Hi! The spec is unclear how to handle this yet, so I'm committing an easy fix, perhaps the right one will be to deal with the wrapping of inner linear into outer firstprivate/lastprivate and/or shared during clause splitting in the FEs.
Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk and 5 branch. 2015-09-10 Jakub Jelinek <ja...@redhat.com> PR middle-end/67517 * gimplify.c (gimplify_scan_omp_clauses): Instead of asserting that decl is not specified in octx->variables, break out of the loop if it is. * c-c++-common/gomp/pr67517.c: New test. --- gcc/gimplify.c.jj 2015-09-09 13:16:14.000000000 +0200 +++ gcc/gimplify.c 2015-09-09 14:20:59.255606297 +0200 @@ -6214,9 +6214,12 @@ gimplify_scan_omp_clauses (tree *list_p, } else break; - gcc_checking_assert (splay_tree_lookup (octx->variables, - (splay_tree_key) - decl) == NULL); + if (splay_tree_lookup (octx->variables, + (splay_tree_key) decl) != NULL) + { + octx = NULL; + break; + } flags = GOVD_SEEN; if (!OMP_CLAUSE_LINEAR_NO_COPYIN (c)) flags |= GOVD_FIRSTPRIVATE; --- gcc/testsuite/c-c++-common/gomp/pr67517.c.jj 2015-09-09 15:36:55.972319265 +0200 +++ gcc/testsuite/c-c++-common/gomp/pr67517.c 2015-09-09 15:35:37.000000000 +0200 @@ -0,0 +1,13 @@ +/* PR middle-end/67517 */ +/* { dg-do compile } */ +/* { dg-options "-fopenmp" } */ + +int +foo (int x, int y, int z) +{ + int i; + #pragma omp parallel for simd linear (y : x & 15) linear (x : 16) linear (z : x & 15) + for (i = 0; i < 256; ++i) + x += 16, y += x & 15, z += x & 15; + return x + y + z; +} Jakub