[llvm-bugs] [Bug 37611] New: nullptr check after pointer dereference not removed
via llvm-bugs
llvm-bugs at lists.llvm.org
Mon May 28 06:47:27 PDT 2018
https://bugs.llvm.org/show_bug.cgi?id=37611
Bug ID: 37611
Summary: nullptr check after pointer dereference not removed
Product: libraries
Version: trunk
Hardware: All
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: Scalar Optimizations
Assignee: unassignedbugs at nondot.org
Reporter: dave at znu.io
CC: llvm-bugs at lists.llvm.org
The following code is the distillation of C++ code expansion. This also
represents a decent performance opportunity for LLVM itself, given that this
bug affects LLVM's casting machinery.
The following code should generate the same assembly, regardless of whether it
is '#if 0' or '#if 1'. The correct code assembly code gen is the '#if 0' case.
The "bad" code generation includes needless nullptr preservation logic *after*
the pointer was dereferenced, thus proving the pointer is not null and need not
be preserved.
struct A { long a; };
struct B { long b; };
struct C : public A, public B { long c; };
struct Other { B *b; };
#if 1
C *cast_wrapper(B *b) {
return static_cast<C*>(b);
}
#else
#define cast_wrapper(b) static_cast<C*>(b)
#endif
long example(Other *o) {
auto b = o->b;
if (b->b != 42)
__builtin_trap();
return cast_wrapper(b)->a;
}
Here is a diff of the IR output. Negative is "bad" and positive is "good":
@@ -31,21 +20,16 @@
unreachable
; <label>:8: ; preds = %1
- %9 = icmp eq %struct.B* %3, null
- %10 = getelementptr inbounds %struct.B, %struct.B* %3, i64 -1
- %11 = bitcast %struct.B* %10 to %struct.C*
- %12 = select i1 %9, %struct.C* null, %struct.C* %11
- %13 = getelementptr inbounds %struct.C, %struct.C* %12, i64 0, i32 0, i32 0
- %14 = load i64, i64* %13, align 8, !tbaa !10
- ret i64 %14
+ %9 = getelementptr inbounds %struct.B, %struct.B* %3, i64 -1, i32 0
+ %10 = load i64, i64* %9, align 8, !tbaa !10
+ ret i64 %10
}
; Function Attrs: noreturn nounwind
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180528/82e428e8/attachment.html>
More information about the llvm-bugs
mailing list