https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127579
Bug ID: 127579
Summary: [16/17 Regression] ICE when building with
-mgeneral-regs -march=armv9-a since
g:b3355804c7054bce288b643ce127ec5e9d0c2061
Product: gcc
Version: 16.2.1
Status: UNCONFIRMED
Keywords: ice-on-valid-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: tnfchris at gcc dot gnu.org
Target Milestone: ---
The following testcase
char a;
int b;
void e() {
char *c = (char *)&b;
int d = 0;
for (; d < b; d++)
a += c[d];
}
compiled with -O2 -mgeneral-regs-only -march=armv9-a ices with a segfault
internal compiler error: Segmentation fault
// 446 | ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
// | ^~~~~~~~~~~~~~~~~~~~
// 0xf3fc23 internal_error(char const*, ...)
// ../../src/gcc/diagnostic-global-context.cc:787
// 0x976543 crash_signal
// ../../src/gcc/toplev.cc:325
// 0x1e43f74 info_for_reduction(_loop_vec_info*, _slp_tree*)
// ../../src/gcc/tree-vect-loop.cc:5060
// 0x1a7946f aarch64_force_single_cycle
// ../../src/gcc/config/aarch64/aarch64.cc:18284
// 0x1a7946f aarch64_vector_costs::count_ops(unsigned int, vect_cost_for_stmt,
_stmt_vec_info*, _slp_tree*, aarch64_vec_op_count*)
// ../../src/gcc/config/aarch64/aarch64.cc:18314
// 0x1a4bfab aarch64_vector_costs::add_stmt_cost(int, vect_cost_for_stmt,
_stmt_vec_info*, _slp_tree*, tree_node*, int, vect_cost_model_location)
// ../../src/gcc/config/aarch64/aarch64.cc:18782
// 0x1da80e3 add_stmt_cost(vector_costs*, int, vect_cost_for_stmt,
_stmt_vec_info*, _slp_tree*, tree_node*, int, vect_cost_model_location)
// ../../src/gcc/tree-vectorizer.h:2099
// 0x1da80e3 add_stmt_costs(vector_costs*, vec<stmt_info_for_cost, va_heap,
vl_ptr>*)
// ../../src/gcc/tree-vectorizer.h:2129
// 0x19103a7 vect_slp_analyze_operations(vec_info*)
// ../../src/gcc/tree-vect-slp.cc:9299
// 0x19cfcaf vect_analyze_loop_2
// ../../src/gcc/tree-vect-loop.cc:2358
// 0x1de8807 vect_analyze_loop_1
// ../../src/gcc/tree-vect-loop.cc:2826
// 0x199ca13 vect_analyze_loop(loop*, gimple*, vec_info_shared*)
// ../../src/gcc/tree-vect-loop.cc:3007
// 0x1de2a2b try_vectorize_loop_1
// ../../src/gcc/tree-vectorizer.cc:1097
// 0x1de2a2b try_vectorize_loop
// ../../src/gcc/tree-vectorizer.cc:1216
// 0x1997683 execute
// ../../src/gcc/tree-vectorizer.cc:1333
since g:b3355804c7054bce288b643ce127ec5e9d0c2061
commit b3355804c7054bce288b643ce127ec5e9d0c2061
Author: Richard Biener
Introduce abstraction for vect reduction info, tracked from SLP nodes
