🚀 go-pugleaf

In <491@mfgfoc.uucp>, exodus@mfgfoc.uucp (Greg Onufer) wrote:

> 	  imull  12(%ebp),%eax,%eax

This has come up many times.  I think the register constraints are OK,
but the C output part is not.  Basically, the 386 has a general two
operand multiply, or a three operand if the destination is a register
and one of the sources is an immediate value and the other either a
register or memory.

Struggling through the Internals document leads me to suspect that
perhaps using "which_alternative" is a better way for this insn to
work.  This change ought to make gcc use the simpler method of
determining which imul form to output instead of trying to figure it
out by looking at operand 2.

This change isn't really tested yet (lack of good test cases), and I
don't understand .md files yet, so reader beware.  Please send/post
corrections!

*** /tmp/,RCSt1a08236	Sun Dec  4 23:41:17 1988
--- i386.md	Sun Dec  4 23:31:58 1988
***************
*** 896,903 ****
  {
!   if (rtx_equal_p (operands[0], operands[1])
!       && (GET_CODE (operands[2]) == MEM
! 	  || GET_CODE (operands[2]) == REG))
!     /* Assembler has weird restrictions.  */
      return AS2 (imul%L,%2,%0);
!   return AS3 (imul%L,%2,%1,%0);
  }")
--- 896,901 ----
  {
!   if (which_alternative == 0)
      return AS2 (imul%L,%2,%0);
!   else
!     return AS3 (imul%L,%2,%1,%0);
  }")
-- 
James R. Van Artsdalen      james@bigtex.cactus.org      "Live Free or Die"
Home: 512-346-2444 Work: 338-8789       9505 Arboretum Blvd Austin TX 78759