[Qemu-devel] FPU Accuracy on i386

[Joel Sherrill]

Hi,

I hope this is the right mailing list.

I am using the precompiled 0.91 image to do
automated testing of i386-rtems.  In the Ada
validation suite, 17 of the 21 failures appear
to be due to FP calculation accuracy.  Here
is a typical test output for these:

,.,. CXG2002 ACATS 2.5 88-01-01 00:00:00
---- CXG2002 Check the accuracy of the complex modulus function.
  * CXG2002 test 1 -- abs(bigreal + 0i) actual:
               +Inf******************* expected:
               1.18973149535723177E+4932 difference:
               -Inf******************* max_err:
               3.86972841582221554E+4913.
  * CXG2002 test 2 -- abs(0 + bigreal*i) actual:
               +Inf******************* expected:
               1.18973149535723177E+4932 difference:
               -Inf******************* max_err:
               3.86972841582221554E+4913.
  * CXG2002 test 8 -- abs(1 + 1*i) actual:  1.41421356237309515E+00
               expected:  1.41421356237309505E+00 difference:
               -9.67108337857069955E-17 max_err:
               4.59988025005012224E-19.
**** CXG2002 FAILED ****************************.

I am running with the "-cpu 486" option on a 32-bit
x86 host. 

Are there known issues with FPU code in qemu that
I should be aware of?

Thanks.

--
Joel Sherrill, Ph.D.             Director of Research & Development
address@hidden        On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
  Support Available             (256) 722-9985