On Wed, Feb 17, 2021 at 2:16 PM Martin Liška <mli...@suse.cz> wrote:
>
> On 2/17/21 9:36 AM, Martin Liška wrote:
> > I'll write it even more robust...
>
> This is more elegant approach I've just tested on the instrumented clang.
>
> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
>
> Ready to be installed?
Isn't it still too complicated? We're asked to write N counters so why
don't we just write N counters? Thus, you already do
for (struct gcov_kvp *node = (struct gcov_kvp *)(intptr_t)start;
- node != NULL; node = node->next)
+ node != NULL; node = node->next, j++)
{
gcov_write_counter (node->value);
gcov_write_counter (node->count);
+
+ /* Stop when we reach expected number of items. */
+ if (j + 1 == list_sizes[i])
+ break;
}
why isn't this then only thing you need (just using pair_count as gathered
previously)? And just count pair_count to zero as IV? No need to
do the node != NULL check?
Note architectures with less nice memory ordering guarantees might
eventually see partially updated pointers and counters so I think
we at least want atomic_read ()s of the values with the weakest
consistency possible. (but that can be done as followup if we agree on that)
Richard.
> Thanks,
> Martin
