[Bug target/16978] New: Inefficient 32x32=>64 multiply and add on x86-32

["tomstdenis at iahu dot ca" <gcc-bugzilla at gcc dot gnu dot org>]

The following code [on an x86-32 target] 
 
typedef unsigned long mp_digit; 
typedef unsigned long long mp_word; 
void test(mp_word *W, mp_digit *x, mp_digit *y, int n) 
{ 
  int iz; 
  mp_word _W; 
  _W = 0; 
  for (iz = 0; iz < n; ++iz) { 
      _W += ((mp_word)*x++)*((mp_word)*y--); 
  } 
  *W = _W; 
} 
 
Demos what I'm talking about ... It produces 
 
.L6: 
        movl    (%ecx), %edx 
        movl    %edx, %eax 
        mull    (%ebx) 
        addl    %eax, %esi 
        adcl    %edx, %edi 
        movl    -4(%ecx), %edx 
        movl    %edx, %eax 
        mull    4(%ebx) 
        addl    %eax, %esi 
        adcl    %edx, %edi 
... 
 
The load of EDX is not required.  Since the multiplication is 32x32 all you 
have to do is load EAX.  This causes annoying stalls on the Athlon since edx 
=> eax => mull is just two stalls in a row. 
 
The correct optimization would be to load the next word into eax directly.

-- 
           Summary: Inefficient 32x32=>64 multiply and add on x86-32
           Product: gcc
           Version: 3.3.4
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tomstdenis at iahu dot ca
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: gcc version 3.3.4 20040623 (Gentoo Linux 3.3.4-r1, ssp-
                    3.3.2-2,
  GCC host triplet: Linux tombox 2.6.7 #1 SMP Mon Jul 26 15:43:24 EDT 2004
                    i686 Inte
GCC target triplet: gcc version 3.3.4 20040623 (Gentoo Linux 3.3.4-r1, ssp-
                    3.3.2-2,


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16978