https://sourceware.org/bugzilla/show_bug.cgi?id=33170
--- Comment #7 from Indu Bhagat <indu.bhagat at oracle dot com> ---
The assertion fail is because the SFrame generation code assumes that a
DW_CFA_restore must be restoring the state of the register to the one at the
beginning of the function. But in this case, the input asm has DW_CFA_restore
in the very beginning of the function itself (and that too together with a
preceding DW_CFA_offset for the same register..) with no initialized state for
SFrame functions to restore to...
$ cat 33170_testcase.s
.type
_ZN4llvm3pdb11SymbolCache31getOrCreateGlobalSymbolByOffsetEj.cold, @function
_ZN4llvm3pdb11SymbolCache31getOrCreateGlobalSymbolByOffsetEj.cold:
.LFSB9553:
.L4226:
.cfi_def_cfa 6, 16
.cfi_offset 3, -56
.cfi_offset 6, -16
.cfi_offset 12, -48
.cfi_offset 13, -40
.cfi_offset 14, -32
.cfi_offset 15, -24
.L4236:
.cfi_restore 12
.cfi_restore 13
...
Something is likely off on the generation side.
--
You are receiving this mail because:
You are on the CC list for the bug.
