On Jan 20, 2014, Richard Biener <[email protected]> wrote:
> local_get_addr_clear_given_value (const void *v ATTRIBUTE_UNUSED,
> void **slot, void *x)
> {
> ! if (*slot != NULL
> ! && vt_get_canonicalize_base ((rtx)*slot) == x)
> *slot = NULL;
> return true;
> }
This part is ok
> *************** val_reset (dataflow_set *set, decl_or_va
> *** 2501,2507 ****
> gcc_assert (var->n_var_parts == 1);
> ! if (var->onepart == ONEPART_VALUE)
> {
> rtx x = dv_as_value (dv);
> void **slot;
> --- 2502,2509 ----
> gcc_assert (var->n_var_parts == 1);
> ! if (var->onepart == ONEPART_VALUE
> ! && local_get_addr_cache != NULL)
> {
> rtx x = dv_as_value (dv);
> void **slot;
But I think this one is wrong. You don't want to treat a one-part value
as if it wasn't one. If we have to discard locs and equivalences for a
one-part value that doesn't have any (because we don't even have a
local_get_addr_cache yet), you can *probably* just return right away,
because your job is already done. So I'd try:
if (var->onepart == ONEPART_VALUE)
{
if (local_get_addr_cache == NULL)
return;
> ! /* Compute reverse top sord order of the inverted CFG
> so that the data-flow runs faster. */
sord?
> ! rc_order = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
> bb_order = XNEWVEC (int, last_basic_block_for_fn (cfun));
> ! int num = inverted_post_order_compute (rc_order);
> ! for (i = 0; i < num; i++)
> bb_order[rc_order[i]] = i;
> free (rc_order);
This looks reasonable to me, but I'm a bit concerned that changes in the
iteration order could break assumptions present in the code. I can't
think of anything specific, but I haven't completed swapping this
code back in ;-)
--
Alexandre Oliva, freedom fighter http://FSFLA.org/~lxoliva/
You must be the change you wish to see in the world. -- Gandhi
Be Free! -- http://FSFLA.org/ FSF Latin America board member
Free Software Evangelist Red Hat Brazil Toolchain Engineer
