[cfe-users] clang x86 assembler 8-bit displacements

[Zenith432 via cfe-users]

This time a question about the integrated assembler
------- code
  .code32
  .data
L1:
  .quad 88
.equ L2, . - L1
  .byte 29

  .text
  mov 55(%ebx), %eax
  mov L2(%ebx), %eax
-------- end code

clang -c -o file.o file.s gives this

---------
0000000000000000	8b4337          	movl	0x37(%rbx), %eax
0000000000000003	8b8308000000    	movl	0x8(%rbx), %eax
---------

Both displacements fit in a signed 8-bit displacement.  When a constant is given, cc1as know how to squeeze the opcode.  The label L2 is local, and known even for a 1-pass assembler.  Yet it doesn't know to squeeze the opcode, and uses a 32-bit signed displacement.

Same result if the symbols L1, L2 are in the .text section instead of .data.

Any way to tell cc1as to use the 8-bit displacement for an absolute symbol?  gnu assembler has a form "mov.d8", and I believe it also knows to choose the 8-bit displacement automatically in this situation.  clang doesn't seem to support .d8.

PS: Where in the clang sources is cc1as found?  I couldn't make it out.