https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103515
Bug ID: 103515
Summary: Unexpected OPTION_MASK_SAVE_TOC_INDIRECT of
rs6000_isa_flag
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: linkw at gcc dot gnu.org
Target Milestone: ---
Test case:
gcc/testsuite/g++.dg/torture/pr81360.C
Option:
-fno-early-inlining -Os
For function rs6000_can_inline_p, I tried to test if the rs6000_isa_flag is
always the same (expected) as the TREE_TARGET_OPTION
(target_option_default_node)->x_rs6000_isa_flags) when the caller_tree is NULL.
The regression testing on LE/BE showed up this failure. By investigating it, I
noticed it's a bug. Since the command line option is -Os, we should not set
this
OPTION_MASK_SAVE_TOC_INDIRECT for most functions there:
/* If we can shrink-wrap the TOC register save separately, then use
-msave-toc-indirect unless explicitly disabled. */
if ((rs6000_isa_flags_explicit & OPTION_MASK_SAVE_TOC_INDIRECT) == 0
&& flag_shrink_wrap_separate
&& optimize_function_for_speed_p (cfun))
rs6000_isa_flags |= OPTION_MASK_SAVE_TOC_INDIRECT;
But after compiler parsing the function main with "#pragma GCC optimize
("-O0")", the rs6000_isa_flags would always get this
OPTION_MASK_SAVE_TOC_INDIRECT on, even if the context is not for the function
main.