From dd34c087ffe7a798835fb95c7eb3dd68afd08f08 Mon Sep 17 00:00:00 2001 From: Jakub Jelinek Date: Fri, 11 Nov 2011 20:56:13 +0100 Subject: [PATCH] re PR tree-optimization/51058 (ICE: gimple check: expected gimple_assign(error_mark), have gimple_call() in gimple_assign_rhs_code, at gimple.h:1992) PR tree-optimization/51058 * tree-vect-slp.c (vect_remove_slp_scalar_calls): New function. (vect_schedule_slp): Call it. * tree-vect-stmts.c (vectorizable_call): If slp_node != NULL, don't replace scalar calls with setting of their lhs to zero here. * gcc.dg/vect/fast-math-vect-call-1.c: Add f4 test. * gfortran.dg/vect/pr51058-2.f90: New test. From-SVN: r181298 --- gcc/ChangeLog | 6 +++ gcc/testsuite/ChangeLog | 6 +++ .../gcc.dg/vect/fast-math-vect-call-1.c | 22 +++++++++- gcc/testsuite/gfortran.dg/vect/pr51058-2.f90 | 20 +++++++++ gcc/tree-vect-slp.c | 42 +++++++++++++++++++ gcc/tree-vect-stmts.c | 6 ++- 6 files changed, 98 insertions(+), 4 deletions(-) create mode 100644 gcc/testsuite/gfortran.dg/vect/pr51058-2.f90 diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 66e6f671767..60d97d15c8f 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,5 +1,11 @@ 2011-11-11 Jakub Jelinek + PR tree-optimization/51058 + * tree-vect-slp.c (vect_remove_slp_scalar_calls): New function. + (vect_schedule_slp): Call it. + * tree-vect-stmts.c (vectorizable_call): If slp_node != NULL, + don't replace scalar calls with setting of their lhs to zero here. + PR tree-optimization/51074 * fold-const.c (vec_cst_ctor_to_array, fold_vec_perm): New functions. (fold_binary_loc): Handle VEC_EXTRACT_EVEN_EXPR, diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 0c9d088f8a5..c96c5406b04 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,9 @@ +2011-11-11 Jakub Jelinek + + PR tree-optimization/51058 + * gcc.dg/vect/fast-math-vect-call-1.c: Add f4 test. + * gfortran.dg/vect/pr51058-2.f90: New test. + 2011-11-10 Jason Merrill * g++.dg/template/linkage1.C: New. diff --git a/gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c b/gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c index 835f8bc2bbe..db6df7e7511 100644 --- a/gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c +++ b/gcc/testsuite/gcc.dg/vect/fast-math-vect-call-1.c @@ -38,6 +38,18 @@ f3 (void) a[i] = copysignf (b[i], c[i]) + 1.0f + sqrtf (d[i]); } +__attribute__((noinline, noclone)) void +f4 (int n) +{ + int i; + for (i = 0; i < 2 * n; i++) + { + a[3 * i + 0] = copysignf (b[3 * i + 0], c[3 * i + 0]) + 1.0f + sqrtf (d[3 * i + 0]); + a[3 * i + 1] = copysignf (b[3 * i + 1], c[3 * i + 1]) + 2.0f + sqrtf (d[3 * i + 1]); + a[3 * i + 2] = copysignf (b[3 * i + 2], c[3 * i + 2]) + 3.0f + sqrtf (d[3 * i + 2]); + } +} + __attribute__((noinline, noclone)) int main1 () { @@ -66,6 +78,12 @@ main1 () for (i = 0; i < 64; i++) if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + i - a[i]) >= 0.0001f) abort (); + else + a[i] = 131.25; + f4 (10); + for (i = 0; i < 60; i++) + if (fabsf (((i & 2) ? -4 * i : 4 * i) + 1 + (i % 3) + i - a[i]) >= 0.0001f) + abort (); return 0; } @@ -76,6 +94,6 @@ main () return main1 (); } -/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 3 "vect" { target { vect_call_copysignf && vect_call_sqrtf } } } } */ -/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target { vect_call_copysignf && vect_call_sqrtf } } } } */ +/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 4 "vect" { target { vect_call_copysignf && vect_call_sqrtf } } } } */ +/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" { target { vect_call_copysignf && vect_call_sqrtf } } } } */ /* { dg-final { cleanup-tree-dump "vect" } } */ diff --git a/gcc/testsuite/gfortran.dg/vect/pr51058-2.f90 b/gcc/testsuite/gfortran.dg/vect/pr51058-2.f90 new file mode 100644 index 00000000000..f647e6cdb4f --- /dev/null +++ b/gcc/testsuite/gfortran.dg/vect/pr51058-2.f90 @@ -0,0 +1,20 @@ +! PR tree-optimization/51058 +! { dg-do compile } +subroutine pr51058(n, u, v, w, z) + double precision :: x(3,-2:16384), y(3,-2:16384), b, u, v, w, z + integer :: i, n + common /c/ x, y + do i = 1, n + b = u * int(x(1,i)) + sign(z,x(1,i)) + x(1,i) = x(1,i) - b + y(1,i) = y(1,i) - b + b = v * int(x(2,i)) + sign(z,x(2,i)) + x(2,i) = x(2,i) - b + y(2,i) = y(2,i) - b + b = w * int(x(3,i)) + sign(z,x(3,i)) + x(3,i) = x(3,i) - b + y(3,i) = y(3,i) - b + end do +end subroutine + +! { dg-final { cleanup-tree-dump "vect" } } diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c index 4606c1922ac..23330b2882f 100644 --- a/gcc/tree-vect-slp.c +++ b/gcc/tree-vect-slp.c @@ -2902,6 +2902,46 @@ vect_schedule_slp_instance (slp_tree node, slp_instance instance, return is_store; } +/* Replace scalar calls from SLP node NODE with setting of their lhs to zero. + For loop vectorization this is done in vectorizable_call, but for SLP + it needs to be deferred until end of vect_schedule_slp, because multiple + SLP instances may refer to the same scalar stmt. */ + +static void +vect_remove_slp_scalar_calls (slp_tree node) +{ + gimple stmt, new_stmt; + gimple_stmt_iterator gsi; + int i; + slp_void_p child; + tree lhs; + stmt_vec_info stmt_info; + + if (!node) + return; + + FOR_EACH_VEC_ELT (slp_void_p, SLP_TREE_CHILDREN (node), i, child) + vect_remove_slp_scalar_calls ((slp_tree) child); + + FOR_EACH_VEC_ELT (gimple, SLP_TREE_SCALAR_STMTS (node), i, stmt) + { + if (!is_gimple_call (stmt) || gimple_bb (stmt) == NULL) + continue; + stmt_info = vinfo_for_stmt (stmt); + if (stmt_info == NULL + || is_pattern_stmt_p (stmt_info) + || !PURE_SLP_STMT (stmt_info)) + continue; + lhs = gimple_call_lhs (stmt); + new_stmt = gimple_build_assign (lhs, build_zero_cst (TREE_TYPE (lhs))); + set_vinfo_for_stmt (new_stmt, stmt_info); + set_vinfo_for_stmt (stmt, NULL); + STMT_VINFO_STMT (stmt_info) = new_stmt; + gsi = gsi_for_stmt (stmt); + gsi_replace (&gsi, new_stmt, false); + SSA_NAME_DEF_STMT (gimple_assign_lhs (new_stmt)) = new_stmt; + } +} /* Generate vector code for all SLP instances in the loop/basic block. */ @@ -2941,6 +2981,8 @@ vect_schedule_slp (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo) unsigned int j; gimple_stmt_iterator gsi; + vect_remove_slp_scalar_calls (root); + for (j = 0; VEC_iterate (gimple, SLP_TREE_SCALAR_STMTS (root), j, store) && j < SLP_INSTANCE_GROUP_SIZE (instance); j++) { diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c index abccd923b44..93e8198678b 100644 --- a/gcc/tree-vect-stmts.c +++ b/gcc/tree-vect-stmts.c @@ -1886,6 +1886,9 @@ vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, it defines is mapped to the new definition. So just replace rhs of the statement with something harmless. */ + if (slp_node) + return true; + type = TREE_TYPE (scalar_dest); if (is_pattern_stmt_p (stmt_info)) lhs = gimple_call_lhs (STMT_VINFO_RELATED_STMT (stmt_info)); @@ -1893,8 +1896,7 @@ vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, lhs = gimple_call_lhs (stmt); new_stmt = gimple_build_assign (lhs, build_zero_cst (type)); set_vinfo_for_stmt (new_stmt, stmt_info); - if (!slp_node) - set_vinfo_for_stmt (stmt, NULL); + set_vinfo_for_stmt (stmt, NULL); STMT_VINFO_STMT (stmt_info) = new_stmt; gsi_replace (gsi, new_stmt, false); SSA_NAME_DEF_STMT (gimple_assign_lhs (new_stmt)) = new_stmt; -- 2.30.2