[LLVMbugs] [Bug 1821] New: llvm spills like crazy on fft code

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=1821

           Summary: llvm spills like crazy on fft code
           Product: new-bugs
           Version: unspecified
          Platform: PC
        OS/Version: All
            Status: NEW
          Keywords: code-quality
          Severity: enhancement
          Priority: P2
         Component: new bugs
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sabre at nondot.org
                CC: evan.cheng at apple.com, llvmbugs at cs.uiuc.edu


On Anton's new oourafft.c example, we run consistently ~20% slower than
GCC-compiled code.  Here's a sample cut-down testcase, we spill far more than
GCC on x86-32.  On x86-64, we generate an identical number of instructions,
presumably due to the extra regs available.

  void test(int n, double *a, double *w)
{
  int j, k1, k2;
  double wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
  double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;

  x0r = a[0] + a[2];
  x0i = a[1] + a[3];
  x1r = a[0] - a[2];
  x1i = a[1] - a[3];
  x2r = a[4] + a[6];
  x2i = a[5] + a[7];
  x3r = a[4] - a[6];
  x3i = a[5] - a[7];
  a[0] = x0r + x2r;
  a[1] = x0i + x2i;
  a[4] = x0r - x2r;
  a[5] = x0i - x2i;
  a[2] = x1r - x3i;
  a[3] = x1i + x3r;
  a[6] = x1r + x3i;
  a[7] = x1i - x3r;
}


-- 
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.