[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665 Dominique d'Humieres dominiq at lps dot ens.fr changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #29 from Dominique d'Humieres dominiq at lps dot ens.fr --- Can this be closed? No answer since almost a year. Closing as FIXED.
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665 Mikael Morin mikael at gcc dot gnu.org changed: What|Removed |Added CC||mikael at gcc dot gnu.org --- Comment #28 from Mikael Morin mikael at gcc dot gnu.org 2012-06-29 17:31:19 UTC --- Can this be closed?
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665 --- Comment #27 from Thomas Koenig tkoenig at gcc dot gnu.org 2010-10-16 16:06:16 UTC --- Author: tkoenig Date: Sat Oct 16 16:06:07 2010 New Revision: 165559 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=165559 Log: 2010-10-16 Thomas Koenig tkoe...@gcc.gnu.org PR fortran/20165 PR fortran/31593 PR fortran/43665 * gfortran.map: Add _gfortran_transfer_array_write, _gfortran_transfer_array_write, _gfortran_transfer_character_write, _gfortran_transfer_character_wide_write, _gfortran_transfer_complex_write, _gfortran_transfer_integer_write, _gfortran_transfer_logical_write and _gfortran_transfer_real_write. * io/transfer.c (transfer_integer_write): Add prototype and function body as call to the original function, without the _write. (transfer_real_write): Likewise. (transfer_logical_write): Likewise. (transfer_character_write): Likewise. (transfer_character_wide_write): Likewise. (transfer_complex_write): Likewise. (transfer_array_write): Likewise. 2010-10-16 Thomas Koenig tkoe...@gcc.gnu.org PR fortran/20165 PR fortran/31593 PR fortran/43665 * trans-io.c (enum iocall): Add IOCALL_X_INTEGER_WRITE, IOCALL_X_LOGICAL_WRITE, IOCALL_X_CHARACTER_WRITE, IOCALL_X_CHARACTER_WIDE_WRIE, IOCALL_X_REAL_WRITE, IOCALL_X_COMPLEX_WRITE and IOCALL_X_ARRAY_WRITE. (gfc_build_io_library_fndecls): Add corresponding function decls. (transfer_expr): If the current transfer is a READ, use the iocall with the original version, otherwise the version with _WRITE. (transfer_array_desc): Likewise. Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/trans-io.c trunk/libgfortran/ChangeLog trunk/libgfortran/gfortran.map trunk/libgfortran/io/transfer.c
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #26 from burnus at gcc dot gnu dot org 2010-09-16 21:30 --- Subject: Bug 43665 Author: burnus Date: Thu Sep 16 21:30:05 2010 New Revision: 164348 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=164348 Log: 2010-09-16 Tobias Burnus bur...@net-b.de PR fortran/43665 * trans-types.c (create_fn_spec): New function. (gfc_get_function_type): Call it. 2010-09-16 Tobias Burnus bur...@net-b.de PR fortran/43665 * gfortran.dg/cray_pointers_2.f90: Disable inlining to avoid optimizations. * gfortran.dg/intent_optimize_1.f90: New test. Added: trunk/gcc/testsuite/gfortran.dg/intent_optimize_1.f90 Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/trans-types.c trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/gfortran.dg/cray_pointers_2.f90 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #24 from burnus at gcc dot gnu dot org 2010-09-12 09:32 --- (In reply to comment #23) I have applied the patch in comment #21 without regression, but the test case from attachment 21265 [edit] fails: - create_fn_spec (sym, type); + type = create_fn_spec (sym, type); as in the original patch. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #25 from dominiq at lps dot ens dot fr 2010-09-12 10:13 --- - create_fn_spec (sym, type); + type = create_fn_spec (sym, type); as in the original patch. With this change the test succeeds. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #23 from dominiq at lps dot ens dot fr 2010-09-11 15:12 --- I have applied the patch in comment #21 without regression, but the test case from attachment 21265 fails: FAIL: gfortran.dg/intent_optimize_1.f90 -O scan-tree-dump-times optimized does_not_exist 0 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #21 from burnus at gcc dot gnu dot org 2010-09-10 12:09 --- Created an attachment (id=21765) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21765action=view) Updated patch Updated patch to fix review issues and the Cray patch issue. I won't be able to work on this (i.e. regtest + submit) for the next days thus I put it here. Cf. http://gcc.gnu.org/ml/fortran/2010-09/msg00198.html and http://gcc.gnu.org/ml/fortran/2010-09/msg00234.html -- burnus at gcc dot gnu dot org changed: What|Removed |Added Attachment #21265|0 |1 is obsolete|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #22 from burnus at gcc dot gnu dot org 2010-09-10 12:12 --- (In reply to comment #21) Created an attachment (id=21765) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21765action=view) [edit] Updated patch Note: I forgot to include the test case from attachment 21265 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #19 from burnus at gcc dot gnu dot org 2010-09-08 06:25 --- Reviewed patch (OK) available at http://gcc.gnu.org/ml/fortran/2010-09/msg00198.html however, it causes regressions as some of the intrinsics (in intrinsic.c) have the wrong intents - which causes wrong code (too much optimized away). Thus, one first needs to audit and fix intrinsic.c before this patch can be committed. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #18 from jamborm at gcc dot gnu dot org 2010-09-07 17:01 --- Subject: Bug 43665 Author: jamborm Date: Tue Sep 7 17:00:44 2010 New Revision: 163960 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=163960 Log: 2010-09-07 Martin Jambor mjam...@suse.cz PR fortran/43665 * ipa-cp.c (ipcp_versionable_function_p): Return false if there are any type attributes. Modified: trunk/gcc/ChangeLog trunk/gcc/ipa-cp.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #15 from jamborm at gcc dot gnu dot org 2010-09-06 14:12 --- I tried compiling the testcase from comment #8 and it did not fail for me either on i686-linux ox x86_64-linux. Can you please check that it still fails for you? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #16 from jamborm at gcc dot gnu dot org 2010-09-06 18:25 --- Created an attachment (id=21714) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21714action=view) Patch to address IPA-CP parameter removal issues This patch makes IPA-CP to refrain from modifying a function when it sees a any type attributes. It fixes the test case. I do not expect it to cause any problems elsewhere but I have not yet bootstrapped or tested it (I have just scheduled both for tonight). As far as I understand it there is already a test in our testsuite that fails (with the patch from comment #7 applied) and so I am not going to add an extra one. I think it's best to check this in separately and I will submit it for approval tomorrow if there are no unforeseen problems. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #17 from burnus at gcc dot gnu dot org 2010-09-06 18:43 --- (In reply to comment #16) This patch makes IPA-CP to refrain from modifying a function when it sees a any type attributes. In a way that's unfortunate: Both fn attr and argument removal are optimization options. As far as I understand it there is already a test in our testsuite that fails (with the patch from comment #7 applied) and so I am not going to add an extra one. OK. I think it's best to check this in separately and I will submit it for approval tomorrow if there are no unforeseen problems. I will also submit my patch - for committal after yours is in. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #14 from burnus at gcc dot gnu dot org 2010-07-22 15:36 --- (In reply to comment #13) (In reply to comment #12) IPA-CP can do that for quite some time please try with -fno-ipa-cp. As expected: It works with -fno-ipa-cp. (I don't have a trunk built with enabled fortran at hand and I am a bit overwhelmed with bugs and other stuff recently so I can have a look at this but it will take at least a few days before I get to it.) Thanks. Some pre-analysis: The fn spec (the space is there to prevent other than internal use - thus I cannot create a C test case) is converted in gimple.c's gimple_call_arg_flags to EAF_* constants. tree-ssa-structalias.c's handle_rhs_call uses them via: for (i = 0; i gimple_call_num_args (stmt); ++i) [...] int flags = gimple_call_arg_flags (stmt, i); thus, the position seems to matter here. The latter is called by find_func_aliases in an if (!in_ipa_mode ... block. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #12 from rguenther at suse dot de 2010-07-21 08:09 --- Subject: Re: INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments On Tue, 20 Jul 2010, burnus at gcc dot gnu dot org wrote: --- Comment #11 from burnus at gcc dot gnu dot org 2010-07-20 16:12 --- (In reply to comment #10) No, this problem was present for ipa-sra, but I thought it was fixed. Does -fno-ipa-sra help? No, it doesn't. But with that option, the number of arguments still reduces from 4 to 2 (before: 1). But again, the code works if one reorders the arguments such that the w attribute matches the x argument (now at position 2 instead of 1). So I wonder what code removes the arguments then. Richard. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #13 from jamborm at gcc dot gnu dot org 2010-07-21 08:27 --- (In reply to comment #12) So I wonder what code removes the arguments then. IPA-CP can do that for quite some time please try with -fno-ipa-cp. (I don't have a trunk built with enabled fortran at hand and I am a bit overwhelmed with bugs and other stuff recently so I can have a look at this but it will take at least a few days before I get to it.) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #7 from burnus at gcc dot gnu dot org 2010-07-20 15:06 --- Created an attachment (id=21265) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21265action=view) Draft patch for external/user procs with INTENT (mostly OK, but breaks existing test) The attached patch mostly works, except for gfortran.dg/allocatable_scalar_4.f90. The latter file works fine with -O1 but with -O2 it crashes. At -O1 with -fdump-tree-original, essentially only one if is optimized away: b = 7482 call checkOptional(.false.,.true., 7482) if (b /= 7482) call abort() ! This line is removed for -O1 call checkOptional(.true., .true., 7482, b) I could not see anything suspicious for -O{2,1} -fdump-tree-optimized; one should check -fdump-tree-optimized-all. I somehow have the feeling that some other DECL is wrong. Cf. PR 44945 for another manifestation of DECL problems (though there for derived types). Note: Using -fwhole-file does not help. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #8 from burnus at gcc dot gnu dot org 2010-07-20 15:26 --- Reduced test case (of allocatable_scalar_4.f90) - fails with ERROR STOP 1: program test implicit none integer, allocatable :: b allocate(b) b = 7482 call checkOptional(.false.,.true., 7482) call checkOptional(.true., .true., 7482, b) if (b /= 46) error stop 1 contains subroutine checkOptional(prsnt, alloc, val, x) logical, intent(in) :: prsnt, alloc integer, allocatable, optional :: x integer, intent(in) :: val if (present(x)) then if (allocated(x) .neqv. alloc) error stop 2 end if if (present(x)) then if (allocated(x)) then if (x /= val) error stop 3 end if end if if (present(x)) then x = 46 end if end subroutine checkOptional end program test -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #9 from burnus at gcc dot gnu dot org 2010-07-20 15:58 --- The procedure 'checkoptional' gets '.rrrw' set as the fn spec. With -O1, the if after the first checkoptional gets properly optimized away (cf. comment 7 and original test case). With -O2, the number of arguments is reduced to one (NULL vs. b) and the condition if (b /= 46) error stop 1 is changed to error stop 1 (cf. -fdump-tree-optimized) For me it looks as if the fn spec information is not moved along when arguments are eliminated but remain at the previous position. It works if one swaps the order (i.e. it works if the optional argument is at position one). Richard, does this make sense or am I completely off track? -- burnus at gcc dot gnu dot org changed: What|Removed |Added CC||rguenth at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #10 from rguenth at gcc dot gnu dot org 2010-07-20 16:01 --- No, this problem was present for ipa-sra, but I thought it was fixed. Does -fno-ipa-sra help? Martin? -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||jamborm at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #11 from burnus at gcc dot gnu dot org 2010-07-20 16:12 --- (In reply to comment #10) No, this problem was present for ipa-sra, but I thought it was fixed. Does -fno-ipa-sra help? No, it doesn't. But with that option, the number of arguments still reduces from 4 to 2 (before: 1). But again, the code works if one reorders the arguments such that the w attribute matches the x argument (now at position 2 instead of 1). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665
[Bug fortran/43665] INTENT(IN) etc. optimization of calls: function annotations for noclobber/noescape arguments
--- Comment #6 from burnus at gcc dot gnu dot org 2010-07-13 17:32 --- Main library calls are done. TODO: - Intrinsic calls such as call ctime() which do not have a function declaration - Nonintrinsic functions with non-pointer INTENT(IN/OUT); handle also (non)clobber [target/pointer ...] - I/O transfer: Split somehow to make it possible to distinguish READ from WRITE. - Possibly, handle somehow unused arguments / not-explicitly given INTENT(in)s by saving this information -- burnus at gcc dot gnu dot org changed: What|Removed |Added Summary|Optimization of libgfortran |INTENT(IN) etc. optimization |calls: function annotations |of calls: function |for noclobber/noescape |annotations for |arguments |noclobber/noescape arguments http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43665