valgrind does not recognize opcode 0x660F3A21 (INSERTPS of SSE 4.1)

[Sebastian Scherer]

Binary package hint: valgrind

I wanted to use valgrind's memcheck to debug a program. Unfortunately, it relies on some SSE intrinsics.

Valgrind stops (unhandled instruction bytes) when it encounters this opcode: 0x66 0xF 0x3A 0x21. The opcode is in this case produced by gcc (4.4.1) from the instrinsic _mm_setr_ps, defined in xmmintrin.h. It seems to correspond to the mnemonic INSERTPS introduced in SSE 4.1.

I am using the 1:3.5.0-2ubuntu2 version of valgrind which is the latest one in Karmic Koala.

[Loïc Minier (lool)]

Would you mind attaching a test case?

[Alessandro Ghedini (ghedo)]

Valgrind has a quite complete support for SSE 4.1 now. Also, I cannot reproduce the problem with the following test case:

% cat insertps.c
int main(void) {
 asm(".byte 0x66, 0xF, 0x3A, 0x21");

 return 0;
}

Which is disassembled by gdb as:

   0x00000000004004a0 <+0>: mov eax,0x0
   0x00000000004004a5 <+5>: mov DWORD PTR [rsp-0x4],0x0
   0x00000000004004ad <+13>: insertps xmm0,xmm3,0x90