Version: 3.6.0 (using KDE 1.2) OS: Linux The application I am attempting to profile is executing a blend packed words (PBLENDW) instruction, which according to http://ref.x86asm.net/coder64.html#x0F3A is part of the sse4.1 instruction set. Valgrind doesn't appear to be handling this instruction and the program runs fine without valgrind. The valgrind 3.6.0 release notes indicate that sse4.2 (two) is supported so I had assumed sse4.1 would be supported as well!? Thanks, Harry. Reproducible: Always valgrind 3.6.0 compiled from source using gcc 4.1.2. The application itself was built on another machine with gcc 4.3.4 (which supports sse4.x) I did not see a test against sse4.1 or sse4.2 in config.log - is that expected!?
Created attachment 54408 [details] c++ source file that, when compiled with gcc-snapshot's g++, reproduces the illegal instruction crash
I see a similar issue when trying to use valgrind 3.6.0 on Ubuntu 10.10 (via the natty packaging of it, here: https://launchpad.net/ubuntu/natty/amd64/valgrind/1:3.6.0-0ubuntu1). When I try to valgrind g++ from the gcc-snapshot package (via the natty version of gcc-snapshot, here: https://launchpad.net/ubuntu/natty/amd64/gcc-snapshot/20101206-1): matt@matt-desktop:~/src/devirt$ valgrind --trace-children=yes -q /usr/lib/gcc-snapshot/bin/g++ -O3 -fwhole-program -flto folding-test.cpp vex amd64->IR: unhandled instruction bytes: 0x66 0xF 0x3A 0x61 0x7 0x0 ==13008== valgrind: Unrecognised instruction at address 0xe0ea54. I have attached the folding-test.cpp to aid in reproducing the issue. I can't continue testing GCC trunk with valgrind until this is fixed. I would prefer to continue using valgrind 3.6.0, as it is *much* faster in 3.5.x in my typical scenarios.
(In reply to comment #2) > I see a similar issue when trying to use valgrind 3.6.0 on Ubuntu 10.10 (via > vex amd64->IR: unhandled instruction bytes: 0x66 0xF 0x3A 0x61 0x7 0x0 Totally unrelated, in fact. This is a PCMPESTRI instruction. Should be a separate bug report.
(In reply to comment #0) > The application I am attempting to profile is executing a blend packed words > (PBLENDW) instruction, which according to Fixed (vex r2075).
Created a new bug, as requested: https://bugs.kde.org/show_bug.cgi?id=262995