[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 cesar at gcc dot gnu.org changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #11 from cesar at gcc dot gnu.org --- Fixed in r234840.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 --- Comment #10 from cesar at gcc dot gnu.org --- Author: cesar Date: Fri Apr 8 21:09:47 2016 New Revision: 234840 URL: https://gcc.gnu.org/viewcvs?rev=234840=gcc=rev Log: gcc/ PR lto/70289 PR ipa/70348 PR tree-optimization/70373 PR middle-end/70533 PR middle-end/70534 PR middle-end/70535 * gimplify.c (gimplify_adjust_omp_clauses): Add or adjust data clauses for acc parallel reductions as necessary. Error on those that are private. * omp-low.c (scan_sharing_clauses): Don't install variables which are used in acc parallel reductions. (lower_rec_input_clauses): Remove dead code. (lower_oacc_reductions): Add support for reference reductions. (lower_reduction_clauses): Remove dead code. (lower_omp_target): Don't remap variables appearing in acc parallel reductions. * tree.h (OMP_CLAUSE_MAP_IN_REDUCTION): New macro. gcc/testsuite/ * c-c++-common/goacc/reduction-5.c: New test. * c-c++-common/goacc/reduction-promotions.c: New test. * gfortran.dg/goacc/reduction-3.f95: New test. * gfortran.dg/goacc/reduction-promotions.f90: New test. libgomp/ * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gang-np-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gw-np-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-2.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-3.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-4.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-vector-p-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-vector-p-2.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-worker-p-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-2.c: New test. * testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-3.c: New test. * testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-1.c: New test. * testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-2.c: New test. * testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-3.c: New test. * testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-4.c: New test. * testsuite/libgomp.oacc-c-c++-common/par-reduction-1.c: Add test coverage. * testsuite/libgomp.oacc-c-c++-common/par-reduction-2.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/parallel-dims.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/parallel-reduction.c: New test. * testsuite/libgomp.oacc-c-c++-common/pr70289.c: New test. * testsuite/libgomp.oacc-c-c++-common/pr70373.c: New test. * testsuite/libgomp.oacc-c-c++-common/reduction-1.c: Add test coverage. * testsuite/libgomp.oacc-c-c++-common/reduction-2.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/reduction-3.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/reduction-4.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/reduction-5.c: Likewise. * testsuite/libgomp.oacc-c-c++-common/reduction-6.c: New test. * testsuite/libgomp.oacc-c-c++-common/reduction.h: New test. * testsuite/libgomp.oacc-fortran/parallel-reduction.f90: New test. * testsuite/libgomp.oacc-fortran/pr70289.f90: New test. * testsuite/libgomp.oacc-fortran/reduction-1.f90: Add test coverage. * testsuite/libgomp.oacc-fortran/reduction-2.f90: Likewise. * testsuite/libgomp.oacc-fortran/reduction-3.f90: Likewise. * testsuite/libgomp.oacc-fortran/reduction-4.f90: Likewise. * testsuite/libgomp.oacc-fortran/reduction-5.f90: Likewise. * testsuite/libgomp.oacc-fortran/reduction-6.f90: Likewise. * testsuite/libgomp.oacc-fortran/reduction-7.f90: New test. Added: trunk/gcc/testsuite/c-c++-common/goacc/reduction-5.c trunk/gcc/testsuite/c-c++-common/goacc/reduction-promotions.c trunk/gcc/testsuite/gfortran.dg/goacc/reduction-3.f95 trunk/gcc/testsuite/gfortran.dg/goacc/reduction-promotions.f90 trunk/libgomp/testsuite/libgomp.oacc-c-c++-common/loop-reduction-gang-np-1.c trunk/libgomp/testsuite/libgomp.oacc-c-c++-common/loop-reduction-gv-np-1.c trunk/libgomp/testsuite/libgomp.oacc-c-c++-common/loop-reduction-gw-np-1.c trunk/libgomp/testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-1.c trunk/libgomp/testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-2.c
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 --- Comment #9 from cesar at gcc dot gnu.org --- The patch I posted for PR70289 a couple of days ago also resolves this issue: https://gcc.gnu.org/ml/gcc-patches/2016-04/msg00202.html
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 Thomas Schwinge changed: What|Removed |Added Assignee|tschwinge at gcc dot gnu.org |cesar at gcc dot gnu.org --- Comment #8 from Thomas Schwinge --- Cesar is now looking into resolving the OpenACC reduction issues.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 Thomas Schwinge changed: What|Removed |Added Assignee|unassigned at gcc dot gnu.org |tschwinge at gcc dot gnu.org --- Comment #7 from Thomas Schwinge --- (In reply to Nathan Sidwell from comment #5) > the intended current implementation is to treat 'sum' as firstprivate. Not actively looking into that one at the moment, but: > however, putting in an explicit firstprivate causes another different ICE > (in lower_oacc_reductions) -- regardless of whether 'sum' is a parameter, > or a local (initialized) variable. Clearly at least two other bugs lurk. I highly suspect (will test) that ICE to be fixed by patch that I'm currently preparing for trunk (already fixed on gomp-4_0-branch). (In reply to Jakub Jelinek from comment #6) > So, given this is a P1, is anyone working on it? So, assigning to me for the moment; will then re-evaluate the status after the ICE's fixed.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 --- Comment #6 from Jakub Jelinek --- So, given this is a P1, is anyone working on it?
