https://llvm.org/bugs/show_bug.cgi?id=24057
Bug ID: 24057
Summary: Missed optimization for explicit little-endian loads
Product: clang
Version: 3.5
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Frontend
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
--- begin code ---
struct eightbytes {
unsigned char bytes[8];
} __attribute__((aligned(8)));
unsigned long long load_le64(const struct eightbytes *p)
{
return ((unsigned long long)p->bytes[0] |
(unsigned long long)p->bytes[1] << 8 |
(unsigned long long)p->bytes[2] << 16 |
(unsigned long long)p->bytes[3] << 24 |
(unsigned long long)p->bytes[4] << 32 |
(unsigned long long)p->bytes[5] << 40 |
(unsigned long long)p->bytes[6] << 48 |
(unsigned long long)p->bytes[7] << 56);
}
--- end code ---
GCC 5.1.1 (and 5.0 as well, I think) generates the obvious optimized code for
load_le64. In fact, GCC 5.1.1 also gets the big-endian version right. clang
generates a bunch of zero-extending loads, shifts, and ors.
Given that this more or less the canonical way of writing portable
serialization/deserialization code without relying on undefined or unspecified
behavior, it would be nice if clang generated good code for it.
Feel free to reassign from clang to LLVM if appropriate.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
