[LLVMbugs] [Bug 9370] [arm] unnecessary MOVW not optimized

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=9370

skoe <skoe at directbox.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|INVALID                     |

--- Comment #2 from skoe <skoe at directbox.com> 2011-03-02 13:05:34 CST ---

Please have a closer look again:

    movw    r0, #32796    @ r0 = 0x0000801c
    mov.w   r1, #-1
    movt    r0, #20480    @ r0 = 0x5000801c
    str     r1, [r0]
    movw    r0, #32796    @ r0 = 0x0000801c
    movt    r0, #20482    @ r0 = 0x5002801c
    str     r1, [r0]

If you remove the second MOVW you have exactly the same result, because the
lower 16 bits of r0 were correct already:

    movw    r0, #32796    @ r0 = 0x0000801c
    mov.w   r1, #-1
    movt    r0, #20480    @ r0 = 0x5000801c
    str     r1, [r0]
    movt    r0, #20482    @ r0 = 0x5002801c
    str     r1, [r0]

Remember:
MOVW: Rd[15:0] := imm16, Rd[31:16] = 0, imm16 range 0-65535
MOVT: Rd[31:16] := imm16, Rd[15:0] unaffected, imm16 range 0-65535

The optimize could check if the lower 16 bits have the correct value already
and replace the upper 16 bits only in this case.

I know, it's only one instruction difference. But if it is in an often called
interrupt handler it makes a different already :)

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.