[Mono-bugs] [Bug 691057] New: Wrong result of SIMD operations on Intel Core/Core 2 while 100% proper on Opteron

[bugzilla_noreply at novell.com]

https://bugzilla.novell.com/show_bug.cgi?id=691057

https://bugzilla.novell.com/show_bug.cgi?id=691057#c0


           Summary: Wrong result of SIMD operations on Intel Core/Core 2
                    while 100% proper on Opteron
    Classification: Mono
           Product: Mono: Runtime
           Version: 2.10.x
          Platform: x86
        OS/Version: All
            Status: NEW
          Severity: Major
          Priority: P5 - None
         Component: JIT
        AssignedTo: lupus at novell.com
        ReportedBy: konrad.kruczynski at gmail.com
         QAContact: mono-bugs at lists.ximian.com
                CC: konrad.kruczynski at gmail.com
          Found By: Community User
           Blocker: ---


Created an attachment (id=427423)
 --> (http://bugzilla.novell.com/attachment.cgi?id=427423)
Original and simdified version of C# n-body benchmark

Description of Problem:
I wanted to contribute a simdified version of n-body benchmark to
http://shootout.alioth.debian.org/u32/performance.php?test=nbody
Program was completed and tested on AMD Opteron (with Mono 2.10.1) and worked
perfectly. Results was 100% ok and was a lot faster than nonsimdified
(original) version. I contributed new version to benchmark (it still waits for
evaluation). In the meantime I also checked program on Intel Core. I was
surprised when not only it worked much slower, but also gave improper results.
On shootout one can see very similar simdified version of n-body written in C,
which gives correct results, so it is not just simd related problem. Also, as I
said, on AMD Opteron, everything is fine. Both programs are short and clear.


Steps to reproduce the problem:
1. Compile attached nbody.cs (original program) and nbody-simd.cs (referencing
Mono.Simd).
2. Run both programs.


Actual Results:
Result of both programs should be identical regardless of the CPU.

Expected Results:
Proper result on AMD Opteron, improper on Intel Core/Core 2 (none others
tested).

How often does this happen? 
Always

Additional Information:
Compiled EXE file was moved from one platform to another, so this is not
related with msc, but rather JIT. Tested on Windows and Linux (Ubuntu, Gentoo)
with same results.

-- 
Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.