on qualifiers on value types do not matter.  */
> -  inner_type = TYPE_MAIN_VARIANT (inner_type);
> -  outer_type = TYPE_MAIN_VARIANT (outer_type);
> -
> -  if (inner_type == outer_type)
> -    return true;
> -
> -  /* If we know the canonical types, compare them.  */
> -  if (TYPE_CANONICAL (inner_type)
> -      && TYPE_CANONICAL (inner_type) == TYPE_CANONICAL (outer_type))
> -    return true;
> -
> -  /* Changes in machine mode are never useless conversions unless we
> -     deal with aggregate types in which case we defer to later checks.  */
> -  if (TYPE_MODE (inner_type) != TYPE_MODE (outer_type)
> -      && !AGGREGATE_TYPE_P (inner_type))
> -    return false;
> -
>    /* If both the inner and outer types are integral types, then the
>       conversion is not necessary if they have the same mode and
> -     signedness and precision, and both or neither are boolean.  */
> -  if (INTEGRAL_TYPE_P (inner_type)
> -      && INTEGRAL_TYPE_P (outer_type))
> +     signedness and precision, and both or neither are boolean.
> +
> +     Do not rely on TYPE_CANONICAL here because LTO puts same canonical
> +     type for signed char and unsigned char.  */
> +  else if (INTEGRAL_TYPE_P (inner_type)
> +           && INTEGRAL_TYPE_P (outer_type))
>      {
>        /* Preserve changes in signedness or precision.  */
>        if (TYPE_UNSIGNED (inner_type) != TYPE_UNSIGNED (outer_type)
> @@ -134,6 +118,25 @@
>        return true;
>      }
>  
> +
> +  /* From now on qualifiers on value types do not matter.  */
> +  inner_type = TYPE_MAIN_VARIANT (inner_type);
> +  outer_type = TYPE_MAIN_VARIANT (outer_type);
> +
> +  if (inner_type == outer_type)
> +    return true;
> +
> +  /* If we know the canonical types, compare them.  */
> +  if (TYPE_CANONICAL (inner_type)
> +      && TYPE_CANONICAL (inner_type) == TYPE_CANONICAL (outer_type))
> +    return true;
> +
> +  /* Changes in machine mode are never useless conversions unless we
> +     deal with aggregate types in which case we defer to later checks.  */
> +  if (TYPE_MODE (inner_type) != TYPE_MODE (outer_type)
> +      && !AGGREGATE_TYPE_P (inner_type))
> +    return false;
> +
>    /* Scalar floating point types with the same mode are compatible.  */
>    else if (SCALAR_FLOAT_TYPE_P (inner_type)
>  	   && SCALAR_FLOAT_TYPE_P (outer_type))
> Index: lto/lto.c
> ===================================================================
> --- lto/lto.c	(revision 224201)
> +++ lto/lto.c	(working copy)
> @@ -303,6 +303,7 @@
>  hash_canonical_type (tree type)
>  {
>    inchash::hash hstate;
> +  enum tree_code code;
>  
>    /* We compute alias sets only for types that needs them.
>       Be sure we do not recurse to something else as we can not hash incomplete
> @@ -314,7 +315,8 @@
>       smaller sets; when searching for existing matching types to merge,
>       only existing types having the same features as the new type will be
>       checked.  */
> -  hstate.add_int (tree_code_for_canonical_type_merging (TREE_CODE (type)));
> +  code = tree_code_for_canonical_type_merging (TREE_CODE (type));
> +  hstate.add_int (code);
>    hstate.add_int (TYPE_MODE (type));
>  
>    /* Incorporate common features of numerical types.  */
> @@ -324,8 +326,13 @@
>        || TREE_CODE (type) == OFFSET_TYPE
>        || POINTER_TYPE_P (type))
>      {
> -      hstate.add_int (TYPE_UNSIGNED (type));
>        hstate.add_int (TYPE_PRECISION (type));
> +      /* Ignore TYPE_SIGNED. This is needed for Fortran
> +	 C_SIGNED_CHAR to be interoperable with both signed char and
> +	 unsigned char (as stadnard requires).  Similarly Fortran FE builds
> +	 C_SIZE_T is signed type, while C defines it unsigned.
> +	 NOTE 15.8 of Fortran 2008 seems to imply that even other types ought
> +	 to be inter-operable. */
>      }
>  
>    if (VECTOR_TYPE_P (type))
> @@ -345,9 +352,9 @@
>        hstate.add_int (TYPE_ADDR_SPACE (TREE_TYPE (type)));
>      }
>  
> -  /* For integer types hash only the string flag.  */
> -  if (TREE_CODE (type) == INTEGER_TYPE)
> -    hstate.add_int (TYPE_STRING_FLAG (type));
> +  /* Fortran FE does not set string flag for C_SIGNED_CHAR while C
> +     sets it for signed char.  To use STRING_FLAG to define canonical types,
> +     the frontends would need to agree.  */
>  
>    /* For array types hash the domain bounds and the string flag.  */
>    if (TREE_CODE (type) == ARRAY_TYPE && TYPE_DOMAIN (type))
> Index: testsuite/gfortran.dg/lto/bind_c-2_0.f90
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-2_0.f90	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-2_0.f90	(working copy)
> @@ -0,0 +1,21 @@
> +! { dg-lto-do run }
> +! { dg-lto-options {{ -O3 -flto }} }
> +! This testcase will abort if C_PTR is not interoperable with both int *
> +! and float *
> +module lto_type_merge_test
> +  use, intrinsic :: iso_c_binding
> +  implicit none
> +
> +  type, bind(c) :: MYFTYPE_1
> +     integer(c_signed_char) :: chr
> +     integer(c_signed_char) :: chrb
> +  end type MYFTYPE_1
> +
> +  type(myftype_1), bind(c, name="myVar") :: myVar
> +
> +contains
> +  subroutine types_test() bind(c)
> +    myVar%chr = myVar%chrb
> +  end subroutine types_test
> +end module lto_type_merge_test
> +
> Index: testsuite/gfortran.dg/lto/bind_c-2_1.c
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-2_1.c	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-2_1.c	(working copy)
> @@ -0,0 +1,36 @@
> +#include <stdlib.h>
> +/* interopse with myftype_1 */
> +typedef struct {
> +   unsigned char chr;
> +   signed char chr2;
> +} myctype_t;
> +
> +
> +extern void abort(void);
> +void types_test(void);
> +/* declared in the fortran module */
> +extern myctype_t myVar;
> +
> +int main(int argc, char **argv)
> +{
> +   myctype_t *cchr;
> +   asm("":"=r"(cchr):"0"(&myVar));
> +   cchr->chr = 1;
> +   cchr->chr2 = 2;
> +
> +   types_test();
> +
> +   if(cchr->chr != 2)
> +      abort();
> +   if(cchr->chr2 != 2)
> +      abort();
> +   myVar.chr2 = 3;
> +   types_test();
> +
> +   if(myVar.chr != 3)
> +      abort();
> +   if(myVar.chr2 != 3)
> +      abort();
> +   return 0;
> +}
> +
> Index: testsuite/gfortran.dg/lto/bind_c-3_0.f90
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-3_0.f90	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-3_0.f90	(working copy)
> @@ -0,0 +1,91 @@
> +! { dg-lto-do run }
> +! { dg-lto-options {{ -O3 -flto }} }
> +! This testcase will abort if integer types are not interoperable.
