Hi!
On 2025-02-08T13:17:55+0100, I wrote:
> pushed to trunk branch commit 6312165650091a4df34668d8e2aaa0bbc4008a66
> "GCN, nvptx: 'sorry, unimplemented: exception handling not supported'"
> For GCN, this avoids ICEs further down the compilation pipeline.
For the record, in case that's helpful later on, here's a note from
~2023-04:
| Before [...], we got a lot of ICEs in 'g++' testing for '-fexceptions' etc.
| For example, 'g++.dg/pr49847.C':
|
| $ build-gcc/gcc/xg++ -Bbuild-gcc/gcc/
source-gcc/gcc/testsuite/g++.dg/pr49847.C -std=gnu++98 -O -fnon-call-exceptions
-Wno-return-type -S -o pr49847.s
| during RTL pass: jump
| source-gcc/gcc/testsuite/g++.dg/pr49847.C: In function ‘int f(float)’:
| source-gcc/gcc/testsuite/g++.dg/pr49847.C:7:1: internal compiler error:
Segmentation fault
| 7 | }
| | ^
| 0x1216aaf crash_signal
| [...]/gcc/toplev.cc:314
| 0x1ef5d23 count_reg_usage
| [...]/gcc/cse.cc:6757
| 0x1ef5f0a count_reg_usage
| [...]/gcc/cse.cc:6797
| 0x1efba4c delete_trivially_dead_insns(rtx_insn*, int)
| [...]/gcc/cse.cc:7028
| 0x1ea4e36 execute
| [...]/gcc/cfgcleanup.cc:3237
|
| ..., and likewise for a lot of other 'g++' test cases, but also
| 'gcc.dg/pr104464.c', 'gcc.dg/uninit-pr106881.c', 'gcc.dg/torture/pr105484.c'.
|
| The SIGSEGV is due to 'REGNO ("reg:DI -1") == INVALID_REGNUM', and
| 'INVALID_REGNUM == (~(unsigned int) 0)', which doesn't work in
| 'count_reg_usage':
|
| counts[REGNO (x)] += incr;
|
| Trying to work around this locally is not sufficient; further ICEs down the
| line.
|
| The 'INVALID_REGNUM' is due to 'gcc/defaults.h:EH_RETURN_DATA_REGNO'.
Grüße
Thomas
