Version: 3.2.3 (using KDE KDE 4.0.0) Installed from: Ubuntu Packages Compiler: gcc-4.1 OS: Linux We've ran into the issue http://bugs.kde.org/show_bug.cgi?id=113849 where the icc runtime library terminates, because the CPU reported by valgrind's emulation of the CPUID instruction return the description of an Athlon64 CPU. This CPU does not support SSE2 nor does is support SSE3. However, according to e.g. https://bugs.kde.org/show_bug.cgi?id=129358 valgrind currently supports SSE3 and supposedly it does support SSE2, too. So, IMHO the CUPID emulation of valgrind should report a CPU, which is capable of all these features making the reported CPU type coherent with valgrind's supported instruction est. The easiest wayto check this is to install the 'cupid' tool and comparing the output of 'cupid' and 'valgrind cpuid'. Regards, Wolfgang
> So, IMHO the CUPID emulation of valgrind should report a CPU, which is > capable of all these features making the reported CPU type coherent with > valgrind's supported instruction est. Yes, I agree. I'll maybe change it to report a Core 2 CPU. BTW, did this happen in 64-bit mode or 32-bit?
Created attachment 22983 [details] The output of cpuid on my core2, x86_64
Created attachment 22984 [details] Output of 'valgrind cpuid' on my core2, x86_64 I've attached the output of my cpuid and 'valgrind cpuid' output on my x86_64 core2 laptop. Since I do not have further insight into valgrind's internal architecture I'm unable to judge, whether the CPUID instruction has be dependent on the underlying hardware or not. (I.e. can valgrind execute an executable with SSE3 instruction on a hardware, that does not support SSE3 ?) TIA for dealing with this issue, Wolfgang
@julian: Now, I've groked your question. Yes, we do use a 64bit kernel and 64bit executables. The questionable progrm is a number crunching application, where we won't miss the 64bit advantages...
SSSE3 support has been added to the trunk, and the CPUID output is updated accordingly, so now it reports I-am-an-Intel-Core2. I believe this should solve the problem. Can you try it out?
Fixed. Fix will be in 3.3.1.