| Message ID | 20100904231844.GF31380@kam.mff.cuni.cz |
|---|---|
| State | New |
| Headers | show |
Jan Hubicka wrote: > fortran, unlike all other frontends, does not set TREE_STATIC flag on static > initializers. This prevents us from folding till the cfgexpand time. To be honest, > I am not quite sure why we insist on the flag (I guess we should assert that all statics > are initializerd with static initializer or ignore it), but this patch brings gfortran > into sync with others. > > Bootstrapped/regtested x86_64-linux, OK? OK. Thanks for the patch. Tobias PS: It helps to CC fortran@ for Fortran patches; sending them only to gcc-patches@ reduces the chance of getting them reviewed.
On Sat, Sep 4, 2010 at 4:18 PM, Jan Hubicka <hubicka@ucw.cz> wrote: > Hi, > fortran, unlike all other frontends, does not set TREE_STATIC flag on static > initializers. This prevents us from folding till the cfgexpand time. To be honest, > I am not quite sure why we insist on the flag (I guess we should assert that all statics > are initializerd with static initializer or ignore it), but this patch brings gfortran > into sync with others. > > Bootstrapped/regtested x86_64-linux, OK? > > Honza > > * trans-expr.c (gfc_conv_initializer): Set TREE_STATIC on static ctors. This caused: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45598
> On Sat, Sep 4, 2010 at 4:18 PM, Jan Hubicka <hubicka@ucw.cz> wrote: > > Hi, > > fortran, unlike all other frontends, does not set TREE_STATIC flag on static > > initializers. This prevents us from folding till the cfgexpand time. To be honest, > > I am not quite sure why we insist on the flag (I guess we should assert that all statics > > are initializerd with static initializer or ignore it), but this patch brings gfortran > > into sync with others. > > > > Bootstrapped/regtested x86_64-linux, OK? > > > > Honza > > > > * trans-expr.c (gfc_conv_initializer): Set TREE_STATIC on static ctors. > > This caused: > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45598 It was previously latent. I am testing fix for that. Honza
2010/9/8 Jan Hubicka <hubicka@ucw.cz>: >> On Sat, Sep 4, 2010 at 4:18 PM, Jan Hubicka <hubicka@ucw.cz> wrote: >> > Hi, >> > fortran, unlike all other frontends, does not set TREE_STATIC flag on static >> > initializers. This prevents us from folding till the cfgexpand time. To be honest, >> > I am not quite sure why we insist on the flag (I guess we should assert that all statics >> > are initializerd with static initializer or ignore it), but this patch brings gfortran >> > into sync with others. >> > >> > Bootstrapped/regtested x86_64-linux, OK? >> > >> > Honza >> > >> > * trans-expr.c (gfc_conv_initializer): Set TREE_STATIC on static ctors. >> >> This caused: >> >> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45598 > > It was previously latent. I am testing fix for that. > Revision 164111: http://gcc.gnu.org/ml/gcc-cvs/2010-09/msg00403.html doesn't fix SPEC CPU 2K: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45634
On Fri, Sep 10, 2010 at 6:20 AM, H.J. Lu <hjl.tools@gmail.com> wrote: > 2010/9/8 Jan Hubicka <hubicka@ucw.cz>: >>> On Sat, Sep 4, 2010 at 4:18 PM, Jan Hubicka <hubicka@ucw.cz> wrote: >>> > Hi, >>> > fortran, unlike all other frontends, does not set TREE_STATIC flag on static >>> > initializers. This prevents us from folding till the cfgexpand time. To be honest, >>> > I am not quite sure why we insist on the flag (I guess we should assert that all statics >>> > are initializerd with static initializer or ignore it), but this patch brings gfortran >>> > into sync with others. >>> > >>> > Bootstrapped/regtested x86_64-linux, OK? >>> > >>> > Honza >>> > >>> > * trans-expr.c (gfc_conv_initializer): Set TREE_STATIC on static ctors. >>> >>> This caused: >>> >>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45598 >> >> It was previously latent. I am testing fix for that. >> > > Revision 164111: > > http://gcc.gnu.org/ml/gcc-cvs/2010-09/msg00403.html Jan, shouldn't /* Fold read from constant string. */ if (TREE_CODE (ctor) == STRING_CST) { if ((TYPE_MODE (TREE_TYPE (t)) == TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) && (GET_MODE_CLASS (TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) == MODE_INT) && GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) == 1 && compare_tree_int (idx, TREE_STRING_LENGTH (ctor)) < 0) return build_int_cst_type (TREE_TYPE (t), (TREE_STRING_POINTER (ctor) [TREE_INT_CST_LOW (idx)])); return NULL_TREE; } in tree-ssa-ccp.c need a similar fix?
> On Fri, Sep 10, 2010 at 6:20 AM, H.J. Lu <hjl.tools@gmail.com> wrote: > > 2010/9/8 Jan Hubicka <hubicka@ucw.cz>: > >>> On Sat, Sep 4, 2010 at 4:18 PM, Jan Hubicka <hubicka@ucw.cz> wrote: > >>> > Hi, > >>> > fortran, unlike all other frontends, does not set TREE_STATIC flag on static > >>> > initializers. This prevents us from folding till the cfgexpand time. To be honest, > >>> > I am not quite sure why we insist on the flag (I guess we should assert that all statics > >>> > are initializerd with static initializer or ignore it), but this patch brings gfortran > >>> > into sync with others. > >>> > > >>> > Bootstrapped/regtested x86_64-linux, OK? > >>> > > >>> > Honza > >>> > > >>> > * trans-expr.c (gfc_conv_initializer): Set TREE_STATIC on static ctors. > >>> > >>> This caused: > >>> > >>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45598 > >> > >> It was previously latent. I am testing fix for that. > >> > > > > Revision 164111: > > > > http://gcc.gnu.org/ml/gcc-cvs/2010-09/msg00403.html > > Jan, shouldn't > > /* Fold read from constant string. */ > if (TREE_CODE (ctor) == STRING_CST) > { > if ((TYPE_MODE (TREE_TYPE (t)) > == TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) > && (GET_MODE_CLASS (TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) > == MODE_INT) > && GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (TREE_TYPE (ctor)))) == 1 > && compare_tree_int (idx, TREE_STRING_LENGTH (ctor)) < 0) > return build_int_cst_type (TREE_TYPE (t), > (TREE_STRING_POINTER (ctor) > [TREE_INT_CST_LOW (idx)])); > return NULL_TREE; > } > > > in tree-ssa-ccp.c need a similar fix? Yes, I am working on patch that wil unify this code, so we won't need duplication... Honza > > > -- > H.J.
Index: fortran/trans-expr.c =================================================================== --- fortran/trans-expr.c (revision 163862) +++ fortran/trans-expr.c (working copy) @@ -3992,19 +3992,23 @@ gfc_conv_initializer (gfc_expr * expr, g gfc_init_se (&se, NULL); gfc_conv_constant (&se, expr); + gcc_assert (TREE_CODE (se.expr) != CONSTRUCTOR); return se.expr; } if (array && !procptr) { + tree ctor; /* Arrays need special handling. */ if (pointer) - return gfc_build_null_descriptor (type); + ctor = gfc_build_null_descriptor (type); /* Special case assigning an array to zero. */ else if (is_zero_initializer_p (expr)) - return build_constructor (type, NULL); + ctor = build_constructor (type, NULL); else - return gfc_conv_array_initializer (type, expr); + ctor = gfc_conv_array_initializer (type, expr); + TREE_STATIC (ctor) = 1; + return ctor; } else if (pointer || procptr) { @@ -4015,6 +4019,7 @@ gfc_conv_initializer (gfc_expr * expr, g gfc_init_se (&se, NULL); se.want_pointer = 1; gfc_conv_expr (&se, expr); + gcc_assert (TREE_CODE (se.expr) != CONSTRUCTOR); return se.expr; } } @@ -4029,14 +4034,21 @@ gfc_conv_initializer (gfc_expr * expr, g gfc_conv_structure (&se, gfc_class_null_initializer(ts), 1); else gfc_conv_structure (&se, expr, 1); + gcc_assert (TREE_CODE (se.expr) == CONSTRUCTOR); + TREE_STATIC (se.expr) = 1; return se.expr; case BT_CHARACTER: - return gfc_conv_string_init (ts->u.cl->backend_decl,expr); + { + tree ctor = gfc_conv_string_init (ts->u.cl->backend_decl,expr); + TREE_STATIC (ctor) = 1; + return ctor; + } default: gfc_init_se (&se, NULL); gfc_conv_constant (&se, expr); + gcc_assert (TREE_CODE (se.expr) != CONSTRUCTOR); return se.expr; } }