https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #18 from Luke Geeson ---
Indeed that is a plus - I have 167k concurrency tests that I feed through
LLVM/GCC -O1/2/3/fast and for GCC -Og that test compilation from the C/C++ to
PPC memory models. If you are interested I can provide t
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #17 from David Edelsohn ---
I'm glad that we have confirmed that the GCC and LLVM code generation for
PowerPC are correct for the memory model. And now your translation tool is more
accurate.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
Luke Geeson changed:
What|Removed |Added
Resolution|--- |INVALID
Status|NEW
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #15 from Luke Geeson ---
I am sorry to you all - I have wasted your time. It was a bug in the
translation tool.
In the future I will make absolutely sure that I check everything before
submitting a bug - this is an embarrassing mis
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #14 from David Edelsohn ---
The conditional branch always will proceed to the next instruction, so the code
that you showed in the PR is a correct "optimization" of the original code, but
the processor does execute the conditional br
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #13 from Luke Geeson ---
Please bear with me whilst I figure out what has happened here.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #12 from Andrew Pinski ---
(In reply to Luke Geeson from comment #11)
> Do you have a trunk build of binutils to hand to test? Godbolt doesn't
> support objdump. I will try to install binutils from source to reproduce.
Yes it does.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #11 from Luke Geeson ---
I'm using
`powerpc64le-linux-gnu-objdump -Dr --disassemble --section=.text
--no-show-raw-insn`.
installed using `binutils-powerpc64le-linux-gnu` under Ubuntu 20:04 (in a
docker container running on an Arm
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #10 from David Edelsohn ---
If I compile your testcase with either GCC 11.3 or GCC trunk, GCC produces
P1:
.LFB1:
.cfi_startproc
.localentry P1,1
pld 9,.LANCHOR0+8@pcrel
sync
lwz 9,0(9)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #9 from Luke Geeson ---
apologies typos:
bc -> bee
bee -> beq
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #8 from Luke Geeson ---
I have not, but I will contact him and link this discussion. In the meantime, I
read that page and provide some more testing. Consider the passage:
```
bc;isync: this is a very low-overhead and very weak form
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #7 from David Edelsohn ---
Have you reached out to Paul McKenney (now at Meta) who suggested the
instruction sequences to implement the C/C++ memory for PowerPC?
https://open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2745.html
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #6 from Luke Geeson ---
Apologies - I've been thinking in syncs and fences too much!
Yes I mean `lwsync`, for clarity I repeat the above so you know it is the
correct fix:
```
lukegeeson@machine:~/Dev/tv-dev/herdtools7$ cat test.li
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
David Edelsohn changed:
What|Removed |Added
Ever confirmed|0 |1
Status|UNCONFIRMED
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #4 from Luke Geeson ---
Add Wilco also addressed this for AArch64 in GCC:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108891
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #3 from Luke Geeson ---
According to the latest C/C++ and PPC models, yes.
If x was non-atomic, then this would be a racy (UB) test, but making x atomic
with relaxed order is well-defined according to the C model (and therefore any
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
Andrew Pinski changed:
What|Removed |Added
Target||powerpc64le
--- Comment #2 from Andrew
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111246
--- Comment #1 from Luke Geeson ---
Edit: The godbolt link points to the same bug in LLVM I am reporting, here is
the GCC godbolt link https://godbolt.org/z/Tan5jrvKa
18 matches
Mail list logo
