diff options
| author | Thomas Schwinge <tschwinge@baylibre.com> | 2024-07-17 23:56:25 +0200 |
|---|---|---|
| committer | Thomas Schwinge <tschwinge@baylibre.com> | 2024-09-05 14:28:39 +0200 |
| commit | 8f5aade15e595b288a2c4ec60ddde8dc80df1a80 (patch) | |
| tree | 5f910d7d047109bf49b204ed26acb3e13f3366c7 /gcc/cp/parser.cc | |
| parent | a1865fd33897bc6c6e0109df0a12ee73ce386315 (diff) | |
| download | gcc-8f5aade15e595b288a2c4ec60ddde8dc80df1a80.zip gcc-8f5aade15e595b288a2c4ec60ddde8dc80df1a80.tar.gz gcc-8f5aade15e595b288a2c4ec60ddde8dc80df1a80.tar.bz2 | |
nvptx: Emit DECL and DEF linker markers for aliases [PR104957]
With nvptx '-malias' enabled (as implemented in
commit f8b15e177155960017ac0c5daef8780d1127f91c
"[nvptx] Use .alias directive for mptx >= 6.3"), the C++ front end in certain
cases does 'write_fn_proto' before an eventual 'alias' attribute has been
added. In that case, we do emit (via 'write_fn_marker') a DECL linker marker,
but then never emit a corresponding DEF linker marker for the alias. This
causes hundreds of instances of link-time 'unresolved symbol [alias]' across
the C++ test suite, which are regressions compared to a test run with (default)
'-mno-alias' (in which case the respective functions get duplicated).
PR target/104957
gcc/
* config/nvptx/nvptx.cc (write_fn_proto_1): Revert 2022-03-22
change; 'write_fn_marker' also for alias DECL.
(nvptx_asm_output_def_from_decls): 'write_fn_marker' for alias
DEF.
gcc/testsuite/
* g++.target/nvptx/alias-g++.dg_init_dtor2-1.C: Un-XFAIL.
* gcc.target/nvptx/alias-1.c: Likewise.
* gcc.target/nvptx/alias-3.c: Likewise.
* gcc.target/nvptx/alias-to-alias-1.c: Likewise.
Diffstat (limited to 'gcc/cp/parser.cc')
0 files changed, 0 insertions, 0 deletions