[llvm-bugs] [Bug 48692] New: Failure to optimize integer conversion pattern

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=48692

            Bug ID: 48692
           Summary: Failure to optimize integer conversion pattern
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: gabravier at gmail.com
                CC: llvm-bugs at lists.llvm.org

int64_t f(int64_t x)
{
    return ((x << 48 >> 48) & -65536) | (x & 65535);
}

This can be optimized to `return (x << 48 >> 48);` (i.e. `(int16_t)x`), see
also Alive2 test here: https://alive2.llvm.org/ce/z/L7z-JQ

Should be generalizable to shifts/and patterns of any size.

Note: I originally found this while investigating how GCC and LLVM optimize
this piece of code:

union reg {
    uint64_t r;
    uint16_t x;
};

reg cbw(reg dest)
{
    dest.r = (int16_t)dest.x;
    return dest;
}

Which (on little endian targets) GCC optimizes to `return (int16_t)x` and LLVM
optimizes to the sample code I gave above.

-- 
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/20210107/67cb2e6b/attachment.html>