Bump GCC to 5.3.0

[palmer-dabbelt]

No description provided.

[palmer-dabbelt]

This doesn't build with "--enable-linux", I'll fix it.

[palmer-dabbelt]

@aswaterman Can you look at the first commit? I think it's OK, but I didn't read that thread. I'm a bit worried that "grep MEMMODEL_SYNC" doesn't find what I'd expect (something in aach64).

[aswaterman]

I read some of the bugzilla thread and I think we're OK. The test program

int test(int* foo, int* bar)
{
  int x = __sync_fetch_and_add(foo, 1);
  return *bar;
}

produces the assembly code

li a5,1
amoadd.w.sc zero,a5,0(a0)
lw a0,0(a1)
ret

and by my reading of the RISC-V memory model, the load cannot be perceived
to have executed before the AMO.

On Sun, Dec 27, 2015 at 8:03 PM, Palmer Dabbelt notifications@github.com
wrote:

@aswaterman https://github.com/aswaterman Can you look at the first
commit? I think it's OK, but I didn't read that thread. I'm a bit worried
that "grep MEMMODEL_SYNC" doesn't find what I'd expect (something in
aach64).

—
Reply to this email directly or view it on GitHub
#111 (comment)
.

[aswaterman]

So, if you've fixed the --enable-linux build, I'm OK to merge.