> +module lto_type_merge_test
> +  use, intrinsic :: iso_c_binding
> +  implicit none
> +
> +  type, bind(c) :: MYFTYPE_1
> +    integer(c_int) :: val_int
> +    integer(c_short) :: val_short
> +    integer(c_long) :: val_long
> +    integer(c_long_long) :: val_long_long
> +    integer(c_size_t) :: val_size_t
> +    integer(c_int8_t) :: val_int8_t
> +    integer(c_int16_t) :: val_int16_t
> +    integer(c_int32_t) :: val_int32_t
> +    integer(c_int64_t) :: val_int64_t
> +    integer(c_int_least8_t) :: val_intleast_8_t
> +    integer(c_int_least16_t) :: val_intleast_16_t
> +    integer(c_int_least32_t) :: val_intleast_32_t
> +    integer(c_int_least64_t) :: val_intleast_64_t
> +    integer(c_int_fast8_t) :: val_intfast_8_t
> +    integer(c_int_fast16_t) :: val_intfast_16_t
> +    integer(c_int_fast32_t) :: val_intfast_32_t
> +    integer(c_int_fast64_t) :: val_intfast_64_t
> +    integer(c_intmax_t) :: val_intmax_t
> +    integer(c_intptr_t) :: val_intptr_t
> +  end type MYFTYPE_1
> +
> +  type(myftype_1), bind(c, name="myVar") :: myVar
> +
> +contains
> +  subroutine types_test1() bind(c)
> +    myVar%val_int = 2
> +  end subroutine types_test1
> +  subroutine types_test2() bind(c)
> +    myVar%val_short = 2
> +  end subroutine types_test2
> +  subroutine types_test3() bind(c)
> +    myVar%val_long = 2
> +  end subroutine types_test3
> +  subroutine types_test4() bind(c)
> +    myVar%val_long_long = 2
> +  end subroutine types_test4
> +  subroutine types_test5() bind(c)
> +    myVar%val_size_t = 2
> +  end subroutine types_test5
> +  subroutine types_test6() bind(c)
> +    myVar%val_int8_t = 2
> +  end subroutine types_test6
> +  subroutine types_test7() bind(c)
> +    myVar%val_int16_t = 2
> +  end subroutine types_test7
> +  subroutine types_test8() bind(c)
> +    myVar%val_int32_t = 2
> +  end subroutine types_test8
> +  subroutine types_test9() bind(c)
> +    myVar%val_int64_t = 2
> +  end subroutine types_test9
> +  subroutine types_test10() bind(c)
> +    myVar%val_intleast_8_t = 2
> +  end subroutine types_test10
> +  subroutine types_test11() bind(c)
> +    myVar%val_intleast_16_t = 2
> +  end subroutine types_test11
> +  subroutine types_test12() bind(c)
> +    myVar%val_intleast_32_t = 2
> +  end subroutine types_test12
> +  subroutine types_test13() bind(c)
> +    myVar%val_intleast_64_t = 2
> +  end subroutine types_test13
> +  subroutine types_test14() bind(c)
> +    myVar%val_intfast_8_t = 2
> +  end subroutine types_test14
> +  subroutine types_test15() bind(c)
> +    myVar%val_intfast_16_t = 2
> +  end subroutine types_test15
> +  subroutine types_test16() bind(c)
> +    myVar%val_intfast_32_t = 2
> +  end subroutine types_test16
> +  subroutine types_test17() bind(c)
> +    myVar%val_intfast_64_t = 2
> +  end subroutine types_test17
> +  subroutine types_test18() bind(c)
> +    myVar%val_intmax_t = 2
> +  end subroutine types_test18
> +  subroutine types_test19() bind(c)
> +    myVar%val_intptr_t = 2
> +  end subroutine types_test19
> +end module lto_type_merge_test
> +
> Index: testsuite/gfortran.dg/lto/bind_c-3_1.c
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-3_1.c	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-3_1.c	(working copy)
> @@ -0,0 +1,78 @@
> +#include <stdlib.h>
> +#include <stdint.h>
> +/* interopse with myftype_1 */
> +typedef struct {
> +  int val1;
> +  short int val2;
> +  long int val3;
> +  long long int val4;
> +  size_t val5;
> +  int8_t val6;
> +  int16_t val7;
> +  int32_t val8;
> +  int64_t val9;
> +  int_least8_t val10;
> +  int_least16_t val11;
> +  int_least32_t val12;
> +  int_least64_t val13;
