https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #21 from jacobly@gmail.com ---
Created attachment 110721
--> https://bugs.kde.org/attachment.cgi?id=110721=edit
Test mem variants of instructions in test-amd64.
Sorry, I had not noticed the switch to git. Here's a git patch against
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #20 from Julian Seward ---
(In reply to jacobly.alt from comment #18)
> Created attachment 110697 [details]
> Test mem variants of instructions in test-amd64.
Is this a patch relative to the git trunk, or a revised version
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #19 from Julian Seward ---
Fix committed, 6ae2edea014669d8082747f0f268e9404e0fd296.
--
You are receiving this mail because:
You are watching all bug changes.
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #18 from jacobly@gmail.com ---
Created attachment 110697
--> https://bugs.kde.org/attachment.cgi?id=110697=edit
Test mem variants of instructions in test-amd64.
I think this triggers a latent bug in the test code with string
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #17 from Julian Seward ---
(In reply to Niels Möller from comment #15)
> But I've run the mini-gmp tests [..]
Good, thanks. That's good enough for me. I'll land the fix.
--
You are receiving this mail because:
You are
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #16 from Julian Seward ---
(In reply to jacobly.alt from comment #14)
> I added mem,reg and reg,mem tests to test-amd64 [..]
Great! Can you pls make the testcase diffs available?
--
You are receiving this mail because:
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #15 from Niels Möller ---
Patch works fine for me.
I haven't been able to run the complete gmp testsuite under valgrind (some
years ago, automake supported that with "make check
TESTS_ENVIRONMENT=valgrind", but now
https://bugs.kde.org/show_bug.cgi?id=384930
jacobly@gmail.com changed:
What|Removed |Added
CC||jacobly@gmail.com
--- Comment #14
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #13 from Julian Seward ---
(In reply to Niels Möller from comment #12)
git clone git://sourceware.org/git/valgrind.git trunk
cd trunk
./autogen.sh && ./configure --prefix=`pwd`/Inst && make -j8 && make -j8 install
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #12 from Niels Möller ---
I've never built valgrind from source before, but I'll give it a try, maybe
already this evening.
--
You are receiving this mail because:
You are watching all bug changes.
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #11 from Julian Seward ---
Niels, the patch below makes your reduced testcase run correctly.
Can you test it more widely on GMP and let me know if it is OK?
Then I'll land it.
$ git diff -U8
diff --git
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #10 from Julian Seward ---
Argh. V doesn't actually write the computed result to the
destination register in the case where one of the sources is
a memory operand, for adox and adcx. Eg
adoxq 48(%rdi),%r12
%r12 is
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #9 from Julian Seward ---
On further analysis and testing, all 3 insns -- adcx, adox and mulx --
appear to be correctly implemented. So now I'm even more mystified.
--
You are receiving this mail because:
You are
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #8 from Niels Möller ---
The large errors are hard to explain just from incorrect carries applied by
adcx and adox (unless internal representation of O and C flag can somehow get
values far beyond {0,1}).
One
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #7 from Julian Seward ---
It seems likely to me, from looking at the implementation, that
V does not correctly preserve the OSZAP flags after ADCX, nor the
SZACP flags after ADOX. At least, that's my current theory.
I'll
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #6 from Julian Seward ---
Reproduced. Thank you very much to whoever reduced the testcase.
--
You are receiving this mail because:
You are watching all bug changes.
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #5 from Niels Möller ---
Created attachment 110629
--> https://bugs.kde.org/attachment.cgi?id=110629=edit
Main program calling assembly mul_basecase.
--
You are receiving this mail because:
You are watching all bug
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #4 from Niels Möller ---
Created attachment 110628
--> https://bugs.kde.org/attachment.cgi?id=110628=edit
GMP assembly mul_basecase.asm after m4 preprocessing
--
You are receiving this mail because:
You are
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #3 from Niels Möller ---
In the mean time, a more self-contained example was posted by the GMP author to
https://gmplib.org/list-archives/gmp-devel/2018-February/004728.html, I'm
copying here for easy reference. To
https://bugs.kde.org/show_bug.cgi?id=384930
--- Comment #2 from Niels Möller ---
Created attachment 110579
--> https://bugs.kde.org/attachment.cgi?id=110579=edit
Disassembly of gmp's mul_basecase.o, for the broadwell cpu family.
--
You are receiving this mail because:
https://bugs.kde.org/show_bug.cgi?id=384930
Niels Möller changed:
What|Removed |Added
CC||ni...@lysator.liu.se
---
21 matches
Mail list logo