http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50155
Bug #: 50155
Summary: [4.7 Regression] AVX2 support broke -mavx
Classification: Unclassified
Product: gcc
Version: 4.7.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: [email protected]
ReportedBy: [email protected]
CC: [email protected], [email protected]
void
foo (int x, double *a, double *b, double c)
{
int i;
for (i = 0; i < x; i++)
*a++ = *b++ * i / c;
}
fails to assemble with -O3 -mavx -m64 on x86_64-linux (at least when using
binutils that don't support AVX2).
vpaddd %ymm6, %ymm1, %ymm5
is AFAIK an AVX2 insn, not AVX, yet it is emitted by the addv8si3 pattern
(*<plusminus_insn><code>3 insn etc. was previously using VI_128 mode iterator,
but now is using VI, guess it wants to use a mode iterator similar to VI, but
which uses TARGET_AVX2 instead of TARGET_AVX guards for the 32 byte integer
modes (VI_AVX2 or VI_1248_AVX2?)).