> +  int_fast8_t val14;
> +  int_fast16_t val15;
> +  int_fast32_t val16;
> +  int_fast64_t val17;
> +  intmax_t val18;
> +  intptr_t val19;
> +} myctype_t;
> +
> +
> +extern void abort(void);
> +void types_test1(void);
> +void types_test2(void);
> +void types_test3(void);
> +void types_test4(void);
> +void types_test5(void);
> +void types_test6(void);
> +void types_test7(void);
> +void types_test8(void);
> +void types_test9(void);
> +void types_test10(void);
> +void types_test11(void);
> +void types_test12(void);
> +void types_test13(void);
> +void types_test14(void);
> +void types_test15(void);
> +void types_test16(void);
> +void types_test17(void);
> +void types_test18(void);
> +void types_test19(void);
> +/* declared in the fortran module */
> +extern myctype_t myVar;
> +
> +#define test(n)\
> +  cchr->val##n = 1; types_test##n (); if (cchr->val##n != 2) abort ();
> +
> +int main(int argc, char **argv)
> +{
> +   myctype_t *cchr;
> +   asm("":"=r"(cchr):"0"(&myVar));
> +   test(1);
> +   test(2);
> +   test(3);
> +   test(4);
> +   test(5);
> +   test(6);
> +   test(7);
> +   test(8);
> +   test(9);
> +   test(10);
> +   test(11);
> +   test(12);
> +   test(13);
> +   test(14);
> +   test(15);
> +   test(16);
> +   test(17);
> +   test(18);
> +   test(19);
> +   return 0;
> +}
> +
> Index: testsuite/gfortran.dg/lto/bind_c-4_0.f90
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-4_0.f90	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-4_0.f90	(working copy)
> @@ -0,0 +1,48 @@
> +! { dg-lto-do run }
> +! { dg-lto-options {{ -O3 -flto }} }
> +! This testcase will abort if real/complex/boolean/character types are not interoperable
> +module lto_type_merge_test
> +  use, intrinsic :: iso_c_binding
> +  implicit none
> +
> +  type, bind(c) :: MYFTYPE_1
> +    real(c_float) :: val_1
> +    real(c_double) :: val_2
> +    real(c_long_double) :: val_3
> +    complex(c_float_complex) :: val_4
> +    complex(c_double_complex) :: val_5
> +    complex(c_long_double_complex) :: val_6
> +    logical(c_bool) :: val_7
> +    !FIXME: Fortran define c_char as array of size 1.
> +    !character(c_char) :: val_8
> +  end type MYFTYPE_1
> +
> +  type(myftype_1), bind(c, name="myVar") :: myVar
> +
> +contains
> +  subroutine types_test1() bind(c)
> +    myVar%val_1 = 2
> +  end subroutine types_test1
> +  subroutine types_test2() bind(c)
> +    myVar%val_2 = 2
> +  end subroutine types_test2
> +  subroutine types_test3() bind(c)
> +    myVar%val_3 = 2
> +  end subroutine types_test3
> +  subroutine types_test4() bind(c)
> +    myVar%val_4 = 2
> +  end subroutine types_test4
> +  subroutine types_test5() bind(c)
> +    myVar%val_5 = 2
> +  end subroutine types_test5
> +  subroutine types_test6() bind(c)
> +    myVar%val_6 = 2
> +  end subroutine types_test6
> +  subroutine types_test7() bind(c)
> +    myVar%val_7 = myVar%val_7 .or. .not. myVar%val_7
> +  end subroutine types_test7
> +  !subroutine types_test8() bind(c)
> +    !myVar%val_8 = "a"
> +  !end subroutine types_test8
> +end module lto_type_merge_test
> +
> Index: testsuite/gfortran.dg/lto/bind_c-4_1.c
> ===================================================================
> --- testsuite/gfortran.dg/lto/bind_c-4_1.c	(revision 0)
> +++ testsuite/gfortran.dg/lto/bind_c-4_1.c	(working copy)
> @@ -0,0 +1,46 @@
> +#include <stdlib.h>
> +#include <stdint.h>
> +/* interopse with myftype_1 */
> +typedef struct {
> +  float val1;
> +  double val2;
> +  long double val3;
> +  float _Complex val4;
> +  double _Complex val5;
> +  long double _Complex val6;
> +  _Bool val7;
> +  /* FIXME: Fortran define c_char as array of size 1.