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 --- Comment #5 from Nathan Sidwell --- the intended current implementation is to treat 'sum' as firstprivate. however, putting in an explicit firstprivate causes another different ICE (in lower_oacc_reductions) -- regardless of whether 'sum' is a parameter, or a local (initialized) variable. Clearly at least two other bugs lurk.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348
Thomas Schwinge changed:
What|Removed |Added
CC||nathan at gcc dot gnu.org
--- Comment #4 from Thomas Schwinge ---
Martin, thanks for the initial analysis.
Nathan CCed for your information; here is a reduced test case (which hopefully
still exhibits the original problem):
void foo (int sum)
{
#pragma acc parallel reduction(+:sum)
;
}
The ICE goes away when adding a data clause, such as copy(sum).
When changing sum from a formal parameter to an (initialized) local variable,
without adding a data clause, there is no ICE, but a "warning: 'sum' is used
uninitialized in this function [-Wuninitialized]". (This probably correct
given the current implementation, which requires you to specify a data clause
in addition to a reduction clause.)
We're currently clarifying with the OpenACC technical committee whether a
reduction clause actually implies a data clause.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 Richard Biener changed: What|Removed |Added Priority|P3 |P1 --- Comment #3 from Richard Biener --- Technically P1 but I guess we'll happily ignore openacc regressions in the end.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 --- Comment #2 from Martin Jambor --- The problem is that there is a PARM_DECL in the IL of the function which is not listed among DECL_ARGUMENTS of the function: The function we are in is: (gdb) call debug_generic_expr(cfun->decl) foo._omp_fn.0 but look at the context of the PARM_DECL in question: (gdb) call debug_tree (op) unit size align 32 symtab 0 alias set -1 canonical type 0x76891888 precision 32 min max context pointer_to_this > addressable used unsigned SI file pr70348.c line 7 col 26 size unit size align 32 context arg-type > Also, if you check the uid of the PARM_DECL: (gdb) p op->decl_minimal.uid $3 = 2907 you will see that it corresponds to a parameter the original function foo, from which OpenACC outlined bits to foo._omp_fn.0: foo (intD.6 nD.2906, unsigned intD.9 sumD.2907) And just by the way, because of the context mismatch, the testcase ICEs in expand_expr_real_1 when compiled at -O0: pr70348.c: In function ‘foo._omp_fn.0’: pr70348.c:11:9: internal compiler error: in expand_expr_real_1, at expr.c:9651 #pragma acc parallel reduction(+:sum) ^~~ 0xa61521 expand_expr_real_1(tree_node*, rtx_def*, machine_mode, expand_modifier, rtx_def**, bool) /home/mjambor/gcc/icln/src/gcc/expr.c:9645 0xa5a4c3 expand_expr_real(tree_node*, rtx_def*, machine_mode, expand_modifier, rtx_def**, bool) /home/mjambor/gcc/icln/src/gcc/expr.c:7962 0xa5178b store_expr_with_bounds(tree_node*, rtx_def*, int, bool, bool, tree_node*) /home/mjambor/gcc/icln/src/gcc/expr.c:5406 0xa5039c expand_assignment(tree_node*, tree_node*, bool) /home/mjambor/gcc/icln/src/gcc/expr.c:5175 0x90e4bc expand_gimple_stmt_1 /home/mjambor/gcc/icln/src/gcc/cfgexpand.c:3618 0x90e8ad expand_gimple_stmt /home/mjambor/gcc/icln/src/gcc/cfgexpand.c:3714 0x9159be expand_gimple_basic_block /home/mjambor/gcc/icln/src/gcc/cfgexpand.c:5720 0x917487 execute /home/mjambor/gcc/icln/src/gcc/cfgexpand.c:6335 All in all, this is an OpenACC lowering/expansion bug.
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 Thomas Schwinge changed: What|Removed |Added Keywords||openacc Status|UNCONFIRMED |NEW Last reconfirmed||2016-03-22 Ever confirmed|0 |1
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 Richard Biener changed: What|Removed |Added CC||jamborm at gcc dot gnu.org Target Milestone|--- |6.0
[Bug ipa/70348] [6 Regression][openacc] ICE in visit_ref_for_mod_analysis, at ipa-prop.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70348 vries at gcc dot gnu.org changed: What|Removed |Added CC||jakub at gcc dot gnu.org Summary|[openacc] ICE in|[6 Regression][openacc] ICE |visit_ref_for_mod_analysis, |in |at ipa-prop.c |visit_ref_for_mod_analysis, ||at ipa-prop.c --- Comment #1 from vries at gcc dot gnu.org --- When removing the gang clause from the loop directive in the testcase, the example compiles with the 5 branch, but ICEs with trunk. So, marking 6 regression.
