https://bugs.llvm.org/show_bug.cgi?id=40845
Bug ID: 40845
Summary: Missed optimization with redundant overflow checks
Product: libraries
Version: trunk
Hardware: All
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: Scalar Optimizations
Assignee: unassignedb...@nondot.org
Reporter: alex.gay...@gmail.com
CC: llvm-bugs@lists.llvm.org
Godbolt link: https://godbolt.org/z/h2lkGE
The subtraction can be known not to overflow because it's derived from an
overflow checked addition. This was reduced/converted to C from this rust bug:
https://github.com/rust-lang/rust/issues/58692
One commenter suggested they thought the issue was in how GVN handled overflow
checked operations.
Code:
#include <stdlib.h>
extern void panic1();
extern void panic2();
extern void panic3();
size_t f(size_t a, size_t b) {
size_t x;
if (!__builtin_add_overflow(a, b, &x)) {
panic1();
__builtin_unreachable();
}
size_t r;
if (!__builtin_sub_overflow(x, a, &r)) {
panic2();
__builtin_unreachable();
}
if (r < 0) {
panic3();
__builtin_unreachable();
}
return x;
}
Assembly:
f(unsigned long, unsigned long): # @f(unsigned
long, unsigned long)
push rax
mov rax, rsi
add rax, rdi
jae .LBB0_3
cmp rax, rdi
jae .LBB0_4
pop rcx
ret
.LBB0_3:
call panic1()
.LBB0_4:
call panic2()
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