> +     char val8;  */
> +} myctype_t;
> +
> +
> +extern void abort(void);
> +void types_test1(void);
> +void types_test2(void);
> +void types_test3(void);
> +void types_test4(void);
> +void types_test5(void);
> +void types_test6(void);
> +void types_test7(void);
> +void types_test8(void);
> +/* declared in the fortran module */
> +extern myctype_t myVar;
> +
> +#define test(n)\
> +  cchr->val##n = 1; types_test##n (); if (cchr->val##n != 2) abort ();
> +
> +int main(int argc, char **argv)
> +{
> +   myctype_t *cchr;
> +   asm("":"=r"(cchr):"0"(&myVar));
> +   test(1);
> +   test(2);
> +   test(3);
> +   test(4);
> +   test(5);
> +   test(6);
> +   cchr->val7 = 0; types_test7 (); if (cchr->val7 != 1) abort ();
> +   /*cchr->val8 = 0; types_test8 (); if (cchr->val8 != 'a') abort ();*/
> +   return 0;
> +}
> +
> Index: tree.c
> ===================================================================
> --- tree.c	(revision 224201)
> +++ tree.c	(working copy)
> @@ -12925,8 +12925,8 @@
>      return TYPE_CANONICAL (t1) == TYPE_CANONICAL (t2);
>  
>    /* Can't be the same type if the types don't have the same code.  */
> -  if (tree_code_for_canonical_type_merging (TREE_CODE (t1))
> -      != tree_code_for_canonical_type_merging (TREE_CODE (t2)))
> +  enum tree_code code = tree_code_for_canonical_type_merging (TREE_CODE (t1));
> +  if (code != tree_code_for_canonical_type_merging (TREE_CODE (t2)))
>      return false;
>  
>    /* Qualifiers do not matter for canonical type comparison purposes.  */
> @@ -12949,14 +12949,19 @@
>        || TREE_CODE (t1) == OFFSET_TYPE
>        || POINTER_TYPE_P (t1))
>      {
> -      /* Can't be the same type if they have different sign or precision.  */
> -      if (TYPE_PRECISION (t1) != TYPE_PRECISION (t2)
> -	  || TYPE_UNSIGNED (t1) != TYPE_UNSIGNED (t2))
> +      /* Can't be the same type if they have different recision.  */
> +      if (TYPE_PRECISION (t1) != TYPE_PRECISION (t2))
>  	return false;
> +      /* Ignore TYPE_SIGNED.  This is needed for Fortran
> +	 C_SIGNED_CHAR to be interoperable with both signed char and
> +	 unsigned char (as stadnard requires).  Similarly Fortran FE builds
> +	 C_SIZE_T is signed type, while C defines it unsigned.
> +	 NOTE 15.8 of Fortran 2008 seems to imply that even other types ought
> +	 to be inter-operable. */
>  
> -      if (TREE_CODE (t1) == INTEGER_TYPE
> -	  && TYPE_STRING_FLAG (t1) != TYPE_STRING_FLAG (t2))
> -	return false;
> +      /* Fortran FE does not set STRING_FLAG for C_SIGNED_CHAR while C
> +	 sets it for signed char.  To use string flag to define canonical
> +	 types, the frontends would need to agree.  */
>  
>        /* Fortran standard define C_PTR type that is compatible with every
>   	 C pointer.  For this reason we need to glob all pointers into one.
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Dilip Upmanyu, Graham Norton, HRB 21284 (AG Nuernberg)