On 8/7/20 8:33 PM, Jakub Jelinek wrote:
On Fri, Aug 07, 2020 at 08:04:57PM +0200, Aldy Hernandez wrote:
--- a/gcc/ipa-fnsummary.c
+++ b/gcc/ipa-fnsummary.c
@@ -82,7 +82,6 @@ along with GCC; see the file COPYING3. If not see
#include "gimplify.h"
#include "stringpool.h"
#include "attribs.h"
-#include <vector>
#include "tree-into-ssa.h"
This part won't apply (Gerald has changed it to #define INCLUDE_VECTOR
earlier I think).
--- a/gcc/vec.h
+++ b/gcc/vec.h
@@ -1281,7 +1281,10 @@ template<typename T, typename A>
inline size_t
vec<T, A, vl_embed>::embedded_size (unsigned alloc)
{
- typedef vec<T, A, vl_embed> vec_embedded;
+ struct alignas (T) U { char data[sizeof (T)]; };
+ typedef vec<U, A, vl_embed> vec_embedded;
+ static_assert(sizeof(vec_embedded) == sizeof(vec), "");
+ static_assert(alignof(vec_embedded) == alignof(vec), "");
s/(/ (/g on the above two lines please (GCC vs. libstdc++ coding
style differences).
I think it would be nice to see e.g. in -fdump-tree-gimple dump
on ipa-fnsummary.c what value_range ctors does it call for the auto_vec and
if that is what we want, other than that LGTM.
I see the following in the dump:
evaluate_properties_for_edge()
{
...
struct auto_vec known_value_ranges;
...
try
{
...
auto_vec<int_range<1>, 32>::auto_vec (&known_value_ranges);
...
It looks like the auto_vec constructor is calling the int_range<1>
constructor correctly:
auto_vec<int_range<1>, 32>::auto_vec (struct auto_vec * const this)
{
...
int_range<1>::int_range (D.130911);
The int_range<1> constructor also looks correct:
int_range<1>::int_range (struct int_range * const this, const struct
int_range & other)
{
*this = {CLOBBER};
{
_1 = &this->D.92184;
_2 = &this->m_ranges;
irange::irange (_1, _2, 1);
_3 = &this->D.92184;
_4 = &other->D.92184;
irange::operator= (_3, _4);
}
}
If y'all agree, I will push the patch with Jakub's suggested changes.
Aldy
