https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117957
Bug ID: 117957
Summary: vectorization pesimises std::vector push/pop test
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: hubicka at gcc dot gnu.org
Target Milestone: ---
my synthetic push/pop minibenchmark regressed due to vectorization:
jh@ryzen3:~> cat t.C
#include <vector>
typedef unsigned int uint32_t;
std::pair<uint32_t, uint32_t> pair;
void
test()
{
std::vector<std::pair<uint32_t, uint32_t>> stack;
stack.push_back (pair);
while (!stack.empty()) {
std::pair<uint32_t, uint32_t> cur = stack.back();
stack.pop_back();
if (!cur.first)
{
cur.second++;
stack.push_back (cur);
}
if (cur.second > 10000)
break;
}
}
int
main()
{
for (int i = 0; i < 10000; i++)
test();
}
jh@ryzen3:~> ~/trunk-install2/bin/g++ -O3 t.C ; time ./a.out
real 0m0.250s
user 0m0.250s
sys 0m0.000s
jh@ryzen3:~> ~/trunk-install2/bin/g++ -O3 t.C -fno-tree-vectorize ; time
./a.out
real 0m0.044s
user 0m0.044s
sys 0m0.000s
This is regression since gcc14
jh@ryzen3:~> g++ -O3 t.C ; time ./a.out
real 0m0.044s
user 0m0.044s
sys 0m0.000s
jh@ryzen3:~> g++ --version
g++ (SUSE Linux) 14.2.1 20241007 [revision
4af44f2cf7d281f3e4f3957efce10e8b2ccb2ad3]
Copyright (C) 2024 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
with clang I get 0.066s.
perf shows:
Percent│ nop ▒
│ 30:┌─→add $0x1,%ebx ▒
0.00 │ │ cmp %rsi,%rdx ▒
│ │↓ je 90 ▒
│ │ movdqa %xmm1,%xmm0 ▒
│ │ movd %ebx,%xmm2 ▒
8.23 │ │ punpckldq %xmm2,%xmm0 ◆
24.84 │ │ movq %xmm0,-0x8(%rax) ▒
│ 49:│ cmp $0x2710,%ebx ▒
│ │↓ ja 73 ▒
│ 51:│ cmp %rbp,%rax ▒
│ │↓ je 150 ▒
│ 5a:│ mov -0x8(%rax),%ecx ▒
58.52 │ │ mov -0x4(%rax),%ebx ▒
0.01 │ │ lea -0x8(%rax),%rdx ▒
│ ├──test %ecx,%ecx ▒
8.39 │ └──je 30 ▒
