[llvm-bugs] [Bug 31777] New: Union store codegen is silly

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=31777

            Bug ID: 31777
           Summary: Union store codegen is silly
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Common Code Generator Code
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-bugzilla at jfbastien.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Consider this code:

#include <inttypes.h>
#include <string.h>

union U {
  uint64_t i;
  uint16_t s;
};

void foo(U* u) {
  u->i = 0;
  u->s = 42;
}

void bar(U* u) {
  memset(u, 0, sizeof(U));
  u->s = 42;
}

void baz(U* u) {
  u->s = 42;
}


Let's ignore that foo could ignore the zero stores (because i isn't the active
member after the store to s): the codegen for this is pretty suboptimal
(https://godbolt.org/g/yS10Ki). The user's intent was to zero out padding bits,
and in many architectures you only need one store to do so (with, in some
cases, one materialization of the value).

x86-64:

foo(U*):                              # @foo(U*)
        movq    $0, (%rdi)
        movw    $42, (%rdi)
        retq

bar(U*):                              # @bar(U*)
        movq    $0, (%rdi)
        movw    $42, (%rdi)
        retq

baz(U*):                              # @baz(U*)
        movw    $42, (%rdi)
        retq

ARMv8:

foo(U*):
        mov     r1, #0
        str     r1, [r0]
        str     r1, [r0, #4]
        mov     r1, #42
        strh    r1, [r0]
        bx      lr

bar(U*):
        mov     r1, #0
        str     r1, [r0]
        str     r1, [r0, #4]
        mov     r1, #42
        strh    r1, [r0]
        bx      lr

baz(U*):
        mov     r1, #42
        strh    r1, [r0]
        bx      lr



I'd expect foo and bar to look like what GCC 7 trunk does:

foo(U*):
        movq    $42, (%rdi)
        ret

-- 
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/20170126/ce23eb27/attachment-0001.html>