gcc-14 bugs, pile 4

In a few weeks gcc-14 branch should see a 14.1.0 release.

Since November 2023 I encountered only 8 more bugs. That makes it 2 bugs per month. 4 time lower rate than I had last time.

summary

Bugs I saw (in discovery order):

• tree-optimization/112711: wrong code on llvm-16, bswap and assume(aligned).
• c++/112869: ICE on libmpt-0.7.3 when built with -std=c++20.
• tree-optimization/112991: ICE on p7zip due to value numbering bugs.
• bootstrap/113132: bootstrap build failure due to -Werror.
• bootstrap/113445: bootstrap comparison failure due to instruction scheduler changes.
• tree-optimization/114249: ICE on lvm2 (wrong type transform in SLP).
• c++/114439: icu4c build failure due to the initialization changes in c++.
• lto/114574: unbound ICE in -flto mode.

fun bug

Only one of 8 bug was a runtime failure on llvm-16 in __builtin_assume_aligned() handling code. It’s extracted from EndiaTest.cpp:

// $ cat EndianTest.cpp
typedef          int i32;
typedef unsigned int u32;

static inline void write_i32(void *memory, i32 value) {
  // swap i32 bytes as if it was u32:
  u32 u_value = value;
  value = __builtin_bswap32(u_value);

  // llvm infers '1' alignment from destination type
  __builtin_memcpy(__builtin_assume_aligned(memory, 1), &value, sizeof(value));
}

__attribute__((noipa))
static void bug (void) {
  #define assert_eq(lhs, rhs) if (lhs != rhs) __builtin_trap()

  unsigned char data[5];
  write_i32(data, -1362446643);
  assert_eq(data[0], 0xAE);
  assert_eq(data[1], 0xCA);
  write_i32(data + 1, -1362446643);
  assert_eq(data[1], 0xAE);
}

int main() {
    bug();
}

The optimization breaks simple store-32 / load-8 / compare sequence:

$ gcc/xg++ -Bgcc EndianTest.cpp -o bug -O0 && ./bug
$ gcc/xg++ -Bgcc EndianTest.cpp -o bug -O2 && ./bug
Illegal instruction (core dumped)

There gcc was too optimistic in assuming that __builtin_assume_aligned() returns address not aliased to input argument. As a result gcc erroneously removed dead-looking stores.

The fix drops overly optimistic assumption.

histograms

• tree-optimization: 3
• c++: 2
• bootstrap: 2
• lto: 1

We are back to the usual tree-optimization at the top of subsystems where bugs lurked.

parting words

This cycle felt easier than the previous three. It feels like gcc-14 is ready for release.

Since the time gcc-13 was released about a year ago I found 53 bugs (past piles 1, 2, 3). This is about one bug a week.

I think the most impactful gcc-14 change will probably be -Werror=implicit-int, -Werror=implicit-function-declaration and friends (see https://gcc.gnu.org/gcc-14/porting_to.html for more details). This will break quite a few old unmaintained but used everywhere C projects like zip, opensp, jam, directfb.

Have fun!