version 1.3, 2019/08/29 13:01:34
|
version 1.9, 2019/08/30 04:49:49
|
Line 1
|
Line 1
|
# Various GCC/vax issues and invesigation # |
# Various GCC/vax issues and invesigation # |
|
|
# libgomp crash with -O2 # |
[WIP tree](gcc-mirror/) |
|
|
|
## libgomp crash with -O2 ## |
Repro: |
Repro: |
|
|
#!/bin/sh |
#!/bin/sh |
Line 18 Repro:
|
Line 20 Repro:
|
|
|
Asserts at emit-rtl.c:2310 gcc_assert (memory_address_addr_space_p (mode, addr, as)); |
Asserts at emit-rtl.c:2310 gcc_assert (memory_address_addr_space_p (mode, addr, as)); |
|
|
NetBSD avoids it with a [local diff](5f534f20bc66738c05c442a8f20088f55335b653/external/gpl3/gcc/dist/libgcc/gthr-posix.h#L418) to reload.c/dse.c. |
NetBSD avoids it with a [local diff](https://github.com/NetBSD/src/commit/e437e96750193b86d0464965661f616e011056fa) to reload.c/dse.c. |
It looks like a real bug in vax_legitimate_address_p not handling some addressing modes. |
It looks like a real bug in vax_legitimate_address_p not handling some addressing modes. |
The code to handle offset(reg)[index] might be off: it doesn't like if neither xfoo0, xfoo1 are constant. |
The code to handle offset(reg)[index] might be off: it doesn't like if neither xfoo0, xfoo1 are constant. |
|
|
|
## unrecognizable insn with any optimization ## |
|
|
|
Repro: |
|
|
|
#!/bin/sh |
|
|
|
cat << EOF > decl.ii |
|
struct a { |
|
long long *b() const; |
|
}; |
|
template <typename> class d : a { |
|
public: |
|
template <typename e> d(e); |
|
long long f() const; |
|
}; |
|
template <typename c> long long d<c>::f() const { |
|
int g; |
|
return b()[g]; |
|
} |
|
int h; |
|
bool i() { |
|
d<int> j(h); |
|
return j.f(); |
|
} |
|
EOF |
|
vax--netbsdelf-g++ -O2 -c decl.ii |
|
|
|
This errors out with: |
|
|
|
$ env PATH=$PWD:$PATH ./xgcc -fno-use-linker-plugin -c -O2 ~/oacc/decl.ii -fno-tree-vectorize |
|
/home/fly/oacc/decl.ii: In function 'bool i()': |
|
/home/fly/oacc/decl.ii:17:1: error: unrecognizable insn: |
|
17 | } |
|
| ^ |
|
(insn 13 12 14 2 (set (reg:SI 33) |
|
(subreg:SI (mem:DI (plus:SI (mult:SI (reg/v:SI 24 [ g ]) |
|
(const_int 8 [0x8])) |
|
(reg/f:SI 23 [ _6 ])) [1 *_10+0 S8 A32]) 4)) "/home/fly/oacc/decl.ii":16:14 -1 |
|
(nil)) |
|
during RTL pass: vregs |
|
/home/fly/oacc/decl.ii:17:1: internal compiler error: in extract_insn, at recog.c:2310 |
|
|
|
Providing a trivial scheduling will avoid subregs of mem. |
|
However, this doesn't work for us. Our crash and mem subregs happen at a too early pass, most likely. |
|
|
|
## DWARF ## |
|
|
|
flag_dwarf2_cfi_asm = 0; |
|
|
|
is kind of a big deal. We should probably get rid of it, but [binutils hates that.](https://github.com/NetBSD/src/commit/1df57dce3744cf4743358b098728a0658b80b8fb) |
|
|
|
ld: warning: dynamic relocation to `?' in readonly section `.eh_frame' |
|
ld: warning: dynamic relocation to `?' in readonly section `.eh_frame' |
|
ld: warning: creating a DT_TEXTREL in a shared object |
|
ld: .eh_frame_hdr refers to overlapping FDEs |
|
ld: final link failed: bad value |
|
|
|
## review comments ## |
|
|
|
Jeff Law [provided some feedback](https://gcc.gnu.org/ml/gcc/2019-04/msg00281.html). More of it should be addressed. |