-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ARM test failures #8402
Comments
Looks like there was a type in |
Probably some of these are the same memory error causing #8314. On Thu, Sep 18, 2014 at 9:09 AM, Viral B. Shah [email protected]
|
It looks slightly better now but still get a failure in numbers, strings, unicode, bitarray, random, math, parallel, and readdlm. The readdlm failure might be due to an old LLVM bug: http://llvm.org/bugs/show_bug.cgi?id=7048 |
The equality test seems broken, which is making the numbers test fail. This should be
|
Hmm, the first operand seems to be getting trucated in the lowering:
Maybe missing some flag to flisp? (works as expected on win32) |
Cc: @vtjnash |
can you try a bit pattern e.g. |
Could you say a bit more about trying the bit pattern? What should I do? |
replicate ihnorton's results but with |
|
(@ViralBShah fyi: I set a breakpoint in |
Never having been one to leave well-enough alone, I switched the crouton version of Ubuntu to utopic on my Chromebook. Partly I wanted to see if it would be possible to use more of the system libraries from Ubuntu 14.10, such as llvm-3.5, to make installation faster. I started with a compilation in the recommended way (i.e. my Make.user only includes ARM.inc) and it did compile. However, but the program dies on a segfault immediately after the splash screen.
|
alright, cool: so someone is definitely truncating that number. any idea who? for example, if you look at the |
False start - I think the half-output is just a I grab the LLVM Values then we see they are lowered correctly.
On the other hand, with the original
|
ok, i see the bug in jl_static_show. what is |
Should that commit have fixed the numbers test? I still get the same error.
What would you want |
Oops, second argument should have been Cdouble and code_native of our test function Apparently I wasn't awake this morning |
No change with Cdouble.
|
Here is something a little bit odd: putting this test inside a function on Win64 gives the wrong answer; doing the same on ARM gives the correct answer. Windows 64:
Arm v7 (master):
|
"Win64 gives the wrong answer", maybe not helping much, just confirming (after fixing UInt32->Uint32, is it a typo or do both work in never versions?) also incorrect on Linux in my 64-bit 0.3.2 (non-ARM): julia> f() = typemax(Uint64) == 2.0^64 julia> f() |
|
I am now using OpenBLAS here and weirdly enough I get this, which does not happen if I multiply 100x100 matrices.
|
Also stuff like |
@ViralBShah This sounds awfully like #9107. |
I am filing individual issues for each failure. |
Ok, I was dreading to do this, but finally got enough courage. Here's where we stand on ARM testing. I am running them one at a time to see which tests run all the way without failure, so that we can focus attention on a smaller set. A few more tests still remain to be run, and I will update with the full list next week.
The text was updated successfully, but these errors were encountered: