[Bug target/57056] New: Missed optimization of finite finite builtin

[neleai at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57056

             Bug #: 57056
           Summary: Missed optimization of finite finite builtin
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: neleai@seznam.cz


Created attachment 29930
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29930
benchmark

Hi,

Current builtin finite is slower on attached benchmark than my version.
It also needs to load 64bit constants which is not friendly to code size. 
My version is:

int finite4(double x)
{
  uint64_t lx;
  EXTRACT_WORDS64(lx,x);
  lx=lx>>52;
  return ((lx&0x7ff)!=0x7ff);
}

Most of uses of isfinite()/finite() function are in condition so I benchmark
finite in condition.

Benchmark is run by

for i in `seq 0 7`; do echo finite$i;  gcc finite_bench.c -O3 -Wall -W
-fno-builtin-finite -Dfinite=finite$i; for j in `seq 1 8`; do /usr/bin/time -f
"%U" ./a.out; done; done