https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90178
Bug ID: 90178
Summary: Missed optimization: duplicated terminal basic block
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: alex.reinking at gmail dot com
Target Milestone: ---
The following short C program,
int* find_ptr(int* mem, int sz, int val) {
for (int i = 0; i < sz; i++) {
if (mem[i] == val) {
return &mem[i];
}
}
return nullptr;
}
compiles to the following on GCC (trunk) with -O3 -march=skylake on Godbolt.
find_ptr(int*, int, int):
mov rax, rdi
test esi, esi
jle .L4 ## Why not .L8?
lea ecx, [rsi-1]
lea rcx, [rdi+4+rcx*4]
jmp .L3
.L9:
add rax, 4
cmp rax, rcx
je .L8
.L3:
cmp DWORD PTR [rax], edx
jne .L9
ret
.L8:
xor eax, eax
ret
.L4:
xor eax, eax
ret
Godbolt link: https://godbolt.org/z/WczJ3J
Here the terminal basic blocks .L8 and .L4 are identical. It seems to me that
there is no benefit to keeping .L4 around, and jumps should be redirected to
.L8. Disabling AVX via -mno-avx eliminates the duplicate. However, a similar
code generation quirk exists in Clang for this program, so I apologize if there
is a microarchitectural subtlety I'm missing here.
Godbolt link for Clang comparison: https://godbolt.org/z/2uVZ8v
