[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 Martin Liška changed: What|Removed |Added Keywords|needs-bisection | CC||marxin at gcc dot gnu.org --- Comment #9 from Martin Liška --- Changed with r8-835-gb7cb2251a32e7af5.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 Andrew Pinski changed: What|Removed |Added Known to fail||4.9.4 Last reconfirmed||2022-01-07 Status|UNCONFIRMED |NEW Ever confirmed|0 |1 Known to work||5.1.0 Keywords||ice-on-valid-code, ||needs-bisection --- Comment #8 from Andrew Pinski --- Seems to have been fixed in GCC 5+. Be interesting to see what revision fixed this.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 --- Comment #6 from torvald at gcc dot gnu.org --- We can keep it open, but my guess is that it would need some volunteer to actively drive the backporting process. I.e., with a patch and testing. Given that TM is experimental, I think backporting fixes for TM has low priority on most people's todo list.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 torvald at gcc dot gnu.org changed: What|Removed |Added Version|5.0 |4.9.1 --- Comment #7 from torvald at gcc dot gnu.org --- Changed version to 4.9.1.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 Clemens Hammacher hammacher at cs dot uni-saarland.de changed: What|Removed |Added Status|RESOLVED|UNCONFIRMED Resolution|WORKSFORME |--- --- Comment #5 from Clemens Hammacher hammacher at cs dot uni-saarland.de --- I confirm that it seems to be fixed in SVN trunk. Changed version to 4.9.1 without further testing - I don't have the corresponsing versions on my machine.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 --- Comment #4 from Patrick Marlier patrick.marlier at gmail dot com --- GCC 4.8.3 and 4.9.1 still fail with an ICE. Please adjust the version in the PR and change the status. (I did not test 4.8.4 and 4.9.2 but I can test it).
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 torvald at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED CC||torvald at gcc dot gnu.org Resolution|--- |WORKSFORME --- Comment #3 from torvald at gcc dot gnu.org --- I can't reproduce this on SVN trunk (r220039), so I'll close this for now. I didn't spot anything bad when briefly looking at the generated code.
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594
--- Comment #1 from Patrick Marlier patrick.marlier at gmail dot com ---
Created attachment 33058
-- https://gcc.gnu.org/bugzilla/attachment.cgi?id=33058action=edit
reduced testcase
$ xgcc -Wall -Wextra -Wfatal-errors -O2 -fgnu-tm -S pr61594.c
pr61594.c: In function ‘fn1’:
pr61594.c:12:40: internal compiler error: in requires_barrier, at
trans-mem.c:1517
__attribute__((transaction_safe)) void fn1() { fn2(); }
[Bug libitm/61594] ICE (assertion failure) in trans-mem.c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61594 Patrick Marlier patrick.marlier at gmail dot com changed: What|Removed |Added CC||aldyh at gcc dot gnu.org, ||rth at gcc dot gnu.org --- Comment #2 from Patrick Marlier patrick.marlier at gmail dot com --- In requires_barrier, the tree is a TARGET_MEM_REF and the operand is a STRING_CST. Because STRING_CST is constant and cannot be modified, it does not need a barrier. So this could be solved like this: Index: gcc/trans-mem.c === --- gcc/trans-mem.c (revision 212229) +++ gcc/trans-mem.c (working copy) @@ -1512,7 +1512,7 @@ requires_barrier (basic_block entry_block, tree x, if (TREE_CODE (TMR_BASE (x)) != ADDR_EXPR) return true; x = TREE_OPERAND (TMR_BASE (x), 0); - if (TREE_CODE (x) == PARM_DECL) + if (TREE_CODE (x) == PARM_DECL || TREE_CODE (x) == STRING_CST) return false; gcc_assert (TREE_CODE (x) == VAR_DECL); /* FALLTHRU */ However I think that we should not instrument for all constants (INTEGER_CST, ...) and even for read-only. So I would prefer solving the issue like this: Index: gcc/trans-mem.c === --- gcc/trans-mem.c (revision 212229) +++ gcc/trans-mem.c (working copy) @@ -1512,7 +1512,7 @@ requires_barrier (basic_block entry_block, tree x, if (TREE_CODE (TMR_BASE (x)) != ADDR_EXPR) return true; x = TREE_OPERAND (TMR_BASE (x), 0); - if (TREE_CODE (x) == PARM_DECL) + if (TREE_CODE (x) == PARM_DECL || TREE_CONSTANT (x) || TREE_READONLY (x)) return false; gcc_assert (TREE_CODE (x) == VAR_DECL); /* FALLTHRU */ Note that 4.8 is also affected.
