From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 2153) id 0E2253852745; Sat, 30 Jul 2022 09:34:51 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 0E2253852745 MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset="utf-8" From: Jakub Jelinek To: gcc-cvs@gcc.gnu.org Subject: [gcc r12-8645] openmp: Fix up handling of non-rectangular simd loops with pointer type iterators [PR106449] X-Act-Checkin: gcc X-Git-Author: Jakub Jelinek X-Git-Refname: refs/heads/releases/gcc-12 X-Git-Oldrev: 0062d8491c20eb95411ac0b112e09ec0cee836d1 X-Git-Newrev: 681c73db9bd156f9b65a73ccc6c4a0a697fe70d6 Message-Id: <20220730093451.0E2253852745@sourceware.org> Date: Sat, 30 Jul 2022 09:34:51 +0000 (GMT) X-BeenThere: gcc-cvs@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-cvs mailing list List-Unsubscribe: , List-Archive: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 30 Jul 2022 09:34:51 -0000 https://gcc.gnu.org/g:681c73db9bd156f9b65a73ccc6c4a0a697fe70d6 commit r12-8645-g681c73db9bd156f9b65a73ccc6c4a0a697fe70d6 Author: Jakub Jelinek Date: Fri Jul 29 09:49:11 2022 +0200 openmp: Fix up handling of non-rectangular simd loops with pointer type iterators [PR106449] There were 2 issues visible on this new testcase, one that we didn't have special POINTER_TYPE_P handling in a few spots of expand_omp_simd - for pointers we need to use POINTER_PLUS_EXPR and need to have the non-pointer part in sizetype, for non-rectangular loop on the other side we can rely on multiplication factor 1, pointers can't be multiplied, without those changes we'd ICE. The other issue was that we put n2 expression directly into a comparison in a condition and regimplified that, for the &a[512] case that and with gimplification being destructed that unfortunately meant modification of original fd->loops[?].n2. Fixed by unsharing the expression. This was causing a runtime failure on the testcase. 2022-07-29 Jakub Jelinek PR middle-end/106449 * omp-expand.cc (expand_omp_simd): Fix up handling of pointer iterators in non-rectangular simd loops. Unshare fd->loops[i].n2 or n2 before regimplifying it inside of a condition. * testsuite/libgomp.c-c++-common/pr106449.c: New test. (cherry picked from commit 97d32048c04e9787fccadc4bae1c042754503e34) Diff: --- gcc/omp-expand.cc | 57 ++++++++++++++------- libgomp/testsuite/libgomp.c-c++-common/pr106449.c | 62 +++++++++++++++++++++++ 2 files changed, 101 insertions(+), 18 deletions(-) diff --git a/gcc/omp-expand.cc b/gcc/omp-expand.cc index ee708314793..19d04a0d959 100644 --- a/gcc/omp-expand.cc +++ b/gcc/omp-expand.cc @@ -6718,7 +6718,7 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) if (fd->loops[i].m2) t = n2v = create_tmp_var (itype); else - t = fold_convert (itype, fd->loops[i].n2); + t = fold_convert (itype, unshare_expr (fd->loops[i].n2)); t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE, false, GSI_CONTINUE_LINKING); tree v = fd->loops[i].v; @@ -6732,7 +6732,7 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) if (fd->collapse > 1 && !broken_loop) t = n2var; else - t = fold_convert (type, n2); + t = fold_convert (type, unshare_expr (n2)); t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE, false, GSI_CONTINUE_LINKING); tree v = fd->loop.v; @@ -6844,7 +6844,7 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) if (fd->loops[i].m2) t = nextn2v = create_tmp_var (itype); else - t = fold_convert (itype, fd->loops[i].n2); + t = fold_convert (itype, unshare_expr (fd->loops[i].n2)); t = force_gimple_operand_gsi (&gsi, t, true, NULL_TREE, false, GSI_CONTINUE_LINKING); tree v = fd->loops[i].v; @@ -6874,17 +6874,25 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) ne->probability = e->probability.invert (); gsi = gsi_after_labels (init_bb); - t = fold_convert (TREE_TYPE (fd->loops[i + 1].v), - fd->loops[i + 1].n1); if (fd->loops[i + 1].m1) { - tree t2 = fold_convert (TREE_TYPE (t), + tree t2 = fold_convert (TREE_TYPE (fd->loops[i + 1].v), fd->loops[i + 1 - fd->loops[i + 1].outer].v); - tree t3 = fold_convert (TREE_TYPE (t), fd->loops[i + 1].m1); - t2 = fold_build2 (MULT_EXPR, TREE_TYPE (t), t2, t3); - t = fold_build2 (PLUS_EXPR, TREE_TYPE (t), t, t2); + if (POINTER_TYPE_P (TREE_TYPE (t2))) + t = fold_build_pointer_plus (t2, fd->loops[i + 1].n1); + else + { + t = fold_convert (TREE_TYPE (fd->loops[i + 1].v), + fd->loops[i + 1].n1); + tree t3 = fold_convert (TREE_TYPE (t), fd->loops[i + 1].m1); + t2 = fold_build2 (MULT_EXPR, TREE_TYPE (t), t2, t3); + t = fold_build2 (PLUS_EXPR, TREE_TYPE (t), t, t2); + } } + else + t = fold_convert (TREE_TYPE (fd->loops[i + 1].v), + fd->loops[i + 1].n1); expand_omp_build_assign (&gsi, fd->loops[i + 1].v, t); if (fd->loops[i + 1].m2) { @@ -6893,14 +6901,19 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) gcc_assert (n2v == NULL_TREE); n2v = create_tmp_var (TREE_TYPE (fd->loops[i + 1].v)); } - t = fold_convert (TREE_TYPE (fd->loops[i + 1].v), - fd->loops[i + 1].n2); - tree t2 = fold_convert (TREE_TYPE (t), + tree t2 = fold_convert (TREE_TYPE (fd->loops[i + 1].v), fd->loops[i + 1 - fd->loops[i + 1].outer].v); - tree t3 = fold_convert (TREE_TYPE (t), fd->loops[i + 1].m2); - t2 = fold_build2 (MULT_EXPR, TREE_TYPE (t), t2, t3); - t = fold_build2 (PLUS_EXPR, TREE_TYPE (t), t, t2); + if (POINTER_TYPE_P (TREE_TYPE (t2))) + t = fold_build_pointer_plus (t2, fd->loops[i + 1].n2); + else + { + t = fold_convert (TREE_TYPE (fd->loops[i + 1].v), + fd->loops[i + 1].n2); + tree t3 = fold_convert (TREE_TYPE (t), fd->loops[i + 1].m2); + t2 = fold_build2 (MULT_EXPR, TREE_TYPE (t), t2, t3); + t = fold_build2 (PLUS_EXPR, TREE_TYPE (t), t, t2); + } expand_omp_build_assign (&gsi, n2v, t); } if (i + 2 == fd->collapse && n2var) @@ -6916,17 +6929,25 @@ expand_omp_simd (struct omp_region *region, struct omp_for_data *fd) tree t2 = fold_build2 (MINUS_EXPR, type, n2, fd->loop.v); if (fd->loops[i + 1].m1 || fd->loops[i + 1].m2) { + tree itype = TREE_TYPE (fd->loops[i].v); + if (POINTER_TYPE_P (itype)) + itype = signed_type_for (itype); t = build_int_cst (itype, (fd->loops[i + 1].cond_code == LT_EXPR ? -1 : 1)); t = fold_build2 (PLUS_EXPR, itype, fold_convert (itype, fd->loops[i + 1].step), t); - if (fd->loops[i + 1].m2) - t = fold_build2 (PLUS_EXPR, itype, t, n2v); - else + if (fd->loops[i + 1].m2 == NULL_TREE) t = fold_build2 (PLUS_EXPR, itype, t, fold_convert (itype, fd->loops[i + 1].n2)); + else if (POINTER_TYPE_P (TREE_TYPE (n2v))) + { + t = fold_build_pointer_plus (n2v, t); + t = fold_convert (itype, t); + } + else + t = fold_build2 (PLUS_EXPR, itype, t, n2v); t = fold_build2 (MINUS_EXPR, itype, t, fold_convert (itype, fd->loops[i + 1].v)); tree step = fold_convert (itype, fd->loops[i + 1].step); diff --git a/libgomp/testsuite/libgomp.c-c++-common/pr106449.c b/libgomp/testsuite/libgomp.c-c++-common/pr106449.c new file mode 100644 index 00000000000..ea3cd827cee --- /dev/null +++ b/libgomp/testsuite/libgomp.c-c++-common/pr106449.c @@ -0,0 +1,62 @@ +/* PR middle-end/106449 */ +/* { dg-do run } */ + +void +foo (void) +{ + int a[1024], *b[65536]; + int *p, *q, **r = &b[0], i; + #pragma omp simd collapse(2) linear(r : 2) + for (p = &a[0]; p < &a[512]; p++) + for (q = p + 64; q < p + 128; q++) + { + *r++ = p; + *r++ = q; + } + for (i = 0; i < 32768; i++) + if (b[2 * i] != &a[i / 64] || b[2 * i + 1] != &a[(i / 64) + 64 + (i % 64)]) + __builtin_abort (); +} + +void +bar (int n, int m) +{ + int a[1024], *b[65536]; + int *p, *q, **r = &b[0], i; + #pragma omp parallel for simd collapse(2) linear(r : 2) + for (p = &a[0]; p < &a[512]; p++) + for (q = p + n; q < p + m; q++) + { + *r++ = p; + *r++ = q; + } + for (i = 0; i < 32768; i++) + if (b[2 * i] != &a[i / 64] || b[2 * i + 1] != &a[(i / 64) + 64 + (i % 64)]) + __builtin_abort (); +} + +void +baz (int n, int m) +{ + int a[1024], *b[8192]; + int *p, *q, **r = &b[0], i; + #pragma omp parallel for simd collapse(2) linear(r : 2) + for (p = &a[0]; p < &a[512]; p += 4) + for (q = p + n; q < p + m; q += 2) + { + *r++ = p; + *r++ = q; + } + for (i = 0; i < 4096; i++) + if (b[2 * i] != &a[(i / 32) * 4] || b[2 * i + 1] != &a[(i / 32) * 4 + 64 + (i % 32) * 2]) + __builtin_abort (); +} + +int +main () +{ + foo (); + bar (64, 128); + baz (64, 128); + return 0; +}