https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96717
--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
#include <vector>
void pop_many(std::vector<int>& v, unsigned n) {
for (unsigned i = 0; i < n; ++i) {
v.pop_back();
}
}
With GCC 10 at -O2 and above this was optimized to simply subtract n from the
v.end() pointer:
_Z8pop_manyRSt6vectorIiSaIiEEm:
.LFB883:
testq %rsi, %rsi
je .L1
salq $2, %rsi
subq %rsi, 8(%rdi)
.L1:
ret
But since r11-2238 it decrements the pointer n times:
_Z8pop_manyRSt6vectorIiSaIiEEm:
.LFB883:
testq %rsi, %rsi
je .L1
movq 8(%rdi), %rdx
xorl %eax, %eax
.p2align 4,,10
.p2align 3
.L4:
subq $4, %rdx
addq $1, %rax
movq %rdx, 8(%rdi)
cmpq %rax, %rsi
jne .L4
.L1:
ret
This regression is still present after the fixes for PR96721 and PR96722.
Should it be marked as [11/12 Regression] ?
