On Mon, 28 Nov 2005, Tim Blechmann wrote:

I should've said this: just because something is a bug in the compiler doesn't mean we don't have to take it into account. Trying to compile devel_0_39 with simd=1 seems to be really, really troublesome with most versions of GCC I encounter. Most people who come to me with problems compiling devel_0_39 have that compiler error problem, so I have to tell them to do scons simd=0, except that sometimes they're trying devel_0_39 *because* of SIMD.

And then, I tried using GCC 4.0.2 for a while, but encountered programs that wouldn't compile anymore and that looked like real trouble to get to compile with GCC 4.x, so I reverted to GCC 3.4, and then encountered fundamental incompatibilities in libstdc++... and I don't know how to keep libstdc++'s from conflicting, short of static linking.

I'd recommend that simd=1 only enables SIMD asm code, and that a new option moresimd=1 would enable GCC options for SIMD generation, and would default to moresimd=0 until versions of GCC that people can actually use also can compile with SIMD enabled!