|
|
There are many possible combinations of math coprocessors and systems, some of which are known to cause problems under specific circumstances.
Because of design defects in the Intel 80386 chip (B1 stepping, part numbers 540344 and 540362), the Intel 80387 math coprocessor may not operate correctly in some computers, causing the CPU to hang during DMA, paging, or coprocessor access. You can work around this problem by changing the tunable kernel parameter DO387CR3 from 0 to 1 using the configure(ADM) utility. For more information, see ``Miscellaneous device drivers and hardware parameters''.
You can replace the 386 chip with a newer release of the 386 chip (a D-step part), or bypass the 387 chip by adding the ignorefpu keyword to the boot command:
Boot : unix ignorefpuThis means that the operating system will not use the 387 chip, but you need not remove it physically; the coprocessor can still be used by other operating systems. To bypass the 387 chip automatically every time you boot your system, add the ignorefpu keyword to the defbootstr option in the /etc/default/boot file. See the boot(HW) manual page for more information.
SCO OpenServer systems are not known to be affected by the reported problems with Pentium math functions. However, users who intend to run math-intensive applications on SCO OpenServer are advised to check with their system supplier.
The Intel RapidCAD Engineering Coprocessor for Intel 386DX PCs is supported.
Weitek numeric coprocessors are also supported. This support extends only to runtime; there is no current development support for creating binaries that take advantage of numeric coprocessors.
In order for the Weitek chip to be recognized by the system, one file needs to be edited in the following way:
weitek N 1 0 0 ...Change the ``N'' (for no) to a ``Y'' (for yes).