I now have access to a i386 without a math coprocessor, and was able
to test the math library I've been working on.

The machine is running NetBSD 0.9, so Wolfgang's fix to the emulator
is not present.  I do not know what effect it would have to the
results.

I am using the math regression tests from P.J. Plauger's "The Standard
C Library".  The tests are not as comprehensive as they might be, but
they do highlight the fact that the old math library has serious
problems with i386 machines without math coprocessors.

Here is the output of the tests with the old library:
  ./tmath1
  assertion "ceil (-5.1) == -5.0" failed: file "tmath1.c", line 30
  assertion "ceil (-4.9) == -4.0" failed: file "tmath1.c", line 32
  assertion "ceil (4.9) == 5.0" failed: file "tmath1.c", line 34
  assertion "ceil (5.1) == 6.0" failed: file "tmath1.c", line 36
  assertion "floor (-5.1) == -6.0" failed: file "tmath1.c", line 40
  assertion "floor (-4.9) == -5.0" failed: file "tmath1.c", line 42
  assertion "floor (4.9) == 4.0" failed: file "tmath1.c", line 44
  assertion "floor (5.1) == 5.0" failed: file "tmath1.c", line 46
  HUGE_VAL prints as Inf
  ./tmath2
  assertion "approx (acos (-rthalf), 3.0 * piby4)" failed: file "tmath2.c", line 29
  assertion "approx (acos (rthalf), piby4)" failed: file "tmath2.c", line 31
  assertion "approx (asin (-rthalf), -piby4)" failed: file "tmath2.c", line 35
  assertion "approx (asin (rthalf), piby4)" failed: file "tmath2.c", line 37
  ./tmath3
  assertion "approx (sqrt (0.5), rthalf)" failed: file "tmath3.c", line 58
  assertion "approx (sqrt (2.0), 1.0 / rthalf)" failed: file "tmath3.c", line 60
  assertion "approx (sqrt (144.0), 12.0)" failed: file "tmath3.c", line 61


Here is the output of the same tests with the fdlibm derived library:
  ./tmath1
  HUGE_VAL prints as Inf
  SUCCESS testing <math.h>, part 1
  ./tmath2
  assertion "approx (asin (-rthalf), -piby4)" failed: file "tmath2.c", line 35
  assertion "approx (asin (rthalf), piby4)" failed: file "tmath2.c", line 37
  ./tmath3
  SUCCESS testing <math.h>, part 3


Note that all three tests pass with "flying colours" on machines that
have a FPU; And that neither library was compiled with any special FPU
support (it exists, but it is not enabled since the emulator doesn't
support some of the instructions).

	--jtc


------------------------------------------------------------------------